APDL Code Thesis [PDF]

Zitiervorschau

EXPERT SYSTEM FOR OPTIMIZATION OF WELDING PROCESS OF THIN WALLED HSLA STEEL STRUCTURES

NAEEM ULLAH DAR 05-UET/PhD-ME-19

Supervisor Prof. Dr. M.M.I Hammouda

Department of Mechanical Engineering Faculty of Mechanical and Aeronautical Engineering University of Engineering & Technology, Taxila, Pakistan November 2009

EXPERT SYSTEM FOR OPTIMIZATION OF WELDING PROCESS OF THIN WALLED HSLA STEEL STRUCTURES

A Dissertation submitted in partial fulfillment of the requirement for the degree of Doctor of Philosophy

NAEEM ULLAH DAR 05-UET/PhD-ME-19

Supervisor Prof. Dr. M.M.I Hammouda

Department of Mechanical Engineering Faculty of Mechanical and Aeronautical Engineering University of Engineering & Technology, Taxila, Pakistan November 2009

DECLARATION

I certify that research work titled “Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures” is my own work. The work has not, in whole or in part, been presented elsewhere for assessment. Where material has been used from other sources it has been properly acknowledged/referred.

Naeem Ullah Dar 05-UET/PhD-ME-19

University of Engineering & Technology, Taxila-Pakistan

Acknowledgments Thanks of Al-Mighty-Allah who enable me to do such kind of research work. I am thankful to all those people who helped and supported me during this research and made it

possible. My deep gratitude goes to my PhD supervisor, respectable Professor Dr. M.M.I Hammouda for his supervision, continuous support, encouragement, research approach and expertise in this research work. The efforts and valuable guidance of Prof. Dr. M.M.I Hammouda made it possible for me not only to serve the purpose but also to enhance my research skills as well. I am very grateful for his inspiration and guidance throughout the research work. I am highly obliged to the invaluable help and guidance of Prof. Dr. M.H. Sahir, Dean (M & AE) for the finalization of all the PhD thesis/degree requirements as a supervisor in place of Prof. Dr. M.M.I Hammouda. My deep gratitude also goes to my Research Committee members, comprising of Dr. M. Zubair Khan, Dr. Saqlain A. Ghumman and Dr. Asif Iqbal, for their support, guidance and encouragement during the research work. I specially owe my deep gratitude to Dr. M. Ejaz Qureshi and Dr. Asif Iqbal for their support, technical discussions and sharing their knowledge & expertise specifically in the field of welding analysis/simulations and expert system respectively throughout the research work. I am really thankful to Engr. Abdul Sammi Khawaja and Engr. Mushtaq Hussain, his welding and testing team, for the support of all extensive experimental work. I am very thankful to respected Prof. M. Anwar Khan, Prof. Dr. M.H. Sahir, Prof. Dr. Shahab Khushnood, Prof. Sageer Ahmed and Dr. Riffat Asim Pasha for their encouragement and supports extended to me in all matters during my research work as well as a long time spent in the UET, Taxila. I am also grateful to Prof. Dr. M.A. Wahab, Louisiana State University, USA and Prof. Dr. Liu Fangjun, BUAA University, China for their guiding/encouraging comments on research work as foreign experts and official thesis evaluators. I would also like to thank of all my valuable teachers, officials of MED & ASR&TD, dear colleagues, my parents and family who always helped me and prayed for my success. I always remember their support and contributions for the completion of this research work. Engr. Naeem Ullah Dar

ii

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Dedication

Dedicated to my respectable parents, adoring wife & children and family members whose prays, love, sacrifices and encouragement is boundless.

iii

University of Engineering & Technology, Taxila-Pakistan

Abstract With the introduction of welding as joining method, the welding technology was applied as major joining technique in hi-tech industries to the welding of steels for manufacturing of different structures like pressure vessels and aerospace applications. Mostly high strength low alloy steels in thin cylindrical shell form are being used for aerospace structures due to high strength and low weight ratio. Despite being high strength and light weight by numerous advantages, the welding of thin walled structure of high strength low alloy steel (also known as HSLA Steel) comes also with a major problems of weld induced imperfections due to high temperatures like residual stresses and distortions with shortening of weld strength and it is a still major challenge for the welding professionals due to the complex nature of the welding phenomenon despite many innovations in welding technology. The most of the weld induced imperfections are the result of transient temperature distributions and subsequent cooling of the welds followed by transient and residual stress fields. Where as, the reliability of thin-walled structures used for any aerospace or pressure vessel application is on the prime importance every time for safe operational. Usually, thin walled cylindrical structures contain two types of weld as longitudinal and circumferential. The major design and industry constraints are weld strength and cost competitive. Gas Tungsten Arc Welding (GTAW) or TIG process is mostly applied due to the excellent weld strength and cost competitiveness. The main aim of this research work is to analyze and experimentally investigate the TIG welding parameters for purpose of minimizing residual stresses and distortion with the requirements of maximizing of weld strength of thin walled structures of HSLA steel respectively. To achieve the aforementioned targets, the following strategy was applied keeping in view the complex phenomena of welding, time and cost of extensive experimentations involved. Weld experiments were subdivided into linear and circumferential welding. Initially for linear welding, TIG welding parameters were analyzed to determine their significance on thin plates of HSLA steel of different thicknesses (3 to 5 mm) by following design of experiments (DOE) with employing 2-level full factorial and response surface method (RSM) designs to have response (weld strength, distortion & residual stress). Whereas for circumferential welding, a hybrid numerical simulation and experimental based analysis approach was employed to model and predict TIG welding process to investigate the transient temperature distributions, transient/residual stress fields and distortion for circumferentially welded thin-walled cylinders of HSLA steel. The simulations strategy was developed and implemented by using commercial available general purpose finite element software ANSYS® enhanced with subroutines. First thermal analysis was completed followed by a separate mechanical analysis based on the thermal history. From the three dimensional FE model developed for TIG welding process of circumferential welding, a series of virtual welding experiments based on statistical designs (DOE) were performed for response (residual stresses and distortion) with different thicknesses by using full factorial and RSM as applied for linear welding. The effects of following six parameters, four numeric and two categorical: welding current, welding voltage, welding speed, sheet/cylinder thickness and trailing (Ar) & weld type (linear and circumferential) were investigated upon following three

iv

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

performance measures: weld strength, residual stresses and distortions for different thicknesses of material of HSLA steel. The experimental results were analyzed using ANOVA and significance of effects of all the tested parameters upon performance measures was determined. Empirical models for weld strength, distortion and residual stresses, in terms of significant parameters, were also developed and numerical optimization was performed according to the desirability for the maximization of weld strength and minimization of distortion & residual stresses. All the statistical analyses were performed by using commercial available statistical software Design-Expert® and MINITAB®. From the results of post-experimental analyses, it was noticed that the effects of welding current, welding voltage and welding speed upon weld strength, residual stresses and distortion are extremely significant, while the effect of trailing and weld type is also considerably significant with respect to material thicknesses. The residual stresses are highly sensitive to heat input (weld temperatures). The residual stresses and distortion in circumferential welding are low as compared to linear welding for the same welding parameters and material thickness respectively. The vital recommendation, in this regard, is to use the parameters of welding resulting low input heat (low current, low voltage and high speed) with application of trailing with respect to material thicknesses for the maximum weld strength and minimum residual stresses and distortion in thin walled structures of HSLA steel. For the trade-off among aforementioned opposing targets and for prediction of values of performance measures at different settings of TIG welding parameters, the expert system tool, employing fuzzy reasoning mechanism, was utilized. Initially, an expert system was developed for the optimization of parameters according to objectives of maximization and/or minimization of weld strength, distortion and residual stresses. The expert system also provided the predicted values of various performance measures based upon the finalized values of the welding parameters. The analyses, simulations, experimental and ANOVA results were utilized for the making of fuzzy rule-base. The fuzzy rule-base was adjusted for maximum accuracy by employing the simulated annealing (SA) algorithm. In the next stage, a machine learning (ML) technique was utilized for creation of a expert system, named as EXWeldHSLASteel, that can: self-retrieve and self-store the experimental data; automatically develop fuzzy sets for numeric variables involved; automatically generate rules for optimization and prediction rule-bases; resolve the conflict among contradictory rules; and automatically update the interface of expert system according to newly introduced TIG welding process variables. The algorithms for these constituents were coded using a pointer-enabled language in C++. The coding involves a data structure named as doubly linked list, which provide the means for fast and efficient processing. The presented expert system is used for deciding the values of important welding process parameters as per objective before the start of actual welding process on shop floor. The user should be absolutely clear about the nature and requirements of any given TIG welding process, e.g., the setting parameters, fixed parameters, and geometric parameters etc. The expert system developed in the domain of welding for optimizing welding process of thin walled HSLA steel structure possesses all capabilities to adapt effectively to the unpredictable and continuously changing industrial environment of mechanical fabrication and manufacturing and to serve the newly emerging field of v

University of Engineering & Technology, Taxila-Pakistan

knowledge management by transforming individual (expert) organizational knowledge i.e. implicit to explicit knowledge.

Key words:

knowledge

into

TIG welding; thin-walled structures; HSLA steel; simulation; FEA; DOE, ANOVA; full factorial; RSM; optimization; maximization/minimization; expert system (ES); fuzzy logic; machine-learning (ML); weld strength; residual stresses; distortion; EXWeldHSLASteel

vi

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Contents Acknowledgements ........................................................................................................... ii Dedication ....................................................................................................................... iii Abstract ............................................................................................................................ iv Contents .......................................................................................................................... vii List of Figures .................................................................................................................. xi List of Tables ................................................................................................................ xvii List of Symbols ................................................................................................................ xx Chapters 1. INTRODUCTION .................................................................................................. 1 1.1 Welding Technology ........................................................................................ 1 1.1.1 Arc Welding ............................................................................................ 1 1.1.2 Physics of Arc Welding .......................................................................... 2 1.1.3 Development of Welding Technology..................................................... 3 1.1.4 Benefits & Threats of Welding ................................................................ 3 1.1.5 Weld Induced Residual Stresses .............................................................. 4 1.1.6 Weld Induced Distortions ........................................................................ 5 1.1.7 Weld Induced Residual Stresses and Distortions in Thin-Walled Structure ........................................................................ 7 1.2 GTAW or TIG Welding ................................................................................... 8 1.3 Variables and Performance Measures in GTAW Process ................................ 9 1.4 Current Challenges in Welding Domain .........................................................12 1.5 Application of Artificial Intelligence and Expert System to Manufacturing.. 14 1.5.1 Expert System ....................................................................................... 14 1.6 Objectives, Research Methodology and Organization of Dissertation............15 1.6.1 Objectives ............................................................................................. 15 1.6.2 Research Methodology ......................................................................... 16 1.6.3 Organization of Dissertation ................................................................. 19 2. LITERATURE REVIEW .................................................................................... 21 2.1 Introduction ................................................................................................... 21 2.2 Literature Survey in Welding Domain ........................................................... 21 2.2.1 Welding Experiments and Optimization ............................................... 22 2.2.2 Welding Simulations ............................................................................. 25 2.2.3 Circumferential Welding Computational Works ................................... 27 2.3 Welding Induced Residual Stresses Measurement ......................................... 33 2.3.1 Residual Stress Measurement by Hole-Drilling .....................................34 2.4 Literature Survey in Artificial Intelligence and Expert System to Manufacturing ................................................................................................ 35

vii

University of Engineering & Technology, Taxila-Pakistan

2.5 Chapter Summary and Conclusions ............................................................... 38 2.5.1 Limitations of Previously Developed AI Tools .................................... 39 3. ANALYZING & OPTIMIZING TIG WELDING PROCESS PARAMETERS .................................................................................................... 41 3.1 Introduction .................................................................................................... 41 3.2 Design of Experiments ................................................................................... 41 3.2.1 Predictor Variables ................................................................................ 42 3.2.2 Response Variables ............................................................................... 47 3.2.3 Fixed Parameters ................................................................................... 47 3.3 Experimental Setup ........................................................................................ 48 3.3.1 Characteristics of Base Metal and Filler Metal ..................................... 53 3.4 Experimental Results, ANOVA, Regression and Optimization .................... 54 3.4.1 Weld Strength ....................................................................................... 55 3.4.2 Distortion .............................................................................................. 62 3.4.3 Weld Induced Residual Stresses ........................................................... 70 3.5 Numerical Optimization and Emperical Modeling using Response Surface Method (RSM) ................................................................. 83 3.6 Chapter Summary and Conclusions ............................................................... 97 4. FE MODELING & SIMULATION OF GTAW PROCESS OF THIN WALLED STRUCTURE FOR CIRCUMFERENTIAL WELDING ............. 99 4.1 Introduction ................................................................................................... 99 4.2 FE Modeling & Simulation Methodology ..................................................... 99 4.2.1 Analytical Model of Arc Welding ...................................................... 100 4.2.2 FE Formulation .................................................................................... 101 4.2.3 Interaction of Different Fields ............................................................. 105 4.2.4 Heat Source Modeling and Efficiency ................................................ 105 4.2.5 Heat Losses Modeling ......................................................................... 108 4.2.6 Material Modeling .............................................................................. 109 4.2.7 Filler Metal Deposition ....................................................................... 113 4.2.8 Simulation Approach in ANSYS® ...................................................... 113 4.2.9 Welding Simulation Numerical Aspects ............................................. 116 4.3 Welding Induced Stresses and Distortions .................................................. 117 4.3.1 FE Discretization ................................................................................ 118 4.3.2 Other Simulation Aspects ................................................................... 120 4.3.3 Experimental Validation ..................................................................... 121 4.3.4 Thermal Effects of Welding ................................................................ 124 4.3.5 Welding Residual Stress Fields .......................................................... 125 4.3.6 Welding Distortions ............................................................................ 131 4.4 Experimental Setup for Validation of FE Models ....................................... 134 4.4.1 Experimental Setup ..............................................................................134 viii

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

4.4.2 FE Models Validation ......................................................................... 135 4.5 Chapter Summary and Conclusions ............................................................. 139 5. VIRTUAL DESIGN OF EXPERIMENTS (DOE) & OPTIMIZATION OF GTAW PROCESS OF THIN- WALLED STRUCTURE FOR CIRCUMFERENTIAL WELDING ................................................................... 141 5.1 Introduction ................................................................................................. 141 5.2 Analyzing the Effects of Welding Parameters on Residual Stresses and Distortions ............................................................................................. 141 5.2.1 Details of Parametric Studies .............................................................. 141 5.2.2 Welding Speed Effects ........................................................................ 144 5.2.3 Input Heat Effects ............................................................................... 146 5.2.4 Cylinder Thickness Effects ................................................................. 147 5.2.5 Root Opening Effects .......................................................................... 150 5.2.6 Tack Weld Orientation Effects ........................................................... 155 5.3 Virtual Design of Experiments (DOE) and Optimization of Circumferential Welding ............................................................................ 159 5.3.1 Virtual DOE ........................................................................................ 159 5.3.2 Predictor Variables .............................................................................. 159 5.3.3 Response Variables ............................................................................. 163 5.4 Virtual Experiments Results, ANOVA, Regression, and Optimization ...... 163 5.4.1 Distortion ............................................................................................ 163 5.4.2 Weld Induced Residual Stresses ......................................................... 171 5.4.3 Numerical Optimization & Emperical Modeling using RSM ............ 184 5.5 Linear & Circumferential Welding Optimization using RSM ..................... 192 5.6 Chapter Summary and Conclusions.............................................................. 206 6. KNOWLEDGE ENGINEERING FOR OPTIMIZING TIG WELDING PROCESS ...................................................................................... 207 6.1 Introduction .................................................................................................. 207 6.2 The Objectives of Expert System and Application to Welding ................... 207 6.3 Expert System Configuration ....................................................................... 208 6.3.1 Optimization and Prediction Modules ................................................ 209 6.3.2 The Expert System Shell ..................................................................... 210 6.3.3 The Procedure ..................................................................................... 210 6.4 Fuzzy Reasoning for the Expert System ...................................................... 211 6.4.1 Fuzzy Sets, Input Fuzzification, and Output Defuzzification ............. 211 6.4.2 Inference for Aggregation of Fuzzy Rules .......................................... 214 6.5 Optimal Formation of Rule-Base ................................................................. 217 6.5.1 Optimal Formation Using Simulated Annealing Algorithm ............... 217 6.5.2 Results of Optimal Formation of the Rule-Base ................................. 219 6.5.3 The Complete Rule-Base .................................................................... 219

ix

University of Engineering & Technology, Taxila-Pakistan

6.6 Application Example ................................................................................... 221 6.7 Chapter Summary and Conclusions ............................................................. 222 7. THE SELF-DEVELOPING EXPERT SYSTEM FOR OPTIMIZING TIG WELDING PROCESS ...................................................................................... 223 7.1 Introduction .................................................................................................. 223 7.2 Self-Development of Expert System ........................................................... 223 7.3 Data Acquisition Module.............................................................................. 224 7.4 Self-Development of Fuzzy Sets Module .................................................... 225 7.5 Self-Development of Prediction Rule-Base Module .................................... 227 7.5.1 Conflict Resolution among Contradictory Rules ................................ 229 7.6 Self-Development of Optimization Rule-Base Module ............................... 229 7.7 Data Structures and Coding ......................................................................... 231 7.7.1 Doubly Linked List ............................................................................. 232 7.8 Application Examples .................................................................................. 234 7.8.1 Example 1: A Fledgling Knowledge-Base ......................................... 235 7.8.2 Example 2: A Veteran Knowledge-Base ............................................ 241 7.8.3 Example 3: Verification of EXWeldHSLASteel Predictions ............. 242 7.8.4 Limitations of EXWeldHSLASteel .................................................... 243 7.9 Chapter Summary and Conclusions ............................................................. 244 8. CONCLUSIONS & RECOMMENDATIONS ................................................ 245 8.1 Conclusions .................................................................................................. 246 8.1.1 Welding Induced Stresses & Distortions and Weld Strength ............. 246 8.1.2 Effect of Welding Process Parameters ................................................ 247 8.1.3 The Expert System .............................................................................. 247 8.1.4 Researcher’s Main Contributions from the present Research Work .. 249 8.2 The Recommendations ................................................................................. 249 8.2.1 Proposals for Future Research ............................................................ 250 List of References ......................................................................................................... 251 List of Publications ...................................................................................................... 267 Appendix Appendix A1 APDL Code for Thermal/Structural Module .............................. 271 Appendix A2 ...................................................................................................... 282 A2-1: Pseudo-code of algorithm for Data Acquisition Module .............................. 282 A2-2: Pseudo-code of algorithm for Self-Development of Fuzzy Sets Module ..... 283 A2-3: Pseudo-code of algorithm for Self-Development of Prediction Rule-Base . 285 A2-4: Pseudo-code of algorithm for Conflict Resolution among Contradictory Rules ...................................................................................... 287 A2-5: Pseudo-code of algorithm for Self-Development of Optimization Rule-Base ... 287 A2-6: Auto-developed Fuzzy Sets & Optimization Rule-Base ............................... 290

x

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

List of Figures Fig. 1.1 Fig. 1.2 Fig. 1.3 Fig. 1.4 Fig. 1.5 Fig. 1.6 Fig. 1.7 Fig. 1.8 Fig. 2.1 Fig. 3.1 Fig. 3.2 Fig. 3.3 Fig. 3.4 Fig. 3.5 Fig. 3.6 Fig. 3.7 Fig. 3.8 Fig. 3.9 Fig. 3.10 Fig. 3.11 Fig. 3.12 Fig. 3.13 Fig. 3.14 Fig. 3.15 Fig. 3.16 Fig. 3.17 Fig. 3.18 Fig. 3.19 Fig. 3.20 Fig. 3.21 Fig. 3.22 Fig. 3.23 Fig. 3.24 Fig. 3.25 Fig. 3.26 Fig. 3.27 Fig. 3.28

Basic physics of arc welding ............................................................................ 2 Schematic view of residual stresses in welded rectangular plate .................... 4 Example of distortion that can occur during welding .................................…..6 Effect of the degree of clamping on the level of distortion & residual stresses ..........................................................................................…..6 a) Schematic view of expansion and shrinkage in circumferentially welded pipe/cylinders. b) Free body diagram of the weld joint …………………. 7 Principle of Tungsten Inert Gas (TIG) welding ............................................... 9 Thin-walled Pressure Vessel failures due to weld induced imperfections .... 13 Scope and methodological strategy framework of the research .................... 18 Axial residual stress plots on outer (left) and inner surface (right) ............... 31 Process showing the factors with multi-inputs and multi-outputs ................. 42 Weld quality upon initial setting of predictors .............................................. 43 Schematic diagram of the welding test plate ................................................. 47 SAF TIGMATE 270 Power Source ............................................................... 48 NERTAMATIC 300 TR Power Source.......................................................... 48 Sample Clamping Arrangement...................................................................... 49 Details of automatic TIG torch and clamping arrangements ......................... 49 Sample before Welding .................................................................................. 49 Sample after Welding .................................................................................... 49 The Distortion Measurement setup ................................................................ 50 Distorted Sample ............................................................................................ 50 Tensile Samples ............................................................................................. 50 Sample after Cutting Tensile Samples............................................................ 51 Tensile Sample (Machined) ............................................................................ 51 Tensile Sample after Testing ......................................................................... 51 Hole-drilling equipment & P-3500 strain indicator from Vishay Group........ 52 Welded specimens with mounted strain gage rosette .................................... 53 Two types of strain gage rosette .................................................................... 53 Residual Stresses Measurement setup ........................................................... 53 Weld Strength of Sixteen Experiments ( t = 3 mm) ...................................... 56 Weld Strength of Sixteen Experiments ( t = 4 mm) ....................................... 56 Weld Strength of Sixteen Experiments ( t = 5 mm) ....................................... 56 Effects of welding parameters upon Tensile Strength (t = 3 mm) ................. 58 Effects of welding parameters upon Tensile Strength (t = 4 mm).................. 59 Effects of welding parameters upon Tensile Strength (t = 5 mm).................. 59 Tensile Strength Predictions w.r.t Predictors (t = 3 mm) ............................... 61 Tensile Strength Predictions w.r.t Predictors (t = 4 mm) ............................... 62 Tensile Strength Predictions w.r.t Predictors (t = 5 mm) ............................... 62 xi

University of Engineering & Technology, Taxila-Pakistan

Fig. 3.29 Fig. 3.30 Fig. 3.31 Fig. 3.32 Fig. 3.33 Fig. 3.34 Fig. 3.35 Fig. 3.36 Fig. 3.37 Fig. 3.38 Fig. 3.39 Fig. 3.40 Fig. 3.41 Fig. 3.42 Fig. 3.43 Fig. 3.44 Fig. 3.45 Fig. 3.46 Fig. 3.47 Fig. 3.48 Fig. 3.49 Fig. 3.50 Fig. 3.51 Fig. 3.52 Fig. 3.53 Fig. 3.54 Fig. 3.55 Fig. 3.56 Fig. 3.57 Fig. 3.58 Fig. 3.59 Fig. 3.60 Fig. 3.61 Fig. 3.62 Fig. 3.63 Fig. 3.64 Fig. 3.65 Fig. 4.1 Fig. 4.2 Fig. 4.3 Fig. 4.4

Distortion of Sixteen Experiments ( t = 3 mm) .............................................. 62 Distortion of Sixteen Experiments ( t = 4 mm) .............................................. 63 Distortion of Sixteen Experiments ( t = 5 mm) ............................................. 63 Effects of welding parameters upon Distortion (t = 3 mm) ............................ 66 Effects of welding parameters upon Distortion (t = 4 mm) ............................ 67 Effects of welding parameters upon Distortion (t = 5 mm) ............................ 67 Distortion Predictions w.r.t Predictors (t = 3 mm) ......................................... 69 Distortion Predictions w.r.t Predictors (t = 4 mm) ......................................... 69 Distortion Predictions w.r.t Predictors (t = 5 mm) ......................................... 69 Residual Stresses of Sixteen Experiments ( t = 3 mm)................................... 70 Residual Stresses of Sixteen Experiments ( t = 4 mm)................................... 70 Residual Stresses of Sixteen Experiments ( t = 5 mm)................................... 70 Effects of welding parameters upon Residual Stresses (t = 3 mm) ................ 73 Effects of welding parameters upon Residual Stresses (t = 4 mm) ................ 73 Effects of welding parameters upon Residual Stresses (t = 5 mm) ................ 74 Residual Stresses Predictions w.r.t Predictors (t = 3 mm).............................. 76 Residual Stresses Predictions w.r.t Predictors (t = 4 mm).............................. 76 Residual Stresses Predictions w.r.t Predictors (t = 5 mm).............................. 76 Response Desirability w.r.t Predictors (t = 3 mm) ......................................... 80 Response Desirability w.r.t Predictors (t = 4 mm) ......................................... 80 Response Desirability w.r.t Predictors (t = 5 mm) ......................................... 80 Effect on Desirability of different Predictors (t = 3 mm) ............................... 81 Effect on Desirability of different Predictors (t = 4 mm) ............................... 81 Effect on Desirability of different Predictors (t = 5 mm) ............................... 81 Response Comparison Scator Plots (t = 3 mm, 4 mm, 5 mm)........................ 82 Effects of welding parameters upon Weld Strength in RSM ......................... 86 Effects of welding parameters upon Distortion in RSM ................................ 87 Effects of welding parameters upon Residual Stresses in RSM .................... 87 Interaction of welding parameters upon Weld Strength in RSM ................... 88 Interaction of welding parameters upon Distortion in RSM .......................... 89 Interaction of welding parameters upon Residual Stresses in RSM .............. 90 Weld Strength Predictions w.r.t Predictors in RSM ...................................... 92 Distortion Predictions w.r.t Predictors in RSM ............................................. 93 Residual Stresses Predictions w.r.t Predictors in RSM .................................. 93 Response Desirability w.r.t Predictors in RSM ............................................. 95 Effect on Desirability of different Predictors ................................................. 95 Response Comparison Scator Plots ............................................................... 96 Goldak's double ellipsoid heat source model for welding heat source ....... 106 Schematic representations of thermal boundary conditions ........................ 109 Thermo-physical properties of HSLA steel ................................................ 111 Thermo-mechanical properties of HSLA steel ........................................... 112

xii

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 4.5 Fig. 4.6 Fig. 4.7 Fig. 4.8 Fig. 4.9 Fig. 4.10 Fig. 4.11 Fig. 4.12 Fig. 4.13 Fig. 4.14 Fig. 4.15 Fig. 4.16 Fig. 4.17 Fig. 4.18 Fig. 4.19 Fig. 4.20 Fig. 4.21 Fig. 4.22 Fig. 4.23 Fig. 4.24 Fig. 4.25 Fig. 4.26 Fig. 4.27 Fig. 4.28 Fig. 4.29 Fig. 4.30

Overview of de-coupled thermo-mechanical simulation approach ............. 114 Detailed sequentially coupled thermo-mechanical simulation strategy ....... 115 (a) 3D FE mesh based on sensitivity analysis (b) "V" groove tack weld and root opening in FE model ...................................................................... 119 Mesh sensitivity analysis based on maximum temperature attained ........... 119 Schematic representations of structural boundary conditions along with geometric parameters ............................................................................ 120 Butt-weld joint geometry ............................................................................. 121 Comparison of computed and measured transient temperature profiles at four different locations on cylinders outer surface ..................................... 122 Computed and measured residual stress values for different locations at cylinder outer surface ............................................................................... 123 Temperature profiles at four different time steps during welding process ... 124 Axial temperature distributions for four different cross-sections at different time steps from the weld start position ......................................... 125 Transient thermal cycles experienced by various points at different cross sections from the weld start position ................................................... 126 Residual axial stresses (MPa) on outer surface at different cross sections from the weld start position .......................................................................... 126 Residual axial stresses (MPa) on inner surface at different cross sections from the weld start position .......................................................................... 127 Residual hoop stresses (MPa) on outer surface at different cross sections from the weld start position .......................................................................... 129 Residual hoop stresses (MPa) on inner surface at different cross sections from the weld start position .......................................................................... 129 Axial and hoop residual stress fields on cylinder outer and inner surfaces on a circumferential path at the WL .............................................................. 130 Measured and predicted axial deformation (face tilt) of the cylinder face ... 131 Axial shrinkage at four different cross sections from the WL on cylinder outer surface.................................................................................................. 132 Schematic representation of transient forces on solidifying weld pool ........ 133 Radial shrinkage at different cross sections from the WL on cylinder outer surface.................................................................................................. 133 Automatic rotary positioner with clamping and strain gages arrangements. 134 Factors affecting the heat distribution during welding ................................. 135 Overall experimental validation approach for circumferential welding ...... 135 Macrograph sample after water jet cutting from cylinder ............................ 136 Low magnification metallographic sample for measurement of FZ and HAZ dimensions .................................................................................... 136 K-type thermocouples used in present research for transient temperature measurement ............................................................................. 137

xiii

University of Engineering & Technology, Taxila-Pakistan

Fig. 4.31 Fig. 4.32 Fig. 4.33 Fig. 4.34 Fig. 5.1 Fig. 5.2 Fig. 5.3 Fig. 5.4 Fig. 5.5 Fig. 5.6 Fig. 5.7 Fig. 5.8 Fig. 5.9 Fig. 5.10 Fig. 5.11 Fig. 5.12 Fig. 5.13

Multi-channels data logging system with thermocouples connected ........... 137 Digital infrared pyrometer (Cyclopes from Minolta/LAND) ....................... 137 Experimental setup used for distortion measurement................................... 138 Experimental setup used for experimental measurement of residual stresses............................................................................................. 139 (a) Experimental macrograph at a section 150o from weld start position (b) Comparison of experimental and simulated FZ and HAZ dimensions... 143 Residual axial stresses on outer and inner surface of the cylinders at a section 150o from the weld start position with different welding speeds .. 144 Residual hoop stresses on outer and inner surface of the cylinders along the circumference at the WL with different welding speeds ............... 145 Residual hoop stresses on outer and inner surface of the cylinders at a section 150o from weld start position for different welding speeds........... 145 Residual hoop stresses on outer and inner surface of the cylinders at a section 150o from the weld start position for different welding current ....... 146 Residual axial stresses on outer and inner surface of cylinders at a section of 150o from the weld start position for different welding current... 147 Axial residual stresses on cylinder outer and inner surfaces at a section of 150o from weld start position for various thicknesses .............................. 148 Hoop residual stresses on cylinder outer and inner surfaces at a section of 150o from weld start position for various thicknesses .............................. 149 Axial and hoop residual stresses on cylinder inner surface at a circumferential path on WL for various wall thickness values ..................... 150 Axial and hoop residual stresses on cylinder outer surface at a circumferential path on WL for various wall thickness values ..................... 151 Axial residual stress variations, circumferentially on the weld line for cylinder inner surface with different root openings ...................................... 152 Hoop residual stress variation, circumferentially on the weld line for cylinder inner surface with different root openings ...................................... 153 Axial residual stress variations, circumferentially on the weld line for cylinder outer surface with different root openings ...................................... 153

Fig. 5.14 Hoop residual stress variation, circumferentially on the weld line for cylinder outer surface with different root openings ...................................... 154 Fig. 5.15 Axial displacement of the restraint free face of cylinder for different root openings .................................................................................. 154 Fig. 5.16 Axial residual stress variations at a longitudinal section 180o from weld line on cylinder inner surface........................................................................ 157 Fig. 5.17 Axial residual stress variations at a longitudinal section 180o from weld line on cylinder outer surface........................................................................ 157

xiv

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 5.18 Axial residual stress variations at a circumferential path on the weld line for cylinder outer surface ....................................................................... 158 Fig. 5.19 Axial residual stress variations at a circumferential path on the weld line for cylinder inner surface ....................................................................... 158 Fig. 5.20 Distortion of Sixteen Experiments ( t = 3 mm) ............................................ 163 Fig. 5.21 Distortion of Sixteen Experiments ( t = 4 mm) ............................................ 164 Fig. 5.22 Distortion of Sixteen Experiments ( t = 5 mm) ........................................... 164 Fig. 5.23 Effects of welding parameters upon Distortion (t = 3 mm) .......................... 167 Fig. 5.24 Effects of welding parameters upon Distortion (t = 4 mm) .......................... 167 Fig. 5.25 Effects of welding parameters upon Distortion (t = 5 mm) .......................... 168 Fig. 5.26 Distortion Predictions w.r.t Predictors (t = 3 mm) ....................................... 169 Fig. 5.27 Distortion Predictions w.r.t Predictors (t = 4 mm) ....................................... 170 Fig. 5.28 Distortion Predictions w.r.t Predictors (t = 5 mm) ....................................... 170 Fig. 5.29 Residual Stresses of Sixteen Experiments ( t = 3 mm)................................. 171 Fig. 5.30 Residual Stresses of Sixteen Experiments ( t = 4 mm)................................. 171 Fig. 5.31 Residual Stresses of Sixteen Experiments ( t = 5 mm)................................. 171 Fig. 5.32 Effects of welding parameters upon Residual Stresses (t = 3 mm) .............. 174 Fig. 5.33 Effects of welding parameters upon Residual Stresses (t = 4 mm) .............. 174 Fig. 5.34 Effects of welding parameters upon Residual Stresses (t = 5 mm) .............. 175 Fig. 5.35 Residual Stresses Predictions w.r.t Predictors (t = 3 mm)............................ 177 Fig. 5.36 Residual Stresses Predictions w.r.t Predictors (t = 4 mm)............................ 177 Fig. 5.37 Residual Stresses Predictions w.r.t Predictors (t = 5 mm)............................ 177 Fig. 5.38 Response Desirability w.r.t Predictors (t = 3 mm) ....................................... 181 Fig. 5.39 Response Desirability w.r.t Predictors (t = 4 mm) ....................................... 181 Fig. 5.40 Response Desirability w.r.t Predictors (t = 5 mm) ....................................... 181 Fig. 5.41 Effect on Desirability of different Predictors (t = 3 mm) ............................. 182 Fig. 5.42 Effect on Desirability of different Predictors (t = 4 mm) ............................. 182 Fig. 5.43 Effect on Desirability of different Predictors (t = 5 mm) ............................. 182 Fig. 5.44 Response Comparison Scator Plots (t = 3 mm, 4 mm, 5 mm)...................... 183 Fig. 5.45 Effects of welding parameters upon Distortion in RSM .............................. 187 Fig. 5.46 Effects of welding parameters upon Residual Stresses in RSM .................. 187 Fig. 5.47 Interaction of welding parameters upon Distortion in RSM ........................ 188 Fig. 5.48 Interaction of welding parameters upon Residual Stresses in RSM ............ 188 Fig. 5.49 Distortion Predictions w.r.t Predictors in RSM ........................................... 190 Fig. 5.50 Residual Stresses Predictions w.r.t Predictors in RSM ................................ 190 Fig. 5.51 Effects of welding parameters upon Weld Strength in RSM ....................... 197 Fig. 5.52 Effects of welding parameters upon Distortion in RSM .............................. 197 Fig. 5.53 Effects of welding parameters upon Residual Stresses in RSM .................. 198 Fig. 5.54 Interaction of welding parameters upon Weld Strength in RSM ................. 198 Fig. 5.55 Interaction of welding parameters upon Distortion in RSM ........................ 199 Fig. 5.56 Interaction of welding parameters upon Residual Stresses in RSM ............ 199

xv

University of Engineering & Technology, Taxila-Pakistan

Fig. 5.57 Fig. 5.58 Fig. 5.59 Fig. 5.60 Fig. 5.61 Fig. 5.62 Fig. 6.1 Fig. 6.2 Fig. 6.3 Fig. 6.4 Fig. 6.5 Fig. 6.6 Fig. 6.7 Fig. 7.1 Fig. 7.2 Fig. 7.3 Fig. 7.4 Fig. 7.5 Fig. 7.6 Fig. 7.7 Fig. 7.8 Fig. 7.9 Fig. 7.10

Weld Strength Predictions w.r.t Predictors in RSM .................................... 202 Distortion Predictions w.r.t Predictors in RSM ........................................... 202 Residual Stresses Predictions w.r.t Predictors in RSM ................................ 203 Weld Strength Desirability w.r.t Predictors in RSM ................................... 203 Distortion Desirability w.r.t Predictors in RSM .......................................... 204 Residual Stresses Desirability w.r.t Predictors in RSM ............................... 204 Main components of an expert system ......................................................... 208 Configuration of the expert system .............................................................. 209 The flow chart representing the operational procedure of expert system .... 211 Fuzzy sets for the numeric input variables .................................................. 212 Fuzzy sets for responses ............................................................................... 213 Fuzzification of input data ........................................................................... 215 Decline of estimation error along number of iterations ............................... 219 Flow chart for data acquisition module ....................................................... 225 Fuzzy sets for maximization and/or minimization of output variable ......... 226 Customized flow chart for auto-development of fuzzy sets ........................ 227 The framework for self-development of prediction rule-base ..................... 228 The framework for self-development of optimization rule-base ................ 230 Mechanism of the linked list Set, consisting of 6 nodes .............................. 231 A portion of 2-D linked list Rule_consequent ............................................. 233 Process of interface of expert system from fuzzy clips ............................... 239 Interface of expert system representing fledgling knowledge-base ............. 240 Interface of expert system representing veteran knowledge-base ............... 242

xvi

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

List of Tables Table 1.1 Table 3.1 Table 3.2 Table 3.3 Table 3.4 Table 3.5 Table 3.6 Table 3.7 Table 3.8 Table 3.9 Table 3.10 Table 3.11 Table 3.12 Table 3.13 Table 3.14 Table 3.15 Table 3.16 Table 3.17 Table 3.18 Table 3.19 Table 3.20 Table 3.21 Table 3.22 Table 3.23 Table 3.24 Table 3.25 Table 3.26 Table 3.27 Table 3.28 Table 3.29 Table 3.30 Table 3.31 Table 3.32 Table 3.33 Table 4.1 Table 4.2 Table 5.1 Table 5.2

A summary of development of welding technology ..................................... 3 Initial settings of predictor variables............................................................ 43 High and Low Settings of Factors (t = 3 mm) ............................................ 45 High and Low Settings of Factors (t = 4 mm) ............................................. 45 High and Low Settings of Factors (t = 5 mm) ............................................. 45 Design of 16 Experiments following Full Factorial (t = 3 mm) .................. 45 Design of 16 Experiments following Full Factorial (t = 4 mm) ................. 46 Design of 16 Experiments following Full Factorial (t = 5 mm) .................. 46 Chemical composition of Base Metal (HSLA Steel)(30CrMnSiA) ............ 54 Chemical composition of Filler Wire (H08) ................................................ 54 Mechanical properties of Base Metal (HSLA Steel) .................................. 54 Max. and Min. Values of Weld Strength Response .................................... 55 ANOVA for Tensile Strength (t = 3 mm) factorial model........................... 57 ANOVA for Tensile Strength (t = 4 mm) factorial model........................... 57 ANOVA for Tensile Strength (t = 5 mm) factorial model .......................... 58 Max. and Min. Values of Distortion Response ........................................... 63 ANOVA for Distortion (t=3mm) factorial model........................................ 64 ANOVA for Distortion (t=4mm) factorial model........................................ 64 ANOVA for Distortion (t=5mm) factorial model........................................ 65 Max. and Min. Values of Residual Stresses Response ................................ 71 ANOVA for Residual Stresses (t=3mm) factorial model ............................ 71 ANOVA for Residual Stresses (t=4mm) factorial model ............................ 72 ANOVA for Residual Stresses (t=5mm) factorial model ............................ 72 Response Desirability Solutions (t = 3 mm)............................................... 77 Response Desirability Solutions (t = 4 mm)............................................... 78 Response Desirability Solutions (t = 5 mm)............................................... 79 High and Low Settings of Factors for RSM ............................................... 83 Historical Data (48 (3x16) observations) including Response Values for RSM ........................................................................... 84 Max. and Min. Values of Responses in RSM ............................................. 85 ANOVA for Weld Strength (2FI Model) of RSM ...................................... 85 ANOVA for Distortion (2FI Model) of RSM.............................................. 85 ANOVA for Residual Stress (2FI Model) of RSM .................................... 86 ANOVA Summary for RSM (2FI model) .................................................. 86 Response Desirability Solutions ................................................................. 94 Welding process parameters ..................................................................... 121 Goldak heat source parameters ................................................................ 121 Heat source parameters ............................................................................ 142 Welding Process Parameters .................................................................... 142 xvii

University of Engineering & Technology, Taxila-Pakistan

Table 5.3 Table 5.4 Table 5.5 Table 5.6 Table 5.7 Table 5.8 Table 5.9 Table 5.10 Table 5.11 Table 5.12 Table 5.13 Table 5.14 Table 5.15 Table 5.16 Table 5.17 Table 5.18 Table 5.19 Table 5.20 Table 5.21 Table 5.22 Table 5.23 Table 5.24 Table 5.25 Table 5.26 Table 5.27 Table 5.28 Table 5.29 Table 5.30 Table 5.31 Table 5.32 Table 5.33 Table 5.34 Table 5.35 Table 5.36 Table 5.37 Table 6.1 Table 6.2 Table 6.3

Cylinder thickness and welding process parameters for parametric studies .............................................................................. 147 Studies for the analysis of effects of root openings ................................. 151 Studies for tack weld orientation analysis ............................................... 155 High and Low Settings of Factors (t = 3 mm) .......................................... 160 High and Low Settings of Factors (t = 4 mm) ........................................... 160 High and Low Settings of Factors (t = 5 mm) ........................................... 161 Design of 16 Experiments following Full Factorial (t = 3 mm) ................ 161 Design of 16 Experiments following Full Factorial (t = 4 mm) ............... 162 Design of 16 Experiments following Full Factorial (t = 5 mm) ................ 162 Max. and Min. Values of Distortion Response ......................................... 164 ANOVA for Distortion (t=3mm) factorial model...................................... 165 ANOVA for Distortion (t=4mm) factorial model...................................... 165 ANOVA for Distortion (t=5mm) factorial model...................................... 166 Max. and Min. Values of Residual Stresses Response .............................. 172 ANOVA for Residual Stresses (t=3mm) factorial model .......................... 172 ANOVA for Residual Stresses (t=4mm) factorial model .......................... 173 ANOVA for Residual Stresses (t=5mm) factorial model .......................... 173 Response Desirability Solutions (t = 3 mm)............................................. 178 Response Desirability Solutions (t = 4 mm)............................................. 179 Response Desirability Solutions (t = 5 mm)............................................. 180 High and Low Settings of Factors for RSM ............................................. 184 Historical Data (48 (3x16) observations) including Response Values for RSM ......................................................................... 185 Max. and Min. Values of Responses in RSM ........................................... 186 ANOVA for Distortion (2FI Model) of RSM............................................ 186 ANOVA for Residual Stress (2FI Model) of RSM .................................. 186 ANOVA Summary for RSM (2FI model) ................................................ 186 Response Desirability Solutions ............................................................... 191 High and Low Settings of Factors for RSM ............................................. 192 Historical Data (48 (3x16) observations) including Response Values for RSM ......................................................................... 193 ANOVA for Weld Strength (2FI Model) of RSM..................................... 195 ANOVA for Distortion (2FI Model) of RSM............................................ 195 ANOVA for Residual Stress (2FI Model) of RSM .................................. 196 Max. and Min. Values of Responses in RSM ........................................... 196 ANOVA Summary for RSM (2FI model) ................................................ 196 Response Desirability Solutions ............................................................... 205 Weld Strength values from all the four rules ............................................ 216 Maximum fuzzy output from Table 6.1 .................................................... 216 List of rules operated by the optimization module ................................... 220

xviii

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 6.4 Table 6.5 Table 7.1 Table 7.2 Table 7.3

List of rules operated by the prediction module ....................................... 220 List of consequents (Residual Stresses) .................................................... 221 Data for the fledgling knowledge-base ..................................................... 234 Welding Parameters for EXWeldHSLASteel Predictions ........................ 243 Comparison of Responses against welding parameters in Table 7.2 ........ 243

xix

University of Engineering & Technology, Taxila-Pakistan

List of Symbols Symbol

Description

A A A af ar AI AISI ANN ANOVA ASTM α b BC BCJ BWR c c C C CF CI CLIPS CFD CONV CSM CVM D E ES EXP ε εel εth εem η Ø ff fr fw F Fa FE FEM FLW

Amp Cross Sectional Area of Column Surface area Length of front ellipsoidal of heat source (mm) Length of rear ellipsoidal of heat source (mm) Artificial Intelligence American Iron & Steel Institute Artificial Neural Network Analysis of Variance American Society for Testing and Materials Co-efficient of thermal expansion (per degree) Half width of heat source (mm) Before Christ Bammann Chiesa-Johnson Boiling Water Reactor Penetration depth of heat source (mm) Specific heat Welding Current (Amp) Concentration Coefficient (m-2) Certainty Factor Computational Intelligence C Language Integrated Production Systems Computational Fluid Dynamics Convection Computational Solid Mechanics Computational Weld Mechanics Material Stiffness Young's Modulus of Column Expert System Experimental Total strain Elastic strain Thermal strain Radiation emissivity of cylinder surface Arc efficiency (%) Outer Diameter of Cylinder Fraction of heat in front ellipsoidal of heat source Fraction of heat in rear ellipsoidal of heat source Wire Feed Rate (cm/min) Circumferential Force Acceleration force vector Finite Element Finite Element Methods or Finite Element Modeling Fine Line Welding

xx

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

FV FZ GMAW GTAW H HAZ HSW HSLA htotal hconvection I imax ISO IGSCC K KBS L L LPHSW M MIG MMAW M(r, z) N NPL Q q(r) q(0) qloss qconvection qradiation Ro RO RSM ri ro r r ρ S SA SEM σbol σ1, σ2, σ3 σ σ1 σ2 σp σy

Finite Volume Fusion Zone Gas Metal Arc Welding Gas Tungsten Arc Welding Enthalpy of material Heat Affected Zone Heat Sink Welding High Strength Low Alloy Combined convection and radiation heat transfer coefficient (Wm-2K) Convective heat transfer coefficient (Wm-2K) Current (amperes) Maximum Number of Iterations International Organization for Standardization Inter-granular Stress Corrosion Cracking Stiffness of Column Knowledge-Based Systems Plate Length Length of Column Last Pass Heat Sink Welding Bending Moment Metal Inert Gas Manual Metal Arc Welding Scalar multiplier as a function of axial and radial position Rotational Speed of Positioner (rpm) National Physics Laboratory Shear Force Surface flux at radius r (Wm-2) Maximum flux at the center of the heat source (Wm-2) Total heat loss Heat loss by convection Heat loss by radiation Outer radius of cylinder (mm) Root Opening Residual Stress Measurement (MPa) Cylinder inner radius (mm) Cylinder outer radius (mm) Cylinder mean radius Radial distance from the center of the heat source (m) Density of material (Kgm-3) Weld Speed (cm/min) Simulated Annealing Scanning Electron Microscopy Stefan-Boltzman constant (5.6703 x 10-8 Wm-2K-4) Three principal stresses (MPa) Stress (MPa) Transverse residual stress Longitudinal residual stress Tensile stress (MPa) Yield stress (MPa)

xxi

University of Engineering & Technology, Taxila-Pakistan

σo σh t t T T0 TS TL ∆T θ T Tamb t TIG TSOFT V V w W WS WL WPP

Weld induced residual stress (MPa) Hoop stress Thickness (mm) Cylinder Wall Thickness Temperature (C˚) Starting Annealing Temperature Solidus temperature Liquidus temperature Difference between the reference & actual temperature Angle from instantaneous arc heat source position Current temperature at cylinder surface Ambient temperature Plate Thickness Tungsten Inert Gas Softening Temperature Welding Voltage (v) Voltage (volts) Displacement vector of a general point Plate Width Weld speed (mms-1) Weld Line Welding Process Parameter

xxii

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

CHAPTER 1 INTRODUCTION The first chapter covers the introduction in brief about welding technology including arc welding and its physics, summary of development of welding technology, welding benefits and threats, and weld induced imperfections (distortions and residual stresses) in thin walled structures. Further, this chapter covers the details of GTAW or TIG welding including the TIG welding process variables (input parameters) and performance measures (output or response), current challenges in welding domain related to thin walled structures and application of artificial intelligence (AI) to manufacturing including the detail of expert system (ES) to welding. At the end of this chapter, the detail of research objectives, research methodology adopted to tackle the objectives defined and organization of this dissertation are presented for the self-development of “expert system for optimization of welding process of thin-walled high strength low alloy (HSLA) steel structures”. 1.1 Welding Technology Welding technology is a major part of any mechanical manufacturing facility in the world. It is considered as the most wide-spread metal joining process in the industries. Generally, welding can be defined as any process in which two or more pieces of metal are joined together by the application of heat, pressure, or a combination of both. Most of the welding processes may be grouped into two main categories [1]: 1. Pressure Welding. The welding in which the weld is achieved by applying the pressure. 2. Heat Welding. The welding in which the weld is achieved by the function of heat. Today, the heat welding is the most common welding type used in the industries. 1.1.1 Arc Welding Arc welding, which is heat-type welding, is one of the most important manufacturing operations for the joining of structural elements for a wide range of applications, including boilers, guide way for trains, ships, bridges, building structures, automobiles, pressure vessels, missiles and nuclear reactors, to name a few. The American Welding Society (AWS) defines arc welding as [1]: “A process that uses an electric arc as a source of heat to melt and join metals” It requires a continuous supply of either direct current (DC) or alternating electric current (AC), which creates an electric arc to generate enough heat to melt the metal and form a weld. The direct additions and indirect evolution of heat is one of the key elements of arc welding process, but it is also the source of tension for most of the welding engineers [2].

1

University of Engineering & Technology, Taxila-Pakistan

1.1.2 Physics of Arc Welding The arc welding physics deals with complex physical phenomena associated with welding, including heat, electricity, magnetism, light, and sound. Upon application of intense heat, metal at the joint between two parts is melted and caused to intermix directly, or more commonly, with an intermediate molten filler metal. It is generally described by an electric field between the positive anode and the negative cathode surrounded by an ionization gas. In arc welding, an electric arc is used to produce the intense heat needed to melt the metal. On metal, there is a thin layer of surface electrons, which are accelerated in the field towards the anode. These electrons collide with the atoms in the gas, causing impact ionization where the atoms are decomposed into the electrons and positive ions, which cause ionization further. The current of electrically charged particles in the arc and the temperature are interrelated as high temperatures increase ionization, causes the temperature rise due to the released energy. The temperature or the current must initially be brought up to a certain level to obtain welding conditions, which is done by igniting the arc. The basic principal of the arc welding process is illustrated in Figure 1.1 [3, 4].

Welding machine AC / DC

Electrode holder

Electrode Arc Work Work cable

Fig. 1.1 Basic physics of arc welding [3, 4] Arc ignition is accomplished by the short circuiting of current, which occurs as the anode and the cathode are brought into brief contact. The short-circuit current shortly increases the temperature and the current and then subsequently the arc can be maintained in the electric field existing under normal welding conditions. The arc is surrounded by a magnetic field that is directed the charged particles towards the center of the arc, causing the arc to localize in spots on the anode and the cathode. When these electrically charged particles impact on the anode and the cathode, the anode and the cathode spots are heated to high temperatures. The high temperature of approximately 3000oC to 5000oC causes to melt of the both, the electrode and the welded metal. Due to suction force of the plasma flow, droplets of the electrode material are deposited on the welded metal. The arc welding process is a remarkably complex operation involving extremely high temperatures, which induces imperfections in the weldments like high levels of residual stresses and severe distortions. These extreme phenomena tend to reduce the strength of a structure, which becomes vulnerable to buckling, corrosion, fracture and other type of failures [5].

2

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

1.1.3 Development of Welding Technology The earliest welding technology has been traced as back as 1000 BC when the forge welding had been utilized into weapons. First time the use of electric fusion process has been reported in 1782 in Germany by G. Lichtenberg [6]. However, the most of the references show commencement of electric arc welding process in late nineteenth century. A brief summary of the development in welding processes is given in Table 1.1 [4, 7]. 1801

Discovery of electric arc by Sir Humphrey Devy

1860-1865

Wilde first intentionally joined metal by electric welding in early 1860 and was granted a patent in 1865 for his work

1885

De Meritens obtained patent of electric arc welding process in England using carbon electrode

1886

E. Thomson obtained a patent on resistance

1887

Benardos, a Russian scientist, got first patent of electric arc welding for slightly different equipment then by De Meritens

1891

Another Russian N. Slavianoff replaced carbon electrode with a metal electrode and obtained a patent on metal arc welding

1908-1940

Kjellberg, a Swedish, got a patent for coated welding electrode Development in joining process continued and major welding process including oxyacetylene, MMAW (manual metal arc welding), GTAW (gas tungsten arc welding), and GMAW (gas metal arc welding) processes were successfully implemented

1960

Advanced welding types such as Electron Beam, Laser and Ultrasonic welding were developed during 1950-1960

2000

Most recent development is magnetic pulse welding introduced Table 1.1 A summary of development of welding technology

1.1.4 Benefits & Threats of Welding Welding represents one of the most complex manufacturing processes in terms of number of variables involved and factors contributing to the final output or response. Welding has been used in the fabrication of structures ranging from conventional industrial applications to high-tech engineering applications in aeronautical, nuclear, aerospace, marine and high-pressure vessel applications. Compared to mechanical joining methods, welding method offers some significant advantages including flexibility of design, improved structural integrity and weight & cost savings [8, 9]. However, the welding method induces 3

University of Engineering & Technology, Taxila-Pakistan

the thermal strains in the weld metal and base metal regions near the weld, resulting in stresses, which in turn combine and react to produce internal forces that cause bending, buckling, and rotation. These displacements are known as welding distortions [10]. As a widely used mechanical manufacturing technique, welding offers a number of technical challenges to the welding community specially shop floor engineers engaged in manufacturing of different welded structures. While joining the components of a structure together by welding, the complex thermal cycles from welding result in formation of residual stresses in the joint region, and deformation of the welded structure. Both weld residual stresses and distortions can significantly impair the performance and reliability of the welded structures [11]. They must be properly dealt with during the product and process design and manufacturing phases, to ensure intended in-service use of the welded structures. Despite the recognition of welding as one of the most important fabrication processes in the engineering industries, there is a little scientific understanding present in the productivity measurement and evaluation of the welding processes. 1.1.5 Weld Induced Residual Stresses Residual stresses are those stresses that would exist in a body if all external loads and restraints were removed. Various technical terms have been used to refer to residual stress, such as internal stress, initial stress, inherent stress, reaction stress and locked-in stress [12]. Mechanical structures suffer from residual stresses during different phases of their life cycle. In engineering structures most of the residual stresses are induced during their manufacturing phase including casting and forging, sheet metal forming and shaping (shearing, bending, grinding, machining etc.) and welding. Plate thickness, t

Lengt h

σ2

σ1

Width

σ1

σ1 = Transverse Residual Stress σ2 = Longitudinal Residual Stress

σ2

Fig. 1.2 Schematic view of residual stresses in welded rectangular plate [4] 4

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Welding residual stresses are produced in a structure as a consequence of local plastic deformations introduced by local temperature history consisting of a rapid heating and subsequent cooling phase. During the welding process, the weld area is heated up sharply compare to the surrounding area and fused locally. The material expands as a result of being heated [13]. The heat expansion is restrained by the surrounding cooler area, which gives rise to thermal stresses. The thermal stresses partly exceed the yield limit, which is lowered at elevated temperatures. Consequently, the weld area is plastically hot-compressed. After cooling down too short, too narrow or too small comparing to the surrounding area, it develops tensile residual stress, while the surrounding areas are subjected to compressive residual stresses to maintain the self-equilibrium [14]. Figure 1.2 [4], shows the calculated longitudinal and transverse residual stresses in center cross sections of rectangular plate at centre of weld. Due to the heating and cooling cycles and constraints from surrounding materials, high longitudinal stress is developed at central section of the plate. As the distance from the weld center increase, the longitudinal stress gradually decreases. Along the transverse direction, the longitudinal stress changes to compressive, whereas along the longitudinal direction it reduces to zero, as dictated by the equilibrium condition of residual stresses. Similar transverse residual stress with minor differences in distribution from the longitudinal stress and smaller magnitude is observed. 1.1.6 Weld Induced Distortions Welding induced distortion can be defined as: change in shape and dimension of a welded, after welding; when the structure is free from any of the external forces of thermal gradients. The interaction of solidifying weld metal with the parent base metal, results in change in dimensions and shape of the weldments, generally referred to as welding distortions. Different types of distortion patterns for plate welding as presented in Figure 1.2 are discussed in detail by [12]. Further, in recent years many researchers presented mechanism involved and the factors affecting different types of welding distortions [15-20]. The temperature distribution is not uniform due to the locally heating of material during welding process. The stresses are released in the melted weld pool and can be assumed to zero. The metal starts to shrink during the solidification of the melted weld pool and to exert stresses on the surrounding weld metal and HAZ. These stresses remain in the material after welding and result in unwanted distortion. A typical example of distortion is given in Figure 1.3. The three different types of residual stress induced distortion can be found in manufactured structures as shown in Figure 1.3. The longitudinal and transverse shrinkage can cause in plane distortion of the work piece whereas plane or axisymmetrical angular shrinkage can cause distortion perpendicular to the plane of the welded component and another distortion is bending due to grids with longitudinal and transverse welds [21]. The residual stresses and the structure deformations are highly affected by the using of welding fixtures during welding process and the amount of restraint determines the control of distortions and residual stress fields on the weldments [22]. Generally, welding residual stresses and strains behave in opposing ways with degree of restraint as shown in Figure 1.4 [4, 23]. Therefore, the type of weld fixtures is used that keep residual stresses low, and those, which reduce residual distortions. 5

University of Engineering & Technology, Taxila-Pakistan

Fig. 1.3 Example of distortion that can occur during welding [21]

Level of stress/distortion

• Risk of process failure • Large deformation

• Risk of in-service failure • Large residual stresses

Deformation

Residual stresses

Minimum clamping

Maximum clamping

Degree of clamping

Fig. 1.4 Effect of the degree of clamping on the level of distortion & residual stresses [4, 23]

6

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

1.1.7 Weld Induced Residual Stresses and Distortions in Thin Walled Structure Thin-walled structures comprises an important and growing proportion of engineering manufacturing with areas of application becoming increasingly diverse, ranging from aircraft, missiles, ships, pressure vessels, bridges and oil rigs to storage vessels, industrial buildings and warehouses due to many factors, including cost and weight economy, new materials and processes etc. Thin-walled structures are designed with advanced numerical analysis techniques and manufactured using sophisticated fabrication processes. There are, however, a number of factors that may effects a structure that is not exactly expected and considered during the design calculations. These effects may be due to the changes in the properties of the structure, in the geometry, and many others and sometimes, even small changes in the structure may produce significant changes in the response. The effects of imperfections in thin-walled structures may introduce changes in the stresses that are nearly equal to the stresses due to the loads. Generally the cylinders are considered as thin-walled if the outer diameter of the cylinder is greater than the twenty times of the cylinder wall thickness i.e. Ø > 20 x t, where Ø is the outer diameter of the cylinder and t is the cylinder wall thickness.

(b) (M is bending moment, Q is shear force, and F is the circumferential force)

Fig. 1.5 (a) Schematic view of expansion and shrinkage in circumferentially welded pipe/cylinders (b) Free body diagram of the weld joint [4, 14]

7

University of Engineering & Technology, Taxila-Pakistan

Incase of circumferentially welded cylinders or pipes, residual stress distribution and distortion pattern is more complex as shown in Figure 1.5 [4, 14]. Two types of dominating distortions axial shrinkage and radial deflection are of critical importance in this case. Figure 1.5(a) presents the possible expansion and shrinkage in a circumferentially welded pipe. The shrinkage of the weld in the circumferential direction induces circumferential force, F, shearing force, Q, and bending moments, M, to the cylinder as show in Figure 1.5(b), which are the resultants of the residual stresses in both circumferential and axial directions [14]. Thus, the state of stress in a circumferential welded pipe may be quite different for that in the flat plate [7, 24]. Distribution of residual stresses in a pipe is affected by many factors such as diameter, wall thickness of the pipe, weld geometry, welding procedure and sequence [6, 25-27]. The importance of effect of axial and radial shrinkage can be accessed depending upon the application and use of the welded structures. Incase of highly pressurized chambers like thrust generating casings in satellite launching vehicles and pressure vessel applications, radial and diametric shrinkage is critically non desirable. In these hi-tech applications, the buckling strengths are very sensitive to the form and amplitude of very minor deviations of geometry from the ideal shape and characterization of these geometric imperfections is important in the scientific design of structures [28]. 1.2 GTAW or TIG Welding Gas Tungsten Arc Welding (GTAW) or Tungsten Inert Gas (TIG) welding is one of the most well established processes of arc welding type. The TIG welding process was invented during the Second World War due to the need of the American aircraft industry for a method of joining magnesium and aluminum. The first TIG process was used for the welding of magnesium using a Tungsten electrode and helium gas in the late 1930´s. GTAW became an overnight success in the 1940s for joining the metals and has played a major role in the acceptance of high quality structural welding and application in industries. GTAW was originally developed for aluminum and stainless steel, which are difficult to weld. The GTAW process is now widely used with other alloys. GTAW has been the most widely accepted welding processes so far in the industry due to its availability and versatility of welding equipment, low cost equipment, excellent quality and skilled welders. The aircraft industry is one of the main users of GTAW. GTAW can be called as a workhorse of industrial production welding. The GTAW process attains a good position in respect of the total cost specifically for thin sections because of the medium equipment cost and mainly due to low wire cost i.e. low deposition rates due to lower wire feed speeds [29]. The principle of the TIG welding process is schematically presented in Figure 1.6 [30]. GTAW is a thermal process depending on conducted heat through the weld joint materials to achieve the penetration. The melting temperature necessary to weld materials in the GTAW process is obtained by maintaining an arc between a tungsten alloy electrode and the work piece and the weld pool temperatures can approach 2500oC (4530oF) [31]. In GTAW process, a non-consumable tungsten electrode of diameter between 0.5 to 6.5 mm is used with an envelope of inert gas shielding (commonly used are argon, helium or their mixture) around it. Since the process uses a non-consumable electrode, extra filler material is usually added. The shielding gas protects both the tungsten electrode and the weld pool from 8

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

the detrimental effects of surrounding atmosphere gases. Argon is commonly used in welding unalloyed, low alloyed and stainless steels. Both power supply sources (AC or DC) are used for GTAW process. Generally, GTAW process uses a direct current (DC) arc, where the tungsten electrode has a negative polarity thus the tungsten electrode becomes the cathode and the work piece becomes the anode and the polarity is called straight polarity or direct current electrode negative (DCEN) [31]. GTAW process has all welding position capabilities while others are limited to one or a few welding positions. GTAW or TIG welding process is well accepted for pressure vessels, aero, rocket, and missile, nuclear and marine industries.

Fig. 1.6 Principle of Tungsten Inert Gas (TIG) welding [30] Many parameters affect GTAW quality, such as base metal, filler wire, weld geometry, electrode type, shielding gas type, welding current, and travel speed of the welding torch. The desired welding parameters are usually determined based on experience or handbook values. However, this does not ensure that the selected welding parameters result in optimal or near optimal welding quality characteristics for the particular welding system and environmental conditions. 1.3 Variables and Performance Measures in GTAW Process Following are some of the basic parameters of welding process besides pre-heating, interpass temperature, post-heating and no. of weld passes etc: 1. Material. Base metal properties like material composition and material properties (like thermal conductivity, coefficient of thermal expansion, reaction with atmospheric oxygen, effect of flux residue, and crack sensitivity) are considered as the most influential parameter.

9

University of Engineering & Technology, Taxila-Pakistan

2. Weld geometry. It is used for the selection of welding process. The joint type may be butt, lap, fillet or T-joint. Bevel may be single-V, double-V or U shape. Weld geometry is directly influential upon weld quality. 3. Welding Position. It can either be flat, horizontal, vertical, or overhead etc. Mainly vertical and horizontal welding position is used. Difficult welding position increases the problems in achieving the required weld quality. Weld bead geometry is affected by the position in which the work piece is held with respect to welding gun. 4. Shielding Gas (lit/min). It is a protective gas used to prevent atmospheric contamination. GTAW process is mostly conducted in shielding. It has been very promising in enhancing weld quality. Shielding Gas Flow Rate has significant effect on weld bead shape which in turn effects the distortion, residual stresses, heat effected zone (HAZ) and mechanical properties of the material to be welded. 5. Welding Speed (cm/min). It is the parameter that varies the weld penetration and width of beads. Maximum weld penetration is at a specific welding speed and decreases as speed varies. The increased input heat per unit length due to reduced speed results increase in weld width and vise versa. Variations in travel speed at a set current and voltage also affect bead shape. As welding speed is decreased, heat input per length of joint increases, and the penetration and bead width increase. Excessively high travel speeds results a crowned bead as well as the tendency for undercut and porosity. 6. Wire Feed Rate (cm/min). It is the parameter that controls the speed of welding filler wire. It is normally attributed to increased resistance heating which itself is increased with the increase in wire feed rate. The welding current varies with the change in wire feeding and the relationship is linear at lower feeding rate. 7. Material Thickness (mm). Material thickness plays a vital role in process selection and parameters setting. Material thickness is used to decide the input heat required and to control the cooling rate. Higher thickness means higher cooling rate resulting increase in heat effected zone (HAZ) and hardness of weld metal. 8. Welding Current (Amp). It is one of the most important parameter that directly affects the penetration and lack of fusion by affecting the speed of welding. Welding current is the current being used in the welding circuit during the making of a weld. If the current is too high at a given welding speed, the depth of fusion or penetration will be too great. For thinner plates, it tends to melt through the metal being joined. It also leads to excessive melting of filler wire resulting in excessive reinforcement. This means additional heat input to the plates being welded leading to increased weld induced distortions and if the welding current is too low, it may result in lack of fusion or inadequate penetration. 9. Welding Voltage (V). It is the parameter that directly affects the bead width. It also influences the microstructure and even the success and failure of the operation. Like current, welding voltage affects the bead shape and the weld deposit composition. 10

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Increase in the arc voltage results a longer arc length and a correspondingly wider, flatter bead with less penetration. Slightly increase in the arc voltage results the weld to bridge gaps when welding in grooves. Excessively high voltage produces a hatshaped concave weld, which has low resistance to cracking and a tendency to undercut. Lower voltages reduce the arc length and increase penetration. Excessively low voltage produces an unstable arc and a crowned bead, which has an uneven contour where it meets the plate. Following are some of the important performance measures of welding process, besides weld quality, toughness, hardness, ductility, HAZ and FZ etc: 1. Weld Strength (MPa). It is the most important performance measure that directly affects the weld efficiency and production cost. Mostly, the weld quality is based and judged by the weld strength and the strength of base metal. Many factors influence the weld strength including the base material, filler metal, weld type, joint type, weld method, heat input, and their interactions. Yield and tensile strength are measured using a standard tensile test from the weld specimen that is to verify the overload failure will occur in the base metal rather than the weld metal or HAZ. 2. Weld Induced Residual Stresses (MPa) & Distortions. Residual stress is the most important in the welding performance. Both, residual stresses and distortions are the major concerns in welded structures. The residual stresses in weld region are normally tensile and close to the material yield stress due to the shrinkage of the weld during cooling. Welding residual stresses not only cause distortion but also significantly affect the performance of welded structures specially, for the failures occurring under low applied stresses (pre-mature failure) such as brittle fracture, fatigue, and stress corrosion cracking. The residual stresses have a significant effect on the process of the initiation and further propagation of the fatigue cracks in welded elements. The fatigue life of the welded elements depends on the possible variations of the residual stress level and in many cases the residual stresses are one of the main factors, determining the engineering properties of structural components, and plays a significant role in fatigue of welded elements. In welding process, low values of residual stresses and distortions are desired. 3. Welding Temperatures (oC). The temperatures experienced by the metal produced by weld torch during the welding process are called as weld temperatures. Base metal and welding wire composition, metal thickness, weld geometry and high settings of welding parameters causes the high values of welding temperatures. The amount of heat input during welding process is very important as the high heat input results increase in heat affected zone (HAZ). The discussion in this thesis is confined to the optimization of GTAW or TIG welding process of thin-walled structure ( linear weld and circumferential weld only) to maximize the weld strength and minimize the weld induced residual stresses and distortion whereas the linear seam welds are not of significant interests in present research.

11

University of Engineering & Technology, Taxila-Pakistan

1.4 Current Challenges in Welding Domain For all the welding manufacturing facilities the following three goals are common, with varying magnitude of desirability: (1) maximize the weld strength; (2) minimize the weld induced imperfections; and (3) minimize the weld cost. Analyzing the arc welding process in this context, it can be asserted that this technology, because of its high values of strength, significantly contributes towards achievement of goal 1, although there are other aspects as well, which need considerable attention for improvement in overall welding process. As far as goal number 2 is concerned, the arc welding technology has come up with mixed outcomes. Though, the arc welding process results in high values of weld strength but due to weld induced imperfections resulting reduction of weld strength and nonconformances or rejection due to pre-mature failures. Optimal settings of welding conditions for minimization of weld residual stresses and distortions are required in order to make arc welding a highly effective technology. For the achievement of goal 3, the arc welding technology already possesses high potentials. The advantages offered by this technology have already been mentioned in previous sections. But there is always need for continuous improvement. Improvisation of welding process for economy, optimal settings of welding conditions for reduction of weld induced residual stresses and distortions & enhancement of weld strength are required in order to make arc welding a highly cost effective technology. Practically speaking, the weld quality needs to be further improved, especially in the case of thin walled structures welding, in order to make welding an unsurpassable metal joining technology. On the other hand, welding has a number of detrimental effects on the structural integrity and in-service performance of the weldments. These detrimental effects are due to imperfections induced by the welding in the weldments, of which the structural shape change behavior, residual stresses and the weld solidification cracks are reported to have very severe degrading effects on the mechanical strengths and possibly can leads to the catastrophic circumstances. A number of catastrophic failures of high pressure vessels of thin-cylindrical shell structures due to longitudinal and circumferential welds are reported in the recent past. An example of such failures of HSLA steel pressure vessel is shown in Figure 1.7 [4]. The figure shows a large diameter high pressure vessel fabricated with linear seam and circumferential welds failed during the proof test. A detailed failure analysis revealed that the premature failure, well below the yield limit occurs at the junction of the linear seam and circumferential welds due to the cumulative effects of welding residual stresses, local deformations at the junction and excessive Heat Affected Zone (HAZ). Tremendous efforts were made in the past to investigate weld induced imperfections by various researchers around the globe but so far now major emphasize has been placed on the prediction of weld induced imperfections like deformations and residual stresses in plate or sheet welds and tee joint structures. A very little significant contribution for the welding residual stress field distribution and its effects on the performance and structural integrity of roll formed circumferentially welded thin-walled cylinders is reported and the critical investigations of these structures is yet to be explored. To ensure the structural integrity of the structures for improved product quality and reliability, it is anticipated that the optimized 12

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

design due to the optimized weld parameters, reduced failure rates, improved product life and most importantly reduction in cost for re-welding / reworking for thin plate and cylindrical welded shell structures can be the major contributions from this research.

Fig. 1.7 Thin-walled Pressure Vessel failures due to weld induced imperfections [4] In the light of discussion provided above, the current challenges of thin walled structures welding process or the research gap present in this field – that have also set the aim of the research presented in this manuscript – can be listed in the following way: 1. Weld induced imperfections like residual stresses and distortions are a major demerit of arc welding technology that adversely affects the weld efficiency. Thus, it is a major need of the present time to search for the welding conditions that could significantly suppress the weld induced imperfections. By the term welding conditions it is meant here the different combinations of welding wire parameters (e.g., weld wire speed, wire composition, wire size etc.), welding parameters (e.g., welding speed, welding current, welding voltage, thickness of base metal & composition, weld type, weld geometry etc.), heating (pre-heating or post heating) and cooling (e.g., air or gas etc.). 2. The parameters that lead to enhanced weld strength do not necessarily provide minimum residual stresses or distortion. In addition, it is also well known that the parameters favorable for low distortion also cause increase in residual stresses. These two facts imply that the challenge sought is two-folded. The researchers are required, not only, to find the ways to minimize residual stresses and distortion but also to make sure that weld strength is not compromised. In other words, researchers have to find the trade-off among the two conflicting objectives: (a) maximize weld strength; and (b) minimize residual stresses and distortions. 3. It is also highly desired to have a fully automated system that should acquire knowledge from the data generated by the research activities and utilize that knowledge to: (a) work out the optimal welding conditions for achievement of desired 13

University of Engineering & Technology, Taxila-Pakistan

objectives in a best possible way; and (b) predict the values of performance measures based upon welding conditions selected. 1.5 Application of Artificial Intelligence and Expert System to Manufacturing The requirement number 3 described in the last section is a hot candidate for application of Artificial Intelligence (AI) tools. AI is a branch of science that imparts to machines the ability to think and reason. Precisely, it can be defined as the simulation of human intelligence on a machine, so as to make the machine efficient to identify and use the right piece of knowledge at a given step of problem solving [32]. The ultimate target of research in field of AI is to construct a machine that can mimic or exceed human mental capabilities including reasoning, understanding, imagination, and creativity [33]. In a very broad sense AI can be subdivided into two categories: (1) Knowledge-Based Systems (KBS); and (2) Computational Intelligence (CI). KBS is a kind of non-conventional computer program in which knowledge is kept explicitly separate from the control module of the program. The module that contains the knowledge, in the form of rules and facts, is called knowledge-base while the control module is called inference engine. The inference engine contains meta-knowledge i.e. the knowledge about how, where, and when to apply the knowledge. Expert System (ES) is a special kind of KBS that contains some extra frills like knowledge acquisition module and explanation module etc [33]. CI is different from KBS in the sense that in CI the knowledge is not explicitly stated in form of rules or facts, rather it is represented by the numbers, which are adjusted as the system improves its efficiency. One of the common forms of CI is the Artificial Neural Network (ANN) [33]. 1.5.1 Expert System It is a computer program designed to simulate the problem solving behavior of a human who is an expert in a narrow domain or discipline or field. An expert system is normally consisted of a knowledge base (information/heuristics etc.), inference engine (analyzing the knowledge base) and the end user interface (accepting inputs/generating outputs). In expert system, the path that leads to the development of expert systems is different from that of conventional programming techniques. Expert systems are capable of delivering quantitative information or for use in lieu of quantitative information. The other one feature is that the expert systems can address imprecise and incomplete data through the assignment of confidence values to inputs and conclusions. The ability to explain reasoning is one of the most powerful attribute of the expert systems. This ability enhances the user confidence in the recommendation and acceptance of the expert system. Further, development of expert system usually proceeds through many phases including problem selection, knowledge acquisition, knowledge representation, programming, testing and evaluation. The power of an expert system is based on the knowledge of the expert. The development of expert systems is in three parts: i) Selecting a problem, ii) Knowledge acquisition and representation, iii) Evaluation and adoption. To identify a suitable problem in developing an expert system is the most critical step. Expert systems are best suited to problems that require experience, knowledge, judgment, 14

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

and complex interactions to arrive at a solution. The knowledge acquisition phase of expert system development is started after the problem selection. The next step is to have the knowledge to solve the problem displayed which the expert used in a systematic way so that it can be coded into the computer. The development of expert systems is expensive. They need resources, expertise and time to build. As the ANN lacks explicit representation of knowledge, so it finds its application, mostly, in fault diagnosis and tool/process condition monitoring within the manufacturing process domain, whereas ES possesses high potentials for optimizing the process parameters and improving the manufacturing efficiency/effectiveness. As the main target of this research work is to enhance the efficiency and effectiveness of thin-walled welding process in form of achievement of objectives, detailed in previous section, the scope of this manuscript will remain concerned mostly with ES. 1.6 Objectives, Research Methodology and Organization of Dissertation 1.6.1 Objectives The scope of the research work in this dissertation is limited to linear butt weld of thin plates and circumferential welding of thin-walled cylindrical shell structures. Mechanical effects of welding such as welding strength, welding induced residual stresses and deformations are primarily focused. The primary objective of this research work is to optimize the welding process to enhance the structural integrity and in-service performance of arc welded thin-walled structures of high strength low alloy steel. The main objective under the scope of the present research work is to find out the ways to reduce distortion and residual stresses to minimum possible levels & maximize the weld strength and to find out the trade-off between them according to requirements of end user. As different welding situations require different settings of parameters for optimization, and considerable amount of effort and time is spent for determination of optimized values of parameters, it is mandatory to develop an automatic tool that would suggest the recommendations of different parameters, within no time, for any kind of welding situation. The potential solution is development of a self-learning expert system for optimization of welding process and prediction of its performance measures. In context of the discussion provided in this chapter and in perspective of the industrial requirements related to the welding technology, the objectives of this research can be enlisted as follows: 1. Find out the options for enhancement of welding strength for any given welding conditions. The term options comprises of the search for best welding parameters, weld wire parameters, and welding environment parameters (e.g., cooling) that could suppress the residual stresses & distortion and ultimately increase the weld strength to a maximum possible level. 2. Work out the best welding, and welding environment parameters that could provide minimum values of residual stresses and distortion for given welding conditions. 15

University of Engineering & Technology, Taxila-Pakistan

3. Find out the ways to satisfy above-mentioned two objectives simultaneously. 4. If objective number 3 cannot be persuaded in absolute manner, then work out the strategy for finding the trade-off among three objectives: (a) maximize weld strength; (b) minimize residual stresses & distortions. The trade-off strategy must care for the priority levels assigned to each objective by the user. 5. Accomplish the objective number 3 or 4 in form of an expert system. The expert system must possess the ability to recommend the optimal values for input welding parameters according to the priorities assigned by the user to the different objectives. It should also have capability to predict the values of welding performance measures according to the values assigned to the input parameters. 6. Impart to the expert system the abilities of self-learning, self-correction, and selfexpansion of its expertise. The expert system should possess the ability to account for newly introduced welding variables, to generate new rules and re-adjust the fuzzy sets according to new data provided, to resolve the conflict among opposing rules, and to include or remove any output variable from the set of objectives according to the user’s need. 1.6.2 Research Methodology Methodology of research for accomplishment of research objectives has been summarized into three stages: 1. First stage consists of determination of effect of each welding parameter upon weld strength, distortion and residual stresses by analysis of welding process parameters. Main welding parameters include welding current, welding voltage, welding speed, sheet thickness, and more. A strategy would be developed in order to quantify the residual stresses and distortion with different parameters and to decide which parameters need to be included in the study. Design of experiments (DOE) techniques would be utilized in order to develop the experimental plan and ANOVA (Analysis of Variance) would be used to isolate the effect of each tested parameter upon response variables (i.e. components of weld strength, residual stresses and distortion). Numerical optimization would be utilized to optimize the influential parameters for minimization or maximization of the response variables. The process utilized would be Gas Tungsten Arc Welding (GTAW) applied to the thin walled circular and linear joints of a high strength low alloy (HSLA) steel. For analysis of residual stresses and distortion in circumferential welding of thin walled structure (cylinder), first finite element based numerical techniques are used for the analysis and model for prediction of response of complex welding phenomenon. Parametric approach is adopted to analyze various contributing factors in welding deformations and residual stresses and the FE models are experimentally validated for temperature gradients, deformations and residual stresses. Virtual DOE based on FE simulations would be applied for the analysis and optimization of circumferential welding process parameters variables and response (i.e. residual stresses and distortions) of thin walled cylinder for saving time and experimental cost effectiveness. 16

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

2. The next stage of research plan will comprise of development of a knowledge-based expert system, utilizing fuzzy reasoning mechanism for optimization and prediction of welding process. The ANOVA results, obtained in first stage, will be used to develop the knowledge-base, which would consist of optimization rule-base and prediction rule-base. The optimization rule-base will be the first one to take charge and it will operate with relevant set of rules for the optimal selection and combination of influential welding parameters (predictor variables), like current, welding speed etc. The prediction rule-base will then make use of the finalized combination of predictor variables and the relevant set of rules in order to estimate the values of performance measures. 3. The third stage of research work will emphasize upon embellishing the expert system with high-level automation. A machine learning algorithm will be developed and utilized that will cause the system to self-develop fuzzy sets for numerical variables, self-generate rules for both of the rule-bases, and self-adjust newly introduced variables and data. End user would need to just enter the experimental data to interface of the ES and system will upgrade itself automatically without need of human intervention for welding. The expert system will improve its accuracy of optimization & prediction as it will be used more and more. The detail research methodology for accomplishment of research objectives has been given in the following: For the accomplishment of objectives 1 and 2, sets of welding experiments have been performed in actual for linear welds and virtual for circumferential welds after the welding analysis in FE with experimental validation, the plans of whose have been governed by the statistical technique called Design of Experiments (DoE). Two kinds of design of experiments have been employed depending upon the targets sought, welding parameters to be tested, results desired, and availability of material. These two kinds of DoE are: (1) Full Factorial design; and (2) Response Surface Method (RSM). The significance of each design will be discussed along with the practical details of each experiment set. First, the complex phenomenon of arc welding process is investigated by using a hybrid experimental and modeling & simulation approach for circumferentially welded thin-walled cylinders. Thermal (temperature profiles) and subsequent structural (transient and residual stress /strain fields, residual deformation patterns) domains are investigated for material of HSLA steel as shown in Figure 1.8 for virtual experiments and they all directly relates to the practical application of the welding process. Primarily modeling and simulation techniques are employed to investigate the insight of arc welding process with moving heat source. Thermo-mechanical behavior of high strength low alloy (HSLA) steel is presented. Further, experimental techniques are employed to validate the thermal and structural fields for better correlation of the process. In the experiments the effects of welding parameters have been tested upon performance measures like weld strength, residual stresses and distortions. The hardware involved for conducting the experiments will be explained in following chapters.

17

University of Engineering & Technology, Taxila-Pakistan

Manufacturing Machining

Metal Forming Arc Welding

EBW

GMAW

Plasma

TIG Welding

Circumferential

Linear Welds

Welds Current, Voltage, Weld Speed, Trailing, Thickness

RSM, Distortion Tensile Sample Testing

Base Metal HSLA Steel (t = 3, 4, 5 mm) Filler Metal – H08

EXPERIMENTS

Virtual DOE for Circumferential Welds based on 3D FE Modeling & Simulation

Experimental Work Modeling & Simulation

Thermal Field

Temp. Field

Mechanical Field

Stress/Strains Deformations

Fuzzy Logic

Fuzzy Sets

Max-Min Inference C++ Coding Data Structures

ANOVA, Regression, Numerical Optimization

Design of Experiments

Factorial Design

Weld Strength, Residual Stresses, Distortion, etc

Rule-Base Optimization Prediction EXPERT SYSTEM

Programming 1D linked list 2-D linked list

Response Surface Method (RSM)

SELF-DEVELOPMENT

IF -THEN Rules

Knowledge Engineering

Optimal Formation

Simulated Annealing Algorithm

Machine Learning New Algorithms

Data Acquisition, Fuzzy Sets, Optimization & Prediction Rule-base Conflict Resolution

EXWeldHSLASteel A self-developing ES Fig. 1.8 Scope and methodological strategy framework of the research 18

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Influence of each welding parameter upon performance measures has been analyzed using statistical technique called Analysis of Variance (ANOVA). ANOVA helps to determine whether the effect of any parameter upon certain performance measure is significant or not. Only those parameters, whose effects are significant upon important performance measures, are used in calculations or further experiments for achievement of best values of those performance measures. Regression has been employed to develop the empirical models of weld strength, distortion and residual stresses in terms of numeric and significant input parameters. The empirical models covering all the significant numeric input parameters for estimating the performance measures were developed on the completion of all the required sets of experiments. Accomplishments of objectives 3 and 4 have been sought using numerical optimization technique. Following on from ANOVA results, the optimal values of input parameters can be determined for the case maximization or minimization of output parameters (performance measures). If the number of output parameters is more than one then the optimization of all of them is worked out simultaneously. The ANOVA results and numerical optimization provided the input data for development of knowledge-base of the expert system that marked the accomplishment of objective 5. The expert system was developed in two modules: optimization module and prediction module. Simulated Annealing Algorithm was employed for the optimization of the rule-base. Fuzzy logic was used as reasoning mechanism in order to cope with the uncertainties in relationships between input parameters and output parameters. Objective 6 has been accomplished by developing innovative algorithms for storing and retrieving variables and data; for developing fuzzy sets; for developing knowledge-base; for resolving conflict among opposing rules; and for updating the expert system interface. C++ language was used to program the pseudo-codes of these algorithms. Figure 1.8 shows, in graphical form, the scope and methodological framework of the research presented in this dissertation. 1.6.3 Organization of Dissertation Manuscript of the dissertation has been divided into eight chapters. Chapter 1 provides the introduction to welding technology, arc welding, TIG welding process, Artificial Intelligence application to manufacturing and Expert System. Based upon this overview the current challenges of welding technology have been pinpointed and stated as targets of this research. Finally, the methodology to accomplish the stated targets has also been elaborated. Chapter 2 includes the overview of previous research carried in welding domain, welding simulations; experimental work pertaining to circumferential welding, weld induced residual stresses and distortions, residual stresses measurement and Artificial Intelligence (AI) to manufacturing & welding and the limitations. Chapters 3–5 cover the analyzing welding parameters, FEM modeling & simulations, and statistically powered experimental techniques of studying the TIG welding process. The main targets of the research have had been the identification of influential parameters upon welding performance measures and the quantification of their effects for the optimization. 19

University of Engineering & Technology, Taxila-Pakistan

Chapter 3 includes the welding experiments for analyzing the TIG welding parameters and their significance and effect on response (i.e. weld strength, residual stresses and distortions) for linear welds of thin plates by using DOE, ANOVA and numerical optimization. 2-Level full factorial and response surface method (RSM) designs based on two factor interaction (2FI) model are used to quantify the effect and significance of input parameters/variables upon the response. Chapter 4 presents the welding simulations containing analytical model, FE formulation, heat models, material model, simulation approach in ANSYS and covered the details of welding induced residual (axial & hoop) stresses fields and distortions (axial & radial shrinkages) and also covers the details of experimental setup for the FE models (thermal and structural) validation for circumferential welding. Chapter 5 includes the effects of welding process parameters (weld speed and heat input), geometric parameters (cylinder thickness), root opening and tack weld orientations on residual stresses and distortion, and includes the detail of virtual design of experiments (DOE) by utilizing the simulations and optimization of welding parameters for thin walled shell structure of high strength low alloy steel and details the use of full factorial technique and response surface method for design of experiments, to investigate the effects of welding upon performance measures in weld experiments. ANOVA was used to analyze the data and optimization was performed using numerical optimization for data collection to have knowledge base for the development of expert system. Chapters 6 – 7 deal with application of expert system for optimization of welding process. Chapter 6 details the expert system configuration, including description of optimization and prediction modules; fuzzy reasoning, including data fuzzification, rules aggregation, and output defuzzification; and utilization of simulated annealing algorithm for optimal formation of the rule-base. Chapter 7 of the dissertation describes the algorithms for imparting the expert system with capabilities of self-learning, self-correction, and selfexpansion. This chapter also includes the description of data structures for programming pseudo-codes of these algorithms, interfacing of expert system and the application examples of expert system. Chapter 8 presents the important conclusions drawn from the analysis and discussion provided in the chapters of the manuscript, the recommendations and usefulness of developed expert system in welding domain regarding applicability of the research conducted, and some of the proposals regarding future research work that can be conducted on the basis of this research.

20

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

CHAPTER 2 LITERATURE REVIEW 2.1 Introduction This chapter presents the details of literature survey conducted in arc welding domain for the analysis of welding residual stresses and distortion followed by the research done in welding process improvement and optimization by applying design of experiments, welding simulations and specifically research related to circumferential welding computational works of thin walled structures for the development of FE model of thin walled HSLA steel structure for circumferential TIG welding of cylinder in this research work. Further, this chapter covers the main research work related to welding induced residual stresses measurement and residual stress measurement by hole-drilling in thin plates and for the experimental validation of FE model to be developed and used for virtual welding experiments for the optimization of welding process. Later, it provides the details of literature survey in artificial intelligence (AI) and expert system (ES) to manufacturing, welding and the limitations of previously developed AI tools for the development of expert system in this research work. 2.2 Literature Survey in Welding Domain Welding process is an important domain of activity of to-day industry, especially in the sector where a lot of assembly of structures has to be done. Generally, the initial design of industrial parts requires revisions, because unpredictable changes occur in the shape or the performance of a component when it is welded because its metallurgical structure was modified by the process. The re-design is very costly since that happen late in the development cycle of the product. It is now possible to avoid this by anticipating the undesirable effects of the manufacturing process early in the design stage, by using numerical simulation, numerical optimization and AI tools. This shows how much virtual welding process can be important in the development of a new component. Welding simulation utilize to optimize process parameters during the earlier stages of a new design cycle avoiding expensive errors that could occur later. The industrial benefits of using welding simulation and optimization are: i)

Minimize distortions: Simulation allows to predict the distortions and to minimize them by optimization. The effects are of increasing the overall quality of the product and of reducing the costs. Increase in heat input during welding can cause the problems of fracture resistance and deformations. The welding cost increases with the increase of welding volume.

ii)

Residual stresses: The goal is to minimize the gradient and to have a smooth distribution of the residual stresses resulting from the welding process. By acting on the welding process, one can have compressive stresses on the surface of the component, which improves its quality and avoids corrosion due to tensile stresses. Residual stresses can be optimized by applying design of experiments and simulations.

21

University of Engineering & Technology, Taxila-Pakistan

iii)

Knowledge of welding process: Welding analysis by ANOVA and simulation allows to define the best welding sequence and to control all the parameters of the welding process. It is important to optimize the amount of heat input brought by the heat source during the process to maximize the weld strength and minimize the deformations. By mastering the process involved, one can use the right parameters to achieve the desired response, and the productivity is improved by applying process knowledge base.

2.2.1 Welding Experiments and Optimization The design of experiments (DOE) is a technique that is used for the planning, conducting and analyzing the experiments to have the efficient and economical conclusions. DOE technique is a proven method used for the improvements of the process yield, process performance and to reduce the process variations. It is a powerful statistical technique used to study and analyze the effect of various process parameters upon the response. In early 1920, it was developed by Sir Ronald Fisher. It was the replacement of traditional approach of experiments i.e. one factor at a time. The selection of experimental design is very important for the success of experiment which depends upon the problem nature, resources and infrastructure available to perform the experiment and budget constraints. The statistical analysis and interpretations of results, effect of significant factors and their interactions on the response/output, is very important to meet the objectives in DOE. Normal probability plot, a graphical tool, is used to identify the real effects by plotting the main and interaction effects along the straight line. Analysis of variance (ANOVA) is used to identify the significance of factors and their interactions. The application of design of experiment (DOE), evolutionary algorithms and computational network are widely used now-a-days to develop a mathematical relationship between the welding process input parameters and the output variables of the weld joint in order to determine the welding input parameters that lead to the desired weld quality and optimization of response. Welding process input parameters play a very significant role in determining the quality of a weld joint. The joint quality can be defined in terms of properties such as weld-bead geometry, mechanical properties, residual stresses and distortion. Generally, all welding processes are used with the aim of obtaining a welded joint with the desired weld-bead parameters, excellent mechanical properties with minimum residual stresses and distortion [34]. In [35], a design of experiments approach was chosen as an efficient technique to maximize the information gained from the experimentation for the reduction of pores in welds by laser at a car production line as case study and an average reduction in the number of pores of 97 per cent was obtained. In [36], the researchers presented the use of response surface methodology (RSM) by designing a four-factor five-level central composite rotatable design matrix with full replication for planning, conduction, execution and development of mathematical models for predicting the weld bead quality and selecting optimum process parameters for achieving the desired quality and process optimization of Submerged arc welding (SAW) of pipes of different diameters and lengths.

22

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

In [37], the authors presented that a software (Design-Expert) was used to establish the design matrix and to analyze the experimental data. The relationships between the laserwelding parameters (laser power, welding speed and focal point position) and the three responses (tensile strength, impact strength and joint-operating cost) were established for butt joints made of AISI304 and optimization of welding process was performed after ANOVA to increase the productivity and minimize the total operating cost. In [38], the same authors further presented the effect of laser power (1.2–1.43 kW), welding speed (30–70 cm/min) and focal point position (−2.5 to 0 mm) on the heat input and the weld-bead geometry (i.e. penetration (P), welded zone width (W) and heat affected zone width (WHAZ)) was investigated using response surface methodology (RSM) based on Box–Behnken design for laser butt-welding of medium carbon steel using CW 1.5 kW CO2 laser and the results achieved indicate that the proposed models predict the responses adequately within the limits of welding parameters being used. Further as in [37 and 38], the same researchers presented and used the design-expert software to optimize the keyhole parameters (i.e. maximize penetration (P), and minimize the heat input, width of welded zone (W) and width of heataffected zone (WHAZ)) in CW CO2 laser butt-welding of medium carbon steel by using the process parameters i.e. laser power (LP), welding speed (S) and focused position (F) to achieve the desirable weld bead quality and to increase the production rate and minimize the total operation cost [39]. In [40], the author introduced an experimental welding technique which is called lowstress non-distortion (LSND) welding for the LSND welding of AL-alloy and steel sheets for aerospace structures to overcome of an uneven temperature distribution and restraining forces applied to the welded parts to prevent out-of-plane buckling caused by the heating and welding processes. The experimental data indicate that the stretching effect following the LSND welding can be used to control welding stress as compared to stretching effect with conventional one-point clamping jig only. LSND welding technique shown to be suitable for materials that are generally fusion-welded with any heat source and the resulting structures are generally free from significant heat distortion. In [41], researchers presented a specially designed test rig which was developed and used for assessment of thermal and residual stresses for given welding conditions characterized by the peak temperature and cooling time of the thermal cycle of high strength low alloy quenched and tempered steel with specified minimum yield strength (690 MPa). An induction coil used for programming the heating and cooling of small specimens for simulation of actual weld thermal cycles. The chosen range of peak temperature and cooling time produced varying effects on the temperature field, microstructural state field, and mechanical field. The longitudinal expansion and contraction of the test specimen was limited during heating and cooling to synthesize the restraint imposed by welding. This technique facilitated the study of important relationships between weld thermal cycles, phase transformations and residual stresses. The thermal stresses are strongly influenced by the transformation, i.e. by the transformation of austenite into ferrite, pearlite, bainite or martensite, since these solid state transformations involve a volumetric expansion. In the aswelded condition for fracture assessment of structures, the tensile residual stresses are assumed to be equal to the room temperature yield strength of the material which exploits the strength benefit of QT steels.

23

University of Engineering & Technology, Taxila-Pakistan

Many welding distortion mitigation methods have been developed by the researchers to eliminate the welding induced imperfections which are the major concerns of welding industry. For this purpose, several researchers have used the trailing heat sink during welding to minimize the distortion. This method is called dynamically controlled low stress no distortion (DC-LSND) welding which was first developed and introduced by Guan et al. [42]. However, still its practical application and implementation is complex. In this method, a trailing heat sink is attached at some short distance behind the welding heat source and moved as the welding heat source. Usually this method is used to control the welding buckling of thin plates as the compresses stresses developed during welding of thin sections exceed the critical level of buckling stress. The welding longitudinal residual stresses effects significantly with the application of trailing heat sink and residual stresses remains below the critical buckling stress level and consequently minimize the buckling. In [43], the two steel plates of AISI 316L of size 250x100x1.5 mm were welded by TIG welding with same parameters (3mm/s, 750 W) with application of trailing heat sink ( at fixed distance of 25 mm from welding torch, CO2 as cooling media of trailing). The plate welded without trailing application was severely buckled whereas the plate welded with trailing application was free of buckling. Three main factors have been investigated i.e. the heat sink influence on temperature, the temperature influence on residual stresses and residual stresses influence on buckling. FE method for thermal mechanical finite element models is used to study the effect of cooling source parameters (with fixed parameters: 580 W Heat input, 75% efficiency, 3 mm/s welding speed, and 293 K heat sink temperature whereas the variables: 0-30 mm heat sink diameter (d), 15-50 mm distance (L) from heat source to heat sink) on temperature field and further the temperature history used for the study of residual stresses. In [44], the researcher presented several approaches that have discussed in this research to analyze the effect of the cooling source parameters. It was determined analytically that the sensitivity of buckling depends upon stress level and their distribution behavior and decreases with the decrease of width of compressive zone at the plate edges that can be achieved with the increase of tension zone width or compressive zone on the weld. It can be done by predicting and using optimum cooling source parameters. Many approaches used for cooling source to have desired effect with some merits and demerits i.e. heat extraction by contact (roll contact with copper shielding box) cooling to have no interference with arc but surface resistance to contact during moving with heat source, width of cooling contact and pressure of contact etc. However, the most useful type of cooling source is a jet of fluid which follows the welding torch at a short distance. Similarly, many cooling media like compressed air, water, argon gas, helium gas and liquid nitrogen were used to study the effects after performing several tests. The water and air are cheap and available very easily. The inert gases (argon and helium) having different thermal properties provide extra shielding for environment interference with material. Deformation was checked after TIG welding of steel plates (250x100x2 mm) with application of cooling and without the application of cooling during welding. Maximum absolute deflection was measured by pressing one end of plates at flat surface plate and plates were buckled with different values. The absolute deflection (in mm) of welded plates was measured with the application of following types of cooling sources: no cooling, air 24

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

cooling (low), water cooling, air cooling (high), and helium cooling whereas the measured values were 22, 18, 17, 16 and 15 mm respectively. The result values shows that the deflection decrease 19% with air cooling (low), 23% with water cooling, 28% with air cooling (high) and 32% with helium cooling respectively. However, the CO2-snow jet is the best cooling source during welding which resulted drastic decrease in temperature and consequently deflection but with the drawbacks of un-stability and practical implementation [44]. 2.2.2 Welding Simulations The main objective in this section would be description of major contributions and current status of the research in the field of welding simulation specially in circumferential welding of axis-symmetric structures specifically of thin walled structures like cylinders or pipe. Welding simulations techniques are mature enough today to be used in an industrial context. However, they are extremely CPU time consuming and that is due to the very fine discretization of the problem needed to observe microscopic phenomena like metallurgical transformations. The modeled problems are also highly non-linear. High quality, low cost and shortened time in welding are now the keys of success for manufacturing industries. Now, thermal-mechanical FEA of the welding process is an emerging and rapidly maturing technique. Computer aided design of the welding process is becoming an efficient and effective approach to achieve high quality weld products with reduced residual stresses and distortions. In recent years, with the development of powerful computing and data storage facilities, finite element analysis (FEA) methods have been applied to the simulation of structural behavior using commercial FE software packages. In welding process, the measurement of transient thermo-mechanical history is very important but it is very expensive and time consuming. Usually it fails to give a full picture of temperature, stress, strains and deformation distribution in the welds. Whereas the detailed experimental measurements of the residual elastic strain distributions in welded parts are typically not feasible due to the consumption of significant resources e.g. man, machine & material. The mathematical modeling of residual stress evaluation is a resource effective method in comparison to the experimental methods but again the development of the modeling requires a very careful experimental data. The analytical approaches were replaced by numerical approaches after the advent of finite element (FE) based numerical simulation techniques for modeling in welding. It is possible to study and analyze the nonlinear effects like temperature-dependent convection and radiation to the surrounding medium, plastic flow and volume expansion during final phase transformation with the use of finite element method (FEM). The welding process simulation technique is very expensive due to the requirement of high computational resources and large data storage capacity, and due to this constraint, some simplification in the FE model made to reduce computational expenses in the past. Two-dimensional finite 25

University of Engineering & Technology, Taxila-Pakistan

element models were analyzed by using the assumptions of lateral and rotational symmetry in most of the studies regarding circumferential welding. The brief discussion of the experimental, analytical and numerical work related to the analysis of arc welding to predict weld induced imperfections like residual stresses and deformation is given below. By the mid of 1940, the analytical determination of welding and its consequences studied. However, the significant contribution related to deformations of welded structures was made in 1950 that led to numerical modeling of heat distributions and structural aspects of welding processes. The thermo-mechanical analysis of welding is non-linear due to multifield interactions, non-linear thermal as well as structural material response. The field of FEA of complex welding phenomena is gaining importance in the past three decades and a significant research work has been contributed and reported in the published literature. The most of the research work on thermal-structural behavior was dependent on experiments in 1960-1970 whereas there were only few series classical finite difference solutions for the non-linear analysis of transient welding heat transfer before 1970. Further, major efforts were made for the development of computer codes to analyze the complex mechanism of heat flow through the weldments in late 1970 and sufficient literature on welding simulation research reported. A detailed review was compiled by Lindgren [45-47] in three parts in the past and another recent comprehensive review was given in [48]. The most of the unwanted consequences (residual stresses and deformation) from welding process are considered due to the non-linear heat flow introduced through a moving heat source. Therefore, the historical development and significant contributions in the heat source modeling are presented in this section. Modeling of moving heat source for the analytical solution of transient temperature distribution in arc welding process presented by Rosenthal [49] was the first step towards the simulation of welding phenomenon. The author presented linear 2D and 3D heat flow in solid of infinite size or bounded by planes and validated the model through experimentally measured temperature distributions during welding of plate of different geometries e.g. length, width and thickness. A predefined temperature at some defined locations of weld was used by Goldak et al. [50]. To overcome the issues in previously presented heat source model, Goldak et al. [51, 52] developed the most dominating heat source model with Gaussian heat source distribution which also known as Double Ellipsoidal Heat Source model and most widely utilized now-a-days. The Goldak heat source model with some necessary adjustments of parameters according to the requirements is used for heat source modeling in this dissertation in chapter 4 for circumferential welding of thin walled cylinder. The application of finite element techniques in welding simulation was pioneered by Ueda and Yamakawa [53] and Hibbitt and Marcal [54]. Further, Friedman [55] and Andersson [56] presented some research work which clarifies the methodologies involved in welding simulations and presented the basic methodology for sequentially coupled analysis technique. They analyzed butt-welding of plates and used plane strain formulation with half sectional model for thermo-mechanical analysis of welding and various experiments were conducted to evaluate the effect of modeling techniques, element types, mesh intensity, modeling of filler materials, type of solvers and numerical integration procedure etc. Both 26

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

studied the material modeling in which the effects of different material properties on thermal and structural response. The most significant contribution was made by McDill et al. [57-61] regarding the efficiency and integrity of the computational technique by developing graded element for dynamic adaptive meshing. At the places of high thermal or stress gradients, the automatic mesh refinement was achieved and reduced the computational expense substantially by decreasing the total number of elements in the model by using this gradient dependant adaptive meshing scheme. Further, for adaptive re-meshing, the same scheme was modified by Runnemalm et al. [62, 63], Lindgren et al. [64], Hyun and Lindgren [65]. 2.2.3 Circumferential Welding Computational Works Many stress analyses were carried out to understand the residual stress distributions induced by welding processes by analytically and numerically [66]. Analysis accuracy depends on the type of model used and the type of computational resource available for solving problem. For simple geometries like welding of plate, there are several available simplified analytical expressions for the determination of distortion and residual stress fields. For butt-welded plates, many researchers [5, 67] summarized such formulas empirically and analytically for the longitudinal, transverse shrinkage and angular distortion. It is assumed in the simplified analytical methods that the welding residual stresses are determined by the local plastic shrinkage strains generated during cooling after the welding process. The weldment undergoes many complex physical changes, involving interactions between microstructural, thermal and mechanical changes during welding processes, and the plastic strains accumulated during the final stages of cooling mostly determine the residual distortion. It means that the temperature and corresponding strain history available during the early stages of welding can be ignored. This simplified method is also known as Inherent Strain Method [69]. Vaidyanathan, Finnie and Todato found residual stress in piping by imposing on a cylinder the residual stress profile generated by a similar weld in a plate through using thinshell theory [70]. There is limited application of this technique to thin-walled pipes with only one weld pass. The Inherent Strain Method in which a combined experimental and analytical approach used to determine the source of residual stress by utilizing the characteristics of the distribution of inherent strains induced in a long welded joint was developed by Ueda et al. [69]. This method was further developed by Hill and Nelson [71]. The inherent strain model assumes an axis-symmetric condition which is incapable of predicting the transient residual stress distributions near the weld start and stop location [72]. Another analytical study by Vaidyanathan et al. [73], the methodology for determination of residual stresses in thin-walled cylindrical shells welded by single passes full penetration welds was described in this study and same material was taken for both i.e. base metal and weld metal. The variation in the calculated and measured stresses values for welding of plate is considered due to the differential thermal expansion of base and filler metals. They extended their study further for variety of the welding conditions, partial penetration of weld, multi-pass weld, and also different materials for base and filler metals. Rybicki et al. [74] presented a numerical study of multi-pass welding regarding the effect of pipe wall thickness on welding residual stresses, which is of significant importance for relating residual stresses with geometrical size of the pipe. Basically it was a parametric study in which basic FE model was validated for residual stresses measured experimentally 27

University of Engineering & Technology, Taxila-Pakistan

and subsequently developed FE model was used for different welding parameters and geometrical dimensions of the pipe. The author concluded the tensile axial residual stresses on the inner surface of four inch diameter pipe and observed higher tensile stresses for relatively thin pipe. Further, as the thickness of pipe increased, the tensile stresses near weld centerline decreased and compressive stresses away from the weld centerline increased. The author also concluded that the zone of influence of stresses increased with both the diameter and the wall thickness increase. The deformation and stresses for butt welded thin walled pipes were analyzed by Lindgren and Karlsson [75]. For moving line heat source for thin plates, analytical solution of Rosenthal [49] was used with the assumptions that temperature through the thickness was uniform and radius of the pipe was much larger than the wall thickness. This model was the first complete three-dimensional models for deformation and stresses in thin-walled welded pipes. To create model geometry, shell element was used which based on the degeneratedsolid concept. This technique was used in two cases i.e. butt-welded plate and butt-welded pipes. The analytical solution for moving line heat source in thin plates presented by Rosenthal [49] was used for thin shells with the assumption that temperature across the thickness was uniform and radius to thickness ratio was very high. The temperature dependant material model which included phase transformation effects was taken from Karlsson and Josefson [76]. The strain hardening was not considered in the model. The comparison between experimental data obtained from literature and calculated axial residual stress on outer surface shown relatively better agreement but there was complete disagreement between the hoop stress profile at outer surface near the weld centerline and the experimental measurement which was considered due to the inadequacies in the material modeling. A numerical study for butt welding of pipes by using shell elements was presented by Karlsson et al. [77]. By considering lateral symmetry, half of the model was analyzed. The analytical model for temperature distribution and temperature dependant material properties were taken from literature along with several other simplifications to reduce computational load like absence of root gap and tack welds and planer symmetry were utilized. A reasonable agreement was achieved by comparing the predicted radial shrinkage with experimentally determined values of a manually welded pipe. The variation in measured and calculated values was considered due to manual welding technique. Due to adequate results for residual stresses by 2D model, the authors concluded that 3D model is essential for transient and residual strains. A complete three-dimensional FE model with solid elements for welding simulation of a 100 mm nominal diameter pipe-pipe joint with 8 mm wall thickness was presented first time by Karlsson and Josefson [76]. To reduce the computational expense, symmetry along the weld centerline was also used. The material used for pipe and filler material was of low carbon steel SIS 2172. A single-pass butt-weld joint geometry was used to model of MIG welding. To model sequentially coupled thermal stress analysis, a finite element code ADINAT was used. The heat was generated in thermal model as consistent nodal heat flow corresponding to the volume of internal heat generation. The intensity of power was assumed to increase linearly in the axial direction and radially inwards. The results of transient and residual strain and stresses were presented and compared with the past experimental, 28

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

analytical and numerical results. However, hoop stresses were almost axis-symmetric except the weld start position. In [78] author presented a finite element study for multi-pass welding of components of piping and pressure vessel. Axis-symmetric 2D FE models for girth weld and plane strain formulation for longitudinal seam were used for the welding of thin pipes of thickness 0.25 inch and thick pipes of thickness 1.3 inch. For FE thermal analysis, Goldak's double ellipsoidal heat source model [51] was used and to model filler material, approach of inactive element was used. In this technique, all the elements of the weld passes were removed from the model before the weld pass and to update FE model for each weld pass, model change option was used. The assumption of elastic-plastic material was made with kinematic work hardening and solid-state transformations were also not included. The computational methodology was validated with good agreement by comparing temperature and stress distribution with experimentally measured data. Center hole-drilling technique was used to determine residual stresses experimentally. Various geometrical and loading parameters were studied like thickness, weld joint geometry, radius to thickness ratio, hydro-test and end restraint effects. The author concluded that the pipe wall thickness has very small effect for an r/t ratio of 10 but for r/t ratio of 25 and 50 with the same joint type (double "V"); the axial residual stress at the inner diameter is higher in thin-walled pipes. Further, as the r/t ratio increases in thin pipes, the magnitude of axial residual stress at the inner diameter reduces and whereas in thick pipes, a little variation in stress is observed. Also, the authors described that the hydro-test reduces the axial residual stress at the inner diameter in girth welds and the reduction is higher in thin pipes than in thick pipes. Further, the hydro-test has negligible effect on the transverse residual stress of seam welds whereas the magnitude of axial residual stress in girth welds of thick pipes of short length with restrained ends is equal to yield stress through thickness. In [79], modeling and experimental validation of residual stresses in 304L stainless steel girth welds in small diameters pipes of 38 mm outer diameter was presented. A 3D decoupled thermo-mechanical simulation with axial symmetry for autogenous Gas Tungsten Arc Welding (GTAW) was used. The volumetric heat source with filler metal addition along with the convection and radiation losses to predict residual stresses was considered. Bammann-Chiesa-Johnson (BCJ) constitutive model was used to model the mechanical material behavior at elevated temperature. To capture the temperature profiles at 6 mm and 12 mm from the weld line for the thermal model validation, thermocouples were used and for the validation of mechanical response, X-ray diffraction method for the measurement of residual stresses was used. The author concluded close agreement by comparing of predicted and measured axial and hoop residual stresses on outer surface of the pipes. A correlation between welding parameters and anticipated welding residual stresses at the inner surface of the pipe based on the data collected from different analytical, numerical and experimental studies reported in the literature was developed by Mohr et al. [80]. Author concluded that residual stresses at the inner surface is the function of heat input in each pass, number of passes and nominal wall thickness. It was also assumed that about all circumferential welds have some proportions to be welded in vertical and overhead position and these positions limits the heat input to maximum value of 50 kJ/in.

29

University of Engineering & Technology, Taxila-Pakistan

A 3D finite element simulation for austenitic stainless steel pipe by using shell elements was performed by Dong et al. [72]. The FE model was similar to the Karlsson et al. [77] and welding was completed in three passes. The temperature dependent material properties were taken from the literature. For experimental measurements of residual stresses at inner and outer surfaces were obtained by blind hole-drilling technique. A qualitative agreement was found between calculated and measured values of residual axial and hoop stresses on the outer surface. A parametric study was also performed to predict the effect of wall thickness and welding speed. Two wall thicknesses of 9.52 mm and 19.04 mm were analyzed for the effect of wall thickness on stress distributions and the results on the inner surface gave a similar trend of stress variation as given by Rybicki et al. [74]. Further, three different welding speeds of 4, 8 and 16 mm/s were used to determine the effect of welding speed. It was concluded that the effect on the stress distribution due to arc travel speed is minimum. In [81], the authors performed a parametric study for the determination of residual stresses in multi-pass girth-butt-welded stainless steel pipes. The main focus of the parametric study was to investigate the sensitivity of residual stresses due to variation in the welding parameters. A comparison was made between the predicted stresses and the expected stress state in ASME Section-IX. 2D FE models were studied by employing rotational symmetry assumption in order to reduce computational time. Various pipe diameters from 76.2 mm to 680 mm with corresponding wall thickness from 7.1 mm to 40.0 mm were studied. The number of weld passes depending on the pipe wall thickness varied from 4 to 36. The temperature dependant material properties with bilinear kinematic hardening and without solid-state transformation were taken from literature. The Von-Mises yield criteria were considered and time step duration was chosen on the basis of welding speed and fraction of circumference included in weld pass volume. To accommodate both the convection and radiation heat losses from the surface, a combined heat transfer coefficient was used. The variation of residual stress in radial direction resulting from different weld parameters were discussed and compared with the available literature results. A sequentially coupled thermal stress analysis for the determination of residual stresses was presented by Teng and Chang [82]. They demonstrated the effects of pipe diameter and wall thickness on residual stresses based on axis-symmetric FE models parametric studies. The temperature dependant material properties were used for the elastic-plastic material model without solid state phase transformations. The results exhibited self balancing behavior of residual stresses. The axial and hoop stresses were tensile on the inner surface near the weld centerline whereas stress reversal occurred in the regions away from the centerline. Increase in wall thickness of pipe reduced residual stresses at the weld centerline and increase in pipe diameter increased influence of residual stresses. Teng and Chang [83] further enhanced their contribution [82] to 3D shell element models and analyze welding residual stresses on the same sized pipes to investigate the effect of pipe wall thickness with same material properties, simulation strategy and most of the welding parameters. Two wall thicknesses i.e. 2.8 mm (schedule 5) and 3.8 mm (schedule 10) were only studied with change of welding speed according to wall thickness. Dong and Zhang [84] presented the effect of strength mismatch welds on residual stress field and fracture behavior. Their work was related to plate welding as well as thick-walled (1.5 inch) pipe welding. The temperature independent material properties with bilinear 30

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

kinematic hardening was used and to model multi-pass girth welding of 304 austenitic stainless steel pipes used 2D axis-symmetric FE model. In [85], the author experimentally evaluated various techniques for the reduction of tensile residual stresses at the inner surface of the pipes. Different welding techniques were used to control in-process of residual stresses. The results were compared of conventional and optimized GTA (gas tungsten arc) narrow gap welding. The authors concluded that the width of fusion zone can be reduced to about 45% with optimized narrow gap welding technique and found relatively less tensile residual stresses due to reduced axial shrinkage and less sensitized zone. The ring core method and XRD method were used for experimental determination of residual stresses. A three-dimensional finite element simulation of sequentially coupled thermo-mechanical analysis of multi-pass welding of pipes was presented by Frieke et al. [86]. The author investigated the weld induced residual stresses and three different techniques for the control of theses residual stresses. Two techniques related to in-process control were application of narrow gap welding and last pass heat sink welding. The third one technique relates to post weld stress mitigation was in-service aging technique. The 8-node linear brick elements were used for temperature dependency of material. A good agreement at discrete locations was concluded after the comparison of simulation results with experimental data but there was disagreement between axial stresses distribution along circumferential direction with experimental results and even did not exhibit qualitative agreement. It was concluded that residual stresses are by no means axis-symmetric on the basis of axial stress variation in circumferential direction. Further, it was presented that tensile axial residual stresses at inner surface change drastically near weld start and end locations.

Fig. 2.1 Axial residual stress plots on outer (left) and inner surface (right) [87] In [87] the authors presented an important contribution in circumferential welding by developing 3-D FE model and 2-D axis-symmetric FE model to analyze the temperature 31

University of Engineering & Technology, Taxila-Pakistan

fields and the residual stress distributions for arc welding of stainless steel (SS304) pipes. The numerical simulations results were compared with the experimental measurements and proved that the proposed computational procedure is an effective method for predicting the thermal cycles and the welding residual stresses. The authors presented that temperature distribution around the heat source is very steady when the welding torch moves around the stainless steel pipe. The simulated results from the 3-D model presented that the residual stress around the circumferential direction almost has a homogenous distribution except for the welding start part. The axial residual stresses on the inside and the outside surfaces showed a contrary distribution. A tensile axial residual stress was produced on the inside surface as shown in Figure 2.1 (right), and compressive axial stress at outside surface as shown in Figure 2.1 (left) in weld zone and its vicinity. Tensile axial stress was formed on the outside surface and compressive axial stress on the inside surface away from the weld centerline. In addition to their work the same authors along with Liang [88] presented hybrid numerical and experimental investigations of welding residual stresses in multi-pass buttwelded of medium thick-walled austenitic stainless steel pipes. Multi-pass welding experiments were performed initially to examine the evolution of temperature and residual stress trends. 2D axis-symmetric FE models and similar modeling & simulation approach as presented in [87] was applied without considering the metallurgical effects. The authors compared the results with experimental measurements and claimed a precise prediction of thermal and stress strain fields. It was also concluded from the results that the yield strength of the weld metal has significant effect on the final welding residual stress especially in the weld zone. Siddique [89] presented the field of arc welding simulations of girth welding. The author focused on the prediction of weld induced imperfections like distortions and residual stresses in pipe-flange joints and presented the effects of welding parameters, welding procedures and applied mechanical constraints on residual stress built up and transient deformations. Some efforts were presented to suggest the mitigation techniques to enhance the in-service life of welded pipe-flange structures. The effects of number and locations of tack welds on the transient and residual stress and strains and deformations on the pipe-flange joint were discussed. The results from some selected simulations were experimentally validated with close co-relation between the deformations and residual stresses. A 3D thermo-mechanical analysis was presented for the study of the effects of welding sequence on welding deformations in pipe joints of AISI stainless steel [90]. Parametric studies were conducted by using single-pass TIG welding with "V" joint geometry in pipes having a diameter of 274 mm and a thickness of 6.2 mm based on nine different welding sequences. Few experimental measurements based on conventional metrology were presented in order to verify the FE simulated data. The authors presented that it is possible to optimize the weld induced deformations by selection of a suitable welding sequence. A 3D finite element simulation of circumferential welding of mild steel pipes was presented by Lee and Chang [19]. Further, parametric studies with pipe inside radius to wall thickness ratio ranging from 10 mm to 100 mm were presented for the study of the effects of 32

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

pipe diameters on residual stresses. Literature based thermo-physical and thermo-mechanical properties of mild steel were used. The volumetric heat input through Gaussian distributed heat patterns to FE model was used in sequentially coupled manner. The simulated data was compared and verified by experimental data from literature. The research work did not included two important aspects i.e. effects of tack welds and effects of root openings on the corresponding residual stresses. The authors concluded that the 3D FE models are mandatory to precisely capture the significant effects at weld start and weld end locations and further concluded that the axial and hoop residual stresses were influenced by the pipe diameters in thin walled pipes. In [91] the author presented experimental and numerical studies on multi-pass, GTA butt-welded Incoloy 800 pipes. The temperature dependence of material properties along with birth and death techniques was used. The study was to investigate the thermal effects of Gaussian distributed heat source on circumferentially welded iron based Incoloy 800H superalloy pipes and to further investigate the structural results. The thermocouples were used for temperature distribution measurement within the HAZ. The temperature measurements were used for the calibration of numerical models and weld bead geometries from experimental micrograph. The authors concluded that fully volumetric heat flow provides best comparative results with experiments due to small thickness of the pipes of 5 mm used in the studies and further showed that 100% increase in heat input results in 100oC increase in maximum temperature of work-piece and the increased heat input results in wider weld zone. 2.3 Welding Induced Residual Stresses Measurement In view the in-service loading conditions of the welded structures, the effects of residual stress may be either supportive or contrary. Mostly residual stresses in structures are created due to mechanically induced plasticity or by thermal effects [92]. In most cases, residual stresses arise from the different production process [93]. A brief description of residual stresses and deformation in steel due to manufacturing process such as casting, welding, rolling, and heat treatment etc. is given in [94]. Mather (1934) developed semi-destructive center hole-drilling technique for measurement of residual stresses and it was further refined by Suite in 1949. A brief introduction and theoretical background of several new techniques can be found in [94-96]. The various experimental methods to evaluate the distribution of residual stresses are: nondestructive methods, such as, X-ray diffraction, ultrasonic and neutron diffraction methods, and destructive method like hole drilling method [97]. The detail classifications of residual stress measurement techniques are given in [89]. The neutron diffraction method is the only non-destructive method that determines residual stresses inside weldments [98] due to deep penetrating capability of neutrons. Shack et al. [99-100] made a substantial effort to experimentally measure the magnitude and distribution of residual stresses in butt-welded austenitic stainless steel pipes. Hayashi et al. [101] experimentally examined the spatial distribution of residual stresses in a socket-welded and a butt-welded pipe joints by neutron diffraction. A brief comparison of these experimental methods and a quick guide for appropriate technique selection for specific application is given in [102].

33

University of Engineering & Technology, Taxila-Pakistan

2.3.1 Residual Stress Measurement by Hole-Drilling The major contributions in the development of this technique are discussed in the following section. The center hole-drilling strain gage method of residual stress measurement is semi destructive technique and is based on measurement of released strain due to drilling of a small hole at the center of the strain gage rosette. This method was invented by Mathar [103] and measured released strain by using mechanical extensometer. Kelsey in 1956 [104], first time determined variation of residual stresses with depth by using this technique and further he developed methodology for blind hole-drilling for non-uniform residual stresses. Rendler and Vigness [105] developed systematic and reproducible procedure for uniform residual stresses and developed geometry of strain gage rosette, and that is being used now-a-days according to ASTM E-837 [106]. Flaman [107] used ultra high speed drilling for this purpose and related fundamentals of center hole-drilling technique can be found in [108] by VishayMeasurements Group (a manufacturer of hole-drilling equipment and strain rosettes). ASTM E-837 provides a standard procedure for measurement of uniform residual stresses through incremental hole-drilling method and a small hole is drilled at the center of special rosette, pasted on the surface of stressed component. Due to hole-drilling, stress profile changes and produce strain in the vicinity of hole. The measured strain corresponding to relieved stress is further used for the determination of residual stresses with the help of data reduction equations and calibration constants which are calculated either experimentally or by using FE formulation. Many variations have been introduced in the hole-drilling method to cater the nature of stress profile with time. There are four types center hole-drilling techniques for the determination of residual stresses as: i) Incremental strain method, ii) Power series method, iii) Average stress method, and iv) Integral method. Incremental strain method based on the assumption that incremental strain relaxation measured during each successive hole-depth is solely due to release of stress in corresponding hole-depth increment. For the determination of non-uniform sub-surface residual stresses, an approximate method called as power series method was introduced by Schajer [109]. This power series method is based on least square procedure to give best fit curve through the measured strain data and is best suited for gradually varying sub-surface residual stresses. Schajer [110] reviewed different stress measurement techniques and presented improved stress calculation procedure for power series method and integral method. The author further presented some practical implementation of integral method of center-hole-drilling technique. Later on Wern [111-112] and Wren et al. [113] used wavelets to solve integral formulation. In his work numerical implementation was done with conjugate-gradient method and succeeded even to get solution of singular matrix. Zuccarello [114] proposed optimized hole-drilling steps to minimize measurement error and proposed depth distribution which gives diagonal coefficient of the coefficient matrix. Aoh and Wei [115-116] further improved the coefficient, determined by Schajer [110] by using 3D FE model.

34

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

2.4 Literature Survey in Artificial Intelligence and Expert System to Manufacturing The artificial intelligence (AI) is related to intelligent behavior i.e. perception, reasoning, learning, communicating, and acting in complex environments, in artifacts having long term goals, both engineering and scientific, of development of machines that can do as human or better [117] . In artificial intelligence (AI) field until early 1970s, the researchers acknowledged that the general purpose problem solving methods developed since 1960s were not capable to tackle the to-day complex research and application oriented problems and felt that there was a need of specific knowledge related to a specific and limited domain of application rather than a general knowledge for many domains. This reason was made a base for the development of knowledge-based systems i.e. expert systems and this technology remained dominant in the field of AI. The history of numerous knowledge-based systems developed earlier can be found in [118]. A good and broad view definition of AI field by Tanimoto is as “Artificial Intelligence is a field of study that encompasses computational techniques for performing tasks that apparently require intelligence when performed by humans. Problems include like diagnosing problems in automobiles, computers and people, designing new computers, writing stories and symphonies, finding mathematical theorems, assembling and inspecting products in factories, and negotiating international treaties. It is a technology of information processing concerned with processes of reasoning, learning, and perception” [118]. The term artificial intelligence was named by John McCarthy in 1956. In brief, the AI is the science which provides the computers with the ability to represent and manipulate the symbols, as used by humans and those can be used to solve the problems that can’t easily solved through algorithmic models. In 1970s, the areas emerged in the AI field were knowledge-based systems (expert systems), natural language understanding, learning, planning, robotics, vision and neural networks. Knowledge-based systems can be defined as “a computerized system that uses knowledge about some domain to arrive at a solution to a problem from that domain. This solution is essentially the same as that concluded by a person knowledgeable about the domain of the problem when confronted with the same problem” [118]. The major distinguish between the knowledge based system and conventional programs is due to the three fundamental concepts i.e. the separation of knowledge from control (how it is used), use of highly specific domain knowledge, and the heuristic rather than algorithmic nature of the knowledge employed. The numerous advantage of knowledge based systems over conventional programs are: wide distribution of scarce expertise, ease of modification, consistency of answers, perpetual accessibility, preservation of expertise, solution of problems involving incomplete data, and explanation of solution, whereas the disadvantage may be as: knowledge limited to the domain of expertise, and lack of common sense [118]. An expert system (ES) that uses a collection of fuzzy rules, facts and membership functions to draw conclusion and uses fuzzy logic for inferencing rather than boolean logic is called a fuzzy expert system (FES) [119]. 35

University of Engineering & Technology, Taxila-Pakistan

In 1975, Lotfi A. Zadeh proposed the fuzzy set theories and fuzzy logic that deals with reasoning with inexact or fuzzy concepts. Fuzzy logic (FL) computes with words rather than with numbers whereas the fuzzy logic controller (FLC) controls with rules (IF-THEN) rather than with equations. Generally, FLCs are used in industrial process, consumer products, utility-services such as traffic control, satellite communication systems and military purpose. Mostly the applications of these are via fuzzy expert systems [119]. Traditionally, AI covers several application areas in manufacturing. Recently developed systems have demonstrated the importance of artificial intelligence based software to produce intelligent engineering software that can make many routine engineering decisions for welding applications and guide a human user to optimum decisions for welding to save cost and human hours. Mostly, these systems utilize expert systems and neural networks technology to provide and predict accurate weld process models and engineering decision making capability [120]. Usually expert systems in welding include the application of to select the suitable filler metal type and size, to determine the pre-heat and post-weld heat-treatment schedules, to determine welding parameters and others [120]. In [121] presented a fuzzy expert system approach for the development of the classification of different types of welding flaws in the radiographic weld domain. The fuzzy rules were generated from the available examples using two different methods and the knowledge acquisition problem was carried by using two machine-learning methods by using a simple genetic algorithm to determine the optimal number of partitions in the domain space. The author concluded that the improved knowledge acquisition method generates better fuzzy rules in terms of classification accuracy and the fuzzy expert system approach shown to perform better than the fuzzy k-nearest neighbor algorithm and the multi-layer perception neural networks approach based on the bootstrap method. Further concluded that the fuzzy models are transparent and human understandable, unlike the two other approaches. In [122] presented that expert system techniques are a more fruitful approach to the automated generation of procedural plans for arc welding than previous algorithmic methods. The main purpose was to evaluate recent computing advances in the context of planning for arc welding and to extract more generic knowledge about the application of expert system techniques to advanced manufacturing problems. A prototype expert system was presented which can generate welding procedures to the specification of BS4870. In [123] the authors developed an expert system for quenching and distortion control in a heat treatment process. The goals of this expert system were predicting results obtained under given quenching conditions and to improve the performance by supporting decision making. Different input parameters (quenchant category, quenchant temperature, agitation velocity, viscosity, geometry and density of part material, speed, carbon content, suspension and fixture type etc) were studied for the analysis of several parameters (desired suspension, cooling rate, cooling nature, hardness, distortion tendency and cracking potential etc). The system responds with predicted values as high, moderate or low and user can alter the input values to change the response.

36

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Artificial neural networks (ANN) are general, multi-variable, and non-linear estimators and the most important benefits of NN are non-linearity and input/output mapping. The most common network used is back-propagation (BP) which is stochastic approximation to nonlinear regression. Artificial neural networks have a wide scope in the field of welding analysis for obtaining optimum welding parameters, residual stresses and bead geometry (width, height and penetration) for a given set of welding parameters and a specific joint configuration. These are used to predict the temperatures distributions in the weld and to predict the size and shape of weld liquid pool on the given parameters like voltage, current, wire speed and travel speed and these can be helpful in selecting the preheat and post weld heat treatment schedules. In [124] a genetic algorithm and response surface methodology used for determining optimal welding conditions and desirability function approach was used for different objective function values. Application of the method proposed in this research revealed a good result for finding the optimal welding conditions in the gas metal arc (GMA) welding process. In [125] researchers have developed direct measuring techniques for welding residual stress. The existing tools are limited in application. This research details the development of intelligent techniques and use of a function-replacing hybrid for predicting the residual stress in butt-welding to meet the demands of advanced manufacturing planning by using neural network, genetic algorithm. In [126] an integrated approach comprising the combination of the Taguchi method and neural networks for the optimization of the process conditions for GTA welding was presented. Taguchi method was used for design of experiments and initial optimization with ANOVA for the significance of parameters of GTA welding (Electrode size, Electrode angle, Arc length, Welding current, Travel speed, and Flow rate). The author concluded that the Levenberg–Marquardt back propagation (LMBP) algorithm neural networks represent an easy and quick method to explore a non-linear multivariate relationship between parameters and responses. By using this technique with Taguchi method only, the average depth-towidth ratio in weld bead of GTA welding process for SS304 improved about 11.96% from the initial optimal parameters to the real optimal parameters. In [127] the authors presented a novel attempt to carry out the forward (the outputs as the functions of input variables) and reverse (the inputs as the functions of output variables) modeling of the metal inert gas welding (a multi-input and multi-output) process using fuzzy logic-based approaches. The statistical regression analysis was used for the forward modeling efficiently. The developed soft computing-based approaches were found to solve the above problem efficiently. In [128] a prototype knowledge based expert system named WELDES was presented. WELDES was developed to identify the aluminum welding defects, to correlate them with the welding parameters, which cause them, and to offer advice regarding the necessary corrective actions for a ship industry. The first step in this expert system was the systematic classification of the aluminum welding defects and the analysis of the reasons, which cause them. The WELDES system consists of two modules: the Diagnostic Module, and the 37

University of Engineering & Technology, Taxila-Pakistan

Adviser Module. The structure of this expert system is ideally suited to the knowledge domain of the welding. Today, the potential of expert systems is now well documented if not unanimously accepted in manufacturing domain. However, the application of expert systems particularly regarding the welding domain, there is comparatively little evidence of practical and regular utilization in the industry. Not much work has been done related to application of AI to the field of welding to predict the response and their optimization. However, some work related to machining process has been done and would be helpful to understand the AI and expert system application to manufacturing process which can be used in the welding domain also. A few papers can be found in manufacturing that describe the utilization of expert system in online diagnosis of machining conditions, selection of appropriate tools and machining parameters. A fuzzy knowledge-base for the diagnosis of turning process states was utilized by Fang [129]. The knowledge-intensive fuzzy feature-state relationships developed with support of expert system were utilized in the diagnosis process. In [130] a fuzzy expert system was developed that contained four modules: a database, a cutter-selection module, a cutting condition design module, and a learning module. The use of fuzzy nonlinear programming for the optimization of process parameters was made in the cutting condition design module. In [131] authors developed an expert system applicable to turning process by utilizing fuzzy logic as the reasoning mechanism to determine the best cutting speeds for given combinations of work piece material hardness and depth of cut. In [132] an expert system presented for the selection and optimization of tool-holder, insert, and cutting conditions of the turning process. Optimization model has utilized for the selection of machining conditions in this system. In [133] an adaptive-network based fuzzy inference system (ANFIS) was utilized to predict the work piece surface roughness by using input parameters in end milling process. The use of triangular membership functions was reported as better than trapezoidal ones in accurately predicting the surface roughness. 2.5 Chapter Summary and Conclusions Rather than concentrating on highly resource intensive welding experiments as in tradition, welding engineers around the globe feels comfortable with modeling & simulation tools to predict the response of welding and to analyze and optimize the welding parameters by systematic design of experiments (DOE), ANOVA and numerical optimization and subsequently their adverse consequences on in-service structural integrity of welded structures. With enhanced computational resources and substantial development in numerical simulation tools, the use of Finite Element techniques in Computational Weld Mechanics (CVM) focusing on structural response of welded structures has been progressively increasing to its maturity level and is being successfully employed in to-day industrial problems. In the preceding discussion related to CVM (in this chapter), efforts are made to encompass the evolution process of this technique with reference to circumferential welding of thin-walled structures specifically. A lot of work pertaining to numerical simulation of 38

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

circumferential welding relates to simplified two-dimensional analysis to reduce computation time by employing lateral and rotational symmetry assumptions due to the availability of limited computational power for a reasonable judgment of residual stress profile in the weldments by elimination the effect of root pass and tack welds etc. Such inherent deficiency of models is their incapability to predict change in deformation and residual stresses due to welding in the circumferential direction. The rapid development in the field of personal computers computational power with necessary data storage capacity attracted researchers to model welding phenomena more realistically by using 3-D FE models with needs of further improvement of commercially available finite element codes by adoption and incorporation of special techniques, such as adaptive mesh management, solid shell elements and better hand-shaking parallel computing algorithms. Hole-drilling strain-gage method is the most widely used technique for experimental determination of residual stress fields in engineering structures because of its established methodology, low equipment cost, versatility and reasonable accuracy. Center hole-drilling technique is selected for experimental verification of stresses by keeping in view its affordable cost, suitability to our specific application and availability of standard procedure for stress measurement. 2.5.1 Limitations of Previously Developed AI Tools Contents of the section 2.4 suggest that application of expert system has remained very limited to welding process and confined mostly to the other manufacturing process like machining process whereas recently, the researchers have started applying this tool for optimization of milling or HSM process. Most of the previously developed AI tools seem to be limited in effectiveness because of following three reasons: 1. The application area is not broad, in the sense that most of the tools do not cover all the influential aspects of the manufacturing process. It can be observed that the recommendation of any controllable process parameter has been provided based upon relationship between two or three given input parameters. In pragmatic conditions there are many more influential parameters that need to be cared for in recommending optimal values of any controllable parameter for desired response. 2. They provide single-purpose consultation. Either they provide recommendations for minimizing or for maximizing the response. They mostly consider one objective at a time for optimization and some expert systems provide just the prediction of some performance measures based upon limited number of input parameters. 3. They lack dynamic characteristics. Most of the expert systems presented are static, in the sense that they lack automated mechanism for expanding their knowledge or increasing the application range with experience. The above mentioned limitations are serious obstacles towards successful utilization of expert system technology on an industrial scale. Against the anticipation, the expert systems have been unable to find their way for full fledge application in most of manufacturing 39

University of Engineering & Technology, Taxila-Pakistan

industries, and metal welding industry is one of them. Mostly the expert systems find their usage only in research institutes, and the reason is the above-mentioned limitations, especially the last one. If an expert system can’t respond, positively and briskly, to the fast changing requirements, targets, and methods then it is likely to be eliminated in a highly dynamic industrial environment. In this research work an expert system for optimizing GTAW or TIG welding process of thin walled structures of high strength low alloy steel based on thin plate welding experimentation and optimization and virtual experiments performed by developing 3D FE model of circumferential welding with experimental validation as per objectives has been presented, that has also addressed all the above limitations or the current challenges or research gap in this area in an effective manner. The details will be provided in the respective chapters.

40

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

CHAPTER 3 ANALYZING & OPTIMIZING TIG WELDING PROCESS PARAMETERS 3.1 Introduction This chapter presents the details of experiments conducted, and analyses performed upon the experimental data of TIG welding of thin high strength low alloy (HSLA) steel sheet, for the purpose of analyzing and optimizing the welding parameters, their practical range and effect. Initially, the experiments performed can be regarded as the preliminary set of experiments, without following a standard design of experiments (DOE) technique. These were performed to identify the potential significant parameters and to design the upcoming sets of experiments based upon the results obtained. Though the emphasis would be upon welding parameters but the effect of thin sheet metal thickness will also be tested. First the parametric range effect will be observed by weld bead and penetration quality of welding samples. Further, this chapter provides in-depth study of effects of welding process parameters upon weld strength, distortion and residual stresses following the standard design of experiments (DOE) by using full factorial (2-level) method and response surface method (RSM). In addition to performing ANOVA of experimental results, the empirical models, for quantifying the effects of welding parameters, numerical optimization and response desirability will also be presented along with the detail of experimental setup used for welding of thin plates of different thicknesses (3, 4 and 5 mm). Mostly, the sheets of thickness between 3 to 5 mm of high strength low alloy steels are used for aerospace applications at high pressures i.e. 100 bars or above. 3.2 Design of Experiments Every experiment has to plan to obtain enough and relevant data so that the science behind the observed phenomenon can be understand. It can be possible by trial-and-error approach or design of experiments. In trial-and-error approach, a series of each experiment performed which gives some understanding. This technique requires measurements after every experiment so that analysis of observed data may be used to decide what to do next. Many times such series of experiments give negative results and cause a discouraging situation for further experimentation with any change. Therefore, such experimentation usually ends well before the number of experiments reach to draw any significant conclusions and solution of the main problem of understanding the phenomenon. Whereas the design of experiments (DOE) is a well planned set of experiments and is a much better approach to obtain systematic data in which all parameters of interest are varied over a specified range to give desired results. The advanced Design of Experiments (DOE) capabilities help to improve the process. The factors can be screened to determine which are important for explaining process variation and to understand how the factors 41

University of Engineering & Technology, Taxila-Pakistan

interact and drive the process to find out the factor settings that produce optimal process performance. The number of runs depends on how many factors and their levels [124,126]. 3.2.1 Predictor Variables Predictor variables are the welding process parameters, whose effects upon the performance measures are tested in the experiments to be conducted. Predictor variables can also be represented as process input parameters or input variables or control factors as shown in the Figure 3.1 describing various factors with multi-inputs and multi-outputs. In every process, there are some inputs which may be converted into desired outputs or response by means of some process through control factors.

Fig. 3.1 Process showing the factors with multi-inputs and multi-outputs By the brainstorming, all the factors are listed that could possibly influence any of the responses to measure and categories according to their influence upon the response. Range and control of each factor that may cause amount of change in response are determined to study. Following is the list of possible TIG welding predictor variables with possible practical parameter values that would be under scrutiny in the actual welding experiments to be performed on HSLA sheets of different thicknesses: 1. Welding Current (Amp) (170-285) 2. Welding Voltage (Volts) (10-15) 3. Welding Speed (cm/min) (15-20) 4. Sheet Thickness (mm) (3, 4, 5) 5. Wire Speed (cm/min) (110-130) 6. Gas Flow Rate (L/min) (20-25) 7. Trailing Flow Rate (L/min) (20-25) 42

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 3.1 displays the initial settings of predictor variables tested in the initial welding experiments to have a practical range of these variables for detailed experiments. Table 3.1 Initial settings of predictor variables Test No. 1 2 3 4 5 6 7 8 9 10 11 12

Welding Current (Amp) 170 170 285 285 170 170 285 285 170 170 285 285

Welding Voltage (V) 10 15 10 15 10 15 10 15 10 15 10 15

Welding Speed (cm/min) 15 20 15 20 15 20 15 20 15 20 15 20

Sheet Thickness (mm) 3 3 3 3 4 4 4 4 5 5 5 5

Wire Speed (cm/min) 110 130 110 130 110 130 110 130 110 130 110 130

Trailing Flow Rate (l/min) 20 25 20 25 20 25 20 25 20 25 20 25

From Table 3.1, four experiments were conducted for each thickness (i.e. 3 mm, 4 mm and 5 mm) with high and low settings of other variables to analyze the effect. The weld quality in respect of weld bead as shown in Figure 3.2 and penetration of test no. 1, 6 and 11 was good and the quality of test no. 2 and 12 was satisfactory but the quality of test no. 3, 4, 5, 7, 8, 9, 10 was not satisfactory. There was no penetration in test no. 9 and 10 due to high thickness and very low current and low values of other parameters whereas the test no. 3 and 4 were burnt at weld line due to less thickness and high current and other parameters. The penetration in test no. 5 was poor and in test no. 7 & 8 was excessive along with the sink of bead. The weld bead quality achieved of twelve tests has presented in Figure 3.2 by given the quality ranking from 1-5 for low to high quality of weld bead which was analyzed by visual inspection. The measurements of weld bead width (10-14 mm), height (1.5-2 mm) and penetration (1-1.5 mm) was observed for a good weld for the thickness of HSLA steel sheet from 3-5 mm. E ffe c ts o f P a r a m e te r s o n W e ld B e a d Q ua lity 5

5

5

4

Weld Bead Quality

4

3

3

3

2

2

2

2

2

1

1

1

8

9

1

0 1

2

3

4

5

6 7 Te s t No .

10

11

12

Fig. 3.2 Weld quality upon initial setting of predictors 43

University of Engineering & Technology, Taxila-Pakistan

The most significant parameter is current with high range of values as compared to others parameters. It shows the practical range of these parameters specifically of welding current should be specific with respect to thickness of material and the heat input (welding current, welding voltage and welding speed) required for the fusion and weld of selected thickness. With the increase in material thickness, the increase in heat input is required accordingly. The parameters and their levels were required to be reviewed. The revised parameters and their practical high and low settings were revised to conduct the experiments as given in detail in the following sub-section. 3.2.1.1 Factorial Design of Experiments The factorial designs allow for the simultaneous study of the effects that several factors may have on a process. During an experiment, varying the levels of the different factors simultaneously rather than one at a time is efficient to save the resources in terms of time and cost, and also allows for the study of interactions between the different factors. Without the use of factorial experiments, important interactions may remain undetected because the interactions are the driving force in many processes. The number of influential input variables is large in many process development and manufacturing applications. The screening process is used to reduce the number of input variables by identifying the key input variables or process conditions that affect product quality and to focus process improvement efforts on the few really important variables and also to suggest the "best" or optimal settings for these factors. In industry, two-level full and fractional factorial designs are often used for screening the actually important factors that influence process output measures or product quality. The responses are measured at all combinations of the experimental factor levels in a full factorial experiment. The factor levels combinations represent the conditions at which responses will be measured. Each experimental condition is called as run and the response measurement as an observation where as the entire set of run is called as design. Each experimental factor in a two-level full factorial design has only two levels. The all combinations of these factor levels are included in the experimental runs. The two-level factorial provide useful information for relatively few runs per factor and is used to indicate the major trends to provide direction for further experimentation and optimal settings. A 24 (4 factors, 2 levels, 16 test) full factorial design model (replicates 1, block 1, centre point per block 0 and order 4FI) was used for the welding experiments. Table 3.2, Table 3.3 and Table 3.4 shows the low and high settings (or levels) for the predictor variables (or parameters) used in sixteen tests for the sheet thickness of 3, 4 and 5 mm respectively. Three of these predictor variables (welding current, welding voltage and welding speed) are numeric while the other one gas trailing is categorical. Complete detail of 16 experiments following full factorial has been presented in Table 3.5, Table 3.6 and Table 3.7 for sheet thickness of 3, 4 and 5 mm respectively. All the statistical analyses were performed using a commercial computing package named Design-Expert® 7.1.6, by Stat-Ease® and statistical software MINITAB® Release 14 throughout this manuscript.

44

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 3.2 High and Low Settings of Factors (t = 3 mm) Factor Name

Units

Type

Low Actual

High Actual

A B C D

A V cm/min

Numeric Numeric Numeric Categoric

170.00 10.50 15.00 nil

210.00 13.50 18.00 Ar

Current Voltage Weld Speed Trailing

Table 3.3 High and Low Settings of Factors (t = 4 mm) Factor Name

Units

Type

Low Actual

High Actual

A B C D

A V cm/min

Numeric Numeric Numeric Categoric

200.00 10.50 15.00 nil

220.00 13.50 18.00 Ar

Current Voltage Weld Speed Trailing

Table 3.4 High and Low Settings of Factors (t = 5 mm) Factor Name

Units

Type

Low Actual

High Actual

A B C D

A V cm/min

Numeric Numeric Numeric Categoric

230.00 10.50 15.00 nil

270.00 13.50 18.00 Ar

Current Voltage Weld Speed Trailing

Table 3.5 Design of 16 Experiments following Full Factorial (t = 3 mm) Std

Run

12 5 1 3 11 7 8 16 13 4 9 10 2 6 15 14

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Factor 1 A:Current A 210.00 170.00 170.00 170.00 170.00 170.00 210.00 210.00 170.00 210.00 170.00 210.00 210.00 210.00 170.00 210.00

Factor 2 B:Voltage V 13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50

Factor 3 C:Weld Speed cm/min 15.00 18.00 15.00 15.00 15.00 18.00 18.00 18.00 18.00 15.00 15.00 15.00 15.00 18.00 18.00 18.00

Factor 4 D: Trailing Ar nil nil nil Ar nil nil Ar Ar nil Ar Ar nil nil Ar Ar

45

University of Engineering & Technology, Taxila-Pakistan

Table 3.6 Design of 16 Experiments following Full Factorial (t = 4 mm) Std

Run

12 5 1 3 11 7 8 16 13 4 9 10 2 6 15 14

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Factor 1 A:Current A 220.00 200.00 200.00 200.00 200.00 200.00 220.00 220.00 200.00 220.00 200.00 220.00 220.00 220.00 200.00 220.00

Factor 2 B:Voltage V 13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50

Factor 3 C:Weld Speed cm/min 15.00 18.00 15.00 15.00 15.00 18.00 18.00 18.00 18.00 15.00 15.00 15.00 15.00 18.00 18.00 18.00

Factor 4 D: Trailing Ar nil nil nil Ar nil nil Ar Ar nil Ar Ar nil nil Ar Ar

Table 3.7 Design of 16 Experiments following Full Factorial (t = 5 mm) Std

Run

12 5 1 3 11 7 8 16 13 4 9 10 2 6 15 14

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Factor 1 A:Current A 270.00 230.00 230.00 230.00 230.00 230.00 270.00 270.00 230.00 270.00 230.00 270.00 270.00 270.00 230.00 270.00

Factor 2 B:Voltage V 13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50

Factor 3 C:Weld Speed cm/min 15.00 18.00 15.00 15.00 15.00 18.00 18.00 18.00 18.00 15.00 15.00 15.00 15.00 18.00 18.00 18.00

Factor 4 D: Trailing Ar nil nil nil Ar nil nil Ar Ar nil Ar Ar nil nil Ar Ar

46

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

3.2.2 Response Variables Response variables are the performance measures, which can also be termed as output variables or output parameters. Following response variables will be measured in order to judge the process performance of thin walled HSLA steel welded structures which are mostly required for pressure vessels welds after the visual inspection and radiographic testing: 1. Weld Strength: Max value of tensile strength - to be measured in MPa by testing of weld tensile sample. 2. Distortion: Max value of weld-induced distortion in the sheet in weld zone – to be measured in mm. 3. Residual Stress: Max value of weld-induced stresses (von-mises) in the weld zone – to be measured in MPa. 3.2.3 Fixed Parameters The welding position used was flat and single V joint geometry including angle of 70 deg was used with 1 mm root face and 1 mm root gap. The electrical characteristics used were DC current and straight polarity. Argon gas (99.999% Liquid) was used for shielding (25 lit/min) and for trailing (25 lit/min). The size of shielding nozzle is Ø 18 mm. The sizes used for trailing are: diameter of trailing nozzle = Ø 1.3 mm, distance from nozzle to sample = 5 mm, distance (centre to centre) between arc and trailing nozzle = 30 mm, effective diameter of trailing = Ø 25 mm. The material of backing fixture used was Cu and alcohol (99%) was used for joint cleaning after mechanical cleaning of both sides (50 mm) of weld joint. Welding conditions used were humidity less than 70%, ambient temperature greater than 18 ºC and no draught in welding area. The material of sheet used as base metal was HSLA steel and sheet sizes used for welding samples were 3 x 130 x 500 mm, 4 x 130 x 500 and 5 x 130 x 500 mm each respectively (i.e. the weld width = 260 mm and weld length = 500 mm) as shown in Figure 3.3. The other welding parameters that were kept constant in all experiments, are: pre-heat temperature = 175 ºC, inter-pass temperature = 150 ºC, tungsten electrode (3% thoriated) size = Ø 3.2, and welding wire (H08) size = Ø 1.6 mm.

Fig. 3.3 Schematic diagram of the welding test plate. 47

University of Engin Engineering & Technology, Taxila-Pakistan

3.3 Experimental Setup The experimental rimental welding equipment for TIG welding, in the present research consists of four major units as follows: •

A clamping unit that supports and clamps the work work-piece.

•

The traverse system that allows a precise and controlled movement of the welding torch unit.

•

The welding power source.

•

A control unit that controls the welding power source and traverse system along with measures the welding parameters.

All the TIG welding experiments, described in this manuscript, have been performed on SAF TIGMATE 270 AC/DC power source, SAF NERTAMATIC 300 TR and fully automatic torch control and movement system is utilized in present research with minimum manual interference. TIGMATE 270 and NERTAMATIC 300 TR welding power sources as shown in Figure 3.4 and Figure 3.5 is a computerized omputerized waveform control technology for high quality TIG welds. The parameters can be controlled as desired. Automatic torch positioning system is used to control / locate the torch movement. Tack welded sheets are properly clamped (as per desired stru structural ctural boundary conditions) with torch aiming at 90o. Refer Figure 3.6 for sample clamping arrangement, Figure 3. 3.7 for details of automatic TIG torch and clamping arrangements. Weld samples before and after welding are shown in Figure 3.8 3. and Figure 3.9 respectively.

Fig. 3.4 SAF TIGMATE 270 Power Source Fig. 3.5 NERTAMATIC 300 TR Power Source

48

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.6 Sample Clamping Arrangement

Fig. 3.8 Sample before Welding

Fig. 3.7 Details of automatic TIG torch and clamping arrangements.

Fig. 3.9 Sample after Welding

Distortion was measured using digital dial indicator (0.001 mm) by placing distorted welded sample on surface plate as shown in Figure 3.10 and Figure 3.11. The measurements were taken along the weld line at 15 mm from centre of weld line with incremental of 50 mm at ten points from end to end of 500 mm plate in length and all the plates are marked according to this division. The maximum distortion was observed at centre of plate (at about 250 mm from end) and minimum at the ends of plate along the weld in all experiments. Tensile test samples were prepared by water jet cutting and then machining from the welded plate of 500 mm in length of thicknesses (3, 4 and 5 mm) according to sizes as given in Figure 3.12. Testing of tensile weld samples (as welded) as shown in Figure 3.12 and Figure 3.13 was performed using Universal Testing Machine EDC (250 KN / 25 Ton), UNLINGEN BC-TestTec GmbH, with MF-A axial extensometer with pretension of 50 N and gripping 49

University of Engineering & Technology, Taxila-Pakistan

length of 110 mm. Test speed was fixed at 5 mm/min and test standard used for welded samples was ASME-IX as shown in Figure 3.14 and Figure 3.15.

Fig. 3.10 The Distortion Measurement setup

Fig. 3.11 Distorted Sample

Fig. 3.12 Tensile Sample

50

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.12 Tensile Samples

Fig. 3.14 Tensile Sample (Machined)

Fig. 3.13 Sample after Cutting Tensile Samples

Fig. 3.15 Tensile Sample after Testing

The hole-drilling method is used for the measurement of residual stresses and the equipment for hole-drilling strain gage along with P3500 strain meter from Vishay Group as shown in Figure 3.16 is used for experimental determination of residual stress fields as shown in Figure 3.17. The milling guide RS-200 can be easily mounted on flat specimens for flat bottomed holes, followed by hole-diameter measurement with microscopic tube. Complete experimental setup for residual stress measurement is shown in Figure 3.19. There are six basic steps involved for measurement of residual stresses by hole-drilling method as

51

University of Engineering & Technology, Taxila-Pakistan

follows [134] and this procedure is relatively a simple procedure standardized by ASTM Standard Test Method ASTM-E837 [135]. •

On the test part at the point where residual stresses are to be determined, a special three (or six) element strain gage rosette is installed. Two different types of strain gage rosette (EA-XX-062RE-120 and CEA-XX-062UM-120) as shown in Figure 3.18 are used with features of 120 ± 0.2% and 120 ± 0.4% in ohms respectively, as given by [136].

•

P-3500 strain indicator (a switch and balance unit) as shown in Figure 3.16 (right) is used for connection of gage grids to a static strain indicator. The accuracy of reading of P-3500 is ± 0.5 % and the resolution sensitivity is ± 1 µε.

•

A precision milling guide Model RS-200, as shown in Figure 3.16 (left) from Vishay Measurement Group is used for accurate placement and centering of a precision milling guide on the drilling target. The range of error of RS-200 is ± 20 MPa.

•

The zero balancing of the gage circuits followed by the drilling of small, shallow hole through the geometric center of the rosette.

•

The relaxed strains measurement, corresponding to the initial residual stress values.

•

The principal residual stresses and their angular orientation are calculated from the measured strains by using special data-reduction relationships.

Due to complexity of the blind-hole geometry and non availability of closed-form solution from the theory of elasticity for direct calculations of the residual stresses from the measured strains, except by the introduction of empirical coefficients [136], finite element techniques are usually used for the calculation of these coefficients. Numerically and experimentally determined calibration coefficients are readily available in the published literature [134-136]. Commercially available data reduction software, H-Drill developed by Schajer [136] is used in the research work.

Fig. 3.16 Hole-drilling equipment & P-3500 strain indicator from Vishay Group

52

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.17 Welded specimen with mounted strain gage rosette

Fig. 3.18 Two types of strain gage rosette

Fig. 3.1 3.19 Residual Stresses Measurement setup 3.3.1 .1 Characteristics of Base Metal and Filler Wire High Strength Low Alloy (HSLA) steels are produced to provide better mechanical properties, and/or greater resistance to atmospher atmospheric ic corrosion. Important applications of the HSLA steels include parts of aircraft and rockets, missiles, and hot forging dies. The HSLA steels have low carbon contents in order to produce adequate formability and the carbon contents of these steels lies bbetween 0.3-0.5% 0.5% and the main alloying elements are Cr, Ni, Mo and V. The hardness of this steel is enhanced by means of heat treatment (quenching) and tempering. When quenched and tempered these steels may attain strength of upto 1700 MPa. However, these steels teels are sensitive to cold cracking due to carbon and alloy contents. Thin sections required preheating before welding and stress relieving after welding to avoid cracking. Table 3.8 shows the chemical composition of base metal (HSLA steel) analyzed by spectrometer on shop floor [137, 138, and 139]. 53

University of Engineering & Technology, Taxila-Pakistan

Table 3.8 Chemical composition of Base Metal (HSLA Steel)(30CrMnSiA)

Content (%)

C

Cr

Si

Mn

V

Mo

Ni

P

S

0.28-0.32

1.0-1.3 1.5-1.7 0.7-1.0 0.08-0.15 0.4-0.55 0.25 ≤0.01

≤ 0.013

Tensile strength is one of the most predominant factor in the application of HSLA steels. Consequently, the filler metal is selected most frequently on the basis of providing a weld metal with minimum yield and tensile strength values that are equal to those of base metal with consideration of alloying contents and corrosion resistance. The low temperature impact properties, as well as the tensile strength of the weld deposit, can be an important factor when selecting the filler metals for the welding of HSLA steels [140]. Table 3.9 shows the chemical composition of filler wire (H08) obtained by wet analysis method [138]. Table 3.9 Chemical composition of Filler Wire (H08) C

Cr

Si

Mn

V

Mo

Ni

P

S

Al

Content 0.08-0.12 1.4-1.7 1.1-1.3 0.9-1.1 0.05-0.15 0.4-0.6 1.8-2 ≤ 0.006 ≤0.005 ≤0.10 (%) When different low alloy steels welded under same welding conditions, showed varying weld induced imperfections and weld strength. This implies that material thickness alone is not an adequate parameter to evaluate weld strength. Chemical composition of base metal and filler wire also play vital role. Table 3.10 presents the mechanical properties of base metal (HSLA steel) [139]. After heat treatment ( i.e. quenching and tempering ), these mechanical properties of base metal reaches to ≤ 1600 MPa (Tensile Strength), ≤ 1300 MPa (Yield Strength), ≤ 8% (Elongation) and ≤ 48 HRC (Hardness). However, the distortion in thin walled structure increases due to heat treatment process. Generally, these metals are stress relieved by heat treatment after welding and practically the weld induced stresses are relieved by this method in ideal condition upto 70-80 %. Therefore, it is required that the weld induced imperfections (i.e. distortion and residual stresses) should be controlled to minimum during welding of thin walled structures of HSLA steel. Table 3.10 Mechanical properties of Base Metal (HSLA Steel) Tensile Strength (MPa) 700-800

Yield Strength (MPa) 500-600

Elongation (%) 20

Hardness (HRC) 20

3.4 Experimental Results, ANOVA, Regression and Optimization In the following sub-sections, discussion related to effects of predictor variables upon the performance measures is provided with description of ANOVA, regression, and optimization applied to the experimental results.

54

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

The analysis of variance (ANOVA) is used to investigate and to model the relationship between a response variable and one or more independent variables. The analysis of variance (ANOVA) was developed by Sir Ronald Fisher in the 1930s as a way to interpret the results of agricultural experiments. ANOVA is not a complicated method and has a large amount of mathematical uniqueness associated with it. The purpose of the ANOVA is to investigate welding process parameters, which can significantly affect the quality characteristics. The percentage contribution in the total sum of the squared deviations can be used to evaluate the importance of the welding process parameter change on these quality characteristics. In addition, the F-test, named after Fisher, can be used to determine which welding process parameters have a significant effect on the quality characteristics. Usually, when the F-test value is greater than 4, it means that a change in the process parameter has a significant effect on the quality characteristics [126]. ANOVA is used to determine the model with the view of the following points: - Model F-Value and associated probability value to confirm model significance. - Predicted R-Squared and Adjusted R-Squared in reasonable agreement. - Adequate Precision (signal to noise ratio) to use model to navigate design space. - Tests on individual terms to confirm they are significant. Regression analysis is also used to investigate and model the relationship between a response variable and one or more predictors. The least squares (for continuous variable), partial least squares (for highly correlated predictors), and logistic (for categorical variable) regression methods are used to estimate the parameters. Numerical optimization is used to optimize any combination of one or more goals. The goals may apply to either factors or responses. The optimization module searches for a combination of factor levels that simultaneously satisfy the requirements placed on each of the responses and factors after performing ANOVA by choosing the desired goal (maximize, minimize, target, within range, and none for responses only) for each factor and response. Optimization of one response or the simultaneous optimization of multiple responses can be performed by numerically, graphically and using the point prediction node. The desirability is an objective function that ranges from zero outside of the limits to one at the goal and the numerical optimization finds a point that maximizes the desirability function. 3.4.1 Weld Strength Figure 3.20, Figure 3.21 and Figure 3.22 show the comparison of weld strength for aforementioned sixteen tests in Table 3.5, Table 3.6 and Table 3.7 respectively. The maximum and minimum values of weld strength obtained with respect to thickness of material are presented in Table 3.11. Table 3.11 Max. and Min. Values of Weld Strength Response Thickness

Response

Units Minimum

Maximum

Mean

3 mm 4 mm 5 mm

Tensile Strength Tensile Strength Tensile Strength

MPa MPa MPa

791 780.4 765.7

754.294 16.9567 743.994 16.3382 735.506 15.7428

730.6 722.3 715

Std. Dev.

55

University of Engineering & Technology, Taxila-Pakistan

R e s p o n s e o f E x p e r im e n ts C o n d u c te d ( t = 3 m m ) 791

7 8 4 .6

800 7 4 1 .7

751

740

3

4

7 5 6 .8

7 4 9 .5

7 7 2 .7

7 5 9 .8 7 3 7 .8

750

7 3 5 .5

7 3 0 .6

7 4 9 .5

7 5 9 .7

7 5 8 .5

14

15

16

Weld Strength (MPa)

700 600 500 400 300 200 100 0 1

2

5

6

7 8 9 10 E x p e r im e n t N o .

11

12

13

Fig. 3.20 Weld Strength of Sixteen Experiments ( t = 3 mm) R e s p o n s e o f E x p e r im e n ts C o n d u c te d ( t = 4 m m ) 800

78 0.4

770.1 74 8.5

73 7.2

730.3

74 2.5

740.1

4

5

6

7 6 0 .1

750.3 72 5.2

722.3

7 36.4

7 2 4 .3

12

13

7 41.2

7 5 5 .4

14

15

16

7 2 9 .5

7 4 2 .8

7 3 0 .5

14

15

16

7 39.6

Weld Strength (MPa)

700 600 500 400 300 200 100 0 1

2

3

7 8 9 10 E x p e r im e n t N o .

11

Fig. 3.21 Weld Strength of Sixteen Experiments ( t = 4 mm) R e s p o n s e o f E x p e r im e n ts C o n d u c te d ( t = 5 m m ) 800

7 5 9 .6 7 2 6 .7

7 3 7 .8

725

7 3 5 .7

7 2 9 .5

4

5

6

7 4 9 .8 7 1 7 .8

7 6 5 .7

7 5 7 .7 730

715

715

Weld Strength (MPa)

700 600 500 400 300 200 100 0 1

2

3

7 8 9 10 E x p e r im e n t N o .

11

12

13

Fig. 3.22 Weld Strength of Sixteen Experiments ( t = 5 mm) 56

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

The combination of welding current, welding voltage, welding speed and gas trailing gave the high weld strength from 765 MPa to 791 MPa with mean value (744 MPa) from 735 MPa to 754 MPa and standard deviation (16.3) of 15.7-16.9 when it was applied to the welding of sheet thickness of 3, 4 and 5 mm. The results of experiment no. 9 and 10 according to design of experiments applying full factorial as given in Table 3.5, Table 3.6 and Table 3.7 shows the high and low values of weld strength at low level of welding current and voltage, high value of welding speed with application of trailing and at high level of welding current and voltage, low level of welding speed without application of trailing for sheet thickness 3, 4 and 5 mm respectively. The weld strength (tensile strength) data were analyzed using ANOVA technique and observations are presented in Table 3.12, Table 3.13 and Table 3.14 for the sheet thickness of 3, 4 and 5 mm respectively. The analysis shows that effects of four parameters (welding current, voltage and welding speed with application of trailing) are significant upon weld strength for sheet thicknesses of 3, 4 and 5 mm. Figure 3.23, Figure 3.24 and Figure 3.25 shows the effects of changing the levels of each parameter upon weld strength while keeping other three parameters fixed for sheet thickness of 3, 4 and 5 mm respectively. It is clear that effects of welding current, voltage and speed are significant. The weld strength decreases with increase of welding current and voltage values and increases with increase of welding speed along with the application of trailing as heat sink. Table 3.12 ANOVA for Tensile Strength (t=3mm) factorial model Source

Sum sqrs

Model 3540.58 A-Current 1258.48 B-Voltage 716.90 C-Weld Speed785.40 D- Trailing 779.81 Residual 772.35 Cor Total 4312.93

DoF

Mean square F-value

4 1 1 1 1 11 15

885.15 1258.48 716.90 785.40 779.81 70.21

12.61 17.92 10.21 11.19 11.11

Prob>F 0.0004 0.0014 0.0085 0.0065 0.0067

Significance Significant Significant Significant Significant Significant

Table 3.13 ANOVA for Tensile Strength (t=4mm) factorial model Source

Sum sqrs

Model 3272.41 A-Current 1423.18 B-Voltage 591.71 C-Weld Speed 633.78 D- Trailing 623.75 Residual 731.64 Cor Total 4004.05

DoF

Mean square F-value

4 1 1 1 1 11 15

818.10 1423.18 591.71 633.78 623.75 66.51

12.30 21.40 8.90 9.53 9.38

Prob>F 0.0005 0.0007 0.0125 0.0103 0.0108

Significance Significant Significant Significant Significant Significant

57

University of Engineering & Technology, Taxila-Pakistan

Table 3.14 ANOVA for Tensile Strength (t=5mm) factorial model Source

Sum sqrs

Model 2827.49 A-Current 1216.27 B-Voltage 435.77 C-Weld Speed 423.33 D- Trailing 752.13 Residual 890.04 Cor Total 3717.53

DoF

Mean square F-value

4 1 1 1 1 11 15

706.87 1216.27 435.77 423.33 752.13 80.91

8.74 15.03 5.39 5.23 9.30

Prob>F 0.0020 0.0026 0.0405 0.0430 0.0111

Significance Significant Significant Significant Significant Significant

Fig. 3.23 Effects of welding parameters upon Tensile Strength (t = 3 mm) 58

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.24 Effects of welding parameters upon Tensile Strength (t = 4 mm)

Fig. 3.25 Effects of welding parameters upon Tensile Strength (t = 5 mm) 59

University of Engineering & Technology, Taxila-Pakistan

Table 3.11 shows the comparison of weld strength values obtained for sixteen experiments. The response values for weld strength (3, 4 and 5 mm thickness) range from 730.6 to 791 MPa, 722.3 to 780.4 MPa and 715 to 765.7 MPa, providing the ratio of maximum to minimum equal to 1.08267, 1.08044 and 1.0709 respectively. The ratio is small and, thus, there is no need to apply any kind of transformation to the data. Table 3.12, Table 3.13 and Table 3.14 presents the ANOVA details for the suggested factorial model. Table 3.12, the Model F-value of 12.61 implies the model is significant. There is only a 0.04% chance that a "Model F-Value" this large could occur due to noise. Values of "Prob > F" less than 0.0500 indicate model terms are significant. In this case A, B, C and D are significant model terms. The model possesses the R2 of 82.09%, R2-adjusted of 75.58%, and R2-predicted of 62.11%. The "Pred R-Squared" of 62.11% is in reasonable agreement with the "Adj R-Squared" of 75.58%. "Adeq Precision" measures the signal to noise ratio. A ratio greater than 4 is desirable. In this model, ratio of 12.617 indicates an adequate signal. This model can be used to navigate the design space. Table 3.13, the Model F-value of 12.30 implies the model is significant. There is only a 0.05% chance that a "Model F-Value" this large could occur due to noise. Values of "Prob > F" less than 0.0500 indicate model terms are significant. In this case A, B, C and D are significant model terms. The model possesses the R2 of 81.73%, R2-adjusted of 75.08%, and R2-predicted of 61.34%. The "Pred R-Squared" of 61.34% is in reasonable agreement with the "Adj R-Squared" of 75.08%. "Adeq Precision" measures the signal to noise ratio. A ratio greater than 4 is desirable. In this model, ratio of 12.305 indicates an adequate signal. This model can be used to navigate the design space. In Table 3.14, the Model F-value of 8.74 implies the model is significant. There is only a 0.20% chance that a "Model F-Value" this large could occur due to noise. Values of "Prob > F" less than 0.0500 indicate model terms are significant. In this case A, B, C and D are significant model terms. The model possesses the R2 of 76.06%, R2-adjusted of 67.35%, and R2-predicted of 49.35%. The "Pred R-Squared" of 49.35% is in reasonable agreement with the "Adj R-Squared" of 67.35%. "Adeq Precision" measures the signal to noise ratio. A ratio greater than 4 is desirable. In this model, ratio of 10.316 indicates an adequate signal. This model can be used to navigate the design space. After experimental and analytical work related to TIG welding of HSLA steel thin plates, the possibility of formulation of empirical models for weld strength in terms of all the significant input parameters (tested in the experiments) was carried out. The models can consist of numeric variables only. The empirical model for the weld strength (MPa), in terms of welding parameters, is as follows: Equation 3.1 (Trailing = nil) and Equation 3.2 (Trailing = Ar) in terms of actual factors for t = 3 mm for weld strength: Weld Strength = +808.04688 - 0.44344*Current - 4.46250*Voltage + 4.67083*Weld Speed

3.1

Weld Strength = +822.00938 - 0.44344*Current - 4.46250*Voltage + 4.67083*Weld Speed

3.2

60

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Equation 3.3 (Trailing = nil) and Equation 3.4 (Trailing = Ar) in terms of actual factors for t = 4 mm for weld strength: Weld Strength = +915.22500 - 0.94313* Current - 4.05417* Voltage + 4.19583* Weld Speed

3.3

Weld Strength = +927.71250 - 0.94313* Current - 4.05417* Voltage + 4.19583* Weld Speed

3.4

Equation 3.5 (Trailing = nil) and Equation 3.6 (Trailing = Ar) in terms of actual factors for t = 5 mm for weld strength: Weld Strength = +822.80313 - 0.43594* Current - 3.47917* Voltage + 3.42917*Weld Speed

3.5

Weld Strength = +836.51563 - 0.43594* Current - 3.47917* Voltage + 3.42917* Weld Speed

3.6

All the models presented in equations 3.1 to 3.6 for weld strength are valid for the following ranges of input parameters: welding current: 170 to 210 A for 3 mm, 200 to 220 A for 4 mm and 230 to 270 A for 5 mm; welding voltage 10.5 to 13.5 V and welding speed 15 to 18 cm/min. The numerical optimization applied to the weld strength data suggests that for any material thickness value lying between 3 and 5 mm, the weld strength in TIG welding of HSLA steel can be maximized if the trailing is used along with low values of heat input i.e. low values of welding current and welding voltage and high value of welding speed. The predicted weld strength values are 783 MPa, 772 MPa and 762 MPa for thickness 3 mm, 4 mm and 5 mm respectively at input parameters as: i)170 A, 10.5 V, 18 cm/min, ii) 200 A, 10.5 V, 18 cm/min and iii) 230 A, 10.5 V, 18 cm/min respectively as shown in Figure 3.26, Figure 3.27 and Figure 3.28 with response desirability of 0.80 for minimization of distortion and maximization of weld strength as shown in Figure 3.35, Figure 3.36 and Figure 3.37.

Fig. 3.26 Tensile Strength Predictions w.r.t Predictors (t = 3 mm)

61

University of Engineering & Technology, Taxila-Pakistan

Fig. 3.27 Tensile Strength Predictions w.r.t Predictors (t = 4 mm)

Fig. 3.28 Tensile Strength Predictions w.r.t Predictors (t = 5 mm) 3.4.2 Distortion Figure 3.29, Figure 3.30 and Figure 3.31 show the comparison of distortion for aforementioned sixteen tests in Table 3.5, Table 3.6 and Table 3.7 respectively. The maximum and minimum values of distortion obtained with respect to thickness of material are presented in Table 3.15. R e s p o n s e

o f E x p e r im e n ts C o n d u c te d

( t =

3

m m )

8 7 .2

7

6 .9 6 .5

6 .4

Distortion (mm)

6

5 .6

6

5 .8

5 .2

5 .7 5 .2

5

5 .7

5 .5

5 .1

4 .9

4 3 .3

3 .2

3 2 1 0 1

2

3

4

5

6

7 8 9 E x p e r im e n t

1 0 No .

1 1

1 2

1 3

1 4

1 5

1 6

Fig. 3.29 Distortion of Sixteen Experiments ( t = 3 mm)

62

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

R e s p o n s e

o f E x p e r im e n ts C o n d u c te d ( t =

4

m m )

7 6 .2

6

5 .9 5 .7 5 .4 5 .2

5 .1

5

4 .9

Distortion (mm)

5

4 .6

4 .5 4 .3

4 .1

4

3 .7 3 .4

3 .5

1 5

1 6

2 .8

3 2 1 0 1

2

3

4

5

6

7 8 9 1 0 E x p e r im e n t N o .

1 1

1 2

1 3

1 4

Fig. 3.30 Distortion of Sixteen Experiments ( t = 4 mm) R e s p o n s e

o f E x p e r im e n ts C o n d u c te d ( t =

6

5

m m )

5 .6 5 .2 5

4 .9

5

4 .7 4 .5

Distortion (mm)

4 .1

4

4

3 .9

3 .8

3 .7

3 .7

3 .6

3 .5

3 .4

3 2 .2

2

1

0 1

2

3

4

5

6

7 8 9 E x p e r im e n t

1 0 No .

1 1

1 2

1 3

1 4

1 5

1 6

Fig. 3.31 Distortion of Sixteen Experiments ( t = 5 mm) Table 3.15 Max. and Min. Values of Distortion Response Thickness

Response

Units Minimum

Maximum

Mean

Std. Dev.

t = 3 mm

Distortion

mm

3.2

7.2

5.5125

1.09293

t = 4 mm

Distortion

mm

2.8

6.2

4.64375 0.96538

t = 5 mm

Distortion

mm

2.2

5.6

4.1125

0.84053

The combination of welding current, welding voltage, welding speed and gas trailing gave the high and low value of distortion at high level of welding parameters without application of trailing and at low level of welding parameters with application of trailing respectively. The results of experiment no. 9 and 10 according to design of experiments applying full factorial as given in Table 3.5, Table 3.6 and Table 3.7 shows the low and high values of distortion at low level of welding current and voltage, high value of welding speed with application of trailing and at high level of welding current and voltage, low level of welding speed without application of trailing for sheet thickness 3, 4 and 5 mm respectively.

63

University of Engineering & Technology, Taxila-Pakistan

As the sheet thickness increases the distortion level decreases although welding current level increases with increase of material thickness. As the sheet thickness increases from 3 mm to 5 mm, the minimum distortion observed decreases from 3.2 mm to 2.2 mm and maximum distortion also decreases from 7.2 mm to 5.6 mm respectively. The mean value (4.75 mm) of distortion is 4 mm to 5.5 mm whereas the standard deviation is 1.0. The results show that with increase of sheet thickness from 30% to 65%, the welding distortion decreases by 15% to 30% in value for 3-5 mm thickness range of HSLA steel sheet in general. The distortion data were analyzed using ANOVA technique and observations are presented in Table 3.16, Table 3.17 and Table 3.18 for the sheet thickness of 3, 4 and 5 mm respectively. The analysis shows that effects of three parameters (welding current, welding voltage and welding speed) are significant upon distortion. The effect of application of trailing on distortion is marginally significant.

Table 3.16 ANOVA for Distortion (t=3mm) factorial model Source

Sum sqrs

Model 13.94 A-Current 8.41 B-Voltage 1.96 C-Weld Speed 2.25 D- Trailing 1.32 Residual 3.98 Cor Total 17.92

DoF

Mean square F-value

4 1 1 1 1 11 15

3.49 8.41 1.96 2.25 1.32 0.36

9.65 23.27 5.42 6.23 3.66

Prob>F

Significance

0.0013 0.0005 0.0399 0.0298 0.0821

Significant Significant Significant Significant

Table 3.17 ANOVA for Distortion (t=4mm) factorial model Source

Sum sqrs

Model 11.62 A-Current 2.64 B-Voltage 4.52 C-Weld Speed 3.52 D-Trailing 0.95 Residual 2.36 Cor Total 13.98

DoF

Mean square F-value

4 1 1 1 1 11 15

2.91 2.64 4.52 3.52 0.95 0.21

13.56 12.32 21.08 16.41 4.44

Prob>F 0.0003 0.0049 0.0008 0.0019 0.0589

Significance Significant Significant Significant Significant

64

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 3.18 ANOVA for Distortion (t=5mm) factorial model Source

Sum sqrs

Model 8.45 A-Current 1.82 B-Voltage 1.10 C-Weld Speed 4.62 D-Trailing 0.90 Residual 2.15 Cor Total 10.60

DoF

Mean square F-value

4 1 1 1 1 11 15

2.11 1.82 1.10 4.62 0.90 0.20

10.82 9.34 5.65 23.68 4.62

Prob>F 0.0008 0.0109 0.0367 0.0005 0.0546

Significance Significant Significant Significant Significant

Figure 3.32, Figure 3.33 and Figure 3.34 shows the effects of changing the levels of each parameter upon distortion while keeping other three parameters fixed for sheet thickness of 3, 4 and 5 mm respectively. It is clear that effects of welding current, voltage and speed are significant. The distortion increases with increase of welding current and voltage values without the application of trailing and decreases with increase of welding speed with the application of trailing. Table 3.15 shows the comparison of weld distortion values obtained for sixteen experiments. The response values for weld distortion (3, 4 and 5 mm thickness) range from 3.2 to 7.2 mm, 2.8 to 6.2 mm and 2.2 to 5.6 mm, providing the ratio of maximum to minimum equal to 2.25, 2.2143 and 2.54545 respectively. The ratio is small and, thus, there is no need to apply any kind of transformation to the data. Table 3.16, Table 3.17 and Table 3.18 presents the ANOVA details for the suggested factorial model. Table 3.16, the Model F-value of 9.65 implies the model is significant. There is only a 0.13% chance that a "Model F-Value" this large could occur due to noise. Values of "Prob > F" less than 0.0500 indicate model terms are significant. In this case A, B and C are significant model terms. The model possesses the R2 of 77.81%, R2-adjusted of 69.75%, and R2-predicted of 53.06%. The "Pred R-Squared" of 53.06% is in reasonable agreement with the "Adj R-Squared" of 69.75%. "Adeq Precision" measures the signal to noise ratio. A ratio greater than 4 is desirable. In this model, ratio of 10.341 indicates an adequate signal. This model can be used to navigate the design space. Table 3.17, the Model F-value of 13.56 implies the model is significant. There is only a 0.02% chance that a "Model F-Value" this large could occur due to noise. Values of "Prob > F" less than 0.0500 indicate model terms are significant. In this case A, B and C are significant model terms. The model possesses the R2 of 83.14%, R2-adjusted of 77.01%, and R2-predicted of 64.33%. The "Pred R-Squared" of 64.33% is in reasonable agreement with the "Adj R-Squared" of 77.01%. In this model, ratio of 12.753 indicates an adequate signal. This model can be used to navigate the design space. In Table 3.18, the Model F-value of 10.82 implies the model is significant. There is only a 0.08% chance that a "Model F-Value" this large could occur due to noise. In this case A, B, C are significant model terms due to values of "Prob > F" less than 0.0500. The model 65

University of Engineering & Technology, Taxila-Pakistan

possesses the R2 of 79.74%, R2-adjusted of 72.37%, and R2-predicted of 57.13%. The "Pred R-Squared" of 57.13% is in reasonable agreement with the "Adj R-Squared" of 72.37%. In this model, ratio of 11.134 indicates an adequate signal. This model can be used to navigate the design space.

Fig. 3.32 Effects of welding parameters upon Distortion (t = 3 mm)

66

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.33 Effects of welding parameters upon Distortion (t = 4 mm)

Fig. 3.34 Effects of welding parameters upon Distortion (t = 5 mm) 67

University of Engineering & Technology, Taxila-Pakistan

The empirical model for the weld distortion (mm), in terms of welding parameters, is as follows: Equation 3.7 (Trailing = nil) and Equation 3.8 (Trailing = Ar) in terms of actual factors for t = 3 mm for weld distortion: Distortion = + 0.23750 + 0.036250* Current + 0. 0.23333* Voltage - 0.25000* Weld Speed

3.7

Distortion = - 0. 33750 +0.036250* Current + 0.23333* Voltage - 0.25000* Weld Speed

3.8

Equation 3.9 (Trailing = nil) and Equation 3.10 (Trailing = Ar) in terms of actual factors for t = 4 mm for weld distortion: Distortion = - 2.73750+ 0.040625* Current + 0.35417* Voltage - 0.31250* Weld Speed

3.9

Distortion = - 3.22500+ 0.040625* Current + 0.35417* Voltage - 0.31250* Weld Speed

3.10

Equation 3.11 (Trailing = nil) and Equation 3.12 (Trailing = Ar) in terms of actual factors for t = 5 mm for weld distortion: Distortion = + 3.94375 + 0.016875*Current + 0.17500* Voltage - 0.35833* Weld Speed

3.11

Distortion = + 3.46875+ 0.016875*Current + 0.17500* Voltage - 0.35833* Weld Speed

3.12

All the models presented in equations 3.7 to 3.12 for weld distortion are valid for the following ranges of input parameters: welding current: 170 to 210 A for 3 mm, 200 to 220 A for 4 mm and 230 to 270 A for 5 mm; welding voltage 10.5 to 13.5 V and welding speed 15 to 18 cm/min. The numerical optimization applied to the distortion data suggests that for any material thickness value lying between 3 and 5 mm, the distortion in TIG welding of HSLA steel can be minimized if the trailing is used along with low values of heat input i.e. low values of welding current and welding voltage and high value of welding speed. The predicted weld distortion values are 3.7 mm, 3.0 mm and 2.7 mm for thickness 3 mm, 4 mm and 5 mm respectively at input parameters as: i)170 A, 10.5 V, 18 cm/min, ii) 200 A, 10.5 V, 18 cm/min and iii) 230 A, 10.5 V, 18 cm/min respectively as shown in Figure 3.35, Figure 3.36 and Figure 3.37 with response desirability of 0.80 for minimization of distortion and maximization of weld strength as shown in Figure 3.47, Figure 3.48 and Figure 3.49 for response desirability with respect to predictors for thickness 3, 4 and 5 mm respectively. Response desirability solutions containing predictors (welding current, welding voltage, welding speed and trailing) and response (weld strength and distortion) with respect to desirability are given in detail for thickness 3, 4 and 5 mm in Table 3.23, Table 3.24 and Table 3.25 respectively. Each table shows fifty different solutions from maximum (0.80) desirability to minimum (0.50) desirability of response and related input parameters. Combine effect of desirability of different predictors is shown in Figure 3.50, Figure 3.51 and Figure 3.52 for thickness 3, 4 and 5 mm respectively.

68

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.35 Distortion Predictions w.r.t Predictors (t = 3 mm)

Fig. 3.36 Distortion Predictions w.r.t Predictors (t = 4 mm)

Fig. 3.37 Distortion Predictions w.r.t Predictors (t = 5 mm) 69

University of Engineering & Technology, Taxila-Pakistan

3.4.3 Weld Induced Residual Stresses Figure 3.38, Figure 3.39 and Figure 3.40 show the comparison of residual stresses for aforementioned sixteen tests in Table 3.5, Table 3.6 and Table 3.7 respectively. The maximum and minimum values of residual stresses obtained with respect to thickness of material are presented in Table 3.19. R e s p o n s e

o f E x p e r im e n ts C o n d u c te d ( t =

3

m m

)

608

6 0 0 54 0

Residual Stresses (MPa)

51 4

547

543

55 3

545

5 42

516

5 0 0

476

47 2

467

4 71

1 1

1 2

468

4 78

1 5

1 6

385

3 90

1 5

1 6

348

357

1 5

1 6

44 8

4 0 0

3 0 0

2 0 0

1 0 0

0 1

2

3

4

5

6

7 8 9 E x p e r im e n t

1 0 No .

1 3

1 4

Fig. 3.38 Residual Stresses of Sixteen Experiments ( t = 3 mm) R e s p o n s e

o f E x p e r im e n ts C o n d u c te d ( t =

4

m m

)

505

5 0 0 45 2

44 8

47 0

459

453

445

Residual Stresses (MPa)

422

4 38

409 39 1

4 0 0

382

3 89

1 1

1 2

36 6

3 0 0

2 0 0

1 0 0

0 1

2

3

4

5

6

7 8 9 E x p e r im e n t

1 0 No .

1 3

1 4

Fig. 3.39 Residual Stresses of Sixteen Experiments ( t = 4 mm) R e s p o n s e

o f E x p e r im e n ts C o n d u c te d ( t =

5

m m

)

5 0 0 452

Residual Stresses (MPa)

425

4 0 0

388

391

404

410

401

398

403

370

357

353

355

1 1

1 2

335

3 0 0

2 0 0

1 0 0

0 1

2

3

4

5

6

7 8 9 1 0 E x p e r im e n t N o .

1 3

1 4

Fig. 3.40 Residual Stresses of Sixteen Experiments ( t = 5 mm) 70

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 3.19 Max. and Min. Values of Residual Stresses Response Thickness

Response

Units Minimum

Maximum

Mean

Std. Dev.

t = 3 mm

Residual Stresses

MPa

448

608

511.75

44.471

t = 4 mm

Residual Stresses

MPa

366

505

425.25

39.211

t = 5 mm

Residual Stresses

MPa

335

452

384.12

32.087

The combination of welding current, welding voltage, welding speed and gas trailing gave the high and low value of residual stresses at high level of welding parameters without application of trailing and at low level of welding parameters with application of trailing respectively. The results of experiment no. 9 and 10 according to design of experiments applying full factorial as given in Table 3.5, Table 3.6 and Table 3.7 shows the low and high values of residual stresses at low level of welding current and voltage, high value of welding speed with application of trailing and at high level of welding current and voltage, low level of welding speed without application of trailing for sheet thickness 3, 4 and 5 mm respectively. As the sheet thickness increases the residual stresses level decreases although welding current level increases with increase of material thickness. As the sheet thickness increases from 3 mm to 5 mm, the minimum residual stresses observed decreases from 448 MPa to 335 MPa and maximum residual stresses also decreases from 608 MPa to 452 MPa respectively. The mean value of residual stresses is 384 MPa to 511 MPa whereas the standard deviation is 32, 39 and 44 for 5mm, 4mm and 3mm thickness respectively. The results show that with increase of sheet thickness from 30% to 65%, the welding residual stresses decreases by 20% to 25% in value for 3-5 mm thickness range of HSLA steel sheet in general. The residual stresses data were analyzed using ANOVA technique and observations are presented in Table 3.20, Table 3.21 and Table 3.22 for the sheet thickness of 3, 4 and 5 mm respectively. The analysis shows that effects of three parameters (welding current, welding voltage and welding speed) are significant upon residual stresses. The effect of application of trailing on residual stresses is also significant.

Table 3.20 ANOVA for Residual Stresses (t=3mm) factorial model Source

Sum sqrs

Model 27779.50 A-Current 2162.25 B-Voltage 1892.25 C-Weld Speed 1225.00 D-Trailing 22500.00 Residual 1885.50 Cor Total 29665.00

DoF

Mean square F-value

4 1 1 1 1 11 15

6944.88 2162.25 1892.25 1225.00 22500.00 171.41

40.52 12.61 11.04 7.15 31.26

Prob>F 0.0001 0.0045 0.0068 0.0217 0.0001

Significance Significant Significant Significant Significant Significant

71

University of Engineering & Technology, Taxila-Pakistan

Table 3.21 ANOVA for Residual Stresses (t=4mm) factorial model Source

Sum sqrs

Model 21376.25 A-Current 2304.00 B-Voltage 3721.00 C-Weld Speed 1190.25 D-Trailing 14161.00 Residual 1686.75 Cor Total 23063.00

DoF

Mean square F-value

4 1 1 1 1 11 15

5344.06 2304.00 3721.00 1190.25 14161.00 153.34

34.85 15.03 24.27 7.76 92.35

Prob>F 0.0001 0.0026 0.0005 0.0177 0.0001

Significance Significant Significant Significant Significant Significant

Table 3.22 ANOVA for Residual Stresses (t=5mm) factorial model Source

Sum sqrs

Model 14439.75 A-Current 1387.56 B-Voltage 1501.56 C-Weld Speed 473.06 D-Trailing 11077.56 Residual 1004.69 Cor Total 15444.44

DoF

Mean square F-value

4 1 1 1 1 11 15

3609.94 1387.56 1501.56 473.06 11077.56 91.34

39.52 15.19 16.44 5.18 121.28

Prob>F 0.0001 0.0025 0.0019 0.0439 0.0001

Significance Significant Significant Significant Significant Significant

Figure 3.41, Figure 3.42 and Figure 3.43 shows the effects of changing the levels of each parameter upon residual stresses while keeping other three parameters fixed for sheet thickness of 3, 4 and 5 mm respectively. It is clear that effects of welding current, voltage and speed are significant. The residual stresses increases with increase of welding current and voltage values without the application of trailing and decreases with increase of welding speed with the application of trailing. Table 3.19 shows the comparison of weld residual stresses values obtained for sixteen experiments. The response values for weld residual stresses (3, 4 and 5 mm thickness) range from 448 to 608 MPa, 366 to 505 MPa and 335 to 452 MPa, providing the ratio of maximum to minimum equal to 1.357, 1.379 and 1.349 respectively. The ratio is small and, thus, there is no need to apply any kind of transformation to the data. Table 3.20, Table 3.21 and Table 3.22 presents the ANOVA details for the suggested factorial model. Table 3.20, the Model F-value of 40.52 implies the model is significant. There is only a 0.01% chance that a "Model F-Value" this large could occur due to noise. Values of "Prob > F" less than 0.0500 indicate model terms are significant. In this case A, B, C and D are significant model terms. The model possesses the R2 of 93.64%, R2-adjusted of 91.33%, and R2-predicted of 86.55%. The "Pred R-Squared" of 86.55% is in reasonable agreement with the "Adj R-Squared" of 91.33%. "Adeq Precision" measures the signal to noise ratio. A ratio greater than 4 is desirable. In this model, ratio of 18.787 indicates an adequate signal. This model can be used to navigate the design space.

72

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.41 Effects of welding parameters upon Residual Stresses (t = 3 mm)

Fig. 3.42 Effects of welding parameters upon Residual Stresses (t = 4 mm) 73

University of Engineering & Technology, Taxila-Pakistan

Fig. 3.43 Effects of welding parameters upon Residual Stresses (t = 5 mm) Table 3.21, the Model F-value of 34.85 implies the model is significant. There is only a 0.01% chance that a "Model F-Value" this large could occur due to noise. Values of "Prob > F" less than 0.0500 indicate model terms are significant. In this case A, B, C and D are significant model terms. The model possesses the R2 of 92.69%, R2-adjusted of 90.03%, and R2-predicted of 84.53%. The "Pred R-Squared" of 84.53% is in reasonable agreement with the "Adj R-Squared" of 90.03%. In this model, ratio of 18.96 indicates an adequate signal. This model can be used to navigate the design space. In Table 3.22, the Model F-value of 39.52 implies the model is significant. There is only a 0.01% chance that a "Model F-Value" this large could occur due to noise. In this case A, B, C and D are significant model terms due to values of "Prob > F" less than 0.0500. The model possesses the R2 of 93.49%, R2-adjusted of 91.13%, and R2-predicted of 86.24%. The "Pred R-Squared" of 86.24% is in reasonable agreement with the "Adj R-Squared" of 91.13%. In this model, ratio of 18.999 indicates an adequate signal. This model can be used to navigate the design space. The empirical model for the weld residual stresses (MPa), in terms of welding parameters, is as follows:

74

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Equation 3.13 (Trailing = nil) and Equation 3.14 (Trailing = Ar) in terms of actual factors for t = 3 mm for weld residual stresses: Residual Stresses = +448.06250 +0.58125* Current +7.25000* Voltage -5.83333* Weld Speed

3.13

Residual Stresses = +373.06250 +0.58125* Current + 7.25000* Voltage - 5.83333* Weld Speed

3.14

Equation 3.15 (Trailing = nil) and Equation 3.16 (Trailing = Ar) in terms of actual factors for t = 4 mm for weld residual stresses: Residual Stresses = +175.87500 + 1.20000* Current + 10.16667* Voltage - 5.75000* Weld Speed

3.15

Residual Stresses = +116.37500 + 1.20000* Current + 10.16667* Voltage - 5.75000* Weld Speed

3.16

Equation 3.17 (Trailing = nil) and Equation 3.18 (Trailing = Ar) in terms of actual factors for t = 5 mm for weld residual stresses: Residual Stresses = + 276.40625+ 0.46562*Current + 6.45833* Voltage - 3.62500* Weld Speed

3.17

Residual Stresses = + 223.78125 + 0.016875*Current + 0.17500* Voltage - 0.35833* Weld Speed

3.18

All the models presented in equations 3.13 to 3.18 for weld residual stresses are valid for the following ranges of input parameters: welding current: 170 to 210 A for 3 mm, 200 to 220 A for 4 mm and 230 to 270 A for 5 mm; welding voltage 10.5 to 13.5 V and welding speed 15 to 18 cm/min. The numerical optimization applied to the residual stresses data suggests that for any material thickness value lying between 3 and 5 mm, the residual stresses in TIG welding of HSLA steel can be minimized if the trailing is used along with low values of heat input i.e. low values of welding current and welding voltage and high value of welding speed. The predicted weld residual stresses values are 443 MPa, 359 MPa and 333 MPa for thickness 3 mm, 4 mm and 5 mm respectively at input parameters as: i)170 A, 10.5 V, 18 cm/min, ii) 200 A, 10.5 V, 18 cm/min and iii) 230 A, 10.5 V, 18 cm/min respectively as shown in Figure 3.44, Figure 3.45 and Figure 3.46 with response desirability of 0.91 for minimization of distortion, residual stresses and maximization of weld strength as shown in Figure 3.47, Figure 3.48 and Figure 3.49 for response desirability with respect to predictors for thickness 3, 4 and 5 mm respectively. Response desirability solutions containing predictors (welding current, welding voltage, welding speed and trailing) and response (weld strength, distortion and residual stresses) with respect to desirability are given in detail for thickness 3, 4 and 5 mm in Table 3.23, Table 3.24 and Table 3.25 respectively. Each table shows about fifty different solutions from maximum (0.90) desirability to minimum (0.50) desirability of response and related input parameters. Combine effect of desirability of different predictors is shown in Figure 3.50, Figure 3.51 and Figure 3.52 for thickness 3, 4 and 5 mm respectively. Comparison of responses achieved is shown in Figure 3.53 which shows that weld strength increases with decrease of distortion & residual stresses and vice versa respectively. 75

University of Engineering & Technology, Taxila-Pakistan

Fig. 3.44 Residual Stresses Predictions w.r.t Predictors (t = 3 mm)

Fig. 3.45 Residual Stresses Predictions w.r.t Predictors (t = 4 mm)

Fig. 3.46 Residual Stresses Predictions w.r.t Predictors (t = 5 mm) 76

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 3.23 Response Desirability Solutions (t = 3 mm) Number Current Voltage Weld Trailing Speed

Weld Strength

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48

783.844 783.767 783.731 783.594 783.639 783.539 783.522 783.473 783.456 783.413 783.32 783.128 783.04 783.104 782.788 782.568 782.395 782.636 782.298 782.003 781.83 780.947 780.592 780.44 778.547 777.143 769.881 769.818 769.795 769.712 769.701 769.673 769.575 769.493 769.511 769.347 769.292 768.896 768.57 767.931 767.851 767.589 767.481 767.227 766.494 766.638 766.286 763.674

170.00 170.00 170.00 170.00 170.46 170.00 170.00 170.00 170.00 170.00 170.26 170.00 170.00 171.67 170.00 170.00 170.00 172.72 170.00 170.04 170.00 170.00 170.00 170.00 181.94 170.00 170.00 170.00 170.19 170.00 170.01 170.47 170.41 170.00 170.00 170.00 170.00 170.00 170.00 170.00 170.02 170.00 170.00 175.99 170.00 170.00 170.00 170.00

10.50 10.52 10.50 10.50 10.50 10.50 10.57 10.58 10.50 10.60 10.50 10.50 10.68 10.50 10.50 10.79 10.82 10.50 10.85 10.50 10.50 11.15 11.23 10.50 10.50 12.00 10.50 10.50 10.50 10.50 10.54 10.50 10.53 10.50 10.58 10.62 10.50 10.50 10.79 10.94 10.95 11.01 11.04 10.50 10.50 11.23 10.50 11.89

18.00 18.00 17.98 17.95 18.00 17.93 18.00 18.00 17.92 18.00 17.91 17.85 18.00 18.00 17.77 18.00 18.00 18.00 18.00 17.61 17.57 18.00 18.00 17.27 18.00 18.00 18.00 17.99 18.00 17.96 18.00 18.00 18.00 17.92 18.00 18.00 17.88 17.79 18.00 18.00 18.00 18.00 18.00 18.00 17.27 18.00 17.23 18.00

Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil

Distortion Residual Stresses 3.775 3.77899 3.78102 3.78836 3.7917 3.79132 3.79184 3.79437 3.79575 3.7975 3.8062 3.81333 3.81705 3.83546 3.83149 3.8417 3.85078 3.87371 3.85581 3.87403 3.88276 3.92649 3.94503 3.95719 4.208 4.12534 4.35 4.35338 4.35702 4.35906 4.35958 4.367 4.37129 4.37075 4.36937 4.37792 4.38149 4.40271 4.41855 4.45197 4.45642 4.46986 4.47551 4.56698 4.53131 4.5196 4.54244 4.67455

443.00 443.124 443.14 443.312 443.268 443.381 443.523 443.602 443.484 443.699 443.661 443.894 444.306 443.97 444.318 445.072 445.354 444.583 445.511 445.3 445.515 447.707 448.283 447.251 449.943 453.886 518 518.079 518.113 518.211 518.291 518.273 518.44 518.484 518.602 518.868 518.744 519.23 520.13 521.168 521.296 521.724 521.899 521.479 522.231 523.27 522.49 528.084

Desirability

0.910 0.910 0.909 0.908 0.908 0.907 0.907 0.907 0.906 0.906 0.905 0.903 0.902 0.901 0.899 0.897 0.895 0.895 0.894 0.891 0.889 0.880 0.876 0.875 0.837 0.829 0.639 0.638 0.638 0.637 0.636 0.636 0.634 0.634 0.634 0.632 0.631 0.627 0.622 0.613 0.612 0.609 0.607 0.600 0.597 0.596 0.594 0.557

77

University of Engineering & Technology, Taxila-Pakistan

Table 3.24 Response Desirability Solutions (t = 4 mm) Number Current Voltage Weld Trailing Speed

Weld Distortion Strength

Residual Desirability Stresses

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53

772.044 771.982 771.974 771.944 771.921 771.894 771.812 771.733 771.801 771.686 771.573 771.543 771.212 771.41 771.14 771.311 770.948 770.644 770.431 769.751 769.89 768.879 768.703 768.41 768.135 765.982 760.955 760.742 759.556 759.504 759.512 759.454 759.359 759.408 759.186 759.253 759.086 759.086 759.119 758.326 757.965 757.568 757.178 756.206 756.509 755.555 754.447 754.28 755.647 753.494 751.049 748.266 747.616

359.625 359.78 359.72 359.752 359.793 360.001 359.92 360.021 360.233 360.115 360.224 360.881 360.684 361.215 360.775 360.629 361.019 361.406 361.835 362.543 365.025 363.652 368.003 368.736 369.427 367.338 374.822 375.353 419.125 419.196 419.236 419.38 419.376 419.496 419.596 419.887 419.723 419.77 420.135 422.209 421.306 421.655 422.382 423.387 423.302 424.608 425.625 425.839 428.927 426.838 429.977 434.597 435.481

200.00 200.00 200.00 200.11 200.00 200.00 200.25 200.33 200.00 200.00 200.50 200.00 200.88 200.00 200.96 200.00 201.16 201.48 200.00 202.43 200.00 203.36 200.00 200.00 200.00 206.43 200.00 200.00 200.00 200.00 200.00 200.00 200.21 200.00 200.39 200.00 200.50 200.00 200.00 200.00 200.00 202.11 200.03 203.55 200.00 200.00 205.42 205.59 200.00 206.43 208.72 200.00 200.08

10.50 10.52 10.50 10.50 10.50 10.54 10.50 10.50 10.56 10.50 10.50 10.62 10.50 10.66 10.50 10.50 10.50 10.50 10.50 10.50 11.03 10.50 11.32 11.40 11.46 10.50 10.50 10.55 10.50 10.50 10.51 10.52 10.50 10.54 10.50 10.57 10.50 10.50 10.59 10.80 10.50 10.50 10.50 10.50 10.50 10.50 10.50 10.50 11.46 10.50 10.50 10.50 10.50

18.00 18.00 17.98 18.00 17.97 18.00 18.00 18.00 18.00 17.91 18.00 18.00 18.00 18.00 18.00 17.83 18.00 18.00 17.62 18.00 18.00 18.00 18.00 18.00 18.00 18.00 15.36 15.36 18.00 17.99 18.00 18.00 18.00 18.00 18.00 18.00 18.00 17.89 17.98 18.00 17.62 18.00 17.44 18.00 17.27 17.05 18.00 18.00 18.00 18.00 17.93 15.31 15.17

Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil

2.99375 2.99915 2.99881 2.99804 3.00289 3.00685 3.00372 3.00715 3.01492 3.02036 3.01404 3.03749 3.0296 3.04912 3.03271 3.04831 3.04096 3.05406 3.11384 3.09252 3.18188 3.1301 3.28559 3.31115 3.33522 3.25488 3.81965 3.83818 3.48125 3.48511 3.48513 3.49012 3.4898 3.49416 3.49718 3.50778 3.50149 3.51629 3.51849 3.58871 3.59979 3.56691 3.65752 3.62555 3.70825 3.77924 3.70131 3.70855 3.82272 3.74238 3.85653 4.32211 4.36817

0.931 0.930 0.930 0.930 0.930 0.929 0.929 0.928 0.928 0.926 0.926 0.924 0.922 0.922 0.922 0.921 0.920 0.917 0.909 0.907 0.899 0.898 0.877 0.871 0.865 0.864 0.758 0.754 0.682 0.681 0.681 0.680 0.679 0.679 0.677 0.676 0.676 0.674 0.673 0.657 0.656 0.656 0.644 0.638 0.633 0.618 0.615 0.612 0.603 0.602 0.569 0.500 0.490

78

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 3.25 Response Desirability Solutions (t = 5 mm) Number Current Voltage Weld Trailing Speed

Weld Distortion Strength

Residual Desirability Stresses

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57

761.444 761.382 761.348 761.194 761.23 761.229 761.193 761.193 761.011 761.134 761.016 760.975 760.479 760.489 759.98 760.496 760.1 759.619 760.281 758.952 758.369 756.135 757.158 754.927 753.381 755.966 751.173 747.323 747.731 747.679 747.61 747.61 747.617 747.364 747.465 747.418 747.296 747.42 747.365 747.376 747.28 747.137 746.909 747.06 746.843 746.859 746.722 745.992 746.114 745.85 745.318 745.239 744.38 742.555 741.002 738.707 738.32

333.438 333.553 333.539 333.704 333.757 333.664 333.753 333.704 333.899 333.765 333.889 333.933 334.468 335.21 335.001 334.44 335.932 335.386 334.666 336.072 339.144 339.107 337.968 345.535 342.049 339.228 352.502 356.747 386.063 386.126 386.192 386.288 386.185 386.455 386.345 386.643 386.528 386.392 386.743 386.512 386.901 387.166 386.94 387.309 387.011 387.681 387.935 387.921 389.064 389.554 390.542 388.697 392.284 391.534 398.554 402.813 403.532

230.00 230.00 230.00 230.57 230.00 230.01 230.00 230.00 230.99 230.00 230.00 230.00 232.21 230.00 233.36 230.00 230.00 234.18 230.00 230.00 230.00 242.18 230.00 230.00 248.49 230.00 230.00 238.45 230.00 230.00 230.28 230.00 230.00 230.84 230.00 230.00 231.00 230.00 230.00 230.00 230.00 230.00 231.89 230.00 232.04 230.00 230.00 233.99 230.00 230.00 230.00 230.00 230.00 230.00 230.00 230.00 230.00

10.50 10.52 10.50 10.50 10.53 10.50 10.52 10.50 10.50 10.50 10.50 10.50 10.50 10.77 10.50 10.50 10.89 10.50 10.50 10.50 11.38 10.50 10.50 12.37 10.50 10.50 13.45 13.50 10.50 10.50 10.50 10.53 10.50 10.50 10.50 10.59 10.50 10.50 10.61 10.53 10.63 10.67 10.50 10.69 10.50 10.75 10.79 10.50 10.96 11.04 11.19 10.50 11.46 10.50 12.43 13.09 13.20

18.00 18.00 17.97 18.00 17.97 17.94 17.94 17.93 18.00 17.91 17.88 17.86 18.00 18.00 18.00 17.72 18.00 18.00 17.66 17.27 18.00 18.00 16.75 18.00 18.00 16.40 18.00 18.00 18.00 17.99 18.00 18.00 17.97 18.00 17.92 18.00 18.00 17.91 18.00 17.92 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00 17.27 18.00 16.49 18.00 18.00 18.00

Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil

2.7375 2.74063 2.74754 2.74716 2.75356 2.75959 2.76036 2.76372 2.75424 2.76982 2.78215 2.78645 2.77486 2.78554 2.79417 2.83657 2.8051 2.80813 2.85896 2.9979 2.89213 2.943 3.18535 3.06531 3.04959 3.30988 3.25413 3.40508 3.2125 3.21748 3.21721 3.21861 3.22433 3.22672 3.24026 3.22824 3.22936 3.24502 3.23094 3.24465 3.23522 3.24241 3.24432 3.24627 3.24687 3.25636 3.26324 3.27984 3.29383 3.30711 3.33388 3.4729 3.38108 3.75338 3.55098 3.6664 3.68587

0.917 0.916 0.915 0.914 0.914 0.913 0.913 0.913 0.912 0.912 0.909 0.909 0.907 0.905 0.901 0.900 0.899 0.896 0.896 0.870 0.869 0.849 0.832 0.811 0.811 0.806 0.748 0.695 0.635 0.634 0.633 0.632 0.632 0.630 0.629 0.629 0.629 0.629 0.628 0.628 0.627 0.625 0.624 0.623 0.623 0.620 0.618 0.611 0.607 0.603 0.594 0.587 0.578 0.534 0.521 0.482 0.475

79

University of Engineering & Technology, Taxila-Pakistan

Fig. 3.47 Response Desirability w.r.t Predictors (t = 3 mm)

Fig. 3.48 Response Desirability w.r.t Predictors (t = 4 mm)

Fig. 3.49 Response Desirability w.r.t Predictors (t = 5 mm) 80

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.50 Effect on Desirability of different Predictors (t = 3 mm)

Fig. 3.51 Effect on Desirability of different Predictors (t = 4 mm)

Fig. 3.52 Effect on Desirability of different Predictors (t = 5 mm) 81

University of Engineering & Technology, Taxila-Pakistan

Response Comparison for t=3 mm

Response Comparison for t=3 mm

Scator Plots

Scator Plots

800 7

Distortion (mm)

Weld Strength (MPa)

780

760

740

720

6

5

4 700 3 450

475

500 525 550 575 Residual Stresses (MPa)

600

625

450

475

Response Comparison for t=4 mm

500 525 550 Residual Stresses (MPa)

575

600

625

Response Comparison for t=4 mm

Scator Plots

Scator Plots 6.5

780

6.0 5.5 760

Distortion (mm)

Weld Strength (MPa)

770

750 740

5.0 4.5 4.0

730 3.5 720 3.0 710 350

375

400 425 450 Residual Stresses (MPa)

475

500

350

375

ResponseComparisonfor t=5mm

400 425 450 Residual Stresses (MPa)

475

500

Response Comparisonfor t=5 mm

Scator Plots

Scator Plots

770

6

5 Distortion (mm)

Weld Strength (MPa)

760 750

740 730

4

3 720

710

2 350

375 400 Residual Stresses (MPa)

425

450

350

375 400 Residual Stresses (MPa)

425

450

Fig. 3.53 Response Comparison Scator Plots (t = 3 mm, 4 mm, 5 mm)

82

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

3.5 Numerical Optimization and Empirical Modeling using Response Surface Method After performing the experimentation and analysis following the 2-level full factorial design (three numeric factors and one categoric factor) as given in Table 3.5, Table 3.6 and Table 3.7 for the sheet thicknesses of 3, 4 and 5 mm respectively for the responses (weld strength, distortion and residual stresses), the effect of thickness as another variable with earlier variables (i.e. four numeric factors and one categoric factor) is analyzed by employing response surface method (RSM). The Response Surface Method (RSM), a design of experiments (DOE) methodology, helps to quantify the relationships between one or more measured responses and the vital input variables/factors. Usually, a two-level factorial method is used for screening design first for five or more factors. The objective is to find a desirable (maximum or minimum for a range of the factors) position in the design to meet a set of parameters for several responses simultaneously. There are a number of RSM designs, including number of numeric or categoric factors, like Central Composite Design, Box-Behnken Design, 3-Level Factorial Design, One Factor RSM Design, Pentagon Design, Hexagon Design, Hybrid Design, RSM D-optimal Design, Distance Based Design, User Defined Design or Historical Data etc. In Historical Data design, any set of numeric or categoric data can be used and analyzed by specifying the number of numeric and categoric factors, the minimum and maximum values for each numeric factor, the number of experiments (equal to the number of rows available in data file/historical data), the levels of any categoric factor(s), and the response names. The factors with low and high settings are given in Table 3.26 and the complete historical data (3x16 observations for each thickness) is given in Table 3.27 with response values. The summary of design is given in the Table 3.28. During model analysis, fit summary suggested to use 2FI model for analysis of variance. ANOVA results are shown in Table 3.29, Table 3.30 and Table 3.31 for weld strength, distortion and residual stresses respectively. The values of R-Squared, Adj R-Squared and Pred R-Squared including Adeq Precision (max to min ratio) are given in Table 3.32. Figure 3.54, Figure 3.55 and Figure 3.56 shows the effects of changing the levels of each parameter upon weld strength, distortion and residual stresses while keeping other parameters fixed respectively. It is clear that effects of welding current, voltage, speed and thickness are significant. The distortion and residual stresses increases with increase of welding current and voltage values without the application of trailing and decreases with increase of welding speed and thickness with the application of trailing. Figure 3.57, Figure 3.58 and Figure 3.59 shows the interaction effect of welding parameters upon weld strength, distortion and residual stresses respectively. Table 3.26 High and Low Settings of Factors for RSM Factor Name

Units

Type

A B C D E

A V cm/min mm

Numeric Numeric Numeric Numeric Categoric

Current Voltage Weld Speed Thickness Trailing

Low Actual 170.00 10.50 15.00 3.0 nil

High Actual 270.0 13.50 18.00 5.0 Ar

Mean 216.66 12.0 16.5 4.0

83

University of Engineering & Technology, Taxila-Pakistan

Table 3.27 Historical Data (48 (3x16) observations) including Response Values for RSM Run

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48

Factor 1 A:Current (A)

Factor 2 B:Voltage (V)

210.00 170.00 170.00 170.00 170.00 170.00 210.00 210.00 170.00 210.00 170.00 210.00 210.00 210.00 170.00 210.00 220.00 200.00 200.00 200.00 200.00 200.00 220.00 220.00 200.00 220.00 200.00 220.00 220.00 220.00 200.00 220.00 270.00 230.00 230.00 230.00 230.00 230.00 270.00 270.00 230.00 270.00 230.00 270.00 270.00 270.00 230.00 270.00

13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50 13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50 13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50

Factor 3 Factor 4 Factor 5 Responses C:Weld Speed D: Thickness D: Trailing Weld Distortion (cm/min) (mm) Strength (MPa) (mm) 15.00 3.00 Ar 696.8 6.5 18.00 3.00 nil 751.6 4.6 15.00 3.00 nil 733.1 5.2 15.00 3.00 nil 704.3 6.5 15.00 3.00 Ar 708.3 5.7 18.00 3.00 nil 715.4 5.8 18.00 3.00 nil 698.5 6.9 18.00 3.00 Ar 702.7 6.2 18.00 3.00 Ar 791 3.2 15.00 3.00 nil 690.7 7.2 15.00 3.00 Ar 767.5 3.8 15.00 3.00 Ar 718.8 5.3 15.00 3.00 nil 711.7 6.2 18.00 3.00 nil 718.6 5.5 18.00 3.00 Ar 733.4 4.7 18.00 3.00 Ar 748.3 4.3 15.00 4.00 Ar 736.4 5.9 18.00 4.00 nil 751.3 3.7 15.00 4.00 nil 741.2 4.4 15.00 4.00 nil 725.2 5.6 15.00 4.00 Ar 739.6 5.2 18.00 4.00 nil 731.4 4.6 18.00 4.00 nil 724.3 5.7 18.00 4.00 Ar 737.2 5.4 18.00 4.00 Ar 780.4 2.8 15.00 4.00 nil 722.3 6.2 15.00 4.00 Ar 760.1 3.4 15.00 4.00 Ar 742.5 4.9 15.00 4.00 nil 727.5 5.3 18.00 4.00 nil 736.3 4.5 18.00 4.00 Ar 750.3 4.1 18.00 4.00 Ar 755.4 3.5 15.00 5.00 Ar 726.7 5.2 18.00 5.00 nil 759.5 3 15.00 5.00 nil 737.8 4.5 15.00 5.00 nil 717.8 5 15.00 5.00 Ar 730.5 4.5 18.00 5.00 nil 729.5 4 18.00 5.00 nil 715 5.3 18.00 5.00 Ar 730 4.9 18.00 5.00 Ar 765.7 2.2 15.00 5.00 nil 715 5.6 15.00 5.00 Ar 757.7 3.4 15.00 5.00 Ar 735.7 3.7 15.00 5.00 nil 725 4.8 18.00 5.00 nil 729.5 4.6 18.00 5.00 Ar 742.8 3.6 18.00 5.00 Ar 749.8 3.5

Residual Stresses (MPa) 514 516 540 547 472 543 553 476 448 608 467 471 545 542 468 478 452 422 448 459 391 453 470 409 366 505 382 389 445 438 385 390 388 391 404 410 357 401 425 370 335 452 353 355 398 403 348 357

84

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 3.28 Max. and Min. Values of Responses in RSM Response Weld Strength Distortion Residual Stress

Units MPa mm MPa

Minimum 690.7 2.2 335

Maximum 791 7.2 608

Mean Std. Dev. 733.752 21.3264 4.80417 1.12741 440.396 65.8586

Table 3.29 ANOVA for Weld Strength (2FI model) of RSM Source

Sum sqrs

Model A-Current B-Voltage C-Weld Speed D-Thickness E-Trailing AB AC AD AE BC BD BE CD CE DE Residual Cor Total

19349.59 4960.11 6068.05 1393.28 5692.48 2953.94 319.13 97.02 3.31 88.66 139.74 6.928E-003 227.51 5.95 67.93 17.11 2026.83 21376.42

DoF

Mean square

F-value

Prob>F

Significance

15 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 32 47

1289.97 4960.11 6068.05 1393.28 5692.48 2953.94 319.13 97.02 3.31 88.66 139.74 6.928E-003 227.51 5.95 67.93 17.11 63.34

20.37 78.31 95.80 22.00 89.87 46.64 5.04 1.53 0.052 1.40 2.21 1.094E-004 3.59 0.094 1.07 0.27

0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0318 0.2248 0.8205 0.2455 0.1472 0.9917 0.0671 0.7613 0.3082 0.6068

Significant Significant Significant Significant Significant Significant Significant

Table 3.30 ANOVA for Distortion (2FI model) of RSM Source

Sum sqrs

Model A-Current B-Voltage C-Weld Speed D-Thickness E-Trailing AB AC AD AE BC BD BE CD CE DE Residual Cor Total

57.66 11.14 18.08 5.43 22.61 7.01 1.03E-003 0.57 0.16 0.011 0.053 0.083 0.70 0.42 0.013 0.027 2.08 59.74

DoF

Mean square

F-value

Prob>F

Significance

15 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 32 47

3.84 11.14 18.08 5.43 22.61 7.01 1.034E-003 0.57 0.16 0.011 0.053 0.083 0.70 0.42 0.013 0.027 0.065

59.23 171.67 278.49 83.74 348.39 108.06 0.016 8.80 2.50 0.18 0.82 1.27 10.80 6.54 0.21 0.42

0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.9003 0.0057 0.1237 0.6767 0.3715 0.2674 0.0025 0.0155 0.6534 0.5236

Significant Significant Significant Significant Significant Significant Significant

Significant Significant

85

University of Engineering & Technology, Taxila-Pakistan

Table 3.31 ANOVA for Residual Stresses (2FI model) of RSM Source

Sum sqrs

Model A-Current B-Voltage C-Weld Speed D-Thickness E-Trailing AB AC AD AE BC BD BE CD CE DE Residual Cor Total

1.977E+005 6810.46 7451.38 2786.90 77338.73 45187.59 751.82 29.60 829.03 0.52 426.02 580.89 188.02 98.54 38.52 372.02 6125.63 2.039E+005

DoF

Mean square

F-value

Prob>F

Significance

15 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 32 47

13181.99 6810.46 7451.38 2786.90 77338.73 45187.59 751.82 29.60 829.03 0.52 426.02 580.89 188.02 98.54 38.52 372.02 191.43

68.86 35.58 38.93 14.56 404.01 236.06 3.93 0.15 4.33 2.736E-003 2.23 3.03 0.98 0.51 0.20 1.94

0.0001 0.0001 0.0001 0.0006 0.0001 0.0001 0.0561 0.6967 0.0455 0.9586 0.1455 0.0911 0.3291 0.4783 0.6568 0.1729

Significant Significant Significant Significant Significant Significant

Significant

Table 3.32 ANOVA Summary for RSM (2FI model) Response

Adeq Precision

R-Squared

Adj R-Squared

Pred R-Squared

Weld Strength Distortion Residual Stress

18.98 34.25 32.269

90.52% 96.52% 97.0%

86.07% 94.89% 95.59%

78.77% 92.81% 93.39%

Fig. 3.54 Effects of welding parameters upon Weld Strength in RSM 86

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.55 Effects of welding parameters upon Distortion in RSM

Fig. 3.56 Effects of welding parameters upon Residual Stresses in RSM 87

University of Engineering & Technology, Taxila-Pakistan

Fig. 3.57 Interaction of welding parameters upon Weld Strength in RSM

88

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.58 Interaction of welding parameters upon Distortion in RSM

89

University of Engineering & Technology, Taxila-Pakistan

Fig. 3.59 Interaction of welding parameters upon Residual Stresses in RSM

90

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

From Table 3.29, Table 3.30 and Table 3.31, the significance of all models is clear from the Model F-value which is greater than 4. As the values of "Prob > F" less than 0.0500 indicate the significance of model terms, in this case A, B, C, D and E terms are significant with some interaction terms. The Table 3.32 shows that the "Pred R-Squared" values are in reasonable agreement with the "Adj R-Squared" values and “Adeq Precision” values indicate an adequate signal that shows the models can be used to navigate the design space. The empirical models for the responses (weld strength (MPa), distortion (mm) and weld residual stresses (MPa)), in terms of welding parameters, are as follows: Equation 3.19 (Trailing = nil) and Equation 3.20 (Trailing = Ar) in terms of actual factors for weld strength: Weld Strength = +727.24584 -0.82644 * Current -14.66111 * Voltage +21.63194* Weld Speed +11.01124* Thickness +0.095761 * Current * Voltage -0.052802* Current * Weld Speed +0.012346*Current * Thickness -0.75833 * Voltage * Weld Speed -0.016595* Voltage * Thickness +0.48614* Weld Speed * Thickness 3.19 Weld Strength = +775.28334 -0.97787* Current -17.56389* Voltage +23.21806 * Weld Speed +13.48516* Thickness +0.095761 * Current * Voltage -0.052802* Current * Weld Speed +0.012346* Current * Thickness -0.75833 * Voltage * Weld Speed -0.016595* Voltage * Thickness +0.48614* Weld Speed * Thickness 3.20

Equation 3.21 (Trailing = nil) and Equation 3.22 (Trailing = Ar) in terms of actual factors for distortion: Distortion = +8.48786 -0.027474 * Current +0.35833* Voltage -0.76667 * Weld Speed +1.94671 * Thickness -1.72414E-004 * Current * Voltage +4.05172E-003 * Current * Weld Speed -2.73160E-003 * Current * Thickness +0.014815 * Voltage * Weld Speed -0.057328 * Voltage * Thickness -0.12989* Weld Speed * Thickness 3.21 Distortion = +5.37119 -0.025750* Current +0.51944 * Voltage -0.78889 * Weld Speed +2.04499 * Thickness -1.72414E-004* Current * Voltage +4.05172E-003* Current * Weld Speed -2.73160E-003* Current * Thickness +0.014815* Voltage * Weld Speed -0.057328* Voltage * Thickness -0.12989 * Weld Speed * Thickness 3.22

Equation 3.23 (Trailing = nil) and Equation 3.24 (Trailing = Ar) in terms of actual factors for weld residual stresses: Residual Stresses = +749.20823 -1.38427* Current +18.50000* Voltage +8.62500* Weld Speed -107.71366* Thickness +0.14698* Current * Voltage -0.029167 * Current * Weld Speed +0.19529* Current * Thickness -1.32407 * Voltage * Weld Speed -4.80532* Voltage * Thickness +1.97917 * Weld Speed * Thickness 3.23 Residual Stresses = +655.16656 -1.39591* Current +15.86111* Voltage +9.81944* Weld Speed -96.17702 * Thickness +0.14698* Current * Voltage -0.029167* Current * Weld Speed +0.19529* Current * Thickness -1.32407 * Voltage * Weld Speed -4.80532* Voltage * Thickness +1.97917 * Weld Speed * Thickness 3.24

All the models presented in equations 3.19 to 3.24 for weld strength, distortion and weld residual stresses are valid for the following ranges of input parameters: welding current: 170 to 210 A for 3 mm, 200 to 220 A for 4 mm and 230 to 270 A for 5 mm; welding voltage 10.5 to 13.5 V and welding speed 15 to 18 cm/min. 91

University of Engineering & Technology, Taxila-Pakistan

The numerical optimization applied to the weld strength, distortion and residual stresses data suggests that for any material thickness value lying between 3 and 5 mm, the distortion and residual stresses in TIG welding of HSLA steel can be minimized (or maximization of weld strength) if the trailing is used along with low values of heat input i.e. low values of welding current and welding voltage and high value of welding speed. The predicted weld strength, distortion and weld residual stresses values are as: i) 781.067 MPa, 3.06 mm and 450.59 MPa, ii) 779 MPa, 2.73 mm and 384.89MPa, and iii) 777.72 MPa, 2.24 mm and 331 MPa at input parameters as: i) 170 A, 10.5 V, 18 cm/min, 3 mm, ii) 200 A, 10.5 V, 18 cm/min, 4 mm and iii) 230 A, 10.5 V, 18 cm/min, 5 mm respectively as shown in Figure 3.60, Figure 3.61 and Figure 3.62 with response desirability for minimization of distortion, residual stresses and maximization of weld strength as shown in Figure 3.63 respectively. Response desirability solutions containing predictors (welding current, welding voltage, welding speed, thickness and trailing) and response (weld strength, distortion and residual stresses) with respect to desirability are given in detail in Table 3.33. Combine effect of desirability of different predictors is shown in Figure 3.64 and the comparison of responses achieved is shown in Figure 3.65 which shows that weld strength increases with decrease of distortion & residual stresses and increase in thickness and vice versa respectively.

Fig. 3.60 Weld Strength Predictions w.r.t Predictors in RSM

92

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.61 Distortion Predictions w.r.t Predictors in RSM

Fig. 3.62 Residual Stresses Predictions w.r.t Predictors in RSM 93

University of Engineering & Technology, Taxila-Pakistan

Table 3.33 Response Desirability Solutions of RSM Number Current Voltage Weld ThicknessTrailing Weld Distortion Residual Desirability Speed Strength Stresses For Goal: Current is equal to 230 A. Thickness is equal to 5 mm. Voltage, Speed and Trailing are in range. 1 2 3

230.00 230.00 230.00

10.50 10.50 10.50

18.00 18.00 17.96

5.00 5.00 5.00

Ar nil nil

777.723 2.24491 330.904 0.951 754.072 3.18198 376.148 0.755 753.932 3.19362 376.222 0.754

For Goal: Current is equal to 200 A. Thickness is equal to 4 mm. Voltage, Speed and Trailing are in range. 1 2 3 4 5

200.00 200.00 200.00 200.00 200.00

10.50 10.50 10.50 10.50 10.52

18.00 17.97 17.83 18.00 18.00

4.00 4.00 4.00 4.00 4.00

Ar Ar Ar nil nil

779.025 778.819 777.865 753.305 753.139

2.73472 2.74535 2.79456 3.82179 3.82828

384.889 384.951 385.238 441.32 441.407

0.863 0.862 0.855 0.636 0.635

For Goal: Current is equal to 170 A. Thickness is equal to 3 mm. Voltage, Speed and Trailing are in range. 1 2 3 4 5

170.00 170.00 170.00 170.00 170.00

10.50 10.53 10.57 10.50 10.50

18.00 18.00 18.00 17.86 18.00

3.00 3.00 3.00 3.00 3.00

Ar Ar Ar Ar nil

781.067 780.543 780.049 780.003 753.279

3.06064 3.08111 3.10036 3.10658 4.29771

450.591 450.682 450.767 451.018 518.21

0.755 0.752 0.749 0.748 0.492

2.17225 310.462 1.82941 329.006 2.08702 291.846 2.09402 295.706 2.11706 294.952 2.09907 309.768 1.52106 282.607 1.56296 295.612 1.57069 286.336 1.4908 317.505 1.81224 295.318 2.06681 296.237 1.99403 307.527 1.48227 284.741 0.845637 305.534 1.97532 334.876 1.21165 312.77 0.810163 301.816 2.18041 308.187 0.827977 295.805 1.8781 305.729 2.17314 302.4 1.6166 307.298 1.77592 321.008 1.93631 331.546 1.84086 296.841 2.10879 322.053 1.65672 297.602 2.15628 309.23 1.50149 300.024 2.10174 287.059 1.84495 304.575 1.31233 306.489

1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000

For Goal: Current, Thickness, Voltage, Speed and Trailing are in range. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

170.08 183.32 173.80 172.19 170.30 175.24 172.61 170.97 172.49 204.32 171.09 173.53 171.37 170.69 177.37 187.54 181.13 178.53 191.87 170.95 173.57 178.72 171.49 186.85 184.16 178.08 170.81 171.97 176.82 173.48 171.29 173.59 177.06

10.52 11.03 12.03 10.77 10.52 11.10 12.78 11.84 12.52 10.52 12.47 11.37 10.55 12.71 10.60 10.60 10.62 10.65 10.65 10.84 10.69 10.54 11.30 11.15 10.60 11.77 10.53 11.45 11.28 11.63 12.11 10.56 11.10

15.30 17.38 16.44 15.37 15.08 16.05 17.99 17.25 17.72 17.98 17.43 15.97 15.61 17.99 17.81 16.99 17.54 17.85 15.89 17.58 15.95 15.34 16.97 17.48 16.81 16.90 15.70 16.78 16.14 17.36 16.30 15.86 17.49

4.79 4.64 4.97 4.97 4.97 4.82 4.99 4.90 4.97 4.97 4.86 4.95 4.83 4.95 4.91 4.61 4.85 4.97 4.99 4.97 4.87 4.95 4.80 4.76 4.63 4.96 4.65 4.91 4.83 4.88 5.00 4.89 4.87

Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar

800.682 797.225 791.492 800.843 802.881 796.413 793.073 800.452 794.663 798.736 791.756 796.546 802.98 794.488 818.009 795.767 810.681 817.928 792.638 819.052 803.195 798.316 803.146 796.62 797.674 795.354 800.35 801.938 793.9 802.105 791.638 804.677 807.428

94

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 3.63 Response Desirability w.r.t Predictors in RSM

Fig. 3.64 Effect on Desirability of different Predictors in RSM

95

University of Engineering & Technology, Taxila-Pakistan

Response Comparisons w.r.t Thickness Scator Plots 5.0

Thickness (mm)

4.5

4.0

3.5

3.0 700

720

740 760 Weld Strength (MPa)

780

800

Response Comparisons w.r.t Thickness S cator Plots 5.0

Thickness (mm)

4.5

4.0

3.5

3.0 2

3

4

5 Distortion (mm)

6

7

Response Comparisons w.r.t Thickness S cator Plots 5.0

Thickness (mm)

4.5 4.0 3.5 3.0 2.5 2.0 300

350

400 450 500 Residua l St re sses (MPa )

550

600

Fig. 3.65 Response Comparison Scator Plots 96

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

3.6 Chapter Summary and Conclusions In this chapter the influence of different welding parameters (welding current, welding voltage, welding speed, weld wire speed, gas flow rate and sheet thickness) was initially investigated upon weld quality to find out the practical parametric range and their effect on weld strength, distortion and residual stresses. The weld quality depended on heat input provided with respect to thickness of material. The experiments employing high and low heat input provided the large effect on weld quality. The welding current with large parametric range was most influential parameter with respect to thickness of sheets besides other parameters. Further, this chapter described the experimental methodology to study and quantify the effects of four welding parameters: welding current (A); welding voltage (V); welding speed (cm/min); and Ar trailing upon the TIG welding of HSLA steel performance measures of weld strength (MPa), distortion (mm) and residual stresses (MPa). Full factorial (2-level) method was utilized to design the experiments and develop the empirical models for weld strength, distortion and residual stress. The results were analyzed using ANOVA technique and numerical optimization was utilized for selection of best values for the four parameters for sheet thickness of 3, 4 and 5 mm respectively. In addition to 2-level factorial design, a response surface method (RSM) design was applied to quantifying the effect of sheet thickness as a variable with other four factors (i.e. total five factors whereas 04 numeric factors (thickness, welding current, voltage, and weld speed) and 01 (Trailing) categoric factor) on the responses (weld strength, distortion and residual stress). Four parameters were found to be the most influential welding parameters upon weld strength. Numerical optimization suggests that weld strength can be maximized if the TIG welding is done at low values of welding current and voltage, high value of welding speed with the application of Ar shielding as trailing. Three parameters were found to be the most influential welding parameters upon distortion, while the Ar trailing application was found slightly influential. Numerical optimization suggests that distortion can be minimized if the TIG welding is done at low values of welding current and voltage, high value of welding speed with the application of Ar trailing. Further, four parameters were found to be significant upon residual stresses and these can be minimized with low heat input with application of Ar as trailing. It was also observed that the application of Ar as trailing improves the welding process by enhancing the weld strength along with decrease in distortion (upto max. 30%) and residual stresses (upto max. 15%). The increase in thickness from 30% to 65% (i.e. 3-5 mm thickness) result decrease in distortion about 15% to 30% and 20% to 25% in residual stresses respectively. The range of increase in weld strength is 10-15% (i.e. 690-790 MPa) only whereas the reduction in distortion is three times (2.2 - 7.2 mm) and two times in residual stresses (335 – 608 MPa) respectively.

97

University of Engineering & Technology, Taxila-Pakistan

Further after applying RSM and taking thickness as variable parameter i.e. total five parameters (four numeric (thickness, current, voltage & speed) and one categoric (trailing)), it was observed that with the increase of thickness the residual stresses and distortion decreases and weld strength increases respectively. At end of the chapter empirical models for weld strength as well as for distortion and residual stresses, in terms of all the significant numeric input parameters were presented. Numerical optimization provided the response values and predictors levels as per the desirability. It was observed from the comparison of responses (weld strength, distortion and residual stresses) results that the low distorted with low residual stresses samples give the high weld strength and vice versa respectively.

98

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

CHAPTER 4 FE MODELING & SIMULATION OF GTAW PROCESS OF THIN-WALLED STRUCTURE FOR CIRCUMFERENTIAL WELDING 4.1

Introduction

In this chapter, FE modeling & simulation methodology, analytical model of arc welding, FE formulation, heat source modeling, heat losses modeling, material modeling, filler metal deposition, simulation approach in ANSYS®, welding simulation numerical aspects used for the development of FE models of thin walled structure for circumferential welding of cylinder are presented for the studies of welding induced residual stresses and distortions followed by FE discretization, other simulation aspects, thermal effects of welding, welding residual stress fields and welding distortions. Further, the details of experimental setup for validation of FE models for circumferential welding of thin walled cylinder are presented. 4.2

FE Modeling & Simulation Methodology

The traditional manufacturing processes were mainly established from trial-and-error experiments approach. Such trial-and-error procedure approach requires tremendous material, energy, labors, as well as produces significant waste, fumes and emissions. The traditional trial and error approach based on costly and time-consuming welding experiments faced hindrance to sound welds due to welding process parameter’s optimization. The appropriate control techniques are mandatory with reliability and cost effectiveness for the application of arc welding process on shop floor level. Whereas the welding simulations simulates an actual welding process based on science and physics and the tests can be performed inside computer without the wastage of resources and hazardous environment impact. A hybrid approach involving both FE modeling and experimental work has proven very beneficial. FE models provide a very suitable tool for analyzing the thermal and mechanical consequences of welding process. FE simulations of welding processes have been a major topic in welding research for several years [141]. The availability of 64-bit high performance computing machines and enhanced FE computational techniques has made it possible to simulate temperature fields developed from the welding process. FE models allow a variety of welding process and heat source parameters studies without considering the practical limitations. FE models can be used for analysis of temperature and stress & strain during and after the welding for the improvement of the process with the experimental validation of FE models. The FE models after validation can be used for welding process optimization by performing virtual design of experiments (DOE). Many commercially available finite element codes such as ANSYS®, ABAQUS®, FEMLAB®, ADINA®, MSC MARC®, and SYSWELD® etc. can be applied to carryout such type of manufacturing processes simulations. However, in expert opinion there is no single model available to realistically account for the arc physics, weld pool phenomenon and finally the deformations and heat conduction in the solid model [2].

99

University of Engineering & Technology, Taxila-Pakistan

The some subroutines are integrated in ANSYS® for the facilitation and the coding comprises elements activation and deactivation functionality, meshing algorithm, heat source modeling, material models, heat flux distribution for analytical model and material properties management depending on the elements temperature etc. Transient thermal analysis followed by elasto-plastic stress analysis is applied to simplify the simulation procedure. In the following section, the simulation strategy for the transient thermal modeling using code(s) from commercially available, analytical model for temperature fields during arc welding simulation, thermal properties of the materials, adaptation of heat source modeling and the filler metal deposition mechanism are presented in detail. 4.2.1

Analytical Model of Arc Welding

Arc welding is a highly non-linear coupled thermo-mechanical phenomenon in which localized heat generation and large thermal gradients results due to the moving heat source and consequently thermal stresses and distortions due to the non uniform temperature distribution. According to the first law of thermodynamics, the energy is conversed. Application of this to a differential control volume “V”, the heat conduction equation ignoring the heat of deformations is given in Equation 4.1 [4]. ρ (T ) c (T )

∂T ( x , y , z , τ ) ∂τ

= ∇.q + Q ( x, y , z ,τ )

(4.1)

Q ( x, y, z,τ ) , is the heat generation per unit volume.

The constitutive equation is the Fourier law of heat conduction as given by Equation 4.2 which relates the heat flux and the temperature distribution. q = − k (T ). A.∇T ( x, y, z ,τ )

(4.2)

From Equation 4.1 into Equation 4.2 ρ (t )c(T )

T ( x, y, z ,τ ) + Ñ (k (T ). A.ÑT ( x, y, z ,τ )) = Q( x, y, z ,τ ) T

(4.3)

Considering thermal conductivity as constant, then ρ (T )c(T )

∂T ( x, y, z ,τ + K (T ). A.∇ 2T ( x, y, z,τ ) = Q( x, y, z ,τ ) ∂τ

(4.4)

Temperature distribution is governed by Equation 4.4. The sum of all the forces and moments acting on a body is zero according to the law of equilibrium. It can be written mathematically as Equation 4.5. ∂ 2ui 1 ∂σ v = + Fl ∂t 2 p ∂x j

(4.5)

100

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Where i = 1, 2, 3 The stress-stain relationship in terms of Lame’s constant is given by Equation 4.6 for linear thermo-elastic problems.

σ v = δ v λε kk + 2µευ − δυ (3λ + 2µ )αT

(4.6)

The strain displacement relationship is given by Equation 4.7. 1 ∂ui ∂u j ) + 2 ∂x j ∂xi

ευ = (

(4.7)

From Equations 4.6 and 4.7 into Equation 4.5 and simplifying into Equation 4.8. ∂ 2ui ∂ε ∂T = (λ + µ ) kk + µ∇ 2ui − (3λ + 2 µ )α + Fl 2 ∂t ∂xi ∂x j

(4.8)

∂T provides a coupling between Equations 4.4 and 4.8. The ∂x j temperatures are calculated from Equation 4.4 and applied as body loads through ∂T (3λ + 2µ )α in Equation 4.8. From displacements, the strains and stresses are calculated. ∂x j

The term (3λ + 2µ )α

4.2.2

FE Formulation

Temperature distribution for isotropic material given in Equation 4.4 can be written in the form as given in Equation 4.9 [4]. ρC

∂T ∂ ∂T ∂ ∂T ∂ ∂T = (K ) + (K ) + (K )+Q ∂t ∂x ∂x ∂y ∂y ∂z ∂z

(4.9)

It can be written in matrix form as ρC

∂T = ( L )T ( D { L} T ) + Q ∂t

Where,

∂    ∂x  ∂ L=   ∂y  ∂    ∂z 

(4.10)

and

K  D= 0 0 

0 K 0

0  0 K 

101

University of Engineering & Technology, Taxila-Pakistan

Applying convective boundary conditions at the surface enclosing the volume “V”, it can be written as given in Equation 4.11.

[ q ]T η = h f (TB − T )

(4.11)

Equations 4.6 becomes by taking the boundary effects into account as ρC

∂T = {L}T ([ D ]{L}T ) + Q + h f (TB − T ) ∂t

(4.12)

Equation 4.10 when multiplied by δ T yields and integrated over the control volume using the boundary conditions ∂T T ∫ (δTρC )dv + ∫ (δT{L} [D]{L}T)dv = ∫ (δTQ)dv + ∫δThf (TB −T)dA ∂ t v v v

(4.13)

Let for any element E, the temperature can be represented by

and

T = [ N ]TE

(4.14)

δ T = [ N ]δ TE

(4.15)

Where, TE is the nodal temperature and [N] is the matrix of element shape functions. This equation is valid for all permissible δ TE . If B = [ L ][ N ]

(4.16)

By substituting Equations 4.14, 4.15 and 4.16 in Equation 4.13 ρ ∫ (C[ N ][ N ]T {t}) dv + ∫ ([ B ]T [ D ][ B ]{TE }) dv = ∫ [ N ]Qdv + ∫ [ N ]h f (TB − [ N ]T {TE })dA v

(4.17)

A

Equation 4.17 containing nodal temperatures can be written in another form as [4]: [C ]{TE } + [ K ]{TE } = {FE }

(4.18)

Where, [C ] = ρ ∫ C[ N ][ N ]T )dv

Specific heat matrix

v

[ K ] = ∫ ([ B ]T [ D ][ B ])dv + ∫ h f [ N ][ N ]T dA v

Thermal conductivity matrix

A

{FE } = ∫ Q[ N ]dv + ∫ h f TB [ N ]dA V

Heat generation and convection matrix

A

In thermal analysis, the temperature fields can be obtained from the Equation 4.14. These results can be used further for structural response in mechanical analysis. By assembling the individual elemental equations, a system of equations is obtained and solved 102

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

for the unknown nodal temperatures TE. By using the principle of virtual work, which states that a virtual change of internal strain energy must be offset by an identical change in external work due to applied load, the finite element form of Equation 4.8 can be obtained and it can be written mathematically as

δU = δ P

(4.19)

U = Internal strain energy (internal work) P = External Work (inertia effect)

Where,

δ = Virtual operator

Whereas the virtual strain energy is given as

δ U = ∫ {δε }T {σ }d {V }

(4.20)

v

ε = Strain Vector, σ = Stress Vector, and V = Volume of element From the theory of basic solid mechanics

and

σ = Dε el

(4.21)

ε = ε el + ε th

(4.22)

Where,

ε = total Strain, ε el = Elastic strain, ε th = Thermal Strain, D = Material Stiffness The thermal strain vector for an isotropic medium with temperature dependent coefficient of thermal expansion is given as:

ε th = ∆Tα (T )

(4.23)

∆T is the difference between the reference temperature and actual temperature.

By substituting Equations 4.20 and 4.21 in Equation 4.19 gives

δ U = ∫ {{δε }T [ D ]{ε } − {δε }[ D ]{ε th }) dV

(4.24)

The strain is related to nodal displacement by the following relations {ε } = [ B ]{u}

(4.25)

For a constant displacement, virtual straining energy is given as:

δ U = {δ u}T ∫ [ B ]T [ D ][ B ]dV {u} − {δ u}T ∫ [ B ]T [ D ]{ε th }dV V

(4.26)

V

103

University of Engineering & Technology, Taxila-Pakistan

The external virtual work due to inertia forces is given as:

δ P = − ∫ {δ w}T v

{F a } dV v

(4.27)

Where w = displacement vector of a general point, {F a } = acceleration force vector According to Newton second law of motion {F a } ∂2 = ρ 2 {w} v ∂τ

(4.28)

If the displacement within the element is related to nodal displacement by

{w} = [ N ]{u}

(4.29)

The Equation 4.27 can be also written as ∂2 δ P = −{δ u} p ∫ [ N ] [ N ]dV 2 {u} ∂τ T

(4.30)

By substituting Equations 4.26 and 4.30 in Equation 4.17 {δ u}T ∫ [ B ]T [ D ][ B ]dV {u} − {δ u}T ∫ [ B ]T [ D ]{ε th }dV V

= −{δ u} p ∫ [ N ]T [ N ]dV

∂2 {u} ∂τ 2

(4.31)

{δ u}T Vector is a set of arbitrary virtual displacement common in all terms, the condition required to satisfy Equation 4.30 gives [4]. [ K c ] − {Fcth } = [ M c ]{u&&}

Where, [ Kc] = ∫ [ B ]T [ D ][ B ]dV

(4.32) Element stiffness matrix

V

{Fcth } = ∫ [ B ]T [ D]{ε th }dV

Element thermal load vector

V

[ M c ] = ρ ∫ [ N ]T [ N ]dV

Element mass matrix

104

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

4.2.3

Interaction of Different Fields

The structural modeling is carried out to study the macroscopic thermal-mechanical behavior of thin-walled cylinders during circumferential welding. The computational domain of the finite element model is only limited to the thin walled structure (i.e. cylinders) i.e. the heat source and clamping fixtures explicitly are not modeled. Further, the finite element model does not predict weld pool geometries that are used as input parameters. Therefore, this model is most reliable outside the weld pool. The thermal field is the motive force behind the changes in mechanical and material fields during welding process. Numerical simulations that are concerned with the mechanical effects of welding require the computation of thermal and mechanical fields. Due to the changing microstructure, the material behavior depends on the temperature and deformation histories. In the structural analysis, stresses and deformations based on temperature are accommodated by incorporating the results of thermal-metallurgical analysis into the structural analysis.

4.2.4

Heat Source Modeling and Efficiency

The problems of residual stresses, distortion, and reduced strength of structures in and near the welded joint are a major concern of the welding industry which is in result directly from the thermal cycle caused by localized intense heat input [51]. Computing the transient temperature fields accurately is the first critical step of creating an efficient welding simulation strategy because the temperature has a first order effect on the microstructure, strain, stress and consequently defects formation in the welds and has a second order effect on the temperature fields [142]. The welding induced imperfections are believed to be due to non-uniform temperature fields arising during the welding and this phenomenon is even more significant in arc welding process like GTAW [143, 144]. Weld induced residual stresses and deformations are highly dependent on transient temperature gradients, a function of the total heat input, and the patterns of heat distribution within the weldments, a critical requirement to determine the temperature gradients in the weldments realistically. Therefore, an accurate moving heat source modeling is mandatory to analyze the exact temperature distributions and accordingly the weld induced imperfections like residual stresses and deformations etc. A double ellipsoidal moving heat source model to incorporate the volume heating was presented by Goldak et al. [51, 52]. The size and shape of the moving heat source can be easily modified to model both the shallow and deep penetrating welding processes. Initially, Goldak presented a semi-ellipsoidal heat source model in which heat flux was distributed in a Gaussian manner throughout the heat source’s volume. The temperature gradients predicted by using this heat source model were less steep in front of the arc and steeper behind the arc as compared to experimental observations. Therefore, a double ellipsoidal heat source model was presented to overcome this problem and here, in the present research, the double ellipsoidal heat source model is opted to model the heat input from the welding torch. In Goldak double ellipsoidal heat source model as shown in Figure 4.1, the front half of the source model is the quadrant of one ellipsoid and the rear half is the quadrant of another ellipsoid. The specific mathematical equation is shown in the following. 105

University of Engineering & Technology, Taxila-Pakistan

The power density distribution of the front half is given in Equation 4.35 [51].

q( x, y, z) =

(−

6 3ηQf f

e

a f bcπ π

3x2 3 y2 3z2 − − ) b2 c2 a f 2

(4.35)

Whereas the power density distribution of the rear half is given in Equation 4.36 [51]. q( x, y, z) =

6 3ηQfr

ar bcπ π

(−

e

Where,

3x2 3 y2 3 z2 − − ) b2 c2 ar 2

Q = VI

(4.36) f f + fr = 2

and

Where af, ar, b c are the shape parameters, qo is the effective heat input, ff and fr are the fractions of the heat deposited both in the front and rear half, and all are the heat input parameters. The 6√3η is heat flux distribution parameter that characterizes the concentration level of heat flux distribution based on the heat flux concentration level or heat flux distribution feature of a welding method to determine its value. The shape of the volume of the power distribution can be selected by varying the parameters af, ar, b and c. By this way, the geometry of the experimental fusion zone can be achieved. From experiments, the data for weld pool geometries can be obtained [145]. However, the methods for estimating the weld pool dimensions for arc welding suggested by Christensen et al. [146] can be used upon the unavailability of such data. A good agreement between actual and modeled weld pool sizes, if the modeled heat source size is approximately 10% smaller than the experimental weld pool size, was presented by Goldak et al. [52]. Further, in the absence of better data, the distance in front of the heat source equal to one-half of the weld widths and the distance behind the heat source equal to twice of the width give better approximations [51]. Y Heat flux (watt m-3) Z

b c af ar

X X

Fig. 4.1 Goldak's double ellipsoid heat source model for welding heat source [52] The parameters for the Goldak model were derived from the experiments and values are given in the Table 4.2 in the respective section. The problems may occur by slight changes in the welding conditions due to lack of a good physical background between the parameters from Goldak’s model. Therefore, a new setting is to be required for every set of 106

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

welding parameters and material composition. Therefore, the model cannot predict any fusion zone geometries that can be only back-calculated. However, the good results were obtained from experiments for the weld pool geometry with estimations. In the heat source geometry, very small changes affect the temperatures very close to the weld pool and in the resultant, the calculated temperature history in and near the weld pool are less accurate. This accuracy affects due to the uncertainties in the heat source geometry as well as in the high temperature thermal material properties and also due to the convective and radiative boundary conditions. In the FE model, the origin of the coordinate system is located at the center of the moving arc in order to simulate the welding torch (heat source) movement with the respective welding speed. To calculate the centroidal distance of elements from the moving arc center corresponding to the welding arc position at any instant, a user-subroutine in APDL is utilized. The welding process parameters and the characteristics of the heat source transient heat fluxes representing the moving of the distributed heat source can be calculated at respective positions in welding areas based on the FE mesh generated by the ANSYS®. It is assumed that the heat source moves through volume and the calculated heat applied to elements is volumetric heat generation. Mainly few researchers reported that the heat introduces into the work-piece from the surface under welding current of 200 amperes due to lack of turbulent motion into the work-piece [147]. Whereas, Kermanpur, Shamanian and Yeganeh [91] proved that the heat flow function depends upon the welding current as well as on the work-piece thickness. For circumferential welding, a modified double ellipsoidal heat source model in cylindrical coordinates is used. For better approximation of the weld pool, the use of superimposed four ellipsoid quadrants is modeled as per recommendations of Goldak. The modified double ellipsoid model used for circumferential welding with a single scalar controlling parameter is given in Equations 4.37 and 4.38 [4].

qf =

6 3 M (r , z ) Q f f

π π a f bc

−3{

e

r 2θ 2 af

r 2θ 2

6 3 M (r , z ) Q f r −3{ ar 2 qr = e π π ar bc

2

+3

+3

z2 b

2

+

Ro2 + r 2 -2 rRo c2

z 2 Ro2 + r 2 -2 rRo + } b2 c2

}

(4.37)

(4.38)

The value of scalar multiplier M ( r , z ) can be recalculated iteratively to match the weld pool shape and dimensions. The origin of the coordinate system is selected at the centre of the heat source for the calculations of spatial heat distribution from the Equation 4.37 and Equation 4.38 and a user subroutine provides the movement of heat source at the defined welding speed. To calculate the centroidal distances of elements from the centre of moving heat source at every load step, another subroutine is used. The heat input to the elements from the heat source parameters and welding process parameters is calculated and projected through the thickness of finite element mesh. 107

University of Engineering & Technology, Taxila-Pakistan

A wide range of variation for the value of heat source efficiency in GTAW process is shown in literature [148]. The temperatures with thermocouples at a specified location are measured and compared with the related values for different efficiencies. Finally, it can be concluded that the efficiency of heat source between the range 75% to 80% provides the best agreement with the experiments. Therefore, an efficiency of 70% to 80% in different studies is used. The simulations for circumferential welding are performed with the assumption of perpendicularly focused welding torch as well as during the experimentation.

4.2.5

Heat Losses Modeling

By means of convection and radiation, a considerable amount of heat is lost through the surfaces of cylinder during the welding process. Except the symmetry surface, the convection and radiation to the environment from all the exposed surfaces is included in the thermal boundary conditions. Figure 4.2 shows the schematic representations of thermal boundary conditions i.e. heat losses from the cylinder surface by both convection and radiation. At higher temperatures, the radiation losses are significant near the weld zone whereas the convection losses are significant away from the weld line at low temperatures. The heat lost is calculated for all heat dissipating surfaces by Equation 4.39 and Equation 4.40 [4]. q loss = q convestion + q radiation

(4.39)

q loss = htotal x A(T − Tamb )

(4.40)

Where A = surface area, T = current temperature at the cylinder surface, Tamb = ambient temperature and htotal = combined convection and radiation heat transfer coefficient, given by the Equation 4.41. 2 htotal =[hconvection + ε em σ bol (T + Tamb )(T 2 + Tamb )]

Where, hconvection ε em σ bol

= = =

(4.41)

Convective heat transfer coefficient (Wm-2K) Radiation emissivity of cylinder surface Stefan-Boltzman constant (5.6703 x 10-8 Wm-2K-4)

In addition to convection and radiation heat losses, some researchers [149] refer that contact heat losses also play a role whereas the other researchers [150] ignore the contact heat losses. However, the contact heat losses are ignored in the present research due to point contact of the clamping fixtures with cylinder through the bolts away from the weld line. There are two unknown parameters, the convective heat transfer coefficient (hconvection) and the radiation emissivity (εem), in the Equation 4.41. The radiation heat losses play a major role in high temperature zones and become in significant in low temperature zones. Whereas the convection heat losses play a major role at lower temperatures only. Mostly the handbooks on heat transfer have listed tabulated data for temperature dependent emissivity for several materials [151, 152]. However, the emissivity is strongly dependent on the surface conditions of the metal from εem < 0.1 (for un-oxidized surfaces) to εem = 0.8 (for 108

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

oxidized surfaces) [153]. The value of the emissivity is estimated during the welding due to a constant changing of the temperatures and surface conditions of a metal surface. The temperature dependent emissivity for common materials including AISI stainless steels and low carbon steels is listed by many internet resources [154, 155]. The emissivity value (εem = 0.51) is used on the surface of the steels in the present research, which is the average value for hot rolled steel plates [156].

CL

Convection boundary conditions Thermal Symmetry boundary conditions Heat source

Fig. 4.2 Schematic representations of thermal boundary conditions In Equation 4.41, the convective heat transfer coefficient is also temperature dependent. Generally, the temperature dependency of the convective heat transfer coefficient is much lower than the radiative heat transfer coefficient but the radiative heat transfer dominates the convective heat transfer at higher temperatures. Therefore, it is assumed that there is no benefit of using a temperature dependent convective heat transfer coefficient [157]. A convective heat transfer coefficient (hconvection) of 7 - 12 Wm-2C-1 is recommended by the researchers [158-161]. For HSLA steel as in this research work, a temperature dependent heat transfer coefficient is used as shown in Figure 4.3.

4.2.6

Material Modeling

In arc welding simulation, the FE heat transfer analysis requires the precise values of thermal conductivity, material density, specific heat and latent heat of fusion upto melting point. The values of low temperature materials are mostly available in different literature like [162-164], but the higher temperature values of materials in the published literature are limited. Therefore, the material values at elevated temperatures are required to be estimated or extrapolated from the available low temperature data. In the present research work, a high strength low alloy steel (HSLA) is used for the experimental investigations and the corresponding finite element predictions. The temperature dependent properties are taken 109

University of Engineering & Technology, Taxila-Pakistan

from the published literature. The weld dilution effects on the property changes are not considered and the materials are assumed as homogenous and isotropic. The materials temperature dependency is fully considered during the welding (a temperature range from room temperature to materials melting point) and prescription of temperature independent properties induces significant errors in the predicted results [46].

4.2.6.1 Material Model for High Strength Low Alloy Steel The temperature dependent thermo-mechanical properties such as conductivity, specific heat and density, and temperature-dependent thermal–structural properties including Young’s modulus, Poisson’s ratio, thermal expansion coefficient, yield strength and strainhardening rate are used for thermal analysis and mechanical analysis, respectively. The analysis and optimization of welded structures of High Strength Low Alloy (HSLA) steel is carried in the present research work. The chemical composition, both from literature [137, 138] and in-house spectroscopic measurement of HSLA under investigation is tabulated in Table 3.8. The standard room temperature properties are readily available from published literature [137, 138]. But the high temperature material data is not available in open literature. Due to scarcity of elevated temperature material properties data, an engineering approach proposed and successfully implemented by ZHU and CHAO [165] is adopted in the present research. It is proposed by the author's that "except for the yield stress, using material properties at the room temperature gives reasonable predictions for the transient temperature fields, residual stress and distortion". Based on this expert opinion the following simplifications are introduced in the material model. • Temperature-dependent thermo-physical and thermo-mechanical (except yield strength) properties of low alloy steel (AH36) grade, previously implemented and closely co-related with the experimental data by [4, 166] are used in the simulation work. For the experimental work, HSLA steel grade with chemical composition previously shown in Table 3.8 is used. • For the yield strength of HSLA, the engineering approach from Zhu and Chao suggests the use of "simplified properties constituted by a piece-wise linear function with temperature for the yield stress for computational weld simulation". Based on recommendations, temperature dependent yield strength of the material from [165] is employed, as shown by Equation 4.42, 4.43 & 4.44 below. (0 ≤ T ≤ 100 oC)

YS = YSRM YS = 5% x YSRM + YS = 5% x YSRM

Tmelt2/3-T x Tmelt2/3-100

95% x YSRM (100 < T < Tmelt2/3) (T ≥ Tmelt2/3)

(4.42) (4.43) (4.44)

Where, YS, YSRM and Tmelt2/3 are yield strength, yield strength at room temperature and 2/3 of melting temperature of HSLA respectively.

110

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

• Similar values of yield strength for both base and weld metal are used. For the experimental work, HSLA steel grade with chemical composition previously shown in Table 3.8 is used. Thermal properties i.e. specific heat and thermal conductivity and density as a function of temperature for HSLA adopted from [166] are given in Figure 4.3. For specific heat, latent heat associated with low temperature solid-solid phase transformation is accounted for both the base metal and the weld metal. Enthalpy formulation is used to avoid the numerical non-convergence. 247 KJ/Kg1 oC-1 of latent heat for solidliquid phase transformation is distributed over the melting and solidification range i.e. between solidus and liquidus temperatures. Due to similar material i.e. low alloy steels, the solidus and liquidus temperatures are taken as 1440oC (1713K) and 1560oC (1833K) respectively. The rmal C onducti vity x 125 (W /m-ºC ) Spe ci fi c He at x 1000 (J/Kg-ºC ) C onve ctive C oe ffe cie nt x 10 (W /m²-ºC )

1.0 0.8 0.6 0.4 0.2 0.0 0

600

1200

1800

2400

3000

o

Temperature ( C)

Fig. 4.3 Thermo-physical properties of HSLA steel To model the weld puddle and to compensate for enhanced convective heat transfer effects caused by the fluid flow within the vicinity of weld metal, the thermal conductivity value of 2-5 times of the solidus; at the liquidus is suggested by the previous researchers [167-170]. In this research a factor of 3.55 is used and at the solidus, an artificial increase to120 Wm-1 oC-1 is given to compensate for fluid flow. A constant density of 8096 Kgm-3 (standard value for most the steels) is used. Temperature dependence of thermo-mechanical properties of HSLA steel is shown in Figures 4.4. The main features are given in the following: • Reduction of elastic modulus at high temperature reflects the material softening behavior of at elevated temperatures and the Poisson’s ratio increases with the increase in temperature. 111

University of Engineering & Technology, Taxila-Pakistan

• The elastic modulus reduces to almost zero values at elevated temperatures equal to or greater than the melting temperature. Some recent research work in the similar area [4] shows that numerical instabilities are encountered when excessively low values of elastic modulus at and above the melting temperatures are used. To overcome this issue, a constant value is set to 15.0 GPa. However even lower values upto 1 GPa are reported in the literature [171]. • The values for both the bulk modulus and Poisson’s coefficient are taken constant after melting temperature of 1000oC. • Volumetric changes associated with low temperature solid-solid phase transformations is not taken into account because [75] and [76] reported stress reversals in hoop stresses at weld centerline, in contradiction to the experimental measurements from [76]. However, later studies showed that volumetric changes may give satisfactory results if the transformation plasticity is included which is not included in the present simulation approach. Therefore, for the thermal expansion, the material in the melt and heat affected zones follows material properties of the base metal during heating and cooling. The material behavior of elastic perfectly-plastic is considered whereas the dislocation hardening effect and the effects of creep and transformation induced plasticity is not considered here. A bi-linear kinematic hardening model (von mises yield criterion with associated flow rule, kinematic hardening rule and bi-linear kinematic hardening material) as given in Equation 4.45 [141] is used with σ1, σ2, and σ3 being the three principal stresses. σv =

1 2

[(σ 1 - σ 2 ) 2 + ( σ 2 - σ 3 ) 2 + ( σ 3 - σ 1 ) 2 ]

(4.45)

El astic Modul us x 2.0E+5 MPa Poi sson Ratio x 1 The rmal Expansion C oe ffe cie nt x 2.0E-05 / ºC

1.1 0.9 0.7 0.4 0.2 0.0 0

400

800

1200

1600

2000

o

Temperature ( C)

Fig. 4.4 Thermo-mechanical properties of HSLA steel 112

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

4.2.7

Filler Metal Deposition

In arc welding, modeling of filler metal deposition with metal addition as Gas Tungsten Arc Welding (GTAW) processes is an important aspect for accurate result prediction in numerical analysis. Presently, three different techniques for the filler metal addition are in practice as the deactivated element or the elements rebirth technique, the quiet element technique and the element movement technique. The conventional quiet element technique is used in the FE models in the present research [175-178], which is relatively easy to implement by utilizing the ANSYS® features of element birth and death. All elements are generated in the start including the filler metal elements to be deposited later. The filler elements are not actually removed from the FE model to achieve the element death effect. Instead, the conductivity, stiffness and other analogous material properties are multiplied by a severe reduction factor to deactivate their contribution in the analysis. Although zeroed out of the load vector, element loads associated with the deactivated elements still appears in the element load lists. During the thermal analysis, all the nodes of deactivated elements except those shared with the base metal are also fixed at ambient temperature until the birth of the respective element and deactivated elements are reactivated sequentially when they come under the influence of the heat source i.e. welding torch. For the sequel mechanical analysis, a similar approach is used where the elements to be welded are first assigned a set of artificial, very soft properties. The actual properties of the metal are reassigned as the elements solidify from the weld pool. 4.2.8

Simulation Approach in ANSYS®

For simulation, a coupled thermo-mechanical simulation approach is divided in two sequentially coupled simulations based on weak thermal to structural coupling. For the analysis of the thermal behavior, a transient non-linear thermal analysis is performed i.e. nodal temperature distribution followed by iterative structural analysis. Figure 4.5 shows an overview of de-coupled thermo-mechanical simulation approach in the present research. Figure 4.6 shows the detailed sequentially coupled thermo-mechanical simulation strategy adopted in the research. The quiet element technique is used in this work for the filler metal deposition. At the start, the nodal temperature of the filler metal elements is equal to the ambient temperature. The nodal constraints are removed at that time when the elements sequentially come under the influence of the heat source. The heat source is supposed to stay at least once on each element along the weld line for better computational results [179]. Equation 4.46 gives the appropriate time step used in the analysis to accomplish the task. Load step

=

Total welding time (sec) No. of circumferential elements along the weld line

(4.46)

Therefore, a constant time step is used during the heating phase and the heat source moves with the specified welding speed quasi-stationary. However, the different time steps are used during the cooling phase. As the cooling of the weldments approaches to the ambient or equal to pre-heat temperature (if any), the time steps continuously increases.

113

University of Engineering & Technology, Taxila-Pakistan

THERMAL ANALYSIS Thermal boundary conditions Moving heat source Non-linear material model

TEMPERATURE HISTORIES FROM THERMAL ANALYSIS

THERMAL POST PROCESSING Temperature fields

3D FINITE ELEMENT MODEL Geometry data Meshing parameters

SEQUEL STRUCTURAL ANALYSIS Structural boundary conditions Non-linear thermo-elastic-plastic material model

STRUCTURAL POST PROCESSING Transient and residual stress profiles Transient and residual deformations

Fig. 4.5 Overview of de-coupled thermo-mechanical simulation approach In structural analysis, computing the stress and strain field is a very important aspect of the computational weld mechanics. The localized heating during the welding causes the non linear distribution of temperature field in the weldments. The main cause of welding induced imperfections like transient and residual stresses and distortions after the completion of the welding process is the uneven temperature fields that generate the transient thermal stresses and deformations. In the present research, the sequel non-linear structural analysis is driven by the application of temperature histories as the thermal strain fields controls the stress fields in the welding. Whereas, the critical link between the thermal and mechanical analysis is the application of nodal thermal history from the transient thermal analysis as the nodal body loads in the structural analysis. The thermal histories computed at all nodes in the thermal analysis are recorded and stored in a thermal analysis result file that is read by using the LDREAD command in the structural analysis for the mapping of the thermal histories onto the nodes in the structural model. For proper data mapping, it is necessary to use ISO meshed with same element topology (8-noded brick elements) in the thermal and structural models. The similar transient fashion from thermal analysis is replicated iteratively by these mapped nodal temperature fields at different times. The similar load steps from thermal analysis are also used for the corresponding structural load steps. Upon solidus temperature, the elements of that particular bead are activated. To track the averaged peak temperature of each element of filler metal and record the time for each element upon attaining the solidus temperature after the peak temperature the thermal analysis, a user APDL subroutine is used. Further in order to simulate the mechanical strains relaxation behavior, the strain history of each element at the time of activation is initiated.

114

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Geometric parameters

Solid Modeling

• Heat source parameters • Thermal material properties • Meshing parameters • Boundary conditions

Next element

Finite element modeling

Elements centroid calculation

• Meshing parameters • Structural boundary conditions • Structural material properties

Finite element modeling Apply heat source

No

Calculate heat input based on heat source parameters and centroidal distance

Next load step

Yes

Elements properties switching (If applicable)

Elements material record file

Elements activation (If applicable)

Elements activation time record file

Application of heat input

Check data mapping completion

Activation of elements under influence of heat source

No Next element

Yes • Elements average temperature calculation • Elements material number switching • Elements material number record file • Elements activation time record file

Iterative transient non-linear thermal solution

Yes

Load steps completed Yes

Transient temperature history

Iterative transient non-linear structural solution

No

Load steps completed

No Next load step

Yes

Structural results file (Stress/strain data)

Fig. 4.6 Detailed sequentially coupled thermo-mechanical simulation strategy [180] To model the enhanced heat losses (localized forced convection) due to the trailing argon gas, a cooling media based on the properties of argon gas is introduced at a fixed trailing distance of 25 mm from the that source. The heat exchange coefficient for forced convection is evaluated by using the correlation shown in Equation 4.47 initially proposed by Steen [68] and later also employed by Shah Alam [5]. 115

University of Engineering & Technology, Taxila-Pakistan

hforced = 13Re1/2 Pr1/3Kar/NPD

(4.47)

Where, Re Pr

νar ρar µar Kar Car D NPD

= Reynolds number of argon = D x νar ρar / µar = Prandlt number of argon = µar Car / Kar = Argon flow rate = Density of argon = Dynamic viscosity of argon = Heat conductivity of argon = Specific heat of argon = Nozzle diameter = Stand-off distance of nozzle

Based on nozzle diameter, an APDL subroutine is developed to select the elements for the application of forced convective heat transfer co-efficient for number of iterations based on movement of heating source (welding torch).

4.2.9

Welding Simulation Numerical Aspects

A general purpose finite element code ANSYS® is used for modeling/simulation and welding phenomena is modeled as sequentially coupled transient non-linear thermal-stress analysis in the present research work. There are generally three types of non-linearties in the structural mechanics as geometric non-linearties (large deformations), boundary nonlinearties (contact), and material non-linearties (hyper-elasticity, plasticity, creep, anisotropy etc.) The incorporation of geometrical non-linearties (NLGEOM, ON) into welding simulations can introduce ill conditioned matrix, resulting in numerical non-convergence issues. Kinematic non-linearties are not included by using small displacement formulation in modeling. No contact/target elements are used in the present research, therefore in most of the work non-linearity due to contact opening and closing between the contact and target elements is not present. Material model with temperature dependency are utilized and properties used in different studies are, however, highly non-linear and are the major source of non-linearties in finite elements studies. Iterative incremental Newton-Raphson (NR) scheme was used to solve the system of equation. The use of this scheme is also essential in the software to adopt quiet elements technique for filler metal deposition. The opted “FULL” NR scheme updates the stiffness matrix at every, equilibrium iteration and thus shows more flexibility to incorporate non-linear behavior of material properties. Though more frequent updating of stiffness matrix needs larger matrix formulations and inversions but gives relatively fast convergence [174]. The matrices obtained from finite element formulations are usually sparsely populated. Therefore, the system of simultaneous equations is solved by using direct sparse matrix solver (elimination solver). Since linear elements are used in the both the thermal and structural analysis, during structural analysis average temperature at the element centroids is used to calculate constant thermal strain within each element. It is considered essential to avoid inconsistency between the thermal strain and displacement strain fields because temperature field directly becomes the thermal strain in mechanical 116

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

analysis. The alternate is to have one degree higher finite element shape function in structural analysis in comparison to thermal analysis. Similarly, when ever elements with linear shape function are in use volumetric strain should be under integrated to avoid undesired locking. Such locking may results in relatively small deformation but significantly high stresses and excessively large computational time. In the present work, reduced integration is performed at the element centroid. However, in contrast reduced integration in solid elements may cause the elements more prone to the zero-energy modes. These modes, commonly referred to as hourglass modes, are oscillatory in nature and tend to have periods that are much shorter than those of the overall structural response. They typically have no stiffness and give a zigzag appearance to a mesh, therefore should always be minimized. Hour-glassing is controlled by adding artificial elastic stiffness to the model. To enhance the convergence, different options available within the ANSYS® such as line search (LNSRCH), adaptive decent, ramped and stepped load (KBC, 0/1) are used. The LINE-SEARCH option attempts to improve a NR solution by scaling the solution vector by a scalar value termed the LINE-SEARCH parameter at the start of equilibrium iterations. The scalar multiplier is automatically determined by minimizing the energy of the system which reduces to find the zero of the nonlinear equation. An adaptive descent is a technique which switches to a stiffer matrix if convergence difficulties are encountered, and switches back to the full tangent as the solution convergence, resulting in the desired rapid convergence rate [174].

4.3

Welding Induced Stresses and Distortions

To produce high strength welded structures, arc welding is mostly used as an effective joining method enabling the welding community around the globe. The thermal stresses occurs in the weld zone and the adjacent areas producing significant residual stress fields due to the non-uniform expansion and contraction of the weld metal and surrounding base metal by heating and cooling cycles during the welding. These high magnitude residual stresses of the order of yield strength of the material within the heat affected zone (HAZ) can be a major threat for the in-service structural integrity of welded structures [181]. The strains produced due to the welding during the heating phase always induce plastic deformation of the metal and in a result of these strains internal forces produced that cause a variety of welding distortions. The shortening strength issue of the structures is a major challenge of the welding industry for decades due to the residual stresses issues in and near the weld zone, and due to poorly fabricated and distorted structures. Therefore, the accurate prediction of transient and residual stress fields and distortions patterns is of critical importance to ensure the in-service structural integrity and reliability of these welded structures. To predict the magnitude and trends of residual stress fields is a complex phenomenon due to the involvement of various factors including short term localized heating and rapid cooling, moving heat source, temperature dependent material behavior and metallurgical transformations. Therefore, the FE based numerical simulations attained a considerable importance for the prediction of adverse consequences of complex welding phenomenon in the last three decades [19, 182-183]. For the analysis of residual stress fields in circumferentially welded structures focusing on pipe and cylinders, a significant contribution is available in literature [184-192]. Mostly, the previous researcher [185-188] reduces 117

University of Engineering & Technology, Taxila-Pakistan

computational power requirements by simplifying with the assumptions of such as rotational symmetry and lateral symmetry in numerical simulations as the computer simulation of welding processes is highly intensive with the requirement of a large computer storage and CPU time. However, these assumptions and simplifications were made to reduce the computational demand but at the cost of results accuracy because the model was over simplified by limiting the solution domain to only a section of the whole with forced symmetry assumptions. Furthermore, these simplified assumptions are not capable to cover the considerable effects of weld start, stop and weld tack modeling. In this regard, an experimental work by Jonsson and Josefson [192] and some three-dimensional finite element studies [189, 190, 193], reported the deviations from rotational symmetry specifically at the beginning and end of the welding cycle for circumferential joint in welding of pipes with lateral symmetry. Dong and Burst [194] and Dong [195], presented that both the moving heat source and weld start and stop effects are inclined to violate the axis-symmetric conditions and the circumferential variation in residual stresses was presented by the authors to strengthen the statement. Later Fricke et al. [196] presents that residual stresses are by no means axis-symmetric by using a full 3D model for multi-pass welding of pipes. The detailed three-dimensional FE models to get insight of this complex phenomenon are still lacking and needs to explore although various three-dimensional FE based numerical investigations are available in the published literature for the circumferential welding of pipes or cylinders. Therefore, initially the investigations in this chapter focusing on the study of thermal and residual stress patterns and the variation in both transient and residual axial and hoop stresses in circumferentially welded cylinders are discussed. Further, the estimation of transient and residual deformation patterns is also discussed. It is supposed that these initial studies based on the FE models developed to get in depth the evolution of stress and deformation patterns in this chapter, as the same modeling and simulation strategy will be further utilized in parametric studies to investigate the different aspects of arc welding phenomenon and then for the optimization purpose.

4.3.1

FE Discretization

For the circumferential welding of two cylinders, a full three-dimensional FE model along with finite element detail with "V" groove developed in ANSYS® is shown in Figure 4.7 [197]. The element types used in modeling are SOLID70 (linear 8-node brick element with one degree of freedom, i.e., temperature at each node) and SOLID45 (linear 8-node brick element with three degrees of freedom at each node: translations in the nodal X, Y, and Z directions.) for thermal analysis and structural analysis respectively. Further details about these elements can be found in [198]. A relatively fine meshing of elements is used within a 10 mm distance on both sides of the weld line (WL) due to high temperature and flux gradients expected in and near the fusion zone (FZ) and heat affected zone (HAZ). The element size increases with an increase in the distance from the WL away from the HAZ. The element size in the weld direction is kept constant equal to 1.96 mm through out the circumference. The element size in transverse direction is used of 1 mm upto 10 mm distance on each side of the weld line (WL) i.e. HAZ area whereas the element size increases with the increase in distance away from weld region. Three elements of 1 mm each in size are used in thickness direction for "V" groove modeling. The two tack welds on the start i.e. 0º and 180º

118

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

"V" groove

Tack Welds

WL

Root Opening Total No. of nodes: 71040 Total No. of elements: 54720

Fig. 4.7 (a) 3D FE mesh based on sensitivity analysis. (b) "V" groove, tack weld and root opening in FE model [197] 2100

o

Temperature ( K)

1864

1836

1900 1730 1700

1860

1865

1790 1525

1500

1655

1300 1305 1100 1009 900 29,000

35,000

41,000

47,000

53,000

59,000

Numbe r of e le me nts

Fig. 4.8 Mesh sensitivity analysis based on maximum temperature attained [197] of the weld are modeled comprising each of 4 elements (7.85 mm) in circumferential direction, 4 elements (4 mm) and 2 elements (2 mm) in two layers in thickness direction. The mesh sensitivity analysis based on maximum temperature attained was performed for successive mesh refinements as shown in Figure 4.8 i.e. maximum temperature (1864 K) at 54720 elements. The tack lengths used in the FE models are according to the physical weld sample. The two cylinders should be considered theoretically as separate parts in the model setup because they are independent units until the moving heat source passes over them and join those. But practically these cylinders are tack welded and mechanically restrained before

119

University of Engineering & Technology, Taxila-Pakistan

the welding. Therefore, the cylinders are modeled as single model in FE modeling because they stay stationary relative to each other during the welding.

4.3.2

Other Simulation Aspects

X

300 mm

The detail of modeling and simulation approach is already discussed in the previous sections. For better understanding of the thermo-mechanical results, the other simulation aspects are given in the following section. The same sequentially coupled thermo-mechanical simulation approach and material model is used. The Goldak double ellipsoidal heat source model along with quiet element technique for addition of filler metal is used for heat source and filler material modeling. The modeling of heat losses from the exposed surfaces (inner and outer) by convection and radiation are considered in thermal analysis. The combined heat transfer coefficient is calculated and applied on all the applicable surfaces. The geometric parameters and joint geometry of two cylinders for circumferential welding by GTAW process are shown in Figure 4.9 and Figure 4.10 respectively. The total heating time along the weld path of 300 mm diameter cylinder with a torch speed of 3 mm/s is about 314.16 sec and the complete welding sequence is divided into 480 load steps of 0.65 sec with equally space increment. For effective application of thermal load during the load step, the available stepped load option in ANSYS® is used. An other additional 47 load steps of different time lengths are used for cooling of the weldments after extinguishing the arc. The total cooling time from the start of the cooling phase to the ambient temperature of 300 K is about 1500 sec (i.e. 25 minutes). Only boundary condition is applied as the constraints to represent the clamping of the cylinders under welding on welding positioner in the structural analysis. All the nodes on a cartesian coordinate axis at the positioner end of the cylinders are constrained in axial direction to match the experimental boundary conditions. In addition, two nodes 180o apart at the positioner end are also constrained in axial radial and circumferential directions for the stability of FE model. The welding process parameters and heat source parameters used in the study are given in Table 4.1 and Table 4.2 respectively. WL 3 mm X Y Z Y Z

X Y Z Legends: X Y

Structural constraints in X, Y and Z

Fig. 4.9 Schematic representations of structural boundary conditions along with geometric parameters [197]

120

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 4.1 Welding process parameters Parameters

Current (I) (amperes)

Voltage (V) (volts)

200

12.5

Efficiency (η) Welding Speed (WS) (%) (mms-1) 75

3

Table 4.2 Goldak heat source parameters Length of ellipsoidal Front (af) Rear (ar) (mm) (mm) 5.0

15.0

Heat source width (2b) depth (c) (mm) (mm) 10.0

3.0

Fraction of heat in ellipsoidal Front (ff) Rear (fr) 1.25

0.75

θ R F

T

R L O Fig. 4.10 Butt-weld joint geometry

4.3.3

Experimental Validation

To conduct the full-scale experiments with proper instrumentation for data measurement for the experimental validation of the developed FE models is a mandatory to ensure the reliability of the developed models before the application on a shop floor. An automatic TIG welding setup with minimum human intervention and skill is also mandatory for the proper validation of FE results due to the possibility of the variations associated with the skill of the operators and rotary synchronization problems in arc welding experiments. A careful data acquisition for proper data measurement, calibrated thermocouples and analysis system are very necessary during the welding experiments for the validation purpose. For the validation of FE models developed, GTAW (TIG) experiments on two thinwalled cylinders with same parameters of geometric and welding process as used in FE models were conducted. The material used is high strength low alloy steel with chemical composition as already given in Table 3.8. Further, argon as shielding gas with 99.999% purity was used with flow rate (25 liters/min). A high-tech fully automatic SAF GTAW welding equipment as shown in Figure 3.4 and Figure 3.5 commercially available along with rotary positioner and required welding fixtures was used to achieve the desired structural boundary conditions. A single pass butt-weld geometry was used with single "V" groove of angle of 90o, 1 mm root face, 1 mm root opening, 3 mm wall thickness and 300 mm length of cylinder as shown in Figure 4.10.

121

University of Engineering & Technology, Taxila-Pakistan

The two cylinders of 300 mm outer diameter, 3 mm wall thickness and 150 mm length were used with placing of two tack welds of ~ 8 mm length at 0o and 180o from the starting position of weld. Further, these tack welds were used to create a root opening prior to welding with the insertion of spacers of 1 mm at some suitable locations during the tack welding and these were removed after the tack welds cooling to room temperature. The areas in and near the tack welds were considered as post weld heated upto 300oC to minimize the effects of initial stress due to tack welds prior to welding. For heating, a conventional gas torch was used with both infrared and touch probe thermocouples for the measurement. However, the stress data was not recorded after the tack welds and the post weld heating and these effects are not considered. Further, the effects of the linear seam weld was not considered as first these cylinders were linearly seam welded after roll forming of sheet metals and stress relieved by heat treatment prior tack welds on cylinders for circumferential welding. o

P1 @ outer surface, 10 mm from WL and 30 from weld start o

P2 @ outer surface, 15 mm from WL and 30 from weld start o

P3 @ outer surface, 10 mm from WL and 90 from weld start o

P4 @ outer surface, 20 mm from WL and 90 from weld start

1000

Temperature (oC)

850 700 550 400 250 100 0

32

64

96

128

160

Time (sec) P1-FE P2-EXP

P2-FE P3-EXP

P3-FE P4-FE

P1-EXP P4-EXP

Fig. 4.11 Comparison of computed and measured transient temperature profiles at four different locations on cylinders outer surface For the sequel structural analysis, the nodal temperature distributions from the thermal analysis are used as a basic input. There is a prerequisite for this purpose whether the experimental data correlation for the FZ and HAZ dimensions or some nodal temperature 122

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

verification for the accurate predictions of subsequent stress fields and distortion patterns. In the present research for the thermal model validation, the later technique is used by using the thermocouples and data acquisition system with computer interfacing. Four thermocouples at different locations were placed and temperatures were recorded through the data logger after the interval of each 10 seconds for the comparison with FE results. Figure 4.11 shows a quantitative comparison of measured and predicted transient temperatures at four thermocouple locations. Three thermocouples at P1, P2 and P3 show a close agreement with the FE data whereas the fourth thermocouple at P4 shows slightly a higher variation of predicted and measured temperatures. However, the overall temperatures are within the maximum variation of about 8% only. The residual stresses are measured at some specified points for comparison through the predicted results for structural model validation,. A centre hole drilling strain gauge method as already discussed in chapter 3 in section 3.3 is used to measure the residual hoop and axial stresses at specified locations i.e. Points P1 to P3 on cylinder outer surface and P4 to P6 on cylinder inner surface. Figure 4.12 shows the the gauge locations from P1 to P6. The further details of the hole drilling residual strain measurement method can be found in [199]. Figure 4.12 shows a quantitative comparison of residual stresses from experiments with predicted data with a good agreement. Figure 4.11 and Figure 4.12 shows the qualitative comparison of nodal temperatures and residual stresses which are an evident that predicted results agreed well with the experimental data showing the experimental validation and the developed FE models can be used further for the research. Legends: o

PX @ Y mm from W L and Z from weld start Where X=1 to 6 Y = 10, 15, 20, 10, 10 and 10 mm for X 1 to 6 respectively Z = 30, 30, 30, 45, 135 and 225 from weld start for X 1 to 6 respectively

200

Stress (MPa)

90 -20 -130 -240 -350 FE

Legends: PX–S = S stress @ point X where X = 1 to 6 S = A (Axial) or H (Hoop) P1-A P1-H P2-A P2-H P3-A P3-H P4-A P4-H P5-A P5-H P6-A P6-H -118.4 70.25 10.24 -36.17 143.2 -63.21 -210.6 115.6 -220 120.5 -226.1 112.2

EXP -88.8 88.42 12.87 -55.01 101 -39.71 -326.7 187.4 -152.9 163.8 -258.7 153.8

Fig. 4.12 Computed and measured residual stress values for different locations at cylinder outer surface

123

University of Engineering & Technology, Taxila-Pakistan

4.3.4

Thermal Effects of Welding

The temperature distributions at four different times (13 sec, 52 sec, 318 sec and 1277 sec) during the welding process of cylinder of 300 mm diameter are shown in Figure 4.13 (a– d). The peak temperatures are observed close to the weld line at the heat source location and steep temperatures gradients ahead of the heat source are observed representing the least heat flow significance ahead of the heat source/welding torch. The cooling phenomenon after the peak temperature achieved is shown by the gradients behind the torch as the torch moves ahead from a certain point. The temperature distribution of the weldments after cooling to almost uniform temperature is shown in Figure 4.13(d) that require some more time steps further to simulate the cooling phase.

(a) t = 13s

(c) t = 318s

(b) t = 52s

(d) t = 1277s

Fig. 4.13 Temperature profiles at four different time steps during the welding process The axial temperature distributions for four different cross-sections from the weld start towards the time progress at different time steps is shown in Figure 4.14 (a–d). The temperature distribution at a section is steep as the arc crosses the section as incase of Figure 4.14 (a), the section is located at an angle of 45o from the weld start position (0o). The welding torch at a speed of 3 mm/s reaches the section after 39.27s around a circumference of 300 x π mm and the maximum temperature is observed at the torch position. As the torch crosses the section, the temperature falls down with slow rate. Figure 4.14 (c) and Figure 4.14 (d) shows the preheating action of the section due to the forward heat flow through the torch just before the torch arrival at a section which is more dominant incase of the sections oriented at 225o and 315o respectively.

124

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

The transient thermal cycles observed at various points at weld line at 5 mm and 15 mm from the weld line and at 0o, 90o, 180o and 270o from the weld start position respectively are shown in Figure 4.15(a–d). When the arc crosses the corresponding section, the thermal cycles shows that temperature at a point reaches a peak value corresponding to that time. The figure shows that a point nearest to the weld line gets heated to a maximum temperature whereas the points away from the weld line show peak temperatures very low. 1800

1800 Te m p-39.27 s e c

1650

Temperature (K)

o

o

Temperature ( K)

Te m p-196.35 s e c

1350

Te m p-39.27 s e c

1650

Te m p-117.81s e c

1500

Te m p-274.89 s e c

1200 1050 900 750 600 450

Te m p-117.81 s e c

1500

Te m p-196.35 s e c

1350

Te m p-274.89 s e c

1200 1050 900 750 600 450

300

300

-85

-68 -51

-34 -17

0

17

34

51

68

85

-85

-68 -51

Distance from WL (mm)

-34 -17

(a)

34

51

68

85

51

68

85

1800 Te m p-39.27 s e c

1650

Te m p-196.35 s e c

1350

Te m p-39.27 s e c

1650

Te m p-117.81s e c

1500

Temperature (K)

Te m p-274.89 s e c

o

o

17

(b)

1800

Temperature (K)

0

Distance from WL (mm)

1200 1050 900 750 600 450

Te m p-117.81 s e c

1500

Te m p-196.35 s e c

1350

Te m p-274.89 s e c

1200 1050 900 750 600 450

300

300 -85

-68 -51

-34

-17

0

17

34

51

Distance from WL (mm)

68

85

-85

-68

-51 -34

-17

0

17

34

Distance from WL (mm)

(c) (d) Fig. 4.14 Axial temperature distributions for four different cross-sections at different time steps from the weld start position.

4.3.5

Welding Residual Stress Fields

4.3.5.1 Axial Residual Stress Fields The stress normal to the direction of the weld bead is called as the axial stress incase of circumferentially welded cylinders. The tensile and compressive axial stress fields are observed in and near the weld region based on different temperature profiles on the inner and outer surfaces of the cylinders respectively. The different temperature gradients results in tensile and compressive residual stress fields and varying shrinkage patterns through the wall thickness near the weld line on inner and outer surfaces respectively. The axial stress distributions on cylinders outer surface at different cross sections (50o, 90o, 150o and 250o) from the weld start position are shown in Figure 4.16.

125

University of Engineering & Technology, Taxila-Pakistan 1800

1800 0 mm from WL

1400

0 mm from WL

1600

5 mm from WL

5 mm from WL

15 mm from WL

1200 1000 800 600

1400

Temperature ( oK)

Temperature ( oK)

1600

15 mm from WL

1200 1000 800 600 400

400

200

200 0

50

100

150

200

250

300

350

400

0

50

100

150

(a)

250

300

350

400

300

350

400

(b) 1800

1800 0 mm from WL

1600

5 mm from WL 1400

Temperature ( oK)

1400

0 mm from WL

1600

5 mm from WL 15 mm from WL

1200 1000 800 600

15 mm from WL

1200 1000 800 600 400

400

200

200 0

50

100

150

200

250

300

350

0

400

50

100

150

200

250

Time (se c)

Time (se c)

(c) (d) Fig. 4.15 Transient thermal cycles experienced by various points at different cross sections from the weld start position 200 100 Outer Stress (MPa)

Temperature ( oK)

200

Ti me (se c)

Ti me (se c)

0 -100 -200 -300 -400 -90

-72

-54

-36

-18

0

18

36

54

72

90

Distance from WL (mm) Axial @ 50 Deg

Axial @ 90 Deg

Axial @ 150 Deg

Axial @ 250 Deg

Fig. 4.16 Residual axial stresses (MPa) on outer surface at different cross sections from the weld start position 126

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

The compressive residual axial stresses near the weld line approaches to zero after 18 mm on both sides of the weld line on the outer surface of cylinder. The stress reversal from compressive to tensile is observed after 18 mm on outer surface. Almost 70 mm away from the weld line, these low values tensile stresses again approach to a zero value. A constant axial stress value near to zero after 70 mm from the weld line is observed as shown in Figure 4.16. Figure 4.17 shows the high tensile residual stresses on the inner surface near the weld line approaching to zero and then reversing to lower compressive residual stresses at 18 mm same as observed in outer surface. Again these lower compressive residual stresses increasing to almost constant value of zero at 70 mm on both sides of weld line observed for cylinder inner surfaces at different cross sections (50o, 90o, 150o and 250o) from weld start position. The general residual axial stresses distribution shows a similar trend as observed in the previous research [4, 185, 187, 189, 192-193, 197].

500

Inner Stress (MPa)

400 300 200 100 0 -100 -200 -90

-72

-54

-36

-18

0

18

36

54

72

90

Distance from WL (mm) Axial @ 50 Deg

Axial @ 90 Deg

Axial @ 150 Deg

Axial @ 250 Deg

Fig. 4.17 Residual axial stresses (MPa) on inner surface at different cross sections from the weld start position The quantitative variation of higher or lower residual stresses in the present research study are due to the different material properties i.e. mechanical properties like yield strength for base and weld filler metals along with other parameters such as weld geometry and heat source parameters etc. The significant related to axial stress fields from Figure 4.16 and Figure 4.17 are given in the following: 127

University of Engineering & Technology, Taxila-Pakistan

•

The compressive and tensile stress fields of high magnitude are observed on and near the FZ for outer and inner surfaces respectively. These stresses are symmetric across the weld line due to the symmetry.

•

Figure 4.16 shows a bulge showing stress variations beneath the weld crown near the weld line at the outer surface of cylinder.

•

Figure 4.16 and Figure 4.17 shows the axial residual stresses on outer and inner surfaces at four different cross sections (50o, 90o, 150o and 250o) and all are almost of the same magnitude and trend with slightly higher tensile axial stresses on inner surface. This previous research [4, 7, 197] shows the same trend. The axial stresses are weakly dependent on the circumferential location and have almost homogeneous distribution around the circumferential direction except the weld start and near region.

4.3.5.2 Hoop Residual Stress Fields Due to the radial expansion and contraction during the welding by heating and cooling sequence of welding process, the residual hoop stresses are developed parallel to the direction of the weld bead. Figure 4.18 and Figure 4.19 shows the residual hoop stresses distribution for inner and outer surfaces of cylinder along the axial directions in different cross sections (50o, 90o, 150o and 250o) from the weld start position respectively. A large tensile and compressive hoop stresses are observed on and near the weld line on the inner and outer surfaces respectively. The stress reduction and stress reversal trends are same as observed for hoop residual stresses in the case of axial residual stresses and also are in a good agreement with the other previous research [4, 185, 187, 189, 192-193, 197]. However, the quantitatively variation is due to the different welding parameters, material properties and heat source parameters respectively. The main observations are as follows: •

The hoop residual stresses are also symmetric due to symmetry across the weld line.

•

High tensile stresses of 146 MPa and 333 MPa are observed near the FZ on outer on inner surfaces respectively. A compressive residual stresses of 230 MPa and 208 MPa are observed away from the HAZ region at about 17 mm from weld line on outer and inner surfaces respectively.

•

The hoop stresses are based on the circumferential location from the weld start to weld end. The hoop residual stresses at three different cross sections (50o, 90o and 250o) varies in magnitude on outer and inner surfaces as shown in Figure 4.18 and Figure 4.19 with almost similar trend as shown by [4, 7, 197].

128

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

150

17 -50 -117 -183 -250 -90

-72

-54

-36

-18

0

18

36

54

72

90

Distance from WL (mm) Hoop @ 50 Deg

Hoop @ 90 Deg

Hoop @ 150 Deg

Hoop @ 250 Deg

Fig. 4.18 Residual hoop stresses (MPa) on outer surface at different cross sections from the weld start position 350 250 Inner Stress (MPa)

Outer Stress (MPa)

83

150 50 -50 -150 -250 -90

-72

-54

-36

-18

0

18

36

54

72

90

Distance from WL (mm) Hoop @ 50 Deg

Hoop @ 90 Deg

Hoop @ 150 Deg

Hoop @ 250 Deg

Fig. 4.19 Residual hoop stresses (MPa) on inner surface at different cross sections from the weld start position 129

University of Engineering & Technology, Taxila-Pakistan

4.3.5.3 Axial and Hoop Residual Stress Fields along the Circumference Figure 4.20 shows the comparison of axial and hoop residual stresses distribution at the weld line for outer and inner surfaces on a circumferential path. The stress distribution profiles are generally in agreement with the other previous research. The important observations are summarized as: •

The hoop stress varies from -234 MPa to 117 MPa on the outer surface. However, some exceptions at weld start and end and tack weld locations at 0o and 180o are observed and almost a zero hoop residual stress is observed at these locations (0o and 180o). For hoop residual stresses on inner surface, a slight variation in magnitude and trend is observed from -95 MPa to 140 MPa, with some exceptions on weld start and end and tack weld locations. Again the stress values almost to zero are observed at tack weld locations.

•

Figure 4.20 shows the compressive axial stresses profile on the outer surface varying from 203 MPa to 505 MPa. Almost a stable stress profile from weld start to weld end with some exceptions near to the weld start and end and tack locations is observed of a low magnitude. The significant effects of weld start and at tack weld points are observed for axial stress on inner surface. The compressive axial stresses varies from -286 MPa to about 490 MPa in magnitude with a reduction upto about 286 MPa and 222 MPa at the weld start of 0o and tack weld locations of 180o respectively. 575

Stress (MPa)

383 192 0 -192 -383 -575 0

30

60

90

120

150

180

210

240 270

300

330

360

Angle from Weld Start (o) Hoop-Inner

Axial-Inner

Hoop-Outer

Axial-Outer

Fig. 4.20 Axial and hoop residual stress fields on cylinder outer and inner surfaces on a circumferential path at the WL 130

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

4.3.6

Welding Distortions

4.3.6.1 Axial Tilt of Cylinders A comparison of predicted and experimentally measured axial tilt of the restraint free end of the cylinder after the welding and cooling of the weldments is shown in Figure 4.21. The accurate measurement of the face tilt of the cylinder experimentally is a difficult task by keeping in view the thickness of the cylinder face. Keeping the welded cylinders fixed in the welding positioner, a digital dial indicator by fixing at an average diameter of the cylinder at 297 mm by using some holding arrangements was used for the measurements by rotating the welding positioner. Five different readings after the tack weld and cooling of the weldments at room temperature are recorded to minimize the error of data acquisition. Figure 4.21 shows the plots of an average value for comparison with predicted results. The maximum range of axial face tilting is observed from -2.61 mm to +1.65 mm during the cylinder face tracking of the dial indicator from 0o to 360o. The degree of axial shrinkage is based on a many factors including welding process parameters, tack weld sizes and orientation. The maximum axial shrinkage of 2.61 mm at 84o is observed whereas the maximum axial deflection of 1.65 mm at 330o is observed near the weld end for the welding process parameters and tack weld geometry used in the this study. The FE predictions shows the lower values as compared to experimental data from 20% to 40% with an average of 20% from the weld start position at 0º to 150º. Whereas again the FE predictions shows the lower values as compared to experimental data with a variation at an average under prediction of 30% from the weld start position at 150º to 359º. The variation is slightly on the higher side, however the results are in a good qualitative agreement from the weld start to weld end. 2.0 1.5 Face Tilt (mm)

1.0 0.5 0.0 -0.5 -1.0 -1.5 -2.0 -2.5 -3.0 0

36

72

108

144

180

216

252

288

324

360

o

Angle from We ld Start ( ) Predicted

Experimental

Fig. 4.21 Measured and predicted axial deformation (face tilt) of the cylinder face 131

University of Engineering & Technology, Taxila-Pakistan

4.3.6.2 Axial Shrinkage The axial shrinkage at four different sections (50o, 90o, 150o and 250o) from the WL at cylinder outer surface is shown in Figure 4.22. The shrinkage on restraint free end is observed as by locating the coordinate axis on the weld line. The maximum axial shrinkage of 1.0 mm and 1.25 mm are observed for axial sections at 50o and 90o from the weld start position near the weld line at 10 mm from weld line towards restraint free end respectively. The minimum axial shrinkage value of 0.05 mm is observed as the axial shrinkage decreases away from the weld line towards the free end. The maximum axial shrinkage of 1.4 mm is observed at a distance of about 10 mm from the weld line for all the sections (50o, 90o, 150o and 250o) on the constrained end. A minimum shrinkage equal to about zero is shown at the restrained end as the axial shrinkage continuously decreases away from the weld line.

2.50

Axial Shrinkage (mm)

2.00 1.50 1.00 0.50 0.00 -0.50 -1.00 -1.50 -2.00 0

25

50

75

100

125

150

175

200

Distance from Weld Line (mm) Axial-50

Axial-90

Axial-150

Axial-250

Fig. 4.22 Axial shrinkage at four different cross sections from the WL on cylinder outer surface

132

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

4.3.6.3 Radial shrinkage Figure 4.23 shows the transient forces primarily responsible for the radial shrinkage during the welding phenomenon incase of circumferentially welded cylinders. This phenomenon can be explained with the concept of advancing solidification front [89]. A common observation during the welding is shrinkage of the weld bead transverse to weld line and along the weld path. A nearly hemispherical solidification front advance at rear of the weld pool exerts three-dimensional forces as shown in Figure 4.23. Figure 4.24 shows the radial shrinkage on outer surface of the cylinder at various cross sections (50o, 90o, 150o and 250o) from the weld start.

Fig. 4.23 Schematic representation of transient forces on solidifying weld pool

Radial Shrinkage (mm)

4.00 3.07 2.14 1.21 0.29 -0.64 -1.57 -2.50 -100

-75

-50

-25

0

25

50

75

100

Distance from Weld Line (mm) Radial-50

Radial-90

Radial-150

Radial-250

Fig. 4.24 Radial shrinkage at different cross sections from the WL on cylinder outer surface

133

University of Engineering & Technology, Taxila-Pakistan

4.4

Experimental Setup for Validation of FE Models

To validate the FE models, Gas Tungsten Arc Welding (GTAW) experiments were carried out. The appropriate way to ensure the reliability of the numerical simulations for the utilization of the research work for applications is by conducting the full-scale experiments in actual with proper instrumentation for data measurement. This section describes the experimental welding set-up, data measuring and acquisition systems used in the present research work for circumferential welding with the methodologies for experimental procedures.

4.4.1

Experimental Setup

Due to the variations associated with the skill of the operators and rotary synchronization problems in the manual arc welding, an automatic welding setup with minimum human intervention is used for arc welding experiments. The skill is mandatory for the proper validation of numerical simulations results because the heat source moves with constant speed i.e. the phenomenon is quasi-stationary in numerical simulations. A proper data acquisition system is required for the careful data measurement and analysis during the experiments. The TIG welding setup consisting of SAF TIGMATE 270 AC/DC power source, automatic rotary positioner, and fully automatic torch control and movement system is used to conduct the experiments. TIGMATE 270 welding power source as shown in Figure 3.1 in chapter 3 is a computerized waveform control technology for high quality TIG welds with the control of required parameters.

(a)

(b)

Fig. 4.25 Automatic rotary positioner with clamping and strain gages arrangements 134

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

To control and locate the torch movement, an automatic torch positioning system is used. Initially, the tack welded cylinders clamped according to the desired structural boundary conditions are rotated by rotary chuck with torch positioning at 90o to the cylinder. For details of automatic rotary positioner along with TIG torch, clamping and strain gages arrangements is given in Figure 4.25.

4.4.2

FE Models Validation

The welded part is subjected to a highly non-uniform and rapidly changing temperature field during welding approaching to temperatures above the melting point and the heat conducts away from the weld by convection and radiation to the surroundings. Numerous factors of the thermal field distribution during welding are the welding heat input, the thermal material properties, the amount of convective flow in the weld pool, the latent heat of melting and solidification, cooling to the surroundings, and contact with the surrounding materials as shown in Figure 4.26 [200]. Figure 4.27 shows the overall experimental validation approach of TIG welding in the present research for circumferential welding.

Fig. 4.26 Factors affecting the heat distribution during welding [200]

TIG WELDING EXPERIMENTS (Circumferentail Welding)

Thermal

Structural

model validation

model validation

Transient temperature measurement

Weld pool measurement by macrograph

Residual stress measurement

Distortion measurement

Fig. 4.27 Overall experimental validation approach for circumferential welding 135

University of Engineering & Technology, Taxila-Pakistan

4.4.2.1 Thermal Model Validation Generally two different types of experimental approaches are used for the validation of weld thermal model. The first approach is based on transient temperature measurements through thermocouples directly mounted on welded surface or by using the infrared pyrometers [201-208] whereas the second approach mostly acceptable is the comparison of FZ and HAZ from experimental macrograph as given by [18, 208-212]. Both temperature measurement approaches at some specified locations by using thermocouples and experimental macrograph for the measurement of FZ and HAZ are used to get data for the calibration of thermal FE models. TIG welding experiments are performed on thin-walled structure (cylinders) for circumferential weld and the sample for macrograph is cut by using water jet cutting process from the cylinders in rectangular cross section to avoid the undesired heat effects by gas or machine cutting process. The sample is cut away from the weld start and end, and tack weld locations in order to avoid the major effects related to these locations. Figure 4.28 shows the sample and the welded cylinder after water jet cutting. The macrograph is prepared as shown in Figure 4.29 (front view of the sample with FZ and HAZ) by doing these steps as sample preparation by wet cutting, mounting of the prepared sample in cast, sequential grinding by using silicon carbide abrasive paper with varying grit sizes in the order of 300, 500, 700, 800 and 1000, diamond paste polishing with particle sizes of 9 µm, 6 µm, 3 µm, and 1 µm, the etching of the sample by 2% nital solution and washing the sample to study the HAZ and FZ dimensions of the sample.

Fig. 4.28 Macrograph sample after water jet cutting from cylinder

Fig. 4.29 Low magnification metallographic sample of FZ and HAZ

K-type thermocouples are mounted directly on the outer surface of the cylinder prior to welding and connected directly to multi-channels data logging system for transient temperature history measurement. The temperature profile during welding at any specified time can be easily stored in computer readable formats like MS Excel for data processing and comparison with FE results later on. The thermocouples and data logger (acuracy and resolution = ± 1oC) used in research work are shown in Figures 4.30 and 4.31 respectively. 136

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 4.30 K-type thermocouples for transient temperature measurement

Fig. 4.31 Multi-channels data logging system with thermocouples connected The thermocouples are not suitable for the accurate measurement of temperature history within the FZ due to the limitations. The high precision general purpose infrared optical pyrometers (Cyclopes) from Minolta/Land are used to validate the temperature within the weld bead. Two Cyclopes with temperature measurement range of 800oC to 3000oC are utilized to measure the temperature with the FZ and on the weld line respectively. The pyrometer used is shown in Figure 4.32.

Fig. 4.32 Digital infrared pyrometer (Cyclopes from Minolta/LAND) 137

University of Engineering & Technology, Taxila-Pakistan

4.4.2.2 Structural Model Validation Two different types of validation i.e. residual stresses and distortion are required to validate the weld structural model as already shown in Figure 4.27. Transient axial and radial distortions are measured on the welded cylinders for distortion measurement. Further, the hoop and axial residual stresses are also measured on the same welded cylinder to compare the results.

4.4.2.2.1 Validation of Distortions Transient axial deformation during the welding at the cylinder face and residual radial deformation near the weld line are measured and compared with the related predicted data from the FE analysis. A digital dial indicator (± 0.001 mm) is carefully located to track the transient distortion at the restraint free face of the cylinder for the measurement of axial shrinkage and precision micrometers (± 0. 01 mm) are used to measure the post weld residual radial shrinkage on the specified points along the entire surface of welded cylinders in the HAZ near weld line as shown in Figure 4.33.

Fig. 4.33 Experimental setup used for distortion measurement 4.4.2.2.2

Validation of Residual Stresses

As already shown in Figure 3.16 in chapter 3, the hole-drilling method is used for the measurement of residual stresses and the equipment for hole-drilling strain gage along with P3500 strain meter from Vishay Group as already discussed is used for experimental determination of residual stress fields in circumferential welding. The milling guide RS-200 as shown in Figure 3.16 can be used for cylindrical surfaces with some special arrangements. A separate fixture to hold the welded cylinder for proper mounting of RS-200 milling guide is used for hole-drilling and measurements. Complete experimental setup for residual stress measurement in circumferential welding is shown in Figure 4.34. The detail of six basic steps involved for measurement of residual stresses by hole-drilling method is discussed in section 3.3 of chapter 3. The two strain gage rosettes (EA-XX-062RE-120 and CEA-XX-062UM120) are used as already shown in Figure 3.18. 138

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 4.34 Experimental setup used for experimental measurement of residual stresses

4.5

Chapter Summary and Conclusions

In this chapter, the theoretical background and finite element modeling aspects of the thermal-mechanical behavior during arc welding was discussed. In order to validate the FE models, Gas Tungsten Arc Welding (GTAW) experiments were carried out to ensure the reliability of the numerical simulations with proper instrumentation for data measurement. Computational methodology and techniques based on finite element analysis (FEA) for the prediction of temperature profiles and subsequent weld induced residual stress fields and distortion patterns in GTA welded thin-walled cylinders of high strength low alloy steel were developed and implemented successfully with close correlation to the experimental investigations. The results related to residual stress fields and distortion was discussed in detail. The significant conclusions from the results are: 1. Due to symmetry across the weld line, the residual stresses (both hoop and axial) are symmetric. Along and near the weld line, a high tensile and compressive axial residual stresses occurs on the cylinder inner and outer surfaces respectively. 139

University of Engineering & Technology, Taxila-Pakistan

Compressive and tensile axial residual stresses produced on inner and outer surfaces away from the weld line. Axial stresses are weakly dependent on the circumferential locations from weld start. 2. Hoop residual stresses are sensitive to the angular location from the weld start position. On the inner surface, the weld start effect is more severe for both axial and hoop stresses and is dominant in the weld start direction. The significant effect of tacks on the axial stress on the inner surface is observed at angular positions of 0° and 180° from the weld start point, whereas, the effect of tacks on hoop stresses is not as prominent. The stress distribution is no more axis-symmetric for a single pass butt circumferential weld with initial tacks. However, if the weld start/end effects are ignored hoop stresses are almost uniform. 3. Maximum axial and radial deflection is observed near the weld line. The axial shrinkage decreases continuously away from the WL and a minimum shrinkage of almost zero shown at the restrained end. However, on the restraint free end some deflection with face tilting is observed. Further, this chapter described the details of experimental welding set-up, data measuring and acquisition systems used for the validation of developed FE models. The FE model developed proves to be very effective and efficient for conducting virtual experiments for the prediction of residual stresses and distortion by using design of experiments (DOE) for optimizing the TIG welding process of thin walled structure (cylinder) and also to provide the data for developing the expert system.

140

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

CHAPTER 5 VIRTUAL DESIGN OF EXPERIMENTS (DOE) & OPTIMIZATION OF GTAW PROCESS OF THIN-WALLED STRUCTURE FOR CIRCUMFERENTIAL WELDING 5.1 Introduction In this chapter, analyzing the effects of welding parameters on residual stresses and distortions by using FE model developed in the previous chapter is presented for parametric studies focusing on the effects of significant welding processes parameters, geometric parameters including joint root openings and tack welds orientation on the residual stresses and distortion. After initial parametric studies for circumferential welding, further parametric studies and their effect on responses are carried out by DOE methodology as applied in chapter 3 for linear welding based upon virtual experiments for the response optimization of welding process by using full factorial and response surface method (RSM) for ANOVA, empirical modeling and numerical optimization. At the end after obtaining the data of all the analysis performed in chapter 3 and chapter 5, RSM is also applied on both i.e. linear (sheets) as well as circumferential (cylinders) welds for empirical modeling containing all the variables analyzed and optimization accordingly.

5.2 Analyzing the Effects of Welding Parameters on Residual Stresses and Distortions In the present research fully 3D FE models are developed for the parametric studies as described later in section 5.2.1.

5.2.1

Details of Parametric Studies

In this section, the research work based on welding parametric studies primarily focus on the effects of significant welding processes parameters on the residual stress formation as analyzed in chapter 3 i.e. welding speed and welding current, cylinder thickness along with joint root openings and tack welds orientation. Whereas, the stick out distance in arc welding processes is believed to have no significant effects on thermal fields (HAZ dimensions) [216], therefore is not considered here. The net heat input to the weldments can be optimized to obtain desired optimum residual stress level by varying the welding current, welding voltage or by controlling the welding speed. The welding parameters of thin-walled cylinders are taken as obtained in chapter 3 as the base parameters for the numerical simulations. To analyze the effects of welding process parameters on residual stress fields, three different values of welding speed and welding current are used in parametric studies for analysis. The values of heat source (fixed) and welding process (variables) parameters are shown in Table 5.1 and Table 5.2 respectively. In these studies, the total heat input to the weldments is used as a reference value to judge the effect of variable parameters of welding process. The total heat input (mega joules per meter (MJm-1)) is as given in Equation 5.1. Heat input =

V *I 1000 * WS

(5.1)

141

University of Engineering & Technology, Taxila-Pakistan

Where, V is the welding voltage (volts), I is the welding current (amperes), and WS is the welding speed (mm/s). The heat input is controlled in such a way that the minimum total heat input (0.625 MJm-1) incase of maximum welding speed (4 mm/s) remains same to the total heat input incase of minimum welding current of 150 amperes.

Table 5.1 Heat source parameters Length of ellipsoidal Front (af) Rear (ar) (mm) (mm) 5.0

15.0

Heat source width (2b) depth (c) (mm) (mm) 10.0

3.0

Fraction of heat in ellipsoidal Front (ff) Rear (fr) 1.25

0.75

Table 5.2 Welding process parameters Parameter WPP-1 WPP-2 WPP-3

Current (I) (amperes) 150 200 300

Voltage (V) (volts) 12.5 12.5 12.5

Efficiency (η) Welding Speed (WS) (%) (mm/s) 80 2 80 3 80 4

Further, three parametric studies to analyze the effects of different geometric parameters, other three additional studies to investigate the effects of joint root openings and tack weld orientations respectively are carried out. At a time one parameter is varied only by keeping all the other parameters as constant to investigate the effects of that changed parameter. Siddique [89] presented the study pertaining to the numerical modeling of the effects of root openings and tack weld modeling for circumferentially welded geometries. Previously in [217-221], the issues of tack welds and root openings in butt-welded plates were discussed. Jonsson et al. [217, 218] presented the effects of tack weld sequence in welded plates by assuming the plane stress problem for purpose of simplification. The temperature dependent interface elements were used in another study for modeling of root gap and tack welds by Shibahara [219, 220]. The conclusion of these studies was that the tack welds and root gaps have significant effects on the structure axial deformations. Further, Jang et al. [221] presented that the root gap has effects on residual stress fields also across the welds by using the plane strain assumptions. The details of the simulation strategy employed, heat source model, material model, FE model (300 mm outer diameter cylinder of thickness 3mm) are already discussed in chapter 4. The same material model for high strength low alloy steel, heat source model and simulation strategy is used for the parametric studies in this chapter. However, different FE models are used as per requirements as these parametric studies based on different geometric parameters including cylinder wall thickness and joint root openings along with the effects of tack weld orientation. The mesh topology used remains same in general for all the FE models in this chapter with necessary modifications as per requirement of different parametric studies. All the FE meshes used in this chapter are based on sensitivity analysis to get the mesh independent results as already discussed in chapter 4. 142

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

A slightly modified and computationally efficient double ellipsoidal heat source model instead of the previously utilized heat source model is used as shown by Equations 5.2 and 5.3 for front and rear ellipsoids respectively [4]. ANSYS® does not have inherent features to model the moving heat source given by Equations 5.2 and 5.3. Therefore, APDL subroutines are used to fulfill the requirement. The coordinate system of the heat source is used in such a way that the origin of the system remains at the centre of heat source defined by the heat source parameters. qf =

6 3 M (r , z ) Q f f

π π a f bc

−3{

e

r 2θ 2 af 2

r 2θ 2

6 3 M (r , z ) Q f r −3{ ar 2 qr = e π π ar bc

+3

+3

z2 b2

z2 b2

+

+

Ro2 + r 2 -2 rRo c2

}

(5.2)

Ro2 + r 2 -2 rRo c2

}

(5.3)

But, the validity of the model experimentally is necessary before the use of the modified heat source model for the parametric studies. The welding experiments are performed to verify the modified heat source model by measuring the fusion zone (FZ) and heat affected zones (HAZ) through micrograph. Figure 5.1 shows the comparison of experimental and simulated FZ and HAZ. To avoid the effects of weld start, stop and tack welds, the macrograph as shown in Figure 5.1 is taken at section 150o from the weld start position. Figure 5.1 shows a very good agreement between the predicted HAZ isotherms and the experimentally measured HAZ. Goldak et al. [51] mentioned that double ellipsoidal heat source model is not suitable to match with some experimentally determined weld pool shape FZ and dimensions. WL o (> 1540 C)

HAZ (> 1400oC)

(a) FZ-EXP

FZ-SIM

HAZ-EXP

HAZ-SIM

Wall thickness (mm)

3 2.5 2 1.5 1 0.5 0 -8

-6

-4

-2

0

2

4

6

8

Distance from WL

(b)

Fig. 5.1 (a) Experimental macrograph at a section 150o from weld start position (b) Comparison of experimental and simulated FZ and HAZ dimensions 143

University of Engineering & Technology, Taxila-Pakistan

5.2.2

Welding Speed Effects

The axial residual stress fields for welding speeds of 2, 3 and 4 mm/s used in the simulations are shown in the Figure 5.2. The maximum tensile stresses and compressive stresses are observed at the lowest welding speed of 2 mm/s on the internal and external surfaces of the cylinder at the weld line respectively. Similarly, the minimum tensile stresses and compressive stresses are observed at the highest welding speed of 4 mm/s on the internal and external surfaces of the cylinder at the weld line respectively. It is technically understandable that the lower welding speed results in more heat input per unit volume which results wider fusion and HAZ zones and increases the stresses on both sides of cylinder surfaces. However, it is reversed away from the weld line i.e. compressive on inner surfaces and tensile on the outer surfaces. The stresses remains almost constant at further away from the weld line. The trend for the residual axial stresses is similar as in the the previous research [4, 222]. The higher residual stresses in the present study are due to the different material properties i.e. mechanical properties of HSLA steel like higher yield stress for base and filler metals, geometry and heat source parameters. 550

Axial Stress (MPa)

425 300 175 50 -75 -200 -325 -450 -85

-68

-51

-34

-17

0

17

34

51

68

85

Distance from WL (mm) Inner (2 mm/sec) Inner (3 mm/sec) Inner (4 mm/sec)

Outer (2 mm/sec) Outer (3 mm/sec) Outer (4 mm/sec)

Fig. 5.2 Residual axial stresses on outer and inner surface of the cylinders at a section 150o from the weld start position with different welding speeds Figure 5.3 shows the hoop residual stresses circumferentially along the weld line for the weld speeds of 2, 3 and 4 mm/s. The minimum residual hoop stresses are observed at the weld start and weld end (i.e. at 0o and 360o) for inner and outer cylinder surfaces respectively. The trend of lower hoop stress is observed at the tack weld location of 180o. The trend for the residual hoop stresses is similar as in the previous research [4, 223]. 144

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

250

Hoop Stress (MPa)

169 88 6 -75 -156 -238 -319 -400 0

36

72

108

144

180

216

252

288

324

360

o

Angle from Weld Start ( ) Outer (2 mm/sec) Outer (3 mm/sec) Outer (4 mm/sec)

Inner (2 mm/sec) Inner (3 mm/sec) Inner (4 mm/sec)

Fig. 5.3 Residual hoop stresses on outer and inner surface of the cylinders along the circumference at the WL with different welding speeds 350

Hoop Stress (MPa)

233 117 0 -117 -233 -350 -85

-68

-51

-34

-17

0

17

34

51

68

85

Distance from WL (mm) Inner (2 mm/sec) Inner (3 mm/sec) Inner (4 mm/sec)

Outer (2 mm/sec) Outer (3 mm/sec) Outer (4 mm/sec)

Fig. 5.4 Residual hoop stresses on outer and inner surface of the cylinders at a section 150o from weld start position for different welding speeds 145

University of Engineering & Technology, Taxila-Pakistan

Figure 5.4 shows the residual hoop stresses at inner and outer surfaces of cylinder at 150 from the weld start position for the weld speeds of 2, 3 and 4 mm/s. Again similar to axial residual stresses at different weld speeds, the maximum residual stresses are observed at lowest welding speed and vice versa. The residual stresses approaches to a minimum level away from the weld line (WL) in a uniform trend. o

5.2.3

Input Heat Effects

Figure 5.5 and Figure 5.6 shows the residual hoop and axial stresses at inner and outer cylinder surface at a section 150o from the start of the weld for the three welding currents of 150, 200 and 300 amperes by keeping all the other parameters constant. Figures shows the maximum hoop and axial residual stress fields due to the more heat input per unit volume as incase of WPP-3 (1.25 MJm-1) and lowest stress fields due to the minimum heat input per unit volume as incase of WPP-1 (0.625 MJm-1). The effects of total heat input per unit volume directly influences the temperature distributions and consequently the residual stress profiles in the welded structures if welding speed and all other parameters are kept constant as per the heat equation 5.1. For both hoop and axial residual stresses, same trend and data values are obtained from the respective parametric studies with same heat input in MJm-1 by varying the welding current whereas the similar trends are observed for axial and hoop residual stress fields for same heat input parametric studies but controlled by varying the welding speed as shown in Figure 5.2 and Figure 5.4. 350

Hoop Stress (MPa)

233 117 0 -117 -233 -350 -85

-68

-51

-34

-17

0

17

34

51

68

85

Distance from WL (mm) Inner (WPP-1) Inner (WPP-2) Inner (WPP-3)

Outer (WPP-1) Outer (WPP-2) Outer (WPP-3)

Fig. 5.5 Residual hoop stresses on outer and inner surface of the cylinders at a section 150o from the weld start position for different welding current

146

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

550

Axial Stress (MPa)

425 300 175 50 -75 -200 -325 -450 -85

-68

-51

-34

-17

0

17

34

51

68

85

Distance from WL (mm) Inner (WPP-1) Inner (WPP-2) Inner (WPP-3)

Outer (WPP-1) Outer (WPP-2) Outer (WPP-3)

Fig. 5.6 Residual axial stresses on outer and inner surface of the cylinders at a section 150o from the weld start position for different welding current 5.2.4

Cylinder Thickness Effects

Three different studies to analyze the effects of varying cylinder wall thickness of 3, 4 and 5 mm on the corresponding stresses are conducted. The outer diameter of cylinder along with other geometric parameters are taken as same from chapter 4 in section 4.3. The heat input in all the three studies is varied in such a way that the penetration with respect to wall thickness remains almost same for a good comparison. Welding process parameters for the parametric studies with respect to three different wall thicknesses are shown in Table 5.3. The details of the FE models are earlier discussed in chapter 4 in section 4.3. The heat source parameters are used as given in the Table 5.1.

Table 5.3 Cylinder thickness and welding process parameters for parametric studies Cylinder Thickness Current (I) (mm) (amperes)

Voltage (V) (volts)

Efficiency (η) Welding Speed (WS) (%) (mm/s)

3

200

12.5

80

3

4

250

14

80

3

5

300

15

80

3

147

University of Engineering & Technology, Taxila-Pakistan

Figure 5.7 shows the axial residual stress fields for three different cylinder wall thickness values on outer and inner surfaces of cylinder at a section 150o from weld start position. In all studies, tensile and compressive stresses are observed at the WL approaching to zero away from the weld line and then changing to compressive and tensile at a distance away from the WL for the inner and outer surfaces of cylinder respectively. There is no considerable variation observed in magnitude and trend for outer and inner surfaces starting from WL and proceeding axially to a distance of ~12 mm. However, more prominent tensile and compressive stresses are observed with the increase of wall thickness for outer and inner surfaces of cylinder respectively. With the increase in wall thickness from 3 to 5 mm, the magnitude of stress zone increases from ~12 mm to ~15 mm for outer and inner surfaces. The increase in stress zone is considered due to the enhanced stiffness for the same outer diameter of cylinder with increase of wall thickness. Due to increase in stiffness of the cylinder with the increase of thickness, there is more resistance to local bending, accounting lesser residual stress and larger tensile and compressive stress zones. Figure 5.8 shows the hoop residual stress fields at an axial section 150o from weld start for outer and inner surfaces of cylinder. The hoop stresses of inner surfaces are almost same in magnitude & trend in HAZ region and show no significant variation for varying the wall thicknesses.

550

Axial Stress (MPa)

425 300 175 50 -75 -200 -325 -450 0

6

13

19

26

32

38

45

51

Distance from WL (mm) Axial-Inner (3 mm) Axial-Inner (4 mm) Axial-Inner (5 mm)

Axial-Outer (3 mm) Axial-Outer (4 mm) Axial-Outer (5 mm)

Fig. 5.7 Axial residual stresses on cylinder outer and inner surfaces at a section of 150o from weld start position for various thicknesses

148

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

A prominent variation in compressive hoop residual stresses is observed away from the HAZ region for lesser thickness. A variation of about 136 MPa is observed in compressive stresses with the variation of thickness from 5 mm to 3 mm. No considerable variation in stress zone of influence is shown for this case. The compressive hoop residual stresses of magnitude 154 MPa and 98 MPa are observed for outer surfaces on the weld line for thickness of 3 and 4 mm respectively and tensile residual hoop stress of magnitude of about 34 MPa is observed for 5 mm wall thickness. 350

Hoop Stress (MPa)

233 117 0 -117 -233 -350 0

6

13

19

26

32

38

45

51

Distance from WL (mm) Hoop-Inner (3 mm) Hoop-Inner (4 mm) Hoop-Inner (5 mm)

Hoop-Outer (3 mm) Hoop-Outer (4 mm) Hoop-Outer (5 mm)

Fig. 5.8 Hoop residual stresses on cylinder outer and inner surfaces at a section of 150o from weld start position for various thicknesses In general, cylinder wall thickness has greater influence on hoop residual stresses as compared to axial residual stresses. As far as the theoretical aspects are concerned, the variation in hoop residual stresses is directly linked to the radial expansion and contraction, during heating and cooling sequence. In Figure 5.8, it is clear that tensile residual hoop stresses are developed for inner surfaces on and near the weld line. Compressive hoop residual stresses on the outer surface are observed at the same locations in general. The degree of restraints against the heat expansion in the hoop direction exceeds to that in the axial direction [82]. The hoop stress distribution differs in terms of stress variation on inner and outer surface. A comparative axial and hoop residual stress on cylinder inner surface along the circumference at weld line is shown in Figure 5.9. The axial and hoop stresses are almost same in terms of magnitude and trend for three wall thickness values but some minor 149

University of Engineering & Technology, Taxila-Pakistan

exceptions are observed at weld start & end and tack weld locations for hoop stresses. The axial and hoop residual stress fields as shown in Figure 5.10 are of the same magnitude with similar trend on cylinder outer surface. A significant variation of hoop stresses is observed at weld start and end and particularly on tack weld locations. A reduction of about 147 MPa in compressive residual stresses following a similar trend variation is observed for an increase of wall thickness from 3 mm to 5 mm. 550 425

Stress (MPa)

300 175 50 -75 -200 -325 -450 0

36

72

108

144

180

216

252

288

324

360

Angle from Weld Start (o) Axial-Inner (3 mm) Hoop-Inner (3 mm) Axial-Inner (4 mm) Hoop-Inner (4 mm) Axial-Inner (5 mm) Hoop-Inner (5 mm)

Fig. 5.9 Axial and hoop residual stresses on cylinder inner surface at a circumferential path on WL for various wall thickness values 5.2.5

Root Opening Effects

To analyze the effects of root opening on the residual stress fields and deformations, four case studies with different root openings are carried by keeping all the parameters constant i.e. geometric parameters (weld joint, cylinders sizes, tack weld sizes and orientation etc.), heat source parameters and welding process parameters (welding current, welding voltage and welding speed etc.). To analyze the effects of varying root opening on residual stresses and deformation fields, same thermal and structural boundary conditions are used. In all four studies, two tack welds with same length in circumferential direction along the weld path are modeled at circumferentially opposite to each other i.e. at weld start of 0o and 180o. By the manipulation of the element size in axial direction in the fusion zone, same model is used as previously discussed with the variation of root opening. The four different values of root opening studied in the case studies are given in Table 5.4.

150

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

150 56

Stress (MPa)

-38 -131 -225 -319 -413 -506 -600 0

36

72

108

144

180

216

252

288

324

360

Angle from Weld Start (o) Axial-Outer (3 mm) Axial-Outer (4 mm) Axial-Outer (5 mm)

Hoop-Outer (3 mm) Hoop-Outer (4 mm) Hoop-Outer (5 mm)

Fig. 5.10 Axial and hoop residual stresses on cylinder outer surface at a circumferential path on WL for various wall thickness values Table 5.4 Studies for the analysis of effects of root openings Studies

1

2

3

4

Reference

RO – 0

RO - 1.6

RO - 2

RO - 2.6

Root Opening

0 mm

1.6 mm

2.0 mm

2.6 mm

The axial residual stresses on a circumferential path at the weld line on cylinder inner surface are shown in Figure 5.11 and observed no variation in axial stress profiles by varying root openings. Figure 5.12 shows the details of hoop residual stress values on the same path. Except for zero roots opening (RO-0), there is no significant variation. However, some higher tensile hoop stresses are observed for zero root opening. At the weld start location, the variation is more significant. In Figure 5.13 and Figure 5.14, similar trends are obtained for residual axial and hoop stresses respectively on the same path at outer surface. In this case a significant variation of about 170 MPa and 200 MPa is observed for axial compressive residual stresses and hoop compressive residual stresses respectively at the tack weld location of 180o for zero root opening.

151

University of Engineering & Technology, Taxila-Pakistan

600

Axial Stress (MPa)

475 350 225 100 -25 -150 -275 -400 0

36

72

108

144

180

216

252

288

324

360

o

Angle from Weld Start ( ) Axial (RO-0.0 mm)

Axial (RO-1.6 mm)

Axial (RO-2.0 mm)

Axial (RO-2.6 mm)

Fig. 5.11 Axial residual stress variations, circumferentially on the weld line for cylinder inner surface with different root openings

Figure 5.15 shows the axial displacement at the restraint free face of the cylinder at outer diameter of cylinder. It is easily observed that root opening has direct influence on axial displacement. There is no significant axial deformation observed with zero root opening and further, there is no significant variation in the axial displacement up to 2 mm root opening. However, significant axial displacement is shown for root opening of 2.6 mm as compared to root opening ≤ 2 mm. Tack welds can be treated as a column as mentioned by Siddique et al. [89]. The stiffness of the column is given by the Equation 5.4. K=

AE L

(5.4)

Where, K is the stiffness, E is the Young's modulus and A, L are the cross sectional area and length of the column respectively. The stiffness of the tack weld is to be inversely proportional to the axial length of the tack by considering the tack weld as a column as discussed above. The stiffness of the tack weld decreases with the increase in root opening i.e. increase in tack weld length. Therefore, higher axial displacement is observed incase of RO-2.6 (2.6 mm root opening) for less stiffened tack weld.

152

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

450

Hoop Stress (MPa)

325 200 75 -50 -175 -300 -425 -550 0

36

72

108

144

180

216

252

288

324

360

Angle from Weld Start (o) Hoop (RO-0.0 mm)

Hoop (RO-1.6 mm)

Hoop (RO-2.0 mm)

Hoop (RO-2.6 mm)

Fig. 5.12 Hoop residual stress variation, circumferentially on the weld line for cylinder inner surface with different root openings 0

Axial Stress (MPa)

-75 -150 -225 -300 -375 -450 -525 -600 0

36

72

108

144

180

216

252

288

324

360

Angle from Weld Start (o) Axial (RO-0.0 mm)

Axial (RO-1.6 mm)

Axial (RO-2.0 mm)

Axial (RO-2.6 mm)

Fig. 5.13 Axial residual stress variations, circumferentially on the weld line for cylinder outer surface with different root openings 153

University of Engineering & Technology, Taxila-Pakistan

200

Hoop Stress (MPa)

113 25 -63 -150 -238 -325 -413 -500 0

36

72

108

144

180

216

252

288

324

360

Angle from Weld Start (o) Hoop (RO-0.0 mm)

Hoop (RO-1.6 mm)

Hoop (RO-2.0 mm)

Hoop (RO-2.6 mm)

Fig. 5.14 Hoop residual stress variation, circumferentially on the weld line for cylinder outer surface with different root openings

1.00 0.66 Face Tilt (mm)

0.33 0.00 -0.34 -0.67 -1.00 -1.33 -1.67 -2.00 0

36

72

108

144

180

216

252

288

324

360

o

Angle from Weld Start ( ) Face Tilt (RO-0.0 mm)

Face Tilt (RO-1.6 mm)

Face Tilt (RO-2.0 mm)

Face Tilt (RO-2.6 mm)

Fig. 5.15 Axial displacement of the restraint free face of cylinder for different root openings

154

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

5.2.6

Tack Weld Orientation Effects

In the following, two case studies are conducted for the effects of a critical geometric parameter i.e. tack welds on the corresponding residual stress fields and distortion patterns. As the tack welds are normally used to restrain the excessive axial shrinkage and maintain the desired root opening during the welding. The number of tack welds, their length and the distance between tacks should be considered for effectiveness [224]. The tack welds provide a considerable resistance to the shrinkage and further, the orientation and size of tacks can change the distortion patterns with in the weldments. For the analysis of effects of tack weld orientation and length of tack welds, same FE model for outer diameter of cylinders with wall thickness of 3 mm is used in the case studies. For these two case studies, two tack welds are modeled at circumferentially opposite to each other with same length in circumferential direction along the weld path. The model is similar as discussed previously but the position of live elements representing tack welds in all the studies is different for the variation of tack weld orientation. The details of the case studies are given in Table 5.5. Heat source parameters are used as given in the Table 5.1 along with welding process parameters as given in the Table 5.3. The effects of tack weld orientation are given in detail in the following sections. Mostly the results given in this domain relates to the effects of tacks expecting to be in and around the weld fusion zone whereas the axial shrinkage is also discussed in detail at the restraint free face of the cylinder.

Table 5.5 Studies for tack weld orientation analysis Studies

1

2

Reference

Case-1

Case-2

Tack welds orientation from weld start at 0o

0o - 180o

45o - 225o

5.2.6.1 Effects on Residual Stresses Figure 5.16 and Figure 5.17 shows the axial residual stresses for different tack weld orientations at a longitudinal section 180o from weld start position for outer and inner surfaces respectively. Most important observations are given in the following.

155

University of Engineering & Technology, Taxila-Pakistan

•

The axial residual stress fields are identical in trend and magnitude for all tack weld orientations on inner surface. However, few exceptions are observed for tack weld orientations of 0o-180o. The lower residual axial stresses are observed due to tack weld at weld start position of 0o. It is observed that the weld tack reduce the stresses as a stress reducer and a gain in stress reduction of about 110 MPa at weld line is observed. But along this, a slightly larger stress zone of influence in this case is also observed.

•

Similarly, compressive residual stress fields both in trend and magnitude on outer surface are shown in Figure 5.17. At weld line for tack welds at 0o-180o, a significant stress increase of about -175 MPa is observed.

•

Figure 5.18 and Figure 5.19 show the variation in axial residual stress fields for different tack weld orientations at a path on the weld line along the circumference for outer and inner surface of the cylinder respectively.

•

The axial residual stresses are compressive on outer surface and tensile on inner surface in general with the following important observations.

•

With some exceptions at tack weld orientations, the axial residual stress fields are same in magnitude and trend on both inner and outer surfaces.

•

The effects of weld start and end on axial residual stresses in all the case studies shows a strong influence. The effects of weld start and end are significant on outer surface. However, the effect is more significant at the weld start on inner surface.

•

A reasonable enhancement in compressive axial residual stress at the tack weld locations is observed on outer surface. The maximum compressive axial residual stress of about 554 MPa at tack weld locations of 90o, 180o and 270o is observed. A residual compressive stresses of 494 MPa and 477 MPa are observed at tack weld locations of 135o and 315o respectively.

•

A slight effect of reduction in tensile axial residual stress at the tack weld locations is observed on inner surface. The maximum reduction of tensile axial residual stress of about 90 MPa at tack weld locations of 90o, 180o and 270o is observed. However at tack weld locations of 135o and 315o, no significant tensile stress reduction is observed.

156

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

550

Axial Stress (MPa)

450 350 250 150 50 -50 -150 -250 -50

-40

-30

-20

-10

0

10

20

30

40

50

Distance from WL (mm) Axial (0-180)

Axial (45-225)

Axial (90-270)

Axial (135-315)

Fig. 5.16 Axial residual stress variations at a longitudinal section 180o from weld line on cylinder inner surface 300

Axial Stress (MPa)

200 100 0 -100 -200 -300 -400 -500 -600 -50

-40

-30

-20

-10

0

10

20

30

40

50

Distance from WL (mm) Axial (0-180)

Axial (45-225)

Axial (90-270)

Axial (135-315)

Fig. 5.17 Axial residual stress variations at a longitudinal section 180o from weld line on cylinder outer surface 157

University of Engineering & Technology, Taxila-Pakistan

0

Axial Stress (MPa)

-75 -150 -225 -300 -375 -450 -525 -600 0

36

72

108

144

180

216

252

288

324

360

Angle from Weld Start (o) Axial-(0-180)

Axial-(45-225)

Axial-(90-270)

Axial-(135-315)

Fig. 5.18 Axial residual stress variations at a circumferential path on the weld line for cylinder outer surface 600 500 Axial Stress (MPa)

400 300 200 100 0 -100 -200 -300 -400 0

36

72

108

144

180

216

252

288

324

360

Angle from Weld Start (o) Axial-(0-180)

Axial-(45-225)

Axial-(90-270)

Axial-(135-315)

Fig. 5.19 Axial residual stress variations at a circumferential path on the weld line for cylinder inner surface 158

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

5.3 Virtual Design of Experiments (DOE) and Optimization of Circumferential Welding This section provides in-depth study of effects of welding process parameters, in virtual experiments based on simulation for circumferential welding of thin walled cylinder, upon residual stresses and distortion following the standard design of experiments (DOE) by using full factorial method. In addition to performing ANOVA of experimental results, the empirical models, for quantifying the effects of welding parameters, numerical optimization and response desirability for circumferential welding of thin walled cylinders of different thicknesses will also be presented in detail.

5.3.1 Virtual DOE It is possible to conduct virtual experiments using existing data and simulations. Simulations are used to predict the output and to support the decision making. In virtual experiments, the experimental design performed as would on a real process. However, instead of making changes to the actual process, make changes to the virtual process as represented by simulations. The cost of doing them is minimum, compared with experiments in the real world and the process of identifying input and output variables. Virtual experiments allow a great deal more ‘what if’ analysis which may stimulate more creative thinking [225]. Modern simulation software can interface with statistical analysis software to allow more detailed analysis of proposed new products and process. In [226] the author demonstrated this capability with iGrafs Process for six sigma and Minitab and these capabilities are also incorporated with other software packages. The process was performed by simulation and analyzed by using DOE techniques for quality and performance improvement [226]. The traditional approach to experimentation require to change only one factor at a time (OFAT) while keeping other as constant and this approach doesn’t provide data on interactions of factors which occurs in most of process. The alternative statistical based approach called “two level factorial design” can uncover the critical interactions that involve simultaneous adjustments of experimental factors at only two levels: high (+1) and low (-1). The two level factorial design offers a parallel testing scheme which is most efficient than the serial approach OFAT. Two level experiments restrict the experiments to minimum and contrast between the levels give the necessary driving force for the process improvement and optimization. The statistical approach to design of experiments (DOE) and analysis of variance (ANOVA), developed by R.A. Fisher in 1920, is an efficient technique for experimentation which provides a quick and cost effective method for complex problem solving with many variables [227].

5.3.2 Predictor Variables Predictor variables are the welding process parameters that can also be represented as process input parameters or input variables. Following is the list of significant TIG welding predictor variables with values that would be under study in virtual welding experiments to be performed on thin walled HSLA steel cylinders of different thicknesses (3, 4 and 5 mm):

159

University of Engineering & Technology, Taxila-Pakistan

1. 2. 3. 4.

Welding Current (Amp) (170-270) Welding Voltage (Volts) (10.5-13.5) Welding Speed (cm/min) (15-18) Ar Trailing (ON/OFF)

The most significant parameter is current with high range of values as compared to others parameters that shows the practical range of these parameters specifically of welding current should be specific with respect to thickness of material and the heat input (welding current, welding voltage and welding speed) required for the fusion and weld of selected thickness as discussed in Chapter 3. With the increase in material thickness, the increase in heat input is required to fuze and weld. The parameters and their levels with high and low settings are given in detail in the following sub-section to conduct the virtual experiments.

5.3.2.1 Factorial Design of Experiments A 24 (4 factors, 2 levels, 16 test) full factorial design model (replicates 1, block 1, centre point per block 0 and order 4FI) was used for the welding experiments. Table 5.6, Table 5.7 and Table 5.8 shows the low and high settings (or levels) for the predictor variables (or parameters) used in sixteen tests for the cylinder thickness of 3, 4 and 5 mm respectively. Three of these predictor variables (welding current, welding voltage and welding speed) are numeric while the other one gas trailing is categorical. Complete detail of 16 experiments following full factorial has been presented in Table 5.9, Table 5.10 and Table 5.11 for cylinder thickness of 3, 4 and 5 mm respectively. All the statistical analyses were performed using a commercial computing package named DesignExpert® 7.1.6, by Stat-Ease® and MINITAB® Release 14.

Table 5.6 High and Low Settings of Factors (cylinder thickness = 3 mm) Factor Name

Units

Type

Low Actual

High Actual

A

Current

A

Numeric

170.00

210.00

B

Voltage

V

Numeric

10.50

13.50

C

Weld Speed

cm/min

Numeric

15.00

18.00

D

Trailing

Categoric

nil

Ar

Table 5.7 High and Low Settings of Factors (cylinder thickness = 4 mm) Factor Name

Units

Type

Low Actual

High Actual

A

Current

A

Numeric

200.00

220.00

B

Voltage

V

Numeric

10.50

13.50

C

Weld Speed

cm/min

Numeric

15.00

18.00

D

Trailing

Categoric

nil

Ar

160

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 5.8 High and Low Settings of Factors (cylinder thickness = 5 mm) Factor Name

Units

Type

Low Actual

High Actual

A

Current

A

Numeric

230.00

270.00

B

Voltage

V

Numeric

10.50

13.50

C

Weld Speed

cm/min

Numeric

15.00

18.00

D

Trailing

Categoric

nil

Ar

Table 5.9 Design of 16 Experiments following Full Factorial (cylinder thickness = 3 mm)

Std

Run

12

1

Factor 1 A:Current A 210.00

Factor 2 B:Voltage V 13.50

Factor 3 C:Weld Speed cm/min 15.00

Factor 4 D: Trailing

5

2

170.00

10.50

18.00

nil

1

3

170.00

10.50

15.00

nil

3

4

170.00

13.50

15.00

nil

11

5

170.00

13.50

15.00

Ar

7

6

170.00

13.50

18.00

nil

8

7

210.00

13.50

18.00

nil

16

8

210.00

13.50

18.00

Ar

13

9

170.00

10.50

18.00

Ar

4

10

210.00

13.50

15.00

nil

9

11

170.00

10.50

15.00

Ar

10

12

210.00

10.50

15.00

Ar

2

13

210.00

10.50

15.00

nil

6

14

210.00

10.50

18.00

nil

15

15

170.00

13.50

18.00

Ar

14

16

210.00

10.50

18.00

Ar

Ar

161

University of Engineering & Technology, Taxila-Pakistan

Table 5.10 Design of 16 Experiments following Full Factorial (cylinder thickness = 4 mm) Std

Run

12 5 1 3 11 7 8 16 13 4 9 10 2 6 15 14

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Factor 1 A:Current A 220.00 200.00 200.00 200.00 200.00 200.00 220.00 220.00 200.00 220.00 200.00 220.00 220.00 220.00 200.00 220.00

Factor 2 B:Voltage V 13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50

Factor 3 C:Weld Speed cm/min 15.00 18.00 15.00 15.00 15.00 18.00 18.00 18.00 18.00 15.00 15.00 15.00 15.00 18.00 18.00 18.00

Factor 4 D: Trailing Ar nil nil nil Ar nil nil Ar Ar nil Ar Ar nil nil Ar Ar

Table 5.11 Design of 16 Experiments following Full Factorial (cylinder thickness = 5 mm) Std

Run

12 5 1 3 11 7 8 16 13 4 9 10 2 6 15 14

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Factor 1 A:Current A 270.00 230.00 230.00 230.00 230.00 230.00 270.00 270.00 230.00 270.00 230.00 270.00 270.00 270.00 230.00 270.00

Factor 2 B:Voltage V 13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50

Factor 3 C:Weld Speed cm/min 15.00 18.00 15.00 15.00 15.00 18.00 18.00 18.00 18.00 15.00 15.00 15.00 15.00 18.00 18.00 18.00

Factor 4 D: Trailing Ar nil nil nil Ar nil nil Ar Ar nil Ar Ar nil nil Ar Ar

162

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

5.3.3 Response Variables Response variables are the performance measures and the following response variables will be measured in order to judge the process performance of all the virtual experiments designed for welding of samples of 3 x Ø300 x 300 mm, 4 x Ø300 x 300 and 5 x Ø300 x 300 mm respectively. 1. Residual Stress: Max value of weld-induced stresses (von-mises) in the weld zone – to be measured in MPa. 2. Distortion: Max value of weld-induced distortion in the cylinder in weld zone – to be measured in mm. Whereas the values of response as weld strength (MPa) achieved by plate welding of 3, 4 and 5 mm thickness followed a full factorial DOE as discussed in chapter 3 are used for the cylinder welding of same thickness in this chapter. Therefore, only residual stresses and distortion will be determined due to change of shape from plate to cylinder or from linear to circumferential welding of same thickness and DOE. Whereas, the weld length in circumferential welding of Ø300 mm is about twice of weld length in linear welding i.e. 500 mm length of plate in each case.

5.4 Virtual Experiments Results, ANOVA, Regression and Optimization In the following sub-sections, discussion related to effects of predictor variables upon the performance measures (residual stresses and distortion) in circumferential welding is provided with description of ANOVA, regression, and optimization applied to the virtual experimental results.

5.4.1 Distortion Figure 5.20, Figure 5.21 and Figure 5.22 show the comparison of distortion for aforementioned sixteen tests in Table 5.9, Table 5.10 and Table 5.11 respectively. The maximum and minimum values of distortion obtained with respect to thickness of cylinders are presented in Table 5.12. R e s p o n s e o f V ir tu a l E x p e r im e n ts C o n d u c te d ( C y lin d e r T h ic k n e s s = 4

3 .8

3 .7 3 .5

3 .4

3 .4

3 .3

3 .2

3 .1

3 .1 2 .9

3 Distortion (mm)

m m )

3 .6

3 .6

3 .3

3

3 .9

2 .7 2 .3

2

1

0 1

2

3

4

5

6

7 8 9 1 0 E x p e r im e n t N o .

1 1

1 2

1 3

1 4

1 5

1 6

Fig. 5.20 Distortion of Sixteen Experiments (cylinder thickness = 3 mm) 163

University of Engineering & Technology, Taxila-Pakistan

R e s p o n s e o f V ir tu a l E x p e r im e n ts C o n d u c te d ( C y lin d e r T h ic k n e s s = 3 .5

4 m m )

3 .3 3 .1 3 2 .9

2 .9

3 .0

2 .8 2 .7

2 .7

2 .7

2 .6 2 .5

2 .5 2 .4

Distortion (mm)

2 .5

2 .3 2 .1

2 .0

1 .8

1 .5 1 .0 0 .5 0 .0 1

2

3

4

5

6

7 8 9 10 E x p e r im e n t N o .

11

12

13

14

15

16

Fig. 5.21 Distortion of Sixteen Experiments (cylinder thickness = 4 mm) R e s p o n s e o f V ir tu a l E x p e r im e n ts C o n d u c te d ( C y lin d e r T h ic k n e s s = 5 m m ) 3 .0

2 .5

2 .9

2 .4 2 .2

2 .2

Distortion (mm)

2 .1 2 1 .9

2 .0

1 .9 1 .8

1 .8 1 .7 1 .5

1 .5 1 .4

1 .5

1 .4

1 .2

1 .0

0 .5

0 .0 1

2

3

4

5

6

7 8 9 10 E x p e r im e n t N o .

11

12

13

14

15

16

Fig. 5.22 Distortion of Sixteen Experiments (cylinder thickness = 5 mm) Table 5.12 Max. and Min. Values of Distortion Response Cylinder Thickness

Response

Units Minimum

Maximum

Mean

Std. Dev.

t = 3 mm

Distortion

mm

2.3

3.9

3.3

t = 4 mm

Distortion

mm

1.8

3.3

2.643 0.379418

t = 5 mm

Distortion

mm

1.2

2.9

1.868 0.434693

0.417931

The combination of welding current, welding voltage, welding speed and trailing gave the high and low value of distortion at high level of welding parameters without application of trailing and at low level of welding parameters with application of trailing respectively. The results of experiment no. 9 and 10 according to design of experiments applying full factorial as given in Table 5.9, Table 5.10 and Table 5.11 shows the low and high values of distortion at low level of welding current and voltage, high value of welding speed with application of trailing and at high level of welding current and voltage, low level 164

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

of welding speed without application of trailing for cylinder thickness 3, 4 and 5 mm respectively. As the thickness of cylinders increases, the distortion level decreases although welding current level increases with increase of material thickness. As the cylinder thickness increases from 3 mm to 5 mm, the minimum distortion observed decreases from 2.3 mm to 1.2 mm and maximum distortion also decreases from 3.9 mm to 2.9 mm respectively. The mean value of distortion is 1.8 mm to 3.3 mm whereas the standard deviation is 0.4. The results show that with increase of sheet thickness from 30% to 65%, the welding distortion decreases by 25% to 45% in value for 3-5 mm thickness range of HSLA steel cylinders in general. The distortion data were analyzed using ANOVA technique and observations are presented in Table 5.13, Table 5.14 and Table 5.15 for the cylinder thickness of 3, 4 and 5 mm respectively. The analysis shows that effects of three parameters (welding current, welding voltage and welding speed) are significant upon distortion. The effect of application of trailing on distortion is also significant in circumferential welding.

Table 5.13 ANOVA for Distortion (cylinder thickness =3mm) factorial model Source

Sum sqrs

Model 2.36 A-Current 0.30 B-Voltage 0.81 C-Weld Speed 0.25 D-Trailing 1.00 Residual 0.26 Cor Total 2.62

DoF

Mean square F-value

4 1 1 1 1 11 15

0.59 0.30 0.81 0.25 1.00 0.023

25.23 12.92 34.60 10.68 42.72

Prob>F 0.0001 0.0042 0.0001 0.0075 0.0001

Significance Significant Significant Significant Significant Significant

Table 5.14 ANOVA for Distortion (cylinder thickness=4mm) factorial model Source

Sum sqrs

Model 1.95 A-Current 0.28 B-Voltage 0.68 C-Weld Speed 0.23 D- Trailing 0.77 Residual 0.21 Cor Total 2.16

DoF

Mean square F-value

4 1 1 1 1 11 15

0.49 0.28 0.68 0.23 0.77 0.019

25.28 14.31 35.34 11.71 39.75

Prob>F 0.0001 0.0030 0.0001 0.0057 0.0001

Significance Significant Significant Significant Significant Significant

Figure 5.23, Figure 5.24 and Figure 5.25 shows the effects of changing the levels of each parameter upon distortion while keeping other three parameters fixed for cylinder thickness of 3, 4 and 5 mm respectively. It is clear that effects of welding current, voltage and speed are significant. The distortion increases with increase of welding current and voltage values without the application of trailing as heat sink and decreases with increase of welding speed with the application of trailing. 165

University of Engineering & Technology, Taxila-Pakistan

Table 5.12 shows the comparison of weld distortion values obtained for sixteen experiments. The response values for weld distortion (3, 4 and 5 mm thickness) range from 2.3 to 3.9 mm, 1.8 to 3.3 mm and 1.2 to 2.9 mm, providing the ratio of maximum to minimum equal to 1.69, 1.83 and 2.416 respectively. The ratio is small and, thus, there is no need to apply any kind of transformation to the data. Table 5.13, Table 5.14 and Table 5.15 presents the ANOVA details for the suggested factorial model.

Table 5.15 ANOVA for Distortion (cylinder thickness=5mm) factorial model Source

Sum sqrs

Model 2.11 A-Current 0.46 B-Voltage 0.53 C-Weld Speed 0.53 D- Trailing 0.60 Residual 0.73 Cor Total 2.83

DoF

Mean square F-value

4 1 1 1 1 11 15

0.53 0.46 0.53 0.53 0.60 0.066

7.97 6.90 7.95 7.95 9.09

Prob>F 0.0029 0.0236 0.0167 0.0167 0.0118

Significance Significant Significant Significant Significant Significant

Table 5.13, the Model F-value of 25.23 implies the model is significant. There is only a 0.01% chance that a "Model F-Value" this large could occur due to noise. Values of "Prob > F" less than 0.0500 indicate model terms are significant. In this case A, B, C and D are significant model terms. The model possesses the R2 of 90.17%, R2-adjusted of 86.60%, and R2-predicted of 79.21%. The "Pred R-Squared" of 79.21% is in reasonable agreement with the "Adj R-Squared" of 86.60%. "Adeq Precision" measures the signal to noise ratio. A ratio greater than 4 is desirable. In this model, ratio of 17.245 indicates an adequate signal. This model can be used to navigate the design space. Table 5.14, the Model F-value of 25.28 implies the model is significant. There is only a 0.01% chance that a "Model F-Value" this large could occur due to noise. In this case A, B, C and D are significant model terms. The model possesses the R2 of 90.19%, R2adjusted of 86.62%, and R2-predicted of 79.24%. The "Pred R-Squared" of 79.24% is in reasonable agreement with the "Adj R-Squared" of 86.62%. In this model, ratio of 17.401 indicates an adequate signal. This model can also be used to navigate the design space. In Table 5.15, the Model F-value of 7.97 implies the model is significant. There is only a 0.29% chance that a "Model F-Value" this large could occur due to noise. In this case A, B, C and D are significant model terms due to values of "Prob > F" less than 0.0500. The model possesses the R2 of 74.36%, R2-adjusted of 65.03%, and R2-predicted of 45.74%. The "Pred R-Squared" of 45.74% is in reasonable agreement with the "Adj R-Squared" of 65.03%. In this model, the ratio of 10.09 indicates an adequate signal. Therefore model can also be used to navigate the design space.

166

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 5.23 Effects of welding parameters upon Distortion (cylinder thickness = 3 mm)

Fig. 5.24 Effects of welding parameters upon Distortion (cylinder thickness = 4 mm) 167

University of Engineering & Technology, Taxila-Pakistan

Fig. 5.25 Effects of welding parameters upon Distortion (cylinder thickness = 5 mm) The empirical model for the weld distortion (mm), in terms of welding parameters, is as follows: Equation 5.5 (Trailing = nil) and Equation 5.6 (Trailing = Ar) in terms of actual factors for cylinder thickness = 3 mm for weld distortion: Distortion = +1.81875+6.87500E-003* Current+0.15000* Voltage-0.083333* Weld Speed

5.5

Distortion = +1.31875+6.87500E-003 * Current+0.15000* Voltage-0.083333* Weld Speed

5.6

Equation 5.7 (Trailing = nil) and Equation 5.8 (Trailing = Ar) in terms of actual factors for cylinder thickness = 4 mm for weld distortion: Distortion = -0.23750+0.013125* Current+0.13750 * Voltage-0.079167* Weld Speed

5.7

Distortion = -0.67500+0.013125 * Current+0.13750 * Voltage-0.079167 * Weld Speed

5.8

Equation 5.9 (Trailing = nil) and Equation 5.10 (Trailing = Ar) in terms of actual factors for cylinder thickness = 5 mm for weld distortion: Distortion = +0.49688+8.43750E-003 * Current+0.12083 * Voltage-0.12083 * Weld Speed

5.9

Distortion = +0.10938+8.43750E-003 * Current+0.12083* Voltage-0.12083* Weld Speed

5.10

168

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

All the models presented in equations 5.5 to 5.10 for weld distortion in cylinders are valid for the following ranges of input parameters: welding current: 170 to 210 A for 3 mm, 200 to 220 A for 4 mm and 230 to 270 A for 5 mm; welding voltage 10.5 to 13.5 V and welding speed 15 to 18 cm/min. The numerical optimization applied to the distortion data suggests that for any material thickness of cylinders value lying between 3 and 5 mm, the distortion in TIG welding of HSLA steel can be minimized if the trailing is used along with low values of heat input i.e. low values of welding current and welding voltage and high value of welding speed. The predicted weld distortion values are 2.56 mm, 1.96 mm and 1.14 mm for cylinder thickness 3 mm, 4 mm and 5 mm respectively at input parameters as: i)170 A, 10.5 V, 18 cm/min, ii) 200 A, 10.5 V, 18 cm/min and iii) 230 A, 10.5 V, 18 cm/min respectively as shown in Figure 5.26, Figure 5.27 and Figure 5.28 with response desirability of 0.90 for minimization of distortion and maximization of weld strength as shown in Figure 5.38, Figure 5.39 and Figure 5.40 for response desirability with respect to predictors for thickness 3, 4 and 5 mm respectively. Response desirability solutions containing predictors (welding current, welding voltage, welding speed and trailing) and response (weld strength and distortion) with respect to desirability are given in detail for cylinder thickness 3, 4 and 5 mm in Table 5.20, Table 5.21 and Table 5.22 respectively. Each table shows fifty different solutions from maximum (0.90) desirability to minimum (0.50) desirability of response and related input parameters. Combine effect of desirability of different predictors is shown in Figure 5.41, Figure 5.42 and Figure 3.43 for cylinder thickness 3, 4 and 5 mm respectively.

Fig. 5.26 Distortion Predictions w.r.t Predictors (cylinder thickness = 3 mm)

169

University of Engineering & Technology, Taxila-Pakistan

Fig. 5.27 Distortion Predictions w.r.t Predictors (cylinder thickness = 4 mm)

Fig. 5.28 Distortion Predictions w.r.t Predictors (cylinder thickness = 5 mm)

170

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

5.4.2 Weld Induced Residual Stresses Figure 5.29, Figure 5.30 and Figure 5.31 show the comparison of residual stresses for aforementioned sixteen tests in Table 5.5, Table 5.6 and Table 5.7 respectively. The maximum and minimum values of residual stresses obtained with respect to thickness of cylinders are presented in Table 5.16. R e s p o n s e o f V ir tu a l E x p e r im e n ts C o n d u c te d ( C y lin d e r T h ic k n e s s = 6 0 0 505

5 0 0 Residual Stresses (MPa)

3

m m )

577

496

501

498

514

497

489

464 447

445

443

444

1 1

1 2

455

441

416

4 0 0

3 0 0

2 0 0

1 0 0

0 1

2

3

4

5

6

7 8 9 1 0 E x p e r im e n t N o .

1 3

1 4

1 5

1 6

Fig. 5.29 Residual Stresses of Sixteen Experiments (cylinder thickness = 3 mm) R e s p o n s e

o f V ir tu a l E x p e r im e n ts

C o n d u c te d

( C y lin d e r T h ic k n e s s

=

4

m m )

5 0 0 4 6 4

Residual Stresses (MPa)

4 3 3

4 2 1 3 9 4

4 0 0

4 0 3

3 9 8

3 8 7 3 7 0

3 6 0

3 5 4

3 4 8

3 5 2

1 1

1 2

3 7 5 3 4 8

3 5 7

1 5

1 6

3 2 4

3 0 0

2 0 0

1 0 0

0 1

2

3

4

5

6

7 8 9 E x p e r im e n t

1 0 N o .

1 3

1 4

Fig. 5.30 Residual Stresses of Sixteen Experiments (cylinder thickness = 4 mm) R e s p o n s e

o f V ir tu a l E x p e r im e n ts C o n d u c te d ( C y lin d e r T h ic k n e s s =

37 0

m m )

36 1 332

Residual Stresses (MPa)

5

398

4 0 0 33 5

340

330

329

320

30 7

302

3 06

1 1

1 2

3 33 3 05 293

3 0 0 26 8

2 0 0

1 0 0

0 1

2

3

4

5

6

7 8 9 E x p e r im e n t

1 0 No .

1 3

1 4

1 5

1 6

Fig. 5.31 Residual Stresses of Sixteen Experiments (cylinder thickness = 5 mm) 171

University of Engineering & Technology, Taxila-Pakistan

Table 5.16 Max. and Min. Values of Residual Stresses Response Cylinder Thickness

Response

Units Minimum

Maximum

Mean

Std. Dev.

t = 3 mm

Residual Stresses

MPa

416

577

477

t = 4 mm

Residual Stresses

MPa

324

464

380.5 36.8221

t = 5 mm

Residual Stresses

MPa

268

398

326.8 31.6106

39.9283

The combination of welding current, welding voltage, welding speed and trailing gave the high and low value of residual stresses at high level of welding parameters without application of trailing and at low level of welding parameters with application of trailing respectively. The results of experiment no. 9 and 10 according to design of experiments applying full factorial as given in Table 5.9, Table 5.10 and Table 5.11 shows the low and high values of residual stresses at low level of welding current and voltage, high value of welding speed with application of trailing and at high level of welding current and voltage, low level of welding speed without application of trailing for cylinder thickness 3, 4 and 5 mm respectively. As the cylinder thickness increases the residual stresses level decreases although welding current level increases with increase of material thickness. As the cylinder thickness increases from 3 mm to 5 mm, the minimum residual stresses observed decreases from 416 MPa to 268 MPa and maximum residual stresses also decreases from 577 MPa to 398 MPa respectively. The mean value of residual stresses is 327 MPa to 477 MPa whereas the standard deviation is 36.8. The results show that with increase of cylinder thickness from 30% to 65%, the welding residual stresses decreases by 30% to 35% in value for 3-5 mm thickness range of HSLA steel cylinder in general. The residual stresses data were analyzed using ANOVA technique and observations are presented in Table 5.17, Table 5.18 and Table 5.19 for the cylinder thickness of 3, 4 and 5 mm respectively. The analysis shows that effects of three parameters (welding current, welding voltage and welding speed) are significant upon residual stresses. The effect of application of trailing on residual stresses is also significant in circumferential welding.

Table 5.17 ANOVA for Residual Stresses (cylinder thickness =3mm) factorial model Source

Sum sqrs

Model 20488.00 A-Current 3136.00 B-Voltage 3136.00 C-Weld Speed 2116.00 D-Trailing 12100.00 Residual 3426.00 Cor Total 23914.00

DoF

Mean square F-value

4 1 1 1 1 11 15

5122.00 3136.00 3136.00 2116.00 12100.00 311.45

16.45 10.07 10.07 6.79 38.85

Prob>F 0.0001 0.0089 0.0089 0.0244 0.0001

Significance Significant Significant Significant Significant Significant

172

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 5.18 ANOVA for Residual Stresses (cylinder thickness=4mm) factorial model Source

Sum sqrs

Model 17019.75 A-Current 3306.25 B-Voltage 5402.25 C-Weld Speed 2070.25 D-Trailing 6241.00 Residual 3318.25 Cor Total 20338.00

DoF

Mean square F-value

4 1 1 1 1 11 15

4254.94 3306.25 5402.25 2070.25 6241.00 301.66

14.11 10.96 17.91 6.86 20.69

Prob>F 0.0003 0.0069 0.0014 0.0238 0.0008

Significance Significant Significant Significant Significant Significant

Table 5.19 ANOVA for Residual Stresses (cylinder thickness=5mm) factorial model Source

Sum sqrs

Model 12081.25 A-Current 2889.06 B-Voltage 2730.06 C-Weld Speed 1314.06 D-Trailing 5148.06 Residual 2907.19 Cor Total 14988.44

DoF

Mean square F-value

4 1 1 1 1 11 15

3020.31 2889.06 2730.06 1314.06 5148.06 264.29

11.43 10.93 10.33 4.97 19.48

Prob>F 0.0007 0.0070 0.0082 0.0475 0.0010

Significance Significant Significant Significant Significant Significant

Figure 5.32, Figure 5.33 and Figure 5.34 shows the effects of changing the levels of each parameter upon residual stresses while keeping other three parameters fixed for cylinder thickness of 3, 4 and 5 mm respectively. It is clear that effects of welding current, welding voltage, welding speed and trailing are significant. The residual stresses increases with increase of welding current and voltage values without the application of trailing as heat sink and decreases with increase of welding speed with the application of trailing. Table 5.16 shows the comparison of weld residual stresses values obtained for sixteen experiments. The response values for weld residual stresses (3, 4 and 5 mm thickness cylinders) range from 416 to 577 MPa, 324 to 464 MPa and 268 to 398 MPa, providing the ratio of maximum to minimum equal to 1.38, 1.43 and 1.48 respectively. The ratio is small and, thus, there is no need to apply any kind of transformation to the data. Table 5.17, Table 5.18 and Table 5.19 presents the ANOVA details for the suggested factorial model. Table 5.17, the Model F-value of 16.45 implies the model is significant. There is only a 0.01% chance that a "Model F-Value" this large could occur due to noise. In this case A, B, C and D are significant model terms. The model possesses the R2 of 85.67%, R2adjusted of 80.46%, and R2-predicted of 69.69%. The "Pred R-Squared" of 69.69% is in reasonable agreement with the "Adj R-Squared" of 80.46%. "Adeq Precision" measures the signal to noise ratio. A ratio greater than 4 is desirable. In this model, ratio of 13.583 indicates an adequate signal. This model can be used to navigate the design space.

173

University of Engineering & Technology, Taxila-Pakistan

Fig. 5.32 Effects of welding parameters upon Residual Stresses (cylinder thickness = 3 mm)

Fig. 5.33 Effects of welding parameters upon Residual Stresses (cylinder thickness = 4 mm) 174

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 5.34 Effects of welding parameters upon Residual Stresses (cylinder thickness = 5 mm)

Table 5.18, the Model F-value of 14.11 implies the model is significant. There is only a 0.03% chance that a "Model F-Value" this large could occur due to noise. In this case A, B, C and D are significant model terms. The model possesses the R2 of 83.68%, R2adjusted of 77.75%, and R2-predicted of 65.48%. The "Pred R-Squared" of 65.48% is in reasonable agreement with the "Adj R-Squared" of 77.75%. In this model, ratio of 13.158 indicates an adequate signal and this model can be used to navigate the design space. In Table 5.19, the Model F-value of 11.43 implies the model is significant. There is only a 0.07% chance that a "Model F-Value" this large could occur due to noise. In this case A, B, C and D are significant model terms due to values of "Prob > F" less than 0.0500. The model possesses the R2 of 80.60%, R2-adjusted of 73.55%, and R2-predicted of 58.96%. The "Pred R-Squared" of 58.96% is in reasonable agreement with the "Adj R-Squared" of 73.55%. In this model, ratio of 11.774 indicates an adequate signal. This model can also be used to navigate the design space. The empirical model for the weld residual stresses (MPa), in terms of welding parameters, is as follows: Equation 5.11 (Trailing = nil) and Equation 5.12 (Trailing = Ar) in terms of actual factors for cylinder thickness = 3 mm for weld residual stresses: 175

University of Engineering & Technology, Taxila-Pakistan

Residual Stresses = +386.00000+0.70000* Current+9.33333 * Voltage-7.66667 * Weld Speed

5.11

Residual Stresses = +331.00000+0.70000 * Current+9.33333 * Voltage-7.66667 * Weld Speed

5.12

Equation 5.13 (Trailing = nil) and Equation 5.14 (Trailing = Ar) in terms of actual factors for cylinder thickness = 4 mm for weld residual stresses: Residual Stresses = +76.50000+1.43750 * Current+12.25000 * Voltage-7.58333 * Weld Speed

5.13

Residual Stresses = +37.00000+1.43750 * Current+12.25000 * Voltage-7.58333 * Weld Speed

5.14

Equation 5.15 (Trailing = nil) and Equation 5.16 (Trailing = Ar) in terms of actual factors for cylinder thickness = 5 mm for weld residual stresses: Residual Stresses = +171.96875+0.67187 * Current+8.70833 * Voltage-6.04167 * Weld Speed

5.15

Residual Stresses = +136.09375+0.67187 * Current+8.70833 * Voltage-6.04167 * Weld Speed

5.16

All the models presented in equations 5.11 to 5.16 for weld residual stresses in cylinders are valid for the following ranges of input parameters: welding current: 170 to 210 A for 3 mm, 200 to 220 A for 4 mm and 230 to 270 A for 5 mm; welding voltage 10.5 to 13.5 V and welding speed 15 to 18 cm/min. The numerical optimization applied to the residual stresses data suggests that for any material cylinder thickness value lying between 3 and 5 mm, the residual stresses in TIG welding of HSLA steel can be minimized if the trailing is used along with low values of heat input i.e. low values of welding current and welding voltage and high value of welding speed. The predicted weld residual stresses values are 410 MPa, 316 MPa and 273 MPa for cylinder thickness 3 mm, 4 mm and 5 mm respectively at input parameters as: i)170 A, 10.5 V, 18 cm/min, ii) 200 A, 10.5 V, 18 cm/min and iii) 230 A, 10.5 V, 18 cm/min respectively as shown in Figure 5.35, Figure 5.36 and Figure 5.37 with response desirability of 0.90 for minimization of distortion and residual stresses as shown in Figure 5.38, Figure 5.39 and Figure 5.40 for response desirability with respect to predictors for cylinder thickness 3, 4 and 5 mm respectively. Response desirability solutions containing predictors (welding current, welding voltage, welding speed and trailing) and response (distortion and residual stresses) with respect to desirability are given in detail for cylinder thickness 3, 4 and 5 mm in Table 5.20, Table 5.21 and Table 5.22 respectively. Each table shows fifty different solutions from maximum (0.90) desirability to minimum (0.50) desirability of response and related input parameters. Combine effect of desirability of different predictors is shown in Figure 5.41, Figure 5.42 and Figure 5.43 for cylinder thickness 3, 4 and 5 mm respectively. Comparison of responses achieved is shown in Figure 5.44 which shows the effect of increase or decrease of distortion & residual stresses respectively.

176

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 5.35 Residual Stresses Predictions w.r.t Predictors (cylinder thickness = 3 mm)

Fig. 5.36 Residual Stresses Predictions w.r.t Predictors (cylinder thickness = 4 mm)

Fig. 5.37 Residual Stresses Predictions w.r.t Predictors (cylinder thickness = 5 mm) 177

University of Engineering & Technology, Taxila-Pakistan

Table 5.20 Response Desirability Solutions (cylinder thickness = 3 mm) Number Current Voltage Weld Trailing Speed

Distortion

Residual Stresses

Desirability

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51

2.5625 2.56354 2.56414 2.56457 2.56693 2.56768 2.57023 2.57362 2.57374 2.57699 2.58088 2.58427 2.58862 2.58967 2.60046 2.61961 2.62153 2.62294 2.6293 2.64247 2.64402 2.64399 2.67057 2.69949 2.69042 2.70302 2.7158 3.0625 3.06366 3.06451 3.06472 3.06547 3.06697 3.06802 3.07102 3.0745 3.07344 3.07459 3.07549 3.0837 3.08953 3.09494 3.09843 3.09796 3.12306 3.13221 3.12929 3.13545 3.19949 3.20639 3.22774

410 410.096 410.163 410.129 410.276 410.477 410.692 410.692 411.144 410.902 411.87 412.002 412.66 412.767 413.493 415.814 415.431 415.561 416.801 414.976 417.5 417.61 419.942 418.524 421.769 422.928 424.104 465 465.118 465.126 465.204 465.302 465.411 465.343 465.53 465.747 466.114 466.231 466.195 466.319 467.486 467.019 467.236 468.262 470.571 469.338 471.8 471.711 473.524 478.237 480.202

0.914 0.914 0.914 0.914 0.913 0.913 0.912 0.910 0.910 0.909 0.908 0.907 0.905 0.905 0.901 0.895 0.894 0.893 0.889 0.887 0.882 0.882 0.866 0.859 0.854 0.846 0.838 0.603 0.603 0.602 0.602 0.602 0.601 0.601 0.599 0.597 0.596 0.596 0.596 0.592 0.587 0.586 0.584 0.582 0.567 0.566 0.561 0.559 0.530 0.516 0.503

170.00 170.00 170.18 170.00 170.00 170.00 170.00 170.00 171.63 170.00 172.67 170.00 173.80 173.95 170.01 178.30 170.00 170.00 179.72 170.00 170.00 171.68 170.00 170.00 170.00 170.00 170.00 170.00 170.17 170.00 170.00 170.43 170.00 170.00 170.00 170.00 171.59 171.76 170.00 170.00 170.00 170.00 170.00 170.00 170.00 170.00 179.71 170.00 170.00 170.00 170.00

10.50 10.50 10.50 10.51 10.53 10.50 10.50 10.57 10.50 10.60 10.50 10.50 10.50 10.50 10.50 10.50 10.50 10.50 10.50 11.03 10.50 10.50 10.50 11.41 10.50 10.50 10.50 10.50 10.50 10.51 10.50 10.50 10.50 10.54 10.56 10.58 10.50 10.50 10.50 10.64 10.50 10.72 10.74 10.50 10.50 10.96 10.50 10.50 11.41 10.50 10.50

18.00 17.99 17.99 18.00 18.00 17.94 17.92 18.00 18.00 18.00 18.00 17.74 18.00 18.00 17.55 18.00 17.29 17.27 18.00 18.00 17.02 17.16 16.70 18.00 16.46 16.31 16.16 18.00 18.00 18.00 17.97 18.00 17.95 18.00 18.00 18.00 18.00 18.00 17.84 18.00 17.68 18.00 18.00 17.57 17.27 18.00 18.00 17.12 18.00 16.27 16.02

Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil

178

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 5.21 Response Desirability Solutions (cylinder thickness = 4 mm) Number Current Voltage Weld Trailing Speed

Distortion

Residual Stresses

Desirability

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46

1.96875 1.96992 1.97027 1.97166 1.97367 1.97723 1.97908 1.97982 1.9804 1.98285 1.98867 1.98912 1.99109 1.99164 1.99204 1.9976 2.00961 2.02617 2.02852 2.03202 2.03266 2.06231 2.06542 2.09432 2.0842 2.09654 2.13271 2.16851 2.40625 2.40861 2.40971 2.41117 2.41148 2.41272 2.41456 2.41552 2.41633 2.41688 2.41777 2.42063 2.42259 2.43855 2.46378 2.47016 2.51169 2.53182

316.625 316.737 316.76 316.928 317.096 317.554 317.545 317.685 317.899 318.169 318.533 318.856 318.615 318.817 319.176 319.785 320.539 322.125 323.171 322.686 323.625 324.966 327.212 327.812 329.27 328.866 334.583 335.76 356.125 356.351 356.456 356.596 356.591 356.834 356.866 357.14 357.194 357.287 357.387 357.406 357.581 359.219 361.636 363.125 366.225 367.312

0.942 0.942 0.942 0.941 0.940 0.939 0.938 0.938 0.938 0.937 0.935 0.935 0.934 0.934 0.934 0.932 0.928 0.922 0.921 0.919 0.919 0.905 0.897 0.884 0.883 0.880 0.848 0.831 0.678 0.676 0.675 0.674 0.674 0.673 0.672 0.671 0.670 0.670 0.669 0.668 0.667 0.656 0.638 0.631 0.606 0.595

200.00 200.00 200.00 200.14 200.00 200.65 200.00 200.00 200.88 201.07 200.00 201.55 200.00 200.00 201.77 202.20 200.00 200.00 204.55 200.00 204.87 200.02 207.36 200.00 208.80 200.00 212.49 200.00 200.00 200.00 200.00 200.00 200.00 200.49 200.00 200.71 200.64 200.80 200.88 200.00 200.00 200.00 200.00 204.87 200.00 200.00

10.50 10.50 10.51 10.50 10.50 10.50 10.58 10.50 10.50 10.50 10.50 10.50 10.66 10.50 10.50 10.50 10.50 10.50 10.50 10.50 10.50 11.18 10.50 11.41 10.50 10.50 10.50 10.50 10.50 10.50 10.50 10.50 10.54 10.50 10.56 10.50 10.51 10.50 10.50 10.60 10.62 10.50 10.50 10.50 10.50 11.41

18.00 17.99 18.00 17.99 17.94 18.00 18.00 17.86 18.00 18.00 17.75 18.00 18.00 17.71 18.00 18.00 17.48 17.27 18.00 17.20 18.00 18.00 18.00 18.00 18.00 16.39 18.00 15.48 18.00 17.97 17.96 17.94 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00 17.59 17.27 18.00 16.67 18.00

Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil

179

University of Engineering & Technology, Taxila-Pakistan

Table 5.22 Response Desirability Solutions (cylinder thickness = 5 mm) Number Current Voltage Weld Trailing Speed

Distortion

Residual Stresses

Desirability

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50

1.14375 1.14519 1.146 1.14761 1.14834 1.14976 1.15056 1.15269 1.15208 1.15479 1.15497 1.16043 1.16364 1.15745 1.16118 1.16318 1.17692 1.17036 1.19074 1.19223 1.23766 1.24255 1.24377 1.25222 1.26025 1.28619 1.53125 1.53428 1.53443 1.53712 1.53994 1.53845 1.54028 1.54209 1.54514 1.54671 1.55035 1.55647 1.55501 1.5644 1.56054 1.56238 1.57524 1.57009 1.63972 1.64775 1.76235 1.80629 1.82298 1.86125

273.313 273.427 273.478 273.506 273.678 273.745 273.75 273.957 273.975 274.055 274.121 274.147 274.307 274.403 274.568 274.86 274.971 275.231 276.699 276.807 278.008 280.454 281.277 281.95 281.709 284.654 309.188 309.345 309.417 309.611 309.622 309.761 309.838 310.051 310.293 310.301 310.143 310.449 310.9 310.845 311.52 311.665 311.387 312.28 317.825 317.584 320.743 322.939 323.774 325.688

0.979 0.979 0.979 0.979 0.978 0.978 0.978 0.977 0.977 0.976 0.976 0.976 0.975 0.975 0.974 0.973 0.973 0.972 0.966 0.966 0.950 0.939 0.935 0.930 0.929 0.910 0.742 0.740 0.740 0.738 0.737 0.737 0.736 0.735 0.733 0.733 0.732 0.730 0.728 0.726 0.724 0.723 0.721 0.718 0.676 0.675 0.631 0.609 0.601 0.583

230.00 230.17 230.05 230.00 230.54 230.00 230.00 230.00 230.99 230.00 230.00 230.00 230.00 231.62 230.00 232.30 230.00 230.00 230.00 230.00 230.00 230.33 241.85 242.86 230.00 246.87 230.00 230.00 230.00 230.02 230.00 230.85 230.00 231.28 231.65 230.00 230.00 230.00 230.00 230.00 233.47 233.67 230.00 234.60 242.86 230.00 230.00 230.00 230.00 230.00

10.50 10.50 10.51 10.50 10.50 10.55 10.54 10.57 10.50 10.57 10.59 10.50 10.50 10.50 10.64 10.50 10.50 10.72 10.89 10.90 10.50 11.29 10.50 10.50 11.46 10.50 10.50 10.50 10.53 10.55 10.50 10.50 10.57 10.50 10.50 10.63 10.50 10.50 10.70 10.50 10.50 10.50 10.50 10.50 10.50 11.46 10.50 10.50 10.50 10.50

18.00 18.00 18.00 17.97 18.00 18.00 17.98 18.00 18.00 17.98 18.00 17.86 17.84 18.00 18.00 18.00 17.73 18.00 18.00 18.00 17.22 18.00 18.00 18.00 18.00 18.00 18.00 17.98 18.00 18.00 17.93 18.00 18.00 18.00 18.00 18.00 17.84 17.79 18.00 17.73 18.00 18.00 17.64 18.00 18.00 18.00 16.09 15.72 15.59 15.27

Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil nil

180

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 5.38 Response Desirability w.r.t Predictors (cylinder thickness = 3 mm)

Fig. 5.39 Response Desirability w.r.t Predictors (cylinder thickness = 4 mm)

Fig. 5.40 Response Desirability w.r.t Predictors (cylinder thickness = 5 mm)

181

University of Engineering & Technology, Taxila-Pakistan

Fig. 5.41 Effect on Desirability of different Predictors (cylinder thickness = 3 mm)

Fig. 5.42 Effect on Desirability of different Predictors (cylinder thickness = 4 mm)

Fig. 5.43 Effect on Desirability of different Predictors (cylinder thickness = 5 mm) 182

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

R e s p o n s e C o m p a r is o n fo r C y lin d e r T h ic k n e s s = 3 m m S c a to r P lo t

Residual Stresses (MPa)

600

550

500

450

400 2 .5

3 .0 D is t o r t io n ( m m )

3 .5

4 .0

R e s p o n s e C o m p a r is o n fo r C y lin d e r T h ic k n e s s = 4 m m S c a to r P lo t 480

Residual Stresses (MPa)

460 440 420 400 380 360 340 320 300 2 .0

2 .5 D is t o r t io n ( m m )

3 .0

3 .5

R e s p o n s e C o m p a r is o n fo r C y lin d e r T h ic k n e s s = 5 m m S c a to r P lo t

Residual Stresses (MPa)

400

375 350 325

300 275 250 1 .0

1 .5

2 .0 D is t o r t io n ( m m )

2 .5

3 .0

Fig. 5.44 Response Comparison Scator Plots (cylinder thickness = 3 mm, 4 mm, 5 mm)

183

University of Engineering & Technology, Taxila-Pakistan

5.4.3 Numerical Optimization & Empirical Modeling using Response Surface Method By performing the virtual experimentation and analysis following the 2-level full factorial design (three numeric factors and one categoric factor) as given in Table 5.9, Table 5.10 and Table 5.11 for the cylinder thicknesses of 3, 4 and 5 mm respectively for the responses (distortion and residual stresses), the effect of cylinder thickness as another variable with earlier variables (i.e. four numeric factors and one categoric factor) is analyzed by employing response surface method (RSM) as discussed in chapter 3. The low and high settings of factors are given in Table 5.23 and the complete historical data (3x16 observations for each thickness) is given in Table 5.24 with response values. The summary of design is given in the Table 5.25. During model analysis, fit summary suggested to use 2FI model for analysis of variance. ANOVA results are shown in Table 3.26 and Table 3.27 for weld distortion and residual stresses respectively. The values of R-Squared, Adj R-Squared and Pred R-Squared including Adeq Precision (max to min ratio) are given in Table 5.28. Figure 5.45 and Figure 3.46 shows the effects of changing the levels of each parameter upon distortion and residual stresses while keeping other parameters fixed respectively. It is clear that effects of welding current, voltage, speed and thickness are significant. The distortion and residual stresses increases with increase of welding current and voltage values without the application of trailing and decreases with increase of welding speed and thickness with the application of trailing. Figure 5.47 and Figure 3.48 shows the interaction effect of welding parameters upon distortion and residual stresses respectively. From Table 5.26 and Table 5.27, the significance of all models is clear from the Model F-value which is greater than 4. As the values of "Prob > F" less than 0.0500 indicate the significance of model terms, in this case A, B, C, D and E terms are significant with some interaction terms. The Table 5.28 shows that the "Pred R-Squared" values are in reasonable agreement with the "Adj R-Squared" values and “Adeq Precision” values indicate an adequate signal that shows the models can be used to navigate the design space.

Table 5.23 High and Low Settings of Factors for RSM Factor Name

Units

Type

A B C D E

A V cm/min mm

Numeric Numeric Numeric Numeric Categoric

Current Voltage Weld Speed Thickness Trailing

Low Actual 170.00 10.50 15.00 3.0 nil

High Actual 270.0 13.50 18.00 5.0 Ar

Mean 216.66 12.0 16.5 4.0

184

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 5.24 Historical Data (48 (3x16) observations) including Response Values for RSM Run

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48

Factor 1 A:Current (A)

Factor 2 B:Voltage (V)

210.00 170.00 170.00 170.00 170.00 170.00 210.00 210.00 170.00 210.00 170.00 210.00 210.00 210.00 170.00 210.00 220.00 200.00 200.00 200.00 200.00 200.00 220.00 220.00 200.00 220.00 200.00 220.00 220.00 220.00 200.00 220.00 270.00 230.00 230.00 230.00 230.00 230.00 270.00 270.00 230.00 270.00 230.00 270.00 270.00 270.00 230.00 270.00

13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50 13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50 13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50

Factor 3 Factor 4 Factor 5 Responses C:Weld Speed D: Thickness D: Trailing Distortion (cm/min) (mm) (mm) 15.00 3.00 Ar 3.4 18.00 3.00 nil 3.1 15.00 3.00 nil 3.3 15.00 3.00 nil 3.7 15.00 3.00 Ar 3.5 18.00 3.00 nil 3.6 18.00 3.00 nil 3.8 18.00 3.00 Ar 3.2 18.00 3.00 Ar 2.3 15.00 3.00 nil 3.9 15.00 3.00 Ar 2.7 15.00 3.00 Ar 3.3 15.00 3.00 nil 3.6 18.00 3.00 nil 3.4 18.00 3.00 Ar 3.1 18.00 3.00 Ar 2.9 15.00 4.00 Ar 2.7 18.00 4.00 nil 2.4 15.00 4.00 nil 2.6 15.00 4.00 nil 3 15.00 4.00 Ar 2.8 18.00 4.00 nil 2.9 18.00 4.00 nil 3.1 18.00 4.00 Ar 2.5 18.00 4.00 Ar 1.8 15.00 4.00 nil 3.3 15.00 4.00 Ar 2.1 15.00 4.00 Ar 2.7 15.00 4.00 nil 2.9 18.00 4.00 nil 2.7 18.00 4.00 Ar 2.5 18.00 4.00 Ar 2.3 15.00 5.00 Ar 2.4 18.00 5.00 nil 1.9 15.00 5.00 nil 2.2 15.00 5.00 nil 2.1 15.00 5.00 Ar 1.8 18.00 5.00 nil 1.5 18.00 5.00 nil 2.2 18.00 5.00 Ar 2 18.00 5.00 Ar 1.2 15.00 5.00 nil 2.9 15.00 5.00 Ar 1.4 15.00 5.00 Ar 1.7 15.00 5.00 nil 1.9 18.00 5.00 nil 1.8 18.00 5.00 Ar 1.5 18.00 5.00 Ar 1.4

Residual Stresses (MPa) 505 464 496 501 445 498 514 447 416 577 443 444 497 489 441 455 433 360 394 403 354 398 421 370 324 464 348 352 387 375 348 357 370 332 335 340 307 330 361 320 268 398 302 306 329 333 293 305

185

University of Engineering & Technology, Taxila-Pakistan

Table 5.25 Max. and Min. Values of Responses in RSM Response

Units

Minimum

Maximum

Mean

Std. Dev.

Distortion

mm

1.2

3.9

2.60417

0.715172

Residual Stress

MPa

268

577

394.771

72.1374

Table 5.26 ANOVA for Distortion (2FI model) of RSM Source

Sum sqrs

Model

22.51

A-Current B-Voltage

DoF

Mean square

F-value

Prob>F

Significance

5

4.50

123.96

0.0001

Significant

0.82

1

0.82

22.59

0.0001

Significant

2.00

1

2.00

55.08

0.0001

Significant

C-Weld Speed 0.96

1

0.96

26.52

0.0001

Significant

D-Thickness 9.76

1

9.76

268.60

0.0001

Significant

E-Trailing

2.34

1

2.34

64.44

0.0001

Significant

Residual

1.53

42

0.036

Cor Total

24.04

47

Table 5.27 ANOVA for Residual Stresses (2FI model) of RSM Source

DoF

Mean square

F-value

Prob>F

2.311E+005

5

46222.18

144.15

0.0001

Significant

A-Current

11546.04

1

11546.04

36.01

0.0001

Significant

B-Voltage

11011.02

1

11011.02

34.34

0.0001

Significant

C-Weld Speed 5440.02

1

5440.02

16.97

0.0002

Significant

D-Thickness 1.141E+005

1

1.141E+005

355.79

0.0001

Significant

E-Trailing

22663.52

1

22663.52

70.68

0.0001

Significant

Residual

13467.60

42

320.66

Cor Total

2.446E+005

47

Model

Sum sqrs

Significance

Table 5.28 ANOVA Summary for RSM (2FI model) Response

Adeq Precision

R-Squared

Adj R-Squared

Pred

R-Squared

Distortion

42.383

93.65%

92.90%

91.68%

Residual Stress

44.193

94.49%

93.84%

92.83%

186

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 5.45 Effects of welding parameters upon Distortion in RSM

Fig. 5.46 Effects of welding parameters upon Residual Stresses in RSM 187

University of Engineering & Technology, Taxila-Pakistan

Fig. 5.47 Interaction of welding parameters upon Distortion in RSM

Fig. 5.48 Interaction of welding parameters upon Residual Stresses in RSM

188

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

The empirical models for the responses (distortion (mm) and weld residual stresses (MPa)), in terms of welding parameters (welding current, welding voltage, welding speed and thickness of cylinders with or without application of trailing), are as follows: Equation 5.17 (Trailing = nil) and Equation 5.18 (Trailing = Ar) in terms of actual factors for distortion:

Distortion = + 4.90833+7.28448E-003* Current + 0.13611* Voltage -

0.094444* Weld Speed - 0.93416* Thickness

Distortion = + 4.46667+7.28448E-003* Current + 0.13611* Voltage - 0.094444 * Weld Speed - 0.93416* Thickness

5.17

5.18

Equation 5.19 (Trailing = nil) and Equation 5.20 (Trailing = Ar) in terms of actual factors for weld residual stresses:

Residual Stresses = + 629.29167+0.86401* Current + 10.09722* Voltage -

7.09722 * Weld Speed - 101.01401* Thickness

Residual Stresses = + 585.83333 + 0.86401* Current + 10.09722* Voltage - 7.09722* Weld Speed - 101.01401* Thickness

5.19

5.20

All the models presented in equations 5.17 to 5.20 for distortion and weld residual stresses in circumferential welding are valid for the following ranges of input parameters: welding current: 170 to 210 A for 3 mm, 200 to 220 A for 4 mm and 230 to 270 A for 5 mm; welding voltage 10.5 to 13.5 V and welding speed 15 to 18 cm/min. The numerical optimization applied to the distortion and residual stresses data suggests that for any material thickness value lying between 3 and 5 mm, the distortion and residual stresses in TIG welding of HSLA steel can be minimized if the trailing is used along with low values of heat input i.e. low values of welding current and welding voltage and high value of welding speed. The predicted weld distortion and weld residual stresses values are as: i) 3.06 mm and 450.59 MPa, ii) 2.73 mm and 384.89MPa, and iii) 2.24 mm and 331 MPa at input parameters as: i) 170 A, 10.5 V, 18 cm/min, 3 mm, ii) 200 A, 10.5 V, 18 cm/min, 4 mm and iii) 230 A, 10.5 V, 18 cm/min, 5 mm respectively as shown in Figure 5.49 and Figure 5.50 with response desirability for minimization of distortion and residual stresses respectively. Response desirability solutions containing predictors (welding current, welding voltage, welding speed, thickness and trailing) and response (distortion and residual stresses) with respect to desirability are given in detail in Table 5.29.

189

University of Engineering & Technology, Taxila-Pakistan

Fig. 5.49 Distortion Predictions w.r.t Predictors in RSM

Fig. 5.50 Residual Stresses Predictions w.r.t Predictors in RSM 190

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 5.29 Response Desirability Solutions of RSM Number Current Voltage Weld ThicknessTrailing Speed

Distortion Residual Desirability Stresses

For Goal: Current is equal to 230 A. Thickness is equal to 5 mm. Voltage, Speed and Trailing are in range. 1 2 3

230.00 230.00 230.00

10.50 10.50 10.50

18.00 18.00 17.67

5.00 5.00 5.00

Ar nil nil

1.20047 257.756 1.000 1.64213 301.214 0.864 1.67366 303.584 0.854

For Goal: Current is equal to 200 A. Thickness is equal to 4 mm. Voltage, Speed and Trailing are in range. 1 2 3 4 5 6

200.00 200.00 200.00 200.00 200.00 200.00

10.50 10.50 10.50 10.50 10.50 10.50

18.00 17.96 18.00 17.89 17.74 17.71

4.00 4.00 4.00 4.00 4.00 4.00

Ar Ar nil nil nil nil

1.91609 1.91953 2.35776 2.36847 2.38259 2.3847

332.85 333.109 376.308 377.113 378.174 378.333

0.762 0.761 0.609 0.606 0.601 0.601

For Goal: Current is equal to 170 A. Thickness is equal to 3 mm. Voltage, Speed and Trailing are in range. 1 2 3 4 5 6

170.00 170.00 170.00 170.00 170.00 170.00

10.50 10.71 10.50 10.50 10.58 10.63

18.00 18.00 18.00 17.90 18.00 18.00

3.00 3.00 3.00 3.00 3.00 3.00

Ar Ar nil nil nil nil

2.63172 2.66096 3.07338 3.08287 3.08365 3.09046

407.944 410.113 451.402 452.115 452.164 452.669

0.507 0.498 0.353 0.350 0.350 0.347

241.877 229.346 238.509 241.414 236.726 226.803 235.217 243.732 249.911 238.568 251.457 252.71 245.96 242.702 247.45 248.359 247.464 233.735 214.292 239.733 252.738 249.335 220.11 232.786 247.043 238.564 235.005 241.301 253.963 241.659

1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000

For Goal: Current, Thickness, Voltage, Speed and Trailing are in range. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

191.61 173.76 174.47 186.41 170.47 170.87 171.66 181.15 217.84 183.61 220.32 175.04 170.23 184.11 170.48 171.96 195.36 173.76 171.84 175.18 214.08 170.23 172.43 174.72 207.82 170.09 176.31 196.91 211.25 171.55

11.15 10.77 10.88 10.52 12.75 10.57 11.43 11.96 10.67 11.59 10.56 10.56 10.56 10.92 11.20 10.75 10.51 12.23 10.56 11.70 10.56 10.84 11.10 11.61 10.81 12.39 12.74 11.54 10.78 10.91

17.24 16.27 15.02 15.33 17.41 16.16 15.70 16.97 17.90 17.06 17.91 17.92 17.49 15.16 17.96 17.26 16.51 17.73 17.56 16.57 17.65 17.97 17.84 17.84 17.87 16.16 17.91 17.85 17.86 16.20

4.95 4.95 4.96 4.98 4.97 4.94 4.98 4.94 5.00 4.97 4.99 4.59 4.65 5.00 4.67 4.67 4.91 4.95 4.97 4.93 4.94 4.61 4.95 4.90 4.96 5.00 5.00 4.99 4.92 4.83

Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar

1.13014 1.03869 1.16358 1.1578 1.1612 1.01188 1.13915 1.19723 1.14671 1.12958 1.15451 1.19767 1.15087 1.19355 1.17809 1.18643 1.17333 1.09964 0.85481 1.16496 1.17795 1.18028 0.92321 1.06062 1.13387 1.19903 1.12607 1.12034 1.19487 1.16213

191

University of Engineering & Technology, Taxila-Pakistan

5.5 Linear & Circumferential Welding Optimization using RSM After performing the analysis and optimization of linear and circumferential welds by full factorial and RSM in chapter 3 and chapter 5 respectively, the effect of linear or circumferential welding as another category parameter with earlier parameters (i.e. four numeric factors and one categoric factor) is analyzed by employing response surface method (RSM). Now the total parameters are six i.e. four numeric (thickness, current, voltage & speed) and two categoric (trailing and weld type i.e. linear or circumferential). The low and high settings of factors are given in Table 5.30 and the complete historical data ( 96 (2x3x16) observations for each thickness and weld type) is given in Table 5.31 with response values. During model analysis, fit summary suggested to use 2FI model for analysis of variance. ANOVA results are shown in Table 5.32, Table 5.33 and Table 5.34 for weld strength, distortion and residual stresses respectively. The summary of design is given in the Table 5.35 The values of R-Squared, Adj R-Squared and Pred R-Squared including Adeq Precision (max to min ratio) are given in Table 5.36. Figure 5.51, Figure 5.52 and Figure 5.53 shows the effects of changing the levels of each parameter upon strength, distortion and residual stresses while keeping other parameters fixed respectively. It is clear that effects of welding current, voltage, speed and thickness are significant. The distortion and residual stresses increases with increase of welding current and voltage values without the application of trailing and decreases with increase of welding speed and thickness with the application of trailing. Figure 5.54, Figure 5.55 and Figure 5.56 shows the interaction effect of welding parameters upon weld strength, distortion and residual stresses respectively. From Table 5.33, Table 5.34 and Table 5.35, the significance of all models is clear from the Model F-value which is greater than 4. As the values of "Prob > F" less than 0.0500 indicate the significance of model terms, in this case A, B, C, D and E terms are significant with some interaction terms. The Table 5.36 shows that the "Pred R-Squared" values are in reasonable agreement with the "Adj R-Squared" values and “Adeq Precision” values indicate an adequate signal that shows the models can be used to navigate the design space.

Table 5.30 High and Low Settings of Factors for RSM Factor Name

Units

Type

A B C D E F

A V cm/min mm

Numeric Numeric Numeric Numeric Categoric Categoric

Current Voltage Weld Speed Thickness Trailing Weld Type

Low Actual 170.00 10.50 15.00 3.0 nil Linear

High Mean Actual 270.0 216.66 13.50 12.0 18.00 16.5 5.0 4.0 Ar Circumferential

192

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 5.31 Historical Data (96 (2x3x16) observations) including Response Values for RSM Run

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48

Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6 Responses A: B: C: D: E: F: Current Voltage Weld Speed Thick. Trailing Weld Type Weld Distortion (A) (V) (cm/min) (mm) Strength (MPa) (mm) 210.00 13.50 15.00 3.00 Ar Linear 696.8 6.5 170.00 10.50 18.00 3.00 nil Linear 751.6 4.6 170.00 10.50 15.00 3.00 nil Linear 733.1 5.2 170.00 13.50 15.00 3.00 nil Linear 704.3 6.5 170.00 13.50 15.00 3.00 Ar Linear 708.3 5.7 170.00 13.50 18.00 3.00 nil Linear 715.4 5.8 210.00 13.50 18.00 3.00 nil Linear 698.5 6.9 210.00 13.50 18.00 3.00 Ar Linear 702.7 6.2 170.00 10.50 18.00 3.00 Ar Linear 791 3.2 210.00 13.50 15.00 3.00 nil Linear 690.7 7.2 170.00 10.50 15.00 3.00 Ar Linear 767.5 3.8 210.00 10.50 15.00 3.00 Ar Linear 718.8 5.3 210.00 10.50 15.00 3.00 nil Linear 711.7 6.2 210.00 10.50 18.00 3.00 nil Linear 718.6 5.5 170.00 13.50 18.00 3.00 Ar Linear 733.4 4.7 210.00 10.50 18.00 3.00 Ar Linear 748.3 4.3 220.00 13.50 15.00 4.00 Ar Linear 736.4 5.9 200.00 10.50 18.00 4.00 nil Linear 751.3 3.7 200.00 10.50 15.00 4.00 nil Linear 741.2 4.4 200.00 13.50 15.00 4.00 nil Linear 725.2 5.6 200.00 13.50 15.00 4.00 Ar Linear 739.6 5.2 200.00 13.50 18.00 4.00 nil Linear 731.4 4.6 220.00 13.50 18.00 4.00 nil Linear 724.3 5.7 220.00 13.50 18.00 4.00 Ar Linear 737.2 5.4 200.00 10.50 18.00 4.00 Ar Linear 780.4 2.8 220.00 13.50 15.00 4.00 nil Linear 722.3 6.2 200.00 10.50 15.00 4.00 Ar Linear 760.1 3.4 220.00 10.50 15.00 4.00 Ar Linear 742.5 4.9 220.00 10.50 15.00 4.00 nil Linear 727.5 5.3 220.00 10.50 18.00 4.00 nil Linear 736.3 4.5 200.00 13.50 18.00 4.00 Ar Linear 750.3 4.1 220.00 10.50 18.00 4.00 Ar Linear 755.4 3.5 270.00 13.50 15.00 5.00 Ar Linear 726.7 5.2 230.00 10.50 18.00 5.00 nil Linear 759.5 3 230.00 10.50 15.00 5.00 nil Linear 737.8 4.5 230.00 13.50 15.00 5.00 nil Linear 717.8 5 230.00 13.50 15.00 5.00 Ar Linear 730.5 4.5 230.00 13.50 18.00 5.00 nil Linear 729.5 4 270.00 13.50 18.00 5.00 nil Linear 715 5.3 270.00 13.50 18.00 5.00 Ar Linear 730 4.9 230.00 10.50 18.00 5.00 Ar Linear 765.7 2.2 270.00 13.50 15.00 5.00 nil Linear 715 5.6 230.00 10.50 15.00 5.00 Ar Linear 757.7 3.4 270.00 10.50 15.00 5.00 Ar Linear 735.7 3.7 270.00 10.50 15.00 5.00 nil Linear 725 4.8 270.00 10.50 18.00 5.00 nil Linear 729.5 4.6 230.00 13.50 18.00 5.00 Ar Linear 742.8 3.6 270.00 10.50 18.00 5.00 Ar Linear 749.8 3.5

Residual Stresses (MPa) 514 516 540 547 472 543 553 476 448 608 467 471 545 542 468 478 452 422 448 459 391 453 470 409 366 505 382 389 445 438 385 390 388 391 404 410 357 401 425 370 335 452 353 355 398 403 348 357

193

University of Engineering & Technology, Taxila-Pakistan

Table 5.31 Historical Data (96 (2x3x16) observations) including Response Values for RSM Run

49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96

Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6 A: B: C: D: E: F: Current Voltage Weld Speed Thick. Trailing Weld Type (A) (V) (cm/min) (mm) 210.00 170.00 170.00 170.00 170.00 170.00 210.00 210.00 170.00 210.00 170.00 210.00 210.00 210.00 170.00 210.00 220.00 200.00 200.00 200.00 200.00 200.00 220.00 220.00 200.00 220.00 200.00 220.00 220.00 220.00 200.00 220.00 270.00 230.00 230.00 230.00 230.00 230.00 270.00 270.00 230.00 270.00 230.00 270.00 270.00 270.00 230.00 270.00

13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50 13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50 13.50 10.50 10.50 13.50 13.50 13.50 13.50 13.50 10.50 13.50 10.50 10.50 10.50 10.50 13.50 10.50

15.00 18.00 15.00 15.00 15.00 18.00 18.00 18.00 18.00 15.00 15.00 15.00 15.00 18.00 18.00 18.00 15.00 18.00 15.00 15.00 15.00 18.00 18.00 18.00 18.00 15.00 15.00 15.00 15.00 18.00 18.00 18.00 15.00 18.00 15.00 15.00 15.00 18.00 18.00 18.00 18.00 15.00 15.00 15.00 15.00 18.00 18.00 18.00

3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 4.00 4.00 4.00 4.00 4.00 4.00 4.00 4.00 4.00 4.00 4.00 4.00 4.00 4.00 4.00 4.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00

Ar nil nil nil Ar nil nil Ar Ar nil Ar Ar nil nil Ar Ar Ar nil nil nil Ar nil nil Ar Ar nil Ar Ar nil nil Ar Ar Ar nil nil nil Ar nil nil Ar Ar nil Ar Ar nil nil Ar Ar

Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential Circumferential

Responses Weld Distortion Residual Strength (MPa) (mm) Stresses(MPa) 696.8 751.6 733.1 704.3 708.3 715.4 698.5 702.7 791 690.7 767.5 718.8 711.7 718.6 733.4 748.3 736.4 751.3 741.2 725.2 739.6 731.4 724.3 737.2 780.4 722.3 760.1 742.5 727.5 736.3 750.3 755.4 726.7 759.5 737.8 717.8 730.5 729.5 715 730 765.7 715 757.7 735.7 725 729.5 742.8 749.8

3.4 3.1 3.3 3.7 3.5 3.6 3.8 3.2 2.3 3.9 2.7 3.3 3.6 3.4 3.1 2.9 2.7 2.4 2.6 3 2.8 2.9 3.1 2.5 1.8 3.3 2.1 2.7 2.9 2.7 2.5 2.3 2.4 1.9 2.2 2.1 1.8 1.5 2.2 2 1.2 2.9 1.4 1.7 1.9 1.8 1.5 1.4

505 464 496 501 445 498 514 447 416 577 443 444 497 489 441 455 433 360 394 403 354 398 421 370 324 464 348 352 387 375 348 357 370 332 335 340 307 330 361 320 268 398 302 306 329 333 293 305

194

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 3.32 ANOVA for Weld Strength (2FI model) of RSM Source

Sum sqrs

Model A-Current B-Voltage C-Weld Speed D-Thickness E-Trailing F-Weld Type AB AC AD AE AF BC BD BE BF CD CE CF DE DF EF Residual Cor Total

38699.18 9920.23 12136.10 2786.56 11384.95 5907.88 0.000 638.25 194.05 6.63 177.32 0.000 279.48 0.014 455.01 0.000 11.89 135.85 0.000 34.21 0.000 0.000 4053.66 42752.84

DoF

Mean square

F-value

Prob>F

Significance

21 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 74 95

1842.82 9920.23 12136.10 2786.56 11384.95 5907.88 0.000 638.25 194.05 6.63 177.32 0.000 279.48 0.014 455.01 0.000 11.89 135.85 0.000 34.21 0.000 0.000 54.78

33.64 181.10 221.55 50.87 207.83 107.85 0.000 11.65 3.54 0.12 3.24 0.000 5.10 2.529E-004 8.31 0.000 0.22 2.48 0.000 0.62 0.000 0.000

0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 1.0000 0.0010 0.0638 0.7290 0.0761 1.0000 0.0268 0.9874 0.0052 1.0000 0.6427 0.1196 1.0000 0.4319 1.0000 1.0000

Significant Significant Significant Significant Significant Significant Significant

Significant Significant

Table 5.33 ANOVA for Distortion (2FI model) of RSM Source

Sum sqrs

Model A-Current B-Voltage C-Weld Speed D-Thickness E-Trailing F-Weld Type AB AC AD AE AF BC BD BE BF CD CE CF DE DF EF Residual Cor Total

195.83 9.11 15.99 5.45 31.04 8.55 118.98 7.040E-004 0.27 0.19 0.022 2.91 0.015 0.087 0.54 4.25 0.27 0.027 1.17 0.016 1.32 0.70 4.10 199.94

DoF

Mean square

F-value

Prob>F

Significance

21 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 74 95

9.33 9.11 15.99 5.45 31.04 8.55 118.98 7.040E-004 0.27 0.19 0.022 2.91 0.015 0.087 0.54 4.25 0.27 0.027 1.17 0.016 1.32 0.70 0.055

168.12 164.21 288.26 98.18 559.57 154.06 2145.12 0.013 4.86 3.49 0.39 52.45 0.27 1.57 9.74 76.63 4.78 0.48 21.10 0.28 23.71 12.63

0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.9106 0.0306 0.0658 0.5321 0.0001 0.6046 0.2146 0.0026 0.0001 0.0320 0.4902 0.0001 0.5984 0.0001 0.0007

Significant Significant Significant Significant Significant Significant Significant Significant

Significant

Significant Significant Significant Significant Significant Significant

195

University of Engineering & Technology, Taxila-Pakistan

Table 5.34 ANOVA for Residual Stresses (2FI model) of RSM Source

Sum sqrs

Model A-Current B-Voltage C-Weld Speed D-Thickness E-Trailing F-Weld Type AB AC AD AE AF BC BD BE BF CD CE CF DE DF EF Residual Cor Total

4.833E+005 16895.72 19540.21 8190.03 1.854E+005 64566.00 47241.36 2614.66 202.05 1527.93 18.35 174.00 1232.67 1914.64 400.17 247.04 333.53 6.00 222.04 444.19 1397.79 2147.04 15082.74 4.984E+005

DoF

Mean square

F-value

Prob>F

Significance

21 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 74 95

23014.79 16895.72 19540.21 8190.03 1.854E+005 64566.00 47241.36 2614.66 202.05 1527.93 18.35 174.00 1232.67 1914.64 400.17 247.04 333.53 6.00 222.04 444.19 1397.79 2147.04 203.82

112.92 82.89 95.87 40.18 909.45 316.78 231.78 12.83 0.99 7.50 0.090 0.85 6.05 9.39 1.96 1.21 1.64 0.029 1.09 2.18 6.86 10.53

0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0006 0.3227 0.0077 0.7650 0.3585 0.0163 0.0030 0.1653 0.2745 0.2048 0.8642 0.3000 0.1441 0.0107 0.0018

Significant Significant Significant Significant Significant Significant Significant Significant Significant

Significant Significant

Significant Significant

Table 5.35 Max. and Min. Values of Responses in RSM Response

Units

Minimum

Maximum

Mean

Std. Dev.

Weld Strength

MPa

690.7

791

733.752

21.2139

Distortion

mm

1.2

7.2

3.70417

1.45073

Residual Stress

MPa

268

608

417.583

72.431

Table 5.36 ANOVA Summary for RSM (2FI model) Response

Adeq Precision

R-Squared

Adj R-Squared Pred R-Squared

Weld Strength

24.614

90.52%

87.83%

84.31%

Distortion

55.145

97.95%

97.36%

96.56%

Residual Stress

46.028

96.97%

96.11%

95.01%

196

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 5.51 Effects of welding parameters upon Weld Strength in RSM

Fig. 5.52 Effects of welding parameters upon Distortion in RSM

197

University of Engineering & Technology, Taxila-Pakistan

Fig. 5.53 Effects of welding parameters upon Residual Stresses in RSM

Fig. 5.54 Interaction of welding parameters upon Weld Strength in RSM 198

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

5.55 Interaction of welding parameters upon Distortion in RSM

Fig. 5.56 Interaction of welding parameters upon Residual Stresses in RSM 199

University of Engineering & Technology, Taxila-Pakistan

The overall empirical models for the responses (weld strength (MPa), distortion (mm) and weld residual stresses (MPa)), in terms of welding parameters, are as follows: Equation 5.21 (Trailing = nil, Weld Type = Linear), Equation 5.22 (Trailing = Ar, Weld Type = Linear), Equation 5.23 (Trailing = nil, Weld Type = Circumferential), and Equation 5.24 (Trailing = Ar, Weld Type = Circumferential) in terms of actual factors for weld strength:

Weld Strength = +727.24584-0.82644 * Current-14.66111* Voltage+21.63194 * Weld Speed +11.01124 * Thickness+0.095761* Current * Voltage-0.052802* Current * Weld Speed +0.012346* Current * Thickness-0.75833* Voltage * Weld Speed -0.016595* Voltage * Thickness+0.48614 * Weld Speed * Thickness 5.21 Weld Strength = +775.28334-0.97787* Current-17.56389* Voltage+23.21806* Weld Speed +13.48516 * Thickness+0.095761 * Current * Voltage-0.052802 * Current * Weld Speed +0.012346 * Current * Thickness-0.75833 * Voltage * Weld Speed -0.016595 * Voltage * Thickness+0.48614 * Weld Speed * Thickness 5.22 Weld Strength = +727.24584-0.82644 * Current-14.66111 * Voltage+21.63194 * Weld Speed +11.01124 * Thickness+0.095761 * Current * Voltage-0.052802 * Current * Weld Speed +0.012346 * Current * Thickness-0.75833* Voltage * Weld Speed -0.016595* Voltage * Thickness+0.48614* Weld Speed * Thickness 5.23 Weld Strength = +775.28334-0.97787* Current-17.56389 * Voltage+23.21806 * Weld Speed +13.48516 * Thickness+0.095761 * Current * Voltage-0.052802 * Current * Weld Speed +0.012346 * Current * Thickness-0.75833 * Voltage * Weld Speed -0.016595 * Voltage * Thickness+0.48614 * Weld Speed * Thickness 5.24

Equation 5.25 (Trailing = nil, Weld Type = Linear), Equation 5.26 (Trailing = Ar, Weld Type = Linear), Equation 5.27 (Trailing = nil, Weld Type = Circumferential), and Equation 5.28 (Trailing = Ar, Weld Type = Circumferential) in terms of actual factors for distortion: Distortion = +4.53993+1.02743E-003 * Current+0.41944 * Voltage-0.43333 * Weld Speed +0.70185 * Thickness+1.00575E-004* Current * Voltage+1.96839E-003* Current * Weld Speed -2.10898E-003* Current * Thickness+5.55556E-003* Voltage * Weld Speed -0.041559 * Voltage * Thickness-0.072593 * Weld Speed * Thickness 5.25 Distortion = +2.34827+2.70846E-003 * Current+0.51944 * Voltage-0.45556* Weld Speed +0.75455* Thickness+1.00575E-004* Current * Voltage+1.96839E-003* Current * Weld Speed -2.10898E-003* Current * Thickness+5.55556E-003* Voltage * Weld Speed -0.041559* Voltage * Thickness-0.072593* Weld Speed * Thickness 5.26 Distortion = +5.36910-0.018369* Current+0.13889 * Voltage-0.28611* Weld Speed +1.18688 * Thickness+1.00575E-004* Current * Voltage+1.96839E-003 * Current * Weld Speed -2.10898E-003 * Current * Thickness+5.55556E-003 * Voltage * Weld Speed -0.041559 * Voltage * Thickness-0.072593 * Weld Speed * Thickness 5.27 Distortion = +3.51910-0.016688 * Current+0.23889* Voltage-0.30833* Weld Speed +1.23957* Thickness+1.00575E-004* Current * Voltage+1.96839E-003* Current * Weld Speed -2.10898E-003* Current * Thickness+5.55556E-003* Voltage * Weld Speed -0.041559 * Voltage * Thickness-0.072593* Weld Speed * Thickness 5.28

200

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Equation 5.29 (Trailing = nil, Weld Type = Linear), Equation 5.30 (Trailing = Ar, Weld Type = Linear), Equation 5.31 (Trailing = nil, Weld Type = Circumferential), and Equation 5.32 (Trailing = Ar, Weld Type = Circumferential) in terms of actual factors for weld residual stresses:

Residual Stresses = +690.13875-1.53586* Current+18.27778 * Voltage+15.25000 * Weld Speed -98.19473 * Thickness+0.19382* Current * Voltage-0.053879* Current * Weld Speed +0.18747* Current * Thickness-1.59259 * Voltage * Weld Speed -6.16882 * Voltage * Thickness+2.57471* Weld Speed * Thickness

5.29

Residual Stresses = +608.72208-1.48715* Current+15.55556 * Voltage+15.58333* Weld Speed -89.28093* Thickness+0.19382* Current * Voltage-0.053879 * Current * Weld Speed +0.18747 * Current * Thickness-1.59259* Voltage * Weld Speed -6.16882 * Voltage * Thickness+2.57471* Weld Speed * Thickness

5.30

Residual Stresses = +673.59708-1.38586* Current+20.41667 * Voltage+13.22222 * Weld Speed -114.00723 * Thickness+0.19382 * Current * Voltage-0.053879 * Current * Weld Speed +0.18747 * Current * Thickness-1.59259* Voltage * Weld Speed -6.16882 * Voltage * Thickness+2.57471 * Weld Speed * Thickness

5.31

Residual Stresses = +611.09708-1.33715* Current+17.69444 * Voltage+13.55556* Weld Speed -105.09343 * Thickness+0.19382 * Current * Voltage-0.053879* Current * Weld Speed +0.18747* Current * Thickness-1.59259* Voltage * Weld Speed -6.16882 * Voltage * Thickness+2.57471 * Weld Speed * Thickness

5.32

All the models presented in equations 5.21 to 5.32 for weld strength, distortion and weld residual stresses in linear and circumferential welding are valid for the following ranges of input parameters: welding current: 170 to 210 A for 3 mm, 200 to 220 A for 4 mm and 230 to 270 A for 5 mm; welding voltage 10.5 to 13.5 V and welding speed 15 to 18 cm/min. The numerical optimization applied to the distortion and residual stresses data suggests that for any material thickness value lying between 3 and 5 mm, the distortion and residual stresses in TIG welding of HSLA steel can be minimized if the trailing is used along with low values of heat input i.e. low values of welding current and welding voltage and high value of welding speed. The predicted weld distortion and weld residual stresses values are as: i) 781.067 MPa, 3.06 mm and 450.59 MPa, ii) 779 MPa, 2.73 mm and 384.89MPa, and iii) 777.72 MPa, 2.24 mm and 331 MPa at input parameters as: i) 170 A, 10.5 V, 18 cm/min, 3 mm, ii) 200 A, 10.5 V, 18 cm/min, 4 mm and iii) 230 A, 10.5 V, 18 cm/min, 5 mm respectively as shown in Figure 5.57, Figure 5.58 and Figure 5.59 with response desirability for minimization of distortion and residual stresses and maximization of weld strength respectively. Figure 5.60, Figure 5.61 and Figure 5.62 show the response desirability for maximization of weld strength and minimization of distortion and residual stresses with respect to predictors respectively. Response desirability solutions containing predictors (welding current, welding voltage, welding speed, thickness, trailing and weld type) and response (weld strength, distortion and residual stresses) with respect to desirability are given in detail in Table 5.37.

201

University of Engineering & Technology, Taxila-Pakistan

Fig. 5.57 Weld Strength Predictions w.r.t Predictors in RSM

Fig. 5.58 Distortion Predictions w.r.t Predictors in RSM 202

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 5.59 Residual Stresses Predictions w.r.t Predictors in RSM

Fig. 5.60 Weld Strength Desirability w.r.t Predictors in RSM 203

University of Engineering & Technology, Taxila-Pakistan

Fig. 5.61 Distortion Desirability w.r.t Predictors in RSM

Fig. 5.62 Residual Stresses Desirability w.r.t Predictors in RSM 204

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 5.37 Response Desirability Solutions of RSM Number Current Voltage Weld ThicknessTrailing Weld Speed Type

Weld Distortion Residual Desirability Strength Stresses

For Goal: Current is equal to 230 A. Thickness is equal to 5 mm. Voltage, Speed and Trailing are in range. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

230.00 230.00 230.00 230.00 230.00 230.00 230.00 230.00 230.00 230.00 230.00 230.00 230.00 230.00 230.00

10.50 10.51 10.50 10.50 10.54 10.50 10.57 10.50 10.74 10.50 10.50 10.51 10.50 10.58 10.50

18.00 18.00 17.97 17.94 18.00 17.98 18.00 17.65 18.00 17.89 18.00 18.00 17.94 18.00 18.00

5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 4.98 5.00 5.00 5.00 5.00 4.99

Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar

Circumf Circumf Circumf Circumf Circumf Circumf Circumf Circumf Circumf Circumf Linear Linear Linear Linear Linear

777.723 777.6 777.534 777.38 777.348 777.505 777.076 775.778 775.515 776.634 777.723 777.587 777.395 776.983 777.38

1.13848 1.14052 1.14394 1.14842 1.14472 1.14583 1.14924 1.1947 1.17521 1.17526 2.29958 2.3064 2.3178 2.33429 2.31985

275.346 275.383 275.437 275.517 275.458 275.72 275.539 276.288 276.004 277.256 331.575 331.672 331.614 331.625 332.449

0.960 0.960 0.959 0.959 0.959 0.959 0.958 0.954 0.953 0.953 0.871 0.870 0.870 0.868 0.867

For Goal: Current is equal to 200 A. Thickness is equal to 4 mm. Voltage, Speed and Trailing are in range. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

200.00 200.00 200.00 200.00 200.00 200.00 200.00 200.00 200.00 200.00 200.00 200.00 200.00 200.00 200.00

10.50 10.50 10.52 10.50 10.55 10.50 10.50 10.50 10.50 10.50 10.50 10.50 10.51 10.50 10.50

18.00 18.00 18.00 18.00 18.00 18.00 17.91 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00

4.00 4.01 4.00 4.03 4.00 4.04 4.00 4.06 4.07 4.08 4.00 4.01 4.00 4.04 4.06

Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar

Circumfer 779.025 Circumfer 779.241 Circumfer 778.789 Circumfer 779.728 Circumfer 778.399 Circumfer 779.999 Circumfer 778.411 Circumfer 780.518 Circumfer 780.719 Circumfer 780.898 Linear 779.025 Linear 779.202 Linear 778.881 Linear 779.924 Linear 780.479

1.78613 1.77782 1.78987 1.75961 1.79604 1.74937 1.79966 1.72978 1.72217 1.71544 2.85035 2.84015 2.85594 2.79866 2.7667

341.409 340.625 341.469 338.943 341.557 337.99 341.741 336.17 335.456 334.835 386.325 385.817 386.337 383.751 382.158

0.888 0.888 0.888 0.887 0.886 0.886 0.886 0.885 0.884 0.884 0.803 0.803 0.803 0.802 0.802

For Goal: Current is equal to 170 A. Thickness is equal to 3 mm. Voltage, Speed and Trailing are in range. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

170.00 170.00 170.00 170.00 170.00 170.00 170.00 170.00 170.00 170.00 170.00 170.00 170.00 170.00 170.00

10.50 10.50 10.50 10.52 10.50 10.50 10.55 10.50 10.58 10.60 10.50 10.50 10.51 10.51 10.50

18.00 18.00 17.98 18.00 18.00 17.95 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00

3.41 3.42 3.42 3.43 3.44 3.43 3.44 3.48 3.46 3.47 3.53 3.53 3.52 3.57 3.44

Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar Ar

Circumf Circumf Circumf Circumf Circumf Circumf Circumf Circumf Circumf Circumf Linear Linear Linear Linear Linear

791.001 791.262 791.006 791.019 791.654 790.998 791 792.751 791.002 791.016 793.772 793.959 793.491 794.752 791.685

1.9528 1.94306 1.94974 1.94502 1.92924 1.94679 1.93597 1.89037 1.92758 1.92062 2.5662 2.55552 2.57725 2.50747 2.68213

381.034 379.969 380.373 379.72 378.51 379.679 378.097 374.395 376.64 375.462 412.44 411.806 412.667 408.78 418.927

0.825 0.825 0.825 0.825 0.825 0.825 0.824 0.824 0.824 0.823 0.756 0.756 0.756 0.756 0.756

205

University of Engineering & Technology, Taxila-Pakistan

5.6 Chapter Summary and Conclusions In parametric studies, the welding current with large parametric range was most influential parameter with respect to thickness of cylinders besides other parameters. Further, this chapter described the virtual experimental methodology to study and quantify the effects of four welding parameters: welding current (A); welding voltage (V); welding speed (cm/min); and Ar shielding as trailing upon the TIG welding of HSLA steel performance measures of distortion (mm) and residual stresses (MPa) in cylinders of different thicknesses. Full factorial method was utilized to design the experiments and develop the empirical models for residual stresses and distortion. The results were analyzed using ANOVA technique and numerical optimization was utilized for selection of best values for the four parameters. Three parameters were found to be the most influential welding parameters upon distortion and residual stresses, while the Ar trailing application was found slightly influential. Numerical optimization suggests that distortion and residual stresses can be minimized if the TIG welding is done at low values of welding current and voltage, high value of welding speed with the application of Ar trailing. It was also observed that the application of Ar trailing improves the welding process by decreasing of distortion and residual stresses (upto max.15%) in circumferential welds. The increase in cylinder thickness from 30% to 65% (i.e. 3-5 mm thickness) result decrease in distortion about 25% to 45% and 30% to 35% in residual stresses respectively. Whereas, the reduction in distortion is three times (1.2 – 3.9 mm) and in residual stresses (268 – 577 MPa) is two times respectively. At end of the chapter empirical models for residual stresses as well as for distortion, in terms of all the significant numeric input parameters were presented. Numerical optimization provided the response values and predictors levels as per the desirability. It was observed from the comparison of responses (distortion and residual stresses) results that the low distorted samples give the low residual stresses and vice versa respectively. After applying RSM and taking thickness of cylinder as variable parameter i.e. total five parameters (four numeric (thickness, current, voltage & speed) and one categoric (trailing)), it was observed that with the increase of thickness of cylinders the residual stresses and distortion decreases accordingly. Further in RSM by taking plate or cylinder (i.e. linear or circumferential weld) as another categorical parameter i.e. total six parameters (four numeric (thickness, current, voltage & speed) and two categoric (trailing & weld type)), it was observed that the response values (residual stresses and distortion) in case of circumferential welds are lower than the linear welds for the same thickness and other parameters even the weld length in circumferential welding of Ø300 mm is double of weld length in linear welding i.e. 500 mm plate length. At end of the chapter empirical models for weld strength as well as for distortion and residual stresses, in terms of all the significant numeric input parameters were presented. Numerical optimization provided the response values and predictors levels as per the response desirability. 206

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

CHAPTER 6 KNOWLEDGE ENGINEERING FOR OPTIMIZING TIG WELDING PROCESS 6.1 Introduction After completion of all welding analysis required to obtain the data by experimental work or virtual experiments by simulations and statistical analysis related to optimization of welding process, the next process is to manage the available welding experimental data and optimization information at a single platform and to utilize some automated means to extract the useful information from that platform in most effective manner as knowledge. The selection of expert system is the best option for this requirement as already mentioned in Chapter 1. Further more, the relationship among welding parameters and response is complex and it is very difficult to represent it using some mathematical model. In this chapter, the objectives of developing expert system and application to welding; the configuration; the utilization of fuzzy logic for reasoning mechanism; and the optimal formation of rule-base of the expert system are presented.

6.2 The Objectives of Expert System and Application to Welding The expert systems are computer programs that embody narrow domain knowledge for problem solving related to that knowledge domain [228]. Generally, an expert system comprises three main elements as a knowledge base, an inference engine, and working memory. The knowledge base is a collection of knowledge which is expressed by using some formal knowledge representation language, normally in form of facts and IF-THEN rules. Whereas the inference engine is a generic control mechanism that uses the axiomatic knowledge present in the knowledge base to the task oriented data to reach at a conclusion. Further, a program that contains meta-knowledge is called inference engine. Usually a knowledge base is very large therefore, it is necessary requirement to have inference mechanisms that search through the database and deduce results in a systematic and organized way [229]. During the execution of the expert system, the working memory is used to temporarily store the values of variables. The main components of an expert system are shown in Figure 6.1. The knowledge is explicitly kept separate from the control module in expert systems, while the knowledge is intertwined with the control mechanism in conventional programs. In this way, the expert system programs are better than conventional programs. It is very easy to add new knowledge in expert system due to the separation of knowledge from the control module during the expert system development phase or by experience of the program throughout use in its lifetime. This feature of mechanism mimics the human brain in which the control processes remain unchanged although individual behavior is continuously changing by addition of new knowledge by experience. This is the main feature that capable the expert system an ideal computer-based replacement of a human expert in the related domain.

207

University of Engineering & Technology, Taxila-Pakistan

The main objective of the research carried out in the welding domain and described in this manuscript is the optimal settings of the welding process input parameters so as to maximize the weld strength and minimize the residual stresses and distortion without compromising the welding quality. The detail of experimental work and simulations presented in chapters 3 – 5 provides the set of crude directives towards achievement of the objective. The highly generalized information generated by the experimental work and virtual based on simulations is very difficult to be utilized by the welder, operator, or engineers for solution of their highly specific welding problems.

Fig. 6.1 Main components of an expert system [33] In short, there is a dire need of a fast-acting informative tool that can recommend the optimal settings of the selected welding process parameters that would lead to accomplishment of desired objectives in best possible and efficient manner. Further, that tool should also be capable of providing highly accurate predictions of the performance measures before the start of the actual process at shop floor to the end user. The expert system developed and presented in this chapter fulfils all these requirements and provides the highly specific information to the user at the expense of few seconds.

6.3 Expert System Configuration To cover the requirements of the current research work, the information available from the experimental data, ANOVA results and numerical optimization was used for the development of knowledge-base or the rule-base. The presented expert system is dual functional as: first it search for the optimal selection and combination of the significant predictor variables in order to satisfy the desired objective; second it provides the predicted values of performance measures or responses for the selected combination of predictor variables or input parameters.

208

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

First consider the selection of five predictor variables only for the purpose of simplicity in description. These five predictor variables are the ones that were tested in the set of experiments explained in Chapter 3, i.e. material thickness, welding current, welding voltage, welding speed, and choice of trailing. The sheet or cylinder thickness will be considered as a parameter that needs not to be optimized. This is so because thickness is the geometric property of the work piece and it cannot be changed unless the work piece is removed from the welding setup and changed. The comprehensive treatment of all the process input parameters will be provided in the Chapter-7 by using the expert system. The configuration of the proposed expert system is shown in Figure 6.2.

TIG welding experiments + Virtual Experiment DOE + ANOVA + Numeric Optimization

Knowledge Base -Welding Current selection rules -Welding voltage/speed selection rule -Trailing selection rules

Shell

-Weld Strength prediction rules -Distortion prediction rules -Weld induced residual stresses prediction rules

Module

User Interface

r

Forward Chaining Inference Engine

Operation Modules Optimization

Use

Data Fuzzifier

Working Memory

Prediction Module

Recommendation

Data Defuzzifier

Fig. 6.2 Configuration of the expert system 6.3.1 Optimization and Prediction Modules The knowledge-base consists of two sets of rules, each one of them being controlled and operated by a separate module as shown in Figure 6.2. The optimization module is the first one that takes charge and operates with relevant set of rules for the optimal selection and combination of four parameters (predictor variables): the welding current, welding voltage, welding speed and the trailing. The selection of the predictor parameters is made in accordance with the objective desired by the user, the material thickness provided, and the predictor variables pre-fixed by the user. After this, the prediction module takes charge and makes use of the finalized combination of predictor variables and the relevant set of rules in order to estimate the values of performance measures, i.e. weld strength, distortion, and weld induced residual stresses. 209

University of Engineering & Technology, Taxila-Pakistan

6.3.2 The Expert System Shell As shown in Figure 6.2, it can be seen that the expert system shell consists of the user interface through which the input is taken from the user. The data fuzzifier fuzzifies the values of numeric parameters (predictor variables) according to the relevant fuzzy templates. The expert system shell also contains the working memory that consists of different variables, while the data defuzzifier is used to defuzzify the fuzzy sets of predictor variables (welding current, voltage, speed) and of performance measures. As the expert system presented is a kind of production system that requires the control of forward-chaining inference mechanism for the extraction of conclusions from its knowledgebase, according to the set of asserted facts and rules. For this purpose, a forward-chaining expert system shell named Fuzzy CLIPS (Fuzzy extension of C Language Integrated Production Systems) – developed by National Research Council, Canada – was utilized for the development of this knowledge based system [229]. Fuzzy CLIPS provides its standard format for defining templates, facts, functions, rules, and modules, and whole of the knowledge-base is the combination of these elements.

6.3.3 The Procedure The flow chart of operating procedure of the expert system is shown in the Figure 6.3. The expert system process starts with the user’s input of desired objective and the values of predictor parameters. It is mandatory for the user to fix the objective as well as the value of sheet or cylinder thickness, while the values of other four variables may or may not be fixed according to the welding problem at hand. The user may choose from following three objectives: 1. Maximize weld strength. 2. Minimize residual stresses or distortion. 3. Achieve 1 and 2 simultaneously. The selection of one objective as given above will lead to recommendation of different values of process input or predictor variables as compared to that of other, and consequentially, it will also lead to prediction of different values of performance measures as per requirements of maximization or minimization. Whereas, the objective number 3 provides the trade-off between the first two objectives. The values of material thickness and welding current (if fixed by user) are fuzzified according to the relevant fuzzy templates. As the welding current has been proved, by ANOVA results, to be the most significant factor for weld strength as well as for the distortion or residual stresses (see chapter 3), this factor is ought to be fixed ahead of others, if not already fixed by the user. The other three variables are also fixed in similar fashion. Welding speed appears before the trailing in the hierarchy list as being the significant variable out of the two. After the fixation of predictor variables as mentioned above, the prediction module takes the charge and the values of three response variables, in accordance with the finalized values of predictor variables, are estimated simultaneously. The next step is data defuzzification, in 210

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

which the fuzzy values of welding current, weld strength, distortion and residual stresses are defuzzified in accordance with preferred defuzzification algorithm. Finally, in the last step, the recommendation of predictor variables and prediction of response variables are printed out.

Start Fix Objective & material Thickness Fuzzify numeric input

Determine current

No

Current fixed?

Yes Voltage fixed?

No

Determine voltage

Yes Determine speed

No

Speed fixed?

Yes Trailing fixed?

No

Determine trailing

Yes Determine weld strength, distortion and residual stresses

Defuzzify current, voltage, speed & weld strength, distortion and residual stresses

Print Output Stop Fig. 6.3 The flow chart representing the operational procedure of the expert system

211

University of Engineering & Technology, Taxila-Pakistan

6.4 Fuzzy Reasoning for the Expert System The fuzzy logic is a discipline that has been successfully used in automated reasoning of expert systems [32]. In the real world system, there are some problems found in relationships between inputs and outputs like uncertainty, vagueness, ambiguity and impreciseness. These input and output relationship problems can be handled effectively by utilizing fuzzy logic treatments.

6.4.1 Fuzzy Sets, Input Fuzzification and Output Defuzzification In the fuzzification, the precise or imprecise input data which is easily understandable by the human minds are converted into a kind of linguistic form, for example very low (weld strength) and highly distort (distortion) etc. The expert system then uses these fuzzified data to give answers to imprecise and vague questions and also describe the reality level of those answers. Figure 6.4 shows the fuzzy sets utilized for four predictor variables: material (sheet or cylinder) thickness, welding current, welding voltage and welding speed; while Figure 6.5 shows fuzzy sets for response (weld strength, distortion and residual stresses).

Fig. 6.4 Fuzzy sets for the numeric input variables

212

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 6.5 Fuzzy sets for Responses Triangular shaped fuzzy sets for the response variables in Fuzzy CLIPS format are as follows: (deftemplate Weld_Strength 680 800 MPa ( (very low (680 1) (700 1) (720 0) ) (low (700 0) (720 1) (740 0) ) (medium (720 0) (740 1) (760 0) ) (high (740 0) (760 1) (780 0) ) (very high (760 0) (780 1) (800 1) ) ) ) 213

University of Engineering & Technology, Taxila-Pakistan

(deftemplate Distortion 1 7 mm ( (vlow (1 1) (2 1) (3 0) ); very low (low (2 0) (3 1) (4 0) ); low (med (3 0) (4 1) (5 0) ); medium (high (4 0) (5 1) (6 0) ); high (vhigh (5 0) (6 1) (7 1) ) ) ); very high (deftemplate Residual_Stresses 100 700 MPa ( (vlow (100 1) (200 1) (300 0) ); very low (low (200 0) (300 1) (400 0) ); low (med (300 0) (400 1) (500 0) ); medium (high (400 0) (500 1) (600 0) ); high (vhigh (500 0) (600 1) (700 1) ) ) ); very high The one predictor variable (trailing) is categorical, therefore cannot be fuzzified. However, this variable is used as crisp variable in the fuzzy knowledge-base. This shows that in this expert system development, both crisp and fuzzy antecedents and consequents are freely mixed for the creation of the rules. The step of application of fuzzy rule provides the recommendation as a crisp value and/or fuzzy set, specifying a fuzzy distribution of a conclusion. But in welding process, the operator or welder needs a single discrete valued direction. Therefore, it is required to select a single point from fuzzy distribution that provides the best value. The process of reducing a fuzzy set to a single point is known as defuzzification [229]. There are two methods commonly used for defuzzifying the fuzzy sets i.e. center of gravity (CoG) or moment method and mean of maxima (MoM) method. The detail of both methods can be referred in [229, 230]. For this expert system development, the centre of gravity (CoG) method is used as defuzzification method for the reason that it provides smoothly varying output of response variables for gradually varying input values of material thickness and welding current. Whereas the utilization of mean of maxima (MoM) method contained the risk of generating highly abrupt output values of response variables for small and gradual variations in material thickness and welding current values that was observable at specific ranges of these two predictor variables.

6.4.2 Inference for Aggregation of Fuzzy Rules Generally, two kinds of methods are commonly used for yielding aggregation of fuzzy rules i.e. max-min inference method and max-product method. The max-min inference method is the default inference method for Fuzzy CLIPS. The application of max-min inference strategy is described in the following example. 214

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Suppose a knowledge-base consists of following set of rules: 1. IF thickness is Small AND current is Low THEN weld strength is Low 2. IF thickness is Small AND current is High THEN weld strength is Medium 3. IF thickness is Large AND current is Low THEN weld strength is Very Low 4. IF thickness is Large AND current is High THEN weld strength is Low Further suppose that it is required to predict the value of weld strength for work piece material thickness of 4.5 mm and welding current of 190 A, utilizing above-mentioned set of 4 rules and fuzzy sets provided in Figure 6.4 and Figure 6.5. Figure 6.6 describes the input fuzzification process in which the welding current value of 190 A has been converted to 2 fuzzy sets: Low (membership function µ Low = 0.8) and High (µ High = 0.2); while the material thickness of 4.5 mm has also been converted into 2 fuzzy sets: Large (µ Large = 0.75) and Small (µ Small = 0.25). The fuzzy membership value for welding current can be expressed as: µ(current) = 0.8/Low, 0.2/High. Similarly the fuzzy membership value for material thickness can be expressed as: µ(thickness) = 0.75/Large, 0.25/Small. All the four rules use AND operator in their antecedent parts. Considering the first rule in the list and applying the max-min strategy, the rule will yield a result (i.e. weld strength is Low) whose degree (or membership function) will be minimum of degrees of current (Low) and of thickness (Small). This can be expressed as follows:

Fig. 6.6 Fuzzification of input data 215

University of Engineering & Technology, Taxila-Pakistan

µ (weld strength) Low = min {µ (current) Low, µ (thickness) Small}

By using all possible combinations of two inputs and applying AND operation, the fuzzy membership values for output variable weld strength (considering application of above listed four rules) can be as given in the following: 1. Low (0.8) and Small (0.25) will yield Low (0.25) 2. Low (0.8) and Large (0.75) will yield Medium (0.75) 3. High (0.2) and Small (0.25) will yield Very Low (0.2) 4. High (0.2) and Large (0.75) will yield Low (0.2) By using the above mentioned four rules according to the max-min strategy, Table 6.1 can be obtained following the procedure of aggregation. By applying OR operation to all the fuzzy set values as given in Table 6.1, the maximum value for the output fuzzy set can be obtained as shown in Table 6.2. The defuzzified output which gives the value of weld strength can be obtained as follows: Weld Strength = 690x0.2+700x0.2+710x0.25+720x0.25+730x0.5+740x0.75+750x0.5+760x0 0.2+0.2+0.25+0.25+0.5+0.75+0.5+0 Weld Strength = 728.49 MPa

Table 6.1 Weld Strength values from all the four rules Fuzzy

Universe of weld strength (MPa)

Subsets

670

680

690

700

710

720

730

740

750

760

Low

0

0

0

0

0.25

0.25

0.25

0

0

0

Medium

0

0

0

0

0

0

0.5

0.75

0.5

0

Very Low

0

0

0.2

0.2

0.2

0

0

0

0

0

Low

0

0

0

0

0.2

0.2

0.2

0

0

0

Table 6.2 Maximum fuzzy output from Table 6.1 670 680 690

700

710

720

730

740

750

760

770

780

790

0

0.2

0.25

0.25

0.5

0.75

0.5

0

0

0

0

0

0.2

216

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

6.5 Optimal Formation of Rule-Base The relationship between inputs and output in a fuzzy system is characterized by set of linguistic statements which are called fuzzy rules [231]. The rules collection is known as rule-base and rule-base combining with the list of facts is called as knowledge-base. In a fuzzy system, the numbers of fuzzy rules are related to the fuzzy sets number for the each input variable. In the present research, there are two fuzzy sets each for material thickness, welding current, voltage and welding speed. In the same way, there are two possible values each for trailing (nil and Ar). However, for four variables (material thickness, welding current, welding speed, and trailing), the maximum possible number of rules for the prediction module of the expert system are 16 (= 2 × 2 × 2 × 2). Here, an important question could be, “which weld strength sets, or distortion sets to be assigned to 16 possible combinations of input sets/values”? Simply, 2-inputs 1-output fuzzy model, the designer will select the most optimum set of fuzzy rules from more than 10,000 combinations. For the output variable of weld strength in the present research, there are 16 fuzzy rules with 7 possibilities each (7 fuzzy sets for weld strength). Thus the total number of possible fuzzy rules combination will be 716 = 3.323 × 1013 for purpose of estimation of weld strength. In similar way, there are more possibilities for formulation of fuzzy rules for estimation of distortion and residual stresses. The simulated annealing algorithm was applied for assigning the most optimum fuzzy set of each of output variables to the 16 rules for the best possible combination of rules. The objective of the rule-base optimization process in the present research is to minimize the estimation error (i.e., difference between the prediction of output variable values and their actual values).

6.5.1 Optimal Formation Using Simulated Annealing Algorithm Simulated annealing (SA) is a stochastic neighborhood search method, which is developed for combinatorial optimization problems [233]. It depends upon the analogy between the annealing process of solids and solution methodology of combinatorial optimization problems. That has the capability of jumping out of local optima for the global optimization and is attained by accepting with a probability the neighboring solutions worse than the current solution. Whereas, the acceptance probability is determined by a control parameter “temperature”, which decreases during the simulated annealing process. The details of SA can be found in [232]. The pseudo-code of the algorithm developed for optimization of fuzzy rules using SA technique is given in the following [233]: [0] Initialize [0.1] Set annealing parameters T0, ATmin, imax, α, Rf [0.2] Initialize iteration counter, i = 0 [0.3] Generate initial rules combination and calculate estimation error value, i.e. rules [0], error [0] [1] Execute outer loop, i.e. steps 1.1 to 1.7 until conditions in step 1.7 are met. [1.1] Initialize inner loop counter l = 0, and accepted number of transitions AT = 0 217

University of Engineering & Technology, Taxila-Pakistan

[1.2] Initialize rules combination for inner loop, rules [i][0] = rules [i] and error [i][0] = error [i] [1.3] Execute inner loop, i.e. steps 1.3.1 to 1.3.5 until conditions in step 1.3.5 are met [1.3.1] Update l = l + 1 [1.3.2] Generate a neighboring solution by changing randomly one rule, and compute estimation error for new rules combination (rules [i][l] and error [i][l]) [1.3.3] Assign q = error [i][l] – error [i][l – 1] [1.3.4] If q ≤ 0 or Random (0, 1) ≤ e-q/To then • Accept rules [i][l] and error [i][l] • Update AT = AT + 1 Else reject generated combination: rules [i][l] = rules [i][l – 1], error [i][l] = error [i][l –1] [1.3.5] If one of following conditions hold true: AT ≥ ATmin; OR l ≥ 5S2 (S – No. of fuzzy sets of output variable), then assign length of Markov chain L [i] = l. Terminate inner loop and go to 1.4, else continue the inner loop and go to 1.3.1 [1.4] Update i = i + 1 [1.5] Update: rules [i] = rules [i – 1][L[i] – 1] and error [i – 1][L[i] – 1] [1.6] Reduce cooling temperature: T [i] = α.T[i – 1] [1.7] If one of following conditions hold true: i ≥ imax; OR (AT / L[i]) ≤ Rf; OR estimation error value does not reduce for last 20 iterations, then terminate the outer loop and go to 2, else continue outer loop and go to 1.1 [2] Print out the best rules combination along with minimum estimation error value and terminate the procedure The C++ was used to generate the algorithm code. Further, the SA parameters are operated using following values: (1) starting annealing temperature (T0) = 1300 MPa; (2) rate of cooling (α) = 0.98; (3) maximum number of iterations (imax) = 100; (4) length of Markov Chain at each iteration (L) = 5 × 7 × 7 = 245; (5) minimum acceptance ratio (Rf) = 0.01; (6) minimum number of accepted transitions at each iteration (ATmin) = 100. The objective function of the “optimization of fuzzy rules” problem is the minimization of estimation error, where the term “estimation error” can be as given in Equation 6.1.

Whereas l, m, n, o = Number of levels (not the fuzzy sets) provided by the user for each of four variables: material thickness, welding current, welding speed and trailing respectively. 218

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

WS* = Actual weld strength WSest = Weld strength estimated by the rule-base.

6.5.2 Results of Optimal Formation of the Rule-Base The optimal formation of fuzzy rule-base related to prediction of weld strength is presented only in this section. In the similar way, the rule-bases for prediction of other output variables (i.e. distortion and residual stresses) can be optimized. Further, the optimization of rules related to the optimizing module of the expert system is not in the scope of this section. Initially, the random combination of fuzzy rules was made and the criteria of termination for algorithm depended upon fulfillment of one of three conditions provided in the algorithm pseudo-code. For estimation error, each transition of the rules of all iterations was tested in order to determine the optimal combination of fuzzy rules by using the data available in Chapter 3 & 5. The program continued processing for 30 iterations based upon SA algorithm and until the criteria of termination was fulfilled. As the estimation error value did not improve for last 20 iterations. The optimal combination of fuzzy rules was printed out at the termination of program as listed in Table 6.4 and the testing values of input variables resulted in least value of estimation error is 5 MPa. Figure 6.7 shows the continuous improvement in estimation error through the iterations of this program run [233]. The optimized rules for prediction of other output variables are listed in Table 6.5.

Fig. 6.7 Decline of estimation error along number of iterations 6.5.3 The Complete Rule-Base In this section, all the rules operated by the optimization module as well as operated by the prediction module are listed. As the target of optimization module is to select the values of predictor variables (welding current, voltage, welding speed and trailing), which will best satisfy the desired objective, so all of the possible values of these variables (fuzzy or crisp) do not appear in the consequent parts of the optimization rules. However, the welding experimental and ANOVA results have shown that these non-appearing values of the variables do not satisfy any of three objectives in any combination of predictor variables. The complete list of rules operated by optimization module is given in Table 6.3. 219

University of Engineering & Technology, Taxila-Pakistan

Whereas the prediction module is assigned to generate the best possible estimate of all the three response variables for any given combination of four predictor variables whether all of the four predictor variables have been fixed by the user or any combination of these has been determined by the optimization module. Table 6.4 presents these 16 rules with two consequents as displayed i.e. weld strength and distortion. Table 6.5 shows the other consequents – residual stresses – for the same 16 rules arranged in same order as in Table 6.4. Table 6.4 and Table 6.5 show the rules that made by using Simulated Annealing Algorithm for maximum precision in predicting the values of output variables.

Table 6.3 List of rules operated by the optimization module Rule Antecedents Consequent No. Objective Thickness Speed Current Trailing 1 1 Any Any Open2 Any Any Speed Low 3 4 2 WS or Both Any Any Open Any Current High 5 Large Any Open Ar or Open Current Low &6 High 3 Dist 4 Dist Large Any Open Nil Current High 5 Dist Small Any Open Any Current Low 6 WS Any Any Any Open Trailing Nil 7 Dist or Both Any Any Any Open Trailing Ar 1 2 3 Fixed with any level of the variable; Not fixed; Maximize weld strength; 4 Achieve 1 & 2 simultaneously; 5Minimize distortion; 6Intersection operator

Table 6.4 List of rules operated by the prediction module [Consequents: Weld Strength & Distortion] Rule No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Antecedents Thickness Small Small Small Small Small Small Small Small Large Large Large Large Large Large Large Large

Current Speed High Low High Low High High High High Low Low Low Low Low High Low High High Low High Low High High High High Low Low Low Low Low High Low High

Trailing Nil Ar Nil Ar Nil Ar Nil Ar Nil Ar Nil Ar Nil Ar Nil Ar

Consequents Weld Strength Distortion very low high & very high very low & low medium medium low & medium medium & high very low very low & low low low medium high low high & very high very low very low high low low& medium medium low high very low extremely low low extremely low very low & low very low low low & medium very low

220

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 6.5 List of consequents (Residual stresses) for the antecedents enlisted in table 6.4 (See the fuzzy sets provided in sub-section 6.4.1) Rule No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Residual Stresses high & very high medium low & medium very low low medium low very low high low& medium low very low low very low & low low very low

6.6 Application Example The application of the presented fuzzy expert system for optimization of parameters and prediction of performance measures in TIG welding process is considered by supposing to find the optimal values of welding current, voltage and welding speed in order to attain the lowest possible distortion during the welding of HSLA steel plates of thickness 5 mm with the application of Ar as trailing. It is also required to have prediction of weld strength, distortion, and residual stresses for the recommended welding conditions. For in this example, user provides following input to the expert system: objective as ‘minimize distortion’; material thickness as 5 mm; and trailing as Ar. After processing, the expert system prints out the following recommendations and predictions:


It is recommended to use welding current of 230 A.
It is recommended to use welding voltage of 10.5 V.
It is recommended to use welding speed of 18 cm/min.
It is predicted that weld strength will be 765.7 MPa.
It is predicted that distortion of plates will be 2.2 mm.
It is predicted that weld induced residual stresses will be 335 MPa.

221

University of Engineering & Technology, Taxila-Pakistan

6.7 Chapter Summary and Conclusions An innovative effort was presented in this chapter regarding the application of AI to the domain of welding process. An expert system based on fuzzy reasoning mechanism was presented for the purpose of optimizing parameters and predicting performance measures in GTAW process of high strength low alloy (HSLA) steel. The developed expert system can optimize the welding process parameters in accordance with the objectives of ‘maximizing weld strength’, ‘minimizing distortion or residual stresses’, and also the attainment of both of these simultaneously. For the reasons of space limitation and simplicity, the role of only four input variables (material thickness, welding current, welding speed, and trailing) in the expert system, was described in this chapter. Based upon the significance of their effects, the other parameters, like gas flow rate, inter-pass temperatures, pre-heating, wire feed rate, and more can also be included in the knowledge-base following the same methodology as described in this chapter. Useful data and information for development of rules was taken from experimental, virtual/simulations and ANOVA results provided in chapter 3 and 5. These rules based on input and output variables were divided into two sets whereas the each set was operated by a separate module known as optimization module and prediction module. The optimization module gives the optimal combination of input parameters (predictor variables) that can best satisfy the required objective whereas the prediction module gives the prediction of performance measures (output variables) according to the finalized values of the input parameters. The Simulated Annealing (SA) algorithm was utilized to form the combination of antecedents and consequents of the rules related to prediction module for the purpose of maximizing the prediction precision of the expert system i.e. to minimize the estimation error. The developed expert system proves to be very effective and efficient for optimizing the welding process of HSLA steel thin walled structures and also to provide important predictions before the start of actual process on the shop floor. The utilization of this expert system by the engineers can help to improve the output quality and reduce production cost of complex welding process at the expense of only few seconds required for expert system to process.

222

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

CHAPTER 7 THE SELF-DEVELOPING EXPERT SYSTEM FOR OPTIMIZING TIG WELDING PROCESS 7.1 Introduction The development of an expert system for optimization of predictor variables (welding current, welding voltage, welding speed and trailing) and prediction of performance measures (weld strength, distortion and residual stresses) of TIG welding process of thin walled HSLA steel structure was presented in the previous chapter. Even the developed expert system consumed a considerable amount of effort and time but still its scope remained limited. The developed expert system deals only the effects of four welding process input parameters after the formulation of 23 rules, 16 of them employing 3 output variables (weld strength, distortion and residual stresses) and also required to be optimized using a complex optimization algorithm. In order to expand the scope of expert system, the developer would have to redo the same hectic efforts in order to incorporate the incoming knowledge from experimental work on welding in knowledge-base. Such type of requirement and situation represents a picture of a major barrier in the way of successful application of expert system at industrial level. In this way, there is strong need to have a computer-based consultation system that can develop and expand its scope of application by itself without requiring knowledge engineering skills of the developers. In this context, this chapter presents this important area of artificial intelligence in detail.

7.2 Self-Development of Expert System Only few research papers are available that have focused and described the ability of self-learning imparted to the expert systems. In broad aspect, the self learning field is called as machine learning in which the computer programs learn from their own experience upon utilization. A self-learning and self-testing fuzzy expert system applicable to control system was presented in [234]. The main feature of the expert system provided is to check the completeness and correction of the knowledge-base. The program was developed based upon the results of actions it performs in such a manner that the system extracts fuzzy rules from the set of input-output data pairs and keeps on correcting its rules. However, the paper does not cover the idea of expanding the scope or applicability of the expert system. In [235] the author presented a general framework for acquisition of knowledge using inductive learning algorithm and genetic algorithm. In manufacturing, a few papers can be found that describe the application of machine learning in field of metal cutting only. In [236] the authors presented a machine learning approach for building the knowledge-base from the numerical data and proved to be useful for classifying the dielectric fluids in Electric Discharge Machining. In [237] the author presented partially the application of pattern recognition and ANN for acquiring the knowledge in order to monitor the condition of tool in a plate machining process. The authors presented the use of ANN for picking up the experience of machinists and data from the machining handbook to predict the values of cutting speed and feed for a given turning process in [238]. The authors presented the utilization of Support Vector Regression, a 223

University of Engineering & Technology, Taxila-Pakistan

statistical learning technique, to diagnose the condition of tool during a milling process in [239]. Now it is obvious that machine learning approach has been utilized on a very limited scale for optimization of few process parameters or for the purpose of tool condition monitoring as given in above review. In this chapter development of a fuzzy expert system for optimizing the welding process will be presented that have the capability of self-learning, self-correcting and also selfexpanding. Following are the salient features of the self developing expert system [240]. 1. To predict the values of output process variables based upon values of input process variables. 2. To suggest the best values of input process variables to maximize and/or minimize the values of selected set of output process variables. 3. Adjustment of newly entered variable at any stage of development automatically. 4. Self learning and correction according to the new data set provided. 5. Generates fuzzy sets for newly entered process variable and regenerates sets for other variables according to newly added data automatically. 6. Generates the rules for the knowledge-base automatically. 7. Solve contradictory rules with conflict resolution facility. 8. Deletion facility of outdated data from database. The first two features represent the main objective of the expert system while the other features describe the automation required for the system to self developing. This self developing expert system offers numerous benefits as given in the following: 1. 2. 3. 4. 5.

Scope of the expert system can be expanded according to the requirements. Minimum human involvement would be required for updating knowledge-base. Higher precision upon more utilization of expert system. No requirement of optimal formation of rule-base and automatic generation of rules. The application of self-developing expert system is expected to be highly adaptive to the rapidly changing industrial environment.

The main components of the self-development mode of the expert system are: data acquisition module; fuzzy sets development module; and rule-base (optimization and prediction) development module [240]. In the following sections, the objectives, functionality, and algorithms for these modules, are described. This chapter presents the details with explanation of data structures and coding techniques for programming these modules. Two comprehensive examples will be presented to show the functioning of the developed expert system at the end of chapter.

7.3 Data Acquisition Module Data acquisition module facilitates the automation of intake, storage, and retrieval of data. The data may be the specifications of a new variable or the values of input and output variables resulted from experiments or empirical models. The data is stored in a file on the hard disk after intake. The pseudo-code of the algorithm that automates the data acquisition and management is given in the Appendix A2-1. 224

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

The flow chart of the data acquisition algorithm is shown in Figure 7.1. The user is asked first of all whether to introduce a new variable or not. If the user says yes then he is further asked to give the details of that variable related to the name, units and input variable or output variable. If the selected variable is input, then the user has to further describe whether it is numeric or categorical input. If the input variable is numerical then a check box is assigned to the interface of the developed expert system asking whether that variable should be prefixed or not. If the input variable is categorical then a choice box, showing all the possible values of that variable, is assigned to the interface of ES. The output variable can be numeric only and for each new output variable, the user is asked whether to include that output variable for optimization purpose or not. If user says yes, then a slider bar for that output variable is assigned to the interface of the ES. The user can specify whether to minimize or maximize that output variable along with the weightage of the objective required from that slider bar. The given specifications of new variable are stored in file Variable.dat.

Fig. 7.1 Flow chart for data acquisition module [240] Next, the user is again asked to enter new data with the choice of the type of variables that user wants to enter. It is mandatory that for each entry, the data should be entered for at least two input variables and one output variable. This newly provided data by the user is appended to the file Data.dat. Next, the user is asked whether to delete any data record or not. If user says yes, then user has to mention the record number and that will be deleted from the file. For, further processing all data records are loaded to a linked list named as Set.

7.4 Self-Development of Fuzzy Sets Module Self development of fuzzy sets module covers three areas: (i) Development of fuzzy sets for newly entered numeric variables, (ii) Rearranging the fuzzy sets for already entered variables according to newly entered data records and (iii) Development of two fuzzy sets (low & high) for each output variable that is marked for optimization purpose. The set low 225

University of Engineering & Technology, Taxila-Pakistan

represents the minimization requirement and the set high represents the maximization. In Figure 7.2, the design of sets for category 3 is fixed whereas the design of first two categories is dynamic and based upon data values of respective variables. Figure 7.2 shows the weightage values that are set by the user using the slider bar available on the interface of the developed expert system. However, any value below 5% means desirability is of totally minimizing the output variable and above 95% means total desirability of maximization. The weightage of 50% means that optimization of that output variable makes no difference. The pseudo-code of the algorithm that deals with development of the fuzzy sets of the numeric variables is given in Appendix A2-2.

Membership function

1

Low

High

0 0

10

20

30

40

50

60

70

80

90

100

Weightage (%)

Fig. 7.2 Fuzzy sets for maximization and/or minimization of output variable [240] The customized flow chart for the methodology used for self-development of fuzzy sets is shown in Figure 7.3. The user has to decide the maximum allowable number of fuzzy sets for input as well as for output variables first of all. For any numeric variable x, copy its values, appearing in all the data records, to a linked list L1 and sort it in the ascending order. If x is input variable, then copy all its distinct values from L1 to another linked list CL1. If any value appears more than once in L1, then don’t copy it more than once but record the number of appearances of that value in CL1. Sort CL1 in descending order of number of appearances of the values. If the total number of values in the CL1 is greater than the maximum number of allowable fuzzy sets (say N1 number of sets) for input variable then copy top N1 number of values to another linked list L2, otherwise copy all the values of CL1 to L2. To each of the values contained by L2 assign a separate triangular fuzzy set in Fuzzy CLIPS format. The logic involved in this methodology is that a value (of input variable), which has higher frequency of appearance, in the data records, has more right to be picked up for allocation of a fuzzy set. If x is an output variable and number of its distinct values appearing in all the data records is less than or equal to maximum number of allowable fuzzy sets (say N2) for an output variable then follow the same strategy described in the last paragraph. On the other hand if the number of distinct values of x is greater than N2 then do the following: Copy all the distinct values from L1 to CL1. For each of the values contained by CL1, compute neighbor distance by using the following Equation 7.1 [240]. 226

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Value[i + 1] − Value[i ]; if (i = first )  Neighbor _ Distance = Value[i ] − Value[i − 1]; if (i = last )  1 (Value[i + 1] − Value[i − 1]) ; otherwise  2

(7.1)

Assign respective neighbor distance to each value in CL1 and sort it in descending order of the neighbor distance and copy top N2 number of values from CL1 to L2. Assign separate triangular fuzzy set to each of the values contained by the list L2. The idea given in this technique is that any value, of an output variable, having greater difference from its previous and next values in the list, possesses more right to be picked up for allocation of a fuzzy set. Print the fuzzy sets for the variable to fuzzy CLIPS file Sets_Rules.clp and repeat the same procedure for all the numeric variables.

Fig. 7.3 Customized flow chart for auto-development of fuzzy sets [240] 7.5 Self-Development of Prediction Rule-Base Module This section covers two parts: (1) automatic development of rules, for prediction of welding performance measures, based upon the data records provided by the users and (2) conflict resolution among self-developed contradictory rules. In the expert system’s execution the priority of rule’s firing is based upon the fulfillment of antecedent part of rule and after that upon salience of the respective rules specified by the rule-base developer. So the sequence of appearance of rules in CLIPS file is absolutely immaterial. In view of this, even though the rules of optimization module are supposed to fire before those of prediction 227

University of Engineering & Technology, Taxila-Pakistan

module, the development of prediction rule-base will be described before that of optimization rule-base. The pseudo-code of the algorithm describing the automatic development of prediction rule-base is given in Appendix A2-3. In the linked list Set there would be lot of data records that contain data values of more than one output variables. First of all, detach these output variables and for each output variable keep record of same set of input variables. The data in this format is copied in doubly linked list named Data_output. Each node of this list contains value of output variable and to each node there is also connected a linked list Data_input that contains respective data of input variables. Now Data_output is the list of data records with one output variable per record. In section 7.7, the structure of doubly linked list is discussed.

Fig. 7.4 The framework for self-development of prediction rule-base [240] 228

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

The graphical description of the above discussed algorithm is shown in Figure 7.4. The objective of algorithm is to convert each node of Data_output (including list of related values of input variables Data_input) into a rule. This is achieved by finding and assigning the most suitable fuzzy sets for all of the values involved per node of Data_output. The list Data_output is navigated from first node to last and for all of its values the closest values in fuzzy sets of respective variables are matched. If the match is perfect then certainty factor (CF) of 1 is assigned to the match of data value and fuzzy set. If the suitable match of any fuzzy set for a given data value is not found then the data value is assigned the intersection of two closest fuzzy sets. The CF for non-perfect matches are calculated according to formulae provided in the pseudo-code. This results in formation of prediction rules-base containing the number of rules equal to number of nodes in Data_output. All the rules are stored in a doubly linked list named Rule_Consequent and each node of represents a rule. Each node contains the assigned fuzzy set of output variable and also a linked list (Rule_antecedent) containing assigned fuzzy sets of all the relevant input variables. To each rule is assigned a priority factor called salience, whose value is in direct proportion to the number of input variables involved in that rule. This emphasizes that a rule containing larger number of variables in its antecedent part enjoys a higher priority.

7.5.1 Conflict Resolution among Contradictory Rules It is expected that new data would be entered according to the wish of the user and therefore, there is always a possibility that some anomalous data might be entered that could lead to development of some opposing rules. Therefore, it is necessary to develop a mechanism that would detect such possible conflict among contradictory rules and would provide a way for its resolution. The pseudo-code of the algorithm that provides the mechanism for conflict resolution among contradictory rules is given in Appendix A2-4. The mechanism of conflict resolution contained by the algorithm can be explained as follows. Compare each and every rule of the prediction rule-base to all the other rules of the same rule-base. If, in the consequent parts of any two rules, following two conditions satisfy: (i) output variables are same and (ii) assigned fuzzy sets are different, then check whether the antecedent parts of both the rules are same (i.e., same input variables with same fuzzy sets assigned). If yes, then these two rules form a pair of contradictory rules. Next the user is inquired which one of the two contradictory rules needed to be abandoned. The CF value of the rule to be abandoned is set to zero. The same procedure is continued for whole of the rule-base. At the completion of the process, all the rules possessing the CF values greater than zero are printed to the CLIPS file: Sets_Rules.clp.

7.6 Self-Development of Optimization Rule-Base Module The objective of this section is to develop an algorithm that would lead to automatic generation of optimization rule-base. The optimization rule-base is responsible for providing the optimal settings of input variables that would best satisfy the maximization and/or

229

University of Engineering & Technology, Taxila-Pakistan

minimization of the selected output variables. The pseudo-code of the algorithm for selfdevelopment of optimization rule-base is given in Appendix A2-5. The idea given in this algorithm is that for maximization of any output variable, select an ideal fuzzy set for each numeric input variable, which, on average, would generate the maximum value of the output variable. For the minimization purpose, select those fuzzy sets for respective input variables that would result in the least possible value of the output variable, available in the data records. The procedural operation for automatic generation of rules for optimization purpose is based upon following outline. For every output variable that has been chosen by the user for optimization purpose, do as given in the following:

Fig. 7.5 The framework for self-development of optimization rule-base [240] First of all, copy all the input variables and corresponding fuzzy sets from list fuzzy to a linked list VariScore. To each and every fuzzy set, in that list, allocate slots Score and Count. Navigate all the rules. For any rule, whose consequent part consists of the output variable 230

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

that is currently under scrutiny, do as given in the following: Find the peak value of the fuzzy set assigned to the output variable in that rule. Suppose it is equal to N1. Next, identify all the input variables and their assigned fuzzy sets involved in antecedent part of that rule, and for all these fuzzy sets of corresponding input variables, listed in VariScore, add N1 to their slots Score and add 1 to their slots Count. Perform the same process for all the rules and at the end calculate the average score for each fuzzy set of all the input variables by dividing the respective value of Score with that of Count. Next, find out for each input variable, which fuzzy set possesses the highest average score and which one possesses the lowest. The fuzzy set with highest average score, for each input variable, is selected for maximization and the one with lowest average score is selected for minimization of that particular output variable. Figure 7.5 shows the framework graphically of the same methodology that was explained. The same procedure is repeated for all the other output variables that have been chosen for optimization purpose. At the end, the optimization rule-base would be ready and all the rules would be printed to the CLIPS file Sets_Rules.clp.

7.7 Data Structures and Coding The nature of the algorithm discussed in the previous sections depicts a complex requirement of data handling and management. It is, thus, very important to design suitable data structures for simplifying the operations to be performed and for keeping the processing time of the algorithm within acceptable limits. C++ has been selected as the programming language for coding the algorithm, because of its high efficiency and flexibility in programming engineering software. The most unique feature of C++, that was fully used in coding of the algorithm, is the availability of pointers. The pointer, defined as address of a variable located in the memory, is kind of hemoglobin for C++. It is mainly used for accessing array elements, obtaining and optimizing memory, and facilitating development of complex but highly efficient data structures, like linked lists. The foremost data structure utilized in this work is also the linked list (1-D and 2-D).

Fig. 7.6 Mechanism of the linked list Set, consisting of 6 nodes [240]

231

University of Engineering & Technology, Taxila-Pakistan

Linked list is a self-referential structure containing not only the items of information but also the pointer to the next structure in the list. The mechanism of a simple linked list Set used in the above mentioned algorithm is shown in Figure 7.6. The each node of the list contains following: (1) Record number; (2) Identity of variable (variable ID 1 represents input variable thickness, 2 represents welding current, and 3 represents weld strength); (3) Value of the variable used in each node; and (4) A link (or pointer) to next node of the list. Figure 7.6 shows the linked list consists of six nodes, i.e. of two record sets only, and three data values per record set. The finally developed expert system for welding process optimization consists of about 112 record sets with up to 18 data values per record set, employing that the linked list Set consisted of more than 2000 nodes. Following is the code, in C++ pattern, representing the declaration of linked list Set: struct Set {

//The name int Record_num; int Variable_ID; double Value; struct Set *next;

//Record number is an integer //Variable ID is also declared as an integer //Value of variable is defined as a floating-point number //*next is pointer to next node in the linked list

}; The coding starts with the declaration of the name of the structure, the ingredients of whose are contained within the braces. The first three ingredients are the declaration of proper data types for items of information. These items of information are record number, variable identity, and value of the variable involved. The fourth ingredient is the pointer that would contain the address, in memory, of next node of the linked list of structure Set. Unlike arrays, there is no limit of amount of data to be added to the linked lists. If the data would have been contained and managed using arrays instead of linked list, it would have led to occupation of major portion of memory, thus, leading to slow processing or even halting of the program.

7.7.1 Doubly Linked List The doubly linked list (2-D linked list) is a data structure developed for containment of data related to representation of fuzzy sets and rules mentioned in the algorithm. The 2-D linked list has been named so because of the fact that the list expands horizontally as well as vertically to an extent depending upon the data involved. The structure of the 2-D linked list Rule_consequent used for representation of rules related to the prediction rule-base is shown in Figure 7.7. The representation of only 2 rules within the list is shown in Figure 7.7. The horizontal chain consists of consequent parts of the rules (in 1 output variable per rule format), which are connected vertically to the respective antecedent parts. The antecedent parts of rules represent 1-D linked list of structure Rule_antecedent. It means that to every horizontal node of list Rule_consquent there is connected a separate 1-D linked list Rule_antecedent, and the whole combination results in a 2-D linked list.

232

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 7.7 A portion of 2-D linked list Rule_consequent [240] The coding of the list declaration, in C++ pattern, can be represented as follows: struct Rule_antecedent { int Variable_id; float Set_Num; float CF; struct Rule_antecedent *down; }; struct Rule_consequent { int Rule_num; int Out_var_id; float Out_set_no; float Out_CF; int Salience; struct Rule_antecedent *down; struct Rule_consequent *next; };

//Structure for input variables //ID of input variable (integer) //Assigned fuzzy set (floating-point) //Certainty Factor //Pointer to next node

//Rule No. //Output variable ID //Assigned set number for output variable //CF for output variable //Salience of the rule //Pointer to first input variable of the rule //Pointer to next rule

233

University of Engineering & Technology, Taxila-Pakistan

The first declaration is that of structure of 1-D linked list Rule_antecedent containing details of input variables involved in making of any rule. The second declaration is that of 2D linked list Rule-consequent, which contains also the pointers to next node of its own (i.e. detail of output variable of next rule) as well as to the first node of the related linked list Rule_antecedent containing details of all of input variables involved in the rule.

7.8 Application Examples The fuzzy expert system presented in this chapter has been named as EXWeldHSLASteel (EXpert system for Welding of High Strength Low Alloy Steel of thin walled structures). This section describes the application examples showing the selfdevelopment of the knowledge-base and interface of EXWeldHSLASteel. The first example illustrates a fledgling knowledge-base that was self-developed from a very limited experimental data provided to it, while the second one portrays a veteran knowledge-base that reached this stage by continuously learning from the data that was supplied to it at different stages. The knowledge-base developed in second example covers all the experimental and statistical results of TIG welding experiments, presented in chapters 3 – 5 of this manuscript. Whereas, the third example covers the verfication of EXWeldHSLASteel predictions by comparing these with the experiments/simulation results. A limited experimental data related to linear welding taken from the chapter 3 is considered here as given in Table 7.1. The values for trailing, the fourth ingredient of experiments of chapter 3, have been intentionally not included in the Table 7.1. If the knowledge-base is developed based entirely upon these data, it is very likely that the expert system may provide anomalous results because of the fact that the other influential welding parameters (e.g., welding current etc.) have not been taken care of.

Table 7.1 Data for the fledgling knowledge-base No.

Thickness (mm)

Voltage

Speed

Weld strength

Distortion

(V)

(cm/min)

(MPa)

(mm)

1

3

10.5

18

784.6

4.9

2

3

13.5

18

749.5

5.2

3

3

10.5

15

751.0

5.1

4

3

13.5

15

740.0

5.6

5

5

10.5

18

759.6

4.1

6

5

13.5

18

729.5

3.7

7

5

10.5

15

737.8

3.6

8

5

13.5

15

725.0

5.0

234

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

7.8.1 Example 1: A Fledgling Knowledge-Base In this example, suppose it is asked to expert system to develop its knowledge-base and update its interface based upon the data given in Table 7.1 and it is also asked to include weld strength, but not the distortion, as output variable for optimization. Following is the detail of triangular fuzzy sets developed itself by expert system in the Fuzzy CLIPS format: (deftemplate Obj_Weld_Strength 0 100 percent ( (Low (0 1) (5 1) (95 0) ) (High (5 0) (95 1) (100 1) ) ) ) (deftemplate Thickness 2 6 mm ( (S1 (2 1) (3 1) (5 0) ) (S2 (3 0) (5 1) (6 1) ) ) ) (deftemplate Welding_Voltage 9 15 V ( (S1 (9 1) (10.5 1) (13.5 0) ) (S2 (10.5 0) (13.5 1) (15 1) ) ) ) (deftemplate Welding_Speed 13.5 19.5 cm/min ( (S1 (13.5 1) (15 1) (18 0) ) (S2 (15 0) (18 1) (19.5 1) ) ) ) (deftemplate Weld_Strength 670 810 MPa ( (S1 (670 1) (725 1) (737.8 0) ) (S2 (725 0) (737.8 1) (749.5 0) ) (S3 (737.8 0) (749.5 1) (751 0) ) (S4 (749.5 0) (751 1) (759.6 0) ) (S5 (751 0) (759.6 1) (784.6 0) ) (S6 (759.6 0) (784.6 1) (810 1) ) ) ) (deftemplate Distortion 0.5 7.5 mm ( (S1 (0.5 1) (3.6 1) (4.1 0) ) (S2 (3.6 0) (4.1 1) (4.9 0) ) (S3 (4.1 0) (4.9 1) (5.1 0) ) 235

University of Engineering & Technology, Taxila-Pakistan

(S4 (4.9 0) (5.1 1) (5.2 0) ) (S5 (5.1 0) (5.2 1) (5.6 0) ) (S6 (5.2 0) (5.6 1) (7.5 1) ) ) ) As explained in section 7.4, the first template is the one defining sets for maximization and minimization of weld strength. Next three templates belong to input numeric variables, namely thickness, welding voltage and welding speed. The maximum allowable number of fuzzy sets for output variable was set to 6, thus, the last two templates have selected the best 6 values out of 8 for assignment of fuzzy sets. Following is the detail of six rules, selfdeveloped by the expert system and operated by its optimization module: (defrule optimization1 (declare (salience 1000)) (Obj_Weld_Strength High) (or (not (Thickness ?)) (Thickness S2))

 (assert (Thickness S2))) (defrule optimization2 (declare (salience 1000)) (Obj_ Weld_Strength High) (or (not (Welding_Voltage ?)) (Welding_Voltage S1))

 (assert (Welding_Voltage S1))) (defrule optimization3 (declare (salience 1000)) (Obj_ Weld_Strength High) (or (not (Welding_Speed ?)) (Welding_Speed S2))

 (assert (Welding_Speed S2))) (defrule optimization4 (declare (salience 1000)) (Obj_ Weld_Strength Low) (or (not (Thickness ?)) (Thickness S1))

 (assert (Thickness S1))) (defrule optimization5 (declare (salience 1000)) (Obj_ Weld_Strength Low) (or (not (Welding_Voltage ?)) (Welding_Voltage S2))

 (assert (Welding_Voltage S2))) (defrule optimization6 (declare (salience 1000)) 236

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

(Obj_ Weld_Strength Low) (or (not (Welding_Speed ?)) (Welding_Speed S1))

 (assert (Welding_Speed S1))) Out of these six rules the first three perform the maximization operation, while the others perform minimization. Let us consider the first rule, whose first line consists of declaration of name of rule and its salience. The salience value is very high because the optimization rules are supposed to fire before prediction rules. The next two lines constitute the IF part of the rule and connected by AND operator. The antecedent part can be read as, “IF the objective is weld strength high AND Thickness is not fixed or Thickness is S2”. The “=>” represents the term “THEN”. The consequent part of the rule can be read as, “Thickness is S2”. Following is the detail of eight rules, self-developed by the expert system and operated by its prediction module: (defrule prediction1 (declare (salience 15) (CF 1)) (Thickness S1) (Welding_Voltage S1) (Welding_Speed S1)

 (assert (Weld_Strength S2 AND S3) CF 0.6918 (Distortion S6) CF 1)) (defrule prediction2 (declare (salience 15) (CF 1)) (Thickness S1) (Welding_Voltage S2) (Welding_Speed S1)

 (assert (Weld_Strength S3) CF 1 (Distortion S4) CF 1)) (defrule prediction3 (declare (salience 15) (CF 1)) (Thickness S1) (Welding_Voltage S1) (Welding_Speed S2)

 (assert (Weld_Strength S6) CF 1 (Distortion S2) CF 1)) (defrule prediction4 (declare (salience 15) (CF 1)) (Thickness S1) (Welding_Voltage S2) (Welding_Speed S2)

 (assert (Weld_Strength S5) CF 1 (Distortion S1) CF 0.7826)) 237

University of Engineering & Technology, Taxila-Pakistan

(defrule prediction5 (declare (salience 15) (CF 1)) (Thickness S2) (Welding_Voltage S1) (Welding_Speed S1)

 (assert (Weld_Strength S2) CF 0.243697 (Distortion S3) CF 1)) (defrule prediction6 (declare (salience 15) (CF 1)) (Thickness S2) (Welding_Voltage S2) (Welding_Speed S1)

 (assert (Weld_Strength S1) CF 1 (Distortion S5) CF 1)) (defrule prediction7 (declare (salience 15) (CF 1)) (Thickness S2) (Welding_Voltage S1) (Welding_Speed S2)

 (assert (Weld_Strength S5) CF 1 (Distortion S2) CF 1)) (defrule prediction8 (declare (salience 15) (CF 1)) (Thickness S2) (Welding_Voltage S2) (Welding_Speed S2)

 (assert (Weld_Strength S4) CF 1 (Distortion S3) CF 0.90625)) Considering 2 fuzzy sets each for thickness, voltage and speed of welding, the total number of prediction rules is 16. Salience of each rule is equal to 15 (= number of input variables in the rule × 5). First line of each rule consists of the name of rule, its salience and calculated certainty factor (CF). The next three lines form the antecedent part of rule, while the last line is the consequent part. In consequent parts of all the rules, two assertions have been made, one for weld strength and other one for distortion. Figure 7.8 shows the process of interface of the expert system from fuzzy clips and Figure 7.9 shows the interface of the expert system related to the fledgling knowledge-base. In Figure 7.9, top of the interface shows two buttons, one is for processing the optimization

238

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Fig. 7.8 Process of interface of expert system from fuzzy clips

239

University of Engineering & Technology, Taxila-Pakistan

and prediction of welding process, while the second one is for self-development of expert system for optimizing welding process according to new data provided to it. The slider bar provides the user whether to maximize or minimize the selected output variable and by how much weightage. Check-boxes are for numerical input variables asking the user whether to pre-fix them or optimize them according to the desired objective(s). These are followed by the choice-boxes for categorical input variables providing the possible choices for respective variables, including the option of leaving them open for optimization (i.e. “Do not know”).

Button for selfdevelopment

Processing button Slider bar for optimization

Check box for numerical variable Information pane Choice box for categorical variable

Fig. 7.9 Interface of expert system representing fledgling knowledge-base At bottom of the interface there is information pane that initially displays the introduction of EXWeldHSLASteel and then, after processing, it displays the results of optimization and prediction processes. Suppose EXWeldHSLASteel is provided with following input: • Objective: maximize weld strength with weightage of 95% • Thickness of material prefixed to 3.5 mm. • Welding voltage and welding speed: open for optimization.

240

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Pressing the Process button starts the processing of expert system and finally following results are displayed in the information pane: • The recommended welding speed is 17 cm/min. • The recommended welding voltage is 11 A. • The predicted weld strength is 755 MPa. • The predicted distortion is 4.3 mm.

7.8.2 Example 2: A Veteran Knowledge-Base The veteran knowledge-base consists of all the data obtained from welding experiments of HSLA steel, their ANOVA results, and empirical models provided in chapters 3 – 5 for weld strength, distortion, and residual stresses fed to the knowledge-base. Automatically developed fuzzy sets and rules related to optimization rule-base (in Fuzzy CLIPS format) have been presented in Appendix A2-6. Figure 7.10 shows the interface of the expert system related to that knowledge-base. Three output variables, namely: weld strength, distortion and residual stresses are selected for simultaneous optimization purpose. The interface contains three slider bars for this purpose. It can be further observed that the expert system at this stage is dealing with six input variables, four of them numeric and two categorical. Suppose the expert system is provided with following input: • Simultaneously maximize/minimize following performance measures: (1) maximize weld strength minimize with weightage of 70%; (2) minimize distortion with weightage of 100%; and (3) minimize residual stresses with weightage of 95%. • Prefix the value of work piece material thickness to 5 mm. • Prefix the value of welding current to 230 A. • Weld Type is Linear. • Leave the other input variables: welding voltage, trailing, and welding speed as open in order to be optimized. EXWeldHSLASteel provides following results, as displayed in information pane of the interface: • The recommended trailing is Ar. • The recommended value of welding speed is 17 cm/min. • The recommended value of welding voltage is 11. V. • The predicted value of weld strength is 780 MPa. • The predicted value of distortion is 2.0 mm. • The predicted value of residual stresses is 350 MPa. 241

University of Engineering & Technology, Taxila-Pakistan

It is to be considered that the maximized value of weld strength is very satisfactory considering the fact that very high value of thickness was prefixed. Residual stresses value minimized by EXWeldHSLASteel seems quite high because of the fact that weightage of this objective was small as compared to other opposing objectives.

Fig. 7.10 Interface of expert system representing veteran knowledge-base

7.8.3 Example 3: Verification of EXWeldHSLASteel Predictions For the verification of EXWeldHSLASteel predictions against the welding parameters that already not fed for the desirability of maximization/minimization of responses of weld strength/distortion & residual stresses as given in Table 7.2 with prefixing the thickness and welding current, the responses are compared with experiments/simulation results as given in Table 7.3. The maximum variations observed between responses values are between 3-8% only.

242

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Table 7.2 Welding Parameters for EXWeldHSLASteel Predictions S.No.

01 02 03 04

Sheet Thickness (mm)

Welding Current (A)

Welding Voltage (V)

Welding Speed (cm/min)

Trailing

Weld-Type

3.5 4.5 3.5 4.5

200 220 200 220

11.5 11.5 11.5 11.5

17 17 17 17

Ar Ar Ar Ar

Linear Linear Circumferential Circumferential

Table 7.3 Comparison of Responses against welding parameters in Table 7.2 EXWeldHSLASteel S.No.

Distortion

01

Weld Strength (MPa) 746

(mm) 3.00

Residual Stresses (MPa) 460.0

Weld Strength (MPa) 724

02

763

2.59

395.4

03

742

2.36

04

763

2.33

Experiment/ Simulation Distortion (mm) 3.25

Residual Stresses (MPa) 485

778

2.48

420

428.0

-

2.21

405

318.6

-

2.15

298

7.8.4 The Limitations of EXWeldHSLASteel The examples mentioned above present the compliance, efficacy, and adaptability of the developed expert system. Besides numerous advantages, EXWeldHSLASteel has also few minor limitations. To ensure the effectiveness and reliability of the expert system, it is utmost important that the welding experimental data provided to EXWeldHSLASteel, for purpose of further self-development, should be based upon some statistical DoE technique. If this is not taken care of then the system may provide anomalous optimization results and it may also fail to provide predictions of some of the welding performance measures desired. By providing more and more welding experimental data (based upon DoE technique) related to the input variables already in use by EXWeldHSLASteel, adds to its accuracy and reliability and providing welding experimental data related to some newly introduced variable, adds to its scope and span of application. If the data related to new welding input variable is based upon some fractional factorial design rather than full factorial design, it might compromise the accuracy of optimization and prediction results. If this situation is unavoidable then the accuracy and reliability of EXWeldHSLASteel can be enhanced to a certain degree by reducing the maximum allowable number of fuzzy sets of input numeric variables.

243

University of Engineering & Technology, Taxila-Pakistan

7.9 Chapter Summary and Conclusions This chapter presented an approach for designing mechanism of a self-developing expert system for optimizing welding process. The developed expert system possesses the ability to manage new TIG welding process variables, to self-develop fuzzy sets, to self-generate rules for optimization and prediction modules, to resolve the conflict resolution among the contradictory rules and to keep updated its interface accordingly. The algorithm pseudocodes of all the modules are presented in detail with necessary explanation. For the coding of algorithm C++ was used. The data structures used are 1-D linked list and the 2-D linked list that provides efficient means for handling the large data and also support to occupy minimum level of the memory required. Three application examples of expert system were provided in detail to understand the functioning. The first example explains the structure and working of a new and inexperienced knowledge-base. The example provides the automatically developed fuzzy sets, automatically generated optimization and prediction rules, and automatically developed expert system interface. The second example shows the working with an experienced knowledge-base that consist four numeric and two categorical input variables whereas the third example presents the verification of predictions of EXWeldHSLASteel. The examples show the optimization of input welding process parameters with respect to simultaneous accomplishment of three opposing objectives (outputs) i.e. maximization of weld strength, minimization of distortion and residual stresses and also provide the predicted values of these maximized and/or minimized output process variables at the optimized levels of input process variables. The contents discussed in this chapter clearly provide the picture of the uniqueness and distinctiveness of the presented self-developing expert system along with its application at high degree for the process of optimizing the TIG welding process of thin walled HSLA steel structures. Further, the feature of automatic development of knowledge-base makes it highly adaptable to the ever-changing environment of manufacturing or fabrication industry.

244

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

CHAPTER 8 CONCLUSIONS & RECOMMENDATIONS Chapter’s summary and conclusion is presented at the end of each chapter pertaining to particular studies conducted within the corresponding chapter. The seven chapters of this manuscript comprehensively dealt with the contemporary need of optimizing the welding process with the use of hybrid technique of experimental, simulation, analysis, optimization and artificial intelligence tools. Summarized concluding remarks of chapters are presented in the following to draw some recommendations and proposals for future research work. The first chapter was of introductory nature and provided the basics of welding technology, arc welding, welding variables, weld induced residual stresses and distortions in thin walled structures, artificial intelligence, expert system, challenges in welding domain, scope and objectives of the research, research methodology and organization of this dissertation. Chapter 2 covered the literature review related to welding technology and issues, welding process and optimization, welding simulation (computational weld mechanics) including computational and experimental work pertaining to circumferential welding, measurement of residual stresses including hole drilling residual stress measurements, application of artificial intelligence to manufacturing and welding along with the AI tools limitations. Chapters 3 – 5 utilized the statistically and FEM powered experimental techniques of studying the TIG welding process. The main targets of the research have been the identification of influential parameters upon welding performance measures and the quantification of their effects. The chapter’s summary detail is as: Chapter 3 included the welding experiments for analyzing and optimizing the TIG welding variable parameters (welding current, welding voltage, welding speed and application of Ar trailing) and their effect and significant on response (weld strength, residual stresses and distortion) on thin plates of HSLA steel of thicknesses 3 to 5 mm by applying DOE, ANOVA and numerical optimization according to desirability statistical techniques by using Design-Expert® and MINITAB®. Chapter 4 presented the welding simulations containing analytical model, FE formulation, heat models, material model, simulation approach in ANSYS, details of welding induced residual (axial & hoop) stresses fields and distortions (axial & radial shrinkages) and covered the details of experiments performed for the FE models (thermal and structural) validation. Chapter 5 included the detail of virtual design of experiments (DOE) by utilizing the simulations and optimization of welding parameters for circumferential welding of thin walled shell structure of high strength low alloy steel of thicknesses 3 to 5 mm and covered the effects of welding process parameters (weld speed and heat input), geometric parameters (cylinder thickness), root opening and tack weld orientations on residual stresses. It also included the detail of analysis and optimization of process parameters for linear and circumferential welding comprising nine variables i.e. six input variables (welding current, welding voltage, welding speed, material thickness, trailing and weld type) and three output variables (weld strength, distortion and residual stresses). Chapters 6 and 7 dealt with application of expert system tool, employing fuzzy reasoning mechanism, for finding the most suitable values of TIG welding parameters for 245

University of Engineering & Technology, Taxila-Pakistan

accomplishment of the objectives of maximizing weld strength and minimizing residual stresses & distortion. The ES was also made capable of predicting the values of performance measures at different combinations of input parameters. Chapter 6 presented a static design of the expert system tool, with much emphasis upon its configuration, constituents, and procedure of operation. The expert system provided promising results and proved beneficial within its limited scope of application. Chapter 7 continued with these basics and presented high level automation for self-development of the expert system. The methodology of selfdevelopment included automatic generation of fuzzy sets, prediction rules, optimization rules, updating of interface, and more with two examples based on knowledge base levels. Following two sections will provide the main conclusions drawn from the research work presented and the recommendations for its practical application at industrial level and also some directions for future research.

8.1 Conclusions The conclusive points related to the welding of thin walled structures of high strength low alloy (HSLA) steel and optimization of the TIG welding process using expert system tool developed after performing experiments (actual and virtual DOE based on simulation) and statistical optimization, have been categorized under following sub-headings of welding induced stresses & distortion and weld strength, effect of welding process parameters, the expert system and the researcher’s main contributions from the present research work:

8.1.1

Welding Induced Stresses & Distortions and Weld Strength

1. Weld strength increases with the reduction of residual stresses and distortion and decreases with the increase of residual stresses and distortion. 2. Weld strength increases with low values of welding current and voltage and high value of welding speed with the application of trailing according to the material thickness. Weld strength decreases above or below the optimal process parameters. 3. Residual stresses and distortion reduces with low values of welding current and voltage and high value of welding speed with the application of trailing according to the material thickness and increases with high values of welding current and voltage and low values of welding speed without the trailing application respectively. 4. From the comparison of responses (distortion and residual stresses), the results show that the low distorted samples give the low residual stresses and vice versa respectively. 5. In circumferential welding, high tensile and compressive axial residual stresses are observed on the thin walled cylinder inner and outer surfaces along and near the weld line respectively. Whereas, away from the weld line, compressive and tensile axial residual stresses observed on inner and outer surfaces respectively. 6. In circumferential welding near the weld line, the maximum axial and radial deflection is observed. Whereas, the axial shrinkage reduces continuously away from the weld line to a minimum or zero level. Similarly, in linear welding, the maximum deflection is observed along the weld line at about the centre of plate length and it reduces to minimum to the ends of plates both in longitudinal and axial direction.

246

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

8.1.2

Effect of Welding Process Parameters

1. Heat input per unit volume (i.e. welding current, welding voltage and welding speed) to the welded structures is the main influential parameter for the occurrence and control of residual stress & distortion levels. Increase in the value directly enhances the residual stress and distortion levels resulting reduction in the weld strength. 2. In parametric studies, the welding current with large parametric range is the most influential parameter with respect to thickness of sheets/cylinders besides other parameters. 3. Generally, plate thickness and cylinder wall thickness has negative effects on the magnitude of residual stress and distortion fields. Increased wall thickness results in reduction of distortion and residual stresses. 4. The increase in plate thickness from 30% to 65% (i.e. 3-5 mm thickness) result decrease in distortion about 15% to 30% and 20% to 25% in residual stresses respectively. The range of increase in weld strength is 10-15% (i.e. 690-790 MPa) only whereas the reduction in distortion is three times (2.2 - 7.2 mm) and two times in residual stresses (335 – 608 MPa) respectively. 5. Similarly, the increase in cylinder thickness from 30% to 65% (i.e. 3-5 mm thickness) result decrease in distortion about 25% to 45% and 30% to 35% in residual stresses respectively. Whereas, the reduction in distortion is three times (1.2 – 3.9 mm) and in residual stresses (268 – 577 MPa) is two times respectively. 6. The application of Ar trailing has negative effects on the magnitude of residual stress & distortion levels and positive effects on weld strength. Application of trailing results in reduction up to 15% of residual stresses and distortion accordingly. 7. The residual stresses and distortion of TIG welding are lower in circumferential welds than in linear welds at same parameters for same thicknesses of HSLA steel thin walled structures.

8.1.3

The Expert System

1. In this research work, expert system tool has been successfully applied for optimization of parameters and prediction of performance measures related to TIG welding process of thin walled HSLA steel structures domain. The optimization of parameters is performed based upon objective(s) of maximization and/or minimization of certain combination of performance measures. At the completion of optimization process the finalized settings of input variables are used to predict the values of the performance measures. This expert system tool possesses high potentials for reducing production cost, cutting down lead-time, and improving the product quality at expense of few seconds that the expert system would take to process. 2. The uncertainty and vagueness in relationship between inputs and outputs of the welding process can be effectively tackled by utilization of fuzzy logic. In this way, the knowledge related to the welding process can be represented by sets of fuzzy rules. The fuzzy rule-base can be optimized for maximum accuracy by applying simulated annealing algorithm. 247

University of Engineering & Technology, Taxila-Pakistan

3. The important feature of this research work is the success in imparting selfdeveloping abilities to the fuzzy expert system for welding process optimization. The presented expert system is capable of auto-managing data, self-developing fuzzy sets, self-generating rule-base, automatically updating expert system interface, and providing conflict resolution among contradictory rules. These abilities make the expert system exceedingly adaptable to continuously changing high-tech industrial environments, without need of human intervention in the field of welding of thin walled structures. 4. A data structure, named as doubly linked list, was introduced for handling the data related to rules for process decision making. This linked list provides efficient way of managing the storage and the processing of data along with allocation of just minute portion of memory. 5. Simulation and optimization of the welding process is becoming an efficient and effective approach to achieve high quality weld products with reduced residual stress and distortion. An expert system, EXWeldHSLASteel, based on experiments for linear welding as well as virtual experiments by performing FEA and simulation of welding process with experimental validation for circumferential welding, has developed to predict welding distortion and residual stress in thin walled structures. Due to its accuracy, computation speed and time, EXWeldHSLASteel can be applied in welding process and product design by engineers on shop floor very easily with minimum know-how. 6. Optimization using expert system, on the other hand, allows engineer to reach optimized process parameters much more efficiently and without hazardous impact to the environment by using the simulation results in real welding process of thin walled cylinders with the quality concept of “do it right at first time”. By using developed ES, the time to have optimized parameters for required response has reduced with cost avoidance and eliminates associated waste, rework, reduced cost of operation, reduced scrap and consumables and energy, as well as fumes and emissions. 7. Usually, an actual welding experiment sample of HSLA steel sheet of 3x260x500 mm (sheet cutting, preparation and welding) including testing (weld strength samples or residual stresses measurement and distortion) related to thin walled structures consume time 3-4 days with use of a good experimental setup and cost of $200 whereas in case of circumferential welding (3 x Ø300 x 300 mm), the cost would be two times. However, in optimization following DOE, the number of experiments required depends upon the number of factors and their levels. The sixteen experiments are required for only four factors with two levels for a full factorial design and time & cost would be sixteen times as mentioned above for only one material thickness. Whereas each simulation time (based on the element topology, thermal & structural boundary conditions, material model, time of load-steps and the corresponding sub-steps etc) required after developing three dimensional FE models is 7-8 hrs for plate and 20-24 hrs for circumferential welding for both thermal and structural analysis with the use of an IBM compatible PENTIUM-IV 2.4 GHz 248

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

computing machine with 6 GB RAM on a 64 bit platform for computational and data storage with extra hard disks whereas the size of one result file is 108 GB. The developed tool for optimization of welding process parameters and prediction of responses consumes only few seconds to give desired solution before the start of process on shop floor and this may be used in shipbuilding, aerospace and nuclear industries, oil and gas engineering and in other areas before the manufacturing of structural elements.

8.1.4

Researcher’s Main Contributions from the present Research Work

1. A fuzzy expert sytem (EXWeldHSLASteel) development for the optimization and prediction of TIG welding process (linear and circumferrential welding) of thin walled HSLA steel strucures. 2. Imparting self-developing capabilities for self-learning, self-correcting and selfexpanding to the developed expert system (EXWeldHSLASteel) for the optimization and prediction of TIG welding process of thin walled HSLA steel strucures. 3. Development of 3D fully parametric FE model with experimental validation for circumferrential welding of thin walled HSLA steel. 4. Development of empirical models for TIG welding process both for linear and circumferrential welds of HSLA steel for maximization of weld strength and minimization of distortions/residual stresses or as per desireability.

8.2 The Recommendations The previous section outlines important conclusions that provide the directives for increasing the viability of thin walled structure welding process with optimized parameters at industrial level. The major area of concern in welding domain is the optimization of welding process of thin walled shell structure of high strength low alloy steel to minimize the residual stresses and distortion for improvement of weld mechanical properties and production rate by using expert system. This research work has considerably contributed towards this requirement, related to TIG welding process of thin walled steel structure for linear and circumferential welding using hybrid technique as experimental and simulation with experimental validation with statistical analysis for numerical optimization and developing expert system. The vital recommendation, in this regard, is to use the parameters of welding resulting low input heat (low current, low voltage and high speed) with application of trailing with respect to material thicknesses for the maximum weld strength and minimum residual stresses and distortion in thin walled structures of HSLA steel for linear and circumferential welding. It is strongly recommended to utilize the presented expert system for deciding the values of important welding parameters as per objective before the start of actual welding process on shop floor. The user should be absolutely clear about the nature and requirements of any given TIG welding process, e.g., the setting parameters, fixed parameters, geometric parameters, structural boundary conditions etc. Finalize the set of objectives with corresponding values of weight-age and input this information to expert system along with the data related to the fixed conditions. For the best possible simultaneous achievement of 249

University of Engineering & Technology, Taxila-Pakistan

these requirements, the expert system will provide the suggestions of most suitable values for the parameters or response predictions under control of the user. Presently, the ES covers the six input parameters and three responses with the capability of automatic enhancement of the scope upon feeding the experimental data after further experimentation at any stage. Today, knowledge management is an emerging area which is gaining interest and importance by both industry and public sectors. For knowledge organizations, knowledge management will play a key role towards the success of transforming individual (expert) knowledge into organizational knowledge. Artificial intelligence is one of the key fields for developing and advancing the field of knowledge management and many knowledge management practitioners and theorists are overlooking this field [241]. This research work related to artificial intelligence by developing expert system in the domain of welding for optimizing welding process of thin walled structure will also serve the newly emerging field of knowledge management.

8.2.1

Proposals for Future Research

1. The comprehensive investigation and optimization of welding process can also be targeted utilizing other welding techniques like EBW, Laser, Plasma or hybrid. The various performance measures can then be compared with those achieved. 2. The developed FE model for TIG welding can also be applied to other welding processes as MIG, Plasma, EBW, or Laser etc. The model will require to be amended according to the nature and requirements of the process and further be used for virtual DOE and optimization. 3. The self-developing expert system can also be applied to other welding processes as well, like EBW, Laser, Plasma or Hybrid etc. The algorithm will require to be modified according to the nature and requirements of the process. 4. The scope and utilization of self-developing expert system for welding process can be extended to other materials, thicknesses and sizes of plates and cylinders etc. The algorithm, FE model and DOE for ANOVA results will require to be modified according to the nature and requirements of the process. 5. The self-developing expert system can also be applied to other manufacturing processes for thin walled structures as well, like heat treatment process, spinning, grinding, and metal forming etc. The algorithm will require to be modified according to the nature and requirements of the process. 6. The optimization of any manufacturing process can also be explored using Hybrid AI systems, for example the combination of expert system and artificial neural networks or FEM simulation with development of auto interfacing to each other as well as online optimization of process.

250

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

References [1] [2]

[3] [4]

[5]

[6] [7]

[8] [9]

[10]

[11] [12] [13] [14] [15] [16]

[17]

Welding Handbook, Volume 1, Eighth Edition; L. P. Connor; American Welding Society, Miami, FL; 1987. Michaela Ancuta Mocanita, “Computational Mechanics of Welding Complex Structures”. M.Sc. Engg. Thesis, Institute for Mechanical and Aerospace Engineering, Carleton University, Ottawa, Canada. http://www.lincolnelectric.com/knowledge/articles/content/arcweldfund.asp. M Ejaz Qureshi, “Analysis of Residual Stresses and Distortions in Circumferentially Welded Thin-Walled Cylinders”. PhD Thesis, National University of Science and Technology (NUST), Pak (Dec-2008). Muhammad Shah Alam, “Structural Integrity and Fatigue Crack Propagation Life Assessment of welded and weld repaired structures”. PhD Thesis, Louisiana State University and Agricultural and Mechanical, Dec-2005. James, F., The Procedure Handbook of Arc Welding, 14th edition. Lincoln Arc Welding Foundation, Cleveland, USA, 2000. M Abid, M Siddique and R A Mufti; Prediction of welding distortions and residual stresses in a pipe–flange joint using the finite element technique, Modeling Simulation in Material Science & Engineering, Vol. 13, pp. 1-16, 2005. D. Yapp. Management of Weld Quality. Lecture Notes on Topic MAN1.5, WERC 2007. Cranfield University, Bedford, UK. O.W. Blodgett. Distortion: How to minimize it with Sound Design Practices and Controlled Welding Procedures plus Proven Methods for Straightening Distorted Members, The Lincoln Electric Company, G261, pp. 1–19. H.D. Hibbitt and P.V. Marcal. A Numerical Thermo-mechanical Model for the Welding and Subsequent Loading of a Fabricated Structure. Journal of Computers and Structures; Vol. 3: pp. 1153–1174, 1073. Z. Feng and O. Ridge. Processes and Mechanisms of Welding Residual Stress and Distortion. Woodhead Publishing Limited, 2005. K. Masubuchi. Analysis of Welded Structures, Pergamon Press, 1980. L. Xiangyang. Influence of Residual Stress on Fatigue Failure of Welded. PhD Thesis, North Carolina State University, 2002. D. Radaj. Heat Effects of Welding, ISBN 0-387-54820-3, Springer-Verlag, 1992. L. Hong and H. Ren. Simulation of Welding Deformations of Ship Structures. Key Engineering Materials; Vol. 324-325, pp. 651-654, 2006. Y. Ueda, K. Fukuda and Y.C. Kim. New Measuring Method of Axis-symmetric Three-Dimensional Residual Stresses Using Inherent Strains as Parameters. Journal of Engineering Materials and Technology; Vol. 108, pp. 328-334, 1986. S. Vaidyanathan, A.F. Todato and I. Finnie. Residual Stress Due to Circumferential Welds. ASME Journal of Engineering Material and Technology; Vol. 95, pp. 233237, 1973.

251

University of Engineering & Technology, Taxila-Pakistan

[18] M. Siddique, M. Abid, H.F. Junejo and R.A. Mufti. 3-D Finite Element Simulation of Welding Residual Stresses in Pipe-Flange Joints: Effect of Welding Parameters. Materials Science Forum; Vol. 490-491: pp. 79-84, 2005. [19] C.H. Lee and K.H. Chang. Three-dimensional finite element simulation of residual stresses in Circumferential Welds of Steel Pipe Including Pipe Diameter Effects. Journal of Material Science & Engineering-A, 2007. doi:10.1016/j.msea.2007.10.011. [20] A.M. Malik, E.M. Qureshi and N.U. Dar. Analysis of Arc Welded Thin-Walled Cylinders to Investigate the Effects of Welding Process Parameters on Residual Stresses. Materials Science Forum; Vol. 575-578, pp. 763-768, 2008. [21] Y. Ueda and T. Yamakawa, Thermal stress analysis of metals with temperature dependent mechanical properties, Proc. of Int. Conf. on Mechanical Behavior of Materials, Vol. III, pp. 10-20, 1971. [22] Fanous IFZ, Younman MYA, Wifi A.S. Study of the Effects of Boundary Conditions on Residual Stresses in Welding Using Element Birth and Element Movement Techniques. ASME Journal of Pressure Vessel Technology; Vol. 125(4), pp. 432439, 2003. [23] H. Porzner-Possibilities of Numerical Simulation for Evaluation and optimization of Welded Designs-Mathematical Modeling of Weld Phenomena 5, Seggau, Graz, Austria, 2001. [24] S.R. Ikeagu. Evaluating the Effects of Different Welding Processes on the Distortion of 4mm Thick DH36 Ship Panels. M. Sc. Thesis, School of Applied Sciences, Cranfield University, 2007. [25] D. Deng, W. Liang and H. Murakawa. Determination of Welding Deformation in Fillet-Welded Joint by means of Numerical Simulation and Comparison with Experimental Measurements. Journal of Materials Processing Technology; Vol. 183(2-3): pp. 219-225, 2007. [26] B. Ahmed. Distortion Response of Synchronized Tandem Wire Gas Metal Arc Welding. M. Sc. Thesis, School of Industrial and Manufacturing Science, Cranfield University, 2006. [27] Y. Takeda. Prediction of Butt Welding Deformation of Curved Shell Plates by Inherent Strain Method. Journal of Ship Production; Vol. 18(2): pp. 99-104, 2002. [28] Teng JG, Lin X, Rotter JM, Ding XL. Analysis of Geometric Imperfections in FullScale Welded Steel Silos. Journal of Engineering Structures; Vol. 27(6): pp. 938950, 2005. [29] Naeem Ullah Dar, Ejaz M.Qureshi, Iqbal Khan and A.M. Malik. Welding Quality and Cost: A Comprehensive Comparative Study. Advanced Design and Manufacture Conference, Harbin, China, 2006. [30] Mikael Ericsson, Simulation of robotic TIG-welding, Division of Robotics Department of Mechanical Engineering, Lund Institute of Technology Lund University, P.O. Box 118, SE-221 00 Lund, Sweden. [31] ASM Handbook (Welding, Brazing and Soldering). Vol. 6, pp.190. [32] Konar, A.: Introduction to Artificial Intelligence and Soft Computing, in: Artificial Intelligence and Soft Computing; CRC Press LLC, Florida, 2000.

252

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

[33] Hopgood, A.A.: Introduction, in: Intelligent Systems for Engineers and Scientists; 2nd Edition; CRC Press LLC, Florida, 2001. [34] K.Y. Benyounis and A.G. Olabi; Optimization of different welding processes using statistical and numerical approaches – A reference guide; Advances in Engineering Software, Vol. 39, Issue 6, pp. 483-496, 2008. [35] Tanco, M; Ilzarbe, L; Viles, E; Alvarez, M J; Applying design of experiments to improve a laser welding process; Proceedings of the I MECH E Part B Journal of Engineering Manufacture, Vol. 222, No. 8, pp. 1035-1042, 2008. [36] V Gunaraj, N Murugan; Application of response surface methodology for predicting weld bead quality in submerged arc welding of pipes; Journal of Materials Processing Technology, Vol. 88, Issues 1-3, pp. 266-275, 1999. [37] K.Y. Benyounis, A.G. Olabi, M.S.J. Hashmi; Multi-response optimization of CO2 laser-welding process of austenitic stainless steel; Optics & Laser Technology, Vol. 40, Issue 1, pp. 76-87, 2008. [38] K.Y. Benyounis, A.G. Olabi, M.S.J. Hashmi; Effect of laser welding parameters on the heat input and weld-bead profile; Journal of Materials Processing Technology, Vol. 164-165, pp. 978-985, 2005. [39] K.Y. Benyounis, A.G. Olabi, M.S.J. Hashmi; Optimizing the laser-welded butt joints of medium carbon steel using RSM; Journal of Materials Processing Technology, Vol. 164-165, pp. 986-989, 2005. [40] Guo, D. L.; Guan, Q.; Li, C. Q., New welding technique for aerospace thin-walled structures-LSND technology, In: ICAS, Congress, 18th, Beijing, China, Sept. 20-25, 1992, Proceedings. Vol. 2 (A93-14151 03-01), pp. 1705-1711. [41] L. Volden, O. Gundersen, G. Rarvik, Development of Residual Stresses in High Strength Low Alloy Steel, Proceedings of the Ninth International Offshore and Polar Engineering Conference Brest, France, May 30-June 4, 1999, ISBN 1-880653-39-7 (Set); ISBN 1-880653-43-5 (Vol. IV); ISSN 1098-6189 (Set). [42] Q. Guan, C.X. Zhang, and D.L. Guo, Dynamic control of welding distortion by moving spot heat sink, Welding in the World, Vol. 33(4), pp. 308-313, 1994. [43] E.M. van der Aa, I.M. Richardson and M.J.M. Hermans: Welding with a Trailing Heat Sink: How to Optimise the Cooling Parameters?, ‘Trends in Welding Research’, Pine Mountain, Georgia USA, ASM International, 2005. [44] E.M. van der Aa, Local Cooling during Welding: Prediction and Control of Residual Stresses and Buckling Distortion, PhD thesis Delft University of Technology, 2007. [45] Lindgren, L.E., Finite Element Modeling and Simulation. Part 1: Increased Complexity, Journal of Thermal Stresses, Vol. 24, pp. 141-192, 2001. [46] Lindgren, L.E., Finite Element Modeling and Simulation. Part 2: Improved Material Modeling, Journal of Thermal Stresses, Vol. 24, pp. 195-231, 2001. [47] Lindgren, L.E., Finite Element Modeling and Simulation. Part 3: Efficiency and Integration, Journal of Thermal Stresses, Vol. 24, pp. 305-334, 2001. [48] Yaghi, A., Becker, A., State of the Review–Welding Simulations using Finite Element Methods, Report No. FENET-UNOTT-DLE-08, School of Mechanical, Materials & Manufacturing Engineering, University of Nottingham, UK, 2004. 253

University of Engineering & Technology, Taxila-Pakistan

[49] Rosenthal, D., The Theory of Moving Heat Source and its Application to Metal Treatment, Transaction ASME, 1946. [50] Goldak, J., Zhou, J., Breiguine, V., and Montoya, F., Thermal Stress Analysis of Welds: From melting Point to Room Temperature, JWRI, Vol. 25, No. 2, pp. 185189, 1996. [51] Goldak, J„ Chakravarti, A. and Bibby, M., A new Finite Element Model for Heat Sources, Metallurgical Transactions B, vol. 15 B, pp. 299-305, 1984. [52] Goldak, J„ Bibby, M, Moore, J., House, R. and Patel, B„ Computer Modeling of Heat Flow in Welds, Metallurgical Transactions B, Vol. 17 B, pp. 587-600, 1986. [53] Ueda, Y., and Yamakawa, T., Analysis of Thermal-Elastic Stress and Strain During Welding by Finite Element Method, JWRI, Vol. 2, No. 2, 1971. [54] Hibbitt, H.D., and Marcal, P.V., A Numerical Thermo-Mechanical Model for the Welding and Subsequent Loading of Fabricated Structure, Journal of Computers and Structures, Vol. 3, pp. 1145-1174, 1973. [55] Friedman, E., Thermo-Mechanical Analysis of the Welding Process using the Finite Element Method, ASME Journal of Pressure Vessel Technology, pp. 206-213, 1975. [56] Andersson, B.A.B., Thermal Stresses in Submerged Arc Welded Joint Considering Phase transformations, ASME Journal of Engineering Material and Technology, Vol. 100, pp. 356-362, 1978. [57] McDill, J.M., Oddy, A.S. and Klien, M.E., Data Transfer for 3-D Adaptive ThermalElasto-Plastic Finite Element Analysis, Simulation of Materials Processing: Methods and Applications, Shen & Dawson (eds), 1995. [58] McDill, J.M., Oddy, and Goldak, J.A., An Adaptive Mesh-management Algorithm for Three Dimensional Automatic Finite Element Analysis, Transactions of CSME, Vol. 15, No. I, pp. 57-70, 1991. [59] McDill, J.M., Goldak, J.A., Oddy, A.S., and Bibby, M J., Isoparametric Quadrilaterals and Hexahedrons for Mesh-Grading Algorithms, Communications in Applied Numerical Methods, Vol. 3, pp. 155-163, 1987. [60] McDill, J.M and Oddy, A.S., A Non-Conforming Eight to 26-Node Hexahedron for Three Dimensional Thermal-Elasto-Plastic Finite Element Analysis, Journal of Computers and Structures, Vol. 54, No. 2, pp. 183-189, 1995. [61] McDill, J.M and Oddy, A.S., Arbitrary Coarsening for Adaptive Mesh Management in Three Dimensional Automatic Finite Element Analysis, Mathematical Modeling and Science Computing, Vol. 2, pp. 1072-1077, 1993. [62] Runnemalm, H. and Hyun, S., Three-Dimensional Welding Analysis using an Adaptive Mesh Scheme, Computer Methods in Applied Mechanics and Engineering, Vol. 189, pp. 515-523, 2000. [63] McDill, J.M, Runnemalm, K.H. and Oddy, A.S., An 8-to-16 Node Solid Graded Shell Elements for Far-Field Applications in 3-D Thermal Mechanical FEA, 12th International Conference on Mathematical and Computer Modeling and Scientific Computing, Chicago, 1999.

254

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

[64] Lindgren L.E., Haggblad H.A., McDill J.M.J. and Oddy A.S., Automatic Re-meshing for Three-Dimensional Finite Element Simulation of Welding, Computer Methods in Applied Mechanics and Engineering, Vol. 147, pp. 401-409, 1997. [65] Hyun, S. and Lindgren, L.E., Smoothing and Adaptive Re-meshing Schemes for Graded Elements, International Journal of Communications in Numerical Methods in Engineering., Vol. 14, pp. 1-17, 2001. [66] H.F. Junejo. Finite Element Modeling of Weld Based Rapid Prototyping, PhD Thesis, Ghulam Ishaq Khan Institute of Engineering & Technology, Pakistan, 2006. [67] Leggatt, R.H., White, J.D., Predicting Shrinkage and Distortion in a Welded Plate, Proceedings of the Conference on Residual Stresses in Welded Construction and their Effects, pp.119-132, 1978. [68] Steen W. M., Laser Materials Processing, 2nd edition, Springer Verlag, London, UK, 1998. [69] Ueda, Y., Kim, Y.C., Measuring Theory of Three Dimensional Residual Stresses using a Thinly Sliced Plate Perpendicular to Welded Line, Transactions of JWRI, Vol.14, No.2, pp.151-157, 1985. [70] Vaidyanathan S., Todato A.T., and Finnic, I., Residual Stresses due to Circumferential Welds, ASME J. Engineering Material and Technology, pp. 239-242, 1973. [71] Hill, M.R. and Nelson, D.V., The Localized Eigenstrain Method for Determination of Triaxial Residual Stress in Welds, PVP 373, pp.397-403, 1998. [72] Dong, Y., Hong, J.K., Tsai, C.L. and Dong, P., Finite Element Modeling of Residual Stresses in Austenitic Stainless Steel Pipe Girth Welds, Welding Journal, Vol.76, No.10, pp.442-449, 1997. [73] Vaidyanathan, S„ Weiss, H. and Finnic, I., A Further Study of Residual Stresses in Circumferential Welds, ASME Journal of Engineering Material and Technology, pp. 239-242, 1973. [74] Rybicki, E.F., McGuire, P.A., Merrick, E., and Wert, B., "The Effect of Pipe Thickness on Residual Stresses due to Girth Welds," Journal of Pressure Vessel Technology, Vol. 104, pp. 204-209, 1982. [75] Lindgren L.E. and Karlsson, L., Deformations and Stresses in Welding of Shell Structures, Int. Journal of Numerical Methods in Engineering, Vol. 25, pp. 635-655, 1988. [76] Karlsson, R.I. and Josefson, B.L, Three Dimensional Finite Element Analysis of Temperature and Stresses in a Single-Pass Butt-Welded Pipe, ASME Journal of Pressure Vessel Technology, Vol. 112, pp. 76-84, 1990. [77] Karlsson, L., Jonsson, M., Lindgren, L,E., Nasstrom, M. and Troive, L., Residual Stresses and Deformations in a Welded Thin-Walled Pipe, Proceedings of ASME Pressure Vessel and Piping Conference, Hawaii, Weld Residual Stresses and Plastic Deformation, PVP-Vol. 17.1, pp. 7-14, 1989. [78] Michaleris, P., Residual Stress Distributions for Multi-Pass Welds in Pressure Vessel and Piping Components, Proceedings of ASME Pressure Vessel and Piping Conference, Montreal, Canada, 1996, Residual Stresses in Design Fabrication, Assessment and Repair, PVP-Vol. 327, 1996. 255

University of Engineering & Technology, Taxila-Pakistan

[79] Dike, J.J., Ortega, A.R. and Cadden, C.H., Finite Element Modeling and Validation of Residual Stresses in 304L Girth Welds, 5th International Conference on Trends in Welding Research, Pine Mountain, GA, June 1-5, 1998. [80] Mohr, W.C., Michaleris, P. and Kirk, M.T., An Improved Treatment of Residual Stresses in Flaw Assessment of Pipes and Pressure Vessel Fabricated From Ferritic Steel, Proceedings of ASME Pressure Vessel and Piping Conference, Orlando, Florida, 1997. [81] Brickstad, B. and Josefson, B.L„ A Parametric Study of Residual Stresses in MultiPass Butt-Welded Stainless Steel Pipes, International Journal of Pressure Vessels and Piping, Vol. 75, pp. 11-25, 1998. [82] Teng, T.L. and Chang, P.H., A Study of Residual Stresses in Multi-Pass Girth Butt Welded Pipes, International Journal of Pressure Vessels and Piping, Vol. 74, pp. 5970, 1997. [83] Teng, T.L. and Chang, P.H„ Three-Dimensional Thermo-mechanical Analysis of Circumferentially Welded Thin-Walled Pipes, International Journal of Pressure Vessels and Piping, vol. 75, pp. 237-247, 1998. [84] Dong, P. and Zhang, J„ Residual Stresses in Strength-Mismatched Welds and Implication on Fracture Behavior, Engineering Fracture mechanics, Vol. 64, pp. 485505, 1999. [85] Engelhard, G., Halrip, L.M., Pellkofer, D., Schmidt, J. and Weber, J„ Optimization of Residual Welding Stresses in Austenitic Stainless Steel Piping: Proof testing and Numerical Simulation of Welding and Post Welding Process, Nuclear Engineering and Design, Vol. 198, pp. 141-151, 2000. [86] Fricke, S„ Keim, E. and Schmidt, J„ Numerical Weld Modeling - A Method for Calculating Weld-Induced Residual Stresses, Nuclear Engineering and Design, Vol. 206, pp. 139-150, 2001. [87] D, Deng and H, Murakawa. Numerical Simulation of Temperature Field and Residual Stresses in Multi-Pass Welds in Stainless Steel Pipe and Comparison with Experimental Measurements. Journal of Computational Materials Science, Vol. 37, pp. 269-277, 2006. [88] D, Deng, H, Murakawa and W. Liang. Numerical and Experimental Investigations on Welding Residual Stress in Multi-Pass Butt-Welded Austenitic Stainless Steel Pipe. Journal of Computational Materials Science, Vol. 42, pp. 234-244, 2008. [89] M. Siddique, Experimental and Finite Element Investigation of Residual Stresses and Distortions in Welded Pipe-Flange Joints, PhD Thesis, Ghulam Ishaq Khan Institute of Engineering & Technology, Pakistan, 2005. [90] I. Sattari-Far and Y. Jawadi. Influence of Welding Sequence on Welding Distortions in Pipes. International Journal of Pressure Vessels and Piping, 2007. doi:10.1016/j.ijpvp.2007.07.003 [91] A. Kermanpur, M. Shamanian and V.E. Yeganeh. Three Dimensional Thermal Simulation and Experimental Investigation of GTAW Circumferentially Butt-Welded Incoloy 800 Pipes. Journal of Materials Processing Technology, 2007. doi:10.1016/j.jmatprotec.2007.08.009.

256

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

[92] Withers P. and Bhadeshia H. Overview-Residual Stress Part 2 - Nature and origins. Materials Science and Technology, Vol. 17(4): pp. 366-375, 2001. [93] Moreira PMGP, Trummer V.R. and Da-silva RAM. Residual Stress Evaluation of a MIG Welded Plate using the Sectioning Method and the Hole Drilling Method. www.fe.up.pt/~pedrom/artigos2/residual_APAET2008.pdf. [94] Totten, G., Howes, M. and Inoue, T., Handbook of Residual Stresses and Deformation of Steel, ASM International, Material Park, USA, 2003. [95] C.W. Tuck. The Measurement of Residual Stresses, BGA Technical Publications, British Gear Association, 1999. [96] Withers, P.J. and Bhadeshia, H.K.D.H, Residual Stress Part 1 - Measurement Techniques, Material Science and Technology, Vol. 17, pp. 355-365, 2001. [97] J. Lu. Handbook of Measurement of Residual Stresses, Society of Experimental Mechanics Inc. The Fairmont Press, Inc. USA. 1996. [98] M. Hayashi and M. Ishiwata. Residual Stress Measurement of Butt Welded Carbon Steel Pipe by Neutron Diffraction, Proceedings of the 13th International Conference on NDE in the Nuclear and Pressure Vessel Industries, pp.477-485, 1995. [99] W.J. Shack, W.A. Ellingson and L.E. Pahis. Measurement of Residual Stresses in Type 304 Stainless Steel Piping Butt Weldments, EPRI NP-1413, 1980. [100] W.J. Shack. Measurement of Through wall Residual Stresses in Large Diameter Type 304 Stainless Steel Piping Butt Weldments, ANL-82-15, Argonne National Laboratory, 1984. [101] M. Hayashi, M. Ishiwata, N. Minakawa and S. Funahashi. Residual-Stress Measurement in Socket Welded Joints by Neutron Diffraction. Proceedings of the 12th International Conference on NDE in the Nuclear and Pressure Vessel Industries, pp.449-454, 1993. [102] Kandil, F.A., Lord, J.D., Fry, A.T. and Grant. P.V., A Review of Residual Stress Measurement Methods - A Guide to Technique Selection, NPL Report MATC(A)04, Project CPM4.5, Measurement of Residual Stresses in Components, NPL Material Center, UK, 2001. [103] G. Focht and K. Schiffner. Determination of Residual Stresses by an Optical Correlative Hole-Drilling Method. Journal of Experimental Mechanics; Vol. 43(1), pp. 97-104, 2003. [104] D. Stefanescu1, C.E. Truman, D.J. Smith and P.S. Whitehead. Improvements in Residual Stress Measurement by the Incremental Centre Hole-Drilling Technique. Journal of Experimental Mechanics; Vol. 46(4), pp. 417-427, 2006. [105] N.J. Rendler and I. Vigness. Hole-Drilling Strain-Gage Method of Measuring Residual Stresses. Journal of Experimental Mechanics; Vol. 6(12): pp. 577-586, 1966. [106] ASTM E-837-94a: Standard Test Method for Determining Residual Stresses by the Hole Drilling Strain -Gage Method, American Society for Testing and Materials, Philadelphia, 1994.

257

University of Engineering & Technology, Taxila-Pakistan

[107] Flaman, M.T., Investigation of Ultra-High Speed Drilling for Residual Stress Measurement by the Center Hole Method, Experimental Mechanics, vol. 22, No. I, pp. 26-30, 1982. [108] Anon: Measurement of Residual Stress by the Hole Drilling Strain Gauge Method, TN-503, Vishay-Measurements Group, Inc. Raleigh, 1993. [109] Schajer, G.S., Application of Finite Element Calculations to Residual Stress Measurements, ASME J. Engineering Material and Technology, vol. 103, pp. 157-163, 1981. [110] Schajer, O.S., Measurement of Non-Uniform Residual Stresses Using the Hole Drilling Method. Part-11 - Practical Implementation of the Integral Method, ASME J. Engineering Material and Technology, vol. 110, pp. 344-349, 1988. [111] Wern, H., Measurement of Non-Uniform Stresses using the Hole Drilling Method, a New Integral Formalism, Strain, vol. 31, No. 2, pp. 63-68, 1995. [112] Wern, H., A New Approach to Triaxial Residual Stress Evaluation by the Hole Drilling Method, Strain, vol. 33, No. 4, pp. 121-125, 1997. [113] Wern, H., Cavelius, R. and Schlafer, D., A New Method to Determine Triaxial NonUniform Residual Stresses from Measurements Using the Hole Drilling Method, Strain, vol. 33, no. 2, pp. 39-45, 1997. [114] Zuccarello, B., Optimal Calculation Steps for the Evaluation of Residual Stresses by the Incremental Hole Drilling Method, Experimental Mechanics, vol. 39, No. 2, pp. 117-127, 1999. [115] Aoh, J.N. and Wei, C.S., On the Improvement of Calibration Coefficients for HoleDrilling Integral method: Part I - Analysis of Calibration Coefficients Obtained by a 3-D FEM Model, ASME J. Engineering Material and Technology, vol. 124, pp. 250258, 2002. [116] Aoh, J.N. and Wei, C.S., On the Improvement of Calibration Coefficients for HoleDrilling Integral method: Part II-Experimental Validation of Calibration Coefficients, ASME J. Engineering Material and Technology, vol. 125, pp. 107-115, 2003. [117] Nils J. Nilsson; Artificial Intelligence: A New Synthesis; Morgan Kaufmann Publishers, USA, pp 1-2, 1998. [118] Avelino J. Gonzalez, Douglas D. Dankel; The Engineering of Knowledge-Based Systems: Theory and Practice; Prentice Hall, New Jersey, pp 16-22, 1993. [119] M. Ganesh; Introduction to Fuzzy Sets and Fuzzy Logic; Prentice Hall, pp 147-174, 2006. [120] ASM Handbook (Welding, Brazing and Soldering). Vol. 6. pp.1025, pp.1059. [121] T.W. Liao, Classification of welding flaw types with fuzzy expert systems, Expert Systems with Applications, Vol. 25, pp. 101–111, 2003. [122] Andrew Taylor, An investigation of knowledge-based models for automated procedure generation in the arc welding domain, International Journal of Production Research, Vol. 27, No. 11, pp. 1855 – 1862, 1989. [123] Aparna S. Vardo, Mohammad Maniruzzaman, Elke A. Rundensteiner and Richard D. Sisson Jr, “The QuenchMinerTM Expert System for Quenching and Distortion Control”, Worcester Polytechnic Institute (WPI), Worcester, MA 01609, USA. 258

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

[124] Dongcheol Kim, Sehun Rhee and Hyunsung Park. Modelling and optimization of a GMA welding process by genetic algorithm and response surface methodology. Int. Journal of Production Research. Vol. 40, pp. 1699 – 1711, 2004. [125] S. Kumanan, R. Ashok Kumar and J. Edwin Raja Dhas. Development of a welding residual stress predictor using a function-replacing hybrid system. Journal of Advanced Manufacturing Technology. Vol-31, 2007. [126] H. L. Lin and C. P. Chou, Optimisation of the GTA welding process using the Taguchi method and a neural network, Institute of Materials, Minerals and Mining, Science and Technology of Welding and Joining, Vol. 11, No. 1, pp.120, 2006. [127] J.P. Ganjigatti, Dilip Kumar Pratihar; Forward and reverse modeling in MIG welding process using fuzzy logic-based approaches, Journal of Intelligent and Fuzzy Systems, Volume 19, Number 2, pp. 115-130, 2008. [128] V. Tsoukalas, M.Kontesis, E.Badogiannis, D.Papachristos, N.Fragiadakis; Prototype of an expert system for aluminum welding; Proceedings of the 5th WSEAS International Conference on Computational Intelligence, Man-Machine Systems and Cybernetics; Venice, Italy, pp.78-83, 2006. ISBN ~ ISSN:1790-5117, 960-8457-56-4. [129] Fang, X.D.: Expert System Supported Fuzzy Diagnosis of Finish Turning Process States; International Journal of Machine Tools & Manufacture, Vol. 35 (6), pp.913924, 1995. [130] Chen, Y.; Hui, A.; Du, R.: A Fuzzy Expert System for the Design of Machining Operations; International Journal of Machine Tools & Manufacture, Vol. 35 (12), pp.1605-1621, 1995. [131] Hashmi, K.; El-Baradie, M.A.; Ryan, M.: Fuzzy Logic Based Intelligent Selection of Machining Parameters; Computers in Industrial Engineering, Vol. 35 (3-4), pp.571574, 1998. [132] Arezoo, B.; Ridgway, K.; Al-Ahmari, A.M.A.: Selection of Cutting Tools and Conditions of Machining Operations Using an Expert System; Computers in Industry, Vol. 42, pp.43-58, 2000. [133] Lo, S.P.: An Adaptive-Network Based Fuzzy Inference System for Prediction of Workpiece Surface Roughness in End Milling; Journal of Materials Processing Technology; Vol. 142, pp.665-675, 2003. [134] J.N. Aoh and C.S. Wei. On the improvements of Calibration Coefficients for HoleDrilling Integral Method: Part-I Analysis of Calibration Coefficients obtained by 3-D FEM Method. ASME Journal of Engineering Materials and Technology; Vol. 124(2): pp. 250-258, 2002. [135] J.N. Aoh and C.S. Wei. On the improvements of Calibration Coefficients for HoleDrilling Integral Method: Part-II Experimental Validation of Calibration Coefficients. ASME Journal of Engineering Materials and Technology; Vol. 125(2): pp. 107-115, 2003. [136] G.S. Schajer. Measurement of Non-Uniform Residual Stresses using the HoleDrilling Method: Part-1 Stress Calculation Procedures. ASME Journal of Engineering Materials and Technology; Vol. 110, pp.338-343, 1998. [137] M.C. Sun, Y.H. Sun, and R.K.Wang, Vibratory Stress Releiving of welded sheet steels of low alloy high strength steel, Materials Letters 58(2004), pp. 1396-1399. 259

University of Engineering & Technology, Taxila-Pakistan

[138] http://www.cnki.net, China Academic Journal Electronic Publishing House, 19942006. [139] Afzaal. M. Malik, Ejaz M. Qureshi, Naeem Ullah Dar and Iqbal Khan, TIG Welding Process: Experimental Validation of Simulated Results, Proceeding of the Int. Conference on Advanced Design and Manufacture, 8-10 Jan, 2006, Harbin, China. [140] ASM Handbook (Welding, Brazing and Soldering). Vol-6. pp-662. [141] S.A. Tsirkas, P. Papanikos and T. Kermanidis. Numerical Simulation of the Laser Welding Process in Butt-Joint Specimens, Journal of Material Processing Technology; Vol. 134(1): pp. 59-69, 2003. [142] V. Kamala, J. Goldak. Error due to Two-Dimensional Approximations in Heat Transfer Analysis of Welds. Welding Journal. vol. 72(9), pp. 440-446, 1993. [143] Welding Handbook Volume 2-Part 1: Welding Processes. American Welding Society 2004, ISBN: 0871717298. [144] R.W. Messler. Principles of Welding: Processes, Physics, Chemistry and Metallurgy. Wiley Interscience, 1st Edition, 1999. [145] N.T. Nguyen, T. A. Ohta, K. Matsuoka, N. Suzuki and Y. Maeda. Analytical Solutions for Transient Temperature of Semi-Infinite Body Subjected to 3-D Moving Heat Sources. Welding Journal; pp. 265-274, 1999. [146] N. Christensen, V.L. Davies and K. Gjermundsen - Distribution of Temperatures in Arc Welding - British Welding Journal, Vol. 12, pp.54-75, 1965. [147] Lin, M.L., Eager, T.W. Influence of Arc Pressure on Weld Pool Geometry, Weld. J. Vol. 64(6), pp. 163-169. [148] Little, G.H., Kamtekar, A.G. The Effect of Thermal Properties and Welding Efficiency on Transient Temperatures during Welding. Computer and Structures, Vol. 68, pp. 157–165, 1998. [149] E.M. Vander. Finite Element Modeling of Temperature Profiles, Distortions and Residual Stresses due to TIG Welding - M. Sc Thesis, Delft University of Technology, 2003. [150] D. Gery, H. Long, and P. Maropoulos. Effects of Welding Speed, Energy Input and Heat Source Distribution on Temperature Variations in Butt Joint Welding. Journal of Materials Processing Technology; Vol. 167(2-3): pp. 393-401, 2005. [151] Kreith, F.; Boehm, R.F. et al. Heat and Mass Transfer" Mechanical Engineering Handbook. CRC Press LLC, 1999. [152] Y.A. Cengel, Heat Transfer: A Practical Approach, WCB/McGraw-Hill, Tokyo, 1998. [153] D.R. Poirier and G.H. Geiger - Transport Phenomena in Materials Processing - The Minerals Metals and Materials Society, 1994. [154] http://ib.cnea.gov.ar/~experim2/Cosas/omega/emisivity.htm [155] http://www.omega.com/literature/transactions/volume1/emissivitya.html [156] M. Abid and M. Siddique. Numerical Simulation of the Effect of Constraints on Welding Deformations and Residual Stresses in a Pipe–Flange Joint. Modeling and Simulation in Materials Science and Engineering; Vol. 13: pp. 919-933, 2005. 260

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

[157] E.M. Vander. Local Cooling during Welding: Prediction and Control of Residual Stresses and Buckling Distortion, Delft University of Technology, 2007. [158] L. Fenggui, Y. Shun, L. Songnian and L. Yongbin. Modeling and Finite Element Analysis GTAW Arc and Weld Pool. Journal of Computational Materials Science; Vol. 29, pp. 371-378, 2004. [159] C.K. Lee, J. Candy and C.P.H. Tan. Measurement and Finite Element Analysis of Temperature Distribution in Arc Welding Process. International Journal of Computer Applications in Technology; Vol. 21(4): pp. 171-177, 2004. [160] I.F.Z. Fanous, M.Y.A. Younan and A.S. Wifi. Study of the Effect of Boundary Conditions on Residual Stresses in Welding using Element Birth and Element Movement Techniques. ASME Journal of Pressure Vessel Technology; Vol. 125, pp. 432–439, 2003. [161] K.T. Chaouch, H. Maza and B. Belkessa. A Finite Volume Analysis of TwoDimensional Transient Heat Transfer in TIG Welding. Journal of Physics IV; Vol. 120, pp. 669-675, 2004. [162] ASM Metals Handbook Volume 2-Properties and Selection Nonferrous Alloys and Special Purpose Materials, ASM International, ISBN 0-87170-378-5 (v. 2). [163] E. A. Brandes, and G. B. Brook., Smithels Metals Reference Book,. Handbook, Seventh Edition, Butterworth-Heinneman, 1992. [164] Materials Handbook (5th edition), George S. Brady, Henry R. Clauser, John A. Vaccari, McGraw-hill Handbooks. [165] X.K. Zhu and Y.J. Chao. Effects of Temperature-Dependent Material Properties on Welding Simulations. Journal of Computers and Structures; Vol. 80(11), pp.967-976, 2002. [166] Xinhai Qi, Numerical Simulation of Buckling in Thin Panels, PhD Thesis, University of Carolina, 2003. [167] Andersson, B. A. B., Thermal Stresses in Submerged Arc Welded Joint Considering Phase Transformations, ASME Journal of Engineering Materials and Technology, vol. 100, pp. 356 – 362, 1978. [168] Mahin, K. W., Winters, W., Holden, T. M., Hosbons, R. R., and MacEwen, S. R., Prediction and Measurement of Residual Elastic Strain Distributions in Gas Tungsten Arc Weld, Welding Journal, 70(9), pp. 245s-260s, 1991. [169] Pardo, E., and Weckman, D.C., Prediction of Weld Pool and Reinforcement Dimensions of GMA Weld using a Finite Element Model, Metallurgical Transactions B, 20B(12), pp. 937 – 947, 1989. [170] Ule, R. L., Joshy, Y., and Sedy, E. B., A New Technique for Three Dimensional Transient Heat Transfer Computations of Autogenous Arc Welding, Metallurgical Transactions B, 21B(12), pp. 1033 – 1047, 1990. [171] Lindgren; L. E. Runnemalm, H. and Nässtrom M. O., Simulation of Multipass Welding of Thick Plate, Int. J. Numerical Methods in Engineering, vol. 44, pp. 1301 – 1316, 1999. [172] Lu. Xiangyang, Influence of Residual Stress on Fatigue Failure of Welded Joints, PhD Thesis, North Carolina State University, 2002. 261

University of Engineering & Technology, Taxila-Pakistan

[173] Brickstad, B., and Josefson, B.L. (1998), “A Parametric Study of Residual Stresses in Multipass Butt-welded Stainless Steel Pipes”, International Journal of Pressure Vessels and Piping, Vol.75, pp.11-25. [174] ANSYS® 10.0 user's documentation, SAS IP Inc, Canonburg, PA 15317, USA, 2005. [175] H. Yu, L. Shouju and L. Yingxi. Finite Element Simulation for Residual Stresses in Welding Process. Key Engineering Materials; 353-358, pp. 1915-1918, 2007. [176] M. Afzaal Malik, M. Ejaz Qureshi, Naeem Ullah Dar, Iqbal Khan; Analysis of Circumferentially Arc Welded Thin-Walled Cylinders to Investigate the Residual Stress Fields; Elsevier's International Journal of Thin Walled Structures, Vol. 46 (2008), pp. 1391-1401, 2008. doi:10.1016/j.tws.2008.03.011. [177] Z. Dong, Y. Wei and X. Yanli. Predicting Weld Solidification Cracks in Multipass Welds of SUS310 Stainless Steel. Journal of Computational Materials Science; Vol. 38(2), pp. 459-466, 2006. [178] M.M. Mahapatra, G.L. Datta, B. Pradhan, and N.R. Mandal. Three-Dimensional Finite Element Analysis to Predict the Effects of SAW Process Parameters on Temperature Distribution and Angular Distortions in Single-Pass Butt Joints with Top and Bottom Reinforcements. International Journal of Pressure Vessels and Piping; Vol. 83(10), pp. 721-729, 2006. [179] L.E. Lindgren. Computational Weld Mechanics: Thermo-mechanical and Microstructural Simulations. Woodhead Publishing, 2007. [180] N U Dar, E M Qureshi, A M Malik, M M I Hammouda, R A Azeem; Analysis of Circumferentially Welded Thin-Walled Cylinders to Study the Effects of Tack Weld Orientations and Joint Root Opening on Residual Stress Fields; Proc. ImechE, Part C: Journal of Mechanical Engineering Science, Vol. 223, Number 5/2009, pp 10371047, DOI: 10.1243/09544062JMES1191. [181] C.C. Silva and J.P. Farias. Non-uniformity of Residual Stress Profiles in Butt Welded Pipes in Manual Arc Welding. Journal of Materials Processing Technology; Vol. 199(6): pp. 452–455, 2008. [182] Hibbitt HD, Marcal PV. A Numerical Thermo-mechanical Model for the Welding and Subsequent Loading of a Fabricated Structure. Journal of Computers and Structures; Vol. 3, pp. 1153–1174, 1973. [183] Jonsson M, Karlsson L. and Lindgren LE. Simulation of Tack Welding Procedures in Deformations in a Welded Thin-Walled Pipe. Journal of Pressure Vessel Technology; Vol. 112(2), pp.76-84, 1990. [184] A. Yaghi, T.H. Hyde, A.A. Becker, W. Sun and J.A. Williams, Residual stress simulation in thin and thick-walled stainless steel pipe welds including pipe diameter effects, International Journal of Pressure Vessels and Piping. Vol. 83 (11-12), pp.864874, 2006. [185] D. Deng and H. Murakawa, Numerical simulation of temperature field and residual stress in multi-pass welds in stainless steel pipe and comparison with experimental measurements, Computational Material Science. Vol. 37 (3), pp. 269-277, 2006.

262

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

[186] E.F. Rybicki, D.W. Schmueser, R.W. Stonesifer, J.J. Groom and H.W. Mishaler, A Finite Element model for residual stresses and deflections in girth-butt welded pipes, ASME Journal of Pressure Vessel Technology. 100, pp. 256–262, 1978. [187] E.F. Rybicki, P.A. McGuire, E. Merrick and J. Wert, The effect of Pipe thickness on residual stresses due to girth welds. ASME Journal of Pressure Vessel Technology. 104, pp. 204–209, 1982. [188] E.F. Rybicki and R.B. Stonesifer, Computation of residual stresses due to multi-pass welds in piping system. ASME Journal of Pressure Vessel Technology. 101, pp. 49– 54, 1979. [189] Y. Dong, J. Hong, C. Tsai and P. Dong. Finite Element modeling of residual stresses in austenitic stainless steel pipe girth welds, Welding Journal, Weld Research Supplement. 442, pp.449–444, 1997. [190] R.I. Karlsson and B.L. Josefson, Three-dimensional Finite Element analysis of temperature and stresses in single-pass butt-welded pipe. ASME Journal of Pressure Vessel Technology. Vol. 112, pp.76–84, 1990. [191] P.H. Chang and T.L. Teng, Numerical and experimental investigation on the residual stresses of the butt-welded joints, Journal of Computational Material Science. Vol. 29 (4), pp.511-522, 2004. [192] M. Jonsson and B.L. Josefson, Experimentally determined transient and residual stresses in the butt-welded pipes, Journal of Strain Analysis. Vol.23 (1), pp.25-31, 1998. [193] L. Karlsson, M. Jonsson, L.E. Lindgren, M. Näsström and L. Troive, Residual stresses and deformations in a welded thin-walled pipe, Proc. ASME Pressure Vessel and Piping Conf. (Hawaii, July 1989) PVP-173, pp.7–11, 1989. [194] Dong P. and Burst BW. Welding Residual Stresses and Effects on Fracture in Pressure Vessel and Piping Components: A Millennium Review and Beyond. Journal of Pressure Vessel Technology; Vol. 122(3), pp.329-338, 2000. [195] Dong P. Residual Stress Analyses of a Multi-Pass Girth Weld: 3-D Special Shell versus Axis-symmetric Models. Journal of Pressure Vessel Technology; Vol. 123(2), pp.207-213, 2001. [196] S. Fricke, E. Keim and J. Schmidt, Numerical weld modeling - a method for calculating weld-induced residual stresses, Nuclear Engineering and Design. Vol. 206 (2-3), pp.139-150, 2001. [197] Naeem Ullah Dar, Ejaz M. Qureshi, M.M.I Hammouda; Analysis of Weld Induced Residual Stresses and Distortions in Thin Walled Cylinders; Journal of Mechanical Science and Technology, KSME, Vol. 23, Number 4/April, 2009, pp. 1118-1131, DOI: 10.1007/s12206-008-1012-6. [198] ANSYS-10.0 user manual. [199] Vishay Group, Measurement of residual stresses by the hole drilling strain gage method, Technical Note No. TN-503. (www.vishay.com/brands/) [200] R. Spina, L. Tricarico, G. Basile and T. Sibillano. Residual Thermo-mechanical Modeling of Laser Welding of AA5083 Sheets. Journal of Materials Processing Technology; Vol. 191(1-3), pp.215-219, 2007. 263

University of Engineering & Technology, Taxila-Pakistan

[201] X. Y. Shan, M. J. Tan and N.P. O'Dowd. Developing a Realistic FE Analysis Method for the Welding of a NET Single-Bead-on-Plate Test Specimen. Journal of Materials Processing Technology; Vol.192-193(1-3): pp.497-503, 2007. [202] A. Kermanpur, M. Shamanian and V.E. Yeganeh. Three-Dimensional Thermal Simulation and Experimental Investigation of GTAW Circumferentially Butt-Welded Incoloy 800 Pipes. Journal of Materials Processing Technology 2007; doi:10.1016/j.jmatprotec.2007.08.009. [203] D. Gery, H. Long and P. Maropoulos. Effects of Welding Speed, Energy Input and Heat Source Distribution on Temperature Variations in Butt Joint Welding. Journal of Material Processing Technology; Vol. 167(2-3), pp.393-401, 2005. [204] P. Groombridge and A. Oloyede. A Control System for Microwave Processing of Materials. ASME Journal of Manufacturing Science and Engineering; Vol. 122(1), pp.253-261, 2000. [205] F. Bardin, R. McBride, A. Moore, S. Morgan, S. Williams, J.D.C Jones and D.P. Hand. Real-Time Temperature Measurement for Process Monitoring of Laser Conduction Welding. Proceedings of 23rd International Congress on Applications of Lasers and Electro-Optics, 2004; http://www.aop.hw.ac.uk/Projects/Papers/LFRef11.pdf. [206] C.S. Wu, J. Chen and Y. M. Zhang. Numerical Analysis of Both Front and Back-Side Deformation of Fully Penetrated GTAW Weld Pool Surfaces. Journal of Computational Materials Science; Vol. 39(3), pp.635-642, 2007. [207] M.A. Quintana, T. DebRoy, J.M. Vitek and S. Babu. Novel Optimization Methodology for Welding Process/Consumable Integration. Final Technical Report #: DOE-ID14204-F1, United States Department of Energy, Scientific and Technical Information,2005-2006; http://www.osti.gov/bridge/servlets/purl/862404DMY9RJ/862404.pdf. [208] H.F. Junejo. Finite Element Modeling of Weld Based Rapid Prototyping. PhD Thesis, Ghulam Ishaq Khan Institute of Engineering Sciences and Technology, 2006. [209] X.K. Zhu and Y.J. Chao. Effects of Temperature-Dependent Material Properties on Welding Simulations. Journal of Computers and Structures; Vol. 80(11), pp.967-976, 2002. [210] A.N. Lari, J. Lu and J.F. Flavenot. Measurement of Residual Stress Distribution by the Incremental Hole-Drilling Method. Journal of Mechanical Working Technology; Vol. 11(2): pp.167-188, 1985. [211] Measurement of Residual Stresses by the Hole-Drilling Strain Gage Method, Application Note TN-503-6, Document No. 11053, Revision January 2005, Vishay Micro-Measurement Group. [212] Determining Residual Stresses by the Hole-Drilling Strain Gage Method, ASTM Standard E837. [213] J.N. Aoh and C.S. Wei. On the improvements of Calibration Coefficients for HoleDrilling Integral Method: Part-I Analysis of Calibration Coefficients obtained by 3-D FEM Method. ASME Journal of Engineering Materials and Technology; Vol.124(2), pp.250-258, 2002.

264

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

[214] J.N. Aoh and C.S. Wei. On the improvements of Calibration Coefficients for HoleDrilling Integral Method: Part-II Experimental Validation of Calibration Coefficients. ASME Journal of Engineering Materials and Technology; Vol. 125(2): pp.107-115, 2003. [215] G.S. Schajer. Measurement of Non-Uniform Residual Stresses using the HoleDrilling Method: Part-1 Stress Calculation Procedures. ASME Journal of Engineering Materials and Technology; 110, pp.338-343, 1998. [216] Gunaraj, V., and Muragan, M., Prediction of Heat Affected Zone Characteristic in Submerged Arc Welding of Structural Steel Pipes, Welding Research Supplement, pp.94-102, 2002. [217] Jonsson, M., Karlsson, L., and Lindgren, L.E., Simulation of Tack Welding Procedure in Butt Joint Welding of Plates, Welding Research Supplement, pp.296301, 1995. [218] Jonsson, M., Karlsson, L., and Lindgren, L.E., Plate Motion and Thermal Stresses in Root-Bead Butt-Welding of Plates, 4th International Conference on Numerical Methods in Thermal Problems (eds. R W Lewis and K Morgan), Swansea, July 1985, Pineridge Press, Swansea. [219] Shibahara, M., Serizawa, H., and Murakawa, H., Finite Element Analysis using Interface Elements for Predicting Deformation during Butt-Welding considering Root Gap and Tack Welds, Transactions of JWRI, 31, No. 1, pp.63-70, 2002. [220] Deng, D., Murakawa, H., and Ueda, Y., Theoretical Prediction of Welding Distortion Considering Positioning and Gap between Parts, The Proceedings of the 12th International Offshore and Polar Engineering Conference, 2002. [221] Jang, G.B., Kim, H.K., and Kang, S.S., The Effects of Root Opening on Mechanical Properties, Deformations and Residual Stresses of Weldments, Welding Research Supplement, pp.80-88, 2001. [222] Tso-Liang, Peng-Hsiang Chang; Three Dimensional Thermo-mechanical Analysis of Circumferentially Welded Thin Walled Pipes. International Journal of Pressure Vessels and Piping, Vol.75, pp.237-247, 1998. [223] Xiangyang, L., Influence of Residual Stress on Fatigue Failure of Welded Joints, PhD Thesis, North Carolina State University, 2002. [224] Allen, J., Distortion caused by welding a plate at the centre of a thin plate before welding into a bridge girder section, http://adhesivetoolkit.com/j32k/protected /band3/ jk36.html. [225] Thomas Pyzdek; Virtual DOE, Data Mining and Artificial Neural Networks; Quality Digest, July 1999, pp.26. [226] Thomas Pyzdek, Virtual DOE using simulation software, Six Sigma Hand Book, McGraw-Hill, pp.699. [227] Martin Tanco, Elisabeth Viles, Lourdes Pozueta, Are All of Experiments Approaches Suitable for Your Company?, World Congress on Engineering (WCE-2008), pp.1202-1207. [228] Pham, D.T.; Pham, P.T.N.: Computational Intelligence for Manufacturing, in: Computational Intelligence in Manufacturing Handbook; CRC Press LLC, Florida, 2001. 265

University of Engineering & Technology, Taxila-Pakistan

[229] Orchard, R.A.: Fuzzy CLIPS, Version 6.04A; User’s Guide, National Research Council, Canada, 1998 [230] http://www.iit.nrc.ca/IR_public/fuzzy/fuzzyJDocs/APIdocs/nrc/fuzzy/FuzzyValue.html [231] Hashmi, K.; Graham, I.D.; Mills, B.; Hashmi, M.S.J.: Adjustment Approach for Fuzzy Logic Model Based Selection of Non-Overlapping Machining Data in Turning Operation; Journal of Materials Processing Technology, Vol. 142, pp.152-162, 2003. [232] Laarhoven, P.J.M.; Arts, E.H.L.: Simulated Annealing: Theory and Applications; Kluwer Academic Publishers, Dordrecht, Netherlands, 1987. [233] Asif Iqbal, Ning He, Liang Li, Naeem Ullah Dar; Simulated Annealing Assisted Optimization of Fuzzy Rules For Maximizing Tool Life in High-Speed Milling Process; Proceedings of 5th IASTED International Conference on Artificial Intelligence & Applications, Innsbruck, Austria, February 13-16, 2006, pp. 335-340. [234] Lekova, A.; Batanov, D.: Self-Testing and Self-Learning Fuzzy Expert System for Technological Process Control; Computers in Industry, Vol. 37, pp.135-141, 1998. [235] Castro, J.L.; Castro-Schez, J.J.; Zurita, J.M.: Use of a Fuzzy Machine Learning Technique in Knowledge-Acquisition Process; Fuzzy Sets and Systems, Vol. 123, pp.307-320, 2001. [236] Filipic, B.; Junkar, M.: Using Inductive Machine Learning to Support Decision Making in Machining Processes; Computers in Industry, Vol. 43, pp.31-41, 2000. [237] Monostori, L.: AI and Machine Learning Techniques for Managing Complexity, Changes, and Uncertainty in Manufacturing; Engineering Applications of Artificial Intelligence, Vol. 16, pp.277-291, 2003 [238] Al Assadi, H.M.A.A.; Wong, S.V.; Hamouda, A.M.S.; Ahmad, M.M.M.H.: Development of Machine Learning Strategy for Acquiring On-Line Machining Skills during Turning Process; Journal of Materials Processing Technology, Vol. 155-156, pp.2087-2092, 2004. [239] Cho, S.; Asfour, S.; Onar, A.; Kaundinya, N.: Tool Breakage Detection using Support Vector Machine Learning in a Milling Process; International Journal of Machine Tools & Manufacture, Vol. 45, pp.241-249, 2005. [240] Asif Iqbal, Naeem U. Dar, Ning He, Muhammad M.I Hammouda, Liang Li; SelfDeveloping Fuzzy Expert System: A Novel Learning Approach, Fitting for Manufacturing Domain; Journal of Intelligent Manufacturing (JIMS), Manuscript Number: JIMS-232R2, DOI: 10.1007/s10845-009-0252-3. [241] J. Liebowitz; Knowledge management and its link to artificial intelligence, Expert Systems with Applications, Vol-20, Issue 1, pp 1-6, Jan 2001.

266

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

List of Publications Journals: 1. Naeem Ullah Dar, Ejaz M. Qureshi, M.M.I Hammouda; Analysis of Weld Induced Residual Stresses and Distortions in Thin Walled Cylinders; Journal of Mechanical Science and Technology, KSME, Vol. 23, Number 4/April, 2009, pp. 1118-1131, DOI: 10.1007/s12206-008-1012-6. 2. N U Dar, E M Qureshi, A M Malik, M M I Hammouda, R A Azeem; Analysis of Circumferentially Welded Thin-Walled Cylinders to Study the Effects of Tack Weld Orientations and Joint Root Opening on Residual Stress Fields; Proc. ImechE, Part C: Journal of Mechanical Engineering Science, Vol. 223, Number 5/2009, pp 10371047, DOI: 10.1243/09544062JMES1191. 3. Asif Iqbal, Naeem U. Dar, Ning He, Muhammad M.I Hammouda, Liang Li; SelfDeveloping Fuzzy Expert System: A Novel Learning Approach, Fitting for Manufacturing Domain; Journal of Intelligent Manufacturing (JIMS), Manuscript Number: JIMS-232R2 (online: 05 March, 2009), DOI: 10.1007/s10845-009-0252-3. 4. E.M. Qureshi, A.M. Malik , N.U. Dar; Residual Stress Fields due to varying Tack Welds Orientation in Circumferrentially Welded Thin-Walled Cylinders; Advances in Mechanical Engineering, Vol. 2009, Article ID351369, 9 pages, 2009. DOI: 10.1155/2009/351369. 5. E.M.Qureshi, A.M. Malik, R.A. Azeem, N.U. Dar, M. Hussain; Analysis of Residual Stress Fields due to Different Tack Welds Orientation in Circumferentially Welded Thin Walled Cylinders; Journal of Welding in the World, IIW (2008) Issue: Special, Vol. 52 (2008), pp. 571-578. 6. M. Ejaz Qureshi, M. Afzaal Malik , Naeem Ullah Dar; Analysis of Arc Welded Thin-Walled Cylinders to Investigate the Effects of Welding Process Parameters on Residual Stresses; Material Science Forum, Vol. 575-578 (2008), pp. 763-768. 7. M. Ejaz Qureshi, M. Afzaal Malik, Naeem Ullah Dar, Iqbal Khan; Analysis of Circumferentially Welded Thin-Walled Cylinders to Investigate the Effects of Varying Clamping Conditions; Journal of Engineering Manufacture, Part-B, Vol. 222 (2008), pp. 901-914. 8. M. Afzaal Malik, M. Ejaz Qureshi, Naeem Ullah Dar, Iqbal Khan; Analysis of Circumferentially Arc Welded Thin-Walled Cylinders to Investigate the Residual Stress Fields; Elsevier's International Journal of Thin Walled Structures (2008), Vol. 46 (2008), pp. 1391-1401. doi:10.1016/j.tws.2008.03.011. 9. A Iqbal, N U Dar, N He, I Khan, L Li; Optimizing Cutting Parameters in Minimum Quantity Lubrication Milling of Hardened Cold Work Tool Steel; Proc. IMechE Vol. 223 Part B: J. Engineering Manufacture, JEM1231 _ IMechE 2009, pp 43-54. DOI: 10.1243/09544054JEM1231. 10. Asif Iqbal, Ning He, Iqbal Khan, Liang Li, Naeem Ullah Dar; Modeling the Effects of Cutting Parameters in MQL-Employed Finish Hard Milling Process Using DOptimal Method; Journal of Materials Processing Technology, vol. 199, Issues 1-3, April 2008, pp. 379-390. 267

University of Engineering & Technology, Taxila-Pakistan

11. G. Hussain, L. Gao, N. Hayat, N.U. Dar; The formability of annealed and pre-aged AA-2024 sheets in single-point incremental forming; The Int. Journal of Advanced Manufacturing Technology, Online published: 15 August 2009, DOI 0.1007/s00170009-2120-x. 12. Hussain G, Gao L, Hayat N., Cui, Z., Pang, Y.C., N.U. Dar; Tool and lubrication for Negative Incremental Forming of a commercially pure-Titanium sheet; Journal of Materials Processing Technology, vol. 203, Issues 1-3, 18 July 2008, pp 193-201. 13. Hussain G, N.U. Dar, Gao L, Chen M.H; A Comparative Study on the Forming Limits of the Aluminum Sheet in Negative Incremental Forming. Journal of Materials Processing Technology, vol. 187-188, 12 June 2007, pp. 94-98. 14. Hussain G, Gao L, N.U. Dar; An Experimental Study on some Formability Evaluation Methods in Negative Incremental Forming. Journal of Materials Processing Technology, vol. 186, Issues 1-3, 7 May 2007, pp. 45-53. 15. Asif Iqbal, Ning He, Naeem Ullah Dar, Liang Li; Comparison of Fuzzy Expert System Based Strategies of Offline & Online Estimation of Tool’s Flank Wear in Hard-Milling Process; Expert Systems with Applications, Vol. 33, Issue 1, July 2007, pp.61-66. 16. Asif Iqbal, Ning He, Liang Li, Naeem Ullah Dar; A Fuzzy Expert System for Optimizing Parameters and Predicting Performance Measures in Hard-Milling Process; Expert Systems with Applications, Vol. 32, Issue 4, May 2007, pp.10201027.

Edited Books: 17. Asif Iqbal, Iqbal Khan, Ning He, Naeem Ullah Dar, Liang Li; Application of Expert Systems in Manufacturing Domain; In: “Progress in Expert Systems Research”; Edited book; Nova Science Publishers, NY, USA, 2007, pp. 73-118. ISBN-13: 9781600216909. 18. Asif Iqbal, Naeem Ullah Dar; "A Self-Progressing Fuzzy Rule-Based System for Optimizing and Predicting Machining Process", Edited book "Advances in Electrical Engineering and Computational Science", Springer Netherlands Publisher, May 2009. Vol. 39, Chapter 37, pp 435-446. DOI: 10.1007/978-90-481-2311-7_37.

Conferences: 19. Naeem Ullah Dar, Ejaz M. Qureshi, Iqbal Khan, Afzaal M. Malik; Welding Quality and Cost: A Comprehensive Comparative Study. (Paper No. 473) Proceedings of ADM-2006 Conference on Design and Manufacture, Harbin, China, January 8-10, 2006. 20. Afzaal M. Malik, Ejaz M. Qureshi, Naeem Ullah Dar, Iqbal Khan; TIG Welding Process: Experimental Validation of Simulated Results. (Paper No. 472) Proceedings of ADM-2006 Conference on Design and Manufacture, Harbin, China, January 8-10, 2006. 268

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

21. M. Afzaal Malik, M. Ejaz Qureshi, Naeem Ullah Dar, Iqbal Khan; Cost Estimation and Quality Comparison for Sound Welds. Proceedings of ASME-2005 International Mech. Engg. Congress and Exposition, Orlando, Florida-USA, Nov 5-11, 2005. 22. Asif Iqbal, Ning He, Liang Li, Naeem Ullah Dar; Simulated Annealing Assisted Optimization of Fuzzy Rules For Maximizing Tool Life in High-Speed Milling Process; Proceedings of 5th IASTED International Conference on Artificial Intelligence & Applications, Innsbruck, Austria, February 13-16, 2006, pp. 335-340. 23. M.Ejaz Qureshi, M. Afzaal Malik, Iqbal Khan, N.U. Dar; Fatigue in Engineering Structures: A three-fold analysis approach; Proceedings of 15th International Conference on Nuclear Engineering (ICONE-15), Nagoya, Japan, April 22-26, 2007. 24. M. Ejaz Qureshi, M. Afzaal Malik, Naeem Ullah Dar; Analysis of Circumferentially Welded Thin Walled Cylinders: Effects of Welding Process Parameters on Residual Stresses; Proceedings of 5th Int. Conference Physical & Numerical Simulation of Materials Processing (ICPNS-23-27 Oct 2007), Zhengzhou, China. 25. M Ejaz M. Qureshi, Afzaal M. Malik, Raja Amer Azeem, Naeem Ullah Dar; Transient non-linear Thermal Analysis of Arc Welding Process to study the effects of Heat Source and Process Parameters; Proc. of 5th Int. Conf. Physical & Numerical Simulation of Materials Processing (ICPNS-23-27 Oct 2007), Zhengzhou, China. 26. M. Afzaal Malik , M. Ejaz Qureshi, Naeem Ullah Dar; Numerical Simulation of Arc Welding Investigation of various Process and Heat Source Parameters; Proceedings of FEMS (22-23 October 2007) conference on Failure of Engineering Materials & Structures, UET Taxila, Pakistan. 27. Asif Iqbal, Naeem Ullah Dar, Iqbal Khan, He Ning; Assessment of Wear Failure Modes of Carbide Cutters in High-Speed Milling of Hardened Steels; Proceedings of FEMS (22-23 October 2007) conference on Failure of Engineering Materials & Structures, UET Taxila, Pakistan. 28. Asif Iqbal, Naeem Ullah Dar, Mubasher Chohan; Effects of Cutting Parameters in Erosion Failure of Ductile Materials in Abrasive Water Jet Machining; Proceedings of FEMS (22-23 October 2007) conference on Failure of Engineering Materials & Structures, UET Taxila, Pakistan. 29. G. Hussain, L. Gao, Wang Hui, N.U. Dar; A fundamental comparison on the formability of a commercially-pure titanium sheet-metal in the incremental and conventional forming processes; Accepted in 2007- ASME Manufacturing Science and Engineering Conference (MSEC2007-31138). 30. Farooq M., Wang Dao Bo, Dar N.U.; A New Approach to Implementation of an Open Architecture Controller for a PUMA Robot; Proceedings of International Conference on Intelligent Automation & Robotics (ICIAR’07). World Congress on Engineering & Computer Science 2007. San Francisco, USA. 24-26 Oct. 2007. 31. Asif Iqbal, Iqbal Khan, Naeem Ullah Dar, Ning He; A Self-Developing Fuzzy Expert System, Designed for Optimization of Machining Process; Proceedings of World Congress on Engineering (WCE 2008) UK, July 2-4, 2008, pp.1728-1733 (ICCIIS). 269

University of Engineering & Technology, Taxila-Pakistan

32. Asif Iqbal, Iqbal Khan, Naeem Ullah Dar, Ning He; Efficacy of Applying Expert System in Monitoring of Milling Process; Proceedings of World Congress on Engineering (WCE 2008) UK, July 2-4, 2008, pp. 1734-1738 (ICCIIS). 33. E.M.Qureshi, A.M. Malik, R.A. Azeem, N.U. Dar, M. Hussain; Analysis of Residual Stress Fields due to Different Tack Welds Orientation in Circumferentially Welded Thin Walled Cylinders; Proceedings of International Conference on Safety and Reliability of Welded Components in Energy and Processing Industry (IIW 2008), Graz Austria, July 10-11, 2008, pp. 571-578. 34. Farooq, M.; Wang, D.B.; Dar, N.U.; Adaptive sliding-mode hybrid force/position controller for flexible joint robot; Mechatronics and Automation, 2008. ICMA 2008. IEEE International Conference on 5-8 Aug. 2008, pp. 724 – 731. 35. A. Iqbal, N.U. Dar, I. Khan; Knowledge-Based Intelligent Diagnostics of Tool Wear States; Proceedings of International Bhurban Conference on Applied Sciences & Technology (IBCAST), Islamabad-Pakistan, January 19 – 22, 2009.

270

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

Appendix A1

APDL Code for Thermal/Structural Module /COM ******************** PARAMETRIC CYLINDER GENERATION ******************** /COM /COM PRESENT STUDY : THERMAL ANALYSIS WITH 300 MM OD /COM PARAMETRIC GEOMETRY : YES /COM PARAMETRIC HEAT SOURCE : YES /COM DIFFERENT BASE/WELD MATERIAL : NO /COM MATERIAL CHANGE FEATURE : YES /COM /COM ****************************************************************************** /COM /CONFIG, NRES, 10E+05 /FILENAME, CYLINDER WELD (THERMAL), 1 /TITLE, CYLINDER WELD (THERMAL ANALYSIS) /PREP7 ET, 1, 70, 0, 1, 0, 0 /COM 8-NODED 3D THERMAL SOLID ELEMENT ET, 2, 57, 0, 0, 0, 0, 0, 0 /COM THERMAL SHELL ELEMENT MPREAD, CYLINDERMAT, TXT /COM READ MATERIAL PROPERTIES FROM FILE /VIEW, 1, 1, 1, 1 /COM CHANGE VIEW SETTING /ANG, 1 /COM SET VIEW ORIENTATION /COM /COM ************************ GEOMETRIC PARAMETERS ************************ /COM CWT=0.003 /COM CYLINDER WALL THICKNESS COR=0.150 /COM CYLINDER OUTER RADIUS CIR=COR-CWT /COM CYLINDER INNER RADIUS L=0.15 /COM CYLINDER LENGTH WZ1=0.008 /COM FIRST ZONE WIDTH WZ2=0.012 /COM SECOND ZONE WIDTH WZ3=0.030 /COM THIRD ZONE WIDTH WZ4=0.040 /COM FOUTH ZONE WIDTH WZ5=0.060 /COM FIFTH ZONE WIDTH /COM /COM ************************* MESHING PARAMETERS ************************* /COM NCWT=3 /COM NODES IN THICKNESS NZ1=4 /COM NODES IN ZONE-1 NZ2=6 /COM NODES IN ZONE-2 NZ3=6 /COM NODES IN ZONE-3 NZ4=5 /COM NODES IN ZONE-4 NZ5=6 /COM NODES IN ZONE-5 NCIR=80 /COM NODES ON CIRCUMFERENCE LEVEL=1 /COM REFINEMENT PARAMETER DEPTH=0 /COM REFINEMENT PARAMETER NFZR=2*NZ1 /COM WELDVEL=0.003 /COM WELDING SPEED

271

University of Engineering & Technology, Taxila-Pakistan

EFF=0.8 WELDI=200 WELDV=15 TOTHTLS=4*NCIR FACT_COR=0 /COM K,,0,0,0 K,,0,0,0.7 *DO,I,1,2,1 X=CIR+(I-1)*CWT K,,X,0,0 K,,X,0,WZ1 K,,X,0,-WZ1 K,,X,0,WZ1+WZ2 K,,X,0,-WZ1-WZ2 K,,X,0,WZ1+WZ2+WZ3 K,,X,0,-WZ1-WZ2-WZ3 K,,X,0,WZ1+WZ2+WZ3+WZ4 K,,X,0,-WZ1-WZ2-WZ3-WZ4 K,,X,0,WZ1+WZ2+WZ3+WZ4+WZ5 K,,X,0,-WZ1-WZ2-WZ3-WZ4-WZ5 *ENDDO /COM A,3,4,15,14 A,4,6,17,15 A,6,8,19,17 A,8,10,21,19 A,10,12,23,21 /COM A,3,5,16,14 A,5,7,18,16 A,7,9,20,18 A,9,11,22,20 A,11,13,24,22 /COM ASEL,ALL VROTATE,ALL,,,,,,1,2,360,2 ALLSEL,ALL /COM LSEL,S,LENGTH,,CWT LESIZE,ALL,,,NCWT LSEL,S,LENGTH,,WZ1 LESIZE,ALL,,,NZ1 LSEL,S,LENGTH,,WZ2 LESIZE,ALL,,,NZ2 LSEL,S,LENGTH,,WZ3 LESIZE,ALL,,,NZ3 /COM LSEL,S,LENGTH,,WZ4 LSEL,R,LOC,X,COR /COMLESIZE,ALL,,,NZ4,0.5 LESIZE,ALL,,,NZ4,1 LSEL,S,LENGTH,,WZ4

272

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

LSEL,R,LOC,X,CIR /COMLESIZE,ALL,,,NZ4,2 LESIZE,ALL,,,NZ4,1 /COM LSEL,S,LENGTH,,WZ5 LSEL,R,LOC,X,COR /COMLESIZE,ALL,,,NZ5,0.5 LESIZE,ALL,,,NZ5,1 LSEL,S,LENGTH,,WZ5 LSEL,R,LOC,X,CIR /COMLESIZE,ALL,,,NZ5,2,1 LESIZE,ALL,,,NZ5,1 /COM LSEL,S,RADIUS,,CIR LSEL,A,RADIUS,,COR LESIZE,ALL,,,NCIR /COM/COM /COM ******************************** MESHING ******************************** /COM ASEL,S,,,11,32,21 ASEL,A,,,13,34,21 ASEL,A,,,52,68,16 ASEL,A,,,54,70,16 /COM MSHKEY,1 AMESH,ALL ESEL,S,TYPE,,2 *GET,TNELEM,ELEM,0,COUNT CM,AREA2UNSEL,AREA ALLSEL,ALL ASEL,S,EXT ASEL,U,,,AREA2UNSEL AMESH,ALL ALLSEL,ALL EREFINE,1,TNELEM,1,LEVEL,DEPTH,SMOOTH,OFF /COM NZ1*NCIR*4 ACLEAR,AREA2UNSEL ASEL,S,,,AREA2UNSEL LSLA,ALL LSEL,U,LENGTH,,WZ1 LSEL,U,LENGTH,,THICK LESIZE,ALL,,,2*NCIR,,1 LSEL,S,LENGTH,,WZ1 LESIZE,ALL,,,NFZR,,1 ALLSEL,ALL AMESH,AREA2UNSEL /COM VSWEEP,1 VSWEEP,6 VSWEEP,11 VSWEEP,16 ESEL,S,TYPE,,1 *GET,TOTELEM,ELEM,0,COUNT /COM

273

University of Engineering & Technology, Taxila-Pakistan

*DO,I,2,5,1 VSWEEP,I VSWEEP,I+10 *ENDDO /COM *DO,II,7,10,1 VSWEEP,II VSWEEP,II+10 *ENDDO /COM ALLSEL,ALL ACLEAR,ALL NUMCMP,ALL SHPP,SUMMARY /COM *DIM,EZCENT,ARRAY,TOTELEM *DIM,EXCENT,ARRAY,TOTELEM *DIM,EYCENT,ARRAY,TOTELEM *DIM,MATREF,ARRAY,TOTELEM *DIM,PEAKTEMP,ARRAY,TOTELEM *DIM,INDICATE,ARRAY,TOTELEM *DIM,EPTIME,ARRAY,TOTELEM *DIM,ELESTATUS,ARRAY,TOTELEM *DIM,MATPCTIME,ARRAY,TOTELEM *DIM,ELEMVOL,ARRAY,TOTELEM *DIM,INPUT_HEAT,ARRAY,TOTHTLS+5 *DIM,ELEMENTNO,ARRAY,TOTHTLS+5 /COM *VGET,EZCENT,ELEM,1,CENT,Z,,2 *VGET,EXCENT,ELEM,1,CENT,X,,2 *VGET,EYCENT,ELEM,1,CENT,Y,,2 /COM /COM ************* HEAT SOURCE AND WELD PROCESS PARAMETERS ************* /COM TLHS=0.02 /COM TOTAL LENGTH OF HEAT SOURCE AF=0.005 /COM LENGTH OF FRONT ELLIPSOIDAL AR=TLHS-AF /COM LENGTH OF REAR ELLIPSOIDAL FR=0.75 /COM FRACTION OF HEAT IN FRONT ELLIPSOIDAL FF=2-FR /COM FRACTION OF HEAT IN REAR ELLIPSOIDAL B=0.005 /COM HALF WIDTH OF HEAT SOURCE C=CWT*1.4 /COM DEPTH OF HEAT SOURCE Q=EFF*WELDV*WELDI /COM NET HEAT INPUT /COM *DIM, CONVEC, TABLE, 49, 1, 1,,, /COM DEFINE TABLE TO STORE TEMP VS /COM CONVECTION *TREAD, CONVEC, HTOTAL, TXT, , 2 /COM READ CONVECTION (htotal) FROM FILE /COM /COM ************************* SOLUTION CONTROL **************************** /COM /SOLUTION /COM

274

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

ANTYPE,4,NEW /COMNLGEOM,ON TRNOPT,FULL NROPT,AUTO LUMPM,OFF LNSRCH,ON TIMINT,ON,THERM TINTP,,,,1,0.5 AUTOTS,ON KBC,1 NEQIT,125 OUTRESS,NSOL,LAST /COMOUTRESS,NSOL,LAST /COM ALLSEL,ALL /COM ESEL,S,,,1,1,1 /COM *DO,K,1,TOTELEM,1 /COM ZCENTER=1000*EZCENT(K) XCENTER=(COR-EXCENT(K))*1000 /COMD_VALUE=2*ZCENTER+XCENTER D_VALUE=2*ZCENTER+2*XCENTER D_VALUE1=-2*ZCENTER+2*XCENTER /COM*IF,D_VALUE,LT,7.0,THEN /COM /COMFOR NZ1=4, D_VALUE AND D_VALUE1 ARE 6.5 /COMFOR NZ1=5, D_VALUE AND D_VALUE1 ARE 5.5 /COMFOR NZ1=6, D_VALUE AND D_VALUE1 ARE 4.5 /COM *IF,NZ1,EQ,4,THEN *IF,D_VALUE,LT,6.5,AND,D_VALUE1,LT,6.5,THEN ESEL,A,,,K *ENDIF *ENDIF /COM *IF,NZ1,EQ,5,THEN *IF,D_VALUE,LT,5.5,AND,D_VALUE1,LT,5.5,THEN ESEL,A,,,K *ENDIF *ENDIF /COM *IF,NZ1,EQ,6,THEN *IF,D_VALUE,LT,4.5,AND,D_VALUE1,LT,4.5,THEN ESEL,A,,,K *ENDIF *ENDIF /COM *ENDDO /COM ESEL,U,,,1,1,1 /COM

275

University of Engineering & Technology, Taxila-Pakistan

EKILL,ALL ALLSEL,ALL /COM CSYS,1 SELTOL,1.0E-6 NSEL,S,LOC,X,CIR NSEL,R,LOC,Z,0 ESLN,S,0 EALIVE,ALL ALLSEL,ALL CSYS,0 /COM ESEL,S,LIVE NSLE,S NSEL,INVE D,ALL,TEMP,300 /COM ALLSEL,ALL *DO,I,1,TOTELEM,1 /COM *GET,ELESTATUS(1),ELEM,I,ATTR,LIVE *GET,ELEMVOL(I),ELEM,I,VOLU /COM *ENDDO /COM /COM ******************** WRITE ELEMENT STATUS TO A FILE ******************* /COM *CFOPEN, ELEMSTAT, TXT, *VWRITE, ELESTATUS(1) (F3.0) *CFCLOSE /COM /COM ************************************************************************** /COM ALLSEL,ALL TUNIF,300 ASEL,S,EXT ASEL,U,,,9,30,21 SFA,ALL,1,CONV,%CONVEC%,300 ALLSEL,ALL /COM SAVE, CYLINDER WELD (THERMAL),,DB,,ALL /COM ALLSEL,ALL PI=4*ATAN(1) OUT_CIR=2*PI*COR CIRTIME=OUT_CIR/WELDVEL DEL_T=CIRTIME/TOTHTLS CIR_THETA=360/TOTHTLS TIMPAR=0 LOOPPAR=TOTHTLS+5 BEGINPAR=1 /COM

276

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

*DO, T, BEGINPAR, LOOPPAR, 1 ALLSEL, ALL /COM *IF,T,GT,REAR_TIME,AND,T,LT,((TOTHTLS/2)-FRONT_TIME),THEN ENDTLOOP=TOTELEM/2 *ENDIF *IF,T,GT,((TOTHTLS/2)+REAR_TIME),AND,T,LT,(TOTHTLS-FRONT_TIME),THEN STARTLOOP=(TOTELEM/2)+1 *ENDIF /COM SOURCE_ZLOC=0 SOURCE_XLOC=COR SOURCE_YLOC = (T-0.75)*CIR_THETA /COM *IF,T,EQ,BEGINPAR,THEN FACT_COR=1.19 *ENDIF *IF,T,EQ,BEGINPAR+1,THEN FACT_COR=1.09 *ENDIF *IF,T,EQ,BEGINPAR+2,THEN FACT_COR=1.033 *ENDIF *IF,T,EQ,BEGINPAR+3,THEN FACT_COR=1.01 *ENDIF *IF,T,GE,BEGINPAR+4,THEN FACT_COR=1.0 *ENDIF /COM INPUT_HEAT(T) = 0 ELEMENTNO(T) = 0 /COM ALLSEL, ALL *DO, K, STARTLOOP, ENDTLOOP, 1 /COM DEL_AXIAL=SOURCE_ZLOC-EZCENT(K) DEL_RAD=SOURCE_XLOC-EXCENT(K) *IF, DEL_RAD,LT, 0, THEN DEL_RAD=-1*DEL_RAD *ENDIF /COM *IF, EYCENT(K), LT,0, THEN EYCENT(K)=EYCENT(K)+360 *ENDIF /COM *IF,SOURCE_YLOC,LT,90,AND,EYCENT(K),GT,270,THEN ANG-POSITION=EYCENT(K)-360 *ELSE ANG-POSITION=EYCENT(K) *ENDIF /COM *IF,SOURCE_YLOC,GT,270,AND,EYCENT(K),LT,90,THEN

277

University of Engineering & Technology, Taxila-Pakistan

ANG-POSITION=360+EYCENT(K) *ENDIF /COM DEL_CIR=EXCENT(K)*(ANG-POSITION-SOURCE_YLOC)*PI/180 *IF,DEL_AXIAL,LT,B,THEN *IF,DEL_CIR,LT,-0.0002,THEN A=AR F=FR ADEL_CIR=-1*DEL_CIR *ELSE A=AF F=FF ADEL_CIR=DEL_CIR *ENDIF /COM *IF, ADEL_CIR, LT, 1.0E-6, THEN ADEL_CIR=0 *ENDIF *IF, ADEL_CIR, LE, A, THEN /COM *IF, DEL_RAD, LE, C, THEN /COM *GET,ELEMSTAT,ELEM,K,ATTR,LIVE *IF,DEL_CIR,GE,-0.0002,AND,ELEMSTAT,LE,0,THEN ELEMSTAT=1 EALIVE,K MPCHG,2,K NSLE,INACTIVE DDELE,ALL,TEMP *ENDIF /COM FACTOR2ADD=1 /COM *IF,DEL_AXIAL,LT,0.006,AND,DEL_AXIAL,GT,0.0032,THEN *IF,DEL_RAD,LT,0.0012,THEN FACTOR2ADD=11.8 *ENDIF *ENDIF *IF,DEL_AXIAL,LT,0.0012,AND,DEL_RAD,GT,0.0045,THEN FACTOR2ADD=4.02 *ENDIF *IF,DEL_RAD,LT,0.0045,AND,DEL_RAD,GT,0.003,THEN *IF,DEL_AXIAL,LT,0.0012,THEN FACTOR2ADD=0.79 *ENDIF *ENDIF *IF,DEL_RAD,LT,0.003,THEN *IF,DEL_AXIAL,LT,0.0006,THEN FACTOR2ADD=0.55 *ENDIF *ENDIF *IF,DEL_AXIAL,GT,0.002,THEN *IF,DEL_RAD,GT,0.0015,AND,DEL_RAD,LT,0.006,THEN

278

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

FACTOR2ADD=0.0 *ENDIF *ENDIF /COM GEN_HEAT=0 Y_FACT=3*ADEL_CIR*ADEL_CIR/(A**2) X_FACT=3*DEL_RAD*DEL_RAD/(C**2) Z_FACT=3*DEL_AXIAL*DEL_AXIAL/(B**2) HEAT_FACTOR=FACTOR2ADD*FACT_COR*6*(3**0.5)*F*Q/(A*B*C*PI*(PI**0.5)) GEN_HEAT=HEAT_FACTOR*EXP(-X_FACT)*EXP(-Y_FACT)*EXP(-Z_FACT) /COM *IF,ELEMSTAT,EQ,1,THEN INPUT_HEAT(T)=INPUT_HEAT(T)+(GEN_HEAT*ELEMVOL(K)) *ENDIF *IF,T,LE,TOTHTLS+1,THEN BFE,K,HGEN,,GEN_HEAT *ENDIF *ENDIF *ENDIF *ENDIF *ENDDO /REPLOT /COM ALLSEL,ALL TIME,DEL_T*T NSUBST,2,4,2 SOLVE ESEL,S,,,1,TOTELEM,1 BFEDELE,ALL,HGEN ALLSEL,ALL SAVE, CYLINDER WELD (THERMAL),DB,,ALL /COM *DO,K,1,TOTELEM,1 ELEMNO=K *IF,SOURCE_YLOC,GE,EYCENT(K),THEN *IF,INDICATE(ELEMNO),LT,1,THEN AVRG_TEMP=0 T1=0 *DO,NNO,1,8,1 *GET,N1,ELEM,ELEMNO,NODE,NNO *GET,T1,NODE,N1,TEMP AVRG_TEMP=AVRG_TEMP+T1 *ENDDO AVRG_TEMP=AVRG_TEMP/8 *IF,PEAKTEMP(ELEMNO),LT,AVRG_TEMP,THEN PEAKTEMP(ELEMNO)=AVRG_TEMP EPTIME(ELEMNO)=TIMPAR *ENDIF /COM *IF,PEAKTEMP(ELEMNO),GT,AVRG_TEMP,OR,T,EQ,LOOPPAR,THEN *IF,AVRG_TEMP,LT,1740,OR,T,EQ,LOOPPAR,THEN /COM *IF,PEAKTEMP(ELEMNO),GE,1740,THEN

279

University of Engineering & Technology, Taxila-Pakistan

MPCHG,6,ELEMNO MNO=6 *ENDIF *IF,PEAKTEMP(ELEMNO),LT,1740,AND,PEAKTEMP(ELEMNO),GE,1423,THEN MPCHG,5,ELEMNO MNO=5 *ENDIF *IF,PEAKTEMP(ELEMNO),LT,1423,AND,PEAKTEMP(ELEMNO),GE,1323,THEN MPCHG,4,ELEMNO MNO=4 *ENDIF *IF,PEAKTEMP(ELEMNO),LT,1323,AND,PEAKTEMP(ELEMNO),GE,1223,THEN MPCHG,3,ELEMNO MNO=3 *ENDIF *IF,PEAKTEMP(ELEMNO),LT,1223,AND,PEAKTEMP(ELEMNO),GE,1083,THEN MPCHG,2,ELEMNO MNO=2 *ENDIF *IF,PEAKTEMP(ELEMNO),LT,1083,THEN MPCHG,1,ELEMNO MNO=1 *ENDIF MATREF(ELEMNO)=MNO INDICATE(ELEMNO)=1 TIMEMPCHD(ELEMNO)=TIMPAR *ENDIF *ENDIF *ENDIF *ENDIF *ENDDO /COM ALLSEL, ALL SAVE, CYLINDER WELD (THERMAL),DB,,ALL /COM *ENDDO /COM *CFOPEN,MATREF,TXT *VWRITE,MATREF(1) (F2.0) *CFCLOSE *CFOPEN,MATPCTIME,TXT *VWRITE,MATPCTIME(1) (F11.8) *CFCLOSE *CFOPEN,EPTIME,TXT *VWRITE,EPTIME(1) (F11.8) *CFCLOSE /COM /COM ***************************** COOLING PHASE **************************** /COM ALLSEL,ALL

280

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

TIMPAR=LOOPPAR*DEL_T T=TIMPAR *DO,I,1,6,1 T=T+0.5 TIME,T NSUBST,2,4,2 SOLVE *ENDDO /COM T=TIMPAR+3 *DO,I,1,10,1 T=T+1 TIME,T NSUBST,2 SOLVE *ENDDO /COM T=TIMPAR+13 *DO,I,1,6,1 T=T+5 TIME,T NSUBST,2 SOLVE *ENDDO /COM T=TIMPAR+43 *DO,I,1,6,1 T=T+15*I TIME,T NSUBST,2 SOLVE *ENDDO /COM T=TIMPAR+358 *DO,I,1,6,1 T=T+100*I TIME,T NSUBST,2 SOLVE *ENDDO /COM T=TIMPAR+2458 *DO,I,1,8,1 T=T+(I+0.5)*200 TIME,T NSUBST,2 SOLVE *ENDDO /COM SAVE, CYLINDER WELD (THERMAL), DB, ALL

281

University of Engineering & Technology, Taxila-Pakistan

Appendix A2

A2-1:

Pseudo-code of algorithm for Data Acquisition Module

[0] Initialize [1] IA ← User Input: Introduce new welding variable? If IA = Yes then execute steps 1.1 to 1.4, otherwise go to step 2. [1.1] Intake following from user in single structure St1: • Name of variable {name} • Whether the variable is used in antecedent or consequent part of rules? {type} • Whether the variable is numeric or categorical? {categ} • Units for numeric variable OR Options for categorical variable {units/options} [1.2] Assign new identity number variable_ID to the variable. Increment the total count of variables, i.e., n = n + 1 [1.3] If, for newly entered variable, variable{type} = input then store details of this variable in between details of first output variable and last input variable in file Variable.dat, ELSE if variable{type} = output then append details of this variable at the end of file [1.4] User Input: Introduce another variable? If Yes then go to step 1.1, else continue [2] IB ← User Input: Enter new data set? If IB=Yes then Open file: Data.dat. AND assign count_records = (number of records already stored in the file) AND execute steps 2.1 to 2.4, ELSE go to step 3 [2.1] Increment count_records= count_records + 1 [2.2] For x = 1 to n, Do: IC ← User Input: Enter value for variable[x]? If IC = Yes then intake value for variable[x], in given units/options AND assign record_num = count_records [2.3] Append data values to file: Data.dat [2.4] User Input: Enter another data set? If Yes then go to step 2.1, ELSE store value of count_records to Data.dat AND go to step 3 [3] Copy all data sets Data.dat → Linked List Set of structure St1 plus record_num [4] ID ← User Input: Erase any data set? If ID=Yes then execute steps 4.1 to 4.5, otherwise go to step 5 [4.1] N1 ← User Input: Number of data set to be erased? [4.2] For x = 1 to count_records, execute loop, i.e., steps 4.2.1 to 4.2.2 [4.2.1] If Set[x]{record_num} = N1 then delete structure Set[x] AND go to step 4.3 [4.2.2] Increment x = x + 1 [4.3] For all the records having Set{record_num} > x, do following: Set{record_num} = Set{record_num} – 1 [4.4] Delete: Data.Dat. Open new file: Data.Dat. Copy the linked list Set to file Data.Dat [4.5] User Input: Erase another data set? If Yes then go to step 4.1, else continue [5] Add field obj to the structure St1. For x = 1 to n, do following, until condition in step 5.1.3 or step 5.1.7 is met: 282

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

[5.1] If variable[x]{type} = output then go to step 5.1.1 ELSE if variable[x]{type} = input then go to step 5.1.4 [5.1.1] IE ← User Input: Use output variable variable[x] as optimization objective? If IE = Yes then execute steps 5.1.1.1 to 5.1.1.2, ELSE assign variable[x]{obj} = N and go to step 5.1.2 [5.1.1.1] Assign variable[x]{obj} = Y [5.1.1.2] Create slider bar at output interface, representing objective-value requirements (Low and High) of variable[x] [5.1.2] Increment x = x + 1 [5.1.3] If x>n then go to step 6, otherwise go to step 5.1 [5.1.4] If variable[x]{categ} = numeric then create check box at output interface, prompting whether value of variable[x] be pre-fixed or not [5.1.5] Else if variable[x]{categ} = categorical then create choice box at output interface, with all the possible options of variable[x] fed to the box [5.1.6] Increment x = x + 1 [5.1.7] If x > n then go to step 6, otherwise go to step 5.1 [6] Move over to Part 2

A2-2:

Pseudo-code of algorithm for Self-Development of Fuzzy Sets Module

[7] Initialize following: • Linked list L1{data_value, next_node} • Linked list L2{data_value, next_node} • Linked list CL1{data_value, count, next_node} • Linked list CL2{data_value, neighbor_distance, next_node} • Linked list FS{set_number, peak_value, down_node} • 2-D linked list Fuzzy{variable_ID, next_node, FS down_node} [8] MAX_SET_IN ← User Input: Maximum number of fuzzy sets for input variable; MAX_SET_OUT ← User Input: Maximum number of fuzzy sets for output variable [9] Initialize x=1. If variable[x]{categ} = categorical then go to step 9.18; ELSE execute outer loop, i.e., steps 9.1 to 9.19 until condition in step 9.19 is met [9.1] Set the linked list Set to its first node. Execute inner loop 1, i.e., steps 9.1.1 to 9.1.4 until the condition in step 9.1.4 is met [9.1.1] If Set{variable_ID} = x then continue ELSE go to step 9.1.3 [9.1.2] Assign L1{data_value }= Set{value}. Append new node to L1 and move: L1 = L1{next_node} [9.1.3] Move: Set = Set{next_node} [9.1.4] If Set{next_node} = NULL then go to step 9.2 ELSE go to step 9.1.1 [9.2] Set the linked lists Set and L1 to their first nodes. [9.3] Sort linked list L1 in ascending order of L1{data_value}. Set L1 to its first node. Initialize cnt_data = 1 [9.4] If (variable[x]{type} = input) then Assign CL1{data_value} = L1{data_value} AND CL1{count} = 1 AND Execute inner loop 2, i.e., steps 9.4.1 to 9.4.3 until condition in step 9.4.3 is met; ELSE go to step 9.5 [9.4.1] If L1{data_value}MAX_SET_IN then copy top MAX_SET_IN number of CL1{data_value} to linked list L2; ELSE copy all of CL1{data_value} to linked list L2. Set L2 to its first node [9.8] If (variable[x]{type}=output & cnt_data n) then go to step 10 ELSE continue with outer loop and go to step 9.1 [10] For all the variables stored in Variable.dat, do the following: [10.1] If (variable[x]{type}=output & variable[x]{obj}=Y) then print objective fuzzy sets to Sets_Rules.clp: “(deftemplate Obj_” variable[x]{name} “0 100 percent” “( Low (0 1) (5 1) (95 0) )” “(High (5 0) (95 1) (100 1)) )” [11] Set 2-D linked list Fuzzy to its first node. Print fuzzy sets from Fuzzy to CLIPS file: Sets_Rules.clp

A2-3:

Pseudo-code of algorithm for Self-Development of Prediction Rule-Base

[12] Initialize the following: • Linked list Data_input{in_variable_ID, data_value, down_node} • 2-D linked list Data_output{serial_num, out_variable_ID, data_value, Data_input down_node, next_node} • Linked list Rule_antecedent{variable_ID, fuzzy_set_num, CF, down_node}; (CF stands for Certainty Factor) • 2-D linked list Rule_consequent{rule_num, variable_ID, fuzzy_set_num, CF, Salience, Rule_antecedent down_node, next_node} [13] Set linked list Set to its first node. For each record set of Set, there are data values for n1 number of output variables. Append n1 number of new nodes to 2-D linked list Data_output and copy data related to each of these output variables to the newly appended nodes of Data_output. For each of these newly appended nodes of Data_output attach a new linked list Data_input and copy the same data of the input variables to those newly attached lists. Do the same for all other record sets of linked list Set. [14] Set 2-D linked list Data_output to its first node. Execute the outer loop i.e., steps 14.1 to 14.14 until condition in step 14.14 is met [14.1] If Fuzzy {variable_ID} = Data_output {out_variable_ID} then go to step 14.2 ELSE move Fuzzy = Fuzzy{next_node} AND repeat step 14.1 [14.2] If Data_output{data_value} < Fuzzy{FS down_node{FS down_node {peak_value}}} then go to step 14.3 ELSE move Fuzzy{FS down_node} = Fuzzy{FS down_node{FS down_node}} AND repeat step 14.2 [14.3] Assign A = Data_output {data_value} – Fuzzy{FS down_node{peak_value}} AND assign B= Fuzzy{FS down_node{FS down_node{peak_value}}} – Data_output {data_value} AND assign A_Set = Fuzzy{FS down_node{set_number}} AND assign B_Set = Fuzzy{FS down_node{FS down_node{set_number}}} [14.4] Assign Ratio = A/(A+B) [14.5] If (Ratio>=0 & Ratio0.35 & Ratio=0.5 & Ratio=0.65 & Ratio= 0 & Ratio2 0.35 & Ratio2 < 0.5) then assign Rule_consequent{Rule_antecedent down_node{fuzzy_set_num}} = (AA_Set + BB_Set) / 2 AND Rule_consequent{Rule_antecedent down_node {CF}} = (Ratio2–0.35)/0.15 AND go to step 14.11.11 [14.11.9] ELSE If (Ratio2 >= 0.5 & Ratio2 < 0.65) then assign Rule_consequent{Rule_antecedent down_node{fuzzy_set_num}} = (AA_Set + BB_Set) / 2 AND Rule_consequent{Rule_antecedent down_node {CF}} = (0.65– Ratio2)/0.15 AND go to step 14.11.11 [14.11.10] ELSE If (Ratio2 >= 0.65 & Ratio2 0 & Rule_consequent{CF} > 0) then continue ELSE go to step 17.2.5 [17.2.2] Initialize Decision=0. Execute the inner loop, i.e., steps 17.2.2.1 to 17.2.2.4 until the condition in step 17.2.2.4 is met [17.2.2.1] If (RuleOut{Rule_antecedent down_node{variable_ID}} = Rule_consequent {Rule_antecedent down_node{variable_ID}} & RuleOut {Rule_antecedent down_node{fuzzy_set_num}} = Rule_consequent{Rule_antecedent down_node{fuzzy_set_num}}) then assign Decision = 1 ELSE assign Decision = 0 AND go to step 17.2.3 [17.2.2.2] Move RuleOut {Rule_antecedent down_node} = RuleOut{Rule_antecedent down_node {Rule_antecedent down_node}} [17.2.2.3] Move Rule_consequent {Rule_antecedent down_node} = Rule_consequent{Rule_antecedent down_node {Rule_antecedent down_node}} [17.2.2.4] If ((RuleOut {Rule_antecedent down_node {Rule_antecedent down_node}} = NULL) OR (Rule_consequent {Rule_antecedent down_node {Rule_antecedent down_node}} = NULL)) then go to step 17.2.3 ELSE go to step 17.2.2.1 [17.2.3] If Decision=1 then PRINT “A conflict between 2 rules has been detected. RULE 1:”. PRINT(Rule_consequent). PRINT “and RULE 2:”. PRINT

287

University of Engineering & Technology, Taxila-Pakistan

(RuleOut). IG ← User Input: “Enter 1 to abandon RULE 1, enter 2 to abandon RULE 2, enter any other key to keep both rules effective” ELSE go to step 17.2.5 [17.2.4] If IG=1 then assign Rule_consequent {CF}=0, ELSE if IG=2 then assign RuleOut{CF}=0 [17.2.5] Move: RuleOut = RuleOut {next_node} [17.2.6] If RuleOut {next_node} = NULL then go to step 17.3 ELSE go to step 17.2.1 [17.3] Move: Rule_consequent = Rule_consequent{next_node} [17.4] If Rule_consequent{next_node} = NULL then go to step 18 ELSE go to step 17.1 [18] Set 2-D linked list Rule_consequent to its first node [19] Execute loop, i.e., steps 19.1 to 19.3 until condition in step 19.3 is met [19.1] If Rule_consequent {CF} > 0 then print rule Rule_consequent to CLIPS file: Sets_Rules.clp [19.2] Move: Rule_consequent = Rule_consequent{next_node} [19.3] If Rule_consequent {next_node} = NULL then go to step 20 ELSE go to step 19.1 [20] Set 2-D linked list Rule_consequent to its first node

A2-5:

Pseudo-code of algorithm for Self-Development of Optimization Rule-Base

[21] Initialize following: • Linked list for recording the score of each fuzzy set for all the input variables VariScore{variable_ID, fuzzy_set_num, score, count, next_node} • Linked list Maxim{variable_ID, fuzzy_set_num, avg_score, next_node} for keeping record of sets that best satisfies objective of maximization of any objective output variable • Linked list Minim{variable_ID, fuzzy_set_num, avg_score, next_node} for keeping record of sets that best satisfies objective of minimization of any objective output variable [22] Copy variable_ID of all the input variables (i.e., the variable for which variable[x]{type} = input) and all the corresponding fuzzy set_numbers from linked list Fuzzy to linked list VariScore, in the fields VariScore{variable_ID} and VariScore{fuzzy_set_num}, respectively [23] Set linked list VariScore to its first node [24] For all the variables stored in Variable.dat, check: if (variable[x]{type} = output & variable[x]{obj} = Y) then, for that variable, execute steps 24.1 to 24.6 [24.1] Execute outer loop, i.e., steps 24.1.1 to 24.1.3 until the condition in step 24.1.3 is met [24.1.1] If Rule_consequent{variable_ID} = variable[x]{variable_ID} then execute inner loop, i.e., steps 24.1.1.1 to 24.1.1.6 until condition in step 24.1.1.6 is met; ELSE go to step 24.1.2 [24.1.1.1] If ((Rule_consequent {Rule_antecedent down_node {variable_ID}} = VariScore {variable_ID}) & (Rule_consequent {Rule_antecedent down_node {fuzzy_set_num}} = VariScore {fuzzy_set_num})) then go to step

288

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

24.1.1.2 ELSE move: VariScore = VariScore {next_node} AND repeat step 24.1.1.1 [24.1.1.2] Assign VariScore{score} = VariScore{score} + Rule_consequent {fuzzy_set_num} [24.1.1.3] Increment VariScore{count} = VariScore{count} + 1 [24.1.1.4] Set linked list VariScore to its first node [24.1.1.5] Move: Rule_consequent{Rule_antecedent down_node} = Rule_consequent{Rule_antecedent down_node{Rule_antecedent down_node }} [24.1.1.6] If Rule_consequent{Rule_antecedent down_node} = NULL then go to step 24.1.2 ELSE go to step 24.1.1.1 [24.1.2] Move Rule_consequent = Rule_consequent{next_node} [24.1.3] If Rule_consequent{next_node} = NULL then go to step 24.2 ELSE continue with outer loop and go to step 24.1.1 [24.2] Set 2-D linked list Rule_consequent to its first node [24.3] Initialize AB = 1. Execute middle loop, i.e., steps 24.3.1 to 24.3.6 until the condition in step 24.3.6 is met [24.3.1] Assign Maxim{variable_ID} = Minim{variable_ID} = AB AND assign Maxim{fuzzy_set_num} = Minim{fuzzy_set_num} = 1 AND assign Maxim{avg_score} = -9999 AND assign Minim{avg_score} = 99999 [24.3.2] Execute inner loop, i.e., steps 24.3.2.1 to 24.3.2.5 until the condition in step 24.3.2.5 is met [24.3.2.1] If VariScore{variable_ID} ≠ AB then go to step 24.3.2.4 ELSE continue [24.3.2.2] If ((VariScore{score} / VariScore{count}) > Maxim{avg_score}) then assign Maxim{fuzzy_set_num} = VariScore{fuzzy_set_num} AND assign Maxim{avg_score} = VariScore{score} / VariScore{count} [24.3.2.3] If ((VariScore{score} / VariScore{count}) < Minim{avg_score}) then assign Minim{fuzzy_set_num} = VariScore{fuzzy_set_num} AND assign Minim{avg_score} = VariScore{score} / VariScore{count} [24.3.2.4] Move VariScore = VariScore {next_node} [24.3.2.5] If VariScore {next_node} = NULL then go to step 24.3.3 ELSE go to step 24.3.2.1 [24.3.3] Set linked list VariScore to its first node [24.3.4] Append new nodes to linked lists Maxim and Minim AND move Maxim = Maxim{next_node} AND move Minim = Minim{next_node} [24.3.5] Increment AB = AB + 1 [24.3.6] If AB > (total count of input variables) then go to step 24.4 ELSE continue with middle loop and go to step 24.3.1 [24.4] Set linked lists Maxim and Minim to their first nodes. Maxim contains the list of input variables and their corresponding fuzzy set numbers that will maximize the value of the output variable for which the optimization process is currently going on (in step 24). Likewise Minim contains the list of input variables and their corresponding fuzzy set numbers that will minimize the value of the same output variable

289

University of Engineering & Technology, Taxila-Pakistan

[24.5] Print the optimization rules, using information contained in linked lists Maxim and Minim, to the CLIPS file Sets_Rules.clp [24.6] Switch over to next output variable and repeat step 24. If all the output variables have been processed then go to step 25 [25] END

A2-6: Auto-developed Fuzzy Sets & Optimization Rule-Base A2-6.1 Fuzzy Sets (Section 7.8.2) (deftemplate Welding_Current 150 290 A ( (S1 (150 1) (170 1) (210 0) ) (S2 (170 0) (210 1) (250 0) ) (S3 (210 0) (250 1) (290 1) ) )) (deftemplate Welding_Voltage 9 15 V ( (S1 (9 1) (10.5 1) (12 0) ) (S2 (10.5 0) (12 1) (13.5 0) ) (S3 (12 0) (13.5 1) (15 1) ) ) ) (deftemplate Welding_Speed 13.5 19.5 cm_per_min ( (S1 (13.5 1) (15 1) (16.5 0) ) (S2 (15 0) (16.5 1) (18 0) ) (S3 (16.5 0) (18 1) (19.5 1) ) ) ) (deftemplate Thickness 2.6667 5.3333 mm ( (S1 (2.6667 1) (3 1) (4 0) ) (S2 (3 0) (4 1) (5 0) ) (S3 (4 0) (5 1) (5.3333 1) ) ) ) (deftemplate Weld_Strength 690 790 MPa ( (S1 (690 1) (710 1) (720 0) ) (S2 (710 0) (720 1) (730 0) ) (S3 (720 0) (730 1) (740 0) ) (S4 (730 0) (740 1) (750 0) ) (S5 (740 0) (750 1) (760 0) ) (S6 (750 0) (760 1) (770 0) ) (S7 (760 0) (770 1) (780 0) ) (S8 (770 0) (780 1) (790 1) ) ) ) (deftemplate Distortion 1 7.5 mm ( (S1 (1 1) (2.5 1) (3 0) ) (S2 (2.5 0) (3 1) (3.5 0) ) (S3 (3 0) (3.5 1) (4 0) ) (S4 (3.5 0) (4 1) (4.5 0) ) 290

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

(S5 (5 0) (5.5 1) (6 0) ) (S6 (5.5 0) (6 1) (6.5 0) ) (S7 (6 0) 6.5 1) (7 0) ) (S8 (6.5 0) (7 1) (7.5 1) ) ) ) (deftemplate Residual_Stresses 100 550 MPa ( (S1 (100 1) (150 1) (200 0) ) (S2 (150 0) (200 1) (250 0) ) (S3 (200 0) (250 1) (300 0) ) (S4 (250 0) (300 1) (350 0) ) (S5 (300 0) (350 1) (400 0) ) (S6 (350 0) (400 1) (450 0) ) (S7 (400 0) (450 1) (500 0) ) (S8 (450 0) (500 1) (550 1) ) ) )

A2-6.2 Optimization Rule-Base (Section 7.8.2) (defrule optimization1 (declare (salience 4000)) (Obj_Weld_Strength High) (or (not (Welding_Current ?)) (Welding_Current S2)) => (assert (Welding_Current S2))) (defrule optimization2 (declare (salience 4000)) (Obj_ Weld_Strength High) (or (not (Welding_Voltage ?)) (Welding_Voltage S2)) => (assert (Welding_Voltage S2))) (defrule optimization3 (declare (salience 4000)) (Obj_ Weld_Strength High) (or (not (Welding_Speed ?)) (Welding_Speed S3)) => (assert (Welding_Speed S3))) (defrule optimization4 (declare (salience 4000)) (Obj_ Weld_Strength High) (or (not (Thickness ?)) (Thickness S3)) => (assert (Thickness S3))) (defrule optimization5 (declare (salience 4000)) (Obj_ Weld_Strength High) (or (not (Weld_Type ?)) (Weld_Type S0)) => (if (>= ?*slid1* 50) then (assert (Weld_Type S0)))) (defrule optimization6 (declare (salience 4000)) (Obj_ Weld_Strength High) 291

University of Engineering & Technology, Taxila-Pakistan

(or (not (Trailing ?)) (Trailing S1)) => (if (>= ?*slid1* 50) then (assert (Trailing S1)))) (defrule optimization7 (declare (salience 4000)) (Obj_ Weld_Strength Low) (or (not (Welding_Current ?)) (Welding_Current S1)) => (assert (Welding_Current S1))) (defrule optimization8 (declare (salience 4000)) (Obj_ Weld_Strength Low) (or (not (Welding_Voltage ?)) (Welding_Voltage S1)) => (assert (Welding_Voltage S1))) (defrule optimization9 (declare (salience 4000)) (Obj_ Weld_Strength Low) (or (not (Welding_Speed ?)) (Welding_Speed S1)) => (assert (Welding_Speed S1))) (defrule optimization10 (declare (salience 4000)) (Obj_ Weld_Strength Low) (or (not (Thickness ?)) (Thickness S1)) => (assert (Thickness S1))) (defrule optimization11 (declare (salience 4000)) (Obj_ Weld_Strength Low) (or (not (Weld_Type ?)) (Weld_Type S1)) => (if (< ?*slid1* 50) then (assert (Weld_Type S1)))) (defrule optimization12 (declare (salience 4000)) (Obj_ Weld_Strength Low) (or (not (Trailing ?)) (Trailing S1)) => (if (< ?*slid1* 50) then (assert (Trailing S1)))) (defrule optimization13 (declare (salience 4000)) (Obj_Distortion High) (or (not (Welding_Current ?)) (Welding_Current S3)) => (assert (Welding_Current S3))) (defrule optimization14 (declare (salience 4000)) (Obj_ Distortion High) (or (not (Welding_Voltage ?)) (Welding_Voltage S3)) => (assert (Welding_Voltage S3))) (defrule optimization15 (declare (salience 4000)) (Obj_ Distortion High) (or (not (Welding_Speed ?)) (Welding_Speed S1)) 292

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

=> (assert (Welding_Speed S1))) (defrule optimization16 (declare (salience 4000)) (Obj_ Distortion High) (or (not (Thickness ?)) (Thickness S1)) => (assert (Thickness S1))) (defrule optimization17 (declare (salience 4000)) (Obj_ Distortion High) (or (not (Weld_Type ?)) (Weld_Type S1)) => (if (>= ?*slid2* 50) then (assert (Weld_Type S1)))) (defrule optimization18 (declare (salience 4000)) (Obj_ Distortion High) (or (not (Trailing ?)) (Trailing S0)) => (if (>= ?*slid2* 50) then (assert (Trailing S0)))) (defrule optimization19 (declare (salience 4000)) (Obj_ Distortion Low) (or (not (Welding_Current ?)) (Welding_Current S1)) => (assert (Welding_Current S1))) (defrule optimization20 (declare (salience 4000)) (Obj_ Distortion Low) (or (not (Welding_Voltage ?)) (Welding_Voltage S1)) => (assert (Welding_Voltage S1))) (defrule optimization21 (declare (salience 4000)) (Obj_ Distortion Low) (or (not (Welding_Speed ?)) (Welding_Speed S3)) => (assert (Welding_Speed S3))) (defrule optimization22 (declare (salience 4000)) (Obj_ Distortion Low) (or (not (Thickness ?)) (Thickness S3)) => (assert (Thickness S3))) (defrule optimization23 (declare (salience 4000)) (Obj_ Distortion Low) (or (not (Weld_Type ?)) (Weld_Type S1)) => (if (< ?*slid2* 50) then (assert (Weld_Type S1)))) (defrule optimization24 (declare (salience 4000)) (Obj_ Distortion Low) (or (not (Trailing ?)) (Trailing S1)) => 293

University of Engineering & Technology, Taxila-Pakistan

(if (< ?*slid2* 50) then (assert (Trailing S1)))) (defrule optimization25 (declare (salience 4000)) (Obj_Residual_Stresses High) (or (not (Welding_Current ?)) (Welding_Current S3)) => (assert (Welding_Current S3))) (defrule optimization26 (declare (salience 4000)) (Obj_ Residual_Stresses High) (or (not (Welding_Voltage ?)) (Welding_Voltage S3)) => (assert (Welding_Voltage S3))) (defrule optimization27 (declare (salience 4000)) (Obj_ Residual_Stresses High) (or (not (Welding_Speed ?)) (Welding_Speed S1)) => (assert (Welding_Speed S1))) (defrule optimization28 (declare (salience 4000)) (Obj_ Residual_Stresses High) (or (not (Thickness ?)) (Thickness S1)) => (assert (Thickness S1))) (defrule optimization29 (declare (salience 4000)) (Obj_ Residual_Stresses High) (or (not (Weld_Type ?)) (Weld_Type S0)) => (if (>= ?*slid3* 50) then (assert (Weld_Type S0)))) (defrule optimization30 (declare (salience 4000)) (Obj_ Residual_Stresses High) (or (not (Trailing ?)) (Trailing S0)) => (if (>= ?*slid3* 50) then (assert (Trailing S0)))) (defrule optimization31 (declare (salience 4000)) (Obj_ Residual_Stresses Low) (or (not (Welding_Current ?)) (Welding_Current S1)) => (assert (Welding_Current S1))) (defrule optimization32 (declare (salience 4000)) (Obj_ Residual_Stresses Low) (or (not (Welding_Voltage ?)) (Welding_Voltage S1)) => (assert (Welding_Voltage S1))) (defrule optimization33 (declare (salience 4000)) (Obj_ Residual_Stresses Low) (or (not (Welding_Speed ?)) (Welding_Speed S3)) => (assert (Welding_Speed S3))) 294

Expert System for Optimization of Welding Process of Thin Walled HSLA Steel Structures

(defrule optimization34 (declare (salience 4000)) (Obj_ Residual_Stresses Low) (or (not (Thickness ?)) (Thickness S3)) => (assert (Thickness S3))) (defrule optimization35 (declare (salience 4000)) (Obj_ Residual_Stresses Low) (or (not (Weld_Type ?)) (Weld_Type S1)) => (if (< ?*slid3* 50) then (assert (Weld_Type S1)))) (defrule optimization36 (declare (salience 4000)) (Obj_ Residual_Stresses Low) (or (not (Trailing ?)) (Trailing S1)) => (if (< ?*slid3* 50) then (assert (Trailing S1))))

295