Empirical Combustion Modeling in SI Engines [PDF]

Zitiervorschau

Empirical Combustion Modeling in SI Engines

FREDRIK LINDSTRÖM

Licentiate thesis Department of Machine Design Royal Institute of Technology SE-100 44 Stockholm

TRITA – MMK 2005:19 ISSN 1400-1179 ISRN/KTH/MMK/R-05/19-SE

TRITA – MMK 2005:19 ISSN 1400-1179 ISRN/KTH/MMK/R-05/19-SE Empirical Combustion Modeling in SI Engines Fredrik Lindström Licentiate thesis Academic thesis, which with the approval of Kungliga Tekniska Högskolan, will be presented for public review in fulfilment of the requirements for a Licentiate of Engineering in Machine Design. The public review is held at Kungliga Tekniska Högskolan, Brinellvägen 23 in room B1 at 13:00 on the 26th of September 2005.

ABSTRACT This licentiate thesis concerns the modeling of spark ignition engine combustion for use in one dimensional simulation tools. Modeling of knock is of particular interest when modeling turbocharged engines since knock usually limits the possible engine output at high load. The knocking sound is an acoustic phenomenon with pressure oscillations triggered by autoignition of the unburned charge ahead of the propagating flame front and it is potentially damaging to the engine. To be able to predict knock it is essential to predict the temperature and pressure in the unburned charge ahead of the flame front. Hence, an adequate combustion model is needed. The combustion model presented here is based on established correlations of laminar burning velocity which are used to predict changes in combustion duration relative to a base operating condition. Turbulence influence is captured in empirical correlations to the engine operating parameters spark advance and engine speed. This approach makes the combustion model predictive in terms of changes in gas properties such as mixture strength, residual gas content, pressure and temperature. However, a base operating condition and calibration of the turbulence correlations is still needed when using this combustion model. The empirical models presented in this thesis are based on extensive measurements on a turbocharged four cylinder passenger car engine. The knock model is simply a calibration of the Arrhenius type equation for ignition delay in the widely used Livengood-Wu knock integral to the particular fuel and engine used in this work. Keywords: spark ignited engines, combustion modeling, knock, 1D simulation, Wiebe, divided exhaust period

iii

SAMMANFATTNING Denna avhandling behandlar modellering av förbränning i ottomotorer med endimensionella simuleringsverktyg. Knackmodellering är av särskilt intresse vid simulering av turbomotorer eftersom dessa motorer oftast begränsas av knack vid höga laster. Det knackande ljudet är ett akustiskt fenomen som uppstår då den obrända bränsle-luftblandningen självantänder framför flamfronten. Knack kan skada motorn. För att förutsäga knack är det av största vikt att känna till tryck och temperatur i den obrända blandningen framför flamfronten, vilket leder till behovet av en förbränningsmodell. Förbränningsmodellen som presenteras är baserad på beprövade korrelationer för laminär flamhastighet. Dessa används för att förutspå förändringar i förbränningsduration relativt en referensförbränning. Turbulensens påverkan på förbränningen fångas genom korrelationer mot tändvinkel och motorvarvtal. Med detta tillvägagångssätt blir förbränningsmodellen prediktiv med avseende på förändringar i temperatur, tryck, restgashalt och bränsle-luftförhållande. Ett referenstillstånd och kalibrering av turbulensens påverkan på förbränningen behövs dock fortfarande. De empiriska modellerna som presenteras i denna avhandling baseras på utförliga mätningar på en fyrcylindrig turbomotor. Knackmodellen är helt enkelt en kalibrering av den Arrhenius-liknande funktionen för tändfördröjning i LivengoodWu’s knackintegral med det bränsle och den motor som användes i testerna. Sökord: ottomotorer, förbränningsmodellering, knack, endimensionell simulering, Wiebe

iv

ACKNOWLEDGEMENTS Where am I to start this display of gratitude towards colleagues and friends? From the beginning of course! It all started with Emil Åberg, who had the patience to guide me, a complete novice in the world of engines and a stranger to essential skills such as welding and soldering, into the fascinating interior of the DEP engine. With the aid of Emil, I could eventually start discussing engines with Hans-Erik Ångström, who never stops to evolve his wonderful engine laboratory. The software support department, i.e. Hans-Erik, has been impeccable; we once clocked the time from failure detection in the Cell4 system to installed and working program update to just over 12 minutes! Christel Elmqvist-Möller put me right into the business of research, as she gave me a good chunk of the knock model to bite into just weeks after my first close encounter with SI engines. I would also like to thank Christel for excellent project management throughout this entire project. The mechanics Henrik Nilsson at KTH and Jon Nilsson at GM Powertrain Sweden kept the engines running even though we did our best to kill them (the engines of course) at times. I thank lab manager Eric Lycke especially for talking me out of disassembling the gear box of my car. As always, there was a much more straightforward solution to the problem… Gautam Kalghatgi has contributed a lot to the work on combustion and knock modeling. Thank you for many interesting discussions and insights and for being such a joyful person. But who is there to talk about knock modeling when Gautam has left the building? Per Risberg and Fredrik Agrell! Thank you also Fredrik for having enough confidence in me to let me present your work in Rio. All colleagues at KTH have made the time here very rewarding. Fredrik Westin’s immense knowledge of racing engines; Andreas Cronhjort for sharing knowledge in filtering and electronics; fellow sailor Fredrik Wåhlin; fellow musician Per Strålin. Thank you Ulrica and Niklas for finally taking the burden of being the department junior off my shoulders. Some people at GM Powertrain Sweden have contributed with valuable comments and support along the way, in particular Börje Grandin, Eric Olofsson, Lennarth Zander and Raymond Reinmann. The link to GM Powertrain has made the work meaningful to me.

v

And finally, thank you to friends and family, nieces and nephews, who encouraged me every now and then along the way. I’ll let you in on a secret. It actually started almost exactly 30 years ago on the outskirts of Rio de Janeiro, Brazil.

Fredrik Lindström Brasilia, August 2005

vi

LIST OF PUBLICATIONS Paper I Divided Exhaust Period – A Gas Exchange System for Turbocharged SI Engines by Christel Elmqvist-Möller, Pontus Johansson, Börje Grandin and Fredrik Lindström. SAE Technical Paper 2005-01-1150 presented by Christel Elmqvist-Möller at the SAE World Congress 2005 in Detroit, USA.

Paper II Optimizing Engine Concepts by Using a Simple Model for Knock Prediction by Christel Elmqvist-Möller, Fredrik Lindström, Hans-Erik Ångström, and Gautam Kalghatgi. SAE Technical Paper 2003-01-3123 presented by Christel Elmqvist-Möller at the SAE Powertrain and Fluid Systems Conference 2003 in Pittsburg, USA.

Paper III An Empirical SI Combustion Model Using Laminar Burning Velocity Correlations by Fredrik Lindström, Hans-Erik Ångström, Gautam Kalghatgi and Christel Elmqvist-Möller. SAE Technical Paper 2005-01-2106 presented by Fredrik Lindström at the SAE Fuels & Lubricants Meeting 2005 in Rio de Janeiro, Brazil.

All three papers are appended to the end of this thesis.

vii

TABLE OF CONTENTS Abstract ..............................................................................................................iii Sammanfattning ................................................................................................. iv Acknowledgements ............................................................................................. v List of publications ........................................................................................... vii Abbreviations, symbols and subscripts............................................................... x Chapter 1 Introduction ..................................................................................1 1.1 Motivation ................................................................................................................2 1.2 Contributions ...........................................................................................................3 1.3 Thesis outline ...........................................................................................................4 Chapter 2 Combustion in spark ignition engines......................................... 5 2.1 Gas Exchange ..........................................................................................................5 2.1.1 Residual gases ......................................................................................................6 2.1.2 Fuel........................................................................................................................6 2.1.3 Turbulence ...........................................................................................................6 2.2 Combustion ..............................................................................................................7 2.2.1 Laminar burning velocity...................................................................................7 2.2.2 Cycle to cycle variations.....................................................................................8 2.3 Knock ........................................................................................................................9 2.3.1 Autoignition chemistry.......................................................................................9 2.3.2 Modes of Autoignition.....................................................................................11 2.3.3 Combustion Chamber Oscillation Modes ....................................................12 2.3.4 Measures of Knock...........................................................................................14 2.4 Combustion simulation ........................................................................................14 2.4.1 The Wiebe Function.........................................................................................17 2.4.2 Knock Simulation .............................................................................................17 Chapter 3 Experimental Method ................................................................ 23 3.1 Measurements ........................................................................................................23 3.1.1 Measurement system ........................................................................................23 3.1.2 Pressure measurement......................................................................................24 3.1.3 Temperature measurement..............................................................................25

viii

3.1.4 Other measurements ........................................................................................27 3.2 Data Acquisition ....................................................................................................29 3.2.1 Signal Conditioning ..........................................................................................30 3.2.2 FIR Low Pass Filter..........................................................................................32 3.2.3 FIR Band Pass Filtering ...................................................................................35 3.2.4 IIR Filtering for Knock Onset Detection.....................................................35 3.3 Heat release calculation.........................................................................................36 3.3.1 Thermodynamic properties of mixture .........................................................37 3.4 Experiment engines...............................................................................................42 3.4.1 Divided Exhaust Period...................................................................................42 3.4.2 Engine specifications........................................................................................43 Chapter 4 Knock Modeling......................................................................... 45 4.1 Experiments ...........................................................................................................45 4.2 Data Evaluation .....................................................................................................46 4.3 Knock Model Calibration.....................................................................................47 4.4 Discussion...............................................................................................................49 Chapter 5 Combustion Modeling Using the Wiebe Function.....................51 5.1 Existing Wiebe models .........................................................................................51 5.1.1 Structure of Existing Models ..........................................................................52 5.1.2 Model Identification Procedure......................................................................53 5.1.3 Csallner ...............................................................................................................53 5.1.4 Witt......................................................................................................................55 5.2 Experiments ...........................................................................................................55 5.3 Data Evaluation .....................................................................................................59 5.3.1 Wiebe Parameter Identification ......................................................................60 5.4 Combustion model calibration ............................................................................62 5.4.1 Modeling Speed Influence ...............................................................................63 5.5 Results .....................................................................................................................64 Chapter 6 Conclusions ................................................................................ 65 6.1 Future work ............................................................................................................66 References ......................................................................................................... 69

ix

ABBREVIATIONS, SYMBOLS AND SUBSCRIPTS Abbreviations SI Spark Ignition HCCI Homogenous Charge Compression Ignition PFI Port Fuel Injection CA Crank Angle TDC Top Dead Center aTDC crank angles after combustion TDC bTDC crank angles before combustion TDC EVO Exhaust Valve Opening IVC Inlet Valve Closing IMEP Indicated Mean Effective Pressure, 720 CA PMEP Pumping Mean Effective Pressure EGR Exhaust Gas Recirculation MBT Maximum Brake Torque spark timing CFR Cooperative Fuels Research octane rating engine RON Research Octane Number MON Motor Octane Number PRF Primary Reference Fuel, iso-octane/n-heptane blend NTC Negative Temperature Coefficient FIR Finite Impulse Response filter IIR Infinite Impulse Response filter FS Full Scale, in measurement system errors A/D Analogue to Digital Symbols A scaling factor in ignition delay correlation and cylinder heat transfer area Ai constant in AVL expression for temperature dependent cp/cv a scaling factor in Wiebe function B cylinder bore and temperature coefficient in ignition delay correlation

Bm, Bλ c

constants in laminar burning velocity correlations speed of sound x

Cp , CV cp, cV Fi, Gi, Hi fi, gi, hi fLP fm,n,k hc Jm Kf m M N n p pm PR Qch q0 , q1 R SL

molar heat capacity at constant pressure / volume specific heat capacity at constant pressure / volume influencing functions in Wiebe correlation normalized influencing functions in Wiebe correlation low pass filter cutoff frequency in Hz combustion chamber natural frequencies heat transfer coefficient in Woschni equation for heat transfer Bessel’s function of the first kind number of periods of sinc function in FIR-filter kernel combustion mode parameter in Wiebe function molar mass engine speed in rpm pressure exponent in ignition delay correlation pressure motored pressure pressure ratio chemical energy released from fuel normalized cutoff frequencies in filters universal gas constant, 8.314 kJ/molK laminar burning velocity

Sp

piston mean velocity

T Tu V Vd xb

gas temperature unburned zone temperature volume, cylinder volume displaced volume mass fraction burned

~ xr

burned gas mole fraction

α, αg

temperature exponent in laminar burning velocity correlations

αm,n

zeros of Bessel’s function of the first kind

β, βg

pressure exponent in laminar burning velocity correlations

β

exponent for air/fuel ratio dependence in ignition delay

γ

ratio of specific heats, cp/cv

xi

λ

normalized air/fuel ratio

λm

constant in laminar burning velocity correlations

θ

crank angle and cylindrical angle coordinate

θ0

start of combustion aTDC

θd

flame development period

∆θ

total combustion duration

τ

ignition delay time

Subscripts EOC SOC b exh k m n norm u 0 and ref

end of combustion ,ref start of combustion burned exhaust longitudinal mode number circumferential mode number radial mode number normalized unburned reference condition

xii

Chapter 1 INTRODUCTION The internal combustion engine as we know it today was invented over a hundred years ago by the likes of Otto and Diesel. Still today, however, there is progress and improvements in the design and operation of internal combustion engines. One of the key factors for the success of the internal combustion engine in the transport of people is the reliability and flexibility as a mobile power source. The focus for research and development has shifted over the years, depending on trends and demand from society as a whole as well as on new enabling technologies. From society, focus has shifted from reductions of the local emissions towards the reduction of greenhouse gas emissions, i.e. CO2 or the fuel efficiency. Introduction of the three way catalyst and close loop fueling control basically solved the problem of local emissions of unburned hydrocarbons, carbon monoxide and nitrogen oxides for spark ignited engines. Today, with the introduction of direct fuel injection with stratified charge to improve part load fuel economy, emissions of nitrous oxides are again coming into focus since the three way catalyst doesn’t work in the overall fuel lean conditions. A rising concern today is the future availability of energy resources suitable for transportation which also brings focus to renewable energy sources and efficiency in using the available energy. State of the art spark ignited engine of today can benefit from technologies such as: variable valve timing, which can replace throttling and reduce part load gas exchange losses and also maximize power output by improving volumetric efficiency; fuel injection with feedback control to maximize after treatment system efficiency; knock sensors for optimal combustion phasing; turbocharging which increases power

1

Empirical Combustion Modeling in SI Engines density and enables downsizing of the engine with less friction losses and improved part load efficiency as a result. Evolution in the fields of electronics and sensors gives new opportunities to optimize the operation of internal combustion engines. Developments in auxiliary units such as water pumps and oil pumps reduce parasitic losses. New manufacturing methods and materials reduce the weight of engines. Electric hybrid systems enable recovery of breaking energy in the vehicle. Many more examples exist of parts of the engine or vehicle where improvements are being made today, all serving to improve fuel efficiency. However, the general trend in passenger cars is that passenger safety and comfort requirements lead to an increased vehicle weight with increased need for power and higher fuel consumption as a result. Statistics from the European Union [1] shows that during the period 1995 to 2002 the average vehicle weight increased by 10 %, average power by over 20 % while average vehicle CO2 emissions decreased by 12,1 % in new vehicles. The average CO2 emissions per kilometer is lower for diesel powered vehicles than for the equivalent gasoline powered vehicle owing to higher average efficiency of the diesel engine An increasing share of diesel powered vehicles explains some of the improvements in average CO2 emissions. The gasoline powered vehicles did however decrease average CO2 emissions per kilometer by 9,1 % in the EU statistics mentioned above. Today, the fuel cell is put forward as an alternative for the internal combustion engine in automotive applications. The technology still has some hurdles to pass before it is a competitive alternative to internal combustion engines. In a recent presentation at SAE Fuels & Lubricants Meeting and Exhibition in Rio de Janeiro by Mitchell [2] the fuel efficiency of a fuel cell powered vehicle with on board fuel reformer was stated to be approximately 46 %, some percentage points over the best diesel powered vehicles. The cost for the fuel cell alone would however be several thousand US dollars per kilowatt of power. The average EU car had 78 kW of power in 2003. The conclusion from this has to be that efforts must be made to improve the currently working technology, i.e. the internal combustion engine, parallel to investigating new and perhaps better alternatives.

1.1 MOTIVATION Simulation tools are becoming increasingly important in the development and improvement of internal combustion engines. One dimensional simulation tools have been used within this project to evaluate a new gas exchange system for SI engines, the Divided Exhaust Period system. In one-dimensional simulation, equations for conservation of mass, momentum and energy are solved in time and in one space dimension along the main flow direction in the engine pipes. However, many 2

Chapter 1 - Introduction phenomena in engines are three-dimensional in their nature. Additional models, correlations or measurements are needed in one-dimensional codes to capture threedimensional phenomena such as flow in pipe bends and junctions, flow over valves and combustion. [3][4] One might think that the increasing use of simulation would decrease the need for expensive engine prototypes and tests. This is not the case today since the simulation models rely on test data to calibrate the sub-models describing three dimensional phenomena. This is especially true for turbocharged engines, where modeling of the turbine is singled out as one of the most difficult tasks [5]. In fact, engine simulation has put new demands on the engine measurement technique, requiring more crank angle resolved measurements of pressures at various positions and more detailed measurements on components that the one dimensional flow calculations fail to describe. For example Westin [5] and Gamma Technologies [4] describe the measurement data needed for calibrating a simulation model. The key focus of this work has been to improve the simulation models for combustion and especially knocking combustion. Knock can be described as an acoustic phenomenon where autoignition of the unburned gas ahead of the propagating flame front triggers pressure oscillations in the combustion chamber. The pressure waves give rise to the characteristic, potentially disturbing, sound which has given knock its name. The pressure pulses can however also damage the engine, why knock must be avoided. The Divided Exhaust Period concept is in part aimed at improving the knock resistance of turbocharged spark ignited engines by reducing the amount of hot residual gases that are trapped in the cylinder when the exhaust valves close. Hot residuals increase the charge temperature and reduce the knock resistance of the engine. To be able to investigate the potential improvement by using the Divided Exhaust Period system, a knock model had to be used in the simulation software. To be able to simulate knock, the in cylinder temperature and pressure have to be predicted accurately which lead to the work with the empirical combustion model. The overall target for this work has been to make the one dimensional simulation model more predictive.

1.2 CONTRIBUTIONS The main contribution of this work is the combustion model presented in Paper III. The presented model combines the empirical approach of using the Wiebe function to describe the heat release in SI engines with established correlations for laminar burning velocity. This makes the presented model predictive in terms of changes in

3

Empirical Combustion Modeling in SI Engines gas properties such as temperature, pressure and composition. The model is still very intuitive and easy to interpret or compare with engine tests. As for the appended papers, my contributions to Paper I and Paper II have been the experimental investigations involved in those papers along with analysis of the experimental results. Paper I concerning Divided Exhaust Period was originally written in two parts, a theoretical and simulation part with fellow licentiate candidate Christel Elmqvist-Möller as main author, and an experimental part with me as main author. In Paper III, I carried out all experiments and analysis with very valuable input regarding how to model combustion from the co-authors. A fruitful team work was developed between the simulation part of the project, i.e. Christel Elmqvist-Möller, and the experimental part of the project.

1.3 THESIS OUTLINE First of all, this thesis contains a short introduction to combustion in SI engines, which serves as a background to the work presented in Paper II and Paper III. Some factors influencing combustion are presented. The knock phenomenon is explored both from an autoignition chemistry viewpoint and from the combustion chamber acoustic viewpoint to help in understanding the measurement technique as well as the knock model presented in Paper II. Description of the measurement technique and measurement data processing tools follows. Signal processing is an important part of combustion engine data analysis. Measurements in internal combustion engines can easily produce several megabytes of data in only a few seconds. With this amount of data automated analysis is preferred. When performing this automated analysis it is important to know what can go wrong and how to handle errors in the measured data. Therefore digital filtering is explored. The algorithm used for calculating in cylinder heat release, i.e. the release of chemical energy from the fuel, from measured cylinder pressure data is also described. Some additional comments to the appended papers are found last in the thesis. These two works give simplified but practical descriptions of how combustion and knock can be simulated in SI engines. With the lack of physical models which are easy to use and calibrate, an empirical approach as in this work can give at least partly predictive simulation tools.

4

Chapter 2 COMBUSTION IN SPARK IGNITED ENGINES The following paragraphs contain a brief overview of the combustion in port fuel injected spark ignition four stroke engines. This overview serves as a base for understanding the simplifications made in the models presented later in this thesis. Parameters that influence the combustion event and cycle-to-cycle variations are summarized. The knock phenomenon is explored. Finally, combustion simulation is discussed.

2.1 GAS EXCHANGE The gas exchange process plays a major role for the combustion in spark ignition engines. The mixture composition in the combustion chamber is set once the inlet valves have closed. Residual gas fraction and air/fuel ratio are important parameters affecting the combustion event. A large part of the residual gases are evacuated from the cylinder during the blow-down phase of the exhaust process, during which the cylinder pressure drops to the pressure in the exhaust manifold. The remaining exhaust is pushed out from the cylinders by the piston during the exhaust stroke. Scavenging of the cylinder is controlled by the pressure difference from intake to exhaust system during the valve overlap period. A positive pressure difference, i.e. higher pressure in the intake system than in the exhaust system, can be accomplished at high load by exhaust and intake system tuning and design and by proper turbocharger matching [6].

5

Empirical Combustion Modeling in SI Engines

2.1.1 Residual gases Residual gases that are trapped in the cylinder when the exhaust valves closes affect the cylinder charge in several ways. The hot residual gases increase charge temperature thereby decreasing charge density and volumetric efficiency. The increased charge temperature reduces the knock resistance of the engine, since knock is highly temperature dependent. Residual gases also have a minor effect on the ratio of specific heats, γ. Residual gases have slightly lower γ than pure air. Vaporized hydrocarbon fuel, on the other hand, has very low γ. Increased residual gases with constant air/fuel ratio decreases the relative amount of fuel in the mixture. The overall result for premixed SI engines is a slight increase in γ when residual gas content increases, according to the frozen mixture gas model described in Chapter 3.1.1. Increased γ leads to increased charge temperature during isentropic compression, but this effect is very small compared to the temperature rise associated with the mixing of hot residual gases with the fresh charge. Calculation of isentropic maximum temperature with the heat release algorithms and assumptions described in Chapter 3 reveal that a 10% increase of residual gases at one high load operating condition cause an increase from 683 K to 803 K in maximum temperature. Only 4 K, or 3,5% of the total increase in temperature, is associated with the increase in γ.

2.1.2 Fuel Fuel is usually injected in the inlet runner towards the inlet valves in port fuel injected engines. Some of the injected fuel is deposited on the inlet runner walls, on the valve stems and on the back face of the inlet valves to form a fuel film and puddles. Fuel enters the cylinder during the intake stroke in vapor phase and in liquid phase. The fuel evaporates and mixes with air and residual gases during the intake and compression stroke.

2.1.3 Turbulence The flow over the inlet valves creates turbulence. Large scale rotating charge motion is created from the intake jet in the form of tumble, swirl or combinations thereof. As shown for example by Söderberg [7], turbulence increases close to top dead center due to tumble breakdown in tumbling engines. Late inlet valve closing combined with low valve lift creates more turbulence around top dead center in the same work. Piston motion during the compression stroke also creates a vortex near the cylinder wall which further increases turbulence at TDC [8].

6

Chapter 2 – Combustion in spark ignited engines

2.2 COMBUSTION The mixture in the combustion chamber is ignited by the spark discharge between the electrodes of the spark plug and a flame kernel is formed. Exothermic chemical reactions take place in the flame kernel. Diffusion of heat and radicals from the flame kernel surface makes the kernel expand and start propagating in the combustion chamber. A thin, smooth reaction sheet, with thickness in the order of 0.1 mm, separates the burned gases from the unburned gases [8]. See Glassman [9] for a thorough discussion about laminar flame propagation. The early flame has been shown to propagate with a speed close to experimentally determined laminar burning velocity [10]. The time between spark discharge and any measurable increase in pressure due to combustion is sometimes referred to as the ignition delay period. The term ignition delay is misleading because the flame kernel will usually have grown to a significant size by this time. For example, Tagalian and Heywood [11] measured flame radiuses of about 5 mm when 0,1 % of the charge mass was burned. A more correct term would be flame development period. When the flame kernel has reached the size of the smallest turbulent eddies, the reaction sheet is wrinkled, resulting in increased surface area of the flame and increased burning velocity. The flame extinguishes when the flame eventually reaches the relatively cold combustion chamber walls

2.2.1 Laminar burning velocity Several correlations exist for laminar burning velocity SL of hydrocarbon/air mixtures. Heywood [8] summarizes the findings of Metghalchi and Keck [12][13] in the equation: α

β

⎛T ⎞ ⎛ p ⎞ ⎟⎟ S L = S L ,0 ⎜⎜ u ⎟⎟ ⎜⎜ (2.1) ⎝ T0 ⎠ ⎝ p 0 ⎠ where T0 = 298 K and p0 = 101,3 kPa is reference temperature and pressure. The exponents α and β are functions of equivalence ratio, i.e. the inverse of the normalized air/fuel ratio λ:

(

)

α = 2,18 − 0,8 λ −1 − 1

(

−1

)

β = −0,16 + 0,22 λ − 1

7

(2.2)

Empirical Combustion Modeling in SI Engines The reference laminar burning velocity SL,0 is a function of equivalence ratio and burned gas mole fraction ~ xr :

(

)

(

)

2 S L , 0 (λ , ~ x r ) = 1 − 2,06 ⋅ ~ x r0, 77 ⋅ ⎛⎜ Bm + Bλ λ −1 − λ −m1 ⎞⎟ ⎝ ⎠

(2.3)

Values for the constants in Equations (2.2) and (2.3) are found in Table 2.1. Table 2.1 Constants for the laminar burning velocity correlations in Equations (2.2) and (2.3) from [8]. Fuel Bm λm Bλ [cm/s] [cm/s] Methanol

1/1,11

36,9

-140,5

Propane

1/1,08

34,2

-138,7

Isooctane

1/1,13

26,3

-84,7

Gasoline

1/1,21

30,5

-54,9

An additional correlation for the temperature and pressure exponents in Equation (2.1) for gasoline from the same source as the other correlations is:

α g = 2,4 − 0,271 ⋅ λ −3.51 β g = −0,357 + 0,14 ⋅ λ − 2, 77

(2.4)

2.2.2 Cycle to cycle variations Variations in the local and global air/fuel ratio, residual gas content, mixing and turbulence characteristics between cycles produce cycle to cycle variations. Turbulence during the intake and compression strokes affects the mixing of air, fuel and residual gases. The charge properties in the vicinity of the spark plug are of particular importance [14]. Air/fuel ratio and residual gas content affect the laminar burning velocity which in turn affects the ignition delay time. The time for developing a turbulent flame is also affected by local variations in turbulence length scales and intensity. Differences in ignition delay affect the turbulent flame speed during the rapid combustion period, since the turbulence varies with time. The large scale charge motion can move the flame in the combustion chamber, affecting the flame interaction with the cylinder walls, hence affecting the flame area and unburned gas temperature.

8

Chapter 2 – Combustion in spark ignited engines

2.3 KNOCK The knock phenomenon has been extensively studied in the past. Grandin [15] gives an interesting historical review of the evolution of knowledge in the field of knock from the 1920’s an onwards. Knock is initiated by autoignition of the unburned charge ahead of the flame front. Autoignition of the end gas leads to a pressure disturbance in the combustion chamber which induces pressure oscillations. The knocking sound associated with the combustion chamber pressure oscillations have given knock its name. It is also the pressure oscillations, together with an increase in heat transfer, that are potentially damaging to combustion chamber components.

2.3.1 Autoignition chemistry The recent interest in HCCI combustion, where a homogenous air/fuel mixture is compressed until it ignites, has caused renewed interest in autoignition research. HCCI related research also improves the understanding of the knock phenomena, since knock and HCCI combustion are practically the same. Westbrook [16] describe the chemical mechanisms leading to end gas autoignition in SI engines. Hydrogen peroxide, H2O2, is singled out as the most important species for autoignition chemistry. The dominating reaction in autoignition chemistry at typical engine temperatures below 1200 K is: H 2 O 2 + M → OH + OH + M

(2.5)

where M is a third body. H2O2 is accumulated during compression from low temperature reactions. Ignition occurs as a result of the chain branching reaction when H2O2 decomposes into two OH radicals. The decomposition of H2O2 is highly temperature dependent and occurs at 900 to 950 K at typical engine conditions. The rapid increase in concentration of the OH radical causes the remaining hydrocarbon to react and ignite. Since the reaction involves a third body, increasing the pressure will increase the probability for collisions and hence lower the critical temperature. Westbrook [16] states that the dominating factor for autoignition is the time at which the mixture reaches the critical temperature and anything that affects this time also affects when autoignition occurs. Low temperature heat release, also called cool flames or Negative Temperature Coefficient behavior (NTC), has strong influence on the time to reach the critical temperature. It is clear from the reasoning in the above paragraphs that the pressure and temperature history of the mixture influences the instantaneous ignition delay time for a given air/fuel mixture. 9

Empirical Combustion Modeling in SI Engines NTC behavior and also low temperature heat release is more pronounced for long unbranched paraffinic hydrocarbon chains such as n-Heptane. Risberg [17] summarizes the principal reaction in the low temperature chemistry of some different hydrocarbons. Iso-octane, or 2,2,4-tri-methyl-pentane, is more branched and displays less low temperature heat release and less pronounced NTC behavior which is also evident from Fieweger et. al. [18] where the ignition delay is measured in a shock tube for different PRF mixtures, see Figure 2.1. An increase in temperature increases the ignition delay time in a certain temperature range in the figure. Commercial multi component fuels contain aromatics, olefins and perhaps also oxygenates and their autoignition chemistry is different from that of paraffins [19].

Figure 2.1 Measured ignition delay times in n-Heptane iso-octane mixtures with stronger NTC behavior for higher fractions of n-heptane. Figure 17 in [18]. A fuel’s resistance to autoignition is usually described by the two octane numbers Research Octane number (RON) and Motor Octane number (MON). Practical fuels are rated by comparing their behavior to a primary reference fuel (PRF) in the RON and MON tests. A primary reference fuel is a mixture of iso-octane and n-heptane and the volume percent of iso-octane is the octane number. The RON and MON tests are carried out in a single cylinder CFR engine and the standardized procedures define engine speed, intake air temperature, ignition angle and compression ratio for the two tests. See for example Swarts et. al. [20] for further 10

Chapter 2 – Combustion in spark ignited engines information on the octane rating of fuels in the CFR engine. However, the RON and MON values alone are not sufficient to describe autoignition quality of a fuel. The engine operating conditions, which influence the temperature and pressure history experienced by the fuel, has to be accounted as well. This can be done by using the Octane Index as described in [17][21][22].

2.3.2 Modes of autoignition Pan and Sheppard [23] distinguishes three distinct combustion modes following end gas autoignition; deflagration, developing detonation and thermal explosion. These three basic modes of combustion are discussed in Glassman [9]. The temperature gradient and inhomogeneity in the vicinity of the autoignition center determines which mode is most likely to occur:

•

In the case of highly inhomogeneous end gas, autoignition might simply result in a second flame front propagating from the autoignition center, i.e. deflagration. • Completely homogeneous end gas should result in a thermal explosion, since all unburned charge would ignite at the same time. • With small end gas temperature gradients and small mixture inhomogeneities, autoignition might result in a developing detonation. In this autoignition mode the reaction front accelerates. Given sufficient time its speed can reach the local speed of sound and there will be a detonation. In engines there is never enough space and time for this to happen. Hence the term “developing” detonation. Nevertheless in this mode very high pressures can be generated. The pressure wave from an initial deflagrative autoignition center might trigger autoignition elsewhere in the end gas leading to secondary autoignition. As described above, low temperature chemistry preceding ignition has been shown to have a great influence on autoignition. Higher end gas temperatures promote the low temperature chemistry which increases the probability of initial deflagrative autoignition transforming into developing detonation. Simulations by Pan and Sheppard [23] indicate that an initial deflagrative autoignition at conditions with high mean end gas temperature is likely to result in secondary autoignition centers developing into detonation even under highly inhomogeneous conditions. This is explained by the low temperature chemistry preceding autoignition. A developing detonation is potentially much more harmful to the engine than deflagration or thermal explosion. The pressure wave amplitudes associated with a developing detonation are high. Conditions in practical SI engines are always 11

Empirical Combustion Modeling in SI Engines inhomogeneous. The charge is cooled close to the wall or heated by hot spots, e.g. soot particles.

2.3.3 Combustion chamber oscillation modes Knock oscillation frequencies have been shown to correspond to the natural frequencies of the combustion chamber, for example in Brunt et. al. [24] and Bengisu [25]. The engines used in the present work were equipped with a pentroof shaped combustion chamber which can be approximated by a cylinder in an attempt to calculate the natural frequencies of the cylinder. The solution to the wave equation:

⎧ ∂ 2u 2 ⎪⎪ 2 − c ∆u = 0 ∂t ⎨ ⎪ ∂u = 0 on the boundary ⎪⎩ ∂n

(2.6)

in a cylindrical model of the combustion chamber with cylindrical coordinates (r, θ, z) has the eigenfunctions (possible vibration modes):

⎛ r ⎞ z⎞ ⎛ ⎟⎟ ⋅ cos(mθ ) ⋅ cos⎜ kπ ⎟ ⋅ sin (2πf m,n, k t ) (2.7) u m,n, k (r , θ , z , t ) = J m ⎜⎜ α m, n B 2⎠ ⎝ h⎠ ⎝ for the vibrational modes m, k = 0, 1, … and n = 1, 2, … where m is the circumferential mode number, n is the radial mode number and k is the longitudinal mode number. Jm denotes the Bessel function of the first kind, αm,n are the zeros of the first derivative of Jm to satisfy the boundary conditions, B is the bore, h is the instantaneous model cylinder height and fm,n,k are the natural frequencies:

f m,n, k

c = 2π

⎛ α m,n ⎜⎜ ⎝B2

2

⎞ kπ ⎞ ⎟⎟ + ⎛⎜ ⎟ ⎝ h ⎠ ⎠

2

(2.8)

The longitudinal modes are often not considered since the natural frequencies of these modes are high close to top dead center, where knock usually occurs. Equation (2.8) then reduces to:

f m,n =

c α m, n πB

(2.9)

For m > 0, the eigenvalues are double with cos(mθ) replaced by sin(mθ) in the corresponding eigenfunctions.

12

Chapter 2 – Combustion in spark ignited engines The speed of sound (transverse propagating wave) in an ideal gas is:

c=

γRT

(2.10)

M

where γ is the ratio of specific heats, R is the universal gas constant, T is temperature and M is molar mass. Since the speed of sound is proportional to the square root of absolute temperature, the natural frequencies of the combustion chamber will decrease during the expansion stroke when temperature decreases. Table 2.2 gives natural frequencies and shows eigenfunctions for the six modes with lowest natural frequency for the engines used in this work with a bore of 86 mm. Speed of sound was estimated to 950 m/s which correspond to a temperature of approximately 2500 K. It should be observed that all modes except the first radial mode have a nodal line in the combustion chamber center. This is important when choosing cylinder pressure transducer position for the purpose of knock detection and analysis. Table 2.2 Predicted vibration modes with natural frequencies below 20 kHz for a cylinder with 86 mm diameter and speed of sound c = 950 m/s. Vibration 1st circumferential 2nd circumferential 1st radial mode m = 1, n = 0 m = 2, n = 0 m = 0, n = 1

α m,n

1,84

3,05

3,83

f m,n [kHz]

6,47

10,7

13,5

3rd circumferential m = 3, n = 0

4th circumferential m = 4, n = 0

1st combined m = 1, n = 1

α m,n

4,20

5,32

5,33

f m,n [kHz]

14,8

18,7

18,75

um,n

Vibration mode

um,n

13

Empirical Combustion Modeling in SI Engines This approximation of the acoustic vibration in the combustion chamber has been shown to give accurate predictions of natural frequencies of pentroof shaped combustion chambers. The first circumferential mode was in good agreement with measured data and the second and third circumferential mode predictions were 10 % higher than measurements in Brunt [24]. Bengisu [25] used a FEM model of a pentroof combustion chamber to predict the natural frequencies and shows that the nodal lines of the circumferential modes are aligned to or perpendicular to the pentroof symmetry axis.

2.3.4 Measures of knock Several definition exist for knock intensity based on either measured cylinder pressure or calculated heat release. Worret et. al. [26] has explored different techniques for detection of knock onset and knock intensity. They conclude that knock intensity should be based on signal energy of the high pass filtered pressure signal or heat release signal. A measure of the signal energy is obtained by integrating the squared high pass filtered signal over a short interval after knock onset. They also conclude that knock onset determined as the point where the signal exceeds a threshold value generally gives too late identified knock onset. A new knock detection algorithm based on high pass filtered heat release is described. Knock onset determined from the cylinder pressure signal can not be more accurate than the propagation time of a pressure wave from the knock center to the pressure transducer. Assuming that the knock center is at a distance of half the bore from the pressure transducer and that the speed of sound is 950 m/s this propagation time can be calculated. With 86 mm bore, the maximum propagation time from the knock center to a centrally located pressure transducer is 0,5 CA at 1000 rpm and 2,7 CA at 5000 rpm.

2.4 COMBUSTION SIMULATION Several approaches to combustion simulations are used in one dimensional simulation software. The simulation software GT-Power provides three predefined ways of modeling SI engine combustion. First of all, a measured combustion profile can be imposed. This is useful when measured data exists. The second approach is to define the combustion profile by the Wiebe function, which will be described in more detail below. The third approach is a turbulent flame model which also uses a model for in cylinder turbulence to estimate flame propagation. Furthermore, a user defined combustion model can be implemented or the one dimensional simulation can be coupled to three dimensional simulation software. The turbulent flame model has the 14

Chapter 2 – Combustion in spark ignited engines highest degree of physicality among the three predefined SI combustion models but requires measured or estimated swirl and tumble coefficients and has several multipliers for calibrating the simulated combustion to measured data. The simpler approaches, measured combustion profile and Wiebe function, rely on measured data but can be very useful in the absence of a physical model. Three dimensional calculation of in cylinder flow and chemical reactions is not practical today, in part because of computer execution time and because of the complex flow field and complex chemical kinetics in the cylinder. State of the art chemical kinetics codes can predict the oxidation of single component fuels, but the research has not yet reached full insight when it comes to practical fuels. It is important to note how the combustion profile, taken from measured data or calculated by the Wiebe function, is handled in GT-Power. The combustion profile in GT-Power defines the rate at which the charge enters a set of chemical equilibrium equations. The equilibrium composition changes with temperature and mixture strength, which causes the heat release rate to lag the burn rate in a GTPower simulation [4]. A simple example is shown in Figure 2.2 below. Wiebe parameters fitted to experimental data were used as input to a GT-Power simulation. The GT-Power burn rate in the figure is identical to the input Wiebe function, except for the scaling which is due to a combustion efficiency set below 1 in the simulation. Comparison of the 50 % heat released point and 10-90 % combustion duration of the input and output heat release is given in Table 2.3. As seen in the table it is necessary to adjust the Wiebe parameters to some extent before simulation since the Wiebe parameters are interpreted as burn rate in GT-Power.

15

Empirical Combustion Modeling in SI Engines

Normalized heat release and burn rate

1.0 0.8 0.6 0.4 0.2

Measured heat release Fitted Wiebe GT-Power burn rate GT-Power heat release

spark timing

0.0 -15

0

15

30

45

60

Crank angle [aTDC]

Figure 2.2 Measured heat release and Wiebe function fitted to measured data used as input to GT-Power together with GT-Power cumulative burn rate and heat release. Table 2.3 Wiebe combustion parameters for measured simulation model input heat release and simulation model output heat release from GT-Power. The difference is due to the interpretation of input combustion profile as burn rate as described above. 50 % heat 10-90 % combustion Wiebe released duration parameter m [aTDC] [CA] Input/measured heat 22,8 22,2 3,97 release Simulation model output 24,8 23,9 3,64 heat release Difference 2,0 1,7 -0,33

16

Chapter 2 – Combustion in spark ignited engines

2.4.1 The Wiebe function One way of specifying the combustion rate in a two-zone combustion model is the Wiebe function [27]. The Wiebe function is commonly used in SI engine simulation. The functional form:

⎡ ⎛ θ − θ 0 ⎞ m +1 ⎤ xb (θ ) = 1 − exp ⎢− a⎜ ⎟ ⎥ ⎢⎣ ⎝ ∆θ ⎠ ⎥⎦

(2.11)

is used to describe the fraction of fuel burnt xb based on considerations of chain reactions in general. θ is the crank angle, θ0 is the start of combustion and ∆θ is the total combustion duration. The parameter m is called the combustion mode parameter and defines the shape of the combustion profile. m was introduced by Wiebe to describe the time dependence of concentration of reaction centers by the function:

ρ = kt m

(2.12)

where k is a constant. In a spherically expanding flame with constant flame speed one would expect m to be 3. Accelerating flame speed should give higher values and vice versa. Wiebe found m to be in the range 2-4 for SI engines. The value of the constant a in Equation 5 follows from the chosen definition of end of combustion. With the mass fraction burned xb,EOC = 99,9% at the end of combustion, a has the value:

a = − ln (1 − x b , EOC ) = − ln 0,001 = 6,90

(2.13)

2.4.2 Knock simulation The knock simulation method used in this work and in Paper II is based on the Livengood-Wu knock integral [28]: tk

1= ∫ 0

dt

τ

(2.14)

where τ is the ignition delay time as a function of temperature and pressure and tk is the time of autoignition. The basic idea behind the Livengood-Wu knock integral is best explained by considering a very simple system with constant autoignition delay time τc at a given pressure and temperature. If the system is exposed to this pressure and temperature for the time τc, it is expected to ignite. The value of the integral in Equation (2.14) at the autoignition instant would be unity. It is assumed that this reasoning holds also for a more complex system where pressure, temperature, composition and ignition delay time vary with time. The ignition delay time in this 17

Empirical Combustion Modeling in SI Engines case is an aggregate ignition delay time for completion of the entire autoignition mechanism as described previously. The instantaneous value of the integral is a measure of the fraction of pre-autoignition reactions that have been completed. It is a way of accounting for the pressure and temperature history of the unburned charge. It is clear from the description of autoignition chemistry above that the pressure and temperature history of the charge determines the current state of the charge and influences the instantaneous ignition delay time for a given mixture. Pressure and temperature history determines to what extent the low temperature chemistry has been completed and also the concentration of the critical H2O2 species. Nevertheless, an ignition delay time with only temperature and pressure dependence has been used in Equation (2.10) by several authors, e.g. Douaud and Eyzat [29]. The functional form used for this aggregate ignition delay time is:

⎛B⎞ (2.15) ⎝T ⎠ The functional form is similar to the Arrhenius expression for chemical reaction rate with a pressure dependence added. The constants in Equation (2.15) have been fitted to several fuels from rapid compression machine test data as well as from engine test data. Values of the constants from several references are summarized in Table 2.4.

τ = Ap − n exp⎜ ⎟

Table 2.4 Reported values for the constants in Equation (2.15) from several authors for temperature in K and pressure in bar. The value of the constant A has been recalculated to metric units in some cases. A n B Fuel [s.barn] [K] Reference PRF 95

1.62e-2

1,7

3800 Douaud, Eyzat [29]

PRF 100

1,87e-2

1,7

3800 Douaud, Eyzat [29]

Commercial RON 93, MON 82 1.02e-4

1,01

6220 Douaud, Eyzat [29]

PRF100 (isooctane)

1.68e-2

1,49

7457 By et.al. [30]

Gasoline/Ethanol, RON95

7.59e-3 1.325 3296 Current study

Measured ignition delay times for the reference gasoline RD387 and a surrogate mixture with similar ignition delay behavior as gasoline from Gauthier et. al. [31] are shown in Figure 2.3. The pressure exponential n was found to be 1,64 for n-Heptane and 1,01 for the reference gasoline. The solid markers are ignition delay without residual gases at λ = 1. Other markers are at various lean, stoichiometric and rich mixtures with or without EGR. λ and EGR seem to affect ignition delay time. 18

Chapter 2 – Combustion in spark ignited engines Gauthier et. al. [31] concludes that richer mixture gives shorter ignition delay at higher pressures and lean mixture gives longer ignition delay. Increased EGR content increases ignition delay. The test data in Gauthier et. al. [31] does not include the region where NTC behavior is expected, compare with Figure 2.1, but it is clear that temperature dependence decreases at lower temperature. Temperature [K]

Ignition delay at 5 MPa [ms]

1250

1111

Douaud, Eyzat 1 (1978) PRF 87

1000

909

833

current study

Douaud, Eyzat (1978) gasoline RON 92/MON 83 0.1 λ = 0,5 to 2, 0 to 30% EGR

Gasoline Surrogate A

λ = 1, no EGR

0.01 0.8

0.9

1.0

1.1

1.2

-1

1000/T [K ]

Figure 2.3 Measured ignition delay time from shock tube experiments for a reference gasoline RD387 with (RON + MON)/2 = 87 and a surrogate mixture as reported in Gauthier et. al. [31]. Solid markers are λ = 1 and no EGR. The lines are estimated aggregate ignition delay time according to Equation (2.12) calibrated by engine tests with constants reported in Table 2.4. Ignition delay time in Equation (2.15) is an attempt to fit a linear curve to represent the data in Figure 2.3 in the region where the engine operates, i.e. at the temperatures and pressures of the end gas at knocking conditions. Figure 2.4 shows the temperature and pressure history from spark timing to detected knock from tests used to calibrate the constants of Equation (2.15) in this work. This data is in the region where the fuel is expected to have low or negative temperature dependence. The estimated ignition delay time at 5 MPa from the calibration in this work as well as from Douaud and Eyzat [29] is also shown as lines in Figure 2.3. The estimate from this work fits the shock tube test data well in the relevant region at low temperatures but can not be expected to predict ignition delay at higher temperatures. 19

Empirical Combustion Modeling in SI Engines It is also noticeable from the pressure and temperature data in Figure 2.5 that knock occurs at temperatures slightly above 900 K which is a critical temperature for the decomposition of H2O2 as described earlier. Also drawn in Figure 2.5 is an isentropic compression line leading to one of the knocking cycles calculated with γ = 1,25 which shows that the operating conditions for these knocking cycles were quite similar, i.e. they are close to the same isentrop. Unburned zone temperature [K] 909

833

769

1.1

1.2

1.3

714

667

625

1.4

1.5

1.6

Cylinder pressure [MPa]

10

8

6

4

2

-1

1000/T [K ]

Figure 2.4 Pressure and temperature history for several knocking cycles at different operating conditions from approximately 30 bTDC to knock onset.

20

Cylinder pressure at knock [MPa]

Chapter 2 – Combustion in spark ignited engines 10.5 10.0 9.5 9.0 8.5 8.0 7.5 890

900

910

920

930

Unburned zone temperature at knock [K]

940

Figure 2.5 Unburned zone temperature and pressure at knock for several knocking cycles at different operating conditions. The dash-dotted line is an isentrop drawn from one of the knocking cycles calculated with γ = 1,25. In a recent work by Yates et. al. [32] an attempt has been made to model the different regions of the ignition delay times by one Arrhenius type expression according to Equation 2.15 for each of the three regions: low temperature region, NTC-region and high temperature region. The total ignition delay time is formed by the expression:

[

τ = (τ 1 + τ 2 )−1 + τ 3−1

]

−1

(2.16)

with values for the constants for a model gasoline found in Table 2.5. The resulting ignition delay surface is shown in Figure 2.6 with ignition delay histories for three knocking cycles. It is evident from the figure that the knocking cycles just barely enter the high temperature region for this test data, which explains why the single stage ignition delay model shown in Figure 2.3 works well. Yates et. al. [32] also show that fuel/air ratio can be modeled by the relationship:

τ (λ ) = τ λ =1λ β where β ≈ 0,67, identified from Figure A.1 in Yates et. al. [32].

21

(2.17)

Empirical Combustion Modeling in SI Engines Table 2.5 Constants for a model gasoline in three part ignition delay model from Yates et. al. [32]. B ln(A) n τ1, low temperature -19,7 -0,101 16196 11,33 -1,623 -3136 τ2, NTC-region τ3, high temperature -11,02 -0,949 15250

Figure 2.6 Ignition delay surface from three part ignition delay model from Yates et. al. [32] for a model gasoline with ignition delay trajectories for 3 knocking cycles.

22

Chapter 3 EXPERIMENTAL METHOD This chapter contains a summary of the measurement system used in the experimental part of this work along with estimated errors in the measurements in order to give a general idea of the measurement accuracy. A few paragraphs are devoted to signal processing, which plays a key role in obtaining high quality measurement data. Finally, the engines used in the experiments are described, including a brief overview of the Divided Exhuast Period system.

3.1 MEASUREMENTS The measurements conducted within this project had several key purposes. One of the purposes was to be able to calibrate a simulation model of the Divided Exhaust Period engine. The second purpose was to gain further understanding of how the Divided Exhaust Period engine responds to different changes in operating conditions and exhaust system geometries. Furthermore, measurements were used as a tool to create and calibrate empirical models for knock and combustion in SI engines, as described in Paper II and Paper III.

3.1.1 Measurement system The test bed control and measurement system used was Cell4, developed by Professor Hans-Erik Ångström. The system is very flexible and accepts analogue and digital input signals which can be measured either as high frequency crank angle resolved data or low frequency time resolved data. Slow measurements are accomplished mainly through Nudam data acquisition modules [33] which accept voltage or 23

Empirical Combustion Modeling in SI Engines thermocouple input depending on model. A sampling frequency of approximately 1 Hz was possible for these measurement channels. Crank angle resolved measurements were accomplished by a 12-bit PowerDAQ A/D card [34] with 1 MHz total sampling frequency divided over a maximum of 16 input channels in the current setup. PIC microcontrollers measured digital input, such as crank angle encoder pulses and turbo speed signal, and produced control signals for test bed and engine control. The PIC:s additionally produced single engine revolution averages for analogue inputs, accomplished by buffering transducer input from external 12-bit A/D converters at 10 kHz and averaging the buffered data once every engine revolutions. Several custom built signal amplifying units were used to drive transducers in the system and condition transducer output signals.

3.1.2 Pressure measurement Pressure was measured at several positions on the engine for the purpose of calibrating the simulation model. On the intake side of the engine, pressures were measured upstream and downstream of each major component, such as compressor, charge air cooler and throttle. Flush mounted GEMS steel diaphragm gauge pressure transducers [35] with 4 bar range were used for these measurements. In cylinder pressures were measured in all four cylinders with near flush mounted AVL GM12D uncooled miniature piezo-elecric transducers [36] and Kistler 5011 charge amplifiers [37]. The transducers have M5 dimensions, which was the largest that could be fitted in the Divided Exhaust Period cylinder head. Kistler 4045A10 piezoresistive pressure transducers [37] were used in the exhaust manifold before and after the turbine, primarily due to the availability of suitable cooling adapters, less thermal sensitivity and higher natural frequency. GEMS transducers with cooling adapters were used at several other positions in the exhaust manifolds. Static calibration of the low pressure transducers, strain gauge and piezoelectric, was accomplished by a traceably calibrated Druck DPI 705 pressure indicator with a hand pump [38]. The entire measurement chain was calibrated at several occasions and the resulting error ranged from ±0,5 - 1 % FS for the 4 bar GEMS transducers and below ±0,08 % FS for the 10 bar Kistler transducers which translates to absolute uncertainty in static measurements of ±2 - 4 kPa for the GEMS transducers and 0,8 kPa for the Kistler transducers. A dead weight tester, Ametek Hydralite [39], was used for cylinder pressure transducer calibration in the range 0 - 10 MPa. Including the uncertainty for the dead weight tester and A/D conversions, the linearity error for the measurement chain was found to be below ±0,4 % FS or 40 kPa. The cyclic temperature shift according to the 24

Chapter 3 – Experimental method transducer manufacturer is < ±60 kPa for the GM12D cylinder pressure transducers. Cyclic temperature shift, i.e. thermal shock, typically leads to too low measured pressure during combustion and during the following expansion stroke [40]. Lee et. al. [41] have quantified the effects of thermal shock in an uncooled transducer with similar properties as the ones used in this work and found that thermal shock persisted through the exhaust stroke, ultimately affecting measured IMEP with up to -4 %. One drawback with using the dead weight pressure tester at ambient temperature, which was the case here, is that the sensitivity of the uncooled GM12D transducers at ambient temperature might be different from the sensitivity at operating temperatures in the engine cylinders. The manufacturer states the thermal sensitivity shift to ±2 % in the temperature range 20 - 400° C. Both cyclic temperature shift and thermal sensitivity shift can be kept lower for cooled transducers.

3.1.3 Temperature measurement Shielded 3 mm type K thermocouples were used for most temperature measurements. Cold junction correction was accomplished in the Nudam 6018 data acquisition modules. Thermocouples measure the temperature of the probe tip, which is not necessarily equal to the temperature of the surrounding gas or liquid. Heat transfer along the stem of the thermometer and radiation to pipe walls decreases the measured temperature in the case of hot fluid in a cooler pipe, which is typically the case in an engine exhaust manifold. The fluid flow rate also affects the heat transfer to the thermocouple. Long insertion lengths were used to minimize errors from conductive heat transfer. [42][43] The response time for a 3 mm thermocouples is several seconds. Hence, the thermocouple signal is some kind of average temperature in typical engine conditions with highly pulsating temperature. According to an investigation comparing measurements and 1-D simulation of the gas temperature in the exhaust manifold of a turbocharged engine in Westin [5], a 3 mm thermocouple with 100 mm insertion length measures a temperature close to the mass averaged temperature of the gas. The accuracy of class 1 type K thermocouples is the larger of 1.5° and 0,004 times the measured temperature. When testing the linearity of some thermocouples in a IsoTech HTQuickCal block calibrator [44], the errors for the measurement chain were found to be within the stated accuracy of the thermocouples. Surface temperature of the aluminium inlet manifold was measured for simulation model calibration with a Testo Quicktemp 860-T3 infrared pyrometer [45]. A pyrometer measures the radiation from an object which depends heavily on the

25

Empirical Combustion Modeling in SI Engines emissivity of the material. The emissivity is a measure of how close to a black body radiator the material is and is a number between zero and unity. Many metals and aluminium in particular has low emissivity. Aluminium has emissivity in the range of about 0,05 to 0,2 depending on oxidation and alloy [46]. A small absolute error in estimated emissivity will give large errors in measured temperature with this low emissivity. Therefore, a small area on the manifold was painted with matte black paint which should have an emissivity around 0,9.

26

Chapter 3 – Experimental method

3.1.4 Other measurements Turbo speed was measured with a Micro-Epsilon eddy current probe [47] mounted in the compressor housing. The probe senses the blade passages of the impeller and the accompanying signal processing unit converts the signal to one digital pulse per completed turbo revolution, which can be up to some 20 crank angles apart depending on engine and turbo speed. The time between pulses is converted to turbo speed and the timestamp of each pulse gives the corresponding crank angle. Any disturbances on the transducer signal might be identified as a blade passage. When this happens, the data analysis system will identify a too high turbo speed. The digital measurement data evaluation algorithms currently used does not handle these errors which, after interpolation to the crank angle basis of the other measurements, gives quite bad results as seen in Figure 3.1.

Turbo speed [1000 x rpm]

91 measured data points interpolated data

90 89 88 87

single cycle (+1000 rpm)

86 85 -180

-90

0

90

average

corrected average

error in average 180

270

360

450

540

Crank angle [aTDC]

Figure 3.1 Error in measured turbo speed from disturbance detected as impeller blade passage make large difference in the averaged data. Single cycle data (top) has been shifted up 1000 rpm.

27

Empirical Combustion Modeling in SI Engines Lambda was measured with an ECM AFRecorder 2000A with the stated accuray ±0.008 for 0,8 < λ < 1,2 [48]. Fuel mass flow was measured by weighing the fuel in a small reservoir with approximately 2,5 dm3 volume. Calibration of the ASE scales was performed by applying known weights to the scales. A measurements accuracy of < ±0,1 % FS was obtained in static calibration, but the measured fuel flow varied significantly over an emptying cycle of the reservoir with the engine running in steady state. Typically, the measured fuel mass flow would decrease during each emptying cycle as in the example in Figure 3.2. This highlights the difference between static and dynamic calibration. The fuel flow measuring system behaved perfectly in static condition, but quite poorly in dynamic conditions. Dynamic calibration has not been performed for any of the measuring systems involved in this work. 1650

6.6

mass

1550

6.4

1450

6.3

1350 mass flow

6.2

1250

6.1 0

20

Fuel mass [g]

Fuel mass flow [g/s]

6.5

40

60

1150 80

Time [s]

Figure 3.2 Measured fuel flow and the fuel mass in the scales during steady state engine operation.

28

Chapter 3 – Experimental method

3.2 DATA ACQUISITION As already mentioned, a 12-bit A/D converter was used for crank angle resolved measurements in the Cell4 measurement system. The input range for the A/D converter was -10 V to +10 V, which was also the output range of the charge amplifiers. The charge amplifiers was set to the physical range 1 V/MPa to be able to measure 10 MPa peak pressure in this set-up. This leads to a quite coarse resolution in the measured cylinder pressure during the gas exchange process, as shown in Figure 3.3. One A/D bit corresponds to 4,88 kPa with these cylinder pressure measurement settings, or 0,049 % FS. The measuring uncertainty introduced by the A/D conversion is small compared to the other error sources described above.

Cylinder pressure [kPa]

180

160

140

120

100 360

405

450

495

540

Crank angle [aTDC]

Figure 3.3 Cylinder pressure from a single cycle during intake stroke sampled with 12-bit A/D converter. A 1,5 kHz FIR low pass filter has been applied to the data. (3000 rpm, 1,49 MPa imep)

29

Empirical Combustion Modeling in SI Engines

3.2.1 Signal conditioning Signal filtering is important to obtain high quality measurements. Anti-aliasing low pass filters should be applied to the analog measurement signal before sampling. The cut-off frequency has to be below the Nyquist frequency, i.e. half the sampling frequency. Transducer natural frequency should also be considered in the choice of cut-off frequency for the anti-aliasing filter. The data displayed in Figure 3.3 was sampled at 45 kHz and the built in 30 kHz filter in the charge amplifier was used to limit aliasing. This means that frequencies between 15 and 22,5 kHz in the sampled signal also contain aliased signals from the 22,5 to 30 kHz range in the real signal. A more detailed study was made of the frequency content of the cylinder pressure signals from one of the engines used in this work. The frequency content was examined over the engine speed range with non-knocking combustion at high load. A discrete Fourier transform of the measured signal from many consecutive cycles shows peaks at every half engine revolution frequency, as should be expected from a four stroke engine with combustion every second revolution. The frequency content along with the envelope of the peak amplitudes is shown in Figure 3.4.

Cylinder pressure [kPa]

1000

Peak amplitude envelope

100

10

1

0

2

4

6

8

10

Frequency [1/revolution]

Figure 3.4 Frequency content for several consecutive cycles at 1000 rpm and the peak amplitude envelope.

30

Chapter 3 – Experimental method The amplitudes of the peaks decrease at higher frequencies. Figure 3.5 shows the peak frequency envelope for several engine speeds. The peak amplitudes on a per engine revolution basis look very similar for all engine speeds and the peaks become buried in noise above 30 times the engine revolution frequency, which leads to the recommendation to low-pass filter the data with the cut-off frequency:

f LP = 30 ⋅

N N = [Hz] 60 2

(3.1)

with the engine speed N given in rpm.

Cylinder pressure [kPa]

1000

1000 rpm 2000 rpm 3000 rpm 4000 rpm 5000 rpm

100

10

suggested cut-off frequency

1

0.1 0

5

10

15

20

25

30

35

40

Frequency [1/revolution]

Figure 3.5 Cylinder pressure frequency envelope per engine revolution at operating conditions from 1000 rpm to 5000 rpm at high load. Several consecutive cycles at steady state operation was used in the analysis. Figure 3.6 shows the cylinder pressure frequency content again, with the useful frequency range according to Equation (3.1) marked. It is noticeable from Figure 3.6 that the natural frequencies of the combustion chamber, see Chapter 2.3.3, show up at high engine speed, although there was very few knocking cycles in the data set. Low amplitude combustion chamber pressure oscillations seem to be the result of the rapid pressure rise with respect to time during combustion at high engine speed.

31

Pressure [kPa]

Empirical Combustion Modeling in SI Engines 100

Cutoff frequency 0,75 kHz

0.01

Pressure [kPa]

0 100

2

4

6

8

10

1,5 kHz

3000 rpm 1 0.01 0

Pressure [kPa]

1500 rpm

1

100

5

10

2.25 kHz

15

20

4500 rpm

1 0.01 0

5

10

15

20

Frequency [kHz]

Figure 3.6 Cylinder pressure frequency content analyzed from several consecutive cycles at steady state operation and suggested cut-off frequencies at 30 times the engine rotational frequency.

3.2.2 FIR low pass filter A Finite Impulse Response (FIR) filter was implemented to filter the sampled pressure data. The filter consists of a windowed sinc function, Equation 3.5, which is convoluted with the data. The idea behind using convolution with the sinc function is that multiplication with a transfer function in the frequency domain corresponds to convolution with the impulse response in the time domain. In the frequency domain, an ideal low pass filter transfer function has zero amplitude for all frequencies above the cut-off frequency fLP and amplitude one with zero phase shift for frequencies below the cut-off frequency. The impulse response of this step-like function in terms of the normalized frequency q0

q0 =

f LP fs

is its inverse Fourier transform: 32

(3.2)

Chapter 3 – Experimental method

1 h(k ) = 2π 1 = 2π

0,5

∫ H (2πq )e

i 2πqk

dq = [H (2πq ), q > q0 ] =

− 0,5 q0

∫e

− q0

i 2πqk

(3.3)

1 dq = sin (2πq0 k ), k = −∞..∞ πk

which is an infinite sequence. The impulse response of the ideal low pass filter is usually truncated with some kind of window function to make computation possible and to limit the computation time. The Hanning window was chosen in this work. A Hanning window with width 2M + 1 centered around k = 0 is given by:

⎛ k−M ⎞ w(k ) = 0,5 − 0,5 cos⎜ π ⋅ ⎟, k = − M ..M M ⎠ ⎝ By introducing the sinc function:

(3.4)

⎧ sin (πk ) ⎪ , k ≠0 sinc(k ) = ⎨ πk ⎪⎩ 1 , k =0 Equation 3.3 can be simplified and the final normalized filter kernel is: ⎡ ⎛ k−M 2q 0 ⋅ sinc(2q 0 k ) ⋅ ⎢0,5 − 0,5 cos⎜ π ⋅ M ⎝ ⎣ g (k ) = h(k ) ⋅ w(k ) = ∑ g (k )

⎞⎤ ⎟⎥ ⎠⎦

(3.5)

, k = − M ..M

(3.6) The filter length determines the steepness of the filter. The length of the filter was chosen to get a filter kernel with Kf periods of the sinc function by setting the half filter length M according to: 2πq 0 M = 2πK f ⇒ M =

Kf q0

(3.7)

Kf = 3 was used for most of the filtering in this work. A higher Kf, i.e. a longer filter kernel, gives steeper filter characteristics at the cut-off frequency but also more pronounced non-causal behaviour which shows up as ringing in the filtered signal prior to steep changes in the raw signal. This can be undesired in for example band pass filtering for knock onset detection, as described further below. Kf can be explained as the number of oscillation periods in the filtered signal before and after a step in the input signal. Figure 3.7 shows an example of cylinder pressure data filtered with the described windowed sinc filter. The suggested cut-off frequency from section 3.2.1 is 33

Empirical Combustion Modeling in SI Engines

Cylinder pressure [MPa]

1,5 kHz, which seems to preserve the characteristics of the data. A lower cut-off frequency removes significant frequency components and produces lower pressure rise rates and lower peak pressure, as seen in the figure. A higher cut-off frequency will attenuate less noise. 1,5 kHz filter

A/D reading

5.2

5.0

1,0 kHz filter

4.8

4.6 25

30

35

Crank angle [aTDC]

Figure 3.7 Cylinder pressure filtered with zero phase shift FIR-filter with 1,0 kHz and 1,5 kHz cut-off frequency. (3000 rpm, 1,49 MPa imep) Convolution of a data sequence of length n with a filter kernel of length 2M + 1 results in a filtered sequence of length n + 2M. 2M points at each end of the data sequence contain data where the original data and the filter do not overlap completely in the convolution. This causes transients in the filtered data sequence. The original data sequence was extended at each end with data from the other end of the sequence to avoid transients. Since the engine pressure data is of periodic nature and all measurements were made at steady state this should be an adequate method of extending the data sequence. A more elaborate method is to mirror, inverse and shift the data set in the end points according to Equation (3.8), which eliminates possible discontinuities due to non-steady state data or transducer drift.

, − M ≤ k ≤ −1 ⎧ 2 p(0 ) − p(− k ) ⎪ , 0 ≤ k ≤ n −1 p extended (k ) = ⎨ p(k ) ⎪2 p(n − 1) − p(2n − k ) , n ≤ k ≤ n + M − 1 ⎩

(3.8)

This extended data sequence can be convoluted with the filter kernel and the 2M points at each end can be discarded. 34

Chapter 3 – Experimental method

3.2.3 FIR band pass filter The frequency content in the cylinder pressure signal changes significantly during knocking combustion. As described above, autoignition of the end gas induces pressure oscillations in the combustion chamber with frequencies corresponding to the natural frequencies of the combustion chamber. Knock amplitude and knock onset had to be determined in the work with the knock model in Paper II. For this purpose, the acquired cylinder pressure data was band pass filtered with the pass band located around the lowest natural frequency of the combustion chamber. A FIR bandpass filter can be designed as the difference between a high pass filter and a low pass filter as described in the previous section. The resulting windowed filter kernel with passband between q0 and q1 is: ⎡ ⎛ k − M ⎞⎤ 2 ⋅ (q1 ⋅ sinc(2q1 k ) − q 0 ⋅ sinc(2q 0 k )) ⋅ ⎢0,5 − 0,5 cos⎜ π ⋅ ⎟ M ⎠⎥⎦ ⎝ ⎣ , k = − M ..M g BP (k ) = ∑ g BP (k )

(3.9) One major drawback with using this type of filter is the non-causal properties of the filter. This will introduce ripple in the signal prior to the actual knock onset. Using a short filter decrease the ripple but the filter steepness also decreases. A FIR filter according to Equation (3.9) is good for evaluating knock level, but not for detection of knock onset.

3.2.4 IIR filtering for detection of knock onset A causal filter will not give any output before the actual knock onset, which should be suitable for knock onset detection. Many different recursive filter structures exist of which many are digital representations of common analogue filter structures such as Bessel or Butterworth. These filters are referred to as infinite impulse respone filters (IIR) since they have an infinitely long impulse response in the time domain. IIR filter have phase distortion as opposed to the FIR filter described above with zero phase distortion. The short filter lengths of recursive IIR filters make them very computationally efficient compared to FIR-filters. With IIR forward and backward filtering the phase distortion can be avoided [49]. No evaluation of the most appropriate IIR-filter for detection of knock onset has been made in this work. A low order filter should be preferable to a higher order filter to minimize phase distortion.

35

Empirical Combustion Modeling in SI Engines

3.3 HEAT RELEASE CALCULATION Heat release was calculated from measures cylinder pressure data based on the first law of thermodynamics with a single temperature zone as described in for example Heywood [8]. Heat transfer to the combustion chamber walls were calculated according to Woschni’s correlations also summarized in Heywood [8]. Crevice effects were not accounted for. The resulting equation for calculating rate of heat release is: dQch dp γ 1 dV V = p + + Ahc (T − Tw ) γ − 1 dθ γ − 1 dθ dθ

(3.10)

where Qch is the chemical energy released from the fuel. Calculation of the mixture and temperature dependent ratio of specific heats in the mixture is described in detail in section 3.3.1 below. T and Tw is the gas and wall temperatures and hc is the heat transfer coefficient given by:

hc = 3,26 ⋅ B −0, 2 p 0,8T −0,55 w 0,8

(3.11)

where B is the bore, p is cylinder pressure, T is average gas temperature and w is the average gas velocity in the cylinder, approximated by:

w = C1 S p + C 2

Vd Tref p ref Vref

( p − pm )

(3.12)

The first term in Equation (1.12) describes the gas velocity from the piston motion and swirl and the second term describes the rise in average gas velocity due to combustion. pm is the motored pressure. The charge temperature and hence the trapped mass was determined at inlet valve closing (IVC) as the weighted average of intake temperature and exhaust temperature at inlet conditions: * TIVC = (1 − x r ) ⋅ Tinlet + x r ⋅ Texh

(3.13)

Residual gas content xr was obtained from simulations. T*exh was calculated by assuming isentropic expansion or compression from exhaust pressure and temperature to inlet pressure: * Texh

⎛ p = Texh ⎜⎜ exh ⎝ p inlet

⎞ ⎟⎟ ⎠

γ exh −1 γ exh

(3.14)

with the average ratio of specific heats of the exhaust gases evaluated at inlet and exhaust temperatures.

36

Chapter 3 – Experimental method Charge temperature was calculated from measured pressure and cylinder volume with the ideal gas law:

T=

pV pV = (mR )IVC p IVC ⋅ VIVC TIVC

(3.15)

Motored pressure pm was calculated from inlet valve closing to exhaust valve opening by assuming isentropic compression and expansion of the charge, with ratio of specific heats evaluated for unburned mixture at each time step. Motored pressure was used in the calculation of heat transfer and the calculation of pressure ratio as described below.

3.3.1 Thermodynamic properties of mixture One key parameter for accurate estimation of heat release from pressure data is the ratio of specific heats [50][51], which is the heat capacity at constant pressure divided by the heat capacity at constant volume, γ = cp/cv,. Thermodynamic theory state that

γ is constant for a ideal monoatomic gas whereas γ is a function of temperature only for a semi-perfect gas. The charge in the combustion chamber is a mixture of several components, each of which can be considered as a semi-perfect gas. Several approaches of modeling the thermodynamic properties of the cylinder charge have been tested with the heat release calculation algorithm described above. First of all the basic relationship between heat capacities is needed. Molar heat capacity at constant volume and constant pressure are related by:

CV = C p − R [J/mol ⋅ K]

(3.16)

where R is the gas constant with appropriate unit. Specific heat capacity is related to molar heat capacity by the molecular mass M: cp =

Cp

[J/kg ⋅ K] M CV of an ideal gas can be predicted by kinetic theory as:

(3.17)

df R (3.18) 2 where df are the degrees of freedom of motion of the molecules in the gas. For monoatomic molecules only translation energy contributes to the internal energy which gives three degrees of freedom. For diatomic gases two rotational degrees of freedom are added and at high temperatures also vibration energy. For more complex molecules such as the hydrocarbons in gasoline the vibration energy becomes more CV =

37

Empirical Combustion Modeling in SI Engines important which gives an even higher CV. A high value of CV gives a low γ. This can help in understanding the general trends γ of in the combustion chamber. During the compression stroke in a PFI engine, the cylinder charge consists of air and fuel. Air consists mainly of diatomic molecules with 5/2 degrees of freedom giving γ = 7/2 = 1,4 at low temperature. The fuel has very high CV and γ close to 1, which means that overall γ is dependent on air/fuel ratio. Residual gases contain carbon dioxide and water, which are triatomic, giving a slightly lower γ than that of pure air. Temperature rise during combustion decrease γ since vibrational energy becomes important. Figure 3.8 shows CV/R for some of the gas molecules found in the combustion chamber over the temperature range relevant to SI engines. Iso-octane is included in the figure as an indication of the heat capacity of a hydrocarbon fuel.

11/2

CV / R

9/2 7/2

CO2

60

H2O

O2

5/2

H2

3/2 1/2

40

N2

H

20

iso-octane

500

CV / R (iso-octane)

80

1000

1500

2000

2500

3000

Temperature [K] Figure 3.8 CV/R for some molecules from JANAF tables [54]. The first investigated approach of estimating γ in the combustion chamber was to consider the trapped charge as pure air, with a temperature dependent γ. A second degree polynomial for the temperature dependence of γ in air is found in Kanury [52]:

c p ,air = 917.4 + 0,2402 ⋅ T + 3,105 ⋅ 10 −5 ⋅ T 2 [J/kg ⋅ K] M air = 28.96 [g/mol]

(3.19)

The second investigated approach is the AVL model for γ [53], which has linear temperature dependence:

38

Chapter 3 – Experimental method

γ=

0,2888 +1 cV (T )

cV (T ) = 0,7 + T ⋅ (0,155 + Ai ) ⋅ 10

(3.20) −3

with the constant Ai = 0,1 for SI engines. The third investigated approach was to calculate frozen mixture equilibrium composition and use NASA polynomials [54] to calculate temperature dependent Cp for air, fuel and residual gases as described in Chapter 4 of Heywood [8].

C p (T ) =

4

∑ ai T i

(3.21)

i = −2

Residual gas content was obtained from simulations. The pressure ratio:

PR(θ ) =

p(θ ) −1 p m (θ )

(3.22)

was used to capture the transition from unburned gases to burned gases during combustion. The normalized pressure ratio weighted average of unburned gas γu and burned gas γb was used in the heat release calculations:

γ = (1 − PR norm ) ⋅ γ u (T ) + PR norm ⋅ γ b (T )

39

(3.23)

Empirical Combustion Modeling in SI Engines Ratio of specific heats and calculated heat release with the different models are shown in Figure 3.9. As seen in the figure, the fuel vapor and residual gases has a major impact on the ratio of specific heats compared to pure air. The AVL model estimate of γ is inbetween the frozen mixture model and the pure air model. Ratio of specific heats from the simulation software GTPower is also included in the figure, which is an average of the burned and unburned zone γ. GTPower predicts higher γ during combustion since dissociation at high temperatures is included in the γ-model. Dissociation introduces smaller molecules with less degrees of freedom of motion and hence higher γ. One conclusion from the tests with different γ-models is that the simple linear and quadratic models fail to capture the high temperature behavior of the charge. temperature dependent air

1.00

AVL model

1.35

0.75

1.30 frozen mixture model

0.50

1.25

0.25

GTPower simulated

1.20

0.00

-135

-90

-45

TDC

45

90

135

Crank angle [aTDC] Figure 3.9 Comparison of different models for the ratio of specific heats γ in the mixture and the resulting calculated heat release. Gas composition: λ = 0,9, gasoline/ethanol fuel and 3 % residual gas content.

40

Fuel normalized heat release

γ = cp/cv

1.40

Chapter 3 – Experimental method Figure 3.10 Shows the calculated motored pressure compared to measured cylinder pressure for the different γ-models. The frozen mixture model predicts motored pressure close to the measure pressure. Motored pressure was calculated without accounting for heat transfer to the cylinder walls, which explains why the calculated motored pressure is higher than measured pressure around top dead center. Cylinder pressure [MPa]

temperature only 3

AVL model

frozen mixture model measured pressure

2

1

-90

-60

-30

TDC

30

60

90

Crank angle [aTDC]

Figure 3.10 Calculated motored pressure with different models for the ratio of specific heats compared to the measured cylinder pressure.

41

Empirical Combustion Modeling in SI Engines

3.4 EXPERIMENT ENGINES The experimental results reported in this thesis were acquired from a modified two liters turbocharged engine. Some key properties of the engine are summarized below.

3.4.1 Divided Exhaust Period One key target for this research project was to evaluate the Divided Exhaust Period (DEP) system on a modern four cylinder turbocharged engine, see Paper I. The main principle behind the DEP system is to divide the exhaust flow from each cylinder into two different exhaust manifolds with different valve lifts for the two exhaust valves in each cylinder. Figure 3.11 shows a schematic view of the DEP engine. Intercooler

Charge air system

Exhaust blow-down system

C

T

Catalyst

Exhaust scavenging system

Trapping valve

Figure 3.11 Schematic view of the Divided Exhaust Period engine. The reason for dividing the exhaust flow into two different manifolds is to decrease exhaust back pressure during the exhaust displacement phase and to prevent interfering pulses between cylinders in four cylinder turbocharged engines. Using DEP results in decreased residual gas content which should improve knock resistance and increase volumetric efficiency. The pumping losses are also decreased. Figure 3.12 shows a comparison between mass flow over the exhaust valves in a standard turbocharged engine and a DEP engine at 5500 rpm full load operation. The energy rich blow-down pulse is fed to the turbine through the exhaust blow-down system and the remaining exhaust is evacuated through the scavenging system, with a much lower exhaust back pressure.

42

Chapter 3 – Experimental method standard t/c

DEP blow-down system

0.2

0.1

0.0

blow-down valve 90

135

DEP scavenging system

Valve lift

Mass flow rate [kg/s]

0.3

scavenging valve 180

225

270

315

360

405

Crank angle [aTDC]

Figure 3.12 Mass flow and valve lift in the DEP engine compared to a standard turbocharged engine at 5500 rpm.

3.4.2 Engine specifications The DEP engine was based on a standard 2 dm3 turbocharged engine with specification according to Table 3.1. The head was modified with separated exhaust runners. The short valve lift durations in the DEP engine puts special demands on the exhaust valves. Exhaust valve diameter was increased from 28 mm to 32 mm to reduce choking in the exhaust ports. Lightweight sodium cooled exhaust valves were used to maximize possible valve lift. Several exhaust and intake camshafts were used in the investigations. Continuously variable camshaft phasing with an adjustment range of 50 CA was used on both intake and exhaust camshafts. A standard head and standard exhaust manifold was also used on the engine in the knock tests reported in Paper I.

43

Empirical Combustion Modeling in SI Engines Table 3.1 Specifications of the test engines. Bore 86 mm Stroke

86 mm

Compression ratio

9.5

No. of cylinders

4

Total displacement

2.0 dm3

Inlet valve diameter

32 mm

Exhaust valve diameter, DEP head

32 mm

Exhaust valve diameter, standard head

28 mm

Cylinder head

4-valve pentroof with central spark plug

Fuel system

port fuel injection

Fuel

95 RON, up to 5% ethanol

Cooled EGR The DEP engine was equipped with a cooled EGR system with a Valeo water/air EGR cooler. The EGR was extracted in the exhaust blow-down system, before the turbine inlet. EGR was inserted upstream of the throttle to assure good mixing with inlet air and mixing balance between cylinders. A butterfly valve was used to control cooled EGR rate.

44

Chapter 4 KNOCK MODELING Paper II contains results from calibration and validation of a knock model based on the Livengood-Wu knock integral as described in Chapter 2.4.2. Some additional information about the experiments and calibration of the knock model is found below together with a new calibration for the ignition delay constants.

4.1 EXPERIMENTS A limited series of experiments were carried out on the standard turbocharged engine, see Chapter 3.4.2 for engine details. The purpose of the tests was to calibrate and validate the knock model at a number of different operating conditions as described in Paper II. The experiment matrix is repeated below for reference. Table 4.1 Operating conditions in the knock model calibration and validation test series. Engine speed 2500, 3000, 3500 rpm

λ @ 2500 rpm

0,92

λ @ 3000 rpm

0,86; 0,99; 1,10

λ @ 3500 rpm

0,84

Fuel

95 RON gasoline w. 0,95 Compression temperature T300 [K]

2,16

Compressions pressure, p300

T300, 0 T300 p300

Residual gas content, xrg

0,088

Engine speed N [rpm]

1+

1,33

T300,0

−0 , 47

x rg ,0 x rg

− 1,16

+ 0,912

T300 p 300

0,237

400 8 ⋅ 10 −5 − N N2

− 0,33

−0 , 28

x rg ,0 x rg

1,33 −

+ 0,763

660 N

1

1 1

750 + 0,625 N

As can be seen in Table 5.1, only air/fuel ratio and engine speed was found to influence the combustion mode parameter in Csallner’s correlations. This is in part due to the method used for fitting Wiebe functions to test data. First of all, the 50 % burned point was used as anchor angle in the identification. The total combustion duration was adjusted so that the mass fraction burned in an 8 CA interval beginning at the 50 % burnt point was equal in the test and the Wiebe function. The combustion mode parameter was then chosen arbitrarily to match measured data. This rather strange identification procedure was chosen to get correct mass burn rate when the burn rate is high and also because start of combustion could not be identified.

54

Chapter 5 – Combustion modeling using the Wiebe function

5.1.4 Witt Witt [57] made two sets of Wiebe parameters correlations to compare throttled and throttleless operation in the simulation and testing of a modern BMW four valve naturally aspirated engine running on part load. Similarly to Csallner, the Wiebe parameter identification method used by Witt focused on getting the 50 % burn point correct in the identified Wiebe function. A very large number of tests were used to identify the combustion model parameters: 221 operating points in throttleless operation and 193 operating points in throttled operation. The parameters investigated were spark timing, residual gas content, indicated work and engine speed. Results from throttled operation are shown in Table 5.2. All identified equations have the form of Equation 5.3, i.e. normalized functions for relative influence. Table 5.2 Wiebe combustion parameter correlations from Witt [57] for throttled operation in a BMW 4-valve engine. Ignition delay, Total combustion Combustion mode parameter, m ∆θd duration, ∆θ

Fi Spark advance θspark [bTDC] Residual gas content, xrg [%] Indicated work wi [kJ/dm3] Engine speed N [rpm]

Gi

0,678 + −4

2,383 ⋅ 10 θ

0,596 +

2 spark

0,879 +

2,48 0,5 θ spark

2 0,429 + 0,031x rg

2 3,648 ⋅ 10 − 4 x rg

1,115 − 0,346 wi

1,112 − 0,545wi1,5

0,992 −

Hi

1,246 ⋅ 10 5 N2

1,355 −

18,49 N

0,964 +

75,56 2 θ spark

1,076 − 2 2,534 ⋅ 10 − 4 x rg

1,007 + 0,004 ln wi

1,046 − 4,075 ⋅ 10 −7 N 1,5

5.2 EXPERIMENTS As described in Paper III, a few engine operating parameters were singled out as candidates that were likely to have an influence on the combustion. The first group of operating parameters are those related state of the combustible mixture of air and fuel, namely air/fuel ratio, intake pressure and temperature and residual gas content. Engine speed and spark timing make up the other group of parameters which are related to the turbulence which the flame encounters as it traverses the combustion chamber. Spark timing actually related to both groups, since the temperature and 55

Empirical Combustion Modeling in SI Engines pressure during the early part of combustion is highly dependent on the spark timing. Coolant temperature was also varied in the tests. The experimental procedure chosen for the combustion model investigation was to vary one parameter at the time while trying to keep other influencing factors constant. As all engine experimentalists know it is not straightforward to change only one parameter in an engine test, since many factors are affected when one control parameter is changed. Retarding the spark timing in a turbocharged engine will for example increase exhaust temperature which leads to increased turbine power producing a change in intake pressure and temperature which in turn affects combustion. The waste gate and throttle were used to control the intake conditions. Some care was also taken to tune the charge air cooler controller. This means that the engine torque was varying quite significantly between the tests. Figure 5.1 shows some test data from a spark timing variation test just to see the large variation in for example turbine inlet temperature and torque. The figure also shows that spark timing were chosen close to MBT timing and later. The pressure after the compressor varies more than the intake pressure since the throttle was used to control intake pressure.

56

Pressure after compr.

40

1.3 Intake pressure Intake temp.

1.2 50

%

1.1 a Sp

30

bu rn t

20

ng im i rk t

Intake temperature [°C] Spark timing [bTDC] 50 % burnt [aTDC]

Lambda Intake pressure [bar] Pressure after compresor [bar]

Chapter 5 – Combustion modeling using the Wiebe function

10

1.0 Lambda

1

2

3

4

5

6

7

Test no. Engine speed

rb in

rq To

e

920

240

in

ue

t le

230

m te p.

900 880

220

860

Turb

840

o spe

210

ed

Torque [Nm] Engine speed / 10 [rpm]

940

250 Tu

Turbo speed / 100 [rpm] Turbine inlet temp. [°C]

960

200

820 1

2

3

4

5

6

7

Test no.

Figure 5.1 Test data from spark timing variation with constant intake manifold pressure and temperature. Other measured parameters vary quite significantly. Waste gate and throttle was varied along with charge air cooling. Two sets of tests were performed with varying engine speed. One test was made with constant spark timing and one test with constant angle for 1 % mass fraction burned. Figure 5.2 shows the measured data from these two tests. The tests with constant 1 % burnt point were made to remove the influence of varying flame development period in terms of crank angles at different engine speeds. 57

Empirical Combustion Modeling in SI Engines 35 Intake temp.

2

30

Intake pressure

1

25 Lambda

0

-1

% 50

1 % burnt

rn bu

t

rk Spa

20 n timi

g

-2

15

Intake temperature [°C] Spark timing [bTDC] 50 % burnt [aTDC]

Lambda 1 % burnt [aTDC] Intake pressure [bar]

3

10 1500 2000 2500 3000

1500 2000 2500 3000

Engine speed [rpm]

Figure 5.2 Test data for speed variation tests with constant spark timing (left) and constant 1 % burnt (right). Interaction effects between parameters might be left out when using a one parameter at the time approach [58]. A second set of experiments was carried out to check for interactions, and also to have a data set for validation of the model. A reduced order two level factorial design with center points, augmented with a so called star composite design, was used. Figure 5.3 shows the basic structure of the design. The factors in the test were air/fuel ratio, spark timing, residual gas content and intake temperature. Engine speed was constant at 3000 rpm and the intake pressure was 125 kPa. The experimental matrix is shown in Table 5.3. The goal with the design of experiments approach was not to fit a response surface to the data but to get a validation data set which spans the experimental range. If there are significant interaction effects between any of the variables, the fit of the identified model in the validation data set will be poor. A total of 30 tests were performed for the validation data set.

58

Chapter 5 – Combustion modeling using the Wiebe function Table 5.3 Operating points for the factorial experimental design at 3000 rpm and 125 kPa intake pressure. Normalized Air/fuel Spark timing Intake EGR level ratio [bTDC] temperature [°C] fraction +1 1,00 18,0 45,0 2,4% -1

0,80

13,0

35,0

0,4%

0

0,90

15,5

40,0

1,4%

+1,41

1,04

19,0

47,1

not possible

-1,41

0,76

12,0

32,9

0,0%

Normalized level, factor 2

Extra points Base points

1 Central points 0 R=2

0,5

-1

-1

0

1

Normalized level, factor 1

Figure 5.3 Basic structure of central composite two level factorial design augmented with a star complement design.

5.3 DATA EVALUATION 200 engine cycles were recorded for each of the four cylinders and each operating condition. Heat release was calculated for each cycle with the algorithm described in Chapter 3.3. Wiebe parameters were identified for each individual cycle, described in more detail below. The average of the identified Wiebe parameters for each operating condition was then used to identify the combustion model parameters.

59

Empirical Combustion Modeling in SI Engines

5.3.1 Wiebe parameter identification As mentioned above in the summary of the works of Csallner [56] and Witt [57], different methods were used for identifying the Wiebe parameters from calculated heat release. Both Csallner and Witt also identified the flame development period for each operating point. Some different methods for identifying the Wiebe function parameters were tested in this work. It was found that the double logarithm method suggested in Wiebe [27] gave quite accurate results when the spark timing was used as start of combustion. After transforming the calculated mass fraction burned and the crank angle by two logarithms:

⎡ ⎛θ −θ ⎞ 0 xb (θ ) = 1 − exp ⎢− a⎜ ⎟ θ ∆ ⎠ ⎢⎣ ⎝

m +1

⎤ ⎥ ⎥⎦ m +1

2 nd

⎛ θ − θ0 ⎞ 1st logarithm : ln(1 − xb ) = −a ⋅ ⎜ ⎟ ⎝ ∆θ ⎠ ⎛ ln(1 − xb ) ⎞ logarithm : ln⎜ ⎟ = (m + 1) ⋅ (ln(θ − θ 0 ) − ln(∆θ )) ⎝ −a ⎠

(5.4)

it is straightforward to fit a straight line to the data and solve for the interesting parameters ∆θ and m. The results are fully comparable to a nonlinear least squares fit of a Wiebe function to measured data as shown in Figure 5.4, but the computational burned for the Wiebe method is only a small fraction of a nonlinear least squares fit. As an example, the time to fit Wiebe parameters for 200 engine cycles with the Wiebe method took about half a second, while a nonlinear least squares solver needed one minute to perform the same task. The crank angle transformation of the data, ln(θ - θ0), is nonlinear which gives an overweight for the later part of combustion in the estimation of Wiebe parameters. The same results could have bee accomplished with a weighted least squares estimate.

60

1.0

3%

Measured Double logarithm Nonlin. least squares

0.8

2%

0.6

1%

0.4

0%

0.2

-1 %

0.0

-2 % -15

0

15

30

45

Error

Mass fraction burned, xb

Chapter 5 – Combustion modeling using the Wiebe function

60

Crank angle [aTDC]

Figure 5.4 Comparison of nonlinear least squares and double logarithm method of fitting a Wiebe function to measured data. One concern for selecting more complicated methods of fitting Wiebe functions mentioned in both Csallner [56] and Witt [57] was to get correct 50 % burn point in the fitted Wiebe function. Another concern was to get correct mass fraction burned in a crank angle interval around the 50% burnt point, which I translate to correct 10-90 % burn duration. Figure 5.5 shows how the Wiebe double logarithm method with spark timing as start of combustion performs in estimating several mass fraction burn points for 89 operating conditions from the combustion model data sets. The standard deviation of the estimated burn points are also shown in the figure. The weighting introduced by the logarithm transformation is evident, since the 50 % burnt point and the 90 % burnt point has the best fit.

61

Empirical Combustion Modeling in SI Engines

Error [CA] Standard deviation [CA]

3

2,66 Mean error

2 1,93

1,26

1

0

0,58 Standard deviation -0,21 -0,68

-1 1%

5%

10%

50%

90%

10-90 %

Mass fraction burned

Figure 5.5 Mean error in mass fraction burned points of fitted Wiebe function compared to measured data. Ensemble averaged data from 89 operating conditions.

5.4 COMBUSTION MODEL CALIBRATION The presented combustion model differs from the existing models described above in two respects: the use of laminar burning velocity and the Wiebe parameters identification method. The identification method with spark timing taken as start of combustion removes the need for separate correlations for the flame development period. Only spark timing and engine speed remains as influencing parameters after the laminar burning velocity influencing function has been removed. For the combustion mode parameter, influencing functions were determined for engine speed and total combustion duration. Equation 5.2 simplifies to:

∆θˆ = ∆θ 0 ⋅ ∏ g i = ∆θ 0 ⋅ g SL ⋅ g spark ⋅ g speed i

mˆ = m0 ⋅ ∏ hi = m0 ⋅ h∆θˆ ⋅ hspeed

(5.5)

i

The first influencing function for total burn duration, gSL, accounts for the laminar burning velocity influence. Paper III shows the good results when using inverse laminar burning velocity as influencing function. In order to identify the remaining influencing functions for total combustion duration, data was normalized with the laminar burning velocity influencing function. The spark timing influencing 62

Chapter 5 – Combustion modeling using the Wiebe function function, gspark, is also described in Paper III. Below I will describe the process of identifying the engine speed influencing function, gspeed, in more detail, which for some cases requires normalizing with both laminar burning velocity and spark timing influencing function.

5.4.1 Modeling speed influence Two different test series were made for identifying the speed influence function, one with constant spark timing and one with constant 1 % mass fraction burnt, as shown in Figure 5.2. The structure of the model for total combustion duration directly gives the influencing function:

∆ θ = ∆ θ 0 ⋅ g SL ⋅ g spark ⋅ g speed ⇒ g speed =

∆θ ∆ θ 0 ⋅ g SL ⋅ g spark

(5.6)

The first step is to normalize the data with a base operating condition, as shown in the top left figure in Figure 5.6. The figure also contains the final identified speed influencing function to see what each step in the normalization does to the data. Since the gas conditions at spark timing for the speed test data varied, the data then had to be normalized by the laminar burning velocity influencing function, shown in the top right figure. Finally, after normalizing with the spark timing influencing function, bottom left, the test data collapses around the identified influencing function. The bottom right figure shows the measured and predicted total burn duration for the speed modeling data. The bottom left figure is identical to Figure 11 in Paper III.

63

Empirical Combustion Modeling in SI Engines

1.2

Constant spark timing Constant 1 % burnt Points from other tests

1.3 1.2

∆θ ∆θ0 x g(SL)

1.3

∆θ ∆θ0

1.1 1.0 0.9 0.8 0.7

1.1 1.0 0.9 0.8 0.7

1500 2000 2500 3000 3500 4000

1500 2000 2500 3000 3500 4000

1.3

Engine speed [rpm] 70

0.584

alt. G(N) = 0.0102 x

1.2 1.1 1.0 0.9 0.8

G(N) = 2,13 - 55,5 x N

-0,5

Measured and estimated ∆θ

∆θ ∆θ0 x g(SL) x g(θign)

Engine speed [rpm]

0.7

65 60 55 50 45

measured model

40 1500 2000 2500 3000 3500 4000

1500 2000 2500 3000 3500 4000

Engine speed [rpm]

Engine speed [rpm]

Figure 5.6 Illustration of the steps of fitting an total combustion duration influence function for the speed influence. Top left shows test data normalized with the base operating condition only. Top right data is also normalized by laminar burning velocity influence. Bottom left is additionally normalized by spark timing influence. All normalized data is plotted together with the identified influencing function for comparison. Bottom right shows the prediction of total combustion duration for the model identification data. The functional form for the speed influencing function was chosen from Wittt [57]. Several other functional forms were tested for the speed influencing function. One of them, a power function, is shown in the bottom left figure of Figure 5.6. The behavior outside of the identification data range has to be considered when choosing the functional form.

5.5 RESULTS The results from calibrating the combustion model along with some validation are summarized in Paper III. All identified influencing functions are summarized in an appendix to the paper.

64

Chapter 6 CONCLUSIONS Combustion simulation in one dimensional software has to be a very simplified description of the actual process, if the software is to maintain reasonably fast execution time. Complete three dimensional simulations of the real physics and chemistry of combustion is not practically possible today. Simulation of in cylinder turbulence requires sub models for turbulence structures smaller than the calculation mesh or an extremely dense mesh which renders the calculation near unfeasible. Chemical kinetic models exist for simple model fuels but are yet to be developed for practical automotive fuels. When three dimensional simulation of SI engine combustion becomes a practical alternative, which I’m sure will happen some time in the future, cycle to cycle variations is the next problem to tackle. Simplified empirical approaches for combustion and knock modeling have been described in this thesis. The combustion model only uses one key property of the flame, the laminar burning velocity, which makes it easy to use and interpret. However the combustion model will not react to changes in turbulence which is the major drawback of this approach. The knock model has been used by many others and this work simply adds to the understanding of the calibration and use of the model. The combustion model does not take cycle to cycle variations into account. The knock model on the other hand is a single cycle model. When using the knock model together with any ensemble average combustion model, it is quite hard to distinguish the meaning of the results. Some arbitrary limits for knock have to be set in order to get any useful results. In Paper II we used 30 crank angles after top dead

65

Empirical Combustion Modeling in SI Engines center and approximately 90 percent mass fraction burned as limits. If knock was predicted within/below these limits, the combustion was phased later until either of the boundaries were reached at simulated knock onset. This is not to be confused with the severity of knock or any other measure of knock. It is just a way to get useful output from the simulations. The use of one dimensional simulation tools in engine development has put new demands on engine measurement technique. Both cycle or time averaged measurements and crank angle resolved measurements fill an important role in simulation model calibration and validation. This thesis has put special focus on measurement data conditioning with some suggestions on filtering of noisy data and filtering for knock detection. These two filtering tasks have separate demands on the filters. Another area of interest is the calculation of heat release from cylinder pressure data. If the data is to be used for calibration or validation of combustion in a simulation model, it is essential to use the same assumptions and models as the simulation software when calculating heat release from measured data. The model for ratio of specific heats described in this thesis is an attempt to get closer to the simulation software in estimating in cylinder gas properties during combustion. The Woschni model for heat transfer is common between the simulations and measured data analysis. The largest discrepancy between measured data analysis and the simulation software is found in the single zone calculation of heat release for the measured data. A two zone model was used in the simulation software in order to be able to draw conclusions regarding knock.

6.1 FUTURE WORK The three dimensional calculated ignition delay surface for a model gasoline fuel developed by Yates and co-workers seems to be a reasonable next step in refining the knock model described in this work. For future development of the presented combustion model, it would be interesting to try to correlate combustion duration to for example swirl- or tumble numbers. This would enable appropriate scaling of the combustion duration when changing for example cylinder head and combustion chamber geometry or, the other way around, to suggest an appropriate turbulence level of a new engine design to avoid for example knock. Additional refinement of the model would be to look into part load and camshaft phasing or fully variable valve trains. Cycle to cycle variations could be incorporated by means of some statistical distribution of the Wiebe parameters, correlated to engine operating conditions in a similar way as the Wiebe parameters themselves. For knock simulation, the most likely knocking cycle could be found from the statistical distribution. After finding the 66

Chapter 6 – Conclusions knock limited combustion phasing for the most likely knocking cycle, the ensemble average cycle could be simulated to give an accurate prediction of engine output at the knock limit.

67

REFERENCES [1]

[2]

[3] [4] [5]

[6] [7] [8] [9] [10] [11] [12]

[13]

[14]

[15]

www.acea.be (May 2005); Monitoring of ACEA’s Commitment on CO2 Emission Reductions from Passenger Cars (2002); Joint Report of the European Automobile Manufacturers Association and the Commission Services, Final Report, 2003. Mitchell, Bill; Advanced fuel cell development for automotive operation; Oral presentation at the SAE Fuels & Lubricants Meeting and Exhibition, Rio de Janeiro, May 12, 2005. Blair, Gordon P.; Design and Simulation of Four-Stroke Engines; Society of Automotive Engineers, 1999. GT-Power User Manual, Version 6.1; Gamma Technologies, August 2004. Westin, Fredrik; Simulation of turbocharged SI-engines – with focus on the turbine; Doctoral Thesis in Machine Design, KTH, Stockholm, 2005, ISSN 1400-1179. Watson, N., Janota, M. S.; Turbocharging the Internal Combustion Engine; The Macmillan Press Ltd, London, 1982. Söderberg, Fredrik, Johansson, Bengt; Fluid Flow, Combustoin and Efficiency with Early or Late Inlet Valve Closing; SAE Technical Paper 972937. Heywood, John B.; Internal Combustion Engine Fundamentals; MCGraw-Hill 1988. Glassman, Irvin; Combustion, third edition; Academic Press, 1996. Kalghatgi, Gautam T.; Early Flame Development in a Spark-Ignition Engine; Combustion and flame 60 (1985) 299-308. Tagalian, Joel, Heywood, John B.; Flame Initiation in a Spark-Ignition Engine; Combustion and Flame 64: 243-246 (1986). Metghalchi, Mohamad, Keck, James C.; Burning Velocities of Mixtures with Methanol, Isooctane and Indolene at High Pressure and Temperature; Combustion and Flame 48 (1982) 191-210. Metghalchi, Mohamad, Keck, James C.; Laminar Burning Velocity of PropaneAir Mixtures at High Temperature and Pressure; Combustion and Flame 38 (1980) 143-154. Johansson, Bengt; On Cycle to Cycle Variations in Spark Ignition Engines; Doctoral thesis, Division of Combustion Engines, Department of Heat and Power Engineering, Lund Institute of Technology, 1995. Grandin, Börje; Knock in Gasoline Engines – the effect of mixture composition on knock onset and heat transfer; Doctoral thesis, Department of Thermo and Fluid Dynamics, Chalmers University of Technology, Gothenburg, 2001.

69

Empirical Combustion Modeling in SI Engines [16] Westbrook, Charles K.; Chemical Kinetics of Hydrocarbon Ignition in Practical Combustion Systems; Proceedings of the Combustion Institute, vol. 28, 2000, 1563-1577. [17] Risberg, Per; A Method of Defining the Auto-Ignition Quality of Gasoline-Like Fuels in HCCI Engines; Licentiate Thesis in Machine Design, KTH, Stockholm, 2005. [18] Fieweger, K., Blumenthal, R., Adomeit, G.; Self-Ignition of S.I. Engine Model Fuels: A Shock Tube Investigation at High Pressure; Combustion and Flame 109 (1997) 599-619. [19] Viljoen, Carl L., Yates, Andy D. B., Swarts, André, Balfour, Gillian, Möller Klaus; An Investigation of the Ignition Delay Character of Different Fuel Components and an Assessment of Various Autoignition Modelling Approaches; SAE Technical Paper 2005-01-2084. [20] Swarts, André, Yates, Andy D. B., Viljoen, Carl L., Coetzer, Roelof; A Further Study of Inconsistency between Autoignition and Knock Intensity in the CFR Octane Rating Engine; SAE Technical Paper 2005-01-2081. [21] Kalghatgi, Gautam T.; Fuel anti-knock quality – Part I. Engine studies.; SAE Technical Paper 2001-01-3584. [22] Kalghatgi, Gautam T.; Fuel anti-knock quality – Part II. Vehicle studies – how relevant is Motor Octane Number (MON) in modern engines?; SAE Technical Papers 2004-01-3585. [23] Pan, J., Sheppard, C. G. W.; A Theoretical and Experimental Study of the Modes of End Gas Autoignition Leading to Knock in S.I. Engines; SAE Technical Paper 942060. [24] Brunt, Michael F. J., Pond, Christopher R. and Biundo, John; Gasoline Engine Knock Analysis using Cylinder Pressure Data; SAE Technical Paper 980896. [25] Bengisu, Turgay; Computing the Optimum Knock Sensor Locations; SAE Technical Paper 2002-01-1187. [26] Worret, R., Bernhardt, S., Schwartz, F., Spicher, U.; Application of Different Cylinder Pressure Based Knock Detection Methods in Spark Ignition Engines; SAE Technical Paper 2002-01-1668. [27] Wiebe, I. I.; The Combustion Speed in Internal Combustion Piston-Engines – fuel combustion rate equation combining an empirical and a theoretical approach; Collected works of piston engine research, Laboratory of Engines, Academy of Science, USSR, Moscow 1956. (Translated from the Russian text by Marek Kiisa, KTH, Stockholm 1993) [28] Livengood, J. C., Wu, P. C.; Correlation of Autoignition Phenomena in Internal Combustion Engines and Rapid Compression Machines; Fifth Symposium (International) on Combustion, 347-356, 1955. [29] Douaud, A. M., Eyzat, P.; Four-Octane-Number Method for Predicting the Anti-Knock Behaviour of Fuels and Engines; SAE Technical Paper 780080. 70

References [30] By, A., Kempinski, B., Rife, J. M.; Knock in Spark Ignition Engines; SAE Technical Paper 810147. [31] Gautier, B. M., Davidsson, D. F., Hanson, R. K.; Shock tube determination of ignition delay times in full-blend and surrogate fuel mixtures; Combustion and Flame 139 (2004) 300-311. [32] Yates, Andy D. B., Swarts, André, Viljoen, Carl L.; Correlating Auto-Ignition Delay And Knock-Limited Spark-Advance Data For Different Types Of Fuel; SAE Technical Paper 2005-01-2083. [33] www.adlinktech.com (May 2005) [34] www.ueidaq.com (May 2005) [35] www.gemssensors.com (May 2005) [36] www.avl.com (May 2005) [37] www.kistler.com (May 2005) [38] www.druck.com (May 2005) [39] www.ametekcalibration.com (May 2005) [40] Pischinger, Rudolf; Engine Indicating – User Handbook; AVL List Gmbh, Austia, 2002. [41] Lee, S., Bae, C., Prucka, R., Fernandez, G., Filipi, Z. S., Assanis, D. N.; Quantification of Thermal Shock in a Piezoelectric Pressure Transducer; SAE Technical Paper 2005-01-2092. [42] Odendall, B.; A Discussion of Errors in the Measurement of Gas Temperature; MTZ 3/2003 pp. 196-199. [43] www.pentronic.se; StoPextra 6/98 [44] www.isotech.co.uk (May 2005) [45] www.testo.com (May 2005) [46] www.monarchinstrument.com [47] www.micro-epsilon.de (May 2005) [48] www.ecm-co.com (May 2005) [49] Smith, Steven W.; The Scientist and Engineer’s Guide to Digital Signal Processing, second edition; California Technical Publishing, San Diego, California 1999; http://www.dspguide.com [50] Brunt, Michael F. J., Rai, Harjit, Emtage, Andrew L.; The Calculation of Heat Release from Engine Cylinder Pressure Data; SAE Technical Paper 981052. [51] Klein, Marcus, Eriksson, Lars; A Specific Heat Ratio Model for Single-Zone Heat Release Models; SAE Technical Paper 2004-01-1464. [52] Kanury, A. Murty; Introduction to combustion phenomena; Table A.3; New York cop. 1975. [53] AVL List GMBH; Operating Instruction AVL Concerto Software Version 3.0; AVL List GMBH, Graz 1999. 71

Empirical Combustion Modeling in SI Engines [54] www.galcit.caltech.edu/EDL/public/thermo/thermo.inp; NASA tables of species thermodynamic properties; January 2005. [55] Bradley, D, Morley, C and Walmsley, H.L.; Relevance of Research and Motor Octane Numbers to the Prediction of Engine Autoignition; SAE Technical Paper 2004-01-1970. [56] Csallner, Peter; Eine Methode zur Vorausberechnung der Ändrung des Brennverlaufes von Ottomotoren bei Geänderten Betriebsbedingungen; München Techn. Univ., Diss., 1981. [57] Witt, Andreas; Analyse der thermodynamischen Verluste eines Ottomotors unter den Randbedingungen variabler Steuerzeiten; Graz, Tech. Univ., Diss., 1999. [58] Box, George E. P., Hunter, Willam G., Hunter, J. Stuart; Statistics for experimenter; John Wiley & Sons, Inc. 1978.

72

PAPER I

2005-01-1150

Divided Exhaust Period – a gas exchange system for turbocharged SI engines Christel Elmqvist Möller, Pontus Johansson, Börje Grandin Fiat-GM Powertrain - Sweden

Fredrik Lindström Royal Institute of Technology Copyright © 2004 SAE International

be hard to meet at the same time and is a major challenge for engine manufactures and researchers.

ABSTRACT The necessity to limit the boost pressure in turbocharged gasoline engines results in higher exhaust pressure than inlet pressure at engine speeds when the wastegate is opened. This imbalance has a negative influence on the exhaust scavenging of the engine and results in high levels of residual gas and consequently the engine is more prone to knock.

Using turbochargers can be a viable solution to this challenge. The major advantage of the turbocharged engine is achieving the same power output with a smaller displaced volume, thereby decreasing the fuel consumption. However, turbocharged engines have some drawbacks. The turbine increases the pressure in the exhaust manifold, thereby increasing the amount of residual gas trapped in the cylinder. This decreases the knock resistance of the engine [1] and also reduces the volumetric efficiency [2]. In a four stroke turbocharged engines with four cylinders and a single turbine, the blowdown pulse from one cylinder can interfere with the previous cylinder in firing order during exhaust-intake overlap. This further impairs the scavenging of the combustion chamber and as a consequence increases the charge temperature and reduces volumetric efficiency and knock resistance. Other drawbacks are the relatively high pumping losses compared to a naturally aspirated engine and delayed catalyst light off due to the turbine acting as a heat sink in the exhaust system. Finding ways to avoid these shortcomings would of course improve the performance of the turbocharged engine.

This paper presents a study of a gas-exchange system for turbocharged SI engines. The concept aims at improving the performance and emissions of a turbocharged SI engine by dividing the exhaust flow from the two exhaust valves into two different exhaust manifolds, one connected to the turbocharger and one connected to a close coupled catalyst. By separating the valve opening period of the two valves and keeping the duration of both valve opening events shorter than 180 crank angle degrees, the disturbance of the exhaust blowdown pressure pulse during valve overlap in a four cylinder engine can be completely eliminated. The study was carried out both experimentally on a four cylinder turbocharged SI engine and with extensive 1-D simulation of the system. A positive pressure difference over the engine could be realized over the entire speed range by using the concept system. Simulations show up to 60 % reduction of residual gas content compared to a standard turbocharged engine. The study also showed that the time to catalyst light off could be reduced with over 35% compared to a standard turbocharged engine with a conventional exhaust system.

One way mentioned as early as 1924 in a British patent [3] has been further investigated on a modern fourcylinder turbocharged engine. In this paper the potential of this concept, which will be called DEP (Divided Exhaust Period), will be examined. The concept has been investigated by means of 1-D simulation as well as by tests on a prototype engine. This paper will deal with the theories behind the concept, the development of a 1-D simulation model and experience gained from a prototype engine. In particular cold start catalyst light-off time and full load performance of the DEP engine with varying exhaust valve lift profiles have been studied. High full power potential of a new engine concept is important to enable a high degree of downsizing.

INTRODUCTION It is a well-known fact that the emission levels from the transport sector needs to be decreased. At the same time customer request is for more powerful engines with less fuel consumption. These different requirements can 1

Charge air system

Exhaust blowdown valve

C

T

Catalyst

Valve lift

Exhaust blow-down system

Intake valves

Exhaust scavenging valve

Exhaust scavenging system

90

Trapping valve

180

270

TDC

450

540

630

Crank angle [aTDC] Figure 2 Valve lift curves for the blowdown, scavenging and intake valves.

Figure 1 Schematic view of the of the exhaust systems configuration in the DEP engine.

duration of the blowdown phase. Figure 3 shows simulated mass flow in both of the DEP engine exhaust systems compared to a standard turbocharged engine at 5500 rpm. Over 60% of the mass flow and the highest enthalpy levels are found in the blow-down pulse at 5500 rpm in the standard turbocharged engine. The remainder of the mass flow is generated by the piston displacement, which can be seen in the figure as a second peak in the mass flow.

TECHNICAL CONCEPT The Divided Exhaust Period (DEP) concept is an alternative way of accomplishing the gas exchange in a turbocharged engine. The aim is to improve the performance of a turbocharged engine, regarding lowend torque, peak power and emissions. In the DEP engine the two exhaust ports from each cylinder have been separated. The blowdown pulse is evacuated through the blowdown valve, which leads to the turbocharger. As the pressure in the exhaust is decreased and the piston displacement phase commence, the scavenging valve open and lead the remaining exhaust gas directly to a close-coupled catalyst (CCC). By bypassing the turbine the high pressure in the manifold connected to the blowdown valve is avoided and the gas exchange is improved. The engine is schematically shown in Figure 1. Figure 2 show the valve lifts for the blowdown valve leading towards the turbine and the scavenging valve connected to the close-coupled catalyst.

In previous sections five main difficulties with a 4cylinder conventional turbocharged engine were described: • Cylinder evacuation • Negative PMEP • Pulse interaction between cylinders • Knock sensitivity • Cold start The same reasoning as for a conventional turbocharged engine holds true for the DEP concept when it comes to choosing between a small or large turbine. The main driving force behind the DEP concept is to avoid or decrease the negative effects coupled to a small turbine. The turbine is restricted in mass flow so that a wastegate needs to be used at high load. Using a

Important parameters for turbine performance are the mass flow, the pressure difference and temperature difference over the turbine, since these are related to the enthalpy. Ideally the high enthalpy blowdown pulse is located around BDC and a positive pressure difference exists between cylinder and exhaust manifold. When the pressure in the cylinder decreases, the positive pressure difference decrease. By opening the cylinder to the scavenging manifold, which has a lower pressure compared to the blowdown manifold, positive pressure difference can be achieved over the engine during the exhaust/intake overlap higher up in the speed range of the engine.

standard t/c

Mass flow rate [kg/s]

0.3

The majority of exhaust gases are evacuated during the blowdown pulse. The crank angle duration and mass fraction evacuated during the blowdown phase depend on engine speed. With increased engine speed the pressure difference between the cylinder and the exhaust manifold decreases and as a consequence the mass fraction evacuated during the displacement phase increases. Choking of the exhaust valves influences the

DEP blow-down system

0.2

0.1

0.0

blow-down valve 90

135

DEP scavenging system

Valve lift

Intercooler

scavenging valve 180

225

270

315

360

405

Crank angle [aTDC] Figure 3 Simulated mass flows in the DEP engine exhaust systems compared to a standard turbo charged engine at 5500 rpm.

2

wastegate is a loss of potential work. With the DEP concept the mass flow is divided already in the cylinder, which has the same effect as a wastegate at high load, i.e. bypassing the turbine.

EXPERIMENTAL SET-UP SIMULATION MODEL - In order to investigate the DEP concept more thoroughly an engine model was developed in the 1-D software GT-Power [4]. The aim was to gain a more in-depth understanding of the gas exchange process and to optimize the engine design with respect to valve timing and pipe design. 1-D simulation is based on the solution of the governing equations; momentum-, energyand massconservation, in 1-D. However, in order to transform a 3-D problem to 1-D some additional information is needed. Figure 4 give some example of areas dependent on input data.

Dividing the mass flow already in the cylinder, as with the DEP concept, has other advantages compared to a wastegate. Cylinder evacuation was mentioned as an important factor for engine performance. By opening up the cylinder to a low pressure exhaust manifold, which in a sense is connected to the atmosphere, the high pressure created by the turbine can be avoided. This enables a better evacuation of the cylinder. By making the duration of the exhaust blowdown valve only slightly longer than 180° CA, the pulse interference between cylinders at blowdown can be eliminated. By eliminating the pulse into the cylinder the DEP engine should be able to decrease the amount of hot residual gases in the engine and decrease the gas exchange pumping losses, thus enabling better volumetric efficiency, knock resistance and overall engine efficiency.

3-D phenomena • Combustion • Pipes • Valves • Turbocharger • Intercooler • …

Catalyst light-off time has a great influence on cold start emissions. Hence, it is important to keep as much heat in the exhaust as possible in order to quickly reach catalyst light-off. Therefore placement of the catalyst close to the exhaust ports is important. By implementing valve de-activation for the blowdown valve, all of the exhaust gas is lead directly to the close-coupled catalyst mounted in the scavenging exhaust system. Consequently the emissions in terms of catalyst light-off time can be improved. The turbine is in this configuration effectively bypassed and the turbine will not act as a heat sink during cold start.

Additional information • Measurement data • Coefficients • Advanced sub-models Figure 4 Schematic image of the transformation from a 3-D heat transfer, flow and chemical kinetics problem to 1-D.

The engine model was built by transferring the geometrical dimensions of the engine into the model. Since the governing equations are only solved in 1-D, in the direction of the flow, 3-D effects must be taken into account in some other manner. Flow loss coefficients are used in order to find the right pressure loss occurring from pipe bends, materials etc. These coefficients are pre-defined in the software and usually not in need of much adjustment. The flow through the intake and exhaust valves was modeled by discharge coefficients as a function of lift height, which was measured in a flow bench [5]. Simulating the combustion is more complicated. Due to the combination of 3-D flow and chemical kinetics it is necessary to use special combustion models or measurement data. In this project the combustion has been simulated with the Wiebe function [4], which can be described as a curve fit to measured heat release. The main drawback with using such a combustion model is its dependence on measured combustion data, i.e. at what crank angle 50 % of the fuel is burnt and the duration of the combustion. Hence, simulating unknown engine concepts relies on the quality of the assumptions made for these combustion parameters.

From the discussion presented above it is clear that valve phasing and duration will have a great influence on the results. In order to investigate and evaluate the DEP concept even further a number of cam profiles were designed. The aim was to minimize the number of cams and to have a clear strategy when evaluating the gas exchange process for the DEP concept. The tests with different camshafts aimed at determining the influence on pumping losses, engine output and cylinder scavenging. Four key factors were identified and are described in Table 1. Table 1 Identified key factors for cam timing and duration evaluation.

Case

Parameter

Objective

1

Duration blowdown valve

Investigate influence of cylinder emptying towards turbine

2

Duration scavenging valve

Investigate required energy to drive the turbine

3

Overlap scavenging valve – intake valve

Investigation of overlap influence between exhaust and intake valve

4

Intake valve closing

Investigate intake valve closing influence with constant TDC overlap

1-D model Representation of 3-D phenomena in 1-D ⇒ Dependence on quality of input data

The turbocharger is another complicated area. Turbocharger modeling is based on performance maps for the compressor and turbine. These performance maps are based on flow bench measurements with 3

stationary flow [6]. The experimental data is then interand extrapolated to obtain the full-range maps. The maps relate speed and efficiency to the pressure ratio and mass flow over the turbine and compressor. Since the turbocharger models are based on stationary measurements in a limited range, these models can require an extensive amount of tuning [7]. In order to simplify the turbocharger simulation it is important to measure pressure, mass flow and turbocharger speed. The turbocharger simulation is difficult to perform when no knowledge of the actual engine performance is available.

28 mm exhaust valve diameter. The cylinder head was modified to house 32 mm exhaust valves with separated exhaust runners to the exhaust pipe flange. The coolant channels were also modified to ensure sufficient cooling of the exhaust ports. Specifications of the DEP engine are found in Table 2. Table 2 DEP engine specifications.

Bore Stroke Compression ratio Combustion chamber Inlet valve diameter Exhaust valve diameter

MEASUREMENT SET-UP - Since little knowledge was available of how the combustion would change with the DEP concept and how the turbine would react to the change in mass flow when the exhaust pulse was divided, a prototype engine was built to run parallel to the simulations. A number of experiments have also been made to determine the potential of the DEP concept. The engine tests were performed at KTH and Fiat-GM Powertrain in-house facilities.

86 mm 86 mm 9.5 Pentroof 32 mm 32 mm

The valve durations are quite short compared to an ordinary turbocharged engine, especially the scavenging valve. In order to handle the dynamics of the valve and yet obtaining maximum valve lift, lightweight materials had to be used for the exhaust valves. TI-Al exhaust valves were originally chosen to maximize possible valve lift area. These valves were replaced with sodium cooled steel exhaust valves due to high thermal load on the blowdown valve. Lightweight conical valve springs and valve spring retainers were also used along with the standard roller rocker arms and hydraulic lifters.

Cylinder pressures have been measured with AVL GM12D piezoelectric pressure transducers at a sample rate of 0.1 to 0.4 crank angles. 200 or more cycles have been recorded for each operating condition as a base for cycle resolved indicated statistics. [8][9] Crank angle resolved pressures in the intake and exhaust system have been measured at several positions with piezoresistive and steel diaphragm strain gauge pressure transducers. These pressure transducers were calibrated in a static test rig, obtaining a total static measurement uncertainty of the measuring chain in the range ± 3 - 8 kPa. Analogue anti aliasing filters and digital zero phase shift low pass filtering was applied to the pressure signals prior to further analysis. Temperatures have been measured time resolved with type K thermocouples, with a typical response time in the order of one second. Emissions have been measured time resolved at different positions in the exhaust systems. A two-channel fast flame ionization detector was used for crank angle resolved hydrocarbon measurements in the cold start tests. Other measurements include fuel mass flow rate, broadband lambda, turbine speed and torque. Air mass flow was estimated from measurements of lambda and fuel mass flow.

A number of exhaust and inlet camshafts were manufactured in order to evaluate different valve overlaps; between the exhaust scavenging valve and the inlet valves as well as the overlap between the blowdown valve and the scavenging valve in the exhaust stroke. The choice of valve lift profiles is described in detail in Table 3. Continuously variable camshaft phasing, with an adjustment range of 50 CAD, was installed on both the exhaust and the intake camshaft. The lift profiles of the different camshafts were measured when mounted on the engine to ensure proper valve lift curves as well as proper positioning relative to the crankshaft. EXPERIMENTS - Several operating modes has been identified and tested on the DEP engine. These include: 1. Low speed, where the scavenging exhaust system is shut off with the trapping valve, see Figure 1, to get sufficient mass flow over the turbine and prevent blow through. 2. High speed, where both exhaust systems are open and assist in emptying the cylinders. 3. Cold start, where the blowdown exhaust system, leading to the turbine, is shut off.

Limiting factors during the maximum torque tests has been 980°C maximum turbine inlet temperature measured with a 6 mm thermocouple, minimum overall relative air/fuel ratio of 0.77, 100 kPa maximum boost pressure and maximum coefficient of variance of IMEP (COV) 5 %. Ignition was advanced towards the knock limit.

Table 3 Valve opening and closing events for the reference camshaft.

Valve

HARDWARE MODIFICATIONS – The DEP engine was based on a standard 2 dm3 turbocharged port fuel injected engine. The standard cylinder head with 4 valves per cylinder had 32 mm inlet valve diameter and

Exhaust blow-down Exhaust scavenging Intake 4

Duration [CAD] 200 159 239

Opening [aTDC] 120 220 341

Closing [aTDC] 320 379 580

Exhaust blowdown valve

Exhaust scavenging valve

process and power potential during cold start with only the exhaust scavenging valve active. The second set of tests focused on optimizing the valve timing for the cold start. The second set of tests were performed in a cold start rig without cylinder pressure measurements and dynamometer, otherwise with a similar set-up as in the other cold start tests. A comparative test with a standard turbocharged engine was also performed. In the tests with the DEP engine, the roller finger follower was removed from the blowdown valve in order to run the engine with only the scavenging valve active.

Valve lift

Intake valves

Reference camshaft Early scavenging 1 Early scavenging 2

90

180

270

TDC

450

540

630

Valve lift

Crank angle [aTDC] Exhaust blowdown valve

90

Exhaust scavenging valve

Two different catalyst systems were evaluated in the cold start rig. One of the systems was a 400 cells per square inch (cpsi) system that was designed for a closecoupled installation behind the turbocharger in the standard engine. The second system had two catalysts, a 900 cpsi close coupled pre-converter and a 400 cpsi main catalyst. This is a system that could be similar to a production DEP system with the pre-converter placed in the exhaust scavenging system and the main catalyst after the junction of the two exhaust systems.

Intake valves

Reference camshaft Late scavenging

180

270

TDC

450

540

630

Crank angle [aTDC]

RESULTS Figure 5 Valve lift profile combinations that have been tested in the Divided Exhaust Period Engine.

In this project initial testing was performed in order to calibrate the simulation model so that it could be used in further studies. Testing and simulation was then performed in parallel. Tests were used for simulation model validation and simulations helped in interpretation of test data. Simulation model data will be presented together with test engine performance for a number of key parameters below. Both cycle average and crank angle resolved results are shown.

Full load performance – A number of camshafts were manufactured and tested on the engine to evaluate the performance of the DEP engine concept at full load. A reference set of camshafts with valve lift events according to Table 3 was selected. The tests with different camshafts aimed at determining the influence of divided exhaust period valve timing on pumping work, engine output and cylinder scavenging. Continuously variable camshaft phasing further increased the possibility to vary the valve timing. Camshaft combinations that have been tested are shown in Figure 5. Results from tests with varying exhaust scavenging valve opening are reported in this paper.

SIMULATION MODEL CORRELATION - One of the key parameters when simulating an engine model is the volumetric efficiency. Air mass flow was obtained from measured fuel mass flow and measured lambda. Since the DEP engine has two separate exhaust manifolds several possibilities for measuring lambda exist. The lambda value used for calculating mass flow was an

Volumetric efficiency

measured

simulated

1.1

800

1.0

600

0.9

400

0.8

200

0

0.7

Two separate tests were performed to evaluate the cold start performance of the DEP engine with active catalysts installed in the exhaust scavenging system. The first tests were focused on the gas exchange

1000

2000

3000

4000

5000

Engine speed [rpm] Figure 6 Comparison of volumetric efficiency and mass air flow between simulation and measurement.

5

Mass air flow [kg/h]

Cold start catalyst light-off – In order to meet future stringent legislation regarding emissions it is imperative to achieve a fast light-off of the catalyst system. It is important to heat up the catalyst substrate as fast as possible to achieve a fast light-off. This is achieved by running the engine with very late ignition timing [10] and sometimes with secondary air injection to produce a secondary reaction in the exhaust manifold [11]. Ideally all of the heat that is generated in this manner should be transferred to the catalyst substrate. A major cause of heat losses for turbocharged engines is the turbocharger. With the DEP system, the turbocharger is excluded during the cold start and a significant improvement in cold start emissions was anticipated.

average from measured values in both the blowdown system and the scavenging system. For some cam settings a significant amount of blow through was obtained, this will inevitably influence the measured lambda value in the scavenging system. Hence the lambda measurement does not necessarily represent the combusted mixture. This will effect the calculation of mass flow. Figure 6 show a comparison between measured and simulated values of air mass flow and volumetric efficiency. The model does not show particularly good accuracy over the speed range, which is surprising since most other engine calibration parameters show good agreement with measurement. This discrepancy might be due to the previously mentioned difficulty in mass flow calculations from measured lambda. Another explanation to the discrepancy in volumetric efficiency is related to the modeling of residual gas. The DEP concept has asymmetrical scavenging of the cylinder since only one exhaust valve is open during the scavenging phase. This will affect the residual gas content. If the residual gas content is not modeled correctly, this will affect the simulated volumetric efficiency. However, 3-D simulation was not available at this stage and 0-D simulation data was considered sufficient.

the DEP concept. At low engine speeds all of the energy in the exhaust has to be transferred via the turbine to the compressor in order to reach the boost pressure target. However, with the DEP concept a part of the exhaust is bypassed the turbine and as a consequence it is difficult to reach the boost target. With the DEP engine concept it would be possible to use a smaller turbine without suffering from increased residuals, as the amount of residual gas is not directly related to the exhaust pressure before the turbine. Even with a considerably smaller turbine it would be very difficult to reach the target with some of the exhaust energy lost through the scavenging system. A thorough matching of the turbocharger was not carried out and the tests were performed with a standard turbocharger with a smaller inlet area. In order to increase the boost pressure at low speed the trapping valve has to be closed, which forces all of the exhaust past the turbine. Thereby the required boost pressure can be obtained. With the trapping valve closed it is beneficial to have a very early scavenging exhaust valve opening, denoted early scavenging 2 in Figure 8, thereby obtaining large overlap between the blow down valve and the scavenging valve. (See Figure 8, compare early scavenging 1 and 2 at 1500 rpm.) A large overlap is necessary in order to have mass transfer from the scavenging system into the blowdown system and subsequently through the turbine. The exhaust scavenging system is emptied into the exhaust blowdown system via the following cylinder in firing order during exhaust blowdown/scavenging overlap. The power delivered to the compressor is increased with an increased mass flow over the turbine. By charging the exhaust scavenging system with intake pressure the mass flow across the turbine can be increased further, hence the boost pressure can be increased as well. Camshaft phasing tests with closed trapping valve

MEAN EFFECTIVE PRESSURES – Simulated BMEP and IMEP, shown in Figure 7, correlate well with measurements. The engine model targets a specified BMEP by adjusting the wastegate diameter. IMEP does not correlate exactly due to uncertainty of simulated FMEP values. Looking at BMEP obtained in engine tests with different camshafts in Figure 8, reveals that the DEP reached slightly higher power than the standard engine in the tests. However, low speed BMEP was in fact below the standard engine when the trapping valve, see Figure 1, was open. The reason for the low BMEP can be found in

2.2 measured

simulated

2.0

IMEP

2,2

1.8

BMEP [MPa]

IMEP and BMEP [MPa]

2,4

2,0 1,8 BMEP

1,6

closed trapping valve

1.6 1.4

standard t/c reference cam early scavenging 1 early scavenging 2

1.2

1,4

1.0

1,2

1000

1,0 1000

2000

3000

4000

5000

2000

3000

4000

5000

Engine speed [rpm]

Engine speed [rpm]

Figure 8 Obtained brake mean effective pressure with different exhaust valve lift curves compared to the standard engine. The solid markers indicate closed trapping valve in the scavenging exhaust system.

Figure 7 IMEP and BMEP comparison between test and simulation. Values for BMEP are on the same curve since the engine model targets BMEP.

6

opening, denoted early scavenging 2 in Figure 9, results in pumping work similar to the standard turbocharged engine at low speed. However at high engine speeds the pumping losses are significantly reduced. The power improvement from the reduced pumping losses in the DEP engine is up to 10 kW at 5000 rpm, which is about 6 % of the engine rated power.

50

PMEP [kPa]

0 -50 -100 -150

RESIDUAL GAS CONTENT - One of the reasons for designing the DEP engine was to decrease residual gas content in the cylinder at full load in order to improve knock resistance and volumetric efficiency. The simulation model proved to be helpful in estimating the residual gas content. Figure 10 illustrate the difference in residual gas content between the DEP engine and a standard turbocharged engine, showing a difference over the entire speed range. At 5500 rpm the residual gas content is decreased with 14 %. However in the lower speed range the decrease is over 60 %.

standard t/c reference cam early scavenging 1 early scavenging 2

-200 1000

2000

3000

4000

5000

Engine speed [rpm] Figure 9 Comparison of pumping mean effective pressure at full load with different divided exhaust period valve lift profiles and standard t/c operation.

Residual gas content was not measured in the engine tests. However, the pressure difference from intake system to exhaust system at the gas exchange TDC gives a hint to how well the exhaust gases are scavenged from the cylinders. Figure 11 shows this pressure drop obtained from engine tests with reference camshaft and with early exhaust scavenging valve opening. The standard turbocharged engine barely reaches a positive pressure difference in the mid speed range whereas the DEP engine has a large positive pressure difference at all speeds up to 4500 rpm for the reference camshaft. Increasing the scavenging exhaust valve duration by early opening, denoted early scavenging 1 in Figure 11, gives more time and valve area for evacuating the exhaust and also a higher pressure difference at higher speeds. A very long exhaust scavenging valve duration, denoted early scavenging 2 in Figure 11, gives a lower pressure difference at lower speeds, primarily due to loss of available energy in the exhaust blowdown system and

showed that increased overlap between the scavenging exhaust valve and intake valves boosted the engine output significantly. A 20 % increase in available torque was observed at 1500 and 2000 rpm when the overlap between exhaust scavenging valve and intake valves was increased by 40 CAD. The DEP engine exhibits positive PMEP over an extended speed range compared to a standard turbocharged engine, i.e. work is added to the crankshaft during the gas exchange. In addition, high speed negative PMEP is reduced compared to the standard t/c engine. Figure 9 shows PMEP over the engine speed range with three different camshaft combinations, varying the opening of the scavenging exhaust valve and blowdown/scavenging valve overlap, see Figure 5 top graph. Advancing the scavenging exhaust valve opening 15 CAD, denoted early scavenging 1 in Figure 9, extends the range of positive pumping work and decreases high speed pumping losses even further compared to the DEP reference camshaft. A very early scavenging exhaust valve

100

standard t/c reference cam early scavenging 1 early scavenging 2

60%

Standard t/c engine

6%

40%

4%

20% DEP engine

2%

DEP improvement

Residual gas content at IVC

8%

Pressure drop over cylinder at 360 aTDC [kPa]

80 60 40 20 0 -20 -40

0%

1000

2000

3000

4000

5000

Engine speed [rpm] 1000

2000

3000

4000

5000

Engine speed [rpm]

Figure 11 Pressure drop over cylinders (from intake runner to exhaust runner in the scavenging exhaust system) at 360 aTDC for several tested divided exhaust period exhaust valve lift profiles.

Figure 10 Comparison of simulated values of residual gas content in the cylinder for the DEP engine vs. the standard turbocharged engine.

7

subsequent loss of boost pressure. Tuning of the scavenging exhaust system also has an effect on the pressure difference. When the exhaust scavenging valve duration is increased the optimum pulse reflection behavior shifts towards higher speeds.

significant pressure drop in the cylinder when the exhaust scavenging valve opens at 230° aTDC. At the same time, the pressure in the exhaust scavenging system rises to the pressure in the cylinder. The cylinder pressure has reached atmospheric pressure a little after 270° aTDC, i.e. after half the exhaust stroke. During overlap between intake and exhaust scavenging valves, there is a 30 kPa pressure drop over the engine, causing large amounts blow-through and poor trapping ratio. There is no evidence of pulse interference between cylinders at blowdown with the 200 CAD duration of the exhaust blowdown valve. However, only moderate 34 kPa boost pressure is produced with open trapping valve.

GAS EXCHANGE IN THE DEP ENGINE – Ideally, the exhaust scavenging system and the cylinder pressure has reached atmospheric pressure when the intake valve opens. However, a very large pressure difference also produces large amounts of blow-through, i.e. fresh charge goes straight through the engine. Excessive blow-through gives poor trapping ratio, especially at low engine speed when valve overlap time is long. Blowthrough can however also be of benefit to increase possible boost pressure when the compressor operates close to the surge limit. To increase the trapping ratio, the trapping valve in the exhaust scavenging system has been introduced after the close-coupled catalyst, see Figure 1. By increasing the pressure in the exhaust scavenging system at low engine speed, the blowthrough can be limited.

With closed trapping valve, the boost pressure reaches 57 kPa, an increase of 23 kPa compared to open trapping valve. All of the exhaust is forced through the turbine causing a higher pressure in the exhaust blowdown system in the second half of the exhaust stroke. The exhaust scavenging system acts as a buffer, filling at the end of the exhaust stroke and during exhaust scavenging/intake valve overlap and emptying its contents into the cylinder during exhaust blowdown/scavenging valve overlap of the following cylinder in firing order. This can be seen in Figure 12 where the pressure in the exhaust scavenging system is higher than the blowdown system pressure shortly after the exhaust scavenging valve has opened. The measured hydrocarbon content in the blowdown system at 2000 rpm with closed trapping valve is 6500 ppm CH4 equivalent, supporting the theory that fresh charge is transported to the exhaust blowdown system via the exhaust scavenging system. By closing the trapping valve, BMEP increases from 1,60 MPa to 1,71 MPa and brake specific fuel consumption decreases 16 %, suggesting better trapping ratio.

Figure 12 shows an example of the cylinder pressure and both exhaust system pressures together with intake pressure in the DEP engine during gas exchange at 2000 rpm with open and closed trapping valve in the exhaust scavenging system. The initial behavior of the blowdown pulse is similar in both cases. The top figure, where the trapping valve is completely open, shows a

Exhaust blow-down Exhaust scavenging Blow-down system

180 160

LIMITING FACTORS – Two key limitations have been identified in the DEP engine; the engine exhibits high turbine inlet temperatures and exhaust valves are choked during the exhaust stroke.

140 120 100

180

Intake system

Scavenging system

Cylinder

Exhaust temperature - As mentioned previously, simulation of the turbocharger is based on performance maps. By comparing measured turbine speed to simulated, the performance of the simulation can be checked. Figure 13 show comparison of turbine inlet temperature and turbine speed between simulation and measurement. As can be seen in the figure the turbine speed correlate well.

160 140 120 100

Valve lift

Pressure [kPa] (closed trapping valve)

Pressure [kPa] (open trapping valve)

Cylinder Intake

Exhaust blow-down valve

180

Exhaust scavenging valve

270

TDC

Intake valves

450

The exhaust gas temperature limit in the engine test was set to 1253 K due to material constraints. For the DEP engine this temperature was reached already at 2250 rpm both in the tests and in the simulation, which can be seen in Figure 13. This is a very low speed compared to a standard turbocharged engine. The most likely explanation to the comparatively high temperature before the turbine can be found in the temperature and

540

Crank angle [aTDC] Figure 12 Pressure in cylinder, exhaust blowdown system, exhaust scavenging system and intake as a function of crank angle after combustion TDC during gas exchange at 2000 rpm with open (top) and closed (middle) trapping valve in the scavenging exhaust system. Valve lift curves are shown in the bottom.

8

130

1100

110

1000

90

900

70

800

50 1000

2000

3000

4000

5000

1200

DEP: 1249 K Std: 1195 K

0.3

0.2

DEP 1000

0.1

Standard t/c 800 90

6000

135

180

225

270

315

0.0 360

Crank angle [aTDC] Figure 14 Temperature and mass flow in the turbine inlet in the DEP engine compared to a standard turbocharged engine at 2500 rpm.

Engine speed [rpm] Figure 13 Comparison of turbine inlet temperature and turbine speed between measurement and simulation.

is also apparent that the scavenging valve is choked during almost the entire displacement phase. The DEP concept would benefit from increased flow area of both exhaust valves; the increase in exhaust valve diameter from 28 mm to 32 mm was not enough to compensate the decrease in total valve lift area. Increased choked period was also seen for the blowdown valve at lower engine speeds however less pronounced. The choking of the scavenging valve is in part explained by the lower temperature in the scavenging system port compared to a standard engine, leading to lower speed of sound and more choking.

mass flow across the turbine for the DEP engine. The gas in the blowdown phase has the highest temperature due to low expansion of the gas. As the pressure decreases, the temperature of the gas also decreases and by the end of the displacement phase the lowest temperature is reached. Due to the fact that the turbine is fed only by the blowdown phase, the comparatively low temperature exhaust gas does not cool down the turbine. Comparing simulation results for a standard turbocharged engine with the DEP concept at 2500 rpm supports this reasoning. In Figure 14 the temperature and mass flow across the turbine is shown. The load in these simulations is around 1.8 MPa BMEP for both engines. There is a slight difference in the position of 50% mass fraction burnt, 31 and 26 CAD aTDC for the standard and DEP engine respectively. Due to the later phasing of combustion the standard engine has a slightly higher peak temperature in the blow down phase. However in the displacement phase the temperatures are comparable. Hence, the problem with the high temperature can be explained by the fact that the mass flow over the turbine for the DEP concept is considerably smaller than for the standard engine during the displacement phase. As a consequence this phase has a much smaller effect in lowering the mass averaged temperature for the DEP concept.

Cylinder pressure [kPa]

250

5000 rpm

200

4000 rpm 3000 rpm

150

2000 rpm 100

1000 rpm

Valve lift

1200

mass averaged temperatures

1400

Temperature [K]

150

Mass flow [kg/s]

simulated

Turbo speed [1000 rpm]

Turbine inlet temperature [K]

measured 1300

50 180

270

360

450

540

630

Crank angle [ATDCf] Figure 15 Cylinder pressure during the gas exchange for increasing engine speeds in the DEP engine at full load with early scavenging exhaust valve opening. Valve lift curves for blow-down and scavenging exhaust valve and the inlet valves are also shown.

Flow over exhaust valves - The engine works as intended at low engine speeds with a significant pressure drop in the cylinder when the exhaust scavenging valve opens. This can be seen in Figure 15 that show measured cylinder pressures at several engine speeds. At speeds over 3000 rpm the piston displacement phase of the exhaust stroke starts to show in the cylinder pressure. This is due to choked flow over the exhaust valves. Figure 16 show Mach number across the exhaust valve at 5500 rpm. The DEP engine shows a substantially longer period of choked flow during the blowdown phase compared to the standard turbocharged engine, shown by the arrow in Figure 16. It 9

1.4

scavenging valve

blow-down valve

0.50 0.25

1.00 0.75 0.50

standard t/c

(4)

0.8

Heat release rate 0.6

(2) (3)

0.4

(1)

0.0 -90

0

90

180

270

360

Crank Angle [aTDC] Figure 18 Cylinder pressure traces from cold start tests with a standard gas exchange system and the DEP system.

0.00 90

180

270

360

Crank angle [aTDC]

The gas exchange was significantly changed during cold start in the DEP engine compared to a standard turbocharged engine. The combustion phasing is very late in order to heat the catalyst. This, in combination with the short valve duration of the scavenging valve, leads to a rather peculiar cylinder pressure trace as shown in Figure 18. Four observations can be seen (numbered in the figure):

Figure 16 Comparison of simulated Mach number across exhaust valve between DEP engine and standard turbocharged engine at 5500 rpm.

COLD START – The power potential during cold start with only the scavenging valve active had to be investigated, since the engine has to run with the turbocharger inactive until both catalysts have reached light-off. The exhaust scavenging valve duration was 174 CAD compared to 240 CAD for the standard turbocharged engine. The most likely scenario during the first 20-30 seconds is low speed driving, similar to the speeds and loads found in for example the EC2005 emission cycle. These loads are typically below 10kW. Therefore the power demand on the engine is rather low. In spite of the extremely short duration of the exhaust period, the engine was capable of producing torque in the same range as a naturally aspirated 1.6 liters engine as seen in Figure 17. Even with a low gearing this torque is sufficient to handle a 1700 kg vehicle.

1. The exhaust emptying period is very short for the DEP concept. 2. The cylinder content is recompressed before the scavenging valve opens. 3. The blow down interference from the following cylinder in firing order is eliminated. 4. The combustion is much faster with the DEPconcept. The short exhaust emptying period is naturally due to the shorter cam duration, which also causes the recompression of the cylinder content. The latter lead to an increased pumping work, which in turn lead to an increased inlet pressure requirement in order to maintain the engine speed. The increased inlet pressure during cold start of the DEP engine, combined with the elimination of the interfering exhaust pressure pulse during the valve overlap lead to decreased residual gas content and increased combustion rate. The combustion phasing is therefore slightly earlier in the DEP engine than for the standard engine with the same ignition timing. The ignition timing that was used during the tests was the same as the calibration for the standard engine. Therefore the temperature in the exhaust manifold was slightly lower for the DEP engine than with the standard engine.

160 860

820

150 145

780 140

BMEP [kPa]

155

Torque [Nm]

1.0

0.2

0.25

Various cam settings 740

135 130 800

DEP Standard t/c

1.2

Pressure [MPa]

DEP engine

0.75

0.00

Standard engine

Mach number

1.00

1200

1600

2000

2400

Engine speed [rpm] Figure 17 Torque with only a 174 CAD scavenging valve activated.

10

Std t/c 1 cat Std t/c 2 cat DEP 1 cat DEP 2 cat

and cooling system in the exhaust blowdown system, from exhaust blowdown valve to the turbine. Since the cold start catalyst light-off is handled by the close coupled catalyst in the scavenging system, aggressive measures can be taken to lower the exhaust gas temperature in the blowdown system without affecting catalyst light off time and cold start emissions. Furthermore, the aging problem of the catalyst, due to high exhaust temperatures, is also of less concern. The close-coupled catalyst is placed in the scavenging system where the exhaust temperature is low and the main catalyst can be placed under the floor of the vehicle where it can be cooled by the airflow underneath the vehicle. This in turn allows a higher temperature out from the turbocharger.

Time [s]

~38%

300°C before cat

300°C after cat

It appears that the scavenging exhaust system plays a major role in the gas exchange even when the trapping valve in the scavenging exhaust system is closed. By varying the trapping valve opening, the exhaust backpressure during the displacement part of the exhaust stroke and during exhaust scavenging/intake valve overlap can be adjusted. This can be used to control the pressure difference over the engine and limit blow-through. Partially closing of the trapping valve can also provide the turbine with increased mass flow and higher enthalpy by increasing the cylinder pressure during exhaust blow-down/scavenging valve overlap.