Vous êtes sur la page 1sur 241
AUTOMOTIVE ENGINE MODELING FOR REAL TIME CONTROL by JOHN J. MOSKWA B.S.E., University of Michigan (1980) MS.E., University of Michigan (1981) SUBMITTED TO THE DEPARTMENT OF MECHANICAL ENGINEERING IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY IN MECHANICAL ENGINEERING at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY May 1988 © John J. Moskwa, 1988 ‘The author hereby grants to M.L'T. permission to reproduc? and to distribute copies of this thesis document in whole or in part. Signature of Author ‘Department of Mechanical Engineering ‘May 13, 1988 Certified by Professor J.K. Hedrick oe ‘Thesis Supervisor Accepted by = a> a Professor A.A. Sonin Chairman, Mechanical Engineering Department Committee ASGAOHUSETTS | STITUTE "OF TECHNOLOGY SEP 06 1988 AUTOMOTIVE ENGINE MODELING FOR REAL TIME CONTROL by John J. Moskwa Submitted to the Department of Mechanical Engineering on May 13, 1988 in partial fulfillment of the requirements for the Degree of Doctor of Philosophy in Mechanical Engineering ABSTRACT In the automotive industry, interactions between the engine and transmission during a shift (torque production and clutch control) are not optimized to provide a smooth shift and minimize clutch energy dissipation. Closed loop control of the engine during a shift (throttle, spark advance, and fuel) can improve the shift in two important ways: © minimize the decrease in transmission output shaft torque as power is transferred to a lower gear ratio (i.e. fill the torque hole), and © decrease energy dissipation as the slip across the oncoming clutch goes to zero. With a part throttle shift, engine control can be used to achieve the “ideal” shift (ie. vehicle jerk approximately zero), within the physical constraints of the system. A nonlinear port fuel-injected dynamic engine model has been developed for the design and implementation of engine control algorithms. This mean torque predictive model has been experimentally validated, and includes: © intake manifold dynamics, © fuel delivery dynamics, and © process delays inherent in the four-stroke engine. ‘The model is compact enough to run in real time, and can be used as an embedded model within a control algorithm or an observer. Although developed and validated for a specific engine, the model can be adapted to represent various types of automotive engines with a limited amount of engine data. Nonlinear engine contro! algorithms are developed using the technique of Vari- able Structures or Sliding Modes. These algorithms are used, in conjunction with clutch control algorithms, to enhance the quality of the shift. Simulation results show that closed-loop engine control can provide the benefits outlined above. Complete algorithms will be implemented on control development vehicles at General Motors Research Laboratories in July, 1988. Thesis Supervisor: J. Karl Hedrick Title: Professor of Mechanical Engineering ACKNOWLEDGMENTS This thesis is dedicated to my parents, Mary and Joseph Moskwa, who have in- stilled in me an appreciation for higher academic achievements. I have been fortunate enough to have the opportunities that were not available to them. My thesis committee has provided invaluable direction, support, and encourage- ment when they were needed most. Special thanks go to Prof. J. Karl Hedrick, Prof. John B. Heywood, and Prof. James C. Keck. The contributions by members of the General Motors Corporation are numerous. My gratitude for support (both financial and moral) go to the leadership of G.M. Research Laboratories, Power Systems Research Department; Dr. Hazem Ezzat, Mr. Gerald Skellenger, and Dr. Alex Alexandridis. This project required pulling together help from many different departments within G.M., but special thanks go to Gerry Pierce and Carol Hutton for providing the engine, instrumentation/control hardware and advice, Glenn Koch and Bill Truby for test cell set-up and support, Don Brown and Reggie DiRezze for advice on how to expedite all manner of things, Herb Vikstrom for valuable consultation, and Kumar Hebbale for help on the mainframe and an enjoyable lunch partner. In addition, significant assistance, in either advice or data, was received from Larry Carrion, Man-Feng Chang, Mike Crenshaw, Keith Dickey, Joe Hart, Brian Heil, Ken Henry, Gerry Januzzi, Dan Lanford, John Lingell, Greg Long, Jim Macey, Dave Marvosh, Ralph Platt, John Priestly, Dave Rule, Mike Rusaw, Greg Schleicher, and Patricia Wombwell. Your help is appreciated. Thave shared the laughs and frustrations with my friends at the Vehicle Dynamics Laboratory, and they have made this an enjoyable, albeit difficult, year. Thank you Bob, Keith, Jahng, Bill, and Radhouan. Finally, my deepest affection goes to my talented family, Kathlyn and Susan, for the love, support, encouragement, and inspiration to make this work possible. I would not have been able to complete this work without your help. We have worked very hard, and have managed to enjoy some fun times, but the best is yet to come. Table of Contents Title Page ee 1 Abstract so ee 2 Acknowledgement... ee ee eee 3 Table of Contents... ee 4 List of Figures 2. ee ee 6 List of Tables... eee 9 Nomenclature 2. oe eee eee 10 Chapter 1 IC. Engine and Control Background... 2... . 15 LL Introduction. 2. ee 15 1.2 Engine History 0.0. ee 7 1.3 Engine Control History... . . 22 1.3.1 Torque/Speed Control 22 1.3.2 Pollution Control... . 25 1.4 Engine Torque Control... . . . 32 1.5 Engine Modeling for Control Development ............- 39 1.6 Thesis Organization ©... 2. ee eee 42 Chapter 2 Model Description 44 2.1 Introduction 44 2.2 Throttle Body Model 44 2.3 Idle Air Control Model». 2... ee 48 2.4 Intake ManifoldModel . 0... eee eee 49 2.5 Fuel Delivery Model... 2... eee ee 56 2.6 Torque Production Model. 2... ee ee ee 63 2.7 Model Summary... 73 Chapter 3 Model Modifications and Validation Tests... ........ 79 3.1 Introduction. 6. ee 79 3.2 Throttle Body, Idle Air Control, and E.G.R.Flows.......... 80 3.3 Intake Manifold . 6 83 3.4 FuelDelivery 2... 87 8.5 Torque Production. 6... 92 3.6 Additional Topics ©... 95 Chapter 4 A Case Study in Engine Validation 4.1 Introduction. ©... 4.3.1 Throttle Body 2... 1. eee ee ee ee 4.3.2 Idle Air Control ©... ee 4.3.3 Exhaust Gas Recirculation ©... 1... eee ee ee 4.4 Fuel Delivery 4.4.1 Fuel Injector Calibration .. 2... 00.0... 2 eee 4.4.2 Dynamic Fuel Parameter Identification 4.5 Torque Production. 2... 2... ee 4.5.1 Spark Influence... 2. ee ee 4.5.2 EGR. EffectonMBT . 2... 2.0.00. 000 eeu ee 4.5.3 Torque Function... 2. ee ee 4.5.4 Friction/Pumping Torque... 2... ee ee 4.5.5 Air/Fuel Influence... 2... 0.0.0... e eee 4.6 Engine-Dynamometer System Analysis... . 2... ...-.0. 4.7 Final Model Comparisons. ©... 1 2 - ee eee ee ee Chapter 5 Nonlinear Engine Control . 2... 0. 2 ee ee ee 173 5.1 Introduction. ©... ee 173 5.2 Engine Control Algorithms . 2... 6... eee ee ee ee mT 5.2.1 Simplified Model Stability ©. 2... 2... eee eee 379 5.2.2 Engine Speed Control, C.L. Spark Advance... 22... 0. 184 5.2.3 Engine Speed Control, C.L. Throttle and Spark Advance... . 187 5.2.4 Gain Weighting ©... 2... eee ee 190 5.2.5 Turbine Speed Control... 2... ee ee ee 194 5.2.6 Coordinated Engine/Transmission Control ..........- 200 Chapter6 Summary ©... 2. -.- 1 eee ee 206 G1 Results... ee 206 References © 6 6 ee ee 21 Appendix 1 Throttle Body Equations. ............0.004 27 Appendix 2 Torque Converter and Transmission Models ........- 222 Appendix 3 Sliding Mode Theory ©... 1.2... ee ee ee 230 Figure 1.1 Figure 1.2 Figure 1.3 Figure 1.4 Figure 1.5 Figure 1.6 Figure 1.7 Figure 1.8 Figure 1.9 Figure 1.10 Figure 1.11 Figure 2.1 Figure 2.2 Figure 2.3 Figure 2.4 Figure 2.5 Figure 2.6 Figure 2.7 Figure 2.8 Figure 2.9 Figure 2.10 Figure 2.11 Figure 2.12 Figure 3.1 Figure 3.2 Figure 3.3 Figure 4.1 Figure 4.2 Figure 4.3 Figure 4.4 Figure 4.5 Figure 4.6 Figure 4.7 Figure 4.8 Figure 4.9 Figure 4.10 Figure 4.11 Figure 4.12 Figure 4.13 Figure 4.14 List of Figures Four-stroke 8.1. Engine Process... 0. ee es Cylinder Pressure Time History . . 2... ee eee ‘Two-stroke S.I, Engine Process... 1 ee ee ee eee Four-stroke Diesel Process 2... eee eee Ball-weight Governor 6... ee ee ee Ball-weight Governor Action ©. 2... ee ee ee eee Catalytic Converter Efficiency 6} ©... 2.0... - eae Oxygen Sensor Output (6) 2. ee Clutch Torques During a1—2Shift (8)... ... 2.0.24. Shaft Torque During a1+2Shift[8]............. Simplified Block Diagram of Air and Fuel Transfer Functions. . Intake Manifold Response to Throttle Step... 2... 5 Fuel Injection Hardware [31]... 2. eee eee Fuel Injector [31]. es Inlet Port [31] 2. 6 ee Inlet Port Fuel Dynamics. 2. ee Fuel Delivery Model ©... 2 2. ee ee eee Simplified Fuel Delivery Model. 2... 0 eee eee Simplified Fuel Delivery Model . 2 2... ee eee Torque Production Model 2... 1-2. eee eee ‘Air/fuel Influence Function (AFI) 2... 0 eee eee Firing and Motoring Pumping Loops... ..-...---- Friction/pumping Torque Model... 2.2.2 ee ae Volumetric Efficiency Breakdown [35]... 0... ee ee Simple Carburetor Schematic [5]... ...-.00-00. Simultaneous Double Fire Fueling Algorithm Transition a Experimental Test Set-up, G.M. Research Laboratories . . . . . Air/fuel Sensor Calibration Set-up .. 2. 2 eee ee eee Air/fuel sensor Response to Air/fuel Ratio Step... .. 2. Throttle body discharge coefficient .. 2... 1... ee eee Pressure ratio discharge coefficient .. 2... 0-0-0 Throttle characteristic (TC)... ee Pressure ratio influence (PRI)... 2 ee ee es LA.C. calibration data 2... ee ee LAC. calibration ©... ee Fuel Injector Calibration ©... ee ee ee Typical Fuel Perturbation Test ©... 1.0... ee eee Air/fuel Influence and Derivative... 2-20. + +05 Typical Estimation Results 2... 2... ee eee ee ee Spark advance test sequence (spark and torque)... ... ~~ ~6- Figure 4.15 Figure 4.16 Figure 4.17 Figure 4.18 Figure 4.19 Figure 4.20 Figure 4.21 Figure 4.22 Figure 4.23 Figure 4.24 Figure 4.25 Figure 4.26 Figure 4.27 Figure 4.28 Figure 4.30 Figure 4.31 Figure 4.32 Figure 4.33 Figure 4.34 Figure 4.35 Figure 4.36 Figure 5.1 Figure 5.2 Figure 5.3 Figure 5.4 Figure 5.5 Figure 5.6 Figure 5.7 Figure 5.8 Figure 5.9 Figure 5.10 Figure 5.11 Figure 5.12 Figure 5.13 Figure 5.14 Figure 5.15 Figure 5.16 Figure 5.17 Figure 5.18 Figure 5.19 Spark influence based on brake torque... ......... 132 Spark influence based on indicated torque .......... 133 MBT dataand model .....-.--...0-.000- 135 MBT Change as a Function of @EGR. ........005 136 Estimation of Torque Function... 2... eee ee ee 138 ‘Motoring friction/pumping torque data... ........ 143 Ideal pumping torque 2 145, Ideal motoring pumping loop with heat transfer... . . . . 146 Gross Indicated Fuel Conversion Efficiency [32] ........ 148 Brake Fuel Conversion Efficiency from Test Data ....... 150 Friction/pumping Torque... ee 151 Intake Manifold Plenum and Runner Pressure... .....- 158 Air/fuel Influence for 3000 rpm and 47 g/sec MAF... . 159 Air/fuel Influence Function .. 2... ..-.020..048 161 Engine-Dynamometer model»... ......2-0-08 162 Shaft Torque Sensor Spectrum, 1200rpm ..... 2... 164 Shaft Torque Response to Spark Step... .. 22-2... 166 Model Comparisons with Test Cell Data... ........ 169 Model Comparisons with Test Cell Data... .......- 169 Model Comparisons with Test Cell Data... 2... 2... 170 Model Comparisons with Test Cell Data... .......- 170 Model Comparison, Unfiltered Shaft Torque ......... 171 Control of First Order Lag with Delay .. 2... 2.0.0. 180 Root-Locus,we=100 . 2 ee ee ee 181 Root-Locus, w, Open-loop Throttle, Closed-loop Spark Control... ..... 187 Closed-loop throttle and spark control»... 2... ee oe 188 Closed-loop throttle and spark control (Sf=1) ........ 190 Closed Loop Throttle and Spark Control (%#=1) ....... 191 Closed Loop Throttle and Spark Control (2=05) ....... 192 Closed Loop Throttle and Spark Control (#=12) ....... 193 Open-Loop Shift Engine Controls and Turbine Speed... . . 193 Open-Loop Shift Slip Speeds and Clutch Torques ...... . 194 Closed Loop Turbine Control...) ee ee ee 195 Closed Loop Turbine Control»... ee ee ee 196 Closed Loop Turbine Control... 2... ee ee 196 Closed Loop Turbine Control... 2. ee ee ee 197 Closed-Loop Turbine Speed Control 24. Closed-Loop Turbine Speed Control =. Shift Algorithm Engine Controls and’ Tndicated Torque ..... 201 Shift Algorithm Turbine Speed .. 0... .0......0% 202 Figure 5.20 Shift Algorithm Carrier Speed . 2.2... 2... eee 202 Figure 5.21 Shift Algorithm Clutch Torques ©... 2... eee ee 203 Figure 5.22 Shift Algorithm Output Shaft Torque ...........~ 203 Figure A1-I Side View of Throttle Plate... 2... ...----00- 218 Figure A1.2 Axial View of Throttle Body Throat... .......... 218 Figure A2.1 Torque Converter and ‘Transmission Schematic [8]... .. . . 223 Figure A2.2 Simplified “clutch-to-clutch” Shift Schematic... .. 2... 224 Figure A2.3 Bond Graph for First Gear... 2... 0 ee ee eee 225 Figure A2.4 Bond Graph for Torque Phase of 1+2Shift 11... 226 Figure A2.5 Bond Graph for Speed Phase of 12Shift .......... 227 Figure A2.6 Bond Graph for Second Gear 2... 02.0... ee 228 Figure A3.1 Illustration of the Sliding Surface 2... 1.0. ....00. 234 Figure A3.2 Signum and Saturation Functions ........-.-00. 237 Table 4.1 Table 4.2 Table 4.3 Table 4.4 Table 4.5 List of Tables Air Flow Gradient Data 22.2.2. 00 ee ee Dynamic Fuel Validaton Results - N=1200 rpm, MAF=18 g/s . . Dynamic Fuel Validaton Results - N=2000 rpm, MAF=32 g/s . . Dynamic Fuel Validaton Results - N=3000 rpm, MAF=47 g/s . . Dynamic Fuel Validaton Results - N=4000 rpm, MAF=55 g/s . . 126 127 127 127 NOMENCLATURE A(@) throttle cross-sectional area (m?) Ac clutch area times effective radius (m*) AFI air/fuel influence on indicated torque A/F air/fuel ratio B/L bore to stroke ratio ¢ shaft damping coefficient (N- m- s/rad) Ca discharge coefficient Cay discharge coefficient as a function of throttle angle Caz discharge coefficient as a function of pressure ratio D throttle bore diameter (m) d throttle rod diameter (m) DEGON injector pulse width in crankshaft degrees E.G.R. exhaust gas recirculation 1.A.C. idle air control IVC intake valve close IVO intake valve open jvt Jq dynamometer polar moment of inertia (kg- m?) Je engine polar moment of inertia (kg- m?) Jez4 effective engine polar inertia (kg- m2) K shaft stiffness (N- m/rad) k ratio of specific heats LQR linear quadratic regulator LQR/LTR linear quadratic regulator loop transfer recovery tines mass air flow into intake manifold (g/sec) tao mass air flow out of intake manifold (g/sec) Titegei mass E.G.R. into intake manifold (g/sec) ~10- Megro hve tye ragga ray ss tinge ringer tatac Ma ma M Mgr Megr MA MAC MAP MBT MFF N ny P Ps Po PB Pm P, Py PCV. mass E.G.R. out of intake manifold (g/sec) flow rate of nondelayed fuel (g/sec) fuel command (g/sec) 2 rev. delayed fast fuel (g/sec) nondelayed fast fuel (g/sec) fuel flow rate entering cylinder (g/sec) lagged slow fuel (g/sec) mass air flow through idle air control valve (g/sec) molecular weight of air (kg/kmole) mass of air in intake manifold (g) molecular weight of air or E.G.R. (kg/kmole) molecular weight of E.G.R. (kg/kmole) mass of E.G.R. in intake manifold (g) maximum flow through throttle, i.e. choked w.o.t. (g/sec) mass of air per cylinder (g) intake manifold absolute pressure (kPa) best torque spark advance (degress before top dead center) mass fuel flow (g/sec) engine speed (rev/min) number of moles of i** constituent intake manifold pressure (kPa) intake manifold air partial pressure (kPa) clutch pressure (kPa) exhaust manifold pressure (kPa) intake manifold pressure (kPa) stagnation pressure (kPa) throttle throat pressure (kPa) positive crankcase ventilation -11- PRI Pw Qhe Pl Ri Te Ra normalized throttle flow as a function of pressure ratio pulse width (sec) lower heating value of fuel (kJ/kg) ideal gas constant (kJ/kg: K) universal gas constant (kJ/kmole- K) gear ratio in gear i (<1) compression ratio differential gear ratio oscillation magnitude (N- m) Laplace variable (sec~*) sliding mode surface spark advance (degrees before top dead center) spark influence on torque temperature brake torque (N- m) first clutch torque (N- m) second clutch torque (N- m) exhaust gas temperature (Kelvin) friction and pumping torque (N- m) indicated torque (N- m) stagnation temperature (Kelvin) torque converter pump torque (N- m) shaft torque (N- m) torque converter turbine torque (N- m) approx. cylinder temperature at top dead center (Kelvin) normalized throttle flow as a function of throttle angle torque function (N- m/g) intake manifold volume (m°) -12- A®linssive c $ ” Ne Ne mt NSA ne Te % cylinder volume (m*) displaced volume of engine (m*) clearance volume (m*) work (N- m) slow fuel split fraction fraction of fuel entering cylinder before current intake valve closes multiplicative E.G.R. error time delay (sec) time delay, intake to torque (sec) time delay, intake to spark (sec) time delay, spark to torque (sec) slip speed on first clutch (rad/sec) slip speed on second clutch (rad/sec) crankshaft angle from start of injection to intake valve closing (degrees) fuel split parameter damping coefficient sliding mode gain sliding mode gain on throttle combustion efficiency fuel conversion efficiency sliding mode gain on spark advance thermal efficiency volumetric efficiency throttle angle from closed (degrees) closed throttle angle w.r-t. throttle plane (degrees) i sliding mode gain ue desired bandwidth (rad/sec) clutch friction coefficient -13- wu parameter vector intake manifold density (kg/m) standard deviation lag time constant (sec) liquid fuel time constant (sec) slow fuel lag time constant (sec) sliding mode boundary layer goodness of fit parameter shaft wind-up angle (radians) reaction carrier speed (rad/sec) dynamometer speed (rad/sec) engine speed (rad/sec) torque converter pump speed (rad/sec) torque converter turbine speed (rad/sec) lagged turbine speed for torque converter calculation (rad/sec) -14- 1 I.C. Engine and Control Background ion The use of automatic or feedback control systems has dramatically changed the world in which we live. These systems make our lives more comfortable and often perform, or assist in performing, tasks which a person would normally be unable to perform. These systems range from thermostat temperature controls, used on most home furnaces, to the exotic algorithms designed to manage the control surfaces of modern jet aircraft or the space shuttle. The application of appropriate control algo- rithms can transform an unmanageable, complex system into a “user-friendly” system whose final task is pleasing and is performed with expediency and accuracy. The focus of this thesis is the management of torque produced by the internal combustion (I.C.) engine of an automobile to perform specific tasks, which either raise the comfort level of the occupants, or enhance the “driveability” of the automobile. Specifically, a dynamic internal combustion engine model will be detailed either for the purpose of control algorithm development (simulation model), or to be used as a real-time embedded model within the control algorithm (control model). An outline is given of changes within the model or the tests, that are necessary to adapt the model to a specific engine of interest. Many of these validation tests will be carried out on a naturally aspirated, port fuel-injected, V-6 engine, in order to develop a usable dynamic engine model for algorithm development. This model is being used by the -15- Chapter 1: I.C. Engine and Control Background Vehicle Dynamics Laboratory of M.LT. and the Power Systems Research Departmentt of General Motors Research Laboratories. Furthermore, engine controi algorithms will be developed, using a nonlinear design technique called “sliding modes”, to control the engine speed, or torque, for a desired objective. This thesis is written for the control engineer. The models are physically based and are intended to capture the major dynamics (lags and delays) which are inherent in the torque production process, and are of interest to the control engineer. Combustion processes are modeled using fuel conversion efficiencies, and the cylinder by cylinder torque pulses are not modeled. Rather, this is a mean torque-predictive dynamic model for control design. The goals of the modeling effort are these: to develop a mean torque predictive dynamic engine model, simplicity is important for real-time simulation within 2 control algorithm, the controls will be the throttle angle, the spark advance, and the fuel flow rate of the injectors, the disturbances will be external torques and E.G.R. (exhaust gas recirculation) flow into the intake manifold. The goals of the validation experiments are driven by the model structure, which dic- tates the parameters of interest. These procedures will be outlined in detail in Chapters 3and 4, tf sponsors of this research -16- Chapter 1: 1.0, Engine and Control Background 1.2__Engine History Control of-the internal combustion engine has been important since its inception, but the inventor or engineer has not always had the control tools which are available today. The four-stroke engine, as we know it today, was invented by Nicolaus A. Otto (1832-1891) [1], and the first prototype was run in 1876. This engine had an overall efficiency of about 14%. The ideal thermodynamic cycle for this engine is called the Otto cycle in honor of the inventor. This engine was an improvement on an atmospheric engine which was invented in 1867 by Nicolaus Otto and Eugen Langen (1833-1895). The unique improvement or innovation of this new engine was that the mixture was compressed prior to combustion, which boosted the power to weight ratio by more than seven times. Actually, a French inventor named Alphouse Beau de Rochas (1815-93) was issued a French patent in 1862, which described the four stroke cycle and detailed the conditions under which maximum efficiency could be achieved. Although these writings predated Otto’s invention, and described the cycle in great detail, no working hardware was produced by Beau de Rochas. Therefore, the credit for the invention of the four-stroke cycle spark ignition (S.L), internal combustion engine is generally given to Otto. An illustration of the four-stroke process is given in Figure 1.1. During the intake stroke, premixed air and fuel are drawn into the cylinder from the intake manifold, past the open intake valve(s). This mixture is often throttled, or below ambient pressure, unless the engine is turbocharged or supercharged. The mixture is then compressed to a volume much smaller than the volume when the intake valve closes. Modern IC. engines typically ignite the mixture anywhere from 50° to 0° before top dead center (B.T.D.C.), to get optimal torque while avoiding knock or precombustion problems Chapter 1: I.C. Engine and Control Background spark iy i oA = oa a LC Intake Compression Expansion Exhaust Figure 1.1 Four-stroke $.I. Engine Process due to the mixture remaining at a high pressure and temperature for an extended period of time. The maximum cylinder pressure occurs during the compression stroke (about 15° after top dead center for optimal spark timing), and the cylinder pressure is translated into torque via a slider-crank mechanism. In the final stroke most of the burnt mixture is expelled from the cylinder past the open exhaust valve and into the exhaust manifold. A small residual fraction of exhaust gases remains in the cylinder, because the cylinder volume does not go to zero at top dead center. Figure 1.2 shows the cylinder pressure history and the timing of events from a modern I.C. engine. Word of Otto’s new engine spread quickly throughout Europe and England, and by 1890 several engineers (James Robson; 1833-1913, Karl Benz; 1844-1929, and Dugald Clerk; 1854-1913) had developed working prototypes of a two-stroke cycle engine. The two-stroke process consists of a compression and a power stroke as shown in Figure 1.3. -18- CYLINDER PRESSURE (KPA R 10! Chapter 1: I.C. Engine and Control Background a a CCRANKSHRET ANGLE (OEGREES) Figure 1.2 Cylinder Pressure Time History l\ ax SN Oo reed valves” ‘Compression Expansion, Exhaust - Scavenging Figure 1.3 Two-stroke S.I. Engine Process -19- Chapter 1: I.C, Engine and Control Background Near the end of the power stroke, ports in the cylinder wall are uncovered by the piston, and the gases start to exhaust. Shortly thereafter, intake ports are uncovered on the opposite side (for cross-scavenged) or the same side (for loop-scavenged) of the cylinder, the burnt gases are purged from the cylinder, and the compression stroke begins. For uniflow scavenging, the intake ports are located around the base of the cylinder, and exhaust valves are located in the cylinder head. A major problem with two-stroke engines is the mixing of burni and fresh mixture during scavenging, so to completely purge the cylinder some fresh mixture must also be exhausted, which increases the hydrocarbon emissions. Finding the optimal port placement so that the exhaust port is covered when the cylinder has just filled with a fresh charge is a difficult task. The available fuels that could be used in these early engines were prone to knocking or precombustion, so the compression ratios had to be low (4:1) compared to today’s spark ignition engines (8-12:1). This in turn limited the exparsion ratio which limited the amount of work, and the efficiency, that could be produced by these engines. Low compression ratio engines, which were fueled by kerosene, became popular throughout Europe and the United States in the late 19th and early 20th centuries as a source of power. ‘The compression ignition engine was another significant step in engine development, and was the result of work by Rudolf Diesel (1858-1913). In 1892 he was issued a patent for an engine in which the fuel was injected directly into the cylinder after the air was compressed. Figure 1.4 shows the four-stroke diesel cycle. During the intake stroke, air is brought into the cylinder past the open intake valve, The intake air is never throttled, so it will be at ambient pressure or above, depending on whether the engine is turbocharged, supercharged, or naturally aspirated. The air is then -20- Chapter 1: I.C. Engine end Control Background Py wy c Fo Eo] OS Intake Compression Combustion, ‘Exhaust Expansion Figure 1.4 Four-stroke Diesel Process compressed to a high pressure and temperature. While compression ratios in modern spark ignition engines are in the range of 8-12:1, compression ignition engines are in the range of 16-24:1. With this higher compression ratio comes a higher expansion ratio, and consequently more work per cycle. The higher compression ratios are possible because only air is being compressed, rather than a combustible mixture, as in the spark ignition engine. As the piston approaches top dead center, fuel is injected directly into the cylinder. The high temperature of the compressed air ignites the fuel as soon as the fuel is sufficiently mixed with the air to within an appropriate air/fuel ratio. The diesel engine always has excess air (i.e. overall air/fuel ratios much greater than stoichiometry) because of the need to mix the injected fuel quickly with air in the combustion chamber for combustion. The power and exhaust strokes follow. Two- stroke diesel engines are also produced; they differ from two-stroke S.I. engines in the same manner as four-stroke engines. That is, the intake is non-throttled air, high -21- Chapter 1: 1.C, Engine and Control Background compression ratios, the fuel is injected directly into the combustion chamber near the end of the compression stroke, and fuel/air mixing occurs in the combustion chamber. Diesel engines can also be classified by the type of injection as direct injection (D.I.) and indirect injection (I.D.I.). Direct injection engines inject the fuel directly into the main combustion chamber. Turbulence and swirl caused by the intake process, squish by the piston, and the injected fuel cause the mixing of fuel and air. In the indirect injection engine, fuel is ignited in a pre-chamber, and the hot gases expanding out of the pre-chamber cause additional turbulence and mixing in the main combustion chamber. This was a significant invention since the knock problem was circumvented, because the fuel did not have to be compressed with the air. As a result, higher compression ratios, and subsequently higher expansion ratios, were achieved which raised the diesel engine efficiency to twice that of the Otto engine. Several practical problems had to be overcome, and the development of a working prototype took five years. 1.3__Engine Control History 1.3.1_Torque/Speed Control Control of these S.I. engines is accomplished through medulation of the fuel flow, air flow, or spark advance. Typically, oil engines were speed limited, and ran at a near constant speed. The speed was limited by means of a ball-weight governor. The governor would spin on a vertical axis in proportion to the engine speed. As the engine speed increased, the weights of the governor would rise due to the “centrifugal force” effect, which is a result of the higher centripetal force required at high speeds to make the balls-weights travel in a circular path. The greater angle ¢ at higher speeds can be translated into a linear displacement through a linkage mechanism, which can then -22- Chapter 1: IC. Engine and Control Background I é 1 ' 1 | High Speed Low Speed 1 1 Centripetal Force me 6 ng Figure 1.6 Ball-weight Governor Action -23- Chapter 1: I.C. Engine and Control Background control either fuel, air, or spark on the engine. Typically, the spark was controlled on the small I.C. “oil? engines. When engine speed rose above some predetermined value, the spark would not fire and the engine speed would decrease until it was below a set-point level. The ball-weight governor is probably the most commonly used engine control device. The output from these governors also controlled steam valves on steam engines. This device is still widely used to control many types of engines, including centrifugal spark advance control on spark ignition engines and fuel control on modern compression ignition engines, although the weights are retracted by springs rather than gravitational forces. ‘Most modern gasoline engines are controlled by throttling the air into the intake manifold. The control on which an operator would have direct influence would be the throttle plate angle. The proper amount of fuel is mixed with this air and atomized, usually by means of a carburetor. In the carburetor the air flows through a venturi, which is a converging-diverging nozzle. The fuel flow rate is limited by a needle valve and driven by the pressure differential, which is a function of the air flow rate, between the throat of the venturi and atmospheric pressure. The spark advance was typically adjusted manually to get the smoothest running engine with the best power. It was not uncommon in early automobiles for the spark advance to be a control adjustment made by the driver as he drove along the road. In a compression ignition engine, the operator typically controls the fuel flow rate of the injectors by controlling a fuel rack at the injector, varying the fuel pressure to the injector, or varying the fuel delivered to the injectors from a jerk pump, depending on the type of engine. The centrifugal governor then limits the highest speed of the engine by limiting the maximum fuel flow rate. —24- Chapter 1: I.C. Engine and Control Background 1.3.2. Pollution Control In 1952 Prof. A.J. Haagen-Smit [2] showed that reactions of hydrocarbons and oxides of nitrogen in sunlight produced smog. It was soon shown that the exhaust from internal combustion engines was a major contributor to these pollutants. These findings, along with an increased awareness of other air pollution problems, led to limits imposed by the federal Environmental Protection Agency on the rate of pollutants that can be given off by vehicles. To meet these requirements several changes were made on the engine. One change is the injection of air into the exhaust manifold to allow any excess hydrocarbon to burn in that manifold. Another change which affects the torque production of the engine is the introduction of burnt gases from the exhaust into the intake manifold or E.G.R. (exhaust gas recirculation) to lower NOx formation. Nitric oxide (NO) is the predominant oxide of Nitrogen produced in the internal combustion engine. With NO being formed from reactions of N, 0, Na, G2, and OH: 0+M=NO+N N+0,=NO+0 N+OH=NO+H, the rate of NO formation for near stoichiometric mixtures [3] is given by: MEO. S20" cap) + [02] - [4a], moles/em® -s, (a) where [ ], denotes the equilibrium concentration. Because of the exponential term in (1.1), the formation rate of NO rises dramatically as cylinder temperatures rise. Small quantities of the exhaust gas (up to 15% of the total inlet mixture) lower the maximum combustion temperature in the cylinder by functioning as a diluent and by increasing the effective heat capacity of the burnt mixture, since no new fuel energy -25- Chapter 1: ILC. Engine and Control Background is added with the E.G.R. (i.e., exhaust gases can absorb heat from combustion). The lower combustion temperatures decrease the formation of oxides of nitrogen, since the formation rate is decreased, and help to meet emission requirements. E.G.R. flow can affect engine response, and is of interest to the control engineer. While this is certainly an oversimplification of the total NOx formation dynamics, it illustrates the temperature dependence of the formation rates. 100 ° NO, « 0 V/ ; © } \ 0 a i 40 0 CONVERSION EFFICIENCY (20 20 10 no 88 WO Wi 2 a HA WS We Te? He 8 15.0 RIR/FUEL RATT Figure 1.7 Catalytic Converter Efficiency [6] A major contribution to lowering the exhaust emissions from the automotive engine is the development of the three-way catalytic converter [4]. The use of the three- way catalytic converter made strict control of the engine's air/fuel ratio a necessity, and ushered in the age of closed-loop electronic fuel control. The catalytic converter uses small amounts of precious metals (platinum, rhodium, palladium) as catalysts to -26- Chapter 1: IC. Engine and Control Background control the concentrations of hydrocarbons (HC), carbon monoxides (CO), and oxides of Nitrogen (NOx) in the engine exhaust. An oxidizing atmosphere is required for CO and HC conversion, and a reducing atmosphere is required for converting NOx. Figure 1.7 shows how the conversion efficiency varies as a function of the exhaust air/fuel ratio. All three of the pollutants can be effectively controlled if the air/fuel ratio of the exhaust is constrained to a narrow band near stoichiometry. Three other pieces of hardware are needed to accomplish this goal; a sensor to measure the air/fuel ratio of the exhaust, an actuator to change the air/fuel ratio going into the engine, and a microprocessor to control the actuator based on sensed quantities. The typical air/fuel ratio sensor, commonly called an oxygen or lambda sensor, consists of a Yttria (¥s0s) stabilized Zirconia (Zr02) ceramic electrolyte, usually plated with platinum [5]. This electrolytic cell generates a potential of about 900 millivolts when oxygen is present at only one side of the cell. Therefore, one side of this cell is exposed to the atmosphere, and the other is exposed to the exhaust gases. With this configuration a voltage will be generated for rich mixtures, which can be used to control the air/fuel ratio going into the engine. Figure 1.8 shows the oxygen sensor output as a function of the exhaust air/fuel ratio. Although this sensor does not indicate the actual air/fuel ratio of the exhaust, the indication of either a rich or lean mixture can be used in a fuel control algorithm. The next piece of hardware used to control the air/fuel ratio is the actuator to change the ratio that enters the cylinder. Historically this hardware took the form of the electronic carburetor. In the General Motors Phase II catalyst system [6] two types of carburetors are used. One carburetor (Holley) has a fuel metering rod that is controlled by vacuum modulated by a solenoid. In the other type of carburetor -27- Chapter 1: I.C. Engine and Control Background 00 — 800 = = sy 600 Ey #500 5 fa Rich Lean = 300 E & 200 100 9, 18.0 14.1 14.2 14.3 14.4 14.5 19.6 18.7 14.8 14.9 15.0 15.1 15.2 15.3 15.4 AIR/FUEL RATION Figure 1.8 Oxygen Sensor Output [6] (Rochester Products) the metering rod is moved directly by the solenoid. In both carburetors the solenoid is pulse-width modulated, so the air/fuel ratio of the mixture is linearly proportional to the duty cycle of the solenoid, which is modulated at 10 Hz.. The pressure drop in the venturi controls the overall fuel demand, with the solenoid and metering valve “trimming” this demand as commanded by the electronic control module with feedback from the oxygen sensor. These carburetors became complicated and expensive. ‘A much simpler and less expensive fuel metering system of fuel injection is used in most automobiles today. There are generally two types of fuel injection setups used today; throttle body injection (T.B.L.) and port fuel injection (P.F.I.). Air entering the engine passes through a throttie body, then into the intake manifold, which usually consists of a plenum to smooth pressure pulses within the manifold, and then runners —28- Chapter 1: £.C. Engine and Corstrol Background which distribute the air to each of the cylinders. The intake port is the part of the cylinder head that is upstream of the intake valve, and is connected to the intake manifold runner. In the throttle body fuel injection arrangement the fuel is injected into the air stream just upstream of the throttle plate, much like the standard carburetor. ‘This system has the advantage of requiring only one or two injectors, which minimizes the cost per vehicle. The disadvantage is that the fuel must travel from the throttle body to the cylinder. This causes a problem that is also present in carbureted engines; fuel wall wetting. A liquid fuel film can build up on the walls of the intake manifold, especially in a cold engine, which can seriously degrade the transient response of the engine due to excursions in the cylinder air/fuel ratio. The accelerator pump is an attempt to compensate for a lean spike when the throttle is opened quickly. The pump injects an additional arnount of fuel into the throttle body in order to compensate for the slow fuel response at the intake valve. With throttle body injection no new fuel hardware is needed for fuel air/fuel enrichment, the injector pulse width is simply lengthened. Port fuel injection is more expensive than throttle body injection, since there is an injector for each cylinder, but the wall wetting problem is diminished by injecting the fuel directly into the inlet port. There is a more uniform fuel distribution between the cylinders, and an increase in volumetric efficiency because little fuel is present in the intake manifold. In port fuel injection usually one of two timing schemes is used. The first is called simultaneous double fire, where groups of injectors (either all injectors or banks of injectors) are fired simultaneously every 360° of crankshaft revolution. ‘Therefore, there are two injections in the intake port per intake for each cylinder. This scheme is simple and injector timing (start of injection) is not extremely important. A -29- Chapter 1: LLC, Engine and Control Background second scheme is called sequential injection where the start of the injection is sequenced with the timing for its respective cylinder (i . usually a fixed number of degrees before top dead center (T.D.C.) of that cylinder). Therefore, no two injectors fire at the same time. While this scheme appears to give the best response, there are some problems when fueling changes are made at high fueling rates. Injectors are sized so that they remain on almost all of the time during wide-open throttle (W.O.T.) and high speed operation (i.e. maximum fueling requirement). This results in the smallest injector and also the best fuel control, especially at short pulse widths during idle conditions. This also causes an additional two crankshaft revolution delay in fuel response at high fueling levels. If the injector starts shortly before the intake valve opens, the intake valve will close before any change in pulse width is made, and this change will not be felt in the cylinder until the next intake. That is the trade-off between simultaneous double-fire and sequential port fuel-injection. The need to measure, or estimate, the air flow rate is an additional requirement of fuel injection that is not needed in the carburetor, since the venturi pressure drop of the carburetor serves this function. Two air flow estimation schemes are generally used. A “speed-density” estimation of the mass air flow rate out of the intake manifold using the intake manifold pressure and tempurature, the engine speed, and an estimate of the volumetric efficiency based on the intake manifold conditions can be made with the equation: thao = Ox pra. We Me Pie T =O (12) -30- Chapter 1: I.C. Engine and Control Background While this method does not require any exotic sensors, it is sensitive to the estimate of volumetric efficiency (n.), which is usually a steady-state measurement, and assumes a homogeneous pressure distribution in the manifold. Another common method of mass air flow estimation is with the use of a mass air flow sensor. The widely used Bosch K-Jetronic system [7] uses a mechanical air flow sensor, which is a moving restricting plate in the inlet air stream, in a cone shaped housing. Most modern systems now use a hot wire anemometer to estimate the air flow. The hot wire anemometer is relatively expensive, but it has the advantage of not depending upon a volumetric efficiency estimate, and the measurement anticipates any transient changes in the air flow out of the intake manifold, since the flow out is nearly a first order lagged version of the flow in. This anticipation is useful in obtaining better air/fuel ratio control, since there is a delay in fuel actuation. The third piece of hardware that is used for closed-loop air/fuel ratio control is the electronic control module (E.C.M.). In a modern vehicle, the E.C.M. may control many engine functions such as the injector actuation signals, spark timing, E.G.R., idle air control (LA.C.), fuel pump, electric cooling fan, ete.. With the electronic carburetor the E.C.M. controls the fuel modulating solenoid in the carburetor. In its simplest form the closed-loop air/fuel ratio controller contains a P-I (proportional-integral) control circuit that drives either a duty cycle generator for the electronic carburetor, or a timing circuit for the fuel injectors [6]. The proportional circuit provides a discrete change in the fuel command when the oxygen sensor goes from lean to rich, or vice versa. The integral correction is always driving the air/fuel ratio to stoichiometry. Around this basic feedback circuit are many additional functions, such as fuel enrichment during cold start, deceleration fuel cut-off, learning devices or adaptive controllers, etc.. Because -31- Chapter 1: 1.C. Engine and Control Background there is a significant delay from fuel command to exhaust air/fuel ratio sensing, the exhaust air/fuel ratio cycles around stoichiometry with approximately a 1 Hz frequency. ‘This cycling has been shown to widen the effective useful air/fuel ratio band of the catalytic converter shown in Figure 1.7, thus improving (decreasing) the overall engine emissions caused by air/fuel ratio errors. 1.4__Engine Torque Control ‘The control of the engine output torque, by modulation of the engine control (throt- tle, spark, and fuel) to perform a desired task, is the thrust of this thesis and ongoing work at the Vehicle Dynamics Laboratory at M.LT.. The desired task is to improve the shift quality of the 1+2 upshift with an automatic transmission. The shifting task is to smoothly transfer power from a high gear ratio to a lower gear ratio by means of clutches connected to each gear reduction set. This is currently accomplished by having an overrunning (one-way) clutch connected to the high gear ratio (first gear), and a simple clutch set connected to the lower gear ratio (second gear). As the second clutch set is brought on, the torque on the first clutch decreases to zero. This phase of the shift is known as the torque phase, since only a transfer of torque has occurred from the overrunning clutch to the second clutch. After the overrunning clutch torque goes to zero it starts to slip, and the slip speed (relative speed between clutch plates) across the second clutch decreases until the clutch is locked. This phase of the shift is called the speed or inertia phase, since the speeds of the two inertias (on each side of the second clutch) approach each other and are finally matched. -32- CLUTCH TORQUES (NM) 650 Chapter 1: I.C. Engine and Control Background 70 DEGREE THROTTLE 600 TT sso ‘00 450 400 350 E 300 E 250 F 200 F sof 100 F port ge T T T T T T T T u T ‘Torque Capacity, Toa q q Torque Phase (WV | j Speed Phase ee eee ee eee 66 68 7.0 7.2 7.4 7.6 7.8 8.0 82 84 .8.6 TIME (SECONDS) Figure 1.9 Clutch Torques During a 12 Shift [8] -33- QUTPUT SHAFT TOROUE (NM) 2800 2600 2400 2200 2000 1800 1600 1400 1200 1000 e800 600 6, Chapter 1: I.C. Engine and Control Background 70 DEGREE THROTTLE I T T T T T T T L L L 1 1 L L 1 1 1 4 66 68 7.0 7.2 7.4 7.6 7.8 &0 82 6.4 6.6 TIME (SEC@NOS) Figure 1.10 Shaft Torque During a 1 +2 Shift (8] -34- Chapter 1: ILC. Engine and Control Background A considerable amount of transmission modeling, done by Rundet [8], clearly indi- cates the parts of the shift and the resulting output torque as shown in Figures 1.9 and 1.10. The output shaft torque during the torque phase of the shift is characterized by 1a decrease in torque because the second gear ratio is much lower than the first. This torque drop is merely a function of the quotient of the two gear ratios, or: Tueo = Ta (3) This torque droop is commonly referred to as the “torque hole”. In the inertia (or speed) portion of the shift, the output torque rises because of the increased torque of slowing down the engine to obtain a synchronous speed across the second clutch. From these plots it is evident that in order to produce a smooth 1~2 shift, the engine torque must be increased during the torque phase to fill the torque hole, and must be decreased during the speed phase so that the slip speed across the second clutch can be brought smoothly to zero. As synchronous speed is obtained, the engine torque must be brought back up to continue the acceleration of the vehicle. These points are discussed in detail by Runde [8]. The engine control useful in modulating torque with a minimum amount of delay is the spark advance. Spark advance can be retarded to decrease engine torque, although an excessive retard can increase HC emissions due to incoraplete combustion of the fuel. Bosch has been successful in demonstrating as improvement in shift quality by modulating spark to reduce torque during the speed phase of the shift [9]. Although spark advance modulation has the advantage of being a fast control (i.e., very little de- lay from control actuation to torque modulation), it has two disadvantages. Typically, t former member of the Vehicle Dynamics Laboratory, M.LT. =35- Chapter 1: ILC. Engine and Control Background spark advance is kept slightly retarded from MBT (minimum best torque), which is the spark advance for maximum torque, so that changes in spark advance can only decrease the engine torque. The other disadvantage is that the magnitude of torque reduction by using spark is limited if one does not want to increase the #C emissions and increase engine roughness, due to misfires or partial burns in the combustion chamber. An al- ternative to changing torque with the spark advance is to modulate the throttle angle and injector fuel flow rate to obtain the desired engine torque. If these two controls are coordinated in such a way that the air/fuel ratio remains near stoichiometry, then the complete range of engine torque, at that engine speed, is available, providing spark advance is kept near MBT. While coordinated throttle/fuel control has the advantage of being able to produce a change in engine torque, this method also has two disadvan- tages. There is a relatively long delay and a lag between throttle actuation and torque response. Figure 1.11 is a block diagram of the elemental engine lags and delays in the Laplace domain. It illustrates the problem of using throttle and fuel instead of spark to change torque. The spark to torque delay At, varies with engine speed and can range from 20° to 190° of crankshaft rotation, depending upon when the spark advance is given with respect to the cylinder firing. By contrast, the intake to torque delay At, is about 300° of crankshaft rotation, and r, is approximately equal to 2, where the engine speed u. is given in radians per second. There is a considerable lag in air flow because of the intake manifold, and an additional delay due to the intake and compression strokes prior to spark ignition. These delays and lags limit the performence of a feedback control system. The second disadvantage of using throttle and fuel to control torque is that there are delays and lags in the fuel delivery that are difficult to quantify accurately -36- Chapter 1: I.C. Engine and Control Background Hatake Manifold Lag Intake—Torque — »| Delay - Throttle Angle 1 Torque rai eats Spark—-Torque Delay Spark Advance ‘Torque ents Figure 1.11 Simplified Block Diagram of Air and Fuel Transfer Functions for all operating conditions. Transient fuel delivery model errors can cause significant air/fuel ratio errors, and closed loop air/fuel ratio control has limitations due to cycle delays and an oxygen sensor lag. While the closed-loop fuel control algorithms work well for most steady-state conditions, the fuel delivery and oxygen sensor lags and delays make it difficult to maintain a stoichiometric air/fuel ratio throughout the rapid transients during a shift. To accomplish the dual task of rapid engine torque modulation while maintaining a near stcichiometric air/fuel ratio, a combination of throttle/fuel and spark control can be used sv as to take advantage of each coatrol’s strengths, and to avoid their weaknesses, This type of control is commonly known as a “drive-by-wire”t control t Flyby of control surfaces, is an analogous term coined by the aircraft industry for microprocessor management -37- Chapter 1: 1.C. Engine and Control Background arrangement, because there is no mechanical connection between the driver and any control. The name originates from microprocessor control of flight control surfaces on aircraft. In most current vehicles the driver's accelerator pedal is mechanically linked to the throttle plate on the carburetor or throttle body. By moving the throttle plate, the driver can control the air flow into the engine, and can control the torque that the engine produces. This type of system has worked well, as evidenced by the number of vehicles produced each year. However, we have seen that during a shift, appropriately timed rapid changes in engine torque can help to smooth that shift, as measured by the transmission output shaft torque or vehicle acceleration. Some torque reduction is possible using spark control, as in the Bosch system, but rapid large changes (increases as well as decreases) can only occur if the throttle is controlied by a microprocessor during the shift. With this system the driver would give a command to the microprocessor, via the accelerator pedal, and the microprocessor would send commands to the throttle, spark advance, fuel injectors, and transmission clutches, hence the title “drive-by-wire”. Another possibility with this arrangement is to achieve a shift that is not detectable by the driver, other than by listening to engine or transmission sounds. In the cur- rent configuration, if the engine produces the same torque before and after the shift, then the output shaft torque and vehicle acceleration drops dramatically after the shift because of the shift to a lower gear ratio. With the driver's accelerator command usu- ally calling for power, in the “drive-by-wire” configuration the throttle angle can be changed to give the driver the power for which he is calling, within the constraints of the maximum engine torque at the respective engine speeds. So, for part-throttle shifts, the possibility exists for a nondetectable shift. Whether this type of shift is possible -38- Chapter 1: I.C. Engine and Control Background will depend on the physical constraints of the system, such as the torque converter lag, the band width of clutch pressure control valves, etc.. Automobile manufacturers have been reluctant to produce a “drive-by-wire” system with an electronic throttle because of safety concerns, although many manufacturers are conducting research in this area. B.M.W. is marketing the first production “drive-by-wire” vehicle with electronic throt- tle on the 750iL, 12 cylinder engine, which is controlled by the Bosch Motronic control system. 1.5__Engine Modeling for Control Development. Dynamic engine modeling for control algorithm development has been underway for approximately the past dozen years. The principal reasons for the development of almost all of these models fall into two main catagories. The first catagory is air/fuel ratio control. With strict emission standards set by the Environmental Protection Agency, the catalytic converter was developed to meet these standards. The three way converter is effective in decreasing the three main pollutants [4,6] (see Figure 1.7). However, the region in which all three pollutants (NOx,CO, HC) ate effectively decreased is a narrow air/fuel ratio band around the stoichiometric ratio, If the air/fuel ratio is allowed to go rich, the efficiency of the converter in decreasing hydrocarbon and carbon monoxide emissions is severely diminished. For lean mixtures, efficiency of the converter in decreasing emissions of oxides of nitrogen reduces dramatically. With these constraints on air/fuel, models of the fuel delivery dynamics were helpful in developing transient fuel control strategies. In these fuel control strategies, the fuel delivery model is inverted and used to minimize air/fuel deviations from stoichiometry. The primary controlled parameter is the command to the fuel injectors. Examples of this type of model are Blumberg, Wu, Auiler (10], Aquino [11], Powell, Wu, Aquino (12], Stivender -39- Chapter 1: I.C. Engine and Control Background (13], and Hires, Overington [14]. Stivender’s approach is rather unique, in which he proposes that the driver should control fuel delivery and the controller should try to add the appropriate amount of air for stoichiometry. With this method, errors tend to occur in a desired direction (i.e. rich spikes from quick torque request increases, etc.). A typical physical application of this type of modeling is a throttle body injected engine with lags and delays of the fuel in the intake manifold. The controller has a delayed and lagged signal, from the oxygen sensor, which indicates if the mixture is lean or rich. The fuel models help to develop controllers which control the fuel open-loop, and trim this control with oxygen sensor feedback. The second category of engine models is used for idle engine speed control. Essen- tially, these models start out as detailed full nonlinear models that are linearized about a desired idle equilibrium condition. In these models the primary controlled parameter is the Idle Air Control Valve (i.e. air flow control). Many of these models are reported in their nonlinear formulation. Examples of this type of model include Dobner [15,16], Coates, Fruechte [17] and Powell [18]. As with any stereotyping, the grouping of engine models into only two groups may be oversimplification, and there is much overlap. Also, there are some models that have no clearly defined goals. A comprehensive engine model summary is given by Powell {19]. In recent years there has been an increased interest in powertrain control, or coordinated engine/transmission control. The control of the engine tends to take the form of a “< -by-wire” system, where the driver’s input (accelerator pedal calling for power) is sent to a microprocessor. The microprocessor then controls the electronic throttle, spark advance, and fuel injectors on the engine, and clutch pressures on the transmission. This type of control is reported by Schwab [9], although -40- Chapter 1: I.C. Engine and Control Background only spark advance control is used on the engine. Simple engine models are often used in powertrain models, such as Tsangarides, Tobler, and Hermann [20]. Most of this powertrain control work is proprietary, and little public information is available. An interesting paper on real time engine modeling and control by Powell, Lawson, and Hogh {21] details work currently being performed at Ford’s Scientific Research Laboratories. ‘The engine models predict cycle by cycle cylinder events, and the engine model can run in real time along with an engine in a test cell. The model requires considerable computaional power, and currently runs on a high-speed Applied Dynamics model AD-10, although it can be a powerful control develepment tool. Powertrain control is the reason for the development of the model outlined in this thesis. The model is to be used for the development of nonlinear control algorithms to be usable throughout the operating range of the engine. A careful analysis of the inher- ent delays of various fueling schemes are presented, and identification of fuel delivery parameters is given for a sequential port fuel-injected engine. Most port fuel injection engine models assume negligible fuel lag or delay, which is not the case. There are inherent delays in fuel as a result of the fueling algorithm and discrete fuel injection, and a torque lag has been identified by means of a nonlinear Leverberg-Marquardt identification technique. A physically based intake manifold model with E.G.R. is pre- sented, and is effective in predicting manifold pressure and the mean flow rate of air out of the intake manifold. -41- Chapter 1: I.C. Engine and Control Background Thesis ion This thesis will detail the development of computer engine models and engine con- trol. Chapter 1 gives an introduction and short history of the subject of engine control. Chapter 2 will describe the engine model in detail with subsections for each part of the engine. This model was developed with a specific engine in mind, but it can be altered to model most automotive S.I, internal combustion engines. The success of modeling other engines depends on the amount of appropriate validation data that is available or that can be gathered. Chapter 3 outlines how the engine model can be altered to represent other engines. That chapter also gives recommendations for validation tests to be performed. Chapter 4 illustrates an actual case study of validating this engine model for a specific engine, That chapter will show how the tests were conducted, what equipment was used, what was discovered from these tests, and what difficulties were encountered in validating the engine model. It will also show how the data was used to extract or estimate the parameters of interest. In Chapter 5 there is a background discussion of the nonlinear control technique of Variable Structures or Sliding Modes [22,23]. Following that discussion ther are examples of the use of this technique to design engine control algorithms. Also included are simulation results with a discus- sion of the issues involved, and the problems encountered, using this technique. The transmission clutch control algorithms that are used in conjunction with this model were designed by Jahng-Hyon Park, and will be given in detail in his Ph.D. thesis. Previous sliding mode engine or transmission control work was done by Runde [8], Allen 24), Harrigan [25], and Cho [26] using models developed by the author. Runde’s work was primarily transmission modeling and control to establish the shape of the desired engine torque profile. Allen and Harrigan’s theses outline a sliding mode fuel ~42- Chapter 1: ILC, Engine and Control Background control algorithm. Cho's work is primarily clutch control algorithm design with some open loop engine control. The final chapter will give thesis conclusions and recommen- dations for future research in this area. Appendix 1 contains details of the throttle body equations in Chapter 2,and a simplified powertrain model for the 1 ~ 2 shift is presented in Appendix 2. Appendix 3 contains a review of the Sliding Mode nonlinear control methodology. -~43- 2 Model Description 2.1__Introduction In this chapter the details of a lumped parameter, physically based engine model are given, which can be adapted to represent many spark-ignition antomotive engines. The engine model is sudivided into subsystems, which will be modeled and validated separately. These subsystems are the throttle body, idle air control, intake manifold, fuel delivery, torque production, and rotational dynamics. The torque converter, trans- mission, and driveline models are given for completeness, and their interaction with the engine will become more apparent in Chapter 5. 2.2 Throttle Body Model ‘The function of the throttle body is to restrict, or throttle, the air flow into the engine in order to control the torque output. The throttle body consists primarily of a cylindrical bore with a throttle rod traversing the bore. A throttle plate is mounted on the rod to restrict the flow through the bore. The throttle plate is typically elliptical, and the plate can completely close off the bore. The angle between the closed throttle and the plane of the bore is greater than zero, to prevent the plate from binding in the bore at closed throttle. The geometry of the throttle body opening changes with varying throttle angle. To accurately predict the air flow through the throttle body, a three-dimensional model would be necessary because of the complex geometry. The —44- Chapter 2: Model Description computational requirements of such a model would be prohibitive. A much simpler model with an empirically determined discharge coefficient has been shown to give reasonable results. The throttle body model is based on the one-dimensional isentropic compressible flow equation for flow across an orifice. This equation (for nonchoked flow) is given as: GQ yo In this equation the mass air flow through the throttle body is a function of the up- P, Fines = Ca- A(9) stream stagnation pressure and temperature. the pressure ratio of throat (minimum area) pressure to upstream stagnation pressure, the ratio of specific heats, ideal gas constant for air, cross-sectional area of the throttle body, and a discharge coefficient. When the flow is choked, or: (2.2) the mass flow rate is no longer an explicit function of the pressure ratio, or: tna 00 a) ae 8h (GE) 3) In the throttle body there is negligible pressure recovery downstream of the minimal cross-sectional area, so the throat pressure can be approximated by the intake manifold plenum pressure. The throttle body is not truly a one-dimensional orifice, and infor- mation reflecting this fact is contained in the discharge coefficient. This coefficient has been shown to be a function of the throttle angle (i.e. throat geometry) and the pres- sure ratio across the throttle body. Furthermore, steady-state flow tests have shown that these two influences are independent within the operating range of the engine. -45— Chapter 2: Model Description The cross-sectional area of the throttle body is a function of the throat diameter, the diameter of the rod on which the throttle plate is mounted, the angle between the plane of the bore and the closed throttle, and the angle of the throttle from the closed position. At large throttle angles the throttle rod diameter has a significant effect on the throttle cross-sectional area. Details of the throttle geometry are given in Appendix 1, The final cross-sectional area of the throttle is given by: aD a\7)} d-D a 2 6-G)P PC QP, a\7yt Fedh-G)T} When further throttle opening does not increase the cross-sectional area (see appendix A) 1), or: o> cor (25) then, the area is constant at: wo Bae(h- (TEE GT e Harrington and Bolt [27] have noted that an error exists between the ideal cross- sectional area at small throttle openings and actual areas, due to machining tolerances. A correction for this can be made by substituting cos(62) for cos(@.) where: c04(82) = cos(0.91 * 6. ~ 2.59), (2) for 4, measured in degrees. Once the upstream stagnation pressure and temperature are established, the flow through the throttle body is a function of two independent parameters; the throttle -46- Chapter 2: Model Description computational requirements of such a model would be prohibitive. A much: simpler model with an empirically determined discharge coefficient has been shown to give reasonable results. The throttle body model is based on the one-dimensional isentropic compressible flow equation for flow across an orifice. This equation (for nonchoked flow) is given as: B)t [2k , ' 2) thai = Ca (0) sen (#) {E b-(R) 1} . (2a) In this equation the mass air flow through the throttle body is a function of the up- stream stagnation pressure and temperature, the pressure ratio of throat (minimum area) pressure to upstream stagnation pressure, the ratio of specific heats, ideal gas constant for air, cross-Sectional area of the throttle body, and a discharge coefficient. When the flow is choked, or: 2 (4) (22) the mass flow rate is no longer an explicit function of the pressure ratio, or: a Pt. (2) thas = Ca A(#) JER k (¢ i) : (2.3) In the throttle body there is negligible pressure recovery downstream of the minimal cross-sectional area, so the throat pressure can be approximated by the intake manifold plenum pressure. The throttle body is not truly a one-dimensional orifice, and infor- mation reflecting this fact is contained in the discharge coefficient. This coefficient has been shown to be a function of the throttle angle (i.e. throat geometry) and the pres- sure ratio across the throttle body. Furthermore, steady-state flow tests have shown that these two influences are independent within the operating range of the engine. -45- Chapter 2: Model Description ‘The cross-sectional area of the throttle body is a function of the throat diameter, the diameter of the rod on which the throttle plate is mounted, the angle between the plane cf the bore and the closed throttle, and the angle of the throttle from the closed position. At large throttle angles the throttle rod diameter has a significant effect on the throttle cross-sectional area. Details of the throttle geometry are given in Appendix 1. The final cross-sectional area of the throttle is given by: an=-*2-h-(5) 5? b- (5 aaa) | Bae hCG) ) Sst oe (l-G a)T D 2” coa(G.) D cos(6. + 8) When further throttle opening does not increase the cross-sectional area (see appendix 1), or: oz coo [$ cou()] (25) then, the area is constant at: ay af 4 4D a {h- G)T}-?-6-G)Y- a Harrington and Bolt [27] have noted that an error exists between the ideal cross- sectional area at small throttle openings and actual areas, due to machining tolerances. A correction for this can be made by substituting os(6:) for cos(6.) where: = c0s(0.91 + 6, ~ 2.59), (2.7) for ¢, measured in degrees. Once the upstream stagnation pressure and temperature are established, the flow through the throttle body is a function of two independent parameters; the throttle -46- Chapter 2: Model Description angle, and the pressure ratio across the throttle body. Stagnation pressure and temper- ature changes will significantly alter the flow through the throttle body. This difference can be seen if a vehicle is driven from high altitude warm climates (Denver, Colorado in the summer) to low altitude cold climates (northern Canada). These influences can be separated into normalized functions and expressed as: thai = MA-TC- PRI, (2.8) where MA is the maximum mass air flow through the throttle body (a function of and geometry), TC is the throttle characteristic (a function of the throttle angle and geometry), and PRI is the pressure ratio influence (a function of the intake manifold plenum and upstream stagnation pressures). Each of these functions are given below. (29) (2.10) > (xh) < (eh). Separating the function in this way speeds the computation of mass air flow. With (2a) fer b-(é)")}. & gE equation (2.8) the second and third terms are put into a look-up table, and MA is a slowly changing parameter which can have slow sampling rate. In the above equations: Ca = Cay “Car, where Ca is only a function of the throttle angle, and Cy: is only a function of the pressure ratio across the throttle. The throttle body air flow is separated into these three parameters to speed computation, A typical electronic control module for engine control has a large memory, but cannot perform lengthy calculations. The normalized functions PRI and TC can be put into look-up tables, which can be accessed quickly -4T— Chapter 2: Model Description and easily. The maximum air flow MA is a function of -S- which changes slowly, and sampling rates for these parameters can be very slow. 2.3 Idle Air Control Model On most modern throttle bodies used on fuel injected engines, there is a small air by-pass of the throttle plate. In this passage is a valve which controls the cross- sectional area of this passage, and limits the air flow through the by-pass. This by-pass and valve is known as the idle air control (I.A.C.). Its function is to control the amount of air flow into the intake manifold during idle operation. It can als. be used to smooth any large air flow change due to a rapid change in the throttle angle. The throttle is typically closed during idle. With the I.A.C., a low engine speed can be maintained at idle. Lowering the engine idle speed can significantly lower the engine fuel consumption since there are high pumping losses at idle, due to the low intake manifold pressure. The idle air control is modeled with the one-dimensional isentropic compressible flow equation, in a manner similar to the throttle body. In place of the throttle characteristic there is an idle air control valve characteristic which is obtained from a choked flow calibration test. The idle air control valve is an axially symmetric valve, so the flow is more closely modeled by the one-dimensional ideal choked flow equation. The discharge coefficient Cuz for the idle air control valve is assumed to be unity, and the one-dimensional choked flow equation is used. The total flow discharge coefficient is only a function of the idle air control valve count. The total flow through the idle air control valve is given as: nga = 4 Meat(count) -y/Bfat pha Yea (count) Test - phe, ec (<:)" —48- Chapter 2: Model Description ‘There are three parts of the I.A.C. air flow equation. The first part is a choked flow calibration of the I.A.C. valve at some known stagnation condition. The second part is an adjustment for a change in the operating stagnation conditions, and the third is an adjustment for nonchoked flow conditions. 2.4__Intake Manifold Model The intake manifold defines the air passage between the throttle body and the cylinder head. This passage typically consists of two sections; the plenum, and the runners. The plenum is immediately downstream of the throttle body, and it is a cavity to which all of the runners are attached. The function of the plenum is to provide a reasonably uniform pressure, throttled air supply for each of the runners, by smoothing out the pressure pulses from the runners, or: ig Potenum * 5 Le Prsaners (2.13) in a lumped parameter model. The runners diverge from the plenum, and supply air to each of the cylinder intake ports in the cylinder head. The runners are often tuned, in conjunction with the plenum, to act as Helmholtz resonators. By tuning the plenum and runners so that the natural frequency of this system coincides with the intake valve opening frequency, additional air can be pushed into the cylinders, thereby increasing the volumetric efficiency near that engine speed, The natural frequency of the Helmholtz resonator is given [28] as: nd ag where ¢ is the sonic velocity, s is the cross-sectional area of the runner, V is the plenum volume, and | is the runner length. ~49- Chapter 2: Model Description The intake manifold model described below is a lumped parameter model, with uniform pressure and temperature. This model does not predict the pressure pulse behavior of the runners. Rather, an empirically determined volumetric efficiency func- tion, based on the intake manifold conditions, is used to describe the flow out of the manifold, if the intake manifold pressure is not known. The intake manifold model also considers the addition of E.G.R., and how the air flow out of the manifold is affected by this diluent. The following assumptions are made in the intake manifold model: (i) the contents of the manifold obey the Ideal Gas Law and Dalton’s Law of Nonreacting Mixtures, (ii) uniformity of temperature and pressure, and (iii) complete mixing of air and E.G.R.. Also, whatever residual gases blown into the intake manifold from the cylinder when the intake valve opens are completely drawn back into the cylinder during the intake stroke. The fuel is assumed to have a negligible effect on the dynamics of the air and E.G.R. in the intake manifold. Combining the Ideal Gas Law and Dalton’s Law with the conservation of mass equation for air and E.G.R. yields: RT (249) 5 (P= Pa) VM, M. tihegro = ages + LP Pa) V Moar op _ VMeor ‘aro rari E77 T- En (P-P,). (2.16) -50- Chapter 2: Model Description Volumetric efficiency, based on the intake manifold density, can be used to estimate the flow of constituents out of the intake manifold.} This estimate is given by: _ KV wet fgg = gente [(a4e-Meor) «Pa t+ Mege (2.17) If the previous three equations are combined, the result is a first order differential equation of the partial pressure of air (P,) in the intake manifold, or: (22) es) Megr RT... M, Pas -K ne (Met) oe BE Cat hs) «Mee ( #2), (2.18) Notice that this equation is a function of the ratio of the molecular weights of air and E.G.R.. If there is a significant difference in the molecular weights, then this equation can be used along with (2.15) to estimate the mass air flow out of the intake manifold. To estimate the molecular weight ratio, consider the stoichiometric complete combustion equation¢ for indolene. (Cr Hys.16 + 10.29(On + 3.773Nz) = 7CO2 + 6.5810 + 38.82NN>. (2.19) Using this equation, the molecular weights of air and E.G.R. are given as: .838 g/mole, (2.20) = 28,884 g/mole. (2.21) Throughout this thesis, the volumetric efficiency will be based on intake manifold conditions (ie. Pm and Tra) $ Air also contains 0.98% argon and 0.039% carbon dioxide which are ignored for this analysis. If considered, the molecular weight of air would be 28,962 g/mole —51- Chapter 2: Model Description Because the molecular weights are essentially equal, (2.18) reduces to the volumetric efficiency relation to be given in (2.25), and is not useful in estimating P, or consequently ‘gg when E.G.R is present. Another approach, to estimating the air flow out of the intake manifold, would be to use the complete mixing assumption: rege _ Mepe _ ers a He (2.22) If this relationship is combined with the two conservation of mass equations, (2.15) and (2.16), the result is a differential equation in the partial pressure of the air in the intake manifold, or: a PRT |. | RT = [5--% 57 (a tnd] thas (2.23) This relation can be combined with (2.15) to estimate the mass flow rate of air out of the intake manifold if the time histories of T, P, rnai, and rneg; are known. Another interesting result is obtained by adding the two conservation of mass equa- tions for air and E.G.R. (2.15),(2.16), and using the volumetric efficiency relation (2.17) to obtain a differential equation in manifold pressure. (2.24) +n (2.28) steady-state Irancient ‘The first term in this equation represents the steady-state volumetric efficiency, based on intake manifold density, typically calculated from steady-state test cell data by —52- Chapter 2: Model Description measuring P,T,w., and rng: + titers. The second term is the dynamic correction term for volumetric efficiency. The steady-state volumetric efficiency relation can be combined with (2.23) to yield: d RT fre (S-nncnl,) eB. With this equation the partial pressure of air can be estimated by knowing the time (2.26) histories of P,T,w., and tna, and the steady-state volumetric efficiency. Ideally tnegr: does not have to be known to estimate P, or rngo. ‘The above relations are helpful in sorting out the physics of a simplified intake manifold. However, from a practical controls perspective there are some problems associated with estimating rm. using these relations. We cannot directly measure the partial pressure of the air in the intake manifold, which is now a state. To use these relations an observert [29] must be used with speed or possibly torque measurements as inputs, as well as friction and torque prediction models within the observer, to estimate the partial pressure pressure state P,. This approach is not promising because of the difficulty of measuring or estimating the engine brake torque from speed measurements, and the multiplicative nature of the models errors involved in the embedded model. One way to avoid these problems is to estimate the mass E.G.R. flow rate out of the intake manifold by using an appropriate time varying first order lag filter on the E.G.R.. With this approach, small errors in the estimation of the E.G.R. flow out of the intake manifold are accepted in order to alleviate the need for an observer and the additional states that go along with them. To see how this alleviates the need for an observer, consider the volumetric efficiency relationship: K-Vewe-neMeP wo Ve te MOP inser 27 RT rs a) f An observer is a model based state estimator. -53— Chapter 2: Model Description which is equation (2.17) where the molecular weights of air and E.G.R. are equal. If the volumetric efficiency of equation (2.25) is substituted into this equation and simplified, the result is an estimation of the air flow out of the intake manifold: thane eM (E 5): (2.28) tao = Fins + aa \t7P Another way of obtaining this relationship is to substitute the last two terms of the conservation of mass equation of air (2.15) into similar terms in the conservation of mass equation for E.G.R. (2.16). Now the flow of E.G.R. out of the intake manifold can be modeled as a first order time varying filtered version of the flow of E.G.R. into the intake manifold, or: Sree t Meare = = (Fears — tegro)- (2.29) Since the time constant of the E.G.R. is a mixing or flow related phenomena, the time constant should be most strongly a function of engine speed and volumetric efficiency. By comparing the response of this filter model with the original partial pressure model, a close approximation can be obtained by choosing the time constant as: 4 = 0.095 ny -we. (2.30) Further simplification can be made by looking at the relative numerical magnitude of each of the terms of equation (2.28). Figure 2.1 shows these terms during a throttle step with no E.G.R. present. Note that, for this 15° to 50° throttle step, the tem- perature derivative term is negligible when compared with the other terms. From a practical standpoint this is a fortunate result, because the temperature derivative is a difficult quantity to obtain. To calculate the temperature derivative during transients, a thermocouple with a small time constant would be required. These thermocouples —54- Chapter 2: Model Description rs c e _ fs & 30 ‘o NO aged Of we ee ee ee oo 2 3 4S 8 7 6 8 LO LI 2 43 TIME (SEC Figure 2.1 Intake Manifold Response to Throttle Step are rather delicate and are too costly to be considered for a production sensor. The final air flow estimate is given as: Fingo = tas + tegrt — Fegro = (2.31) The E.G.R. valve which is modeled consists of three calibrated orifices that are opened and closed by means of solenoids. With various combinations of activated and deactivated solenoids, eight different flow rates are possible for any one set of intake and exhaust manifold conditions. Based on a flow calibration supplied by the Buick- Oldsmobile-Cadillac Division of General Motors [30], and engine operating conditions, the flow of E.G.R. gases into the intake manifold can be estimated. ‘The theoretical choked flow characteristic is assumed, which is a good approximation. The equation —55— Chapter 2: Model Description for the E.G.R. flow is similar to the idle air control model, except that the exhaust gas conditions at the valve are now the stagnation conditions. The flow is given by: t ‘aye sen -Y Rt, of) I fe> (ze) (votoes)-«/ Te hc, es (s) Because the exhaust gas stagnation conditions can vary over a large range, the stag- nation condition adjustment (Vee witz) can be significant. This will be discussed = a (2.32) further in Chapter 4. 2, fodel The physical processes of fuel delivery in the fuel injected spark ignition engine are complex.t This fuel delivery model is an attempt to capture the dynamic effects that are important to the control engineer from the perspective of a mean torque predictive engine model. The approach used by the author is to examine the physical equipment involved, understand the physics of the processes, and determine which physical processes would affect mean engine output torque deviations. Figure 2.2 is an illustration of the typical physical arrangement of the fuel delivery equipment on the engine. Fuel (gasoline) is supplied under pressure to a fuel rail which is connected to each of the fuel injectors. More fuel is supplied to the fuel rail than is used by the injectors. The excess fuel is routed back to the fuel tank. After this excess fuel passes by the injectors, and before it leaves the fuel rail, it is sent through a fuel pressure regulator. The control side of the pressure regulator is connected to the intake “"} This discussion will be limited to the port fuel-injected engine. Other fuel delivery systems and algorithms will be discussed in Chapter 3. —56- Chapter 2: Model Description IMuECTOR ¢ TEST POINT FUEL PRESSURE PRESSURE GAGE TEST POINT REGULATOR FUEL INLET ‘TEST POINT it i Ae f PRESSURE, REGULATOR INJECTORS INJECTORS. INTAKE MANIFOLD 3.0L LN7 3.8L UNO asLie Figure 2.2 Fuel Injection Hardware [31] manifold plenum. This regulator maintains a near constant pressure differential across the injectors. Large deviations in this pressure differential would vary the fuel flow rate through the open injector. By maintaining a near constant pressure differential, the fuel flow through the injector is a function only of the injector pulse width, albeit a nonlinear function. A typical fuel injector is shown in Figure 2.3. The excited solenoid draws the needle valve towards the solenoid, and when the solenoid is deactivated, the return spring closes the pintle and the fuel flow stops. In the port fuel injected engine there is one fuel injector for each cylinder. ‘The physical acrangement of the inlet port is shown in Figure 2.4. The injectors are mounted on the intake manifold, with the fuel spray aimed at the intake valve, which is in the cylinder head. The fuel spray from the injectors is not completely atomized. The high temperature of the walls of the inlet port, and the closed intake valve,helps a Chapter 2: Model Description oistRIBUTOR oe URE UPPER O-RING FILTER ELECTRICAL CONNECTION SAFETY cue MAGNETIC WINDING SOLENOID’ NEEDLE VALVE LOWER O-RING Nozze Figure 2.3 Fuel Injector (31] SEAL, 4770 AF RATIO Figure 2.4 Inlet Port (31] -58- Chapter 2: Model Description to vaporize much of the liquid fuel, which is normally sprayed onto them. Because the injector is sized to be on almost all of the time at the maximum fuel delivery rates, the fuel is more often injected into the inlet port when the intake valve is closed. In this situation the hot inlet port and valve have time to vaporize much of the injected fuel. Figure 2.5 shows some of the complex fuel phase change dynamics which occur in the inlet port, To capture all of these dynamics requires a time, temperature, and geometry dependent fuel deposition and evaporation model. While this is not consistent with the simplicity and real-time goals identified at the beginning of the thesis, an adequate fuel delivery model is possible for a warmed-up engine, with respect to the mean torque predictive criteria. Another important consideration in the fuel delivery model is the the algorithm of the injection (i.e, sequential injection versus simultaneous double-fire injection). This model represents the sequential fuel injection algorithm used on many modern fuel-injected automobiles. Changes in this model to represent simultaneous double fire injection are detailed in Chapter 3. Figure 2.6 illustrates the fuel delivery model. This is 2 model of the fuel delivery from when the demand for fuel is calculated (rny,) to when the intake valve closes (rny.). Since this is essentially a continuous model, many of the delays are estimates of the mean delays. First, consider the case of no liquid fuel film on the port walls. The first delay, after the desired fueling has been calculated by the E.C.M., is from the output calculation to when the following injector is activated. For the sequentially fired six cylinder engine this can be anywhere from 0° to 120° of crankshaft rotation. The average time delay, assuming % 0 during this time period, would then be given as: seconds. (2.33) —59- Chapter 2: Model Description - z Fuel Vapor Fuel Droplets Liquid Fuel Film Time Figure 2.5 Inlet Port Fuel Dynamics 7} +r mH ‘ ~{beley}4 Figure 2.6 Fuel Delivery Model -60- Chapter 2: Model Description 4-5 ngs Delay: ye A= 5p 4 Mtiasarve a Figure 2.7 Simplified Fuel Delivery Model In the sequentially fired fuel system, the start of injection for each cylinder is timed with respect to top dead center of that cylinder. So, added to this delay is the time from the start of injection to when the intake valve closes, or: Abs seconds (2.34) assuming °s* +0 during injection. If the injector pulse has ended by the time that the intake valve has closed, then all of the injected fuel will be in the cylinder (assuming no liquid film and negligible transport delay of the gaseous fuel). If the injector pulse is still occurring when the intake valve closes, then the remaining fuel will be delayed by two crankshaft revolutions or $ seconds. A simplified model of this process is shown in Figure 2.7. The parameter 7 is given by: 1 if £PW < Linjaive { Buse, if CPW > Linjaivo? (39) -61- Chapter 2: Model Description where snyivo is the net fuel flow rate to the engine if the completion of the. injector pulse coincided with the intake valve closing, or: xf mvs inal = hype “Ata, (2.30) Sel where thy is the fuel flow rate to the engine through and open injector. The actual fuel injection process is further complicated by a liquid fuel film which can build up on the inlet port walls, especially when injection occurs with a closed intake valve, and slow the fuel delivery response. This liquid fuel lag is noticeable on throttle body injected engines, but is also reported to occur on port fuel injected engines (13]. Fuel having the transport characteristics of gaseous fuel continues on a similar path as shown in Figure 2.6, and the liquid fuel goes through delays that are multiples of two crankshaft revolutions, or At = £2. A representation of this process was shown in Figure 2.5. Epsilon () is called the fuel split parameter, and represents the fraction of fuel having the transport properties of gaseous fuel, and the f’s are defined as: (2.37) a { Afi yexp(-2)av, i=1,2,.. 4. J2,eap(-£)a¥, i=5 This represents a discretized first order lag of the liquid fuel film. If the engine speed is assumed to change very little when compared to the fuel dynamics, then: watt, (2.38) and + represents the time constant of the liquid fuel film lag. The above model is a: attempt to describe the discretized nature of the fuel delivery when it remains in the port after the intake valve closes. The value (1 -<) will turn out to be less than the fraction of fuel that is deposited on the walls, to compensate for the evaporation of -62- Chapter 2: Model Description some of this fuel. In practice this fuel split parameter is a complex parameter which depends on the port temperature, engine speed, mass flow rates, etc... For this modeling exercise, « and r are empirically determined parameters extracted from transient engine tests. The series of delays is truncated at i+ 5-r, which represents more than 99% of the lagged fuel. The last fraction includes all remaining fuel so that no fuel is “lost”. If the discrete fuel delivery is not apparent in the torque response data, then the slow fuel portion of the fuel model may be furthe: simplified as shown in Figure 2.8. Equations for this process are given as: tingo = tiga + tings + tinge (2.39) viyr2= lingele-any © O- Deas (2.40) tngt9 = Fpl an) 7 (241) ragel(s-any) (1-9 ~ try aa % {Ypeoow, Sikerwae “*"? re) ‘The subscripts (t— At) indicate that the parameter has been delayed by time At. The more complete discretized model is available to understand any anomalies which may occur in the data used for validation. rqui tion Model ‘The combustion and torque production model is a variation on a structure by Dob- ner [15]. This is a steady-state model and does not contain any dynamic elements, except for the process delays associated with the four-stroke combustion process. This approach is consistent with the real time control goals, and is justified since the com- bustion dynamics are much faster than the air, fuel, or W.G.R. transport dynamics. A -63— Chapter 2: Model Description {=} Figure 2.8 Simplified Fuel Delivery Model block diagram of this process is shown in Figure 2.9. The maximum possible indicated torque, for a stoichiometric mixture, is assumed to be a function of the mass of air per cylinder. The torque function used to estimate this maximum torque comes from com- bustion simulations by Chang [32], and will be discussed in detail in Chapter 4. The torque function is a constant times the gross indicated stoichiometric fuel conversion efficiency which, for a given engine, is a function of the intake manifold pressure and the engine speed. TF= Ot Faas Mlnste = 1439 nylare (244) 100 - ny] soi = —157.5 + 0.926 - nor + 0.854 - nacap + 0.647 nw +0647 - mp1 + 1.087 - ny, + 1.016 -ng (2.45) ny = 44.9 15.1-.N-0088 ) 64 Chapter 2: Model Description non = 0.729 nig 0.226 re (2.47) nia = [1 = r5-Y)] 100 (2.48) ng = 87.896 (2.49) nap = 42.1 - 4.36 (2.50) 0.020 na ju = ~108.6-+ 145.7- (2) (251) fy, = 45.7 — 13.8 Vz 008 (2.52) Figure 2.9 Torque Production Model For the engine of interest: P, \~0288 a 100 -nylyepe = 1.09 [us-szs. (#5) 971 wz (2.53) Details of the fuel conversion efficiency model can be found in Chang's thesis [32]. -65— Chapter 2: Model Description ‘The mass air flow out of the intake manifold is a sum of the flows into each.cylinder or: es fae = Jo toa (t)- (2.54) & So, the mass of air in cylinder & is given by: swe mae [sesh (255) or in the steady-state case: (2.58) This steady-state calculation is used to estimate the mass of air per cylinder at all times. This is necessary because the intake manifold model does not estimate the instantaneous air flow to each of the individual cylinders, but rather, an averaged flow as if the engine had many cylinders. The actual air/fuel ratio in cylinder & is given by: ' a air) _ S202 tad ‘asn Fuel ley Sel" tyoydt However, here again the steady-state estimation is used, or: (258) ‘The maximum possible torque (i.e. with an equivalence ratio of 1.1, and spark advance at MBT) is reduced by two normalized functions. The first one is an air/fuel ratio influence function. This function represents the decreased torque when there is not enough fuel to utilize all of the air in the cylinder, or if there is insufficient air to burn all of the fuel. In the extreme, it represents the lean burn limits and decrease in torque at rich mixtures. This is an empirically determined function derived from fuel perturbation data. Figure 2.10 shows a typical air/fuel ratio influence function -66- Chapter 2: Model Description for the engine modeled in Chapter 4. The second normalized function is the spark influence function. This function decreases the indicated torque as a function of how far the spark advance is from the MBT spark timing. In this case the spark advance is always retarded from MBT so as to avoid the significant knock problems associated with highly advanced spark. An MBT map is derived from test data, and is considered to be a function of engine speed, charge density (or mass of air per cylinder), and percent E.G.R.. For the engine modeled in Chapter 4, the spark influence function is given as: 8x 10-*.($A~ MBT)?, (259) where: MBI = 88: MAC? ~ 99.59 -MAC+4761+0.08- N+ 58. — Tar. (2.60) AAIR/FUEL, INFLUENCE (AFI) AND DERIVATIVE o 2 4 6 6 10 12 14 16 18 2 2 7% 26 2 90 AUR/FUEL RATIO Figure 2.10 Air/fuel Influence Function (AFI) -67— Chapter 2: Model Description ‘The functional relationship between air/fuel ratio and MBT is not modeled since air/fuel excursions from stoichiometry exist for only short periods of time, and neglect- ing this association decreases testing time by a significant amount, since the functional map has one less independent parameter. There is a process delay from intake to spark (ate ), and another from spark to torque production (At ~ £4343*.=), which turn out to limit the response of any closed loop engine control algorithm. In effect, they provide nonminimum phase zeros, or zeros in the right hand half of the LaPlace domain, which limit the performance of any closed loop control system that is applied to this engine. These delays can potentially provide stability problems for feedback systems, which will be discussed in Chapter 5. To obtain an estimate of the actual brake torque produced by the engine, an esti- mate of the engine’s friction and pumping torque must be subtracted from the gross indicated torque. This friction/pumping torque is assumed to be a function of the in- take manifold pressure, exhaust manifold pressure, and engine speed. If a relationship can be found between the exhaust pressure and the intake manifold pressure and engine speed (utilizing a volumetric efficiency function), then the friction/pumping torque can be estimatea as a function of intake manifold pressure and engine speed, or: Pe= filtinaos we) Fira = Lawes Pst) p= Pe = Lal fale, Ps fal Pmsie)]i a fa(wes Pm)» (2.61) ‘tte = fo Pye) While these are not exclusively functions of the variables shown tte = {Pm We; Tm, Toort; Teootane residual fraction, geometry,...]), strong functional re- lationships do exist, and estimates of these quantities besed on the variables shown -68— Chapter 2: Model Description prove to be useful in torque estimation. ‘he combustion /torque model is given as: aon Te = { lets TF -aFi]| st} Tp Feo (eaten Seated a ari © Heul-Wele ate) =Trp. (282) Vacated AP tit Aeon E.G.R. is treated as a diluent [33] and affects torque production in two ways: « by decreasing ‘nao from the intake manifold, and « by altering the MBT spark advance. There are two parts of the torque production model that are not directly measure- \ able with the test set-up that was used. They are che friction pumping torque, and the torque function (or the gross indicated stoichiometric fuel conversion efficiency). While a careful analysis of accurate cylinder pressure indicator diagrams, along with flow measurements, could give considerable insight into the indicated fuel conversion efficiency, such data was not obtained due to time and equipment constraints. In the first attempt at resolving both of these unknowns, the actual firing friction/pumping torque is assumed to be equal to the motoring friction/pumping, and then to calculate the resulting torque function from engine test data. This assumption produced fuel conversion efficiencies that were higher than ideal Otto cycle estimates, and that ran counter to the expected trends which will be discussed in later chapters. The idealized firing and motoring pumping loops are shown in Figure 2.11. The most significant difference between these two cycles is the state of the gases in the cylinder just prior to the opening of the exhaust valve. In the firing loop, the cylinder gases are quite hot, and still at a pressure significantly higher than the exhaust pressure. When the exhaust valve is opened, there is a blow-down of these hot gases -69- Chapter 2: Model Description Pressure Pressure Pe P Pm Pm Volume Volume Firing Motoring Figure 2.11 Firing and Motoring Pumping Loope into the exhaust manifold, and then most of the remaining gases are displaced during the exhaust stroke, In the motoring loop, just before the exhaust valve opens, the cylinder contains relatively cool, dense gases that are significantly below either the exhaust or intake manifold pressures. During the compression and expansion strokes, there is some heat transfered to the cylinder walls and the engine coolant. Because of this energy loss, the cylinder gases after expansion will be cooler and at a lower pressure than they were at intake (approximately intake manifold conditions). So, when the exhaust valve is opened there will be a blow-down of gases from the exhaust manifold into the cylinder (opposite direction from that of the firing loop). Then cool dense gases will be displaced from the cylinder during the exhaust stroke. The inertia effects of this reversed flow blow-down, and the increased discharge losses through the exhaust valve due to the denser gases account for most of the discrepancies between the firing and motoring pumping loops. Friction differences between these two loops are -T0- Chapter 2: Model Description caused by higher temperatures and greater bearing loading and ring pressures during firing loops. Assuming the motoring friction/pumping torque can be used to model the firing friction/pumping torque was not feasible, since this approach gave unrealistic fuel con- version efficiences. The approach that was finally used was to estimate a torque func- tion (or fuel conversion efficiency), and then infer the friction/pumping torque from this estimate and engine test data. Engine test data and and engine simulation model by Paulos and Heywood [34], which was later expanded by Nitschke [35] and Chang [32], was used to estimate the torque function. An estimate of the friction/pumping torque is derived from this data and simulation which is shown in Figure 2.12. The friction /pumping torque model used in the engine of interest is given as: Tyjp= TF MAC: AFI-SI~Try (2.63) = (ap) maces 75) mace (K+ 2 (2.64) TP, P vp) P = 0.36 x 107° - N + 0.2651 x 1077, (2.65) K = 0318 x 10°. N? 0.99 x 107? +33, (2.60) H = 0.182 x 1077 N? 40.514 x 107*- +042. (287) With this model an estimate of the brake torque can be made, by: Te -Typ- (2.08) The dynamics of the engine inertia are modeled in a constant inertia, lumped parameter model. In an actual engine the effective polar moment of inertia of the crankshaft, connecting rod, piston, and valve train assembly changes cyclically, due -n- Chapter 2: Model Description FRICTIEN/PUNPING T@ROUE HADEL (het 6 8 8 8 8 8 70510 «1S .20 2830S AOS SOSH «CO RSS OF AIR PER CYLINDER (G) Figure 2.12 Friction/pumping Torque Model to the varying geometry of the slider-crank mechanism. Also, the crankshaft is not a rigid member, and torsional vibrations are set up in the crankshaft due to the pressure pulses from the firing cylinders. Typically a torsional vibration damper is mounted on the front end of the crankshaft to dampen out any torsionals excited by the cylinders. Because of the level of lumped parameter modeling used in the engine model, which is dictated by the simplicity goal, neither the variable inertia nor the crankshaft torsionals are modeled. Instead, a rigid crankshaft is modeled with a mean effective crankshaft, connecting rod, piston, valve train inertia, and the rotational dynamics are modeled, using Newton’s second law, as: «= 5) [Sas (2.69) ltt dhe The effective inertia (J.j,) would include any rotating inertias whose rotational veloc- ities are assumed to be linearly proportional to the engine's speed, such as accesories -72- Chapter 2: Model Description connected to the crankshaft pulley (alternator, power steering pump, etc.). All effective inertias are calculated by summing the actual inertias divided by the connecting gear ratios squared, or: eee (2.70) Re where: (27) The external torques (7;) include loading on the engine from the torque converter, and any accessories mounted on the front of the engine. ‘These include electrical loading from the alternator, and loading by the power steering pump and the air conditioning compressor, etc... lodel This engine model can be used in three ways. It can be used as a simulation model to assess controller designs or system responses, as a validation model to compare model and actual outputs for parameter identification (given the inputs), and it can be used as a part of the control algorithin within a vehicle. These models vary by what inputs are available to the model (such as intake manifold pressure, engine speed, etc.), and what is the required output of the model (i.e. brake torque, engine/dyno shaft torque, ete.). ‘The simulation model will be presented first, and then changes to the model for validation or control will be discussed, Chapter 2: Model Description Throttle body: wt 0-0) a | eat: oe (2 em es(xh)" -Ma( Fe) TC(8) pai( 3) ; (2.73) a= £2- (BYP oA? (Gast) Zl (6)T) D cos(G. +8) 2 ee Idle air control: t sayy nace | neta YF ry IMEC) Tg (a) ‘nea (count) -y/ Ties - pF, ges(ch)™ E.G.R. valve: tng = 4 Peat valves fF te Helle tinea (valves) «yf Test - phe, se-<(mh) Intake manifold: dpe (F-« wom) P+ BZ (ings + tag) (2.77) (incre) = Ki (nears tnegra) (2.78) 4 Ginare) = Ki Mote reget ~ taps . tnce = FAY sayy E taero (2.19) -14- Chapter 2: Model Description Fuel delivery: tga = tga ty ys + tyes tings = [rhgelg_any © (1-Die-any tigga = thgela any 7 tapel-any(h~9 — fag 7 fy if DEGON < kinso 1= Peson, otherwise Torque production: aon T= (Fare Fre TR APD 6,24) SD) e-ateaz) ~ l0 oe can Mle = 1439-17 lose = -157.5 + 0.926 - nor + 0.854 nar + 0.647 -ny +0.647 -npyz + 1.087 ny, + 1.016-ny = 15.1-.N-0088 Nor = 0.729 m4 — 0.226 «re 1 = 56-2) 100 ng = 37.396 a nope sono. (2) 45.7 - 13.8 - V8 -15- (2.80) (2.81) (2.82) (2.83) (2.84) (2.85) (2.86) (287) (2.88) (2.89) (2.90) (2.91) (2.99) (2.93) (2.04) Chapter 2: Model Description For the modeled engine: p, \-7a88 100-1 Ijote = 1.09 [ise -3:723. (#5) 0.77 wz S1=1-3.3x10"*-(SA~MBT)? MBT = 88 - MAC? - 99.59 MAC + 47.61 + 0.003 -N +58 tha + Tear AFI = (Air/Fuel) Ty {p= TF -MAC- AFI-S1— Toe = (cs) «MAC? + (j *) MAC+ (+ P =0.36 x 10-* - N + 0.2651 x 107? K = 0.315 x 107° - N? 0.99 x 1077. + 33, H = 0.182 x 107 -N? +0514 x 1074. +0.42 Engine (converter pump) dynamics: 4, T= Typ- are Je (2.95) (2.96) (2.97) (2.98) (2.99) (2.100) (2.101) (2.102) (2.103) (2.104) The torque converter inertia is combined with the engine inertia as a lumped param- eter. The other inertias ans models (i.e. torque converter, transmission, etc.) can then be added on this model, which effectively provides an additional torque on the engine/pump inertia, to vary the engine speed. The validation model will often include the test set-up, such as the engine and dynomometer system, and dynomometer and/or engine speeds may be provided as -76— Chapter 2: Model Description inputs to the model. The output may be the torque in the shaft connecting the engine and the dynomometer. dus = (2.105) dun Bath Tye a (2.106) €- (Gea) + K- (ue 02) (2.107) Depending upon the measurements of the test set-up, two of the previous three equa~ tions may be used for validation. Bear in mind that the indicated torque will not predict the firing torque pulses, and the engine speed (u,) will deviate from predicted values as a consequence. For the test cell set-up used in this study, the shaft parameters are given as: K-=2965 N-m/rad (2.108) .08 N-m-s/rad (¢=0.035) (2.109) Various inputs may be used, depending upon what is being validated. Throttle angle and IAC count may be inputs to estimate the air flow into the intake manifold, or a mass air flow measurement may be available. Intake manifold pressure may be measured or predicted. The control model is the simplest model, which utilizes all available measurements to estimate flows and torques. Air flow out of the intake manifold can utilize the mass flow measurement, with the pressure measurement, to give: ngs + thrac = mass air flow sensor output (2.110) Pm = measured (2411) -17- Chapter 2: Model Description as *) Specds are usually inputs into the model, and the brake torque is the model output tinge = thas + tiheges — Tegra + tings + Fnegrs — tegro — (2.112) » 4 (or turbine torque in the torque converter). Chapter 5 shows how these simple engine models can be inverted to give an estimate of the desired engine control (throttle or spark advance), and how they affect the total system performance. ~78- 3 Model Modifications and Validation Tests 8.1 Introduction When the control engineer attempts to develop internal combustion engine models for control algorithm development, he is often confronted with two problems: + the engine of interest is sufficiently different from any previously established model such that @ new structure or new parameters must be developed, or the engineer has a limited data set and does not have the means to undertake a complete validation program. The thrust of this chapter is to discuss how the engine model can be adapted to various engine configurations, and to suggest what to do in the case of incomplete validation data sets. The author will indicate which parameters are the most important in making an accurate engine torque prediction, and which parameters play a less important role. ‘The chapter is subdivided into engine subsystems, with a short section discussing topics not included in the original engine model. As with any modsl, the more high quality test data the engineer has at his disposal, the better will be the resulting engine model. Ideally, the test engineer would have at his/her disposal an engine in a test cell with high speed data acquisition equipment. The test engineer would know his/her model well, and be able to proceed with his model validation activities concurrent with -79- Chapter 8: Model Modifications and Validation Tests engine testing. While this leads to a busy achedule, it gives the engineer the luxury of iterating on a solution and the ability to investigate areas in which questions or uncertainties arise. Also, validation procedures sometimes suggest methods of testing or specific types of required data. By doing the validation concurrently, the engineer can be flexible and responsive in his/her testing so as to accommodate the particular needs of the validation procedures being used. However, most of the time circumstances dictate the amount of available data or the timing of the testing and validation. In this chapter the author attempts to give the reader direction on how he/she can synthesize a reasonable real time mean torque estimation model with some reasonable amount of available test data. 3.2 Throttle Body, Idle Air Control, and E.G.R. Flows Models representing flows into the intake manifold (equations (2.72) through (2.76)) are based on the one-dimensional isentropic compressible flow equation. This equation has the advantage of representing the basic properties of the flow with a simple math- ‘ematical form. The problem is that actual flow is neither one-dimensional [37,38], nor steady. Errors in assuming unidimensionality must be folded into the discharge coef- ficient (C4). This discharge coefficient is the parameter which must be identified, as a function of throttle angle and pressure ratio across the throttle. The remaining por- tion of the equation is known for a given pressure ratio, throttle angle, throttle bore diameter, and throttle rod diameter. It is assumed that the control engineer knows the geometery of the throttle, in order to determine cross-sectional area. If there are mul- tiple throttle plates, the areas of the throttle openings (equation (2.74)) are summed. The discharge coefficient can be empirically determined by measuring mass air flow —80- Chapter : Model Modifications and Validation Tests rate through the throttle body while it is mounted on the engine, a flow bench or cali- bration apparatus. It is important to measure the flow rate in as much of the throttle angle-pressure ratio plane as possible. This is much easier to do when the engine is in a test cell, coupled to a dynamometer. However, it is feasible to do this test with an engine in an automobile, provided there is a good measurement of the flow rate. There can be a pressure drop across the air cleaner at high flow rates, so use of the intake system which will be used on the engine is recommended. Using a flow bench to measure flows at various throttle angles and pressure ratios van be convenient, provided the experimental set-up is available. An important point, both for identification tests and for flow estimation, is that the pressure ratio must be measured carefully when it approaches £ Taking the derivative of the mass air flow rate with respect to pressure ratio, for nonchoked flow, yields: (3a) and: (8.2) Small variations in the pressure ratio (near 1) yield large variations in mass air flow tates. So, if flow through the throttle body is to be estimated with a throttle body model, careful measurements of pressure ratio must be made. This is the biggest practical shortcoming of estimating flow with equation (2.72). The transformation of equation (2.72) into equation (2.73) is useful, since calculation times can be decreased —81- Chapter 3: Model Modifications and Validation Tests considerably with two look-up tables and a stagnation condition adjustment (which can have a slow sample rate). If there is a good air flow sensor on the throttle, it may be used to calibrate the throttle model. If it is fast, it may be used in place of the model to estimate air flow. Beware of air flow sensors with long time constants, or those that are positioned a considerable distance from the throttle. Estimating air flow through the Idle Air Control valve is usually simpler than the throttle body, although the one dimensional compressible flow equation is still used. ‘Typically, the geometry of the flow is much simpler than that of the throttle body, and the valve often has radial symmetry, such as with a cone valve. Usually, though not always, the discharge coefficient is most strongly a function of the valve geometry (not pressure rati |, and the ideal choked flow characteristic can be assumed. The discharge coefficient can be lumped with the cross-sectional area term. The only other flow adjustment is for changes in stagnation conditions. Idle Air Control valve geometry may vary widely, so flow characteristics should be checked carefully. The method used for determining the discharge coefficient /cross-sectional area term is to measure the flow while varying the input to the LA.C. valve, and maintaining a constant pressure ratio by changing engine speed. This is easily accomplished with a dynamometer. Various pressure ratios can be used, for different tests, to check the ideal choked flow assumption. A estimate of E.G.R. mass flow rate into the intake manifold can be made, if all parameters in equation (2.76) are known. Typically, this is not the case. Exhaust temperature and pressure are seldom measured, but are needed for the estimate, since the exhaust conditions (at the E.G.R. valve) are now the stagnation conditions (as- suming low velocities). An exhaust pressure model can be developed, based on engine ~82- Chapter $: Model Modifications and Validation Tests parameters (such as air flow, torque, etc.) from engine test data, or from other sources such as Takizawa [39] or Kay [40]. Chapter 6 of Heywood’s text [3] contains a good set of references on the gas exchange process. Since exhaust temperature and pressure are correlated, simple phenomenological correlation models can often be developed, which simplifies the flow estimates considerably (see Chapter 4). If exhaust restrictions raise the exhaust pressure a considerable amount, the flow through the E.G.R. valve may be considered to be choked at all times, which considerably simplifies the flow estimation equations. A typical E.G.R. ve ve has a single orifice, controlled by modulating the vacuum to a diaphragm, which moves the valve. With this type of valve it is difficult to calibrate the flow through the valve, as a function of valve position. At minimurn, the position of the valve must be sensed in order to get a calibration of flow. This is not a trivial task with this type of valve. Secondly, these valves tend not to be closely calibrated, and considerable variation of flow amoung different valves can exist. Some automotive manufacturers are using solenoid operated E.G.R. valves with multiple, calibrated orifices. With this type of valve, the flow at ambient conditions, as a function cf open orifices (combined flow area), can be estimated from calibration tests and solenoid signals. This is the type of valve used on the engine described in Chapter 4, and can be useful in estimating flow. 3.3 Intake Manifold The purpose of the intake menifold model is to estimate mean mass air flow rate into the cylinders, given air flows through the throttle body and I.A.C. valve, and E.G.R. flow. This is an important function, since this flow rate will differ from the -83- Chapter $: Model Modifications and Validation Tests flow rate into the manifold. This flow rate out is what the fuel rate is trying to match (for a 14.7 air/fuel ratio), and it has a strong influence on the indicated torque produced by the engine. For the control model (for use on the engine), air flow out can be estimated by knowing the flows in, manifold pressure and temperature, and an estimate of the E.G.R. flow out of the manifold (equation (2.112). In the current model, the estimate of E.G.R. flow out is a lagged version of the flow in, with a time varying time constant (equation (2.78)), based on the engine speed, and an estimate of the volumetric efficiency (based on intake manifold density). This time constant is the only place where a volumetric efficiency estimate is needed. Since the E.G.R. fraction is small, only a rough estimate of the volumetric efficiency is sufficient. Except in extreme conditions. temperature variations are aot great enough (on an absolute scale) to have a major effect on flow estimates, so the temperature derivative term in equation (2.112) was dropped with negligible loss of accuracy (see Figure 2.1). The simulation model (equations (2.77) through (2.79)) is a different matter. Vol- umetric efficiency estimates (based on intake manifold density) play a crucial role in estimating air flow out of the intake manifold, and can have a profound effect on in- dicated torque estimates. Error in estimating volumetric efficiency is one of the major contributing factors in discrepancies between measured and estimated torque values shown in Figures 4.32 through 4.36. It is strongly suggested, if the control engineer has any way of empirically determining volumetric efficiency for his engine, that 1nea- surement be taken and used in the simulation model. Figure 3.1 shows a breakdown of the major factors affecting volumetric efficiency. This plot is taken from a Master of Science thesis by Nitschke [35]. This thesis is use- ful in understanding the effects of various engine parameters on volumetric efficiency. -84— Chapter $: Model Modifications and Validation Tests Induction Flow Friction! Volve Flow Friction 10. Flow Choking 65 60. ss. Ros Effect (Viv Timing) VOLUMETRIC EFFICIENCY (%) a 3000 3500 4000 4500 sd00 sS006000 ENGINE SPEED (rpm) Figure 3.1 Volumetric Efficiency Breakdown [35] Estimating an engine's volumetric efficiency from engine timing, intake geometry, heat transfer models, etc. usually leads to a crude estimate at best. Complex intake, ex- haust, and valve geometries have strong influences on friction losses, heat losses, exhaust gas backflow, and ram effects which are difficult to predict a priori without an extremely detailed model. For the models discussed in this thesis, an empirical volumetric effi- ciency model is much more accurate, and time and cost effective to acquire than are a priori predictive models. However, if the control engineer does not have access to an engine for these tests, Nitschke’s thesis can be helpful in making an estimate of the volumetric efficiency. Taylor [41] compares the volumetric efficiency characteristics of several engines, and discusses cause and effect relationships of the efficiencies. There are also several papers that are useful in estimating efficiency [39,42,43,44]. For the simulation model, a constant manifold temperature estimate is sufficient. The intake manifold volume is assumed to be known. -85- Chapter 8: Model Modifications and Validation Tests The assumption of uniformity appears to work well with the intake manifold of the engine tested in Chapter 4 (relatively large plenum and short runners). For an engine with long runners and a small plenum this model may not work as well, since there would be a greater pressure difference between the plenum and the runners at the valves. For the controi model with this configuration a multi-zone homogeneous model may be used. The runners and plenum may each be considered a homogeneous system, with their respective pressures (and temperatures) measured. Pressure drops in the runners would indicate flow through the runner. With flows into the plenum known, and continuity at the plenum-runner interface, as well as total flow out, defined by volumetric efficiency, the system would be well defined. This model would be able to use the additional runner pressure information to give a pulsed estimate of air out of the intake mausifold, and the possibility exists of extending the entire model to estimate the pulsed torque response of the engine. A further extension would be a finite element model with heat transfer and intake geometry, but run time problems with these extended models exist. To be useful as a control model, it must run in eal time, and while these extended models would provide more information, they also require much faster processors. Another important consideration in modeling the intake system is to note where the Positive Crankcase Ventilation (P.C.V.) gases enter the intake system. If they enter directly into the intake manifold, at low manifold pressures this P.C.V. flow is sufficient to give large errors in air/fuel ratio if not modeled. If these gases enter the induction system prior to the throttle throat, the flow never gets large in magnitude, and subsequent errors are negligible. -86- Chapter $: Model Modifications and Validation Tests 3.4 Fuel Delivery Fuel delivery characteristics can vary considerably, depending upon the physical arrangement of hardware and type of fuel algorithm that is used. The engine modeled in Chapter 2 utilizes sequential port fuel injection. The types of engines that can be considered, by making modifications to this model, are carbureted, throttle body injected, and port fuel injected engines. Modeling fuel metering of a carburetor can be complex, but there are several good papers an‘ texts on this subject that can help the control engineer in his formulation. Hartington and Bolt (27] give a detailed analysis of fuel metering in a carburetor, as well as a list of important related work on modeling of carburetor subsystems and physical processes. Papers by Shinoda, Koide, and Yii [45], Bolt and Soerma [46] give the reader greater physical insight into processes of carburetion, as well as an appreciation of the difficulties of modeling the system. Taylor [41], Khovakh [47], and Heywood [5] all present the basics of carburetor operation. In the carburetor, a pressure drop in a venturi in the air stream is the potential which entrains fuel into this airstream. Figure 3.2 shows a schematic of a carburetor in its simplest form. The task of modeling carburetion involves predicting fuel flow rate, given air flow rate by means of the throttle equations. The main characteristic of carburetion is that the air/fuel ratio increases at decreasing air flows. If the pressure drop at the venturi is less than the pressure head of the fuel: APS py gh, (3.3) then no fuel will be delivered, since the pressure potential cannot overcome the gravity potential. As the pressure drop increases (from increased air flow), more fuel will be -87- Chapter 3: Model Modifications and Validation Tests fuel supply Pate a throttle plate hd Figure 8.2 Simple Carburetor Schematic (5] delivered until the air/fuel ratio remains nearly constant. A fuel metering orifice is set so that the desired air/fuel ratio is achieved at some intermediate flow value. Several auxiliary devices have been added to this basic carburetor to compensate for its many weaknesses. A seperate idle fuel delivery system has been added for light loads, to avoid leaning of the mixture in this region. An accelerator pump adds jonal fuel when the throttle is opened quickly (essentially proportional to the throttie rate), to compensate for fuel lags that occur in the intake manifold. Many systems enrichen the mixture at high air flow rates to give the maximum torque in this region (an equivalence ratio t of approximately 1.1). The air/fuel ratio is usually decreased by a choke when starting the engine, to aid in combustion. This device provides an air restriction between the air bleed orifice and the venturi. With the t equivalence ratio = —88— Chapter 8: Model Modifications and Validation Tests choke closed, a large pressure drop can occur at the venturi, with little air flow, thereby enriching the mixture for start-up. There is often a small relief valve in the choke to prevent the air flow from being completely stopped. Most modern carburetors provide a device which compensates for large changes in ambient pressure (i.e. altitude compensation). All of these compensations are attempts to correct for shortcomings in the basic carburetor. A sufficient carburetor model for control development. depends highly upon the specific device used on the engine to be controlled. Unless it is an electronic carburetor, the control input would be the throttle angle. Fortunately for the control engineer, most modern automobiles are being equipped with fuel injection systems instead of carburetors, which makes the modeling task easier. Once carburetor characteristics have been modeled, fuel transport to the intake valve must be modeled (i.e. fuel transport across the intake manifold). Aquino [11] has done considerable modeling of these processes, and has developed fuel transport models which have been correlated with data from a Ford § liter engine. This model is summarized below. Liquid fuel film in the intake manifold: da Fao HMA (Lt Emp (1= 6) ge &. (3.4) Fuel vapor in the intake manifold: 4 1 treo ; GMO ME Me FE Ts. (3.5) Total fuel flow out of the intake manifold: thyeo + tito (3.8) Chapter 3: Model Modifications and Validation Tests In this model ¢ is the fuel split parameter (as in Chapter 2), or (1) ray: is the fuel flow rate that goes into a liquid film. Aquino has modeled ¢ as a linear function of throttle angle, or: aes at pes. (32) For large throttle angles, a greater percentage of the fuel forms a film on the floor of the intake manifold. The constants « and f must be identified for a specific intake geometry. n is the boiling time constant, and is modeled as a function of intake manifold temperature (middle front skin temperature). Aquino gives estimates of the boiling time constant and fuel split parameter for a 5 liter engine. Other models, which are similar to equations (2.80) through (2.83), with 7=1 and long time constants r, have also been used to simulate fuel transport across the intake manifold. Fuel injection has many advantages over carburetion, and is being used extensively ‘on modern automobile engines. In the carburetor, the pressure drop in the venturi drives the flow rate of fuel. In fuel injection, the air flow rate must be measured (with a mass flow sensor) or estimated (speed-density equation (2.79)). Based on this information, the electronic control module controls the injector pulse width. Injector dynamics are typically much faster than fuel flow dynamics, so fuel injection response is assumed to be instantaneous, although this flow should be calibrated as is Figure 4.10. With throttle body injection, fuel must still traverse the intake manifold. Equations (3.4) through (3.6), or a similar model, can be used for this purpose. With throttle body injection, the type of injection algorithm determines the model. A sequential injection scheme is modeled in Chapter 2. Another widely used algorithm is called simultaneous double fire. In this aigorithm all injectors are fired simultaneously once -90- Chapter 8: Model Modifications and Validation Tests every crankshaft revolution. The fuel response to the cylinders differs from sequential fuel injection. The first delay of simultaneous double fire is from the output fuel calculation of the E.C.M. to the subsequent start of injection. Since injectors now fire every 2.x radians, the mean delay is given by: At = Fe seconds. (3.8) ‘The next delay is from start of injection to when the first intake valve closes following the end of injection (A@;,;+), or: A= ed seconds, (3.9) This is the first time a cylinder’s contents will be fixed at a new level. However, if the pulse width has been lengthened, the cylinder will contain the fuel from one short pulse and one long pulse. As a result, the fuel in this cylinder (assuming no fuel lag) is halfway between the old fueling rate and the new fueling rate. One crankshaft revolution later (Aty = 2% seconds) there will be a cylinder with two long fuel pulse widths of fuel. Fueling transitions are stepwise with this algorithm, which is shown in Figure 3.3. The slow fuel algorithm can be modeled as before, so the fuel model can be given as: Thyo = tayya t tyyat tmyany (3.10) tay = [es e-tnge| -an-anl-as! (ean fingys = 0.5 tye] (3.12) Ie-atn—ata)” d. rrgele-any - 9 - gia= a 7 (3.13) ‘The nonlinear estimation method described in Chapter 4 can be used to estimate fuel parameters. -91- Chapter 3: Model Modijications and Vatidation Tests Fuel In alge Width —— Pulse Wit — 1 —" — Intake 2 nO = Valve is 4 Open Figure 3.3 Simultaneous Double Fire Fueling Algorithm Transition 5 Torque Production The estimate of mean torque production is divided into friction/pumping and in- dicated torque. The maximum stoichiometric, MBT indicated torque is based on an estimate of gross indicated stoichiometric fuel conversion efficiency. If the control engi- ner is able to collect cylinder pressure records at many different operating conditions, indicated and pumping torques can be evaluated from these plots by measuring the areas of the compression-expansion loop, and the pumping loop, on @ pressure-volume diagram. Typically this information is not available, so indicated fuel conversion effi- ciencies must be estimated by other means. Engine cycle simulation models developed at M.LT. by Paulos and Heywood [34], Nitschke [35] and Chang [32] can be useful in estimating this function. Chang developed a model for gross fuel conversion efficiency -92- Chapter 8: Model Modifications and Validation Tests based on cycle simulations using Woschni’s heat transfer model [48], and the burning and entrainment models developed by Beretta, Rashidi and Keck (49]. This efficiency model is given in equations (2.87) through (2.94). If a simpler model is desired, fuel conversion efficiency can be approximated as a constant, since it does not vary by a large amount (see Figure 4.23 or Chang (32]). Estimates of air flow rates out of the intake manifold, fuel conversion efficiencies, and equation (2.86) can be combined to yield the mean stoichiometric MBT indicated torque. This maximum torque can be adjusted as a function of the air/fuel ratio by defining the air/fuel influence (see Figure 4.28). Chapter 4 explains how this function was measured and what equipment was used. If a fuel setpoint controller is not avail- able, this function can be approximated near stoichiometry by Edson and Taylor's [33] data. Air/fuel influence is calculated by: AFI= % / Be Bow (3.14) Farther from stoichiometry the air/fuel function reflects misfires and partial burns for both lean and rich regimes. These boundaries may be estimated using the data of Quader (50,51]. In the engine tested in Chapter 4, torque dropped quickly between air/fuel ratios of 7 and 6 (i.e. AFI 0.8 to AFI 0.0, respectively). The exact values of these burn limits are not crucial, since there is a large coefficient of variation in torque at these points, and the engine spends little time at these air/fuel ratios. ‘The spark influence function is easy to determine, given spark and torque data. Most electronic control modules have controllers that can independently step spark advance. If this type of data is not available, data such as equation (2.96) may be used. Haider and Griffin [52] provide similar data on the spark influence on torque. Most -93- Chapter $: Model Modifications and Validation Tests engine map data provides spark influence, as it is a commonly derived parameter. Determining the MBT point is usually a little more difficult. However, if one has the spark data for many different flows and speeds, MBT data can be extracted by the process defined in Chapter 4, equations (4.64) through (4.68). The sequence of identifying the spark parameters is important. Spark influence data that is measured directly from spark perturbation tests is based on brake torque, or: Te si| . lorake Theor (3.18) The brake torque is measured, and the MBT torque can be estimated by a quadratic fit of the data. Spark influence used in the engine model is based on indicated torque, or: (3.16) Spark influence based on indicated torque can be expressed in terms of known param- eters. B+ Tp Tot Blyar~ Fe Baer Ter (3.7) Ti] pp 204 Ts are known from spark advance and torque measurements, and 7, ear calculated from the engine model as: Tuer = He (o.18) ‘This analysis assumes that friction/pumping torque is not a function of spark advance. Estimating friction/pumping torque can be one of the most difficult aspects of the model. If cylinder pressure data is available, indicated and pumping torque can be estimated from this data, and friction torque can be calculated as: Ty =T-Ty- To (2.19) 94. Chapter 3: Model Modifications and Validation Tests If an estimate of the fuel conversion efficiency is used to estimate indicated torque, then friction/pumping torque may be calculated as: Typ = Te~ To (8.20) If brake torque data is not available, it is difficult to get an estimate of friction and pumping torque which can be subtracted from indicated torque estimates to yield brake torque. Wide-open throttle motoring torque is often used as an estimate of friction torque, but this estimate may be off by several percent, since influences of heat and pressure during firing are not directly included. Nitschke [35] has a rather thorough treatment of the functional relationships of mechanical losees, and can be useful in estimating friction/pumping torque. A paper by Bishop [53] is a classical investigation into the contributions to friction/pemping losses. Delays associated with the four- stroke process can be calculated from valve timing diagrams which are assumed to be available to the control engineer. 3.6 Additional Topics Additional $.1. engine configurations, not discussed so far, are engine supercharg- ing and turbocharging, which are used on some automobiles. The addition of a super- charger model to the current engine is straight forward. The supercharger’s speed is some ratio times engine speed. Flow through the supercharger is a function of this speed and the pressure ratio across the supercharger. A performance map of the supercharger of interest would be required to estimate this flow. Outlet pressure of the supercharger can be estimated by knowing the outlet plenum volume, estimating temperature, inte- grating the flows, and using the ideal gas law. Effectively, the supercharger raises the ambient air pressure of the current model. Chapter 8: Model Modifications and Validation Tests Modeling turbocharging is more complicated. The turbine and compressor are linked together so that they rotate at the same speed. A pressure ratio across the turbine forces a flow through the turbirie, and directional changes of the flow cause a torque on the turbine rotor. In the compressor, changes in flow direction caused by the rotating blades generate a torque on the compressor rotor. Integrating the torques and dividing by the moment of inertia of the turbine/compressor yields the rotor speed. Steady-state efficiency maps may be used to determine the flows as a function of rotor speed and pressure ratio. Intake and exhaust pressures are determined by the ideal gas law and integration of the flows. Again, the turbocharger effectively raises the inlet pressure of the engine model, as well as the exhaust pressure (due to the flow restriction of the turbocharger). The rotor speed is determined by the torques and inertia of the rotor. Consequently, there can be lags and surges in the intake manifold pressure, which result in torque lags and surges. Van Wylen and Sonntag [54] provide an analysis of flow through turbine or compressor blades. Watson and Marzouk [55] give details of a nonlinear dynamic model of a turbocharged diesel engine, which can be adapted to an 8. I. engine. Also, Watson and Janota [56] is a good source of information on internal combustion engine turbocharging. -96- 4 A Case Study in Engine Validation 4.1 Introduction Although there is much analysis of steady-state engine operating conditions, very little actual dynamic engine model data, for control development, is available to the control engineer. Since the engine model outlined in Chapter 2 was crucial to the development of control algorithms, the author was given the opportunity to set up and conduct dynamic engine validation tests cn a port fuel-injected engine. These tests were conducted at the General Motors Research Laboratory, Power Systems Research Department test facility from May 198€ to December 1986, with additional tests in March of 1987. A short summary of the tests is given below. « Friction/Pumping Test - Steady-state Engine Motoring = Tyrie = fey Pn) » Steady-state Engine Maps - Torque = (0,u.) - Volumetric Efficiency = (Pm ; Tmive) - Throttle Body Validation -LA.C. Calibration + Fuel Injector Calibration - Net Fuel to Engine = f(Injector Degrees On) -97- Chapter 4: A Case Study in Engine Validation « Throttle Step/Ramp Dynamics - Variable Fuel and Spark Advance Controlled by E.C.M. + Spark Advance Dynamics - Fixed A/F = 14.7 - Torque Response to Spark + Fuel Perturbation Dynamics - Fixed Spark Advance Set to E.C.M. Map - A/F Influence on Torque - Fuel Delivery Dynamics + E.G.R. Dynamics Test - Torque Response to Steps in E.G.R. - Torque Response to Spark at Various E.G.R Levels Each of these tests (except for throttle step/ramp dynamics) was used to identify parameters in at least one of the engine subsystems defined in Chapter 2. The throttle step and ramp data was then used as an independent data set, to validate the entire model. This chapter describes the approach taken to identify unknown parameters in the engine. dw Engine model validation tests were performed at the General Motors Research Lab- oratories, Power Systems Research Department test facility. The tests were conducted on a naturally aspirated V-6 sequential port fuel injected engine. A load torque was imposed on the engine by a dynamometer, coupled to the engine with a compliant shaft. -98- Chapter 4: A Case Study in Engine Validation Figure 4.1 Experimental Test Set-up, G.M. Research Laboratories -~99- Chapter 4: A Case Study in Engine Validation A dynamic analysis of the engine-dynamometer system is discussed in Chapter 6. Fig- ure 4.1 shows the test cell set-up. For the tests that involve identifying steady-state parameters (i.e. throttle discharge coefficients, volumetric efficiency, etc.), an electronic control module was used to control the spark advance and fuel injector pulse width. The throttle was controlled by the operator. For all tests, except the E.G.R./spark advance tests, E.G.R. flow was set to zero by closing all of the orifices on the E.G.R. valve. During the E.G.R. tests, the E.G.R. valve solenoids were controlled by an in- dependent source. Engine oil temperature and coolant temperature were controlled by a closed loop heating/cooling circuit. Desired oil and coolant temperatures were specified by the operator, which minimized the effect of these variables on the data. The fuel delivery system can be seen in the lower left hand corner of Figure 4.1. This system consists of a float bowl, which maintains a constant head to the fuel flow meter, and then to a fuel “make-up” box. This “make-up” box contains a float and valve to maintain a near constant volume of fuel inside of the box. There is a small limit cycle to the float device, so steady-state fuel flow measurements are averaged over long periods of time for an accurate reading. From the box the fuel goes through a pump, pressure regulator, and shut-off valve to the fuel rail on the engine. Fuel returning from the rail goes directly to the “make-up” box. The throttle is controlled by the operator via a stepper motor connected to the throttle linkage, and a potentiometer is used to measure the throttle angle from closed. The velocity of the throttle is controlled by a joy stick, or the stepper motor can be controlled by an indexer, with velocity or acceleration profiles set by the operator. An estimate of the exhaust air/fuel ratio was made with an analyzer, shown in the lower right-hand corner of Figure 4.1. This sensor was quite useful for steady- -100- Chapter 4: A Case Study in Engine Validation Exhaust Manifold Solenoid Operated Oz Shut-Off Valve © 1 O2 Supply Pressure Regulator TT apr sensor 34" (68") ———+ Figure 4.2 Air/fuel Sensor Calibration Set-up state data, but had limited success in measuring transient air/fuel data. A test was devised to calibrate the dynamic response of the air/fuel sensor. Figure 4.2 shows the test apparatus. Additional oxygen is added to the sample tube that draws exhaust gases from the exhaust manifold and leads to the air/fuel ratio sensor. This additional oxygen leans out the exhaust sample to an air/fuel ratio of 21. The solenoid suddenly closes the valve, and the air/fuel sensor response is recorded, along with voltage to the solenoid. Figure 4.3 shows a typical sensor response, along with validation results. The model of the air/fuel sensor is a delay followed by a first order lag, or in the Laplace domai (4.1) air/fuel), G57] Full cosurea -101- Chapter 4: A Case Study in Engine Validation a 20 ND MODEL IN/@UT? 19 AIR/FUEL RATIC (ACT.. o 8 «10 1S 20 25 30 35 40 45 5.0 5.5 6.0 TIME (SEC) Figure 4.3 Air/fuel sensor Response to Air/fuel Ratio Step ‘The mean delay was calculated as Ats,” = 0.234 seconds. If this delay was simply a result of the sample velocity in the tube, then changing the length of the tube should give a delay proportional to the tube length. By replacing the 34” tube with a 68” tube, the new delay was calculated to be Ates» = 0.435 seconds, or 1.85 times longer. If the ratio of the distance from the valve to the sensor is calculated, the result is the same, varifying the sample velocity assumption. ea" 46" sarsem "5 Ree (4.2) ‘The time constant of the sensor and/or gas mixing process, was estimated by using a nonlinear estimation technique called the Levenberg-Marquardt method, which will be discussed in the dynamic fuel delivery calibration section of this chapter. The resulting time constant was estimated as r= 0.432 seconds. ‘This model was helpful in assessing the exhaust air/fuel ratio, but differed from the dynamic engine model exhaust air/fuel -102- Chapter 4: A Case Study in Engine Validation catio output by up to 20%. This difference is associated with unmodeled mixing of exhaust gases of differing air/fuel ratios in the exhaust manifold during transients. ‘The exhaust sample is taken from the exhaust manifold ruaner of the first right hand bank cylinder, about 4” from the cylinder head, as shown ia Figure 4.1. Whenever possible, engine parameters were recorded from sensors on the engine, rather than from calibrated estimates used by the electronic control module. There were two exception to this rule. Spark advance was recorded using a digital tc analog converter coupled through an interface module to the electronic control module. When the electronic control module specified the fuel pulse width (i.e., not controlled by the set point controller), then that pulse width was outputed in a similar manner. These varibles were output to a fast update rate analog port whose update rate is a function of the electronic control module parameters (up to 300 samples per second update). To measure this data delay, spark perturbation data was analyzed by comparing the point of spark change (from the converter) with shaft torque change (from the inline torque sensor). If we assume there is no delay from spark advance to torque, then the mean delay is calculated as 0.076 seconds (standard deviation of 0.0054 seconds). Since these spark steps are done at four different speeds (1200, 2000, 3000, 4000 rpm), the expected spark to torque delays can be added to these values, giving a mean D.A.C. delay of 0.083 seconds (standard deviation of 0.0033 seconds). Note that the coefficient of variation has decreased from 0.07 to 0.039 with the estimated engine delays, and the total delay of the digital to analog converter is estimated as 0.083 seconds. The output data is compensated by time shifting, to alleviate this measurement delay. Fuel perturbation tests to identify dynamic fuel parameters, and the air/fuel influ- ence function, require that the fuel injector pulse width be changed independent of all —203- Chapter 4: A Case Study in Engine Validation other engine parameters. To accomplish this task, a fuel injector set point controller, developed by the Instrumentation Department of General Motors Research Laborato- ries, was utilized. This controller uses its own injector drivers, and can vary injector pulse width throughout the operating range of the engine. In conjunction with this controller, an output circuit was constructed to give a voltage that is proportional to the crankshaft degrees that the injector is on, which is proportional to net fuel used by the engine. The change in this output voltage occurs when the first injector changes its pulse width, so this output is an accurate representation of the fuel command to the injectors. The start of the injector pulse, with respect to top dead center, was unchanged from the production engine. Only the duration was altered to avoid adding another input parameter. Spark advance perturbations were accomplished by means of a “heads-up” display/controller, which was connected to the electronic control module through an interface module. A voltage proportional to thespark advance was outputed through a digital to analog converter coupled to the electronic control module. The output shaft torque was measured by an inline torque transducer, which is shown in Figure 4.1 at the dynamometer end of the shaft. Dynamometer speed was outputed from the dynamometer hub by a 60 pulse per revolution signal. A 360 pulse per revolution encoder was attached by a short toothed belt to the damper on the front end of the crankshaft. During the engine tests there were problems getting a good signal from this transducer, so all speed measurements used in these analyses are based on the dynamometer speed measurement. -104— Chapter 4: A Case Study in Engine Validation In summary, the controls to this system are: throttle angle degrees spark advance degrees before T.D.C. u= 4 net fuel flow to engine grams/sec command to E.G.R. solenoids volts Idle Air Control count unitless and the outputs are: shaft torque N*m dynamometer speed rpm fuel injector pulse width msec fuel injector pulse width crankshaft degrees fuel flow to “make-up” box cc/sec air/fuel ratio of exhaust unitless mass air flow sensor g/sec (Hz from sensor) E.G.R. solenoids volts spark advance degrees before T.D.C. _ } intake manifold plenum pressure kPa ¥=) intake manifold runner pressure kPa air pressure measured btw. air cleaner and throttle kPa exhaust manifold pressure kPa oil pressure kPa fuel rail pressure kPa air temperature btw. air cleaner and throttle °C intake manifold plenum temperature °C exhaust manifold temperature °C oil temperature °C coolant temperature °c ‘The transducer used to measure intake manifold temperature, was a grounded 0.010" diameter thermocouple with a 0.039 second time constant. Data from engine tests was recorded in two ways, depending on if the data was to be used for dynamic or steady-state validation. A digitizer was available that had a maximum sample rate of 50 Hz. This was adequate for all of the steady-state data, such as flow measurements of the throttle body, or motoring friction measurements. However, for dynamic measurements of torque during spark or fuel perturbations, this sampling rate is wholly inadequate. To alleviate this problem, an F.M. tape recorder was used to effectively increase the sampling rate. Data was recorded on F.M. tape in -105— Chapter 4: A Case Study in Engine Validation real time. The data was later played back at a slower speed, and digitized at 50 Hz. Since the ratio of tape recording speed to playback speed was 8, the effective sampling rate of the dynamic data was 400 Hz. The frequency response of the recorder was much higher than any signals of interest, and analog filters in the sampling system prevented any aliasing when the signals were digitized. ‘ottle, 1 an w_Identificatio DiscHRRGE COEFFICIENT bbs a3 cc 4. 8 6 7 8 9 LO PRESSURE RATIO. (PH/PAY Figure 4.4 Throttle body discharge coefficient Parameter identification experiments for the throttle body were conducted on the engine using the mass flow sensor on the throttle body in conjunction with a calibration from B.O.C. (Buick, Oldsmobile, Cadillac Group of General Motors), and a throttle -106- Chapter 4: A Case Study in Engine Validation ISCHARGE COEF PRESSURE RAT aT 2 3 4@ 5 6 «7 8.9 PRESSURE RATIO’ (PH/PAD Figure 4.5 Pressure ratio discharge coefficient position potentiometer mounted on the throttle rod. The experiment involved setting the throttle angle and varying engine speed from 5000 to 1000 rpm. This experiment was repeated for throttle angles from 5° from closed to 75° (wide open throttle). The Idle Air Control valve was closed, using the “heads up” display, for these tests. The measurements of intake manifold and upstream stagnation pressures, stagnation tem- perature, throttle angle, and mass air flow can be used with equations (2.72) and (2.74) to calculate the discharge coefficient. This coefficient as a function of the pressure ratio is shown in Figure 4.4. Each group of points represents a different throttle setting. If this discharge coefficient is actually a product of two independent discharge coefficients (one a function of throttle geometry and the other a function of the pressure ratio), then these individual curves can be collapsed into one curve by adjusting for the throttle geometry term, Indeed this is the case as shown in Figure 4.5. The throttle geometry term (Cj) is obtained by calculating the quotient of the discharge coefficients of two -107- Chapter 4: A Case Study in Engine Validation adjacent throttle setting curves where they overlap with respect to the pressure ratio. The functionally related discharge coefficients and the parts of the one-dimensional compressible flow equation can be grouped together (ie. Ca: - A(9)) and normalized to produce the throttle characteristic and pressure ratio influence which are shown in Figures 4.6 and 4.7. The derivation is given previously in equations (2.9), (2.10), and (2.11). The final calculation of the maximum air flow through the throttle is given as: Pa MA= 5157 (43) THROTTLE CHARACTERISTIC (TC) a 10 20 30 40 30 60 70 20 THROTTLE ANGLE (OEGREES FROM CLASED) Figure 4.6 Throttle characteristic (7C) For the experiments P, and 7, were given as 100 kPa and 298 Kelvin. The air flow rate into the throttle body (which supplies all of the cylinders) is measured by an air flow sensor. The output of this sensor is a voltage frequency which -108— Chapter 4: A Case Study in Engine Validation PRESSURE RATIB INFLUENCE (PRI o 0. 2 2 @ 5 6 .? 8 .9 1.0 PRESSURE RATIO. (PH/PAD Figure 4.7 Pressure ratio influence (PR/) is a function of the mass air flow rate. A calibration for this air flow sensor was supplied by the B.O.C. Group of General Motors [57]. This calibration is known to be sensitive to the intake system upstream of the sensor, so a stock air cleaner housing and element was used throughout the tests. 4.3.2_Idle Air Control The choked flow characteristic for the idle air control appeared to be quite close to the theoretical values, so ideal choking is assumed (i.e. Cx: = 1 for the I.A.C. model). ‘The parameter identification experiment consisted of defining the flow as a function of the ILA.C. count. The opening of the A.C. valve is a function of the digital counts to the I.A.C. motor. In the experiment, a pressure ratio of 0.45 was used throughout. ‘Three different air flow ranges were examined by first setting the throttle, The pressure ratio was then maintained by changing engine speed as the I.A.C. count was changed. -109- a WAS AIR FLOM (G/SEC) 8 Chapter 4: A Case Study in Engine Validation estimation of the data. ad = 2 4 6 0 100 120 10 160 160 TALC, CauNT Figure 4.8 I.A.C. calibration data ‘This data is shown in Figure 4.8. The mass air flow shown in this figure is the total flow of the I.A.C. and throttle. The throttle flow can be removed from the data by letting zero I.A.C. count represent zero mass air flow through the I.A.C. valve. This adjusted data is shown below in Figure 4.9 along with the least squares polynomial 4.3.3 Exhaust Gas Recirculation In March of 1987 several E.G.R. tests were conducted at the Power Systems Re- search Department of General Motors Research Laboratories. From this data two important engine characteristics can be determined: (i) validation of the digital E.G.R. valve flow calibration, and -110- Chapter 4: A Case Study in Engine Validation TLALC. MASS AIR FLOM (G/SEC) 0 a a a a TALC. CBUNT Figure 4.9 A.C. calibration (ii) determining the effect of E.G.R. on the MBT spark timing. ‘The E.G.R. valve which is used on the engine consists of three different sized calibrated orifices, which can be opened by solenoids in any combination to control the E.G.R. flow into the intake manifold. E.G.R. is used at part load engine operation as a diluent to lower combustion temperatures in the cylinder. Lower combustion temperatures decrease the amount of NOx emissions, so E.G.R. is used to help meet the emission requirements for the engine. There are two types of E.G.R. valves used on the 3800 engine. These valves are designated as 1.3 Ib/min or 0.9 Ib/min valves, which indicate the approximate maximum flow (with all orifices open). An orifice calibration was supplied by the Buick-Oldsmobile-Cadillac Division for the two types of E.G.R. valves [31]. These calibrations are performed at room temperature with theoretical chocked flow (pressure ratio of 0.5), and indicate the flow (in pounds per minute) for each of -111- Chapter 4: A Case Study in Engine Validation the seven combinations of open orifices. Minimum air, seal leakage, and pull-in voltage were also provided. The valve used during the E.G.R. tests is a very early version of the 0.9 Ib/min valve. While looking at the data from these tests, it appeared that the actual mass flow of E.G.R. was somewhat lower than the values from the calibration. It has been shown previously that the molecular weights of air and E.G.R. are approximately equal (equations (2.20) and (2.21). Since mass air flow and intake manifold pressure are measured during the E.G.R. experiments, any change in mass air flow plus mass E.G.R. flow will show up as a change in manifold pressure, given a constant engine speed. Because molecular weights are equal, the intake manifold pressure will change the same amount whether air flow or E.G.R. flow is changed (it does not know the difference). The relationship between flow and manifold pressure can be found using previous air flow data, and compared with theoretical values. Consider the speed-density calculation for air flow into the engine: (44) we mou. (45) During each of the E.G.R. experiments the intake manifold temperature remained essentially constant, and can be estimated as Tw300 K. The speed-de: jity equation (4.4) can be differentiated with respect to manifold pressure, assuming a constant speed, to give the gradient of mass air flow with respect to intake manifold pressure, or: (4) Chapter 4: A Case Study in Engine Validation From previous flow data it can be shown that *sf4e| is essentially a constant, he=constane which implies that: This constant can be estimated as: any nha S| ae ta The following table shows a comparison between the theoretical values of 2MAF ly cone. and the actual slope calculated from mass air flow data. Table 4.1 Air Flow Gradient Data N (rpm) 1200 2000 3000 4000 MAE equa. (5.85) [0.397 0.662 0.993 1.324 In=cona. Hts from data 0.3675 0.6525 0.975 1.34 ln=conat. ‘The estimate of: aMAF Te Mmuoaa 7 08M (49) appears to give very good results and can now be used to estimate the mass E.G.R. flow, given the equality of molecular weights. OMAF _ alna + thege) Pm Pm fas +1 Pa yr — Frege (4.10) This equation can be rearranged to give an estimate of the change of mass E.G.R. flow as: tga ~ tees = SHAE (= Py) ~ (trea = tras). (4.11) -113- Chapter 4: A Case Study in Engine Validation Now, the actual E.G.R. mass flow rate is assumed to be some unknown value times the calibration value, or: Fea laceut (4.12) Combining equations (4.11) and (4.12) gives: a - OMAF . 12 Fhegea = 14 Tne = SEAR. (P ~ Py) ~ (Gea ~ (4.19) where thrge2 and riggrs are calculated from the B.O.C. calibration. If 7 is assumed to be aconstant, then this expression can be used to solve for the multiplicative error 7 as: SHAE «(Py ~ Pi) - ea — ts) [Freora = ears oa, . (4.4) 1 was calculated from the E.G.R. data and, while there was some scatter in the data, there were no apparent patterns to the variations. Rather, it looks like experimental or measurement variability. From this analysis 7 was found to have a mean of 0.75 with a standard deviation of 0.13. A better understanding of why the actual E.G.R. flow values appear to be about 75% of the calibration values can be acquired by looking at the one-dimensional compressible flow equation for choked flow (equation (2.3)). During calibration of the E.G.R. valve, stagnation temperature and pressure are given as 300 K and 94.6 kPa, respectively. Relating this equation to the multiplicative error 7 gives: tala Terk F = we Saat =0,18127- He (4.15) -114- Chapter 4: A Case Study in Engine Validation While exhaust temperature at the exhaust manifold was monitored during the ex- periment, this parameter was not recorded for each case. Exhaust temperature was monitored in order to avoid causing any damage to the pistons or rings with excessive heat. Two points were recorded which can be used in the previous equation. One low flow point gave an exhaust pressure of ~100 kPa and an exhaust temperature of ~573 K. A higher flow value yielded an exhaust pressure of 115 kPa, and an exhaust tempera- ture of 973 K. Substituting these points into equation (4.15) yields values of 7=0.75 and 7=0.69 respectively. For the second point (higher flow), there would be considerable cooling of the exhaust gases from the exhaust manifold (where the exhaust temperature is measured) to the E.G.R. valve (stagnation temperature T.), which would raise the value of 7 closer to the 0.75 point calculated from the data. In summary, higher exhaust pressures and temperatures lower the actual E.G.R. flow values from those given in the calibration. If exhaust pressure and temperature near the E.G.R. valve are known, a fairly accurate estimate of the E.G.R. flow can be made using the flow equation (2.76). In lieu of these measurements a value of 7=0.75 can be used as a reasonable estimate, and E.G.R. flow can be calculated using equation (4.12). 4.4 Fuel Delivery 4.4.1_Fuel Injector Calibration Early in the planning phase of the validation tests, the need to measure the net fucl flow to the engine was identified. It was not feasible to have the fuel flow meter, that measures the fuel flow to the “make-up” box, do this task. The limit cycle of the float valve assembly introduces significant errors in this measurement. Instead, an averaged net fuel flow (measured with the flow meter) was determined as a function of -115- Chapter 4: A Case Study in Engine Validation the pulse width of the injectors, and the engine speed. By averaging the flow over a long period of time, the errors introduced by the limit cycle are integrated out. During these tests, 41 different speed and pulse width points were run, with pulse widths from 2 to 18 msec, and engine speeds from 1200 to 5000 rpm. Fuel rail pressures and intake manifold pressures were monitored during the tests to see if the pressure differential across the injector was constant, and to see if any pressure differential change affected the flow of fuel through the injector. As it turned out, there was a decrease in the pressure differential across the injector at high pulse widths, of up to 25% (from 2 to 18 msec). The surprising result is that this pressure drop had a negligible effect on the fuel flow through the injector. In theory, the net fuel flow to the engine should be proportional to the number of crankshaft degrees rotation that the injector is cn, or: oe a W(t) (Sei) = constant -crankshaft degrees of injection. (4.16) ‘At a constant crankshaft angle of injection, the pressure differential across the injector varied up to a maximum of 21%, but the variation in net fuel flow to the engine between these two points was less than 1%. Figure 4.10 shows the final calibration of net fuel flow to the engine as a function of injector pulse width in crankshaft degrees. Fuel flow decreases slightly at very short pulse widths, due to inertia effects of the fuel, otherwise the plot is nearly linear. Injector calibration is given by: rng = -0.13263 - 10-4 - DEGON? +0.02199. DEGON - 0.36, (4.17) where DEGON is the injector pulse width in crankshaft degrees. With this calibration, a net fuel flow to the engine can be inferred by measuring the injector pulse width and the engine speed. -116- Chapter 4: A Case Study in Engine Validation oso 100 150 200 250 200 390 400 450 500 INJECTOR PULSE WIDTH (CRANK DEGREES? Figure 4.10 Fuel Injector Calibration 4.4.2 Dynamic Fuel Parameter Identification There has been considerable interest in modeling fuel transport characteristics in carbureted and throttle body injected engines [11,58,14,59,60,61]. The reason for this interest is that a portion of the fuel forms a liquid fuel film on the walls of the intake manifold, especially with a cold manifold, This film slows the transport c* fuel to the intake valves, and significantly alters the air/fuel ratio ingested into the cylinders, and consequently, engine brake torque. If a fairly accurate model of this process could be developed, then a compensation scheme could also be developed to maintain a desired air/fuel ratio in the cylinders. An accelerator pump, or acceleration enrichment, is used on these engines. The introduction of port fuel injected engines alleviated the need for the fuel to traverse the intake manifold. Instead, fuel is injected directly into the intake port, upstream of the intake valve. Since the injectors are sized as small as possible, for -11T- Chapter 4: A Case Study in Engine Validation good fuel control, injector pulse widths at high fueling rates can be equivalent to nearly two revolutions of the crankshaft. Thus, fuel is not always injected when the intake valve is open, but can sit in the inlet port for a period before the intake valve opens. Fortunately, this inlet port is quite warm, so vaporization does occur, but there is still evidence that a fuel lag exists {13] even with port fuel injected engines. The problem with this evidence is that the models used to estimate these lags and delays are not always clearly defined, and often one does not know which physical delays are included as part of the final lag value. In this section the author will use the fuel delivery model of Chapter 2, equations (2.80) through (2.84), and estimate the fuel split parameter «, and the slow fuel lag time constant r. Note that the two revolittion delay in fuel, for fuel injected after the intake valve closes, is included in the model, and is not part of the slow fuel lag time constant. Sequential fuel injection is used, with a constant start of injection point with respect to top dead center. ‘The nonlinear estimation technique used in this section (and in the exhaust air/fuel sensor estimation) is the Levenberg-Marquardt method [62]. This is an iterative method that tries to minimize a x? merit function which is defined by the input data, model, and output data. The Levenberg-Marquardt method varies smoothly between the steepest descent method (used when y? is far from a minimum), and the Inverse Hessian method (used near the minimum). Consider a function f(z) which can be approximated by its Taylor series expan- sion as: oe f(a) = 1e)+ 34 Bayt (4.18) Chapter 4: A Case Study in Engine Validation we eat bat Ag, (4.19) where, = 42), (4.20) ~VWee (4.21) (4.2) The matrix [4], is called the Hessian matrix of f at B. We can solve for the gradient of f, ast vie Az-h (423) so if we want to find an extremum (or V/ =0), we simply solve: Aaah (4.24) by inverting the Hessian matrix A. This is accomplished by using Gauss-Jordan Elim- ination. This method can be applied to minimizing the x? function by assuming that our model outputs (y) are a function of the model inputs (u), and a parameter vector (9) or: y= y(usé)- (4.25) ‘The x? function is defined as: _ Sw - vung]? v= re] : (426) Near the minimum, the x? function is approximated by the quadratic portion of the Taylor series expansion, or: MQ MIE AG (427) -119- Chapter 4: A Case Study in Engine Validation If the approximation is correct, we can find the minimum of x? by solving: Ag=b (4.28) by Gauss-Jordan elimination. To accomplish this we need to find the Gradient and the Hessian. These can be calculated by taking partial derivatives of the x? function, or: 8 og Folmcasel ad EG = ee (4.29) and, Aulus 8) aylu8) ari ee a and k=1,2,.4M, (4.30) where N is the number of data points in the data set, and M is the number of parameters to be fitted. The second derivative term in equation (4.30) can be destabilizing if the data is contaminated by outlier points, or if the model fits the data poorly. So the Hessian will be approximated by: ax? aed [ees eeu) 568 oF | Oey Oe s2,0yM and k=1,2,.. (4.31) If we let aj. = 4- $24, and A, =-4- 9, then the solution oft aM Vay 664 = Bi, (4.32) yields the increment in the parameters of 6¢. The steepest descent method simply translates into: 56; = constant -B;. (4.38) -120- Chapter 4: A Case Study in Engine Validation ‘The Marquardt method first sets the constant of the Steepest Descent method as the al of the diagonal Hessian element, or: Sem Be (4.94) where \ is some nondimensional scaling factor to adjust the step size. This method also combines the Steepest Descent and Inverse Hessian methods by defining: 5 a if; oie eae) ithe: (4.35) Substituting equation (4.35) into equation (4.32) yields: a Vale 66 = 8 (4.36) a As A gets large, equation (4.36) approaches equation (4.34), and as A goes to zero, equation (4.36) approaches equation (4.32), thus combining the two methods. To solve the validation problem, the partial derivatives of the model output, with respect to each of the parameters to be identified, must be calculated. This requires taking these partials within the time domain. The model will utilize equation (2.106), with the assumption that dynamometer speed is equal to the engine,t or: (437) The test data consists of several different fuel perturbation tests, each done at a constant air flow, and a nearly constant speed (i.e. dynamometer in speed control mode). This type of data set significantly simplifies the model. Since the mass air flow, torque function, and the spark influence are constant, the indicated torque model reduces to: AFi| constant temas (4.38) t This aseumption, and the engine-dynamometer model, are discussed near the end of this chapter. -121- Chapter 4: A Case Study in Engine Validation where: constant = 20.0 thay TFSI, (4.39) and the intake torque delay is given as: At = (4.40) Since friction/pumping torque is modeled as a function of mass air flow and engine speed, it is also a constant for each data set. The inputs to the model include injector pulse width in crankshaft degrees, dynamometer speed, and current time. The output is the shaft torque. Time shifting the data to eliminate the delays, and diffentiating the shaft torque with respect to the fuel split parameter, yield: (4.41) 140 120 a8 8 AVE AND BRAKE TOROUE (Net) Bo 8 8 8 0 2 4 6 6 10 i 10 60 60 TIME (SEC) Figure 4.11 Typical Fuel Perturbation Test -122- Chapter 4: A Case Study in Engine Validation A typical data test sequence is shown in Figure 4.11. This data set can be analyzed to yield an air/fuel influence function by looking at the torque drop as a function of the air/fuel ratio. An estimate of the partial derivative of aiz/fuel influence with respect to the air/fuel ratio can be made by calculating the average of the finite difference of AFI on each side of the point of interest, or: aaFl| Sar (4.42) IR/FUEL INFLUENCE (RFI) AND DERIVATIVE 2 4 6 @ 10 12 14 16 18 2 2 28 26 2 30 AIR/FUEL RATIO Figure 4.12 Air/fuel Influence and Derivative This data can be seen in Figure 4.12. Since the air/fuel ratio is estimated as the ratio of air flow to fuel flow out of the intake manifold, then the rate of change of the air/fuel ratio with respect to the fuel flow out is given by: (4.43) Chapter 4: A Case Study in Engine Validation Finally, the rate of change of the fuel out of the intake manifold, with respect to the fuel split parameter, must be calculated. From equation (2.80) we see that: dre re Sriyya , Dringys , Orpen (44) le ae ae Since ¢ is assumed to be constant within each data set: Orn, Be lac an-on Maca (sas) and: Bring ys ; a ae “ To solve for 2%4, first solve for rny,. from equation (2.83), or: x(t) = eAlee-tad «2(te) +f eAle-n Be u(s) db, (447) where: 4 (4.48) emt (4.48) a(t) = thyer, (4.49) u(t) = Yel any” (4.50) Consider the discrete sequence modeled by a zero order hold, then: Ms siege re , =H t+ () ty, (4.51) since u is a constant between samples for a zero order hold. The integral can now be evaluated, yielding: zy t(1-¢> [i - | (te). (4.52) Fag =e -124- Chapter 4: A Case Study in Engine Validation This expression can now be differentiated with respect to ¢: (4.53) or: Brngar| —_ stat Sng ae ae (4.54) ~ [ire where delt is the current time step of the data. Ina similar manner, the rate of change of the shaft torque with respect to the slow fuel time constant can be calculated as: an, (4.55) where 245 is computed from the air/fuel influence data, 24/ is given in equation (4.43), and: = OR (4.56) is solved as is shown previously for «. This type of analysis, using the zero order hold approximation, yields: (1-4 trae i}: (4.57) Equations (4.41) through (4.46), (4.54), (4.55), and (4.57) can be combined to cal- culate the Gradient and Hessian matrices of x, and nonlinear estimation of the fuel parameters can continue. Figure 4.13 shows a typical plot of fuel pulse width (in crankshaft degrees), measured shaft torque, and estimated shaft torque using the parameters from nonlinear estimation procedures. Shaft torque is left unfiltered to show what the estimation -125- Chapter 4: A Case Study in Engine Validation 200 120 60 40 % wer ‘SHAFT TOROUE (WM) AND PH (OECREES) -100 aan eI o . 2 38 @ § 6 «7 8 9 10 TIME: (SEC) Figure 4.13 Typical Estimation Results procedure is trying to fit. Oscillations in measured shaft torque are the result of the engine and dynamometer oscillating with respect to each other, due to the compliant shaft connecting them together (x 35 Hz). The validation results are summarized in Table 4.2. Table 4.2 Dynamic Fuel Validaton Results - N=1200 rpm, MAF=18 g/s A/E « . Large AT,? _ Comments 86 0.22 0.50 Yes to large o os 0.0 0.078 Yes from large o MouT (0.48 0.19 Yes from large o 147-18 0.56 19 No small 8-147 0.85 14 No small o M122 (0.64 0.45 Yes to large o -126- 221.7 14.7 +8 0.42 0.85 Chapter 4: A Case Study in Engine Validation 0.083 ‘Yes from large o 12 No small o Table 4.3 Dynamic Fuel Validaton Results - N=2000 rpm, MAF =32 g/s 36 68 3-10 asa 14.722 2214.7 78 229 14.7 0.55 0.0 0.9 0.58 0.60 0.52 0.33 0.50 0.57 ‘Yes to large o 0.10 Yes from large © 0.087 No small ¢ 0.51 No small ¢ 0.24 ‘Yes to large o 0.19 Yes from large o 0.98 No small ¢ 0.072 ‘Yes from large o Table 4.4 Dynamic Fuel Validaton Results - N=3000 rpm, MAF=47 g/s 6-8 1618 18 + 20 22424 24147 14.748 14.722 2214.7 147-8 6.047 0.39 0.35 0.25 0.21 0.67 0.56 0.13 0.66 0.060 Yes from large o 0.24 ~Yes ~to large o 0.23 ~Yes ~to large o 0.059 No large @ 0.13, Yes from large 23 No small 0.44 Yes to large o 0.099 Yes from large 29 No small Table 4.5 Dynamic Fuel Validaton Results - N=4000 rpm, MAF =55 g/s -127- Chapter 4: A Case Study in Engine Validation 10-8 0.38 3.1 No small 22-24 0.31 0.080 ~Yes large ¢ Mout 0.44 0.16 Yes from large @ 7 0.62 19 No small o 8147 0.10 0.23 No from large @ M742 © 0.61 0.31 Yes to large o mat 0.0 0.071 Yes from large o 47-8 0.65 25 No small ¢ Although there tends to be a large standard deviation in the data, two observations can be made: + where + is small, ¢ is not a good measurement, and + where cis large and A7, is small, r is not a good measurement. The first observation comes from equation (4.54). Taking the limit of “+ as + ap- proaches zero, yields: Brngo _ : Orgs dingo = Pe (Lt) + He + Jims “SE = tinge (L= 1) + tye — tye =0. (4.58) ‘This means that for small values of r, changes in « have little effect on the shaft torque, and these «’s are not good measures of the actual parameter. The second observation comes from equation (4.52). Frpetigg, OH stingy, HL 9) f = seas: (4.50) -128- Chapter 4: A Case Study in Engine Validation Where « gets large (ie. approaches 1), then rnysx grows very small over time, and tiny. is the only fuel branch that + affects. So, with « large, r has little effect on the shaft torque. When small torque changes occur with a large «, the probl’ m is compounded. To evaluate «, only points where r > 0.19 were used. This gave a mean and standard deviation of: n= El (4.60) a= VE ()P} = 0.19. (4.61) Similarly, for large AT, and where « < 0.7, the mean and standard deviation of the time constant r are: seconds, (4.62) 2, = YE((r—E(x))?} =0.17 seconds, (4.63) ‘The construction of the model attributes these transient torque lags exclusively to slow fuel lags, where they may have some contribution from other processes, such as heat transfer transients or other physical processes. This is why some of the time con- stant estimates, especially for small torque changes, can be quite long when compared to the mean. However, the estimated time constant does describe the phenomemologi- cal transients of torque as a function of changes in fuel flow, which is what the control engineer requires. 4.5 Torque Production -129- Chapter 4: A Case Study in Engine Validation 4.5.1 Spark Influence In the engine literature, there is evidence that the relationship between spark advance from MBT (minimum best torque) and brake mean effective pressure, or brake torque, is a parabola [41,52]. To check this assumption and to identify the parame- ters which define the relationship, an experiment was conducted on the engine. This experiment consisted of first setting an engine speed and mass air flow rate. At this condition the electronic control module would set the spark advance. This spark ad- vance setting was manually retarded in five degree increments until zero spark advance was achieved (i.e. spark at top dead center). After each increment the spark advance was returned to the initial setting, so a typical test spark advance sequence might be: 40°, 35°, 40°, 30°, 40°, 25°, 40° Figure 4.14 shows a plot of spark advance from top dead center and engine brake torque versus time for one of these experiments (4000 rpm, 55 g/sec air flow). The first step in developing a spark advance influence function. from this data is to remove transitional data points and divide the data into spark advance sets. The mean, standard deviation, and higher moments can be calculated to determine the distribution of the data within each data set. Finally, a least squares polynomial fit was made to the means in order to determine polynomial coefficients for each speed/air flow experiment. This process was then repeated for eleven speed /air flow data groups. These polynomial fits to the spark data were quite good, with a mean index of determination of 0.992 and a standard deviation of 0.008075. ‘The equation of the parabola can be represented as: (z- A)? =4-p-(y—h), (4.64) -130-- Chapter 4: A Case Study in Engine Validation IMs ent utah te tly fal why iH ' SPARK AOVRANCE (DBTOC) AND TOROUE (Net) Cr eT ) TIME (SECI Figure 4.14 Spark advance test sequence (spark and torque) where (h,) are the coordinates of the vertex, and |4-p] is the length of the latus rectum, or P is the directed distance from vertex to focus. In terms of spark influence function; h=MBT spark advance, (4.65) k= torque at MBT, (4.66) and p defines the shape of the spark influence function. Equation (4.64) can be expanded into polynomial form to give: -() (Be) eo Using the results from the least squares polynomial fit, values of p, h, and & can be calculated for each speed/air flow data set. Equation (4.67) can be normalized by -131- Chapter 4: A Case Study in Engine Validation dividing by k, or the MBT torque, to give: (a i) #+ ~(ca) #3) =H) Wn This equation now represents a normalized spark influence function with respect to brake torque. Plots of the spark influence curves show that they vary somewhat with respect to brake torque. Low air flow data represents the lowest curves on this plot, which is quite reasonable. Low air flow data represents the lowest brake torque values. At these torque levels the friction and pumping torque are a greater percentage of the total torque, or the brake torque is more nearly zero. Any change in brake torque at the low torque levels would produce a greater change in a normalized torque function than at a higher brake torque. 1.0 NORMALIZED SPARK INFLUERCE s 0 IS 2 22 9% 3 4 a 50 SA-YOT (CRANKSHAFT DEGREES) Figure 4.15 Spark influence based on brake torque -182- Chapter 4: A Case Study in Engine Validation 1.0 2 8 “4 NORHAL.IZED SPARK. INFLUENCE 5s 0 IS 0 2 90 3 4 45 50 SALYGT (CRANSHRFT DEGREES) Figure 4.16 Spark influence based on indicated torque ‘To remove this influence from the data, it makes sense to look at the normalized spark influence with respect to indicated torque. This would give a non-negative torque influence, and more fairly compare high and low air flow (and torque) levels. Since friction and pumping torque are not known exactly at this point (it will be discussed further in this chapter), measured motoring torque is assumed to be identical to firing friction/pumping torque. The above calculations for k, h, and p are now repeated, except that the ordinate in the non-normalized torque function is an assumed indicated torque calculated from brake torque and motoring friction/pumping torques. The figure below shows the plots of the eleven speed /air flow experiments. Note that, as expected, the plots fall into a more uniform pattern than with the brake torque, because the spark advance changes the cylinder pressure-slider crank geometry relationship which causes changes in the indicated torque. These curves can be modeled as a single spark influence function, and all points would fall within one standard deviation of -133- Chapter 4: A Case Study in Engine Validation the original curves. An estimation of the MBT spark advance can be made from this analysis. Since no E.G.R. is used in this experiment, and the air/fuel ratio is held constant at stoichiometry, MBT spark timing is most strongly a function of charge density in the cylinder and engine speed. For lower charge densities in the cylinder, laminar flame speed is decreased, s0 MBT spark timing should be more advanced. At higher engine speeds the spark timing must be advanced to achieve maximum torque, since there is less time to burn the mixture. This effect is offset somewhat because of the higher turbulence in the cylinder, which increases the laminar flame speed. Overall MBT spark timing should increase slightly with increased engine speed. Except for the very lowest speed and air flow point, all of the above trends are reflected in the data. At the slow engine speed and very low charge density point there is very little turbulence in the cylinder, and it takes a little longer than normal to get the flame front developed. This shows up in the data as a slightly more advanced MBT spark timing than would be expected looking at the rest of the data. Figure 4.17 shows this MBT data along with the model. 5.2 MB’ Earlier in this chapter, equations for the spark influence contour were estab- lished (equations (4.67) and (4.68)). In that section a least-squares second order polyno- mial was fit to the spark perturbation data, and the MBT point (A) could be estimated from that analysis. Using a similar polynomial fit to the spark/E.G.R. data produced results which were not consistent with the earlier data, and the results were not consis- tent within each data set. The problem with this approach, when E.G.R. is added, is a result of the spark advance range covered by the data. When no E.G.R. was added, spark advance ranged from near MBT to approximately zero spark advance. Similar -134- Chapter 4: A Case Study in Engine Validation 8 BT SPARK AOVANCE (08TOCY a 8 8 8 01S 20.25 3088 40S S~SO SS ED MASS OF AIR PER CYLIMOER (G) Figure 4.17 MBT data and model numerical values of spark advance were used for the E.G.R. test. However, MBT spark timing has advanced as a result, so E.G.R. test data is now farther from MBT than it was when no E.G.R. was added. When this data is fit to a second order polynomial, the data is much more linear than when no E.G.R. was added. Consequently, small errors or deviations in the E.G.R. torque data can result in large variations of h (MBT). These variations result in inconsistencies in the MBT results. One way to get around this problem is to look at the horizontal (spark advance) shift of the spark influence function as more E.G.R. is added to the intake manifold. This method is most accurate where the slope of the spark influence function is steepest. A least-squares fit of this is a good indicator of the change of MBT as a result of added E.G.R.. Figure 4.18 shows the change of MBT (i.e. crankshaft degrees of spark -135- Chapter 4: A Case Study in Engine Validation advance) as a function of the percent E.G.R. which is given as: REG.R. = — Fra + Fear 100.0. (4.69) CHANGE IN MOT (DEGREES) 1 2 14 $6 18 20 2 PERCENT €.6.R. Figure 4.18 MBT Change as a Function of % E.G.R. ‘This data appears to be consistent with itself, since the change of MBT with respect to a change in percent E.G.R. is nearly constant. However, some of the data does not extrapolate to the origin. The two experiments from which this data is extracted (ie. MBT without E.G.R, and MBT with E.G.R.) were conducted several months apart, and many hours of engine run time were logged between them. Some variability is to be expected over such a span, and this might explain the small variance in the extrapolated data. Investigations by Rhodes and Keck into the effect of the addition of E.G.R. on laminar burning speeds in indolene-air-diluent mixtures [63] shows that ~136- Chapter 4: A Case Study in Engine Validation change in laminar burn speeds is a nearly linear function of percent E.G.R., as the above data suggests. Furthermore, the work of Rhodes and Keck shows that this near linear relationship is independent of pressure at the time of combustion, and of the air/fuel ratio of the mixture. This result suggests that the mean gradient from the above E.G.R. analysis could be multiplied by the percent E.G.R. and added to the previous (non-E.G.R.) functional relationship for MBT to now include the effect of E.G.R.. The gradient mean and standard deviation are given by 0.58 Se and 0.28 sSESE, respectively. This yields an MBT of: MBT = 88. MAC? — 99.59 MAC +4761 +0.003-.N +58 = (4.70) which is complete except for the air/fuel ratio effects. 4.5.3 Torque Function In the torque production portion of the engine model, the maximum possible MBT indicated torque is modeled as a nearly constant torque function times the mass of air per cylinder. This function can now be evaluated since the air flow is measured, fuel flow is known through a calibration, air/fuel influence on torque can be estimated from fuel-air analysis (33), spark influence is known (spark advance is very close to MBT in these tests), and friction/pumping torque is assumed to be the same as motoring torque. Also, an air/fuel ratio sensor using exhaust gases is used to verify the air/fuel ratio calculated from mass flow rates. With air flow and fuel flow calibrations, air/fuel ratio and mass of air per cylinder can be calculated. All of the measurements are now available to estimate torque function TF as: = et Ty “ ST-AFT- MAC?’ ~137- 1F (4.71) where, 3 TOROUE FUNCTION ESTIMATE (N=H/G) 8 8 20 Chapter 4: A Case Study in Engine Validation 20.0: MAP MAC ¥ 50. 70 INTAME HANIF@LD PRESSURE (KPA) 20 Figure 4.19 Estimation of Torque Function (4.72) The figure below shows calculated torque function, from a data set, as a function of manifold pressure. This data suggests that torque function is independent of engine speed, and that engine efficiency decreases with increasing manifold pressure. Since brake torque is roughly proportional to manifold pressure in steady-state operation, this data suggests that the engine efficiency at high torque levels is about 80% of the efficiency at low torque levels. The expected result from the torque function calculation is that the torque function remain nearly constant, and possibly increases slightly with increased engine speed due to the lower percentage of heat transfer at these higher speeds, The issue here is to analyze this calculated result and compare it to results in -138- Chapter 4: A Case Study in Engine Validation the engine literature to either confirm the previously stated results, or find out where errors lie and correct them. An effective way of analyzing engine performance is by calculating the fuel conversion efficiency, which is given by: (4.73) This efficiency is the work per cycle (W.) produced by the engine divided by the amount of fuel in each cylinder (my) times the lower heating value (Qn.) of the fuel. This measure of efficiency indicates how much work (power) is produced by the engine for the amount (rate) of energy that is put into the engine as fuel. W, in equation (4.73) can represent several different quantities. If W. represents the engine output work per cycle, then ny is the brake fuel conversion efficiency. If W, is the net indicated work per cycle (the area of the compression-expansion loop minus the area of the exhaust-intake loop of the P-V diagram) then n, is the net indicated fuel conversion efficiency or nya. Similarly, if W, is represented only by the area of the compression-expansion loop, then ny is the gross indicated fuel conversion efficiency or ny,. The fuel conversion efficiency is related to thermal efficiency by combustion efficiency as: ny = Nees (4.74) For lean mixtures the combustion efficiency is very close to unity (3], so for these mixtures: mM (4.75) In the ideal Otto cycle where compression and expansion occur isentropically, and com- bustion and blow-down are constant volume processes, thermal efficiency is a function -139- Chapter 4: A Case Study in Engine Validation of the compression ratio and the ratio of specific heats. When an indicator diagram is plotted with log-log axes, the polytropic exponent that tends to fit the data well is 11.3725. Using this exponent along with a compression ratio of 8.5 gives: = 54.9%. (4.76) ‘This value can be used as an upper limit of fuel conversion efficiency for lean mixtures, although the actual value will be somewhat lower. A more limiting upper bound may be obtained by looking at fuel-air cycle data computed by Edson and Taylor [33] for iso-octene. For a stoichiometric mixture and a compression ratio of 8.5, this analysis yields a fuel conversion efficiency of 43.3%. Again, this may be considered an upper limit of efficiency. What needs to be defined is the relationship between gross indicated fuel conversion efficiencies, given above, and the torque function of the engine model. From the engine model, torque function can be calculated as: Tor + Trp Timer TP = STAPF: MAG ~ MAC: AFT’ (4.77 where Tiler represents gross indicated MBT torque (81 = 1). The power produced by the engine is given by: P=2* tor -N, (4.78) 60 where 1 is given in revolutions per minute. With respect to fuel conversion efficiency, torque is assumed to be MBT torque. The mass fuel flow rate can be represented as: rae (4.79) Using equations (4.73), (4.78), and (4.79), gross indicated fuel conversion efficiency is given by: x Tilwer-N-$ ye a (4.80) -140- Chapter 4: A Case Study in Engine Validation Now, since the mass of air per cylinder is given by equation (4.72), fuel conversion efficiency is represented in terms of mass of air per cylinder by: x Timer: $ (4.81) UF One MAC” Combining equations (4.77) and (4.81) gives an expression for torque function in terms of fuel conversion efficiency as: = 8 Qne my Te ay (4.82) Air/fuel influence (AF/) is defined as the normalized effect that the air/fuel ratio has on MBT gross indicated torque for a constant mass of air per cylinder. Rearranging equation (4.81) gives: 3: Qe ny MAC Tilor = SSe 4 MAC (4.83) Dep If the mass of air per cylinder is held constant, the gross indicated MBT torque is a function of rs The air/fuel influence is consequently given as: AFI = ac (4.84) #) mae The quantity 4f, as a function of the air/fuel ratio, is very flat at its maximum, and obtaining an accurate value of the air/fuel ratio at the maximum is rather difficult. The task can be simplified by using the expression: AFI= Br (AF Nstoic.- (4.85) In this expression (A/F), is known, and (ny)stoic. and (AF) oe, can be easily evaluated from the fuel-air analysis data [33]. Combining this expression with equation (4.82) -141- Chapter 4: A Case Study in Engine Validation gives the final relationship between torque function and gross indicated fuel conversion efficiency: (nus Faas (AF lina” (20) If the lower heating value of the fuel is estimated as 44x10° 44, then: 7 1439+] (4.87) Previous calculations from the fuel-air cycle analysis gave a stoichiometric fuel conver- sion efficiency of 43.3%, which would yield an upper bound on torque function of 623 Comparing these results with the results in Figure 4.19 reveals two problems with results previously calculated from the data. The magnitude of the torque function goes above the theoretical fuel-air analysis limit of 623 4 (which is actually higher than would be expected in practice, since heat transfer is neglected), and since ny is nearly constant, equation (4.87) would indicate that torque function should be nearly constant. The main problem lies in the friction/pumping torque model. Motoring friction torque cannot be assumed to be the same as firing friction torque. Mass air and mass fuel flow calculations tend to agree with exhaust air/fuel measurements, and engine maximum torque levels are as expected for this engine, which lends confidence to these measurements. Motoring friction also agrees with data from B.O.C. (64), but, the assumption that motoring and firing friction/pumping torques are identical is not valic The motoring friction/pumping torque used in the previously described model is the steady-state torque that is required to maintain a given speed with a specified -142- Chapter 4: A Case Study in Engine Validation ae ° 36 cee ate z 2 60 ds pe 250 olka is 240 Sea : 2. 00 "Faas 2 25 S g one Bs s000r—m 2 10 2 INTE WRNIFOLD PRESSURE (KPRY Figure 4.20 Motoring friction/pumping torque data intake manifold pressure. Pumping torque is varied by changing the throttle angle which, in turn, varies manifold pressure. This friction /pumping data is shown below as a function of manifold pressure, The pumping portion (a function of manifold pressures) can be examined by comparing ideal firing and motoring pumping torques. While this model is not an exact representation of what happens in the cylinder during actual engine operation, it is useful in understanding the basic processes and in developing an appropriate friction/pumping torque. The ideal processes most closely represent actual processes at low engine speeds, where gas transport dynamics are negligible. The area of the pumping loop represents the ideal pumping work done by the piston on the cylinder gases. This ideal process assumes that exhaust gases are displaced from the cylinder at a constant exhaust pressure, and the exhaust valve closes at top dead center. The intake valve opens and cylinder pressure decreases to the intake manifold pressure at a constant volume (clearance volume), so there is a blow-down of residual -143- Chapter 4: A Case Study in Engine Validation gases into the intake manifold. Cylinder pressure during intake is maintained at the intake manifold pressure, and the intake valve closes at bottom dead center. For the firing engine the gases are then compressed, ignited, and expanded. Next a blow-down process takes place into the exhaust manifold when the exhaust valve opens (ideally at bottom dead center). In the motoring engine no ignition takes place, so when the piston reaches bottom dead center, the gases will be at the intake manifold pressure if compression was isentropic. Ideal motoring and firing pumping loops can be modeled as a rectangle, and in the firing engine the lower corner of the ideal compression-expansion loop will cancel the overlapping portion of the pumping loop. An ideal pumping loop, assuming no heat transfer, forms a rectangle with area: w = (Pe~ Pn) Wate Vis) (438) ‘This work per cylinder per cycle can be related to pumping torque by first calculating the power per cylinder as: . (4.89) and, for a six cylinder engine: Ww = 0™ (4.90) ‘The power can be calculated as: Pater Neh, (4.91) Combining equations (4.90) and (4.91) yields an engine torque of: a a-(8) gop UP Pn) Vine ~ Vea) (4.92) -144- Chapter 4: A Case Study in Engine Validation Cr a a rT) INTAKE MANIFOLD PRESSURE (KPA) Figure 4.21 Ideal pumping torque Figure 4.21 shows pumping torque as a function of intake manifold pressure for various exhaust pressures. Motoring pumping data tends to follow these trends. At low engine speeds, the slopes of ideal and measured motoring torques, with respect to manifold pressure, are very close. The measured pumping data shows a flatter pumping curve, with respect to manifold pressure, at higher engine speeds. Again, this trend is con- sistent with the ideal curve, since exhaust pressure tends to increase with increased manifold pressure at high engine speed, due to increased air flow. Thus, there is less of a change in pumping torque, at high speeds, from high to low manifold pressures than at low engine speeds, because of a greater change in exhaust pressure. Additional work is done by the piston during the compression-expansion stroke because of heat transfer. -145- Chapter 4: A Case Study in Engine Validation If compression is assumed to be isentropic, then the temperature of the com- Pressed gases at top dead center can be estimated. Combining the polytropic relation- ship P-V7 =constant with the ideal gas law yields: _ Pde Vise mR Trae = = 8.5° 300 =570 K =297 °C. (4.93) £ & Pn Vese Volume Vote Figure 4.22 Ideal motoring pumping loop with heat transfer -146-

Vous aimerez peut-être aussi