Preprints of the 18th IFAC World Congress Milano (Italy) August 28 - September 2, 2011

LQR control for speed and torque of internal combustion engines

Jos e David L opez Jairo Jos e Espinosa John Ramiro Agudelo School of Mechatronichs, Universidad Nacional de Colombia, Medell n (e-mail: jodlopezhi@unal.edu.co). School of Mechatronichs, Universidad Nacional de Colombia, Medell n (e-mail: jairoespinosa@ieee.org) Department of mechanical engineering, Universidad de Antioquia, Medell n, Colombia, (e-mail: jragude@udea.edu.co) Abstract: This paper presents a robust automation model for internal combustion engines test beds. A Linear Quadratic Regulator (LQR) allows setting the desired engine speed and torque on both compression and spark ignition engines. With this methodology, the user can change the engine by another one of dierent characteristics with few adjustments on the controller parameters. The controller was implemented using a microcontroller in order to guarantee operation in real time. The LQR controller performance has been validated in a wide range of engine operating modes, from low to high speeds and variable loads showing a good response. The description of the model using rst order transfer functions with delay has proven to be a good approximation, despite of the nonlinearities caused by the turbocharger and the electronic control unit (ECU) incorporated in the engines. This low cost automation system has been tested for the last three years in a test rig at a university laboratory showing a good performance. Keywords: Internal Combustion Engine, LQR control, MIMO model 1. INTRODUCTION Since the late nineteenth century, oil became the main source of energy in transportation systems, but by the early twenty-rst century the shortage of this product and its polluting eects force researchers to develop environmentally friendly fuels. Research and development tasks for Internal Combustion Engines (ICE) requires precise and reliable control systems in order to guarantee consistent experimental conditions, especially when small dierences are expected. This is the case in which several fuels, blends of fuels, additives, fuel saving devices or postreatment systems (among others) are going to be compared in an engine test bed. The test bed is composed by an internal combustion engine whose main shaft is connected to a braking system in order to simulate and test variable load conditions for the engine. An internal combustion engine is a machine that involves thermodynamic processes, uid mechanics and chemical reactions, its modeling is still dicult and a subject of discussion. Several authors agree that the dynamic of the combustion chamber is the dominant process of the ICE (Powers (1981); Heywood (1988); Plint and Martyr (1997); del Portillo Valdes et al. (2007)), which implies the use of
This project was nancially supported by the Colombian Ministry of Agriculture and Rural Development, Sofasa-Renault, El Area Metropolitana del Valle de Aburr a, and the Comit e para el Desarrollo de la Investigaci on (CODI) from the University of Antioquia. Project No. 003 2007D3608-67 (Bioethanol E-20 project) and project 001 2007D3347-499 (Biodiesel).

dierent models according to the engine ignition system (compression or spark), this prevents the use of a generic model for control design. The braking system is a magnetic brake based on Eddy currents where the braking force depends on the current and the coil inductance, that is dened by the geometry and material characteristics (Lee and Park (1999); Gonzales (2004); Gosline et al. (2006)). In absence of manufacturer information, it is possible to retrieve the parameters from experimental data, in order to obtain a proper model since torque is directly proportional to the current owing through the brake and to the shaft speed (Anwar (2004)). The inherent dicult of modeling the engine-brake assembly using phenomenological rst principle models oriented to control systems, has made that some authors only focus on control variables. Powers (1981) stated that for a digital control, a relatively low order model (quasi-linear) was enough, due to the restrictions in control implementation; i.e. an approach of transfer functions can be successful in the vicinity of a given speed-torque point. Bunker et al. (1997) and Cook and Powell (1988) have followed this approach and designed models with two inputs and two outputs with transfer functions of rst and second order with delays. In this work, a control system for engine speed and torque, based on a LQR, has been developed. It was designed with rst order transfer functions with delay, and it was successfully implemented and validated in the whole

Copyright by the International Federation of Automatic Control (IFAC)

2230

Preprints of the 18th IFAC World Congress Milano (Italy) August 28 - September 2, 2011

operating range of a diesel and a spark ignition enginedynamometer assemblies (L opez (2009)). 2. METHODOLOGY 2.1 Equipments and systems Two engine test beds were instrumented and automated; one was used for research in diesel engines and the other in spark ignition engines. Table 1 presents the technical specications of the engines used in this work. Table 1. Technical specications of the engines used
Value/Engine Cylinders Engine size (cm3 ) Fuel Fuel injection Air supply Max speed Max torque Max power ISUZU R 4 2.400 Diesel Direct Turbocharger 5.000 rpm 170Nm @ 2.500rpm 55kW @ 3.000rpm RENAULT R 4 1.600 Gasoline Electronic Natural 7.000 rpm 150Nm @ 3.000rpm 65kW @ 5.500rpm

Fig. 1. Ton e main window, a real time monitoring and recording software developed for the project the brake, and by the engine speed, i.e. both variables are strongly related. For the throttle valve opening, a direct current stepper motor was used. A direct current owing through the brake coil, similar to those used to control the speed in DC motors was used to generate rotation resistance (load). In order to guarantee the appropriate operating mode, an 80 V 20 A power supply with a chopper based in MOSFET transistor was developed and controlled by a Pulse Width Modulation (PWM). 2.2 Automation model description Several authors agree that the process inside the combustion chamber is the dominant dynamic of the engine, so it is advisable to model it with the ow of air or fuel entering the chamber (Powers (1981); Cook and Powell (1988); Van Lith (2002); Ritz en (2003); del Portillo Valdes et al. (2007)). But this approach divides the models by type of injection system, preventing to use one model on two engines with dierent characteristics, and in many cases, the model is too complex for real time control. Due to the diculty to obtain a phenomenological model suitable for all engine-dynamometer assemblies, besides the enormous eort involved, some authors have chosen to focus the model towards the most signicant control variables. Hopkins and Borcherts (1980) and Morris et al. (1981) have created models with seven state variables, later Powers (1981) reduced it to ve. The three authors studied spark ignition engines and they included the spark dynamics on the control variables, but that is not a modiable parameter in serially manufactured engines. A more generic approach is desirable, Powers (1981) stated that if the target is a digital control, then relatively low order models in discrete time and quasi-linear are wanted. Some authors have followed this philosophy (Bunker et al. (1997); Cook and Powell (1988)), and have designed models of two inputs and two outputs with transfer functions of rst and second order with delay. Based on their work, the internal combustion engines used in this project were modeled with independent transfer functions according to the following matrix:

The diesel engine was coupled to a Schenck W230 Eddy currents magnetic brake, while the spark ignition engine was coupled to a Schenck E90 Eddy currents magnetic brake. Both test beds have been instrumented with hotwire air ow sensors (Magnetrol TA2), K-type thermocouples (environment temperature, inlet-outlet engine water, oil, hot gases, turbo compressor inlet and outlet temperatures) and strain gauge pressure transducer sensors (ambient pressure, turbo compressor inlet and outlet and oil pressures). The diesel fuel ow was measured with an electronic weight scale, while the gasoline was measured with a Danfoss Masso 6000 Coriolis-type mass ow sensor. The engine speed sensors were mounted on the electromagnetic brakes. The Schenck E90 brake had an optical sensor, while the Schenck W230 had a magnetic reluctance sensor. A frequency-voltage converter was implemented in order to generate an analog signal proportional to the speed. The engine torque was measured with load cells attached to the brake stators. This signal reached few milliamperes and it was very sensitive to noise. In order to solve these problems an instrumentation amplier was implemented. It rejects the noise that enters through the casing and other spurious interferences while amplies the signal between 1000 and 5000 times. A low-pass active lter with a cuto frequency below the minimum engine speed (idle speed) was implemented in order to avoid undesirable variations generated by piston movement. High speed data were acquired using a LabViewTM based software together with a National InstrumentsTM data acquisition system (Model PCI-MIO-16E-4 board). The program developed for real time monitoring and recording was called Ton e (Figure 1). The engine speed system is mainly governed by the fuel entering the combustion chambers which is controlled by the throttle valve position of the acceleration system and by the torque generated by the brake. The braking torque is mainly governed by the electric current through

2231

Preprints of the 18th IFAC World Congress Milano (Italy) August 28 - September 2, 2011

K11 e1 s K12 e2 s 1 b s + 1 Y (s) = a s + U (s) K21 e3 s K22 c s + 1 d s + 1 with: Y (s) = Rpm(s) T orque(s) U (s) = %Opening (s) %Brake(s)

(1)

where: Rpm(s) : Engine speed in rpm T orque(s) : Braking torque in Nm %Opening (s) : Percentage of the throttle opening Brake(s) : Current entering the brake represented in divisions of PWM Kij : Gains i : Delay in seconds a , b , c , d : Subsystems time constants The system inputs are the percentage of throttle opening and the current through the magnetic brake, and the outputs are the engine speed and torque. In order to nd the parameters of the matrix of transfer functions, the following procedure was implemented covering a wide range of engine operating modes: (1) A test of acceleration steps (each 5%) without any brake load was carried out from idle to maximum engine speed. (2) With the throttle valve xed in the maximum engine speed point, current incremental steps (each 5%) were applied on the dynamometer coil, increasing the load until engine speed reached the minimum. (3) Alternating the acceleration and the load, steps 1 and 2 were carried out until the engine reached its maximum power, in order to obtain relevant data from the main operating modes. The IDENT toolbox of Matlab R was used after obtaining these experimental data in order to nd the parameters of each curve. An average of eight curves located throughout the operating range (speed and torque) was used in order to obtain each subsystem model. All sub-models were validated with dierent datasets from those used on the identication. The rst-order model with delay obtained for the spark ignition engine was: 350es 27es 1 s+1 YR (s) = s + U (s) (2) 0.4e s 0.8 s+1 s+1 and the model for the diesel engine was: 100es 30es 1 s+1 U (s) (3) YI (s) = s + es 3.3 s+1 s+1 The controller response was aected by disturbances associated with each engine, i.e. the turbocharger and pump for the diesel engine and the electronic control unit for the spark ignition engine. The turbocharger turn-on and the fuel pump turn-o were the most signicant non-linearities of the diesel engine.

Fig. 2. Turbocharger turn on, note that both acceleration steps had the same percentage of amplitude, but the second one (17 - 24 s) started the turbocharger with the consecuently larger speed increase (About ten times)

Fig. 3. Fuel pump turn o. At 6 s the engine reached enough speed to feed the fuel, the fuel pump turned o causing a momentary loss of speed, even the acceleration continued increasing The rst one generated a sudden increase of speed, while the fuel pump turn-o caused a temporally loose of power (ISUZU (1991)). Figure 2 shows the turbocharger turn-on and the Figure 3 shows the fuel pump turn-o processes for the diesel engine. On the spark ignition engine, the non-linearities were introduced by the electronic control unit (ECU), governing the fuel supply while guaranteeing a stoichiometric fuel-air mixture. This means that the fuel supply does not depend exclusively on the throttle valve position (accelerator). Figure 4 shows the ECU eect at 2.550 rpm (without load) where the ECU changed the speed with no throttle action. 2.3 State-space model selection A Linear Quadratic Regulator (LQR) was implemented in order to fulll the test bench requirements (speed and torque control in transient and steady state). It is a linear control technique that oers a simple state control rule, robust to non-linearities of the engines and simple to implement in digital systems (Franklin et al. (1997); Espinosa (2003)). The LQR requires a state-space model of the engine. A transformation to the state-space with a transfer function can be performed using computational tools, but multipleinput and multiple-output systems have a special proce-

2232

Preprints of the 18th IFAC World Congress Milano (Italy) August 28 - September 2, 2011

Fig. 5. Spark ignition engine on the linear region. Several xed size speed and toque steps were performed between 1.500 2.000 rpm and 10 20 Nm with an expected response on each case Fig. 4. ECU eect on the engine. On the electronic spark engine the ECU controlled the throttle valve based on the acceleration pedal position and the lambda sensor. At 38 s the internal conditions of the engine changed and the ECU increased its speed even the accelerator was xed dure for each situation. In this case a simple and practical state-space transformation method developed by Pota (1996) was implemented. The transformation procedure began by introducing state-space entries into the matrix of transfer functions of equation (1) with a change of variable, and applying the inverse Laplace transformation, resulting: x 1 1 0 x1 1 0 u1 = + x 2 0 1 x 2 0 1 u2 (4) y1 K11 K12 x1 = y2 K21 K22 x2 With this procedure, the states lost some of their physical meaning by making the transformation to time, because they could not be represented as physical variables, but they retained a very simple structure that allowed to recognize the origin of each parameter. The gain matrix values (Kij ) were replaced by each engine gains. As the control algorithm was implemented on a microcontroller, the matrix was corrected to include the actuators gains, resulting: 17.54 3.18 CRenault = (5) 1.16 2.32 CIsuzu = 0.493 4.44 0.147 14.52 (6) 3. RESULTS AND VALIDATION 3.1 Spark ignition engine validation The performance of the controller was tested through dierent operating regions of the engine operating mode. Initially several steps were made in the linearized region, Figure 5 shows speed steps between 1.500 rpm and 2.000 rpm with changes in torque of 10 Nm to 20 Nm, with a good response of the controller. Figure 6 shows the controller response to speed steps from 3.000 rpm to 3.400 rpm. Responses with damping were expected, because in this region the gain was larger than that obtained on the model linearization. which penalizes the error in the system by applying a correction proportional to the magnitude of the error. The matrix K (t) corresponds to the time invariant control law: K = R 1 B T P (9) this means that calculating the matrix K in steady state, its coecients are constant (Menendez and Simon (2004)), then the controller only needs to nd appropriate values for the Q and R matrices. An integral action was added to compensate the steadystate error. The matrices Q and R were expanded to match the structure of the cost function. On the ICE model proposed, the gain of the engine speed was punished because it was larger, as the same as the torque response time because it had a faster actuator. The engine speed and torque did not varied signicantly with dierent values of Q and R, but the actuators were sensitive to this changes and tended to oscillate, even when the system seemed stable, then those matrices must be carefully chosen. Adjusted the curves of the model to meet the design conditions, R matrices on both engines were identity, and the Q matrices were: 0.01 0 0 0 0.1 0 0 0 0 8 0 0 0 1 0 0 QR = QI = 0 0 0.05 0 0 0 0.1 0 0 0 0 1 0 0 0 0.1

In both engines most of the transfer functions coincided in the delay, it was chosen an average value for each one: Renault = 0.15 s Isuzu = 0.6 s 2.4 LQR controller LQR technique calculates a matrix K , which being admitted to a feedback control must meet x =Ax+Bu (7) u = K x furthermore, this technique minimizes a cost function given by 1 T J= x Qx + uT Ru dt (8) 2 0

2233

Preprints of the 18th IFAC World Congress Milano (Italy) August 28 - September 2, 2011

Fig. 6. Spark ignition engine out of the linear region. Two speed steps performed with a xed torque of 46 Nm show how at higher speed and torque values the controller has an accurate response

Fig. 8. Compression ignition engine with high amplitude steps. At 65 s was performed a speed step from 3.100 rpm to 2.200 rpm, the speed changed its inclination three times until it reached its nal value causing same number of peaks on torque

Fig. 7. Compression ignition engine without load. Several increments on speed (1.000 rpm each) show the controller response in the entire operating range of the engine. Note the turbocharger turn on (20 30 s) and the fuel pump turn o (128 140 s) 3.2 Compression ignition engine validation The controller performance was tested around the nonlinearities exposed before. Figure 7 shows speed steps without load from 1.000 rpm up to 4.000 rpm. Between 1.200 rpm and 1.800 rpm the turbocharger turn-on is observed, the controller did not show oscillation or steadystate error, leading the engine to the 2.000 rpm set point. When the ICE switched from 3.000 rpm to 4.000 rpm, there was a drop in the speed produced by the fuel pump turn-o, the controller response was accurate without overdamping on the nal speed value. Figure 8 shows steps from the region close to the turbocharger turn-on until 3.000 rpm, with variations of torque between 20 Nm and 40 Nm. The controller could get the nal value with an expected overshoot due the large set-point change of about 30 % of the span. 3.3 Transient control (Follower of trajectories) The procedure for conducting this test was to remove the manual control of the set-point, replacing it with a trajectory providing the controller the desired engine speed and torque at each instant of time. Figure 9 shows the test result, the ideal trajectories were superimposed (straight lines), the engine speed and torque were followed by most of the trajectory. However, as can be seen at 20 s of testing, the change in the slope of torque aected the speed trajectory, because there was not a fast actuator on the fuel injection (as expected).

Fig. 9. Controller response to transient tests. A road condition was simulated in order to verify the controller response: the user increases the engine speed before arriving to a hill, then the torque has two increases and nally the hill ends and the vehicle begins the descent. The controller had an accurate response, a short oscillation appeared at 20 s due the hard torque demand 3.4 Other results This control system has been used during the last three years in the engine laboratory at the University of Antioquia in Medell n (Colombia). Both engine test beds have been used in order to explore the performance and emissions of diesel and spark ignition engines using several biofuels and its blends with conventional fuels over a wide range of engine operating modes. The use of blends such as E20 (20% ethanol + 80% gasoline) and blends such as B5 and B20 (blends of diesel fuel with 5 and 20% of several indigenous biodiesels) have been analyzed using the instrumentation, controller and software herein presented. Those results have been published in several scientic journals (Agudelo et al. (2009a,b); Benjumea et al. (2009); Agudelo et al. (2010)). 4. CONCLUSIONS A low cost, powerful and robust multivariable control system based on transfer functions was developed, implemented and tested on two internal combustion engines (diesel and spark ignition) test beds. The control did not depend on engine or fuel type. The controller was successfully tested in a wide range of engine operating modes. It could pass through the engine non-linearities such as

2234

Preprints of the 18th IFAC World Congress Milano (Italy) August 28 - September 2, 2011

turbocharger turning on and some ECU points reaching the stabilization in neighboring points. The process of designing a LQR controller from a matrix of rst order transfer functions with delay for engine speed and torque, met the requirements of control for engine test beds used for research activities. ACKNOWLEDGEMENTS Authors would like to thank professor Orlando Carrillo for his valuable help during engine test beds instrumentation and control system development. REFERENCES Agudelo, J., Agudelo, A., and Perez, J. (2009a). Energy and exergy analysis of a light duty diesel engine operating at dierent altitudes. Revista Facultad de Ingenieria, 48, 45 54. Agudelo, J., Benjumea, P., and Villegas, A. (2010). Evaluation of nitrogen oxide emissions and smoke opacity in a HSDI diesel engine fuelled with palm oil biodiesel. Revista Facultad de Ingenieria, 51, 62 71. Agudelo, J., Gutirrez, E., and Benjumea, P. (2009b). Experimental combustion analysis of a HSDI diesel engine fuelled with palm oil biodiesel-diesel fuel blends. Dyna, 159, 103 113. Anwar, S. (2004). A parametric model of an eddy current electric machine for automotive braking applications. IEEE transactions on control systems technology, 12(3). Benjumea, P., Agudelo, J., and Agudelo, A. (2009). Eect of altitude and palm oil biodiesel fuelling on the performance and combustion characteristics of a hsdi diesel engine. Fuel, 88, 725 731. Bunker, B., Franchek, M., and Thomason, B. (1997). Robust multivariable control of an engine dynamometer system. IEEE transactions on control systems technology, 5(2), 189 199. Cook, J. and Powell, B. (1988). Modeling of an internal combustion engine for control analysis. IEEE power control systems magazine, 20 26. del Portillo Valdes, L.A., Tinaut Fluixa, F.V., Melgar Bachiller, A., and Gimenez Olavarria, B. (2007). Validaci on de un modelo fenomenol ogico para el estudio de motores diesel de inyecci on directa. In Octavo congreso iberoamericano de ingenier a mec anica, Cuzco. Espinosa, J.J. (2003). Control lineal de sistemas multivariables. Corporaci on Universitaria de Ibagu e. Franklin, G., Powell, J.D., and Workman, M. (1997). Digital control of dynamic systems. Addison Wesley Longman Inc. Gonzales, M.I. (2004). Experiments with eddy currents the eddy current brake. European Journal of Physics, 25, 463 468. Gosline, A., Campion, G., and Hayward, V. (2006). On the use of eddy current brakes as tunable, fast turn-on viscous dampers for haptic rendering. Proc. Eurohaptics, 229 234. Heywood, J. (1988). Internal combustion engine fundamentals. McGraw Hill. Hopkins, H.G. and Borcherts, R.H. (1980). Discrete Time Modeling of the Torque Response of a Spark-Ignited Fuel-Injected Engine: Applications of Adaptive Control. Academic press, New York.

ISUZU (1991). Technical manual of the diesel engine ISUZU 4JA1. ISUZU. Lee, K. and Park, K. (1999). Optimal robust control of a contactless brake using an eddy current. Mechatronics, 9, 615 631. L opez, J.D. (2009). Automatizaci on de un banco de ensayos para MCIA. Masters thesis, Universidad de Antioquia. Menendez, A. and Simon, S. (2004). Aportacio al control del convertidor CC/CA de tres nivells. Ph.D. thesis, Universitat Politecnica de Catalunya. URL http://www.tdx.cat/TDX-0104105-081901. Morris, R.L., Borcherts, R.H., Warlick, M.V., and Hopkins, H.G. (1981). Spark ignition engine model building - an identication approach to throttle-torque response. In Workshop on adaptative control. Plint, M. and Martyr, A. (1997). Engine testing: Theory and practice. Oxford: Butterworth Heinemann. Pota, H.R. (1996). Mimo systems, transfer function to state space. IEEE transactions on education, 39(1), 97 99. Powers, W.F. (1981). Internal combustion engine control system research at ford. In 20th IEEE Conf. on Decision and Control, volume 20, 1447 1452. Ritz en, J. (2003). Modelling and xed step simulation of a turbo charged diesel engine. Masters thesis, Linkopings Universitet. Van Lith, P. (2002). Hybrid fuzzy rst principles modeling. Ph.D. thesis, University of Twente.

2235