University of Cambridge. Control Strategies in HCCI Engines

Transcription

1 University of Cambridge Department of Chemical Engineering Certificate of Postgraduate Studies Dissertation Control Strategies in HCCI Engines Ali M Aldawood Girton College Supervised by Dr Markus Kraft Submitted in June 2007

2 Preface This dissertation is submitted for the Certificate of Postgraduate Studies (CPGS) in Chemical Engineering at the University of Cambridge and is based on work carried out at the Department of Chemical Engineering of the University of Cambridge between October 2006 and June No part of the work contained in this dissertation has been submitted for any other degree. The main text of the dissertation (chapters one through six) consists of 6775 words (±5%), excluding captions, headings, citations, abbreviations and numbers. The author would like to thank Dr Markus Kraft and Dr Sebastian Mosbach for their continuous guidance and support. Appreciation and gratitude is also extended to Saudi Arabian Oil Company (Saudi Aramco) and Cambridge Overseas Trust for their financial support. ii

3 Summary The present study investigates two potential strategies for controlling the combustion process in Homogenous-Charge Compression-Ignition (HCCI) engines. The investigated strategies follow a dual fuel approach and utilize variable octane number or variable hydrogen ratio to modify the rate and timing of combustion as necessary. The study uses a modelling approach to simulate steady-state and transient operation of a single-cylinder HCCI engine. A detailed-chemistry stochastic reactor model was coupled to a one-dimensional gas dynamics model to account for the full engine cycle. The model was calibrated over a wide range of engine operation conditions, and validated against experimental data for motored and fired operation. The two control strategies were implemented using closed-loop control systems built within the gas dynamics model. Limited tests indicated that the two strategies were able to adjust the combustion phasing during load and speed transients. These tests also gave a useful insight on the performance of the employed control systems and the setup as a whole. It was found that the change needed in octane number to adjust the combustion phasing decreased substantially with increase in load, indicating a need for scheduling different gains for different operating conditions. Also, the tests showed that the responses of the controllers were affected by the significant variations in temperature at IVC with increasing speed. The controllers, therefore, need to be recalibrated taking temperature variation into consideration. Additionally, the combustion phasing was very sensitive to changes in hydrogen ratio. This sensitivity is significantly higher than what is reported in the literature for reformed gases containing mostly hydrogen. Further examination is, therefore, needed to identify the reasons behind such high sensitivity. iii

4 Table of Contents Preface...ii Summary...iii Table of Contents...iv List of Figures...v Chapter 1. Introduction HCCI versus conventional engines HCCI control problem Aim of this study...3 Chapter 2. Literature Review Effects of octane number on HCCI combustion Effects of hydrogen on HCCI combustion HCCI combustion control studies and setups...6 Chapter 3. Engine and Model Description The model Modelled engine Model calibration and validation The controller...18 Chapter 4. Results and Discussion Model settings Octane-number based control Response to load step changes Response to speed step changes Hydrogen based control...24 Chapter 5. Conclusions...26 Chapter 6. Future Work Model enhancements Future studies...29 References...30 List of Symbols...33 iv

5 List of Figures Figure 1: Combustion in (a) spark-ignition, (b) compression-ignition and (c) homogeneous charge compression-ignition (HCCI) engines...1 Figure 2: Feedback closed-loop control of combustion timing using variable octane number...6 Figure 3: A closed-loop system with a PID controller...7 Figure 4: Response of an octane number controller to positive and negative step changes in CA50 (Olsson et al. 2001)...8 Figure 5: Response of a compression ratio controller to changes in load and speed (Haraldsson et al., 2003)....9 Figure 6: Response of an inlet temperature controller to step changes in load and speed (Haraldsson et al., 2004) Figure 7: A simplified block diagram of the test setup Figure 8: The engine s full-cycle is simulated by GT-Power coupled with a stochastic reactor model...12 Figure 9: A schematic of Sandia's engine set-up (Sjöberg and Dec, 2004) Figure 10: The GT-Power model of Sandia s single-cylinder HCCI engine Figure 11: Comparison of cylinder temperature at 10 o after BDC in the compression stroke as predicted by Sandia s model and current GT-Power model Figure 12: Temperature variations of firedeck, piston, and cylinder wall with the engine speed during motored and fired operation...15 Figure 13: Model validation against experimental data of motored operation at an engine speed of 1200 rpm...16 Figure 14: Exhaust temperature versus engine speed as predicted by the GT-Power model compared to experimental readings at motored operation Figure 15: Model validation of fired operation at an engine speed of 1200 rpm and equivalence ratio (Φ) of Figure 16: A flow diagram of the control process...18 Figure 17: HCCI operating window bounded by knocking and misfire Figure 18: GT-Power map of octane number and hydrogen ratio controllers Figure 19: Response of the octane number controller to step changes in load at 1200 rpm Figure 20: Response of the octane number controller to variations in engine speed at Φ= Figure 21: Response of the hydrogen ratio controller to step changes in load at an engine speed of 1200 rpm and with base fuel octane number of Figure 22: Replacing real-time computations with pre-processed tabulated results...28 v

6 Chapter 1. Introduction The interest in the homogeneous-charge compression-ignition (HCCI) concept has been growing steadily since its first introduction in 1979 (Onishi et al., 1979). The concept has very attractive emission and fuel economy characteristics, making it a potential alternative to combustion regimes used in current internal combustion engines. Successful application of this concept, however, is still dependent on finding an effective strategy for controlling its combustion over a wide range of operating conditions HCCI versus conventional engines The low efficiency of the spark-ignition (SI) engine results in poor fuel economy and high carbon dioxide emissions. Other undesirable emissions (such as carbon monoxide, unburned hydrocarbons and nitrogen oxides) are also produced, but these can be treated effectively by exhaust aftertreatment systems. The compression-ignition (CI) engine, despite its high efficiency, is constrained by its hard-to-treat emissions of nitrogen oxides (NO x ) and particulate matter (PM). Figure 1: Combustion in (a) spark-ignition, (b) compression-ignition and (c) homogeneous charge compression-ignition (HCCI) engines. In SI engines, a premixed air/fuel charge is compressed and then ignited by an externally controlled spark. As a result, combustion takes place in the vicinity of the spark plug electrodes, initiating a flame front that travels across the combustion chamber consuming the combustible charge (Figure 1a). In order for the spark to efficiently initiate the combustion and for the flame to travel across the chamber in reasonable time, the air-to-fuel ratio is kept close to stoichiometric. Additionally, compression temperatures and pressures should be relatively low to avoid spontaneous ignition of the charge that results in the knocking phenomenon. These limitations negatively affect the overall efficiency of SI engines. 1

7 Compression ignition (CI) engines, on the other hand, operate with overall lean to very lean charge and at much higher compression temperatures and pressures. In this engine, fuel is injected at high pressure into highly compressed air (Figure 1b). Stratified charge is formed in the injection area, with air-to-fuel ratio ranging from very rich to very lean. Due to the high air temperature, the rich region of the charge auto-ignites, initiating a flame front that propagates through the remaining charge regions. The formation of NO x occurs mainly in the rich regions of the stratified charge, where the flame is initiated and the temperature is highest. Particulate matter is formed through pyrolysis processes in the very rich regions where oxygen is scarce. The HCCI engine can be seen as a hybrid between SI and CI engines. In an HCCI engine, fuel is premixed with air at relatively lean ratios. The nearly homogeneous charge is then compressed to a pressure and temperature sufficient to cause the charge to auto-ignite throughout the combustion chamber. Normally, combustion starts simultaneously at many locations where temperature is highest (Figure 1c). The utilization of lean combustion and the operation at relatively high compression pressure gives the HCCI engine the advantage of high efficiency, and the homogeneity of air/fuel charge eliminates the formation of particulate matter and minimizes the formation of nitrogen oxides HCCI control problem The HCCI combustion characteristics make it a promising combustion concept for environmentally compliant engines. However, several problems must be resolved before this concept can be developed into a practical engine. One of the most difficult problems in HCCI is the lack of control over the start and rate of combustion process. In SI engines, the start of combustion depends on the externally-controlled spark timing, and the burn rate is limited by the flame propagation rate. In CI engines, combustion takes place after a short delay following the injection of fuel, and the burn rate depends on the externally-controlled fuel injection rate. In HCCI, the combustion of the premixed charge starts spontaneously without any external control. The start and rate of combustion are solely dependent on the thermo-chemical conditions inside the combustion chamber. It has been well established that such intrinsically controlled spontaneous combustion is dominated by chemical kinetics rather than thermodynamic effects. The lack of external control in HCCI combustion results in difficulties at low and high load operation. The low combustion temperature at low loads becomes insufficient to sustain a complete combustion of the air-fuel charge which results in increased emission of carbon monoxide and hydrocarbons. Misfire eventually takes place limiting the low load operating range of the engine. At high loads, the rapid heat release causes the engine operation to become very noisy and unstable. Knocking starts to occur and 2

8 combustion efficiency deteriorates as a result setting a limit to the upper range of operation. Misfire at low load and knocking at high load result in an impractically narrow HCCI operating envelope along the speed range. In order to solve the control problem, several combustion control strategies have been suggested and studied. Examples of such strategies include using intake air heating, internal/external exhaust gas recirculation, variable compression ratio, variable dual fuel mixtures, and charge/thermal stratification. In general, these methods try to control the chemical reactions by controlling the combustion temperature and pressure or by altering the species involved in the reactions Aim of this study The current study uses a modelling approach to investigate the feasibility of using a dual fuel strategy to control the combustion in HCCI over a wide range of speed and load. The study uses a full-cycle model to simulate steady-state and transient operation of a single-cylinder HCCI engine fuelled with two primary reference fuels: iso-octane and normal-heptane. The combustion timing is controlled by altering the ratio between the two fuels (i.e. the octane number). Also, a second controlling strategy based on hydrogen addition is investigated. In this latter case, the ratio of the primary reference fuels is kept constant while the hydrogen ratio is varied as necessary to control the combustion rate and timing. Preliminary results obtained using these two strategies are described in this report. 3

9 Chapter 2. Literature Review 2.1. Effects of octane number on HCCI combustion The octane number is defined as the volume percentage of iso-octane (C 8 H 18 ) in a mixture of iso-octane and normal-heptane (C 7 H 16 ). This number, which has a scale of 0 to 100, is used as a measure of a fuel s resistance to knock in the context of sparkignition engines. Lower octane number means lower resistance to knock and vice versa. Iso-octane, a highly branched-chain paraffinic compound, has by definition an octane number of 100 while normal-heptane, a straight-chain paraffinic compound, has an octane number of 0. A mixture which contains 70% iso-octane and 30% n-heptane has, therefore, an octane number of 70 (Heywood, 1988). For commercial fuels, which usually comprise a very large number of hydrocarbon compounds, the octane number is determined by matching their knocking intensity at certain test conditions to an appropriate mixture of iso-octane and n-heptane. The tested fuel is then given an octane number equal to that of the matching mixture. Octane number played, and continues to play, an important role in defining combustion characteristics of spark-ignition engine fuels. However, it is widely agreed that octane number is not an appropriate indicator of the fuel combustion quality in HCCI engines as it only represents fuel auto-ignition characteristics at certain operating conditions. Fuels which have different compositions may exhibit similar characteristics at the octane test conditions but have completely different characteristics at other conditions. This variability is called fuel octane sensitivity. For fuels which do not exhibit this sensitivity, the octane number can be a good indicator of auto-ignition characteristics at different conditions. Studies have shown that the octane number of primary reference fuels, characterized by no octane sensitivity, can be used as an indicator of the low temperature oxidation reactions preceding the main combustion event in HCCI engines. These reactions, referred to as first-stage ignition, play an important role in determining the rate and timing of combustion in HCCI engines (Tanaka et al., 2003). Primary reference fuels, such as iso-octane and normal-heptane, have different firststage ignition characteristics. Normal-heptane exhibits a much larger heat release during the first-stage ignition than iso-octane. The larger heat release increases the rate of pressure rise and shortens the delay before the main combustion event (i.e. second-stage ignition) resulting in advanced combustion timing (Tanaka et al., 2003). These effects suggest that octane number can be potentially employed to control the combustion process in HCCI engines and extend the HCCI operating range. Such conclusion is supported by several studies. For instance, Lu et al. (2005) found that increasing the octane number retards the start of heat release and decreases the pressure 4

10 rise rate in both combustion stages. They also found that the cumulative heat release in the first stage was strongly dependent on the concentration of n-heptane in the mixture. Atkins and Koch (2005) suggested that octane number variation could be employed effectively to extend the operating range of the HCCI engine. Bhave et al. (2004a) found that increasing octane number retarded the start of combustion and prolonged the combustion duration Effects of hydrogen on HCCI combustion Many HCCI control strategies based on the dual fuel approach have been suggested and investigated in the recent years. A common problem in these strategies is the need for establishing new infrastructures to supply two different fuels. For this reason, there has been a tendency to search for a dual fuel system that only needs one primary fuel to be supplied to the engine, and the secondary fuel is produced on-board by reforming part of the primary fuel (Hosseini and Checkel, 2007a and 2007b; Narioka et al., 2006; Shudo et al., 2003; Bromberg and Ravinovich, 2001). A good example of this is the case where a primary hydrocarbon fuel (such as gasoline) is used to produce hydrogen-rich reformed gases on-board the engine. These gases are then introduced to the engine as necessary to control the combustion of the primary fuel. The main constituent in these reformed gases is hydrogen. Other constituents include carbon monoxide, carbon dioxide and traces of light hydrocarbon gases. These gases can be introduced to the engine cylinder along with the intake charge or with the recirculated exhaust gas, or injected directly into the cylinder. The latter case requires a dedicated injection system but it allows for a fast-response (cycle-to-cycle) control. Hosseini and Checkel (2007a and 2007b) added reformed gas containing 75% hydrogen and 25% carbon monoxide to low octane and high octane primary reference fuels. They found that reformed gases addition caused the combustion timing to retard, and consequently, the maximum combustion pressure and pressure rise rate to decrease. The same researchers also examined the effect of adding reformed gas to a compressed natural gas (CNG) HCCI engine (Hosseini and Checkel, 2006). They found that adding the hydrogen-rich gas expanded the operating range of the CNG HCCI engine and increased the combustion efficiency. But, in contrast to the case of primary reference fuels, it was found that addition of reformed gas to CNG advanced the combustion start. This opposite effect was also reported in other studies. Using a chemical kinetics model, Kongsereeparp and Checkel (2007) found that addition of reformed gas advances the start of combustion in the case of natural gas while retarding it in the case of n-heptane. Yap et al. (2004 and 2006) also found that adding hydrogen to natural gas advances the start of combustion. The observed opposite effect was attributed to the difference in combustion characteristics between natural gas and n-heptane. The effect of reformed gas is 5

11 dominantly thermodynamic in the case of the single-stage ignition natural gas. It was shown that addition of reformed gas alters the ratio of specific heats and increases the pre-combustion temperature causing the combustion to start earlier. In the case of n- heptane (a two-stage ignition fuel), the effects are dominated by chemical kinetics. Addition of reformed gas delays the onset of the first stage ignition and hence delays the start of combustion and decreases the rate of heat release (Kongsereeparp and Checkel, 2007). The existence of carbon monoxide in reformed gases may have an influence on hydrogen effect. Sato et al. (2005) found that adding only hydrogen to methane caused the ignition timing to advance while adding CO caused it to retard. When a mixture of these gases is added however, the ignition timing advanced but at a slower rate than the hydrogen case HCCI combustion control studies and setups In order to evaluate a control strategy, a test setup capable of representing engine behaviours at a wide range of steady-state and transient conditions is necessary. This setup must be capable of adjusting the controlled parameter(s), on which the strategy is based, in response to variations in operation conditions. For example, if the control strategy is based on controlling the combustion timing by varying the octane number, then the setup must be capable of changing the octane number based on the combustion timing reading. If a feedback closed-loop control system is used in this case, the controller will use the combustion timing reading as a feedback signal and compare it to a target value to calculate the difference, or the error. The controller then decides the change necessary in the octane number based on this error and commands an actuator, dual fuel injectors in this case, to implement the change (Figure 2). Figure 2: Feedback closed-loop control of combustion timing using variable octane number. This process continues until the actual combustion timing becomes equal to the target value, i.e. the error becomes zero. Of course this is true only in the operating window where octane number adjustment is capable of bringing the combustion timing to the target value. The boundaries of the operating window using a specific strategy are 6

12 drawn at the points where a well-tuned controller fails to converge to the target value. Controllers of a PID (Proportional, Integral, and Differential) type are often used in control strategy evaluation studies because of their simplicity and effectiveness. A PID controller calculates the necessary change in the controlled parameter based on the error (i.e. the difference between target value and actual value). This calculated change is called the gain, and it consists of three terms: a proportional gain, an integral gain, and a differential gain (Figure 3). Figure 3: A closed-loop system with a PID controller. Finding appropriate gains for a certain application is called controller tuning. There are several well-established methods for tuning PID controllers. The most famous of these is the Z-N method developed by Ziegler and Nickols in Other popular methods include the CHR method developed in 1952 by Chien, Hrones, and Reswick, and the C- C method developed in 1953 by Cohen and Coon (Coughanowr, 1991; Astrom and Hagglund, 1995). These methods evaluate the response of the controlled system to a certain implemented change and calculate the gains accordingly. It is not necessary to consider the three gain terms of a PID controller. In systems where the first-order dynamics are dominant, the differential term is often ignored to simplify the tuning process and to decrease the sensitivity of the controller to fluctuations in the system. Different approaches have been used to evaluate various potential HCCI combustion control strategies. Some researchers have carried out their studies on actual HCCI engines while others used sophisticated full-cycle engine models to simulate HCCI behaviours. The modelling approach, which is adopted in this current study, provides significant flexibility in building the test system but understandingly offers only an approximation to actual engine behaviours. Olsson et al. (2001) implemented a closed-loop control strategy based on the octane number to control the combustion phasing of an actual turbo-charged HCCI engine. A 7

13 dual fuel port injection system injects iso-octane and n-heptane in varying ratios controlled by a PID controller. Combustion phasing, represented by CA50 (the crank angle of 50% heat release), was calculated in real time for each cycle and used as a feedback signal to the controller. The sensitivity of combustion phasing to change in octane number varied significantly, and therefore, different gains were used for different operating conditions. The approach was found to work on a wide range of speeds and loads. The controller s response time to a step-change of 5 degrees in CA50 was in the range of 10 to 30 engine cycles depending on the direction of the change (Figure 4). Figure 4: Response of an octane number controller to positive and negative step changes in CA50 (Olsson et al. 2001). Haraldsson et al. (2003) applied a closed-loop control strategy to control combustion phasing in a real five-cylinder engine using variable compression ratio. Combustion phasing, represented by CA50, was calculated in real time using pressure traces from individual cylinders, and the mean value was used as a feedback signal. An effort was made to minimize the difference in CA50 between different cylinders by varying the fuel input to these cylinders. They used cascaded coupled PID controllers for both compression ratio and CA50, and a target CA50 of 7 o after TDC was set throughout the tests. Their system was found to be fast and effective and handled step changes in load in a matter of few cycles. It was not so effective though in handling ramps of speed and inlet temperature (Figure 5). The time constant for the cascaded controllers was about 14 cycles at an engine speed of 2000 rpm, which corresponds to about 0.84 seconds. The same research group also studied a closed-loop control system that uses fast thermal management (Haraldsson et al., 2004) instead of variable compression ratio. This system utilized an electric heater and heat recovered from exhaust along with an unheated cold stream to control the air inlet temperature. A PID controller controls the inlet temperature by adjusting the ratios of cold and hot air streams to the engine. Similar to the previous study, the target CA50 was set to 7 o after TDC throughout the tests (Figure 6). This system provided a comparable and in some cases better 8

14 performance than the previous system which used variable compression ratio. The controller handled step changes in load satisfactorily with a time constant of 5 to 6 cycles at an engine speed of 2000 rpm ( seconds). The performance of the controller was less satisfactory during speed ramps. However, they found that changing the target CA50 with speed improves the controller response during these ramps. They also found that choice of fuel and its low temperature reactions properties had a large impact on the controller s performance. Figure 5: Response of a compression ratio controller to changes in load and speed (Haraldsson et al., 2003). Chang et al. (2007) studied a control strategy that uses modulated residual gas fraction to compensate for transient variations in cylinder wall temperature. They simulated a multi-cylinder HCCI engine using the GT-Power platform, a simulation tool based on one-dimensional fluid dynamics, along with a simple auto-ignition model coupled with empirical correlations for burn rate and combustion efficiency. The use of simple empirical correlations instead of chemical kinetics models was motivated by the need for fast calculations. The control strategy was implemented by a PID Simulink controller (using only the PI terms) linked to the GT-Power engine model. The residual gas fraction is used as the controlled parameter in order to control the ignition timing. The employed control strategy was able to bring the ignition timing (represented by the start of combustion) to the target window throughout the imposed transient profile which spanned a load range from 2 to 4.5 bar and engine speed from 1400 to 2100 rpm. 9

15 Figure 6: Response of an inlet temperature controller to step changes in load and speed (Haraldsson et al., 2004). 10

16 Chapter 3. Engine and Model Description 3.1. The model The current study uses a modelling approach to investigate the feasibility of two control strategies. One is based on octane number variation and the other is based on hydrogen addition. The study utilizes a full-cycle model to simulate steady-state and transient operation of a single-cylinder HCCI engine fuelled with primary reference fuels (mixtures of iso-octane and normal-heptane). The combustion rate and timing are controlled by varying either the octane number of the fuel (i.e. the ratio between the two reference fuels) or the hydrogen ratio. A simplified block diagram of the test setup is given in Figure 7. Two independent closed-loop control systems are used to implement the control strategies. Each of these systems can be switched on or off as desired, and in principle, they can be used simultaneously. Figure 7: A simplified block diagram of the test setup. The engine model was built using the GT-Power platform, a one-dimensional gas dynamics code capable of representing flow and heat transfer in internal combustion engines. GT-Power also contains advanced capabilities for engine operation analysis and control at both steady-state and transient conditions. A stochastic reactor model (SRM) is coupled with the GT-Power model to account for the closed-volume portion of the engine cycle. This SRM is based on the probability density function (PDF) approach and has been used and validated in many previous HCCI combustion studies (Bhave et al., 2004a; Bhave et al., 2004b; Bhave et al., 2005; Mosbach et al., 2006). The SRM accounts for detailed chemistry kinetics associated with combustion, as well as physical interactions such as turbulent mixing and heat transfer between the fluids and surrounding walls. In this current setup, the SRM does not account for residual gases left out from previous cycles. 11

17 Cylinder pressure Figure 8: The engine s full-cycle is simulated by GT-Power coupled with a stochastic reactor model. As illustrated in Figure 8, the closed-volume period starts with the closing of the intake valve (IVC) and ends with the opening of the exhaust valve (EVO). GT-Power simulates the open-volume portion of the cycle and passes the closed-volume initial conditions to the SRM at the IVC point. From there, the SRM marches through the closed-volume period in pre-defined time steps until the EVO point, accounting for the compression, ignition, combustion and expansion processes in the engine cycle. At each time step, the SRM passes back to GT-Power progress information so that GT-Power can correctly represent operation progress and calculate global engine parameters. Figure 9: A schematic of Sandia's engine set-up (Sjöberg and Dec, 2004). 12

18 3.2. Modelled engine Figure 9 shows a schematic of the HCCI experimental setup used by the engine group at Sandia National Laboratories. The engine used in this setup is a six-cylinder mediumduty diesel engine converted to a single-cylinder HCCI research engine. The five remaining cylinders were deactivated but kept for dynamic balancing of engine rotation. The engine is equipped with two different fuelling systems, a direct injection system and a pre-mixed fuelling system. A detailed description of the setup, which has been used extensively to study HCCI combustion, is given in Sjöberg and Dec (2004). The engine s main specifications are given in Table 1. Table 1: Specifications of Sandia s single-cylinder HCCI engine (Sjöberg and Dec, 2004). Cylinder displacement (litres) Bore Stroke (mm) Connecting rod length (mm) 192 Compression ratio Number of valves 4 IVO (CA) IVC (CA) EVO (CA) EVC (CA) Valve timing crank angle (CA) is measured with respect to the firing TDC (end of compression) which is set to 0 o in this case Figure 10: The GT-Power model of Sandia s single-cylinder HCCI engine. 13

19 For the current study, a GT-Power model of Sandia s HCCI engine was built and calibrated. The model accounts only for the active cylinder and part of the intake and exhaust systems, and only considers the fully pre-mixed fuelling option (Figure 10). Future extensions planned for the model will account for external exhaust gas recirculation and direct fuel injection Model calibration and validation An extensive effort was made to ensure that the model is capable of representing the responses of the modelled engine over a wide range of engine operation conditions. For this purpose, the actual variations in intake/exhaust temperature and pressure at various conditions were taken into consideration during the calibration of the model. The intake pressure in the actual engine is maintained at a time-average of 1 bar by adjusting the air flow rate to the engine. Instantaneous pressure at the instant of intake valve opening (IVC) changes slightly depending on the engine speed. In the model, the pressure at the IVC changes from about 0.98 to 1.01 over the speed range from 600 to 2400 rpm. A PID controller to keep the pressure at IVC constant at all conditions was included in the engine model. It was thought however that the small changes in IVC pressure could be tolerated, and hence the controller was not activated. As in the engine, the exhaust pressure in the model maintained a value of 1 bar at all conditions Sandia model Current GT-Power model BDC Temperature (K) Engine Speed (rpm) Figure 11: Comparison of cylinder temperature at 10 o after BDC in the compression stroke as predicted by Sandia s model and current GT-Power model. In contrast to the pressure, the temperature at IVC varies significantly with engine speed. Figure 11 compares the temperature variations in the current GT-Power model with those predicted by Sandia s model for the same engine (Sjöberg and Dec, 2004) in 14

20 motored operation (i.e. equivalence ratio is equal to zero). The temperature readings in this case are taken 10 o after the bottom dead centre (BDC) during the compression stroke, which corresponds to 15 o before the IVC. The temperature at this instant in the cycle increases by more than 20 degrees Kelvin as the engine speed increases from 600 to 2400 rpm. The results from Sandia (Sjöberg and Dec, 2004) suggest that variation in equivalence ratio has a weaker effect on the temperature at IVC than that of the engine speed. The temperature taken 10 o after BDC increases only by about 7 degrees Kelvin as the equivalence ratio is increased from 0.1 to 0.3. Because of this and the lack of experimental data for the equivalence ratio range targeted in this study, the calibration was only made to a single operating point for which information is available. This point occurs at an engine speed of 1200 rpm and equivalence ratio of 0.44 at which IVC temperature is known to be 358 K. The variation in internal cylinder-wall temperatures with engine speed was also accounted for in the GT-Power engine model. These variations affect the IVC temperature and the heat transfer process during the closed-volume period. However, because the stochastic reactor model (SRM) handles the heat transfer interactions between the fluids inside the cylinders and the cylinder walls independently from GT- Power, the variations in wall temperatures have no effect on these interactions. This is resolved by passing a weighted-average temperature for the three walls (firedeck, cylinder, and piston-top) to the SRM as it accounts only for one internal surface temperature. 470 Temperature (K) Fired Motored Fired Motored Piston Firedeck 390 Fired Motored Cylinder wall Engine Speed (rpm) Figure 12: Temperature variations of firedeck, piston, and cylinder wall with the engine speed during motored and fired operation. 15

21 Experimental data for firedeck (cylinder head) temperatures against engine speed during motored operation are imposed on the model through a speed-temperature lookup table. The cylinder and piston temperatures are calculated using empirical approximations suggested by Sandia for the same engine (Sjöberg and Dec, 2004). For the fired operation, experimental temperature information is only available for one operating condition (1200 rpm and Φ=0.44). This information was used to calculate the fired operation temperature curves by elevating the curves of the motored case so that to match the known temperatures at 1200 rpm. The wall temperature variations with engine speed in motored and fired conditions are shown in Figure 12. Wall temperature variations due to changes in equivalence ratio were not considered because of lack of experimental data. Instead, the temperature information at Φ=0.44 was used across the tested equivalence ratio range. The built GT-Power model was validated against experimental data obtained directly from Sandia or from data they published in the literature. Figure 13 shows the model results of cylinder pressure trace versus experimental results in motored operation at 1200 rpm. Exhaust temperatures at various speeds in motored operation are also compared in Figure Experimental data, Motoring GT-Power model, Motoring Pressure (bar) Crank Angle (deg) Figure 13: Model validation against experimental data of motored operation at an engine speed of 1200 rpm. Performance of the coupled GT-Power/SRM model at fired operation was also validated against a set of experimental data obtained form Sandia. Figure 15 compares the cylinder pressure trace as predicted by the model to experimental results at equivalence ratio of Φ=0.44 and engine speed of 1200 rpm. Both the engine and the model use a fuel comprised of iso-octane and normal-heptane only. The shown fit was obtained with a model octane number of 62 as opposed to 83 in the experiment. 16

22 Exhaust Temperature (K) Experimental data GT-Power model Engine Speed (rpm) Figure 14: Exhaust temperature versus engine speed as predicted by the GT-Power model compared to experimental readings at motored operation. The results also show that the model does not predict the low temperature reactions properly. This and the discrepancy in octane number can be partially attributed to the deficiency in the chemical kinetic mechanisms employed in the model. Also, the fact that the SRM in the current setup does not account for residual gases (i.e. gases left out from previous cycle) is another possible source of these discrepancies. Residual gases usually contain active trace species that enhance low temperature reactions and hence affect ignition timing and pressure rise rate. Using a lower octane number fuel (which contains more normal-heptane) compensates for the lack of these active species Experimental data GT-Power/SRM model Pressure (bar) Crank Angle (deg) Figure 15: Model validation of fired operation at an engine speed of 1200 rpm and equivalence ratio (Φ) of

23 3.4. The controller As described before (refer to Figure 7), the current setup is designed to test two different control strategies: one is based on octane number and the other on hydrogen ratio. Either of the two strategies is used at any given time. The two strategies are implemented using two simple independent closed-loop control systems. PID controllers are utilized to adjust the combustion phasing by varying the octane number of the fuel (i.e. the ratio between the two reference fuels) or varying the ratio of hydrogen while keeping the ratio between the two reference fuels constant. Figure 16: A flow diagram of the control process. The combustion phasing reference point considered here is CA50, the crank angle at which 50% of heat has been released. The CA50 is calculated by the SRM upon the end of each cycle and passed to the PID controllers which use it to determine the necessary change in octane number or hydrogen ratio. This change is implemented by the SRM in the following cycle (Figure 16). The SRM also passes pressure rise rate (PRR) information to the controllers to indicate whether the cycle is operating in the normal, knocking, or misfiring region. High PRR indicate that the cycle is operating at or beyond the knocking boundary of the operating window, and low values indicate that the cycle is operating at or beyond the misfiring boundary (Figure 17). Based on the PRR information, the controllers make adjustments to keep the next cycle within the normal region. Normally, several cycles are needed before the target combustion phasing is reached and the controller s output stabilizes. This is called settling time and depends highly on the calibration process in which the different gain terms are determined. The PID controllers used in this setup were calibrated using a procedure suggested by Gamma Technologies Ltd. (2006) and based on an open-loop step response analysis method. Controller parameters are calculated by analysing the open-loop system s response to a small step-change of the controlled parameter (octane number or hydrogen ratio in this case). The procedure simplifies the calculations by assuming a first-order system response and hence ignores the differential gain term. This simplification is reasonable since most responses in engine applications are dominated by first-order dynamics. 18

24 Figure 17: HCCI operating window bounded by knocking and misfire. Figure 18 shows the GT-Power map for the control system used in the current study. At the end of each cycle, the SRM passes CA50 and PRR signals to the two controllers through limiting switches. These switches pass the received CA50 signal only if PRR is falling within the acceptable range which is chosen to be between 15 and 70 bar/ms (corresponding to about 2-7 bar/degree at 1200 rpm). If the PRR value falls below 15 bar/ms indicating a misfiring or motored cycle, a CA50 value slightly higher than the target CA50 is passed to the controllers which respond by trying to advance the combustion and push the next cycle back into the operating windows. If the PRR is higher than the maximum allowed value which indicates a knocking cycle, a CA50 value slightly lower than the target CA50 is passed to the controllers to retard the combustion timing and hence decrease the pressure rise rate of next cycle. Figure 18: GT-Power map of octane number and hydrogen ratio controllers. 19

25 Because optimum CA50 varies with operating conditions, it might not be appropriate to use a fixed CA50 value across the operating range. Extensive experimentation with the current model showed that a CA50 between 6-9 o after TDC would suit most of the tested operating range for this engine. Variation in the engine speed has the most significant effect on the optimum CA50. To account for this effect, the controller is equipped with a lookup table which determines the target CA50 according to engine speed. 20

26 Chapter 4. Results and Discussion A limited set of preliminary results are presented and discussed in this chapter. These results aim only to give a sense of the performance of the developed setup rather than demonstrating the overall validity of the two control strategies under investigation. The results below were obtained using the full setup described in the previous chapter which includes the GT-Power engine model, the stochastic reactor model and the GT- Power based controller. Analysis of these results gives a useful insight on the performance of the employed control system and the setup as a whole, and shows the areas where improvement is necessary before more comprehensive tests can be undertaken Model settings The SRM was set to use a relatively low computational resolution to speed up the simulation time. With this setting, the model takes about 10 minutes to simulate one full engine cycle. This however affects the accuracy of the model s predictions. In particular, the pressure rise rate is generally over-predicted necessitating a relaxation of the maximum pressure rise rate limit set in the controller. The limit, which defines the knocking boundary of the operating window and which was set originally to 70 bar/ms, was changed to 120 bar/ms to compensate for the model s over-prediction. The simulation runs consisted of 60 to 90 cycles, of which the first nine cycles were run in motored operation (no combustion) using GT-Power only. The SRM and the controller are activated in the tenth cycle until the end of the run, and step changes in engine speed or load are applied in ten-cycle intervals. For these preliminary tests and based on extensive pre-experimentation, the conditions and ranges of the runs were chosen carefully to fall within the operating window of the engine. Since these tests span only a limited operating range, the variation in cylinder wall temperatures with engine speed was only considered by the GT-Power model. The SRM was set to use a fixed average cylinder wall temperature of 420 K in all test conditions. Also, a fixed combustion phasing angle (between 7 o and 9 o after TDC) was used in each run Octane-number based control For the case of octane-number based control, the performance of the controller was tested with step changes in load and speed. No hydrogen was added to the primary reference fuels and the hydrogen controller was inactive. In the load test, the speed was fixed at 1200 rpm and a load change was implemented by varying the equivalence ratio from to 0.35 in steps of The load, measured in terms of brake mean effective pressure (BMEP), changed correspondingly from about 2.9 to 4.0 bar. A combustion phasing angle (CA50) of 7 o after TDC was used throughout the test. 21

27 Response to load step changes Figure 19 shows the response of the octane number controller to the implemented step changes in load. The controller was able to bring and keep the CA50 within a range of ±0.5 o from the target value throughout the test range. The needed change in octane number to adjust CA50 decreased substantially with increase in load. This agrees with the observations reported earlier by Olsson et al. (2001). It also indicates the need for using different controller gains to suit the different sensitivities at different conditions. Cycles BMEP (bar) BMEP Equivalence ratio Equivlalence ratio CA50 (deg ATDC) Target CA50 Actual CA50 Controller output (octane ratio) Controller output (octane ratio) Pressure at IVC (bar) Pressure at IVC Temp at IVC Temperature at IVC (K) Time (seconds) Figure 19: Response of the octane number controller to step changes in load at 1200 rpm. The current controller uses a fixed set of gains at all test conditions. This may explain the relative instability in the controller s response at the two ends of the test. This instability will most likely escalate further at loads lower and higher than the tested range. 22

28 A smooth performance is noticed at the second load step, which indicates that the used gains are suitable for the conditions at this operating point. The settling time, which is defined as the time required to reach the target value and stay within acceptable variations around it, was 7 cycles on average. This corresponds to about 0.68 seconds at 1200 rpm. Engine speed (rpm) Engine Speed BMEP BMEP (bar) CA50 (deg ATDC) Controller output (Octane ratio) Target CA50 Actual CA Controller output (octane ratio) Pressure at IVC (bar) Temp at IVC Pressure at IVC Temperature at IVC (K) Time (seconds) Figure 20: Response of the octane number controller to variations in engine speed at Φ= Response to speed step changes The results from the speed test are shown in Figure 20. In this test, the target CA50 was set to 8 o after TDC and the equivalence ratio was kept at 0.44, which corresponded to a BMEP ranging from 4.9 to 5.2 bars depending on speed. Positive and negative step changes of speed (200 rpm) were applied at each ten-cycle intervals. Interestingly, the results show that at these conditions, the variations in IVC temperature due to speed 23

29 change seem to be significant enough to adjust and keep the combustion phasing within the target value without the need to change the octane number. Only a slight change was needed at the highest speed step. This effect may have also benefited from the slight increase in cylinder pressure at IVC with increasing speed. This observation confirms the findings of previous studies which suggested that intake temperature could be used effectively to control the combustion process in HCCI engine (Haraldsson et al., 2004). This can be investigated further using the current setup which already contains a facility to control the IVC temperature by adjusting intake temperature. The temperature control could be investigated as a stand alone strategy or in conjunction with octane number or hydrogen ratio strategies. In any case, the significant effect of IVC temperature on combustion phasing indicates the necessity of re-calibrating the controller parameters taking such effect into consideration. This might not be a trivial task given that IVC temperature is not dependent only on engine speed Hydrogen based control In the hydrogen based control case, the hydrogen ratio controller was activated and the octane number controller was switched off. The octane number of the base fuel (isooctane and n-heptane) was kept constant at 44 throughout the test. From the previous section, it was seen that the controller responses to speed changes were highly affected by the variations in temperature at IVC. Since these effects were not considered in the calibration of the hydrogen ratio controller, a re-calibration will need to be performed before responses to speed changes can be properly tested. Because of this, the results here only consider responses to change in load. Figure 21 shows the response of the hydrogen ratio controller to step changes in load at an engine speed of 1200 rpm. The change in load was applied by changing the equivalence ratio from 0.4 to 0.5 using increments of after each 10-cycle interval. The conditions used in this test correspond to the upper-intermediate load range for this engine as can be judged from the relatively high BMEP ( bar). The pressure rise rate at these conditions will be the limiter of the engine s operating window. For this reason, the target value for CA50 was set to the relatively late angle of 9 o after TDC to keep the pressure rise rate within the set limits. The results show that the controller was able to bring and keep the CA50 within ±0.5 o from the target value for all implemented steps. The CA50 settling time in this test was approximately 7 cycles (0.68 seconds at 1200 rpm) as in the case of octane number controller. A slight overshoot and fluctuation is noticed in most load steps indicating a need for a better tuning of the controller gains. 24