hcci combustion control using dual-fuel approach: experimental
TRANSCRIPT
HCCI Combustion Control Using Dual-Fuel Approach: Experimental and Modeling Investigations
Preprint Cambridge Centre for Computational Chemical Engineering ISSN 1473 – 4273
HCCI Combustion Control Using Dual-FuelApproach: Experimental and Modeling
Investigations
Ali Aldawood 1, Sebastian Mosbach 1, Markus Kraft 1
released: 16 January 2012
1 Department of Chemical Engineeringand BiotechnologyUniversity of CambridgeNew Museums SitePembroke StreetCambridge, CB2 3RAUnited KingdomE-mail: [email protected]
Preprint No. 114
Keywords: HCCI control, dual-fuel HCCI, multiobjective optimization, stochastic reactor model
Edited by
CoMoGROUP
Computational Modelling GroupDepartment of Chemical Engineering and BiotechnologyUniversity of CambridgeNew Museums SitePembroke StreetCambridge CB2 3RAUnited Kingdom
Fax: + 44 (0)1223 334796E-Mail: [email protected] Wide Web: http://como.ceb.cam.ac.uk/
Abstract
A dual-fuel approach to control combustion in HCCI engine is investigated in thiswork. This approach involves controlling the combustion heat release rate by adjust-ing fuel reactivity according to the conditions inside the cylinder. Experiments wereperformed on a single-cylinder research engine fueled with different ratios of pri-mary reference fuels and operated at different speed and load conditions, and resultsfrom these experiments showed a clear potential for the approach to expand the HCCIengine operation window. Such potential is further demonstrated dynamically usingan optimized stochastic reactor model integrated within a MATLAB code that simu-lates HCCI multi-cycle operation and closed-loop control of fuel ratio. The model,which utilizes a reduced PRF mechanism, was optimized using a multi-objective ge-netic algorithm and then compared to a wide range of engine data. The optimizationobjectives, selected based on relevance to this control study, were the cylinder pres-sure history, pressure rise rate, and gross indicated mean effective pressure (IMEPg).The closed-loop control of fuel ratio employed in this study is based on a searchalgorithm, where the objective is to maximize the gross work rather than directlycontrolling the combustion phasing to match preset values. This control strategyproved effective in controlling pressure rise rate and combustion phasing while notneeding any prior knowledge or preset information about them. It also ensured thatthe engine was always delivering maximum work at each operation condition. Thisis in a sense analogous to the use of maximum brake torque timing in spark-ignitionengines. The dynamic model allowed for convenient examination of the dual-fuelapproach beyond the limits tested in the experiments, and thus helped in performingan overall assessment of the approach’s potential and limitations.
1
Contents
1 Introduction 3
2 Engine Experiments 4
3 Model Setup and Optimization 9
4 Results and Discussions 19
5 Summary and Conclusions 23
6 Acknowledgement 28
7 Definitions, Acronyms and Abbreviations 28
References 29
2
Page 2 of 21
was feasible. Similar results were also reported by Atkins and Koch [4] who concluded that varying of fuel octane rating (i.e. the ratio of iso-octane to n-heptane) could be employed to increase the load range of the HCCI engine.
Figure 1 - A schematic of the dual-fuel engine concept. A pressure sensor fitted inside the cylinder provides a pressure feedback signal to a controller, which based on this signal decides the best mixing ratio of the two fuels.
The use of commercial fuels or mixtures of single-component fuels and commercial fuels have been investigated in several more recent studies. Examples of fuels considered include mixtures of ethanol and n-heptane [5]; mixtures of n-heptane and diesel [6]; and mixtures of gasoline and diesel [7,8]. All of these studies reported some positive results and varying levels of feasibility.
The current paper aims to shed more light on the potential of using primary reference fuels in a dual-fuel strategy to control HCCI combustion. The paper presents new HCCI engine results as well as simulation results that examine wider operating windows, and which are based on a modeling approach that combines reduced chemistry representation with genetic algorithm optimization. The paper also suggests a new way for controlling HCCI combustion based on maximization of obtained work rather than the conventional phasing control.
Table 1 – Engine specifications and valve timing information. Crank angles here are measured with respect to firing TDC.
Number of cylinders 1 Operation cycle 4-stroke Combustion mode HCCI Number of valves 4 Displacement (litre) 0.5 Bore (mm) 84 Stroke (mm) 90 Connecting rod length (mm) 159 Crank radius (mm) 45 Compression ratio 12:1 Fuel delivery system PFI Cooling water temperature 90oC Lubrication oil temperature 90oC IVO (CAD) -356 EVC (CAD) -352 IVC (CAD) -156 EVO (CAD) 170
Intake port Exhaust port
Low reactivity fuelHigh reactivity fuel
Pressuredata
HCCI
Air
Controller
Fuel System
Controller
Figure 1: A schematic of the dual-fuel engine concept. A pressure sensor fitted inside thecylinder provides a pressure feedback signal to a controller, which accordinglydecides the best mixing ratio of the two fuels.
1 Introduction
The Homogeneous-charge compression-ignition (HCCI) concept exhibits attractive emis-sions and fuel economy characteristics. This makes it a potential alternative to combustionmodes currently used in internal combustion engines. Successful application of this con-cept, however, is still dependent on finding an effective strategy for controlling HCCIcombustion over a wide range of operating conditions.
Many HCCI control strategies based on a dual-fuel approach have been suggested and in-vestigated in recent years. Among these is the use of two hydrocarbon fuels with largelydifferent reactivity and combustion characteristics. The combustion in this case is con-trolled by adjusting fuel reactivity, by the means of mixing of the two fuels, accordingto the conditions inside the cylinder. Such approach is schematically illustrated in Figure1. A pressure sensor fitted inside the cylinder provides a pressure feedback signal to acontroller, which accordingly decides the best mixing ratio of the two fuels.
Fuels that exhibit multiple stages ignition are known to be more reactive than those ex-hibiting only a single stage. Characteristics of first-stage ignition also differ among fuelswith multiple stages ignition. For instance, n-heptane has a much larger heat release dur-ing the first-stage ignition than iso-octane. The larger heat release increases the rate ofpressure rise and shortens the delay before the main combustion event (i.e. second-stageignition) resulting in advanced combustion timing [12, 16].
The above effects exhibited by primary reference fuels (PRF) and many other hydrocarbonfuels could be potentially employed to extend the operating range of the HCCI engine.This potential has been examined in several previous studies. For example, Olsson et al.
3
[15] investigated a dual-fuel approach that uses primary reference fuels to control HCCIengine, and concluded that such approach was feasible. Similar results were also reportedby Atkins and Koch [4] who concluded that varying of fuel octane rating (i.e. the ratioof iso-octane to n-heptane) could be employed to increase the load range of the HCCIengine.
The use of commercial fuels or mixtures of single-component fuels and commercial fuelshave been investigated in several more recent studies. Examples of fuels considered in-clude mixtures of ethanol and n-heptane [18]; mixtures of n-heptane and diesel [13]; andmixtures of gasoline and diesel [8, 11]. All of these studies reported some positive resultsand varying levels of feasibility.
The current paper aims to shed more light on the potential of using primary referencefuels in a dual-fuel strategy to control HCCI combustion. The paper presents new HCCIengine results as well as simulation results that examine wider operating windows, andwhich are based on a modeling approach that combines reduced chemistry representationwith genetic algorithm optimization. The paper also suggests a new way for controllingHCCI combustion based on maximization of obtained work rather than the conventionalphasing control.
2 Engine Experiments
The experiments for this work were carried out on a half-liter single-cylinder researchengine running in HCCI mode. The specifications of this engine are listed in Table 1.The engine has a pent-roof combustion chamber fitted with four valves. Fuel is deliveredintermittently via a port fuel injector located at the end of the intake port just above theintake valves. Intake air passes through a conditioner to adjust its temperature, humidity,and pressure as necessary. All experiments were performed at a boosted intake pressureof 1.5 bar and an intake air temperature of 75oC. The engine was fueled with primaryreference fuels at three different volume ratios: PRF40 (i.e. 40% iso-octane and 60%n-heptane), PRF60 and PRF80. Load sweeps were performed at three constant speedsof 1200, 1500 and 1800 rpm. Upper and lower bounds of possible load range are firstidentified at each test condition, and then test points are selected at reasonably-distancedintervals. The load range is bounded by knocking and misfire limits, identified in thiswork respectively by a maximum pressure rise rate of 10 bar/deg and a CoV in IMEP of5%. Experimental data for 300 consecutive cycles are collected at each test point afterallowing sufficient operation settling time. The results reported in this section representaverage values for the 300 cycles.
The results from these experiments show a clear potential for using the dual-fuel approachto extend the operating range at the three tested speeds. Figure 2 illustrates how gross in-dicated mean effective pressure (IMEPg) varies with varying equivalence and PRF ratios.Increasing the PRF octane rating extends the upper operation boundary and preserves theengine ability to deliver more work as equivalence ratio increases. The figure also showsthe advantage of varying the PRF ratio on the engine positive work at similar operatingconditions as well as the critical points when the change must take place to realize suchadvantage. Gross IMEP, which accounts only for the indicated work during the closed
4
Table 1: Engine specifications and valve timing information. Crank angles here are mea-sured relative to firing TDC.
Parameter ValueNumber of cylinders 1Operation cycle 4-strokeCombustion mode HCCINumber of valves 4Cylinder displacement (liters) 0.5Bore x Stroke (mm) 84 x 90Connecting rod length (mm) 159Crank radius (mm) 45Compression ratio 12:1Fuel delivery system PFICooling water temperature (oC) 90Lubrication oil temperature (oC) 90IVO (CAD) -356EVC (CAD) -352IVC (CAD) -156EVO (CAD) 170
volume, was used here to exclude pumping variations and facilitate comparison with sim-ulation results from the closed-volume model described in the next section.
The boundaries of operation at the tested speeds can be inferred from Figure 3 whichshows the readings of maximum pressure rise rate and variation in IMEP. To protect theengine from knocking effects, all test points were selected so that the maximum pressurerise rate didn’t exceed 10 bar/deg. The point at the highest equivalence ratio, beforereaching the set knocking limit, is chosen to represent the upper boundary of the operationenvelope. The lower boundary is determined by a maximum coefficient of variation inIMEP of 5%. The lowest equivalence ratio, before the variation coefficient exceeds 5%,is considered here as the lower boundary of the operation envelope at the respected speed.
Using the conventions described above, the engine operation envelopes for the differ-ent PRF ratios are constructed as shown in Figure 4. While the envelopes shrink as thePRF octane rating increases, they also move upward allowing the engine to operate athigher equivalence ratios and to produce more work. These effects generally becomemore prominent as the speed increases. For example, while changing the PRF octane rat-ing from 40 to 80 at 1200 rpm increases the equivalence range from 0.32 to 0.34 and theIMEPg by about 1 bar, the equivalence ratio increases from 0.31 to 0.38 and the IMEPgincreases by about 2 bars at 1800 rpm.
5
Page 3 of 21
ENGINE EXPERIMENTS
The experiments for this work were carried out on a half-liter single-cylinder research engine running in HCCI mode. The specifications of this engine are listed in Table 1. The engine has a pent-roof combustion chamber fitted with four valves. Fuel is delivered intermittently via a port fuel injector located at the end of the intake port just above the intake valves. Intake air passes through a conditioner to adjust its temperature, humidity, and pressure as necessary.
All experiments were performed at a boosted intake pressure of 1.5 bar and an intake air temperature of 75oC. The engine was fueled with primary reference fuels at three different volume ratios: PRF40 (i.e. 40% iso-octane and 60% n-heptane), PRF60 and PRF80. Load sweeps were performed at three constant speeds of 1200, 1500 and 1800 rpm. Upper and lower bounds of possible load range are first identified at each test condition, and then test points are selected at reasonably-distanced intervals. The load range is bounded by knocking and misfire limits, identified in this work respectively by a maximum pressure rise rate of 10 bar/deg and a CoV in IMEP of 5%.
Figure 2 – Gross mean effective pressure for different PRF ratios and operating conditions. Increasing the PRF octane rating extends the upper operation boundary and preserves the engine ability to deliver more work as equivalence ratio increases.
The results from these experiments show a clear potential for using the dual-fuel approach to extend the operating range at the three tested speeds. Figure 2 illustrates how gross indicated mean effective pressure (IMEPg) varies with varying equivalence and PRF ratios. Increasing the PRF octane rating extends the upper operation boundary and preserves the engine ability to deliver more work as equivalence ratio increases. The figure also shows the advantage of varying the PRF ratio on the engine positive work at similar operating conditions as well as the critical points when the change must take place to realize such advantage. Gross IMEP, which accounts only for the indicated work during the closed volume, was used here to exclude pumping variations and facilitate comparison with simulation results from the closed-volume model described in the next section.
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.40
2
4
6
8
10IM
EP
g (b
ar)
1200 rpm
PRF40PRF60PRF80
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.40
2
4
6
8
10
IME
Pg
(bar
)
1500 rpm
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.40
2
4
6
8
10
IME
Pg
(bar
)
Φ
1800 rpm
Figure 2: Gross mean effective pressure for different PRF ratios and operating condi-tions. Increasing the PRF octane rating extends the upper operation boundaryand preserves the engine ability to deliver more work as the equivalence ratioincreases. Error bars on experimental data are based on ±1 standard deviation
6
Page 4 of 21
Figure 3 – Maximum pressure rise rate and variation of mean effective pressure for all test points. The point at the highest equivalence ratio, before hitting the set knocking limit of 10 bar/deg, is chosen to represent the upper boundary of the operation
envelop. The lower boundary is determined by a maximum coefficient of variation in IMEP of 5%. Error bars on maximum pressure rise values represent ±1 standard deviation.
The boundaries of operation at the tested speeds can be inferred from Figure 3 which shows the readings of maximum pressure rise rate and variation in IMEP. To protect the engine from knocking effects, all test points were selected so that the maximum pressure rise rate didn’t exceed 10 bar/deg. The point at the highest equivalence ratio, before hitting the set knocking limit, is chosen to represent the upper boundary of the operation envelop. The lower boundary is determined by a maximum coefficient of variation in IMEP of 5%. The lowest equivalence ratio, before the variation coefficient exceeds 5%, is considered here as the lower boundary of the operation envelop at the respected speed.
Using the conventions described above, the engine operation envelops for the different PRF ratios are constructed as shown in Figure 4. While the envelops shrink as the PRF octane rating increases, they also move upward allowing the engine to operate at higher equivalence ratios and to produce more work. These effects generally become more prominent as the speed increases. For example, while changing the PRF octane rating from 40 to 80 at 1200 rpm increases the equivalence range from 0.32 to 0.34 and the IMEPg by about 1 bar, the equivalence ratio increases from 0.31 to 0.38 and the IMEPg increases by about 2 bars at 1800 rpm.
0.1 0.2 0.3 0.40
0.1
0.2
0.3
0.4
0.5IM
EP
CoV
1200 rpm
PRF40PRF60PRF80
0.1 0.2 0.3 0.40
2
4
6
8
10
12
Max
. Pre
ssur
e R
ise
(bar
/deg
) 1200 rpm
0.1 0.2 0.3 0.40
0.1
0.2
0.3
0.4
0.5
IME
P C
oV
1500 rpm
0.1 0.2 0.3 0.40
2
4
6
8
10
12
Max
. Pre
ssur
e R
ise
(bar
/deg
) 1500 rpm
0.1 0.2 0.3 0.40
0.1
0.2
0.3
0.4
0.5
IME
P C
oV
Φ
1800 rpm
0.1 0.2 0.3 0.40
2
4
6
8
10
12
Max
. Pre
ssur
e R
ise
(bar
/deg
)
Φ
1800 rpm
Figure 3: Maximum pressure rise rate and variation of mean effective pressure for alltest points. The point at the highest equivalence ratio, before reaching theset knocking limit of 10 bar/deg, is chosen to represent the upper boundaryof the operation envelope. The lower boundary is determined by a maximumcoefficient of variation in IMEP of 5%. Error bars on maximum pressure risevalues represent ±1 standard deviation.
7
Page 5 of 21
Figure 4 - Engine operation envelops for different PRF ratios. While the envelops shrink as the PRF octane rating increases, they also move upward allowing the engine to operate at higher equivalence ratios and to produce more work. These effects generally
become more prominent as the speed increases. Error bars on IMEPg values represent ±1 standard deviation.
MODEL SETUP AND OPTIMIZATION
The results from engine experiments showed a clear potential for the dual-fuel approach to expand the HCCI engine operation window. In order to investigate this potential further, a stochastic reactor (SRM) model (see [9-12]) is used after being optimized for the current engine.
The model uses a fixed residual gas composition extracted from simulation results for an arbitrary but representative operating condition. The ratio of residual gas is determined through the optimization step discussed below. Initial pressures and temperatures at IVC for wide operating ranges, including and extending beyond experimental ranges, were estimated using a GT-Power gas dynamics model developed and calibrated for the current engine. Figure 5 shows the obtained maps of IVC temperature and pressure at 1200 rpm for different PRF octane ratings and equivalence ratios. The maps for different speeds have been integrated within the model to allow simulation of transient operation. These maps indicate that both equivalence ratio and speed had a moderate effect on IVC temperature, but a weak effect on IVC pressure. PRF ratio, on the other hand had little or no effect on IVC pressure and temperature.
1200 1400 1600 18001
2
3
4
5
6
7
8
IME
Pg
(bar
)
Engine Speed (rpm)
1200 1400 1600 18000.15
0.2
0.25
0.3
0.35
0.4Φ
PRF80
PRF60
PRF40
PRF60
PRF40
PRF80
Figure 4: Engine operation envelopes for different PRF ratios. While the envelopesshrink as the PRF octane rating increases, they also move upward allowingthe engine to operate at higher equivalence ratios and to produce more work.These effects generally become more prominent as the speed increases. Errorbars on IMEPg values represent ±1 standard deviation.
8
3 Model Setup and Optimization
The results from engine experiments showed a clear potential for the dual-fuel approach toexpand the HCCI engine operation window. In order to investigate this potential further,a stochastic reactor (SRM) model (see [2, 6, 7, 14]) is used after being optimized for thecurrent engine.
The model uses a fixed residual gas composition extracted from simulation results for anarbitrary but representative operating condition. The ratio of residual gas is determinedthrough the optimization step discussed below. Initial pressures and temperatures at IVCfor wide operating ranges, including and extending beyond experimental ranges, were es-timated using a GT-Power gas dynamics model developed and calibrated for the currentengine. Figure 5 shows the obtained maps of IVC temperature and pressure at 1200 rpmfor different PRF octane ratings and equivalence ratios. The maps for different speedshave been integrated within the model to allow simulation of transient operation. Thesemaps indicate that both equivalence ratio and speed had a moderate effect on IVC tem-perature, but a weak effect on IVC pressure. PRF ratio, on the other hand had little or noeffect on IVC pressure and temperature.
The IVC maps incorporate steady-state temperature and pressure values. This preventsthe model from considering the thermal inertia effects during transients. To minimize theimpact of this on the simulation results, a sufficient settling time will need to be allowedbetween transient steps. In the current study, a 20-cycle settling time is used in all transientcontrol simulations.
The model utilizes a highly-reduced PRF mechanism that has only 33 species and 38reactions. Description of this mechanism is given in [17]. The use of this small mecha-nism was motivated by the need for speedy computations that allow extensive simulationof transient, multi-cycle HCCI engine operation. Two more detailed PRF mechanisms(with 767 and 157 species respectively) were considered initially, and they both performedsignificantly better than the 33-species mechanism in terms of predicting key engine re-sponses. However, the improved performance of the more detailed mechanisms camewith a higher computational time that was deemed not practical for the purpose of thisstudy.
The use of the reduced mechanism along with the other model simplifications (such asusing a fixed residual gas ratio and composition) necessitated an optimization step to en-hance the model predictivity of engine responses relevant to the current control study.These responses, which constitute the optimization objectives, are the cylinder pressurehistory, the pressure rise rate, and the gross indicated work (or IMEPg). For this purpose,the model was optimized using a multi-objective genetic algorithm, where the best solu-tions are searched for using the principles of evolution. Further information about thisapproach can be found in [3].
The optimization variable set consisted of the three coefficients of Arrhenius equation(i.e. the pre-exponential factor, the temperature exponent and the activation energy) forall reactions, as well as four selected variables from the SRM (average wall tempera-ture, residual gas fraction, turbulent mixing time and heat transfer coefficient). Objectivefunctions for the selected responses were formulated in the form of sum of squared differ-
9
Page 6 of 21
Figure 5 – IVC temperature and pressure maps for the engine speed of 1200 rpm. The maps (including those for other speeds) indicate that both equivalence ratio and speed had a moderate effect on IVC temperature, but a weak effect on IVC pressure. PRF
ratio had little or no effect on IVC pressure and temperature.
The model utilizes a highly reduced PRF mechanism that has only 33 species and 38 reactions. Description of this mechanism is given in [13]. The use of this small mechanism was motivated by the need for speedy computations that allow extensive simulation of transient, multi-cycle HCCI engine operation. The limited chemistry representation along with the other model simplifications (such as using a fixed residual gas ratio and composition) necessitated an optimization step to enhance the model predictivity of engine responses relevant to the current control study. These responses, which constitute the optimization objectives, are the cylinder pressure history, the pressure rise rate, and the gross indicated work (or IMEPg). For this purpose, the model was optimized using a multi-objective genetic algorithm, where the best solutions are searched for using the principles of evolution. Further information about this approach can be found in [14].
The optimization variable set consisted of the three coefficients of Arrhenius equation (i.e. the pre-exponential factor, the temperature exponent and the activation energy) for all reactions, as well as four selected variables from the SRM (average wall temperature, residual gas fraction, turbulent mixing time and heat transfer coefficient). Objective functions for the selected responses were formulated in the form of sum of squared differences between experiment and model values, and the optimization was carried out against results from 12 simultaneous experiments run at different operating conditions.
Optimization results are shown in Figures 6 and 7. Figure 6 shows the history of objective function values throughout the optimization run, which involved more than 1500 iterations (with each iteration taking about 10 minutes in average on a fairly fast machine). The optimization run was terminated after the compound objective function had fairly stabilized. The set of optimum solutions were then determined by constructing the Pareto front from the solution set, as shown in Figure 7. The selection of a particular solution from the Pareto set is down to the user and based on the weights he assigns to the different objectives. The results showed a low sensitivity of maximum pressure rise rate to changes in the optimization variables, and all Pareto solutions gave almost the same value for maximum pressure rise rate objective. As such, a Pareto front solution that gives the best compromise between cylinder pressure history and gross indicated work was selected.
0.20.3
0.40.5
0
50
100360
362
364
366
Equivalence RatioOctane Number
IVC
Tem
pera
ture
(K)
0.20.3
0.40.5
0
50
100
1.572
1.5722
1.5724
Equivalence RatioOctane Number
IVC
Pre
ssur
e (b
ar)
Figure 5: IVC temperature and pressure maps for the engine speed of 1200 rpm. Themaps (including those for other speeds) indicate that both equivalence ratioand speed had a moderate effect on IVC temperature, but a weak effect on IVCpressure. PRF ratio had little or no effect on IVC pressure and temperature.
10
Page 7 of 21
Figure 6 – History of objective function values throughout the multi-objective optimization run. A termination criterion was set based on settling of compounded objective function. The results show a low sensitivity of maximum pressure rise rate to changes in
the optimization variables.
Figure 7 – Solution space with Pareto front set of optimum solutions. All Pareto solutions gave almost the same value for maximum pressure rise rate objective. As such, a Pareto front solution that gives the best compromise between cylinder pressure
history and gross indicated work was selected.
0 200 400 600 800 1000 1200 1400 16000
0.5
1
Pre
ssur
e O
bjFu
nctio
n
0 200 400 600 800 1000 1200 1400 16000
0.5
1
max
PR
R O
bjFu
nctio
n
0 200 400 600 800 1000 1200 1400 16000
0.5
1
IME
Pg
Obj
Func
tion
0 200 400 600 800 1000 1200 1400 16000
0.5
1
Iteration
Com
pund
ed O
bjFu
nctio
n
00.2
0.40.6
0.81
0
0.5
10
0.2
0.4
0.6
0.8
1
Pressure ObjFunmaxPRR ObjFun
IME
Pg
Obj
Fun
Solution setPareto front set
Selected solution
Figure 6: History of objective function values throughout the multi-objective optimizationrun. A termination criterion was set based on settling of compounded objectivefunction. The results show a low sensitivity of maximum pressure rise rate tochanges in the optimization variables.
11
Page 7 of 21
Figure 6 – History of objective function values throughout the multi-objective optimization run. A termination criterion was set based on settling of compounded objective function. The results show a low sensitivity of maximum pressure rise rate to changes in
the optimization variables.
Figure 7 – Solution space with Pareto front set of optimum solutions. All Pareto solutions gave almost the same value for maximum pressure rise rate objective. As such, a Pareto front solution that gives the best compromise between cylinder pressure
history and gross indicated work was selected.
0 200 400 600 800 1000 1200 1400 16000
0.5
1
Pre
ssur
e O
bjFu
nctio
n
0 200 400 600 800 1000 1200 1400 16000
0.5
1
max
PR
R O
bjFu
nctio
n
0 200 400 600 800 1000 1200 1400 16000
0.5
1
IME
Pg
Obj
Func
tion
0 200 400 600 800 1000 1200 1400 16000
0.5
1
Iteration
Com
pund
ed O
bjFu
nctio
n
00.2
0.40.6
0.81
0
0.5
10
0.2
0.4
0.6
0.8
1
Pressure ObjFunmaxPRR ObjFun
IME
Pg
Obj
Fun
Solution setPareto front set
Selected solution
Figure 7: Solution space with Pareto front set of optimum solutions. All Pareto solutionsgave almost the same value for maximum pressure rise rate objective. As such,a Pareto front solution that gives the best compromise between cylinder pres-sure history and gross indicated work was selected.
12
Page 8 of 21
To check the performance of the selected solution, several comparisons with experimental data for indicated work rate (which reflects cylinder pressure/volume changes), maximum pressure rise rate, and gross indicated work were made. These comparisons are shown in Figures 8-13. A very good fit can be seen across the tested engine operating window. There is a consistent over prediction of pressure value during the expansion process (as can be inferred from Figures 8 and 9), which in turn leads to consistently higher values of predicted gross indicated work (as can be seen in Figures 10 and 11). This resulting accumulated error in IMEPg is in the range of 20%.
Figure 8 – Comparing model predicted work rate with experimental data at 1200 rpm. A very good fit can be seen across the tested engine operating window. However, there is a slight, but consistent, over prediction of pressure value during the expansion
process. The accumulation of this error however leads to a significant, but fairly consistent, deviation in predicted IMEPg values.
-200 0 200-10
-5
0
5
10
15PRF40, Φ=0.32
-200 0 200-10
-5
0
5
10
15PRF40, Φ=0.19
W
ork
Rat
e (N
m/d
eg)
-200 0 200-10
-5
0
5
10
15PRF60, Φ=0.32
-200 0 200-10
-5
0
5
10
15
Wor
k R
ate
(Nm
/deg
)
PRF60, Φ=0.21
-200 0 200-10
-5
0
5
10
15
CAD (deg)
PRF80, Φ=0.35
-200 0 200-10
-5
0
5
10
15
CAD (deg)
Wor
k R
ate
(Nm
/deg
)
PRF80, Φ=0.29
ExperimentOptimized model
Figure 8: Comparing model predicted work rate with experimental data at 1200 rpm. Avery good fit can be seen across the tested engine operating window. However,there is a slight, but consistent, over prediction of pressure value during theexpansion process. The accumulation of this error leads to a significant, butfairly consistent, deviation in predicted IMEPg values.
13
Page 9 of 21
Figure 9 – Comparing model predicted work rate with experimental data at 1500 rpm. As in the case of 1200 rpm, the fit with experiments is excellent despite the slight over prediction of pressure value during the expansion process.
-200 0 200-10
-5
0
5
10
15
Wor
k R
ate
(Nm
/deg
)PRF40, Φ=0.20
-200 0 200-10
-5
0
5
10
15PRF40, Φ=0.33
-200 0 200-10
-5
0
5
10
15PRF60, Φ=0.32
-200 0 200-10
-5
0
5
10
15
Wor
k R
ate
(Nm
/deg
)
PRF60, Φ=0.23
-200 0 200-10
-5
0
5
10
15
CAD (deg)
Wor
k R
ate
(Nm
/deg
)
PRF80, Φ=0.34
-200 0 200-10
-5
0
5
10
15
CAD (deg)
PRF80, Φ=0.36
ExperimentOptimized model
Figure 9: Comparing model predicted work rate with experimental data at 1500 rpm. Asin the case of 1200 rpm, the fit with experiments is good despite the consistentover prediction of pressure value during the expansion process.
14
Page 10 of 21
Figure 10 – Comparing model predicted gross IMEP with experimental data at 1200 rpm. The model consistently over predicts the gross work, with an estimated average error of about 20%. Error bars on experimental data are based on ±1 standard deviation.
0.1 0.15 0.2 0.25 0.3 0.35 0.4
2
4
6
8IM
EP
g (b
ar)
Exper. PRF40SRM PRF40,
0.1 0.15 0.2 0.25 0.3 0.35 0.4
2
4
6
8
IME
Pg
(bar
)
Exper. PRF60SRM PRF60
0.1 0.15 0.2 0.25 0.3 0.35 0.4
2
4
6
8
φ
IME
Pg
(bar
)
Exper. PRF80SRM PRF80
Figure 10: Comparing model predicted gross IMEP with experimental data at 1200 rpm.The model consistently over predicts the gross work, with an estimated aver-age error of about 20%. Error bars on experimental data are based on ±1standard deviation.
15
Page 11 of 21
Figure 11 – Comparing model predicted gross IMEP with experimental data at 1500 rpm. Again, we see here a consistent over prediction of gross work across the tested range. Error bars on experimental data are based on ±1 standard deviation.
0.1 0.15 0.2 0.25 0.3 0.35 0.4
2
4
6
8
Gro
ss IM
EP
(bar
)
Exper. PRF40SRM PRF40
0.1 0.15 0.2 0.25 0.3 0.35 0.4
2
4
6
8
Gro
ss IM
EP
(bar
)
Exper. PRF60SRM PRF60
0.1 0.15 0.2 0.25 0.3 0.35 0.4
2
4
6
8
φ
Gro
ss IM
EP
(bar
)
Exper. PRF80SRM PRF80
Figure 11: Comparing model predicted gross IMEP with experimental data at 1500 rpm.Again, a consistent over prediction of gross work across the tested range isseen here. Error bars on experimental data are based on ±1 standard devia-tion.
16
Page 12 of 21
Figure 12 – Comparing model predicted maximum pressure rise rate with experimental data at 1200 rpm. Error bars on experimental data are based on ±1 standard deviation.
0.1 0.15 0.2 0.25 0.3 0.35 0.40
5
10
Max
. PR
R (b
ar/d
eg)
Exper. PRF40SRM PRF40
0.1 0.15 0.2 0.25 0.3 0.35 0.40
5
10
Max
. PR
R (b
ar/d
eg)
Exper. PRF60SRM PRF60
0.1 0.15 0.2 0.25 0.3 0.35 0.40
5
10
φ
Max
. PR
R (b
ar/d
eg)
Exper. PRF80SRM PRF80
Figure 12: Comparing model predicted maximum pressure rise rate with experimentaldata at 1200 rpm. Error bars on experimental data are based on ±1 standarddeviation.
17
Page 13 of 21
Figure 13 – Comparing model predicted maximum pressure rise rate with experimental data at 1500 rpm. Error bars on experimental data are based on ±1 standard deviation.
RESULTS & DISCUSSIONS
In order to assess the dual-fuel control approach in transient operating conditions and beyond the limits tested in the experiments, a dynamic control study was conducted using a simulation code that incorporates the optimized HCCI model (explained in the previous section) and allows for multi-cycle transient operation and closed-loop control of fuel ratio. This setup simulates a dual-fuel HCCI engine similar to that depicted in Figure 1, where the operation range is extended as possible by varying the ratio of the low-reactivity and high-reactivity fuels. Specifically, the fuels used here are the n-heptane (which has an octane number of 0) and the iso-octane (which has an octane number of 100). Varying the ratio between these two primary reference fuels affects the fuel reactivity and thus the heat release rate, which eventually affects the engine operability range.
An appropriate control strategy is needed to ensure that optimum fuel ratio is delivered for each operation condition. A PID controller is typically used for this purpose (see [3,12,15,16,17]). Such controller, however, requires prior knowledge about the system responses and the optimum target values of controlled quantities. Also, the non-linearity of system responses (which exist in the engine case) will normally require careful tuning of the controller and non-trivial gain scheduling. This has been avoided in the current study by using a search algorithm that looks for a fuel ratio that maximizes the gross indicated work. Such control strategy, as the results suggest, appears to be sufficient to control pressure rise rate and combustion phasing while not needing any prior knowledge or preset information about them. It also ensures that the engine is always delivering maximum work at each operating condition. This is in a sense analogous to the use of maximum brake torque timing in spark-ignition engines.
The search for optimum PRF ratio is carried out in the current model using MATLAB’s “fminbnd” algorithm, which is based on golden section search and parabolic interpolation [18]. It works by successively narrowing the range inside which the sought minimum or maximum is known to exist. As will be seen below, this algorithm proved to be sufficiently efficient for the purpose of this study. It normally took 2-6 iterations to locate the optimum PRF ratio.
0.1 0.15 0.2 0.25 0.3 0.35 0.40
5
10
Max
. PR
R (b
ar/d
eg)
Exper. PRF40SRM PRF40
0.1 0.15 0.2 0.25 0.3 0.35 0.40
5
10
Max
. PR
R (b
ar/d
eg)
Exper. PRF60SRM PRF60
0.1 0.15 0.2 0.25 0.3 0.35 0.40
5
10
φ
Max
. PR
R (b
ar/d
eg)
Exper. PRF80SRM PRF80
Figure 13: Comparing model predicted maximum pressure rise rate with experimentaldata at 1500 rpm. Error bars on experimental data are based on ±1 standarddeviation.
18
ences between experiment and model values, and the optimization was carried out againstresults from 12 simultaneous experiments run at different operating conditions.
Optimization results are shown in Figures 6 and 7. Figure 6 shows the history of objectivefunction values throughout the optimization run, which involved more than 1500 itera-tions (with each iteration taking about 10 minutes in average on a fairly fast machine).The optimization run was terminated after the compound objective function had fairlystabilized. The set of optimum solutions were then determined by constructing the Paretofront from the solution set, as shown in Figure 7. The selection of a particular solutionfrom the Pareto set is down to the user and based on the weights he assigns to the differentobjectives. The results showed a low sensitivity of maximum pressure rise rate to changesin the optimization variables, and all Pareto solutions gave almost the same value formaximum pressure rise rate objective. As such, a Pareto front solution that gives the bestcompromise between cylinder pressure history and gross indicated work was selected.
To check the performance of the selected solution, several comparisons with experimentaldata for indicated work rate (which reflects cylinder pressure/volume changes), maximumpressure rise rate, and gross indicated work were made. These comparisons are shown inFigures 8-13. A very good fit can be seen across the tested engine operating window.There is a consistent over prediction of pressure value during the expansion process (ascan be inferred from Figures 8 and 9), which in turn leads to consistently higher values ofpredicted gross indicated work (as can be seen in Figures 10 and 11). This resulting accu-mulated error in IMEPg is in the range of 20%. The specific causes of this over-predictionhave not been examined in this work. However, the original mechanism (i.e. before op-timization) exhibited significant over-prediction of pressure throughout the reactive partof the closed-volume cycle. Over-prediction during the expansion process persisted evenafter the optimization step. This could be simply an indication of a deficiency in the chem-istry, but could also indicate that more emphasis should have been put on the expansionside when formulating the optimization objective function of cylinder pressure . The cur-rent function accounts for a total of five points on the pressure curve with only two on theexpansion side (more discussion on how the objective functions were formulated can befound in [3]).
4 Results and Discussions
In order to assess the dual-fuel control approach in transient operating conditions andbeyond the limits tested in the experiments, a dynamic control study was conducted usinga simulation code that incorporates the optimized HCCI model (explained in the previoussection) and allows for multi-cycle transient operation and closed-loop control of fuelratio. This setup simulates a dual-fuel HCCI engine similar to that depicted in Figure 1,where the operation range is extended as possible by varying the ratio of the low-reactivityand high-reactivity fuels. Specifically, the fuels used here are the n-heptane (which hasan octane number of 0) and the iso-octane (which has an octane number of 100). Varyingthe ratio between these two primary reference fuels affects the fuel reactivity and thus theheat release rate, which eventually affects the engine operability range.
An appropriate control strategy is needed to ensure that optimum fuel ratio is delivered
19
for each operation condition. A PID controller is typically used for this purpose (see[2, 5, 9, 10, 15]). Such controller, however, requires prior knowledge about the systemresponses and the optimum target values of controlled quantities. Also, the non-linearityof system responses (which exist in the engine case) will normally require careful tuningof the controller and non-trivial gain scheduling. This has been avoided in the currentstudy by using a search algorithm that looks for a fuel ratio that maximizes the grossindicated work. Such control strategy, as the results suggest, appears to be sufficient tocontrol pressure rise rate and combustion phasing while not needing any prior knowledgeor preset information about them. It also ensures that the engine is always deliveringmaximum work at each operating condition. This is in a sense analogous to the use ofmaximum brake torque timing in spark-ignition engines.
The search for optimum PRF ratio is carried out in the current model using MATLAB’s”fminbnd” algorithm, which is based on golden section search and parabolic interpolation[1]. This algorithm proved to be efficient as it normally took only 2-6 cycles to locatethe optimum PRF ratio. However, because the algorithm works by cornering the solutionwithin increasingly smaller intervals, large spikes in engine operation are expected duringthe initial intervals (as can be seen in the current results). Although different search algo-rithms may perform differently, the need for real time search could potentially make theapproach very difficult to implement on a real engine.
Figures 14-17 show the results for running the model at engine speeds of 1200, 1500, and1800 rpm. In all cases, the model simulates an HCCI engine subjected to an equivalenceratio sweep and where the controller at each condition seeks to maximize the gross indi-cated work (IMEPg) by varying the iso-octane between 0 and 100% in a mixture of iso-octane and n-heptane. Each equivalence ratio step lasts for 20 full cycles, during whichthe controller should have located the optimum PRF ratio and operation has stabilized.
Figure 14 demonstrates how the used control strategy ensures delivery of maximum workat each operation condition. The case where a controller is used is compared to three fixedPRF ratio fuels (PRF20, PRF50, and PRF80). The controlled case was consistently ableto deliver maximum work at all conditions. The performance of other fuels varied acrossthe sweep, and in most cases the delivered work was by far lower than the controlled case.
The benefit of dual-fuel control approach on HCCI operation range is demonstrated inFigure 15. Here, the simulation results at 1200 rpm show how using a fixed PRF ratiofuel adversely limits the operating range of this HCCI engine. The operating limits are setbased on maximum pressure rise rate, where the upper limit (i.e. the knocking limit) is atabout 10 bar/deg and the lower limit (i.e. the misfire limit) is about 2 bar/deg (this agreeswell with the experimental data presented in Figure 3 where 5% IMEP CoV correspondsto about 2 bar/deg). While fixed PRF ratio fuels resulted in significantly limited operatingranges, varying the PRF octane rating between 0-100 kept the HCCI combustion withinthe set limits almost over the whole sweep. It can also be seen that, without any priorknowledge about optimum combustion phasing, the employed control strategy kept theCA50 under control almost over the whole sweep.
Similar results have been obtained for the engine speeds of 1500 and 1800 rpm. Theseresults are shown respectively in Figures 16 and 17. By combining the results for thedifferent speeds, a map of optimum PRF ratio for the different speeds and equivalence
20
Page 14 of 21
Figure 14 –Dynamic simulation of octane ratio control to maintain maximum work at 1200 rpm. Here, the controlled case is compared to three fixed PRF ratio fuels (PRF20, PRF50, and PRF80). The controlled case was consistently able to deliver
maximum work at all conditions. Performance of other fuels varied across the sweep, and in most cases the delivered work was by far lower than the controlled case.
Figures 14-17 show the results for running the model at engine speeds of 1200, 1500, and 1800 rpm. In all cases, the model simulates an HCCI engine subjected to an equivalence ratio sweep and where the controller at each condition seeks to maximize the gross indicated work (IMEPg) by varying the iso-octane between 0 and 100% in a mixture of iso-octane and n-heptane. Each equivalence ratio step lasts for 20 full cycles, during which the controller should have located the optimum PRF ratio and operation has stabilized.
Figure 14 demonstrates how the used control strategy ensures delivery of maximum work at each operation condition. The case where a controller is used is compared to three fixed PRF ratio fuels (PRF20, PRF50, and PRF80). The controlled case was consistently able to deliver maximum work at all conditions. The performance of other fuels varied across the sweep, and in most cases the delivered work was by far lower than the controlled case.
0 50 100 150 200 250 3000
0.2
0.4Φ
0 50 100 150 200 250 3000
0.5
1
Oct
ane
Rat
io
0 50 100 150 200 250 3000
2
4
6
8
10
Gro
ss IM
EP
(bar
)
Cycle
With controlPRF50PRF80PRF20
Figure 14: Dynamic simulation of octane ratio control to maintain maximum work at1200 rpm. Here, the controlled case is compared to three fixed PRF ratio fuels(PRF20, PRF50, and PRF80). The controlled case was consistently able todeliver maximum work at all conditions. Performance of other fuels variedacross the sweep, and in most cases the delivered work was noticeably lowerthan the controlled case.
21
Page 15 of 21
Figure 15 –Effects of octane ratio control on CA50 and maximum pressure rise rate at 1200 rpm. While fixed PRF ratio fuels resulted in significantly limited operating ranges, varying the PRF octane rating between 0-100 kept the HCCI combustion within
the set limits almost over the whole sweep.
The benefit of dual-fuel control approach on HCCI operation range is demonstrated in Figure 15. Here, the simulation results at 1200 rpm show how using a fixed PRF ratio fuel adversely limits the operating range of this HCCI engine. The operating limits are set based on maximum pressure rise rate, where the upper limit (i.e. the knocking limit) is at about 10 bar/deg and the lower limit (i.e. the misfire limit) is about 2 bar/deg (this agrees well with the experimental data presented in Figure 3 where 5% IMEP CoV corresponds to about 2 bar/deg). While fixed PRF ratio fuels resulted in significantly limited operating ranges, varying the PRF octane rating between 0-100 kept the HCCI combustion within the set limits almost over the whole sweep. It can also be seen that, without any prior knowledge about optimum combustion phasing, the employed control strategy kept the CA50 under control almost over the whole sweep.
0 50 100 150 200 250 3000
0.5
Φ
0 50 100 150 200 250 300-40
-20
0
20
CA
50
0 50 100 150 200 250 3000
0.5
1
Oct
ane
Rat
io
0 50 100 150 200 250 3000
5
10
15
20
Cycle
Max
. PR
R (b
ar/d
eg)
With controlPRF50PRF80PRF20
Upper limit
Lower limit
Figure 15: Effects of octane ratio control on CA50 and maximum pressure rise rate at1200 rpm. While fixed PRF ratio fuels resulted in significantly limited oper-ating ranges, varying the PRF octane rating between 0-100 kept the HCCIcombustion within the set limits almost over the whole sweep.
22
ratios can be constructed. This map, shown in Figure 18, can now replace the controllerfor further investigations within the operating ranges and conditions considered in thisstudy.
The HCCI operating envelope as predicted by the model with PRF ratio control and workmaximization objective is compared to those operating envelopes obtained from experi-ments with fixed PRF octane rating of 40, 60, and 80. This comparison is shown in Figure19. These envelopes, as mentioned before, represent the possible operating windows be-tween misfire and knocking limits. It should be noted that the lower part of the figurerepresents the maximum gross work when optimum PRF ratio is used at each equivalenceratio. The size of the operating envelope is therefore better represented by the equivalenceratio map on the upper part of the figure.
In terms of equivalence ratio and gross indicated work (IMEPg), the predicted expansionin the operating windows was significant especially with respect to the upper operatinglimit. For example, increasing the octane rating from 80 to 100 at 1200 rpm allowed for anincrease in the equivalence ratio range from 0.34 to 0.42. This in turn increased the grossindicated work by about 50% (that is with the consideration of the error in the modelprediction of gross indicated work). In general, the results indicate that the dual-fuelapproach, using iso-octane and n-heptane, could potentially allow an HCCI load rangeequivalent to about 60% of a typical boosted spark-ignition engine. However, as the studyonly considered a narrow speed range, more work is needed to examine this potential athigher speeds.
5 Summary and Conclusions
A dual-fuel approach using primary reference fuels to control combustion in HCCI engineis investigated in this work both experimentally and through a simulation study. Experi-ments were performed on a single-cylinder research engine fueled with different ratios ofprimary reference fuels and operated at different speed and load conditions.
The results from these experiments showed a clear potential for the dual-fuel approach toextend the HCCI engine operating window. Engine operation envelopes constructed fromthe results showed a significant effect of PRF octane rating on HCCI operation. While theenvelopes shrink as the PRF octane rating increases, they also allow the engine to operateat higher equivalence ratios and to produce more work. These effects generally becomemore prominent as the speed increases. The results also showed a clear advantage ofselecting an optimum PRF ratio on the engine positive work for each operating condition.
In order to assess the dual-fuel control approach in transient operating conditions andbeyond the limits tested in the experiments, a dynamic control study was conducted usinga simulation code that incorporates a stochastic reactor HCCI model and allows for multi-cycle transient operation and closed-loop control of fuel ratio.
The HCCI model utilizes a highly reduced PRF mechanism that has only 33 species and38 reactions. This limited chemistry representation along with the other simplifications(such as using a fixed EGR ratio and composition) necessitated an optimization step toenhance the model predictivity of those responses relevant to the current study. The model
23
Page 16 of 21
Figure 16 – Dynamic simulation of octane ratio control to maintain maximum work at 1500 rpm.
0 50 100 150 200 250 3000
0.5
Φ
0 50 100 150 200 250 300-20
0
20
CA
50
0 50 100 150 200 250 3000
0.5
1
Oct
ane
Rat
io
0 50 100 150 200 250 3000
5
10
15
20
25
Max
. PR
R (b
ar/d
eg)
0 50 100 150 200 250 3000
5
10
IME
Pg
(bar
)
Cycle
With controlNo control
ControlledFixed
Low er limit
Upper limit
Figure 16: Dynamic simulation of octane ratio control to maintain maximum work at1500 rpm.
24
Page 17 of 21
Figure 17 – Dynamic simulation of octane ratio control to maintain maximum work at 1800 rpm.
Similar results have been obtained for the engine speeds of 1500 and 1800 rpm. These results are shown respectively in Figures 16 and 17. By combining the results for the different speeds, a map of optimum PRF ratio for the different speeds and equivalence ratios can be constructed. This map, shown in Figure 18, can now replace the controller for further investigations within the operating ranges and conditions considered in this study.
The HCCI operating envelop as predicted by the model with PRF ratio control and work maximization objective is compared to those operating envelopes obtained from experiments with fixed PRF octane rating of 40, 60, and 80. This comparison is shown in Figure 19. These envelopes, as mentioned before, represent the possible operating windows between misfire and knocking limits. In terms of equivalence ratio and gross indicated work (IMEPg), the predicted expansion in the operating windows was significant especially with respect to the upper operating limit. For example, increasing the octane rating from 80 to 100 at 1200 rpm allowed for an increase in the equivalence ratio range from 0.34 to 0.42. This in turn increased the gross indicated work by about 50% (that is with the consideration of the error in the model prediction of gross indicated work).
0 50 100 150 200 250 3000
0.5
Φ
0 50 100 150 200 250 300-20
0
20
CA
50
0 50 100 150 200 250 3000
0.5
1
Oct
ane
Rat
io
0 50 100 150 200 250 3000
5
10
15
20
Max
. PR
R (b
ar/d
eg)
0 50 100 150 200 250 3000
5
10
IME
Pg
(bar
)
Cycle
ControlledFixed
With controlNo control
Upper limit
Low er limit
Figure 17: Dynamic simulation of octane ratio control to maintain maximum work at1800 rpm.
25
Page 18 of 21
Figure 18 – Optimum PRF octane rating map for different engine speeds and equivalence ratios.
Figure 19 – Expansion of HCCI engine operating envelop by varying the PRF octane rating from 0 to 100. The resulting envelop is compared to those obtained from engine experiments run with three constant PRF ratios. Predicted expansion in the operating
windows is significant especially with respect to the upper operating limit.
12001400
16001800
0.20.3
0.4
0.2
0.4
0.6
0.8
Engine Speed (rpm)Equivalence Ratio
Oct
ane
Num
ber
1200 1400 1600 18000
2
4
6
8
10
IME
Pg
(bar
)
Engine Speed (rpm)
1200 1400 1600 18000.1
0.2
0.3
0.4
0.5
Φ
PRF60(exp.)
PRF40(exp.)
PRF80(exp.)
PRF80(exp.)
PRF60(exp.)
PRF40(exp.)Model lower limit
Model upper limit
Model upper limit
Model lower limit
Figure 18: Optimum PRF octane rating map for different engine speeds and equivalenceratios.
was optimized using a multi-objective genetic algorithm, where the best solutions aresearched for using the principles of evolution. To check the performance of the optimizedmodel, several comparisons with experimental data for indicated work rate, maximumpressure rise rate, and gross indicated work were made.. Except for the gross indicatedwork where a large but consistent error persisted, a good fit with the experiment was seenacross a wide range of operating conditions.
The closed-loop control of fuel ratio employed in this study is based on a search algo-rithm, where the objective is to maximize the gross work rather than directly controllingthe combustion phasing to match preset values. The use of this search algorithm, whileproviding an efficient performance, allowed for avoiding the difficulties associated withtuning a PID controller. It also proved effective in controlling pressure rise rate and com-bustion phasing (measured by CA50) while not needing any prior knowledge or presetinformation about them. Above all, this control strategy ensured that the engine is alwaysdelivering maximum work at each operating condition. However, the need for real-timesearch could potentially make this approach very difficult to implement on a real engine.
While fixed PRF ratio fuels resulted in significantly limited operating ranges, varying thePRF octane rating between 0-100 kept the HCCI combustion within the set limits overa wide equivalence ratio range. Results from the dynamic control model allowed formapping the optimum PRF ratio for the different speeds and equivalence ratios, and forconstructing a full operating envelope for a dual-fuel case where PRF octane rating isvaried from 0 to 100. Comparing this model-predicted envelope to those obtained fromthe experiments provided a useful insight on the potential of the duel-fuel approach as ameans for controlling HCCI combustion and extending its operating range.
26
Page 18 of 21
Figure 18 – Optimum PRF octane rating map for different engine speeds and equivalence ratios.
Figure 19 – Expansion of HCCI engine operating envelop by varying the PRF octane rating from 0 to 100. The resulting envelop is compared to those obtained from engine experiments run with three constant PRF ratios. Predicted expansion in the operating
windows is significant especially with respect to the upper operating limit.
12001400
16001800
0.20.3
0.4
0.2
0.4
0.6
0.8
Engine Speed (rpm)Equivalence Ratio
Oct
ane
Num
ber
1200 1400 1600 18000
2
4
6
8
10
IME
Pg
(bar
)
Engine Speed (rpm)
1200 1400 1600 18000.1
0.2
0.3
0.4
0.5
Φ
PRF60(exp.)
PRF40(exp.)
PRF80(exp.)
PRF80(exp.)
PRF60(exp.)
PRF40(exp.)Model lower limit
Model upper limit
Model upper limit
Model lower limit
Figure 19: Expansion of HCCI engine operating envelope by varying the PRF octane rat-ing from 0 to 100. The resulting envelope is compared to those obtained fromengine experiments run with three constant PRF ratios. Predicted expansionin the operating windows is significant especially with respect to the upperoperating limit.
27
In general, the results indicated that the dual-fuel approach, using iso-octane and n-heptane, could potentially enable an HCCI load range equivalent to about 60% of a typi-cal boosted spark-ignition engine. However, as the study only considered a narrow speedrange, more work is needed to examine this potential at higher speeds. It is also importantto note that, while the use of primary reference fuels is convenient to generally representgasoline-like fuels, the performance of actual gasoline or gasoline-like fuels could be im-mensely different in HCCI engine. While the current study tries to provide a useful insighton the potential of the dual-fuel approach, the fuel effects will certainly need to be takeninto account when quantifying this potential.
6 Acknowledgement
The authors would like to thank Yoann Viollet, Marcin Plociniczak and Hassan Babiker ofSaudi Aramco’s R&D Center for their help in carrying out the HCCI engine experiments.
7 Definitions, Acronyms and Abbreviations
ATDC After top dead centreBTDC Before top dead centreCAD Crank angle degreeCoV Coefficient of variationEVC Exhaust valve closureEVO Exhaust valve openingHCCI Homogeneous charge compression ignitionIMEPg Gross indicated mean effective pressureIVC Intake valve closureIVO Intake valve openingPFI Port fuel injectionPID Proportional, integral, and differentialPRF Primary reference fuelSRM Stochastic reactor modelTDC Top dead centreΦ Equivalence ratio
28
References
[1] MATLAB 7 User Guide. The MathWorks Inc., 2009.
[2] A. Aldawood, S. Mosbach, and M. Kraft. HCCI combustion phasing transient con-trol by hydrogen-rich gas: Investigation using a fast detailed-chemistry full-cyclemodel. SAE Paper No. 2009-01-1134, 2009.
[3] A. Aldawood, S. Mosbach, M. Kraft, and A. Amer. Multi-objective optimization ofa kinetics-based hcci model using engine data. SAE Paper No. 2011-01-1783, 2011.
[4] M. Atkins and C. Koch. The effect of fuel octane and dilutent on homogeneouscharge compression ignition combustion. Proceedings of the Institution of Mechan-ical Engineers, Part D: Journal of Automobile Engineering, 219, 2005.
[5] J. Bengtsson, P. Strandh, R. Johansson, P. Tunestl, and B. Johansson. Closed-loopcombustion control of homogeneous charge compression ignition (hcci) engine dy-namics. International Journal of Adaptive Control and Signal Processing, 2004.
[6] A. Bhave, M. Kraft, L. Montorsi, and F. Mauss. Modelling a dual-fuelled multi-cylinder HCCI engine using a PDF based engine cycle simulator. SAE Paper No.2004-01-0561, 2004.
[7] A. Bhave, M. Kraft, F. Mauss, A. Oakley, and H. Zhao. Evaluating the EGR-AFRoperating range of a HCCI engine. SAE Paper No. 2005-01-0161, 2005.
[8] J. Cha, S. Kwon, and S. Park. An experimental and modelling study of the combus-tion and emission characteristics for gasoline-diesel dual-fuel engines. Proceedingsof the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineer-ing, 225, 2011.
[9] P. T. G. Haraldsson and B. Johansson. Hcci combustion phasing with closed-loopcombustion control using variable compression ratio. SAE Technical Paper 2003-01-1830, 2003.
[10] G. Haraldsson, P. Tunestl, and B. Johansson. Hcci closed-loop combustion controlusing fast thermal management in a multi cylinder engine. SAE Technical Paper2004-01-0943, 2004.
[11] S. Kokjohn, R. Hanson, D. Splitter, and R. Reitz. Experiments and modeling ofdual-fuel hcci and pcci combustion using in-cylinder fuel blending. SAE TechnicalPaper 2009-01-2647, 2009.
[12] X. Lu, W. Chen, Y. Hou, , and Z. Huang. Study on the ignition, combustion andemissions of hcci combustion engines fuelled with primary reference fuel. SAEPaper No. 2005-01-0155, 2005.
29
[13] J. Ma, X. L, L. Ji, and Z. Huang. An experimental study of hcci-di combustionand emissions in a diesel engine with dual fuel. International Journal of ThermalSciences, 47, 2008.
[14] S. Mosbach, M. Kraft, A. B. F. Mauss, J. Mack, and R. Dibble. Simulating a ho-mogenous charge compression ignition engine fuelled with a DEE/EtOH blend.SAE Paper No. 2006-01-1362, 2006.
[15] J. Olsson, P. Tunestl, and B. Johansson. Closed-loop control of an hcci engine. SAEPaper No. 2001-01-1031, 2001.
[16] Tanaka, F. Ayala, J. Keck, and J. Heywood. Two-stage ignition in hcci combustionand hcci control by fuels and additives. Combustion and Flame, 132, 2003.
[17] T. Tsurushima. A new skeletal prf kinetic model for hcci combustion. Proceedingsof the Combustion Institute, 2009.
[18] L. Xingcai, H. Yuchun, Z. Linlin, and H. Zhen. Experimental study on the auto-ignition and combustion characteristics in the homogeneous charge compression ig-nition (hcci) combustion operation with ethanol/n-heptane blend fuels by port injec-tion. Fuel, 85, 2006.
30