Fuel 89 (2010) 3416–3430

Contents lists available at ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

Optimization of a heavy-duty compression–ignition engine fueled with diesel

and gasoline-like fuels
Yu Shi *, Rolf D. Reitz
Engine Research Center, University of Wisconsin-Madison, 1500 Engineering Dr., Madison, WI 53706, USA

a r t i c l e i n f o a b s t r a c t

Article history: Optimal injection strategies for a heavy-duty compression–ignition engine fueled with diesel and gaso-
Received 17 December 2009 line-like fuels (#91 gasoline and E10) and operated under mid- and high-load conditions are investigated.
Received in revised form 15 February 2010 A state-of-the-art engine CFD tool with detailed fuel chemistry was used to evaluate the engine perfor-
Accepted 17 February 2010
mance and pollutant emissions. The CFD tools feature a recently developed efficient chemistry solver that
Available online 2 March 2010
allowed the optimization tasks to be completed in practical computer times. A Non-dominated Sorting
Genetic Algorithm II (NSGA II) was coupled with the CFD tool to seek optimal combinations of injection
Keywords:
system variables to achieve clean and efficient combustion. The optimization study identified several key
Optimization
Compression–ignition engine
parameters that influence engine performance. It was found that the fuel volatility and reactivity both
Diesel play important roles at the mid-load condition, while the high-load condition is less sensitive to the fuel
Gasoline reactivity. However, high volatility fuels, such as gasoline and E10, were found to be beneficial to fuel
E10 economy at high-load. The study indicates that with an optimized injection system gasoline-like fuels
are promising for heavy-duty CI engines due to their lower NOx and soot emissions and higher fuel econ-
omy compared to conventional diesel fuels. However, the high in-cylinder gas pressure rise rate associ-
ated with Partially Premixed Combustion of gasoline-like fuels can become problematic at high-load and
the low-load operating limit is also a challenge. Potential solutions are discussed based on the present
optimization results.
Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction a CI engine fueled with gasoline-like fuels has great potential to

meet more stringent emission standards while maintaining fuel
Due to its superior durability, drivability and fuel efficiency, the economy.
compression–ignition (CI) engine has found broad application in However, as also indicated by Shi et al. [1], in general CI engines
both heavy-duty and off-highway vehicles and equipment. In a fueled with gasoline-like fuels have higher in-cylinder gas pressure
comprehensive parametric study, Shi et al. [1] demonstrated that rise rate (thus engine noise) and Unburned Hydrocarbons (UHC)
emissions than those of conventional diesel engines. An injection
system that is calibrated or optimized for a CI engine fueled with
diesel requires modifications for gasoline-like fuels due to their
Abbreviations: AMC, Adaptive Multi-grid Chemistry; BDC, bottom dead center; distinct spray characteristics and fuel reactivity. Therefore, it is of
CA10, crank angle of 10% total heat release; CA50, crank angle of 50% total heat
much interest to compare optimal injection parameters of a CI en-
release; CA90, crank angle of 90% total heatrelease; CI, compression–ignition; CFD,
computational fluids dynamics; COSSO, Component Selection and Smoothing gine fueled with diesel and gasoline-like fuels, to provide guidance
Operator; DI, direct injection; EGR, exhaust gas recycle; EDAC, extended dynamic for engine design.
adaptive chemistry; EOI, end of injection; EVO, exhaust valve open; GDI, gasoline The increasing capacity of computers has enabled wide applica-
direct injection; GISFC, Gross Indicated Specific Fuel Consumption; HCCI, homoge- tion of CFD modeling to engine design. Many researchers have
neous charge compression–ignition; HD, heavy-duty; IMEP, indicated mean
effective pressure; IVC, intake valve closure; MOGA, Multi-Objective Genetic
shown that multi-dimensional CFD engine models are able to pre-
Algorithm; NMHC, non-methane hydrocarbons; NVO, Negative Valve Overlap; PPC, dict both combustion characteristics and pollution formation rea-
Partially Premixed Combustion; PPRR, Peak Pressure Rise Rate; PRF, Primary sonably well, and that the major trends of engine processes can
Reference Fuel; SOGA, Single Objective Genetic Algorithm; SOI, start of injection; be captured adequately [2,3]. Therefore, numerical simulation is
TDC, top dead center; UHC, Unburned Hydrocarbons.
becoming a reliable way to evaluate real engine performance and
* Corresponding author. Present address: Department of Chemical Engineering,
Massachusetts Institute of Technology, USA. emissions trends. This motivates the application of optimization
E-mail address: yushi@mit.edu (Y. Shi). methods to engine design. Compared to traditional experimental

0016-2361/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.fuel.2010.02.023
 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430 3417

tests for engine optimization, numerical investigation saves cost 2. Description of optimization and regression methods
and time, while maintaining acceptably high reliability and
accuracy. 2.1. Optimization method
Genetic Algorithm (GA) search techniques have become one
of the main-stream optimization methods and are now increas- Engine optimization is a multi-objective problem (MOP), as nor-
ingly used in engine design [4–10]. For example, Wickman et al. mally more than one objective function, such as emissions and fuel
[7] applied a Single Objective Genetic Algorithm (SOGA) to opti- consumption, are of interest. So, it is necessary to understand if a
mize a small passenger-car diesel engine and achieved an opti- multi-objective solution is optimal since trade-offs always exist
mal piston design. Bergin et al. [8] used the same SOGA method in MOPs. The notion of ‘‘optimum” is normally referred as the Par-
to optimize spray variables and discovered a new concept called eto optimum [17]. As illustrated in Fig. 1, Cases A to D are Pareto
‘‘spin-spray” combustion. However, SOGA demonstrates depen- optimal cases because neither of them is out-performed by other
dency on the choice of merit function and, unfortunately the cases in this problem with the aim of minimizing both objectives.
definition of an appropriate merit function is usually unclear Motivated by the fact that elitism helps achieve better conver-
to the decision maker prior to the optimization process of a gence in Multi-Objective Evolutionary Algorithms (MOEAs), Deb
multi-objective problem, such as engine optimization. This et al. [11] proposed a new elitist non-dominated sorting GA. NSGA
motivates interest in use of Multi-Objective Genetic Algorithms II employs the crowding tournament selection in which two rules
(MOGA). Shi and Reitz [9] assessed three widely used MOGAs are applied to the selection operator: (1) Solutions with higher
for optimizing a heavy-duty diesel engine operated at high-load. ranks are given preference to be selected; (2) If they have the same
In their research, the Non-dominated Sorting Genetic Algorithm rank, the less crowded cases are assigned higher priority. The small
II (NSGA II) [11] with a large population size was found to per- crowding distance indicates that the case is similar to other exist-
form better for use in engine optimization, and an optimal ing cases and is thus redundant. Recently, Shi and Reitz [18] added
injection strategy and better matching of the piston geometry new features to the original NSGA II code [19] for application to en-
with the spray plume were also identified. In Shi and Reitz’s la- gine optimization. They applied a design-niching method to the
ter study [10], the NSGA II method coupled with multi-dimen- optimization process, in which the crowding distance was calcu-
sional engine CFD modeling with simplified combustion models lated based on design parameters instead of objective functions.
was successfully applied to optimize the piston design for a It was shown that the evolutionary process of their engine optimi-
heavy-duty diesel engine operated at low- and high-load condi- zation generated more diversified design parameters instead of
tions. They showed that the high-load operating condition is objective functions. In the same study, a convergence metric was
more sensitive to the combustion chamber geometry compared defined in order to dynamically monitor the engine optimization
to at the low-load condition, which suggests that engine opti- process and decide the termination time. These two additional fea-
mization studies for all operating loads should start with an tures of NSGA II were thus employed in this study.
optimization study of the piston geometry and spray targeting
at the high-load condition. 2.2. Non-parametric regression analysis method
In modern engines, the vast number of variables that control
the combustion process results in a large number of iterations to Because hundreds of individual evaluations of engine design are
achieve optimal designs. In order to save computational time, all typically conducted in engine optimization problems, the large vol-
the aforementioned engine optimization studies used either phe- ume of resulting data necessitates the use of regression methods
nomenological models (e.g., Ref. [5]) or relatively simple combus- for post-process analysis. In order to create response functions of
tion models (e.g., Refs. [6–10]). Obviously, the use of simple the parameter values that are mapped to objective values, a non-
combustion models is not applicable in the present research, be- parametric regression (NPR) method was used in this work as a
cause the focus of this study is to compare the influence of differ- post-processing and optimization data analysis tool. A recently
ent fuels on engine optimal designs. Therefore, this necessitates developed method for model selection and model fitting in non-
the use of more computationally expensive engine CFD tools with parametric regression models, the COSSO [15,16] was applied.
detailed fuel chemistry, and the KIVA3v2-Chemkin code was used COSSO is based on a smoothing spline analysis of variance (SS-AN-
in this study. Since with detailed chemistry over 90% of the compu- OVA) model, and within this framework, the response function can
tational time is spent on the chemistry solver, acceleration of the be expressed as one-to-n-way interactions among input parame-
solver is critically important. As discussed in Refs. [12,13] in detail, ters, where n represents the number of design parameters of inter-
the Adaptive Multi-grid Chemistry (AMC) model [12] and the Ex- est. In this work, only one-way interaction (namely, the effect of
tended Dynamic Adaptive Chemistry (EDAC) scheme [13] were one parameter on objective functions) was considered. After the
employed to achieve efficient calculations with detailed fuel chem- response function is determined, the magnitude and response
istry. The feasibility and reliability of the AMC model for engine
optimization was validated by Ge et al. [14].
The present paper starts with a description of multi-objective
optimization problems. The NSGA II method is summarized and
new features that were used in this research are also discussed.
In addition, a non-parametric regression analysis method, i.e., the
COmponent Selection and Smoothing Operator (COSSO) [15,16]
that was used to post-process the optimization results is briefly de-
scribed. The numerical models were first validated against experi-
ments on a CI heavy-duty engine with an SOI sweep study.
Optimization results of the same CI heavy-duty engine fueled with
diesel, gasoline, and E10 (10% volume percent ethanol blended
with 90% gasoline) and operated under mid- and high-load condi-
tions, are then discussed for both non-parametric and parametric
studies. Finally, guidelines for optimal engine design with the dif- Fig. 1. Definition of Pareto solutions and Pareto front.
ferent fuels are provided.
3418 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430

Table 1 Table 2
Engine specifications. Operating conditions of the Caterpillar DICI engine.

Engine Caterpillar DI diesel Mid-load High-load

Combustion chamber Quiescent, direct injection Speed (rev/min) 1300 1300
Bore  stroke (mm) 137.16  165.1 Fuel amount (g/cycle) 0.135 (0.141 for E10) 0.270 (0.282 for E10)
Bowl width (mm) 97.8 IMEP (bar) 11 21
Displacement (L) 2.44 EGR (%) 30 30
Connection rod length (mm) 261.6 Global equivalence ratio 0.6 0.8
Geometric compression ratio 16.1:1 IVC temperature (K) 435 435
Boost pressure (bar) 2.0 3.0
IVC (°aTDC) 85 85
shape that each design parameter has on the objective-space can EVO (°aTDC) 130 130
be quantified. It is also worth noting that a design needs to be se-
lected as the reference point (center id) for the non-parametric
study, and in this sense, the study can be regarded as sensitivity The validity of COSSO was proven by Shi and Reitz [9,10], and Refs
analysis of all parameters about a reference design of interest. [15,16] provide more details about the regression method.

Fig. 2. Comparison of simulation and experiment [21] (gasoline main SOI sweep (70% injected fuel) with 0% EGR and 30% pilot injection at 137 °C).
 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430 3419

3. Model validation were set according to Shi et al. [9,10], and a population size of 32
was used. Six tasks (two operating conditions and three fuels) were
The present work adopted a modified version of the KIVA3v2 conducted, which used 192 computer nodes running in parallel on
code with an efficient chemistry solver (AMC model [12] and EDAC the University of Wisconsin Condor system [22], which dynami-
scheme [13]) to evaluate the performance of a CI heavy-duty en- cally collects idling computers throughout the university campus.
gine fueled with diesel, gasoline and E10 under mid- and high-load The entire optimization was completed in approximately six weeks
conditions. The engine specifications and operating conditions are (estimated time would be more than six months with the original
summarized in Tables 1 and 2, respectively. The CFD tool with de- chemistry solver).
tailed fuel chemistry [20] was validated against the experimental The optimization focused on selecting injection system param-
data by Hanson et al. [21] on the same engine. Fig. 2 compares eters with the different fuels. In-cylinder air motion due to swirl
the pressure traces and emissions for a SOI sweep from 8 to 2 was also considered due to its large impact on the air–fuel mixing
°aTDC. It is seen that the CFD simulations agree with the experi- process. Therefore, eight parameters were optimized, which are
mental trends fairly well. In addition, Shi et al. [1] has shown that summarized in Table 3 together with their lower and upper bound-
the present numerical models with the efficient chemistry solver aries. Six objective functions were selected: soot, NOx, UHC, and
are able to predicate the engine performance and pollutant emis- CO emissions, fuel consumption (indicated by Gross Indicated Spe-
sions as functions of design parameters adequately well. cific Fuel Consumption (GISFC)), as well as an engine noise indica-
tor, the Peak Pressure Rise Rate (PPRR). In addition, three feasibility
constraints were defined: maximum cylinder pressure and PPRR of
4. Results and discussion
20 MPa and 30 bar/CA, respectively, and lowest maximum average
temperature of 1200 K, below which the engine would misfire. If
The optimization studies were conducted using a Multi-Objec-
one or more of the constraints were violated, a penalty was as-
tive Genetic Algorithm, NSGAII. The NSGA II related parameters
signed to its objectives, and thus the design was given the lowest
Table 3 priority for selection in the evolution. This simple penalty mecha-
Optimization parameters and their ranges. nism proved to be very effective to remove infeasible designs.
Parameter Range
Pilot SOI (°aTDC) 85.0 to 55.0 4.1. Mid-load case
Injection pressure of pilot injection (bar) 300–2000
Amount of pilot injection (%) 0–50 The optimization was terminated at the 25th generation (i.e., 32
Main SOI (°aTDC) 35.0–10.0
(population)  25 (generation) = 800 individual evaluations for
Injection pressure of main injection (bar) 500–2000
Spray included angle (°) 60.0–85.0 each fuel) based on the convergence metric [18] shown in
Swirl ratio 0.0–2.0 Fig. 3(a). The figure shows that further GA evolutions only produce
Number of holes 6–12 more cases to fill the existing Pareto front (optimum solution set).

Fig. 3. Convergence metric and Pareto solutions for mid-load.

3420 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430

Since it is impossible to visualize Pareto solutions with six objec-

Case 4

28.9

75.3
16.8
78.0
tives in a single plot, the values of the six objectives of the Pareto

6.978

211.6
0.163

9.051

2.099
0.006
1658

1166

23.1
20.9

81.0
0.0
solutions are mapped onto three plots, as shown in Fig. 3(b–d)

8
for soot and NOx, UHC and CO, as well as GISFC and PPRR,
respectively.
Fig. 3(b) shows the trade-off relation of soot and NOx emissions

Case 3

33.8

63.5
24.3
67.0

0.144
4.703
4.061

4.208
200.8
for diesel (squares), while for gasoline (circles) and E10 (triangles),

0.004
1783
73.9

18.4
24.0
883

0.7
most of the Pareto solutions have extremely low soot emissions, as

7
well as NOx. As seen in Fig. 3(c), the optimum diesel solutions tend
to produce more CO emissions while gasoline and E10 generate
more UHC emissions. CO indicates the combustion completeness

Case 2

33.6

79.0
20.8
80.0

7.616

3.136
0.169
7.630

206.6
0.004
with diesel, but UHC is the indicator for gasoline and E10 under

1670
79.7

19.4
648

1.5
0.5
the present operating condition. The optimal diesel fuel designs

8
have lower PPRRs than those of gasoline and E10 and the fuel con-
sumption of the optimum designs with the three fuels is distrib-
uted widely in Fig. 3(d). In general, designs with higher PPRR

Case 1

66.1
13.9
25.0
70.0

2.432

189.7
4.038

30.39
9.100
0.000
have better fuel economy, and for gasoline and E10 it is more dif-

1150

1250
72.5
25.0

10.4
E10

1.0
ficult to obtain a good compromise between fuel economy and

9
PPRR.

Case 4

79.3

17.5

59.3
2.321
4.437
4.422
183.2
29.42
0.000
1873

1113

5.0
73.7
0.17

10.7
Table 4

1.7
11
Effect of individual design parameters on objectives for mid-load.

Soot NOx UHC CO GISFC PPRR

0.116 1.582 2.716 10.22 206 8.58

Case 3

56.1

50.6
(a) Diesel

2.339
3.539
183.2
12.49
0.810
0.003
1798

7.7
46.9

12.5
66.0
974

0.5

0.9
SOI1 68.0

10
Pressure1 1070
First pulse 0.102

SOI2 2.66
Case 2

55.2

49.5
12.33
12.59
218.5
4.189
0.515
0.003
1926

1238
8.4

1.2
49.9

18.3
66.0
0.7
Pressure2 982

6
Spray angle 73.0
Swirl ratio 1.39
Gasoline
Case 1

66.3
14.4
25.0
70.0

Hole 8.61

3.583
1.532
4.516
180.6
69.04
0.006
1150

1250
72.5
25.0

3.3
1.0

number
9

Soot NOx UHC CO GISFC PPRR

0.00335 0.882 14.35 8.61 215.0 9.94
(b) Gasoline
Case 4

SOI1 69.1
64.2

63.4
1.151
2.836
181.7
8.179
2.260
0.046
1352

9.3

76.6
648

Pressure1 1260
6.6

1.8

6.8
5.0
8

First pulse 0.319

SOI2 12.3
Case 3

84.6

78.9
10.7

Pressure2 1280
1.961
1.146

187.5
4.360

10.42
0.049
76.3
578

965

1.6

2.8
2.8
5.0

Spray angle 70.6

7

Swirl ratio 1.47

Representative optimal designs for the mid-load condition.

Hole 8.80
number
Case 2

72.8
78.0

3.876
9.947
223.6
4.982
0.124
0.368

16.6
19.5
31.0

60.4
699

954
6.8

0.4

Soot NOx UHC CO GISFC PPRR

10

0.00578 0.6909 11.06 6.363 205.0 11.0

(C) E10
SOI1 68.6
Case 1
Diesel

66.8
15.7
25.0

13.0

Pressure1 1120
70.0

8.491
7.581

226.1
11.60

30.14
0.055
1150

1250
72.5
25.0

1.0

First pulse 0.301

9

SOI2 29.9
(a) Design parameters

(b) Simulation results

Pressure2 1520
GISFC (g/kW hr)
Pressure1 (bar)

Pressure2 (bar)

UHC (g/kW hr)

Spray angle (°)

Soot (g/kW hr)

NOx (g/kW hr)

Spray angle 77.2

PPRR (bar/CA)
First pulse (%)

CA50 (°aTDC)
CO (g/kW hr)
Hole number
SOI1 (°aTDC)

SOI2 (°aTDC)

Swirl ratio

Swirl ratio 0.83

EOI1 (°C)
EOI2 (°C)
Table 5

Hole 9.27
number
 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430 3421

The EPA on-highway HD 2010 emissions regulations [23] man- ter. Vertical arrows indicate primary influences and the tilted ar-
date dramatically low engine exhaust PM, NOx, and non-methane rows represent secondary effects. An upward directed arrow
HC (NMHC) levels, which are 0.0136, 0.27, 0.19 g/kW h, respec- indicates that the value of the objective increases with the design
tively. Searching all Pareto solutions reveals that relatively few parameter, and downward arrows indicate a monotonic decrease.
optimal diesel fuel designs satisfy the NOx regulation, and none Horizontal arrows represent negligible influences, while arc signs
meet the PM and NMHC standards. For gasoline and E10, 95% of indicate that there is either a minimum value (downward) or a
optimal designs are below the soot limit, and a few designs with maximum value (upward) of objectives in the range of the design
both soot and NOx emissions lower than the standards are also variables. A combination of downward and upward arcs means
found. However, none of the optimal gasoline and E10 designs that both a local minimum and a maximum value were found.
meet the NMHC limit, which indicates that further oxidation of Filled arcs are for primary effects as opposed to hollow arcs for
the exhaust gases is needed. A similar conclusion was reached by the secondary effects.
Manente et al. [24] whose experiments suggested that a conven- The numbers represent the mean values of the design parame-
tional oxidation catalyst should be used to oxidize UHC and CO ters and objectives of all the Pareto solutions, which provide the
emissions when a CI engine is fueled with gasoline-like fuels. general performance for each fuel. The tabulated results are consis-
In the present work benchmark designs (i.e., Case 1 in all para- tent with the observations of Fig. 3(b–d). Gasoline and E10 fuels
metric studies) for further regression analysis were set to be de- significantly favor the reduction of soot and NOx, while producing
signs with central values of the investigated ranges of the design more UHC emissions than diesel, but with slightly lower CO emis-
parameters listed in Table 5, except that the main SOI was ad- sions. It is also seen that the PPRR level increases with the fuel oc-
vanced to 25 °aTDC instead of its central value 12.5 °aTDC tane number as the optimum designs with E10 fuel have the
(since the E10 engine was found to misfire). The results of the highest PPRR, followed by gasoline and diesel fuel.
benchmark designs for the three fuels are indicated by stars in Interestingly but not surprisingly, the primary design parameter
Fig. 3(b–d). Due to the early injection timing, the Peak Pressure with diesel fuel is the amount of fuel in the first injection pulse, as
Rise Rate of E10 for the benchmark design is very high. four objectives show a strong dependency on this parameter, fol-
The effects of each individual parameter on the objectives are lowed by the second injection timing. This is different from the en-
summarized in Table 4. In the table each symbol designates the gine with gasoline and E10, where the second injection timing
predicted change of an objective as a function of a design parame- predominantly affects engine performance and emissions, fol-

Fig. 4. In-cylinder images of representative diesel cases for mid-load.

3422 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430

lowed by the second injection pressure and spray included angle. in-cylinder image processing and parametric studies are used to
The principle reason is that the combustion phasing with diesel further verify and investigate the optimal designs.
fuel is controlled by both the pilot injection and the main injection Accordingly, in addition to the benchmark design, Case 1, three
due to the high reactivity of diesel. However, as indicated in Fig. 10 optimal designs from the Pareto solutions were reevaluated. They
of Ref. [1], a rich mixture and a high ambient pressure are both were chosen to further discuss the effects of primary design
important to ignite gasoline and E10 within a reasonable time scale parameters on engine objective functions with each fuel. The de-
to avoid engine misfire. Under the mid-load condition of this study, sign parameters and simulation results of the selected optimal de-
the amount of the first injection gasoline and E10 is not able to signs are shown in Table 5(a) and (b), respectively. In addition to
form locally a rich enough mixture. Therefore, it is the second the value of the six objectives, Table 5(b) also lists the end of injec-
injection that triggers the combustion and affects the subsequent tion timings for the first and second injections, as well as the loca-
emission formation. It is also found that E10 needs a very early tion of 50% accumulated heat release, CA50.
main injection timing as compared to gasoline due to its lower Results are visualized at four different representative timings
reactivity under the mid-load condition. This is also indicated by during the engine cycle of each case in Figs. 4–6 for the three fuels.
the averaged values in Table 4. The first image of each case illustrates the contours of fuel mass
Since the amount of fuel in the first injection is critically impor- fraction from 0% to 1% in a vertical cut-plane through the spray axis
tant, its timing also plays a role in diesel fuel performance. Retard- at the end of the first injection (EOI1). The second image shows the
ing the first injection timing with diesel reduces UHC and CO fuel distribution (0–1%) at the time when 10% of the total energy is
emissions, as well as GISFC and PPRR. For gasoline and E10, a high- released (CA10). The third image visualizes temperature contours
er injection pressure of the main injection shortens the injection (1000–2500 K) on the cut-plane at CA50. The last image demon-
duration with promoted premixed combustion, which is found to strates the distribution of CO mass fraction (0–1%) on the plane
benefit the reduction of soot, UHC, CO and GISFC. Spray included at CA90. The colored spheres in the images represent spray drop-
angle, swirl ratio, and the number of nozzle holes have only mod- lets and their colors indicate the size distribution of the droplets
erate influence on the objectives for the different fuels. The regres- from 0 to 100 lm diameter. In these images, blue (light) represents
sion analysis reveals that there are complex relationships between the low boundary while red (dark) indicates the high boundary.
the design parameters and engine performance and emissions. To The superimposed arrows on the CA50 and CA90 images represent
gain insightful understandings about the causes of the influences, the directions of the bulk flow.

Fig. 5. In-cylinder images of representative gasoline cases for mid-load.

 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430 3423

The regression analysis indicates that diesel combustion was lized, which is also consistent with the regression analysis. So,
most sensitive to the amount of fuel injected in the first pulse. the location of CO formation and soot formation (not shown here)
The reason is revealed by comparing Cases 1 and 2 with Cases 3 for these cases is close to the stagnation-point. The centrifugal ef-
and 4 in Fig. 4. It is seen that a large portion of the diesel fuel in- fect of the swirling flow further confines the stagnation flow. With
jected at early timings enters the crevice region in Cases 1 and 2. a higher swirl ratio as for Case 4, the location of CO and soot forma-
The fuel resides in the crevice region and is released during the tion is closer to the piston bowl outer wall. For small spray in-
expansion stroke. If the main combustion occurs before TDC (with cluded angles, the air motion due to the spray jet is guided by
higher likelihood for diesel), the escaped fuel contributes to high the combustion chamber walls, as shown in Case 2 at CA50, which
UHC emissions and poor fuel economy. But if the combustion oc- results in more CO and soot being directed into the squish region as
curs late in the cycle, such as in Case 2 with CA50 of 19.5 °aTDC, it is not co-located with the high temperature region of the bowl.
the escaped fuel can be oxidized further, as illustrated in the tem- Most of the optimal solutions of the gasoline and E10 cases had
perature distribution of Case 2 at CA50. This also explains why for a relatively large amount of fuel in the first injection. The reason is
gasoline and E10, a larger amount of fuel in the first injection does that a locally rich mixture is needed in order to ignite those two
not necessarily lead to higher UHC emissions since they usually fuels. This is confirmed by observing that most cases start combus-
burn after TDC. If the first injection forms a combustible air–fuel tion in the region at the piston bowl edge and the squish region
mixture, the combustion phasing will be largely determined by where the first injection fuel and the second injection fuel overlap,
its associated parameters, such as injection timing and pressure. as seen in Figs. 5 and 6. Therefore, together with the regression
For example, for cases with only a small amount of pilot fuel that analysis, the second injection timing is found to be the most influ-
is not able to trigger the ignition, the combustion characteristics ential parameter for gasoline-like fuels under the mid-load condi-
are mainly determined by the second injection, as for Cases 3 tion. According to the fuel distribution at CA10, most of the fuel is
and 4 in Fig. 4. prepared (mixed) prior to combustion for gasoline and E10, and the
Within a proper window of injection timings, a spray with in- mixing level increases as the fuel reactivity decreases. Retarding
cluded angle of around 75° results in a stagnation-point flow field the second injection timing results in higher UHC and CO emis-
in the combustion chamber, as indicated by the flow direction for sions, as well as fuel consumption, but on the contrary, this bene-
Cases 1, 2, and 4 at CA50 for diesel in Fig. 4. This benefits air–fuel fits NOx reduction and also lowers PPRR. As a result of the
mixing as the air in both squish and bowl regions are better uti- premixed combustion, no stagnation-point flow field is found in

Fig. 6. In-cylinder images of representative E10 cases for mid-load.

3424 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430

the studied cases at CA50. It is seen that the flow directions are pri- NOx emissions. Different from the mid-load condition, some of the
marily determined by the location of the highest volumetric heat optimal designs for gasoline and E10 produce high soot emissions
release during the combustion. At CA90, CO is distributed more that are comparable to the diesel fuel case. However, designs with
widely for the gasoline and E10 cases as compared to diesel. This simultaneously reduced soot and NOx are still found in Fig. 7(b) for
favors CO oxidation with the ambient oxygen. The higher UHC gasoline and E10. As illustrated in Fig. 7(c), the combustion com-
emissions of gasoline and E10 are mainly attributed to their later pleteness is represented by the higher CO emissions compared to
combustion phasing. the UHC for all three fuels different from the mid-load condition
Overall, by filtering the optimal solutions with criteria, practical discussed previously. The high octane gasoline and E10 fuels have
engine optimal designs are summarized in Appendix A for each good fuel economy at high-load, as seen in Fig. 7(c) by the lower CO
fuel at mid-load in order to provide design guidance. The diesel en- emissions and in Fig. 7(d) by the lower GISFC. However, the trade-
gine requires both late first (-55 °aTDC) and second injection (9 off between GISFC and PPRR becomes more problematic for the
°aTDC) timings. In addition, the amount of fuel injected in the first high-load engine with gasoline and E10. None of the optimal diesel
pulse should be limited below 10%. The preferred nozzle numbers fuel designs was found to have soot and NHMC emissions lower
are from 8 to 10 with narrow injection angles from 60° to 70°. For than the EPA 2010 regulations. About 35% of the optimal designs
gasoline and E10, up to 50% first injected fuel is seen in the practi- had NOx levels below the limit, but they also were accompanied
cal cases. The first injection timings of gasoline-like fuels are about with very high GISFC. For gasoline, only a few designs had both
10° earlier ( 65 °aTDC) than those of diesel, and the second soot and NOx emissions lower than the regulated values. For E10,
injection timings are about 20 ( 10 °aTDC) to 30 ( 20 °aTDC) several optimal designs with soot emissions slightly higher than
degree earlier. The number of holes for gasoline-like fuels spreads the regulated limit were seen and about 30% optimal designs had
widely in the studied range, i.e., from 6 to 12, and the injection an- NOx emissions below the regulation. It should be noted that for
gles are from 65° to 75°. Injection pressures and swirl ratios for gasoline and E10, low NOx does not necessarily indicate high GIS-
these cases depend on the combinations of the other design vari- FC, as in the case of diesel. However, the low NMHC emission reg-
ables, which distribute widely for all three fuels. ulation is still not reachable with gasoline and E10 under the high-
load condition.
4.2. High-load case Following the same convention as for the mid-load case, the
influence of the design parameters on engine performance and
An optimization study at high-load was also conducted follow- emissions are listed in Table 6. The average values of CO and GISFC
ing the same procedure as that used for the mid-load condition. of all Pareto solutions confirm that the engine operating under
The simulations were again terminated at the 25th generation, high-load has better fuel economy with gasoline and E10 than die-
based on the convergence metric shown in Fig. 7(a). sel although the engine noise level indicated by the PPRR is higher.
Similar to the mid-load case, it is seen in Fig. 7(b) that the opti- This is primarily due to the better mixing characteristics of the gas-
mal designs with diesel fuel show a trade-off between the soot and oline and E10 sprays because the air–fuel mixing process becomes

Fig. 7. Convergence metric and Pareto solutions for high-load.

 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430 3425

Table 6 more important as the engine load increases. However, for gasoline
Effect of individual design parameters on objectives for high-load. and E10, air–fuel mixing-related parameters are not as important
Soot NOx UHC CO GISFC PPRR as they are for diesel. For diesel, the fuel amount in the first injec-
0.4473 0.4697 8.96 33.6 255 4.21 tion is no longer the sole dominant parameter. As seen in Table
(a) Diesel 6(a), more parameters, including the second injection timing, pres-
SOI1 72.9 sure, swirl ratio, and the number of holes that influence the air–
Pressure1 898 fuel mixing process play more important roles. This again confirms
First pulse 0.14 that focus needs to be placed on improving spray mixing at high-
load. The second injection timing is found to be still critically sig-
SOI2 2.15 nificant for gasoline and E10. In addition, the second injection pres-
Pressure2 932 sure becomes more important at high-load with gasoline, while
E10 is more sensitive to the spray included angle.
Spray angle 76.9 As for the mid-load condition, four individual cases, including
Swirl ratio 1.27 the benchmark case were selected for each fuel and the design
parameters and the corresponding results are given in Table 7(a)
Hole number 8.06
and (b), respectively. The regression analysis showed different ef-
Soot NOx UHC CO GISFC PPRR
fects of each design parameter on engine performance and emis-
0.1876 0.5201 5.81 16.24 223 9.77 sions, compared to at mid-load. As for the mid-load cases, cases
(b) Gasoline
at four representative timings during the engine cycle were chosen
SOI1 69.4 to visualize the in-cylinder flow fields (the fuel mass fraction range
Pressure1 862 was increased to 2% and that of CO was changed to 3%).
As with the mid-load cases, early injection of the diesel fuel pi-
First pulse 0.264
lot spray results in fuel film which eventually enters the crevice re-
SOI2 1.03 gion and is directly correlated with the UHC emissions. Retarded
main injection is needed to prevent the engine from violating the
Pressure2 1450
peak pressure and PPRR constraints (20 MPa and 30 bar/CA,
Spray angle 69.8 respectively). However, this deteriorates fuel economy as the time
Swirl ratio 1.13 allowed for air–fuel mixing shortens. Therefore, enhanced mixing
Hole number 8.53
is needed at high-load, as also pointed out in the regression analy-
sis. Case 2 in Fig. 8 has 10 nozzle holes with the smallest hole area
Soot NOx UHC CO GISFC PPRR of the four cases studied. Since the spray penetration scales linearly
0.217 0.4312 6.93 20.398 230.73 8.12
with the hole area [25], Case 2 has the shortest spray penetration,
(c) E10 although it also has the highest injection pressure (tip penetration
SOI1 64.0
scales as the 1=4 power). The smaller hole size produces smaller
Pressure1 1039 droplets downstream of the nozzle, as seen in Fig. 8(b) at CA10,
First pulse 0.344 and consequently the fuel spray evaporates faster. Hence, decreas-
SOI2 2.58 ing nozzle size (increasing the number of holes) reduces soot, UHC,
CO emissions with improved fuel economy, as shown in Table 6(a).
Pressure2 1194
Compared with the mid-load results, a significant amount of CO is
Spray angle 69.0 formed under the high-load condition, and the CO is distributed
Swirl ratio 1.16 along the combustion chamber walls and in the squish region. This
highlights the importance of swirl for this operating condition. As
Hole number 8.46 in the mid-load cases, the flow patterns at CA50 again are found to
be driven by the spray injection in the CI engine with diesel fuel.

Table 7
Representative optimal designs for the high-load condition.

Diesel Gasoline E10

Case 1 Case 2 Case 3 Case 4 Case 1 Case 2 Case 3 Case 4 Case 1 Case 2 Case 3 Case 4
(a) Design parameters
SOI1 (°aTDC) 70.0 65.1 75.4 74.2 70.0 62.4 71.3 70.8 70.0 66.3 58.1 67.1
Pressure1 (bar) 1150 1189 1866 1523 1150 888 1088 345 1150 744 1170 1697
First pulse (%) 25.0 13.2 23.0 19.1 25.0 8.5 15.3 23.6 25.0 31.6 37.1 37.
SOI2 (°aTDC) 25.0 1.5 8.8 5.3 25.0 9.7 8.8 3.4 25.0 1.5 1.2 3.3
Pressure2 (bar) 1250 1850 1301 1119 1250 1588 1891 1883 1250 997 1616 1748
Spray angle (°) 72.5 61.1 78.5 82.6 72.5 70.7 60.0 79.2 72.5 67.5 73.0 68.5
Swirl ratio 1.0 1.7 1.9 1.6 1.0 1.6 0.6 1.5 1.0 1.2 1.3 0.9
Hole number 9 10 8 6 9 11 6 7 9 9 6 11
(b) Simulation results
Soot (g/kW h) 0.104 0.183 0.120 0.277 0.006 0.034 0.032 0.063 0.013 0.060 0.060 0.107
NOx (g/kW h) 3.506 0.375 1.773 0.264 3.369 0.988 0.648 0.830 2.769 0.411 0.523 0.735
UHC (g/kW h) 12.17 4.265 5.051 9.394 0.109 1.771 4.309 4.092 0.321 3.296 6.081 3.096
CO (g/kW h) 17.96 24.39 8.190 20.11 1.185 27.09 7.042 8.295 3.611 7.007 6.181 12.32
GISFC (g/kW h) 227.5 227.8 201.2 246.8 190.3 206.9 212.2 205.9 191.0 215.3 215.2 200.9
PPRR (bar/CA) 27.38 4.454 10.11 3.706 174.4 29.0 11.1 27.0 159.0 8.060 17.22 8.333
EOI1 (°C) 63.5 61.8 70.8 69.9 62.6 59.5 66.7 58.0 62.3 54.1 46.7 57.6
EOI2 (°C) 6.4 19.2 9.9 26.5 3.7 13.3 10.7 14.3 2.7 21.2 17.7 19.1
CA50 (°aTDC) 10.1 18.2 6.2 24.1 0.7 18.9 18.9 15.1 3.3 22.8 16.7 13.5
3426 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430

Fig. 8. In-cylinder images of representative diesel cases for high-load.

For the high-load condition with high in-cylinder pressure in general, as also shown by the averaged values in Table 6. Since
and global equivalence ratio, auto-ignition is no longer a prob- auto-ignition of both E10 and gasoline is not an issue for the
lem for both gasoline and E10. This is reflected by the location high-load condition, the main consideration in the GA evolutionary
of the high temperature region at CA50 in Figs. 9 and 10, which process of the optimization study is meeting the constraints. Too-
is not necessary to be the overlapping region of the first and sec- well-premixed air–fuel charge (CA50-EOI2 > 5 °CA) and too early
ond spray plumes, as seen in the mid-load cases. Examination of combustion phasing (CA50 < 15 °aTDC) results in unacceptably
all optimal solutions of the gasoline and E10 cases reveals that high pressure rise rates. This trend can also be observed from Cases
the second injection timing is around TDC for most cases, which 1, 2, and 4 of gasoline and Cases 1 and 3 of E10 in Table 7(b). Under
is very different from the mid-load cases. This is most likely dri- the high-load condition the reactivity of E10 is slightly lower than
ven by the requirement of meeting the peak pressure constraint gasoline, which allows for more time for the mixing processes.
of 20 MPa in the optimization process. The larger area over Therefore, it is necessary to retard the combustion phasing later
which fuel is distributed at CA10 for gasoline and E10 signify than for gasoline in order to achieve similar PPRR levels. This also
the larger proportion of premixed combustion as compared to explains why, in general, E10 has lower fuel efficiency than
the diesel fuel case. But the difference at high-load is less than gasoline.
that for the mid-load case, as seen in Figs. 4–6. This is obviously Both spray-driven stagnation-point flow patterns and volumet-
due to the late main injection timing under this operating con- ric-heat-release-driven flow patterns are seen in Figs. 9 and 10.
dition and also the longer injection duration. However, optimal Due to the late CA50 of E10, the spray included angle and swirl ra-
designs with low soot emissions are still found for gasoline tio play more important roles than for gasoline. As can be seen in
and E10. Different from the diesel cases, the sensitivity of the Fig. 10, larger spray included angles ensure better utilization of
gasoline and E10 cases to the main injection pressure is primar- ambient air in both the piston bowl and squish regions and thus
ily through the injection parameters. In the optimization study, enhance combustion. However, smaller spray included angles help
the total nozzle area was fixed, and thus increasing injection to slow down the combustion and to lower PPRR levels, even
pressure is equivalent to decreasing the injection duration, though this sacrifices oxidation of soot, UHC and CO, as illustrated
which increases the time for air–fuel mixing. in Table 6(c). Spray targeting and the flow patterns during the
Distinguished from the mid-load cases where E10 required combustion profoundly influence the distribution of CO and soot
much earlier main injection timing than gasoline, the main SOI during the late engine cycle, as represented by the CO mass frac-
of E10 for the high-load condition is slightly later than gasoline tions in Figs. 8–10.
 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430 3427

As for the mid-load, practical engine optimal designs are sum- tageous over either HCCI or conventional diffusion combustion
marized in Appendix B for each fuel at high-load in order to facil- modes with diesel-like fuels. For the pure HCCI mode, control of
itate engine design. For the diesel fuel, the first injection amount the onset of auto-ignition is problematic. As seen in the optimiza-
should be limited below 20%. The first injection timings are from tion study, the second injection timing with gasoline-like fuels at
-65 °aTDC to -85 °aTDC, and the main injections range from the mid-load condition provides a mechanism to control engine
 10 °aTDC to 0 °aTDC. High swirl level is required, which is combustion. The high volatility and low ignitability of gasoline-like
seen above 1.5 for many cases. Nozzle hole numbers are from 6 fuels promote air–fuel mixing and allows a longer mixing time and
to 10 with wide injection angles from 70° to 80°. For gaso- late but still reasonable combustion phasing, which is essential for
line-like fuels, the first injection amounts are between 30% and low soot and NOx emissions and high efficiency. However, this is
40%, and the first injections are located between  65 °aTDC hardly achievable for diesel-like fuels. The premixed combustion
and  80 °aTDC. The swirl ratio is close to 1 for most cases. Nozzle mode with diesel fuel results in high UHC and CO emissions due
hole numbers are from 8 to 11 with widened injection angles from to the fact that the early injection leads to wall-wetting while
70° to 80°. Similar to the mid-load cases, the injection pressure the late injection has poor air–fuel mixing and combustion effi-
varies over a large range for all three fuels, depending on other de- ciency, such as seen in PCCI combustion [26] and MK-type combus-
sign variables. tion [27,28], respectively. In addition, conventional diesel diffusion
combustion exhibits the well-known trade-off relationship be-
tween soot and NOx emissions, as also shown in the previous opti-
5. Discussion
mization study.
For the mid-load case, it was also found that both the fuel phys-
The above optimization results indicate that it is possible to run
ical (e.g., volatility) and chemical (e.g., ignitability) properties are
a CI heavy-duty engine with gasoline-like fuels and to achieve clea-
important. But the chemical properties are more influential, as
ner and more efficient combustion than with diesel-like fuels un-
seen in the comparison of gasoline and E10 cases. E10 has different
der mid- and high-load conditions. In addition, it is noted that
preferred optimal injection variables than gasoline, especially the
compared to GDI engines CI engines fueled with gasoline-like fuels
second injection timing. Furthermore, the higher octane number
have higher compression ratio, thus better thermal efficiency.
E10 more likely features unacceptably high pressure rise rates than
The combustion mode of gasoline-like fuels under the mid-load
gasoline and diesel. This suggests that lower octane number gaso-
condition is Partially Premixed Combustion (PPC), which is advan-

Fig. 9. In-cylinder images of representative gasoline cases for high-load.

3428 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430

Fig. 10. In-cylinder images of representative E10 cases for high-load.

line-like fuels (e.g., PRF80) may be better, or more injection pulses ing the low ignitability of lean mixtures. Therefore, the use of HCCI
may be needed. Nevertheless, the high volatility of gasoline-like with advanced engine thermal management system (such as fuel
fuels is very beneficial to achieve clean and efficient CI engine com- reformation in the Negative Valve Overlap (NVO) period [30,31])
bustion, which indicates that the volatility requirement of fuels may be necessary to extend the operating limit to low-load for a
will most likely increases in future, as discussed by Kalghatgi CI engine fueled with gasoline-like fuels. Also, the use of dual-fuel
et al. [29]. Although, the optimal octane number of gasoline-like as proposed by Kokjohn et al. [32] could extend the operating limit
fuels in CI engines should be emphasized in future. of gasoline CI engines. Finally, it should be pointed out that the
Compared to the mid-load condition, the high-load condition high volatility of gasoline-like fuels is helpful for an HCCI engine
was found to be less sensitive to fuel reactivity. The second injec- in order to better prepare a fully premixed charge.
tion timings of optimal solutions were all close to TDC with the dif-
ferent fuels. However, mixing-related parameters, such as injection 6. Conclusions
pressure and swirl strength were found to be more important for
the high-load cases. It is noted that due to the long injection dura- The present study presents a comprehensive optimization study
tion, diffusion combustion was not avoidable for all three fuels of a heavy-duty CI engine operated under mid- and high-load con-
investigated at high-load. Also, considering the maximum pressure ditions and fueled with diesel, gasoline, and E10. The focus of the
and Peak Pressure Rise Rate constraints, the degree of premixed present paper was optimization of injection system parameters,
combustion should be limited. At high-load, the size of the nozzle including pilot and main injection timings, pressures, and
holes and the injection pressure were found more influential than amounts. Conclusive remarks are as follows.
for the mid-load case. The total area of the nozzle holes was fixed
in the present study, so the injection duration was only a function (1) Due to the large amount of individual engine CFD evalua-
of injection pressure. However, the optimization study suggests tions that are required for optimization with detailed fuel
that increasing the total nozzle hole area may be helpful for gaso- chemistry, an efficient chemistry solver was necessary and
line-like fuels since reduced injection duration was found to be was successfully applied in the research.
beneficial. (2) Gasoline-like fuels exhibit great potential for cleaner com-
Although optimization of a low-load condition was not con- bustion than with conventional diesel fuel. For the mid-load
ducted, general guidance is offered from the results of the present condition, the ignitability of gasoline-like fuels significantly
study. Igniting gasoline-like fuels at low-load is difficult consider- influences the specification of the injection-related design
 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430 3429

parameters for engine performance and emissions. As a due to the diffusion combustion, the diesel fuel exhibits
result, the engine performance with gasoline-like fuels is stagnation-point flow fields in many optimal cases, while
greatly affected by the second injection timing, while for gasoline-like fuels show more volumetric-heat-release-dri-
the diesel fuel the first injection amount was found to be ven flows due to the premixed combustion. The results of
critical. For the high-load condition, mixing-related parame- the optimization study also indicate that lower octane num-
ters dominate the family of optimal designs as the ignitabil- ber gasoline-like fuels may be more helpful to improve the
ity is no longer a major factor, and injection pressure, swirl, controllability of CI engines in PPC mode and reduce engine
and nozzle designs are more influential. Consequently, noise.
higher injection pressure, swirl ratio and smaller nozzle
holes that promote air–fuel mixing are desirable but they Acknowledgement
are also subject to requirements of meeting the constraints.
(3) Different in-cylinder flow patterns were identified in the The financial support from the US Department of Energy under
optimal engine designs with the different fuels. For example, contract DEFC26-06NT42628 is greatly acknowledged.

Appendix A. Practical optimal designs at mid-load

No. Soot NOx UHC CO GISFC PPRR SOI1 Pre1 Inj (%) SOI2 Pre2 Angle Swirl Hole
(a) Diesel (soot < 0.068 g/kW h, NOx < 1.35 g/kW h, PPRR < 10 bar/CA)
1 0.05 0.85 7.60 31.77 250.19 4.30 55.02 738.57 6 8.92 1415.0 68.21 0.01 10
2 0.04 0.61 6.89 13.35 252.76 2.35 65.83 1598.2 3 8.96 1165.1 60.08 1.97 10
3 0.02 0.68 12.00 22.93 275.47 2.36 55.17 738.57 6 8.92 1415.0 68.43 0.00 8
4 0.04 1.03 8.90 22.62 234.14 5.96 55.03 738.57 6 8.92 1415.0 68.43 1.76 10
5 0.05 0.93 5.31 25.41 233.45 10.11 55.17 1525.6 6 8.45 1415.7 68.42 0.00 9

(b) Gasoline (soot < 0.0136 g/kW h, NOx < 1.35 g/kW h, GISFC < 210 g/kW h, PPRR < 10 bar/CA)
1 0.04 0.30 13.77 13.33 197.39 4.90 64.75 351.78 49 11.55 1431.8 72.66 1.66 12
2 0.00 0.95 26.34 11.17 205.26 9.20 84.54 825.18 46 15.69 1650.6 74.61 1.50 10
3 0.00 1.02 24.68 11.44 206.84 9.29 71.07 1395.4 49 6.75 776.94 76.21 1.00 11
4 0.00 1.06 19.92 13.48 208.24 6.73 70.17 1720.4 43 7.44 1660.6 64.71 1.51 6
5 0.01 0.95 4.14 6.57 185.02 9.22 56.51 1933.1 47 7.70 756.31 66.24 0.52 10
6 0.00 0.93 4.32 6.74 185.70 7.51 56.51 1933.1 47 7.62 767.37 66.41 0.52 10
7 0.01 0.95 7.94 7.79 207.72 2.34 84.78 1949.1 13 16.29 805.68 66.38 0.53 7

(c) E10 (soot < 0.0136 g/kW h, NOx < 0.27 g/kW h, GISFC < 210 g/kW h, PPRR < 10 bar/CA)
1 0.01 0.20 12.03 7.98 204.38 3.71 68.30 865.03 45 34.57 1712.8 73.30 0.72 8
2 0.01 0.19 10.88 10.37 200.99 5.00 68.24 1173.0 49 34.20 1409.5 77.57 0.06 10
3 0.00 0.20 7.76 7.80 199.77 4.62 57.69 1180.2 50 17.04 1752.1 71.90 1.61 11
4 0.01 0.25 7.54 5.03 189.70 8.38 57.29 854.16 34 34.97 1752.1 71.50 1.60 8

Appendix B. Practical optimal designs at high-load

No. Soot NOx UHC CO GISFC PPRR SOI1 Pre1 Inj (%) SOI2 Pre2 Angle Swirl Hole
(a) Diesel (soot < 0.272 g/kW h, NOx < 1.35 g/kW h, GISFC < 220 g/kW h, PPRR < 10 bar/CA)
1 0.24 0.83 3.34 26.87 218.86 5.89 75.43 1890.8 1 7.36 1306.2 73.64 0.68 6
2 0.22 1.33 2.89 19.67 202.31 5.43 63.06 930.53 8 8.41 936.47 79.14 1.59 6
3 0.26 1.09 3.61 24.76 208.08 3.76 83.27 1594.5 7 7.35 876.15 82.43 1.19 7
4 0.26 0.78 3.10 28.07 217.60 7.51 68.12 364.73 1 0.96 1794.6 80.89 1.77 10
5 0.20 1.03 4.51 10.33 207.70 3.89 55.81 305.47 20 3.27 924.59 79.99 1.60 7
6 0.21 1.21 1.95 19.15 204.61 4.28 83.09 1625.6 5 6.60 1013.7 81.31 1.84 7
7 0.27 0.88 3.92 22.61 210.43 3.35 68.29 520.23 8 5.14 926.46 82.66 1.18 8
8 0.27 0.63 4.54 23.69 219.94 3.64 82.27 1505.5 10 4.11 1563.0 82.06 1.22 9
9 0.23 0.90 2.51 24.92 213.28 9.53 59.15 333.63 5 2.13 1983.8 79.41 1.92 10
10 0.24 1.11 9.57 15.59 212.05 3.60 80.35 1328.6 25 3.75 978.76 82.61 1.57 7
11 0.24 0.89 2.18 16.74 210.96 3.61 71.12 834.36 1 5.06 1085.7 82.39 1.45 8

(b) Gasoline (soot < 0.272 g/kW h, NOx < 1.35 g/kW h, GISFC < 210 g/kW h, PPRR < 10 bar/CA)
1 0.22 0.92 12.81 17.50 209.12 5.06 76.19 1004.3 34 3.23 1543.4 76.34 1.75 8
2 0.25 0.72 6.98 17.64 207.87 7.88 78.39 1176.2 34 6.51 1543.4 76.27 0.96 11
3 0.26 1.16 5.38 23.48 200.50 5.61 81.86 1175.2 34 0.54 1543.2 78.39 0.96 10
(continued on next page)
3430 Y. Shi, R.D. Reitz / Fuel 89 (2010) 3416–3430

Appendix B (continued)

No. Soot NOx UHC CO GISFC PPRR SOI1 Pre1 Inj (%) SOI2 Pre2 Angle Swirl Hole
4 0.21 0.72 5.25 22.35 202.42 6.85 75.59 819.02 34 0.68 1396.1 60.06 1.02 9
5 0.16 1.07 5.29 14.55 202.55 8.88 76.43 318.46 34 5.30 1904.9 73.63 1.26 11
6 0.23 0.88 4.51 16.75 201.46 5.76 71.35 997.99 35 2.07 1860.2 73.62 0.73 8
7 0.24 0.74 5.05 19.26 202.75 9.80 72.54 1365.7 34 4.93 1395.7 78.39 0.96 11
8 0.16 1.08 4.28 13.75 200.42 8.39 75.60 302.94 36 5.30 1904.8 73.37 1.26 11
9 0.20 0.94 5.13 16.16 205.41 7.43 70.77 531.85 31 5.18 1896.8 74.50 1.57 9

(c) E10 (soot < 0.272 g/kW h, NOx < 1.35 g/kW h, GISFC < 210 g/kW h, PPRR < 10 bar/CA)
1 0.19 1.23 4.11 23.38 201.36 9.11 57.93 1171.4 34 3.23 1543.4 72.61 0.81 11
2 0.23 1.17 5.22 27.15 203.50 7.24 66.19 1601.6 34 3.30 1615.8 72.52 0.80 11
3 0.22 0.69 8.05 28.28 209.33 9.78 69.01 1337.4 40 3.08 1550.0 63.03 0.86 10
4 0.15 0.62 5.13 17.72 209.50 9.72 55.28 1202.3 35 6.25 1623.3 62.96 0.89 9
5 0.15 1.15 5.01 11.90 202.92 8.42 64.86 842.69 37 3.51 1622.2 78.24 1.03 11
6 0.19 1.02 3.17 17.64 202.78 7.39 55.38 1134.0 34 3.19 1416.9 79.29 0.87 11
7 0.12 1.29 4.20 14.92 198.53 9.61 62.82 1103.2 39 3.06 1610.7 69.99 0.38 11

References [16] Liu Y, Lu F, Reitz RD. The use of non-parametric regression to investigate the
sensitivities of HSDI diesel emissions and fuel consumption to engine
parameters. Int J Engine Res 2006;7:167–80.
[1] Shi Y, Wang Y, Reitz RD. CFD modeling a heavy-duty compression–ignition
[17] Deb K. Multi-objective optimization using evolutionary algorithms. New
engine fueled with diesel and gasoline-like fuels. Int J Engine Res 2010 [under
York: Wiley; 2001.
review].
[18] Shi Y, Reitz RD. Assessment of multiobjective genetic algorithms with different
[2] Kong SC, Han ZY, Reitz RD. The development and application of a diesel
niching strategies and regression methods for engine optimization and design.
ignition and combustion model for multidimensional engine simulations. SAE
J Eng Gas Turbines Power;in press:132. doi: 10.1115/1.4000144.
Paper 950278; 1995.
[19] <http://www.iitk.ac.in/kangal/codes.shtml>, accessed on Dec. 2009.
[3] Singh S, Reitz RD, Musculus MPB, Lachaux T. Validation of engine combustion
[20] Ra Y, Reitz RD. A reduced chemical kinetic model for IC engine combustion
models against detailed in-cylinder optical diagnostics data for a heavy-duty
simulations with primary reference fuels. Combust Flame 2009;155:713–38.
compression–ignition engine. Int J Engine Res 2007;8:97–126.
[21] Hanson R, Splitter D, Reitz RD. Operating a heavy-duty direct-injection
[4] Verma R, Lakshminarayanan PA. A case study on the application of a genetic
compression–ignition engine with gasoline for low emissions. SAE Paper
algorithm for optimization of engine parameters. Proc Inst Mech Eng Part D-J
2009-01-1442; 2009.
Automob Eng 2006;220:471–9.
[22] Thain D, Tannenbaum T, Livny M. Distributed computing in practice: the
[5] Hiroyasu T, Hiroyasu H, Miki M. Reduction of heavy-duty diesel engine
CONDOR experience. Concurr Comput-Pract Exp 2005;17:323–56.
emission and fuel economy with multi-objective genetic algorithm and
[23] <http://www.dieselnet.com/standards/us/hd.php>, accessed September; 2009.
phenomenological model. SAE Paper 2004-01-0531; 2004.
[24] Manente V, Tunestal P, Johansson B. Effects of different type of gasoline fuels
[6] De Risi A, Donateo T, Aforgia D. Optimization of the combustion chamber of
on heavy duty partially premixed combustion. SAE Paper 2009-01-2668; 2009.
direct injection diesel engines. SAE Paper 2003-01-1064; 2003.
[25] Shi Y, Reitz RD. Study of diesel engine size-scaling relationships based on
[7] Wickman DD, Senecal PK, Reitz RD. Diesel engine combustion chamber
turbulence and chemistry Scales. SAE Paper 2008-01-0955; 2008.
geometry optimization using genetic algorithms and multi-dimensional spray
[26] Opat R, Ra Y, Gonzalez MA, Krieger R, Reitz RD, Foster DE, Durrett RP, Siewert
and combustion modeling. SAE Paper 2001-01-0547; 2001.
RM. Investigation of mixing and temperature effects on HC/CO emissions for
[8] Bergin MJ, Hessel RP, Reitz RD. Optimization of a large diesel engine via spin
highly dilute low temperature combustion in a light duty diesel engine. SAE
spray combustion. SAE Paper 2005-01-0916; 2005.
Paper 2007-01-0193; 2007.
[9] Shi Y, Reitz RD. Assessment of optimization methodologies to study the effects
[27] Kimura S, Aoki O, Ogawa H, Muranaka S, Enomoto Y. New combustion concept
of bowl geometry, spray targeting and swirl ratio for a heavy-duty diesel
for ultra-clean and high-efficiency small DI diesel engines. SAE Paper 1999-01-
engine operated at high-load. SAE Trans, J Engines 2008;117:537–57.
3681; 1999.
[10] Shi Y, Reitz RD. Optimization study of the effects of bowl geometry, Spray
[28] Kimura S, Aoki O, Kitahara Y, Aiyoshizawa E. Ultra-clean combustion
targeting, and swirl ratio for a heavy-duty diesel engine operated at low and
technology combining a low-temperature and premixed combustion concept
high load. Int J Engine Res 2008;9:325–46.
for meeting future emission standards. SAE Paper 2001-01-0200; 2001.
[11] Deb K, Pratap A, Agarwal S, Meyarivan T. A fast and elitist multiobjective
[29] Kalghatgi GT, Risberg P, Angstrom HE. Partially pre-mixed auto-ignition of
genetic algorithm: NSGA-II. IEEE Trans Evol Comput 2002;6:182–97.
gasoline to attain low smoke and low NOx at high load in a compression
[12] Shi Y, Hessel RP, Reitz RD. An adaptive multi-grid chemistry (AMC) model for
ignition engine and comparison with a diesel fuel. SAE Paper 2007-01-0006;
efficient simulation of HCCI and DI engine combustion. Combust Theory Model
2007.
2009;13:83–104.
[30] Hiray K, Kakuhou A, Urushihara T, Itoh T. A study of gasoline-fueled
[13] Shi Y, Liang L, Ge H-W, Reitz RD. Acceleration of the chemistry solver for
compression ignition engine – effect of fuel reformation during negative
modeling DI engine combustion using Dynamic Adaptive Chemistry (DAC)
valve overlap. SAE Paper 2002-08-0319; 2002.
schemes. Combust Theory Model; in press. doi:10.1080/13647830903548834.
[31] Cao L, Zhao H, Jiang X. Analysis of controlled auto-ignition/HCCI combustion in
[14] Ge H-W, Shi Y, Reitz RD, Willems W. Optimization of a HSDI diesel engine at
a direct injection gasoline engine with single and split fuel injections. Combust
low-load operation using CFD with detailed chemistry and multi-objective
Sci Technol 2008;180:176–205.
genetic algorithm. Proc Inst Mech Eng Part D-J Automob Eng; in press.
[32] Kokjohn SL, Hanson R, Splitter D, Reitz RD. Experiments and modeling of dual-
doi:10.1243/09544070JAUTO1351.
fuel hCCI and PCCI combustion using in-cylinder fuel blending. SAE Paper
[15] Lin Y, Zhang HH. Component selection and smoothing in smoothing spline
2009-01-2647; 2009.
analysis of variance models. Institute of Statistics Mimeo Series 2556, NUCS;
2003.