CFD based optimization of the mixture formation in...

Scientia Iranica B (2014) 21(5), 1621{1634

Sharif University of TechnologyScientia Iranica

Transactions B: Mechanical Engineeringwww.scientiairanica.com

CFD based optimization of the mixture formation inspark ignition direct injection CNG engine

I. Chitsaza;b, M.H. Saidia;� and A.A. Mozafaria

a. Center of Excellence in Energy Conversion (CEEC), School of Mechanical Engineering, Sharif University of Technology, Tehran,P.O. Box 11155-9567, Iran.

b. Department of Laboratory, Iran Khodro Power Train Co. (IP-CO), Tehran, P.O. Box 1398813711, Iran.

Received 12 April 2012; received in revised form 6 November 2013; accepted 25 January 2014

KEYWORDSCNG-DI;Mixture formation;Optimization;Genetic algorithm;Bowl geometry.

Abstract. This paper describes optimization of the combustion chamber geometry andinjection timing of a new generation of EF7 engine, where CNG is directly injected to thecombustion chamber, with the aim of providing the best mixture at low and high speeds.The Multi-Objective Genetic Algorithm (MOGA) is coupled with the KIVA ComputationalFluid Dynamics (CFD) code, with grid generation, in order to maximize the ammablemass of the mixture. This would result in better combustion and improved fuel economy.The optimization variables related to the combustion chamber are seven geometry variablesand injection timing. Through the present optimization, a great improvement in mixturedistribution is achieved. The optimization results show that early injection with a shallowbowl shape can be advantageous at high speeds, while late injection would result in betterresults at low speeds.© 2014 Sharif University of Technology. All rights reserved.

1. Introduction

Compressed Natural Gas-Direct Injection (CNG-DI)engine development has now become a challenging andinnovative technology. Due to the growing importanceof future emission restrictions, manufacturers of in-ternal combustion engines are forced continuously tooptimize combustion chamber design for improvementof the mixture formation and combustion process [1].Direct injection in internal combustion engines enablesrealization of strati�ed charges and seems to be themost promising way [2-4] of improving fuel economyin spark-ignition engines. Mixture distribution ofstrati�ed direct injection is very important for morestable combustion and lower emissions.

High capacity computers have enabled wider ap-plication of computational uid dynamics modeling in

*. Corresponding author. Tel.: +98 21 66165522;Fax: +98 21 66000021E-mail address: [email protected] (M.H. Saidi)

engine design. Mixture distribution a�ects combustione�ciency, engine output and emissions. Thus, inorder to maximize fuel economy, optimization of thefuel-air mixture formation is essential for di�erentengine operating conditions or various combustion bowlshapes. This motivates engine designers to combineoptimization processes with numerical work.

Some previous studies were conducted on naturalgas direct-injection combustion using a rapid com-pression machine or engine [5-11]. Zeng et al. [5,6]investigated the combustion characteristics of a directinjection natural gas engine under various fuel injectiontimings. The e�ect of fuel injection timing relativeto ignition timing for natural gas direct-injection ina rapid compression machine [6,7,9,10] were studied,and the characteristics of combustion and emissionsof CNG-DI combustion [8] were evaluated. It wasshown that the compression ratio should be higher than10, and a compression ratio of 12 provided the bestresults for the engine, employed by Zheng et al. [11].Hassan et al. [12] and Kalam and Masjuli [13] presented

1622 I. Chitsaz et al./Scientia Iranica, Transactions B: Mechanical Engineering 21 (2014) 1621{1634

a converted CNG-DI engine that has some bene�tsin BSFC, NOx, while a great penalty is observedfor HC and CO emissions. This penalty is due tothe same compression ratio for CNG-DI, CNG-BI andgasoline-PI. Sen et al. [14] investigated the time seriesof the indicated mean e�ective pressure of CNG-DIusing wavelet to �nd the dominant oscillatory modes.Tadesse and Aziz [15] presented the e�ect of boostpressure on engine performance and exhaust emissions.An experimental study was conducted on a 4-strokedirect injection compressed natural gas spark ignitionengine with compression ratio of 14. They concludedthat partial direct injection gives better performancefor speeds of 2000 compared to early injection timings.They also found that early injection timing with boostpressure gives better performance for engine speedsof 4000 to 5000 RPM, compared to partial directinjection, due to su�cient available time for mixturepreparation.

Abdullah et al. [16] performed a numerical simu-lation on a single cylinder 1.6 liter engine running atwide open throttle. They concluded that fuel injectiontiming plays an important role in engine performance,combustion and emissions, because it in uences thetime of fuel air mixing shortly before spark ignitionbegins. The retardation of fuel injection reduces CO inthe exhaust but increases NO emission, due to excessair in the mixture, and decreases combustion pressure.Agarwal and Assanis [17] undertook a numerical workto investigate heat release and pollutant formationprocesses in a CNG-DI engine. They showed thatthe use of detailed chemistry was extremely impor-tant for predicting the correct ignition delay periodunder di�erent engine operating conditions. Barattaet al. [18] applied a Star-CD based model for inves-tigation of the direct natural gas injection processfrom a poppet-valve injector into a bowl-piston enginecombustion chamber. The numerical procedure wasvalidated through comparison of numerical results withSchlieren images of the jet pro�le in a constant-volumebomb. Several studies have been focused on CFDbased optimization in engine design. Mohd Ali etal. [19] presented the gas-jet ignition method to extendthe lean combustible limit of CNG engines. Theirresults showed that a CNG engine can be operated atequivalence ratio ranges from 0.3 and 0.8 by applyinga two-stage injection combined with gas-jet ignition.Their experimental and numerical e�orts showed thatcombustion stability and emissions are a�ected by fueldistribution. Andreassi et al. [20] applied a singlecylinder direct injection for the direct injection ofa small quantity of natural gas through the sparkplug. They also performed numerical work to improvethe micro-direct injection PSC process. Andreassi etal. [21] also presented an analysis of di�erent injectionstrategies to demonstrate that these tools are suitable

for optimization of direct injection gas engines. In theirnumerical e�ort, the main performance parameters,such as maximum pressure, maximum temperature,heat rate curve and burned fuel mass fraction, havebeen considered to achieve a better comprehensionof the physical phenomena involved in the engine.Douailler et al. [22] performed a 3D CFD simulationand an experimental test on the converted dieselengine. They modi�ed and designed a new cylinderhead and piston crown to optimize combustion velocityafter their simulations. Mohamad et al. [23] useda spark plug fuel injector for the conversion of portfuel injection to direct injection of natural gas. Theyshowed that there is pressure loss in the fuel injection ofnatural gas but a spark plug gaseous injector improvesvolumetric e�ciency. Research into the e�ects ofdi�erent injection strategies, injector types and pistoncrown geometries on the mixture formation of a directinjected turbocharged CNG-engine was carried out byChiodi et al. [24]. They followed homogeneous andstrati�ed mixture formation to realize the e�ects ofmixture formation on combustion.

Kurniawan et al. [25,26] applied a neural networkas an arti�cial intelligence tool to determine the per-formance and emissions in a compressed natural gasdirect injection engine. They used 15 data acquiredfrom CFD combustion simulation to characterize thecombustion behavior of such engines used for thetraining process of arti�cial neural networks. Theyconcluded that the best strategy for achieving higherpower, lower CO, and NO emissions was to advance thestart of injection timing and delay the end of injectiontiming with earlier spark advanced timing. They [27]also optimized the combustion process of compressednatural gas direct injection using the coupled CFDand multi-objective genetic algorithm code, in orderto improve engine performance and emissions. Theirobjectives were optimized by a variation in start ofinjection, end of injection and spark ignition timing at aspeed of 2000 rpm. Engine power output and exhaustemissions were also considered important parametersas objective functions.

For development of the CNG-DI engine operatedunder di�erent loads, the mixture formation variesconsiderably, which results in di�erent injection timingand combustion chamber shapes. It can be anticipatedthat di�erent injection strategies and matching ofpiston geometry are needed for engines under di�erentoperating conditions in order to get better results. Inprevious work, the optimized geometry and operatingcondition of CNG-DI has not yet been studied, andall optimization research is con�ned to engine output,while mixture distribution has a critical role. In thepresent study, a CFD based optimization is used tosimulate mixture formation of a CNG-DI engine in EF7and to study the e�ects of bowl geometry and injection

I. Chitsaz et al./Scientia Iranica, Transactions B: Mechanical Engineering 21 (2014) 1621{1634 1623

timing on the engine. The optimized combinationsof piston bowl geometry and injection timing areconsidered to improve fuel economy at low and highspeeds. This investigation into the CNG-DI engine hasnot been undertaken in existing literature. Moreover,the optimization process applied here allows severalobjectives to be contemporarily optimized; the e�ectsof injection timing and combustion bowl geometry arealso considered.

2. Mathematical model

In the present study, simulations were performed usingthe KIVA3V [28] code, which was coupled with a ge-netic algorithm and mesh generation with KIVA-Prepin order to optimize the combustion bowl geometry andinjection timing. The volume of the piston bowl waskept constant for all cases to keep the compression ratiounchanged. A second order upwind scheme was usedto discretize both momentum and continuity equations.The closed-loop of the engine cycle was simulatedand the turbulent ow was modeled by a modi�edrenormalization group (RNG) k-" model [29]. NaturalGas injection in KIVA is modeled by the Ra et al. [30]procedure. In this method, the in ow conditions werecalculated using the under-expanded jet model and thesimilarity solution [31]. The e�ective nozzle diameterfor the similarity solution was assigned to the size ofthe Mach disk, and the velocity at the e�ective nozzleexit was assumed to be sonic. The mass ow rate ofthe jet was calculated at the distance where the jet'shalf-velocity width was equal to the size of the in owboundary cell of the computational grid.

2.1. Engine description and operatingconditions

A single-cylinder, direct injection, four-stroke SI en-gine, based on an EF7, was used in the modeling.The main geometric speci�cations and fuel injectorparameters are summarized in Table 1. Due to the

Table 1. Engine and injector speci�cation.

Baseline engine EF7 SI engine

Bore � stroke (mm) 78.6 � 95

Fisplacement (liter) 2.0

Connecting rod length (mm) 134.5

Guel injector nozzles Axisymmetric nozzle

Tail pressure (bar) 30

Nozzle ori�ce diameter (mm) 0.225 mm

Compression ratio 12.24

Equivalence ratio Stoichiometric

Swirl ratio 2.8

limited space for mounting the injector in the EF7cylinder head, a spark plug fuel injector is selectedfor the injection system. This means that the injectoris one hole, centrally mounted in the cylinder head.The base piston bowl geometry is determined, based onBaratta et al. [18]. However, slight changes were madein the present modeling work, such as axisymmetricsimulations. Bowl geometry was modi�ed and also thestroke was increased from 82.5 to 95 to increase thecompression ratio. The piston geometry and cylinderhead are shown in Figure 1, where z is axisymmetricof the geometry.

2.2. Grid-size independencyTo avoid grid-size dependency, four mesh sizes with thesame conditions were used to check the mesh depen-dency of the problem. The methane mass fractions ofthe axisymmetric line (AH in Figure 1) for di�erentmesh sizes are given in Figure 2. As shown, themethane mass fraction along the axis approximately

Figure 1. Computational grid based on EF7 engine andBaratta et al. simulations [18].

Figure 2. Methane mass fraction of four mesh size atTDC.


Table 2. Design variables and objectives.

Design variablesRange

800 rpm 2000 rpm

zGp (mm) -30 to -26 -30 to -26

zG (mm) -30 to -26 -30 to -26

zF (mm) -25.5 to -21.5 -25.5 to -21.5

zE (mm) -22.9 to -18.9 -22.9 to -18.9

xEe (mm) -22 to -20 -22 to -20

xE (mm) -20.8 to -19.5 -20.8 to -19.5

xG (mm) -18.7 to -17.7 -18.7 to -17.7

xGp (mm) -18 to -16.7 -18 to -16.7

Injection timing 180 BTDC to 20 BTDC 180 BTDC to 20 BTDC

Objectives:

Flammable mixture in the vicinity of spark plug

Flammable mixture in the cylinder

does not change after re�nement of 55000 meshes. Asa compromise between computational e�ciency andmodel accuracy requirements, the cell size was speci�edas 0.2 mm in this work.

2.3. Optimization parameters and objectivesIn order to analyze a wide range of piston crown shapes,a parametric grid generator is developed. In this gridgeneration, �ve parametric points of H, G, F , E andEe, in Figure 1, are varied to construct new pistoncrown shapes to keep the compression ratio constant.In addition, the start of injection which is one ofoperating conditions, is optimized.

There are a number of methods applied to quan-tify the mixing performance of a given gas injec-tor [32,33]. The mixture that �lls each computational�nite volume can belong to three categories: toolean, ammable and too rich. Too lean and richmixtures cause the ame to quench or can be ledto mis�re at the start of combustion. Therefore, a ammable mixture would result in stable combustion.It is also notable that unburned hydrocarbon andNOx emission would be increased, due to the toorich and too lean mixtures in the engine operatingcondition, respectively. Therefore, ammable massdistribution increases the performance of the engineand also reduces the engine emissions. In other words,the ammable mass fraction of the fuel is related tocombustion e�ciency, as it indicates the relative massof fuel that is combustible [33].

The objectives of engine optimization includeboth the ammable mass fraction of the mixture at thetime of ignition in the cylinder, and the ammable mix-ture in the vicinity of the spark plug. The ammable

mass fraction of the mixture is de�ned as the ratioof the ammable mass of the mixture to the totalmass of the mixture in the cylinder, which is writtenas:

fm =Flammable mass of mixture

Total mass of mixture: (1)

Optimization parameters and objectives are summar-ized in Table 2.

2.4. Micro genetic algorithmIn the present study, the optimization process is per-formed with the multi objective micro genetic algo-rithm that was already tested for engine application byPark [34]. This optimization process is used to optimizecombustion bowl geometry as well as injection timing inthe CNG-DI engine. The multi-objective optimizationapproach, based on a micro genetic algorithm (micro-GA), a genetic algorithm with a very small population(�ve individuals), was used in our experiment. Thealgorithm starts from a random initial population ofbinary strings and calculates the �tness of each indi-vidual ranked according to Pareto optimally criterion.The Pareto optimal solutions consist of solutions thatare not dominated by any other solutions, where theconcept of domination is de�ned as:

m < n, 8i(mi � ni) ^ 9i(mi < ni): (2)

For an optimization where the minimization of eachobjective is sought (such as minimizing emissions andfuel consumption), when Eq. (2) is met, m is said tobe dominated by n [35]. Figure 3 [36] shows thatdesigns A to D are Pareto optimal solutions. Thebest individuals, those that have strings with high


Figure 3. Pareto optimal solutions [36].

Figure 4. Computational mesh of the baseline engine atthe top dead center.

overall �tness values, are selected for reproduction,while their chromosomes are exchanged with the singlepoint crossover technique. In this optimization, elitismis also applied, by replacing some strings from the newpopulation with individuals from the Pareto optimalsolution. The new generation is used as startinggeneration until convergence is met. In order toperform the optimization process, the baseline bowlgeometry needs to be generated based on variablesrelated to the chamber geometry. Therefore, the pistonbowl geometry of the baseline engine was modeled asthe mesh shown in Figure 4.

2.5. Injection modelDue to the focus of this research being on mixtureformation, numerical gaseous injection should also bevalidated. Gaseous injection is the start of mixtureformation, and the accuracy of numerical simulationhas a great e�ect on the results. In the present study,it is assumed that the uid is compressible and Newto-nian with temperature-dependent uid properties. Anumerical solution of the mean ow and thermal �eldsrequires resolving the continuity equations: Reynoldsaveraged Navier-Stokes equation, time-averaged energyequation, species transport equation and ideal gasrelation [37]:

@(��ui)@xi

= 0; (3)�@(��ui�uj)@xj

�=� @P

@xi+

@@xj

"�

@�ui@xj

+@�uj@xi

� 23�ij@�ul@xl

!#� @@xj

��u0iu0j

�; (4)

@@xi

��ui��Yjhj + �

�u2

2

��=

@@xi

�K@ �T@xi� �u0iT 0

�+

@@xi

�hj��Dj;k +

�tSct

�@Yj@xi

�; (5)

@(��ujYi)@xj

=@@xj

��Di;k +

�tSct

�@Yi@xj

�; (6)

P = �RT: (7)

Enthalpy in Eq. (5) is de�ned by Eq. (8):

�hj =Z T

298:15Cp;jdT: (8)

As shown in Eqs. (4) and (5), additional terms, ��u0iu0jand �u0iT 0, were entered in the energy and Navier-Stokes equations. Additional terms should be modeled.Boussinesq represent a fundamental equation that isshown in Eq. (9). Some turbulence models, such ask-", use this fundamental equation:

�u0iu0j = ��t�@�ui@xj

+@�uj@xi

�+

23

��k + �t

@�ul@xl

��ij ;

u0iT 0 = �Cp�t�Prt

�@T@xi

�: (9)

In Boussinesq approximation, Reynolds stress is relatedto the local velocity gradients introducing the eddyviscosity, �t. The turbulence quantities (k and ")used to calculate �t are determined from the followingmodeled transport equations. The presented turbulent ow was modeled by a modi�ed renormalization group(RNG) k-" model [29]:

�@k@t

=@@xi

��k�eff

@k@xi

�+ �t

p2

2

�@�ui@xj

+@�uj@xi

�� "� 2�"

k RT

; (10)

�@"@t

=@@xi

��"�eff

@"@xi

�+ 1:42

"k�t

�p

22

�@�ui@xj

+@�uj@xi

�� 1:68�

"2

k�R; (11)


Figure 5. Numerical and experimental data of gaseousinjection: (a) Helium density of simulated andexperimental results in pressure ratio 1.31 and d = 1 mmat 4 ms; and (b) tip penetration comparisons ofexperimental and numerical data at d = 1 mm.

�t = C��k2

"; (12)

where C�, �k and �" are constants that equal 0.0845,1.393 and 1.393, respectively.

The main di�erence between the RNG and stan-dard k-" models lies in the additional term in the "equation, given by:

R =C��3(1� �=�0)

1 + ��3"2

k; (13)

where:

� � k="� �t �p

22

�@�ui@xj

+@�uj@xi

�;

�0 = 4:38; � = 0:012:

In regions where � < �0, the R term makes apositive contribution. As a result, for weakly tomoderately strained ow, the RNG model tends togive more turbulence in comparison to the standardk-" model.

To evaluate model accuracy, tip penetration of thehelium injection is validated and shown in Figure 5.The density contours of the present computations arealso compared to the experimental ones extracted fromthe Schlieren [38]. As shown in this �gure, tip penetra-tion is predicted quite accurately in the numerical sim-ulation. It is notable that the diameter of the injectedjet is slightly larger than the geometrical diameter usedin the engine simulations. As far as the gas-jet widthand penetration are concerned, an overall trend canbe observed, that is, both become smaller when thenozzle diameter decreases. In the present study, asecond order upwind scheme is used to discretize bothmomentum and continuity equations. Gas injection inKIVA3V was modeled by the Ra et al. [30] procedure

Figure 6. Schematic of the hybrid combination oftheoretical and underexpanded jet models [30].

Figure 7. Numerical and experimental in-cylinderpressure of EF7.

which is shown in Figure 6. The present modelemployed a simpli�ed procedure for under-expanded jetbehavior and a theoretical correlation for the turbulentgas jet to obtain the corresponding conditions at thenumerical in ow boundaries. As shown in Figure 6,the �rst part of the jet emerging from the Mach diskis approximated using theoretical gas jet results, andthe CFD code only simulates the subsequent gas jetmixing.

3. Results and discussion

In the lack of experimental data for mixture distribu-tion, the in-cylinder pressure of the EF7 engine versuscrank angle is compared with numerical simulation andshown in Figure 7. The experiments were performedfor the port fuel injection of CNG in the EF7 engineat 1250 rpm and full load operation. Present com-


putations predict in-cylinder pressure with quite goodaccuracy.

For the evaluation of the mixture, the ammabil-ity limits should be de�ned in terms of mixture RelativeAir-Fuel Ratio (RAFR). Based on previous experiencewith port-injected CNG SI engines [39], in the presentwork, 0.8 to 1.4 RAFR limits were taken. Morespeci�cally, these boundaries should be considered asthose de�ning a stable condition of ame propagation.

The optimization process of an EF7 engine isperformed in order to have the best mixture formationto increase the fuel economy of the CNG-DI engine.Our objectives are the percent of ammable mixturenear the spark and in the total volume of the cylin-der. The spherical volume with a radius of 0.5 mmis selected as the vicinity of spark. A ammablemixture near the spark is essential for the start ofignition. It is also notable that the continuity of themixture should be kept in order to stable combustion.Operating conditions of direct injection engines over2500 rpm, whether at high or low loads, are inclined toa homogeneous mixture that is prepared by early fuelinjection at the intake process. In this study, we havefocused on the strati�ed stoichiometric mixture alreadyproposed by Bai et al. [40]. This strategy leads tohigher CO and lower HC and NOx emissions, which canbe e�ectively after-treated using a three-way catalyst.The strati�ed stoichiometric mixture can also reducethe knocking phenomena, due to the lean mixture nearthe walls at the end of combustion. Therefore, we havefocused on low engine speeds where this strategy can beimplemented. Our operating conditions are con�ned tothe boundaries of the strati�ed stoichiometric mixturecondition. Therefore, 800 rpm (idle) and 2000 rpm areselected.

The results of the optimization process are shownin Figure 8, as points in the plane of objectives accord-ing to the ammable mixture close to the spark plugversus total in-cylinder ammability. These optimal

designs are found to be able to increase the ammablemixture to improve fuel economy, compared to a base-line engine under the same operating conditions. Asshown in this �gure, the Pareto front at 2000 rpm is awide range, while, at 800 rpm, it is narrow. It indicatesthat objectives at 800 rpm have more connectivity toeach other than those of 2000 rpm. Therefore, they cansimultaneously improve during optimization.

Some of the designs selected from the Pareto frontat 2000 rpm are reported in Table 3. The geometry andinjection timing of the Pareto front gave a signi�cantimprovement for all cases, with respect to the baselineengine. The results of Table 3 show that goal functionsare a�ected by injection timing, while the geometry isthe same for some of the optimum cases. This impliesthe importance of the start of the injection on goalfunctions. Moreover, the percent of ammable areanear the spark plug has more impact than other goalfunctions at the start of combustion.

In order to see the sensitivity of the optimizationvariables to the objective functions, the correlationsbetween the ammable mass of the mixture versus thegeometry variable (Ee) are studied for 2000 and 800rpm, and the results are shown in Figures 9 and 10.Two curves are �tted to the data to understand thee�ects of the variable on mixture preparation. It is seenfrom Figure 9(a) that decreasing xEe would increasethe ammable mixture near the spark, while it doesnot have an important e�ect on total ammability.This trend shows that shallow wide bowl geometryprepares the mixture more appropriately than a deepnarrow bowl at 2000 rpm, which leads to enhancedfuel composition. Cubic �tting implies that decreasingcup depth reduces the ammability almost linearly.It is believed that the shallow type bowl geometryutilizes the in-cylinder oxygen more easily, which leadsto better mixture distribution. Figure 10 implies that ammability near the spark in the deep narrow bowlis improved, while total in-cylinder ammability does

Figure 8. Pareto front and results of the optimization process: a) 2000 rpm; and b) 800 rpm.


Table 3. Some of Pareto design at 2000 rpm.

Paretofront

zGp zF zE xEe xE xG xGpStart ofinjection

Percentage of ammable

area

Percentage of ammablearea nearspark plug

Baseline -29 -24.5 -21.9 -21 -21.5 -18 -17 130CA BTDC 37.5% 0%1 -26 -21.5 -18.9 -21.9 -22.4 -18.9 -17.9 154CA BTDC 82.5% 62%2 26 21.5 18.9 21.9 22.4 18.9 17.9 163CA BTDC 67.5% 74%3 26 21.5 18.9 21.9 22.4 18.9 17.9 155CA BTDC 82.5% 64%4 26 21.5 18.9 21.9 22.4 18.9 17.9 158CA BTDC 80% 68%5 26.3 21.8 19.2 21.8 22.3 18.9 17.8 160CA BTDC 74.75% 70%6 26.3 21.8 19.2 21.8 22.3 18.9 17.8 166CA BTDC 60% 78%7 26.3 21.8 19.2 21.8 22.3 18.9 17.8 168CA BTDC 55% 78%8 26 21.5 18.9 21.9 22.4 18.9 17.9 171CA BTDC 45% 80%

Figure 9. E�ect of combustion bowl geometry on objectives at 2000 rpm: a) Flammability near spark plug; and b)in-cylinder ammability.

Figure 10. E�ects of combustion bowl geometry on objectives at 800 rpm: a) Flammability near spark plug; and b)in-cylinder ammability.

not signi�cantly vary with variation in geometry at 800rpm. The linear trend in this �gure for total in-cylinder ammability shows a 10 percent improvement, whilecubic �tting has a descending trend. Data is scatteredin all parts of the plane, as shown in Figure 10, andthis distribution does not show a speci�c trend. This

distribution shows that the geometry parameter doesnot have a distinctive in uence on the objectives at 800rpm.

The e�ect of piston crown height that correspondsto zF is shown in Figure 11. This �gure veri�es thata shallow piston has better mixture distribution than


Figure 11. E�ect of combustion bowl geometry on objectives: a) In-cylinder ammability at 2000 rpm; b) ammabilitynear spark plug at 2000 rpm; c) in-cylinder ammability at 800 rpm; and d) ammability near spark plug at 800 rpm.

Figure 12. E�ect of start of injection on objectives at 2000 rpm: a) Flammability near spark plug; and b) in-cylinder ammability.

a deep bowl shape at 2000 rpm. It is notable thatgeometry does not have an important e�ect on amma-bility near the spark plug at 2000 rpm, while it a�ectsat 800 rpm, and decreasing zF increases ammabilitynear the spark. Total in-cylinder ammability at 800rpm was not in uenced by geometry in the optimumcases, due to early injection of fuel, and is evident from

Figure 11(c). When gas is injected too early, thereis enough time for mixing, which leads to a nearlyhomogenous mixture. Therefore, the e�ect of geometrybecomes negligible.

Correlations between mixture formation and in-jection timing are shown in Figures 12 and 13. Inthe present study, injection durations of 56� and 23�


Figure 13. E�ect of start of injection on objectives at 800 rpm: a) Flammability near spark plug; and b) in-cylinder ammability.

Figure 14. Comparison of optimum and baseline cases at 2000 rpm: a) Methane mass fraction of baseline at TDC; b)velocity magnitude of baseline at TDC; c) methane mass fraction of optimum case at TDC; and d) velocity magnitude ofoptimum case at TDC.

correspond to speeds of 2000 and 800 rpm, respectively.Figure 12(a) and (b) indicate that the optimum in-jection strategy for the best mixture distribution ofthe strati�ed engine is to advance injection timing,while the late injection at 800 rpm would result inbetter mixture distribution. In the late injection,the gas jet is not in uenced by geometry, due tothe limited time of di�usion. Therefore, the start ofinjection has the greatest e�ect on the results. Thisjusti�es the weak independency of the total in-cylinder ammability of Figure 10(b) to geometry constraints.It is also notable that the percent of ammabilitynear the spark is more sensitive to injection timingthan total in-cylinder ammability. As shown in this�gure, in some �nite injection timing, the ammablemass of the mixture is available at the vicinity of thespark, due to the high pressure near the spark plugand special con�guration of the piston crown shape.At the appropriate injection timing, natural gas has

enough momentum to overcome the pressure of thecylinder head, di�uses to the vicinity of the spark plug,and the ammable mass fraction would be available forthe start of stable combustion. Figure 13 also impliesthat both objectives were simultaneously optimized atspeci�c injection timing. Therefore, both objectives at800 rpm can be merged into a single function. Figure 13indicates that there is an optimum for the start ofinjection at 800 rpm, and it is close to 46 CA BTDC.

Species distribution and the in-cylinder velocityof the optimum case at 2000 rpm were compared tothose of the baseline case in Figure 14. As shownin this �gure, a signi�cant improvement can be seenin methane distribution, while velocity magnitude isnot greatly changed. Methane distribution is moreuniform for the optimum case, but a rich mixture wasaccumulated in the piston crown due to the generatedvortex in the combustion chamber. Optimum caseshave a shallow type bowl shape and advanced injection


Figure 15. Comparison of optimum and baseline cases at 800 rpm: a) Methane mass fraction of baseline at TDC; and b)methane mass fraction of optimum case at TDC.

timing to get the best mixture distribution. Opti-mum shape decreases the rich mixture and distributesmethane in the whole volume of the cylinder head.A narrow bowl piston combustion chamber is notappropriate for charge strati�cation, because the richmixture is con�ned to the piston crown shape, withoutmeeting compactness requirements. Such a combustionchamber is suitable for homogeneous mixture prepa-ration, with advanced SOI. In fact, Figure 14 showsthat the resultant fm level at the ignition crank angleincreases with larger and more uniform bowl geometry.This could be not true for combustion chambers with at head pistons, because the recirculation motion ofthe methane is hindered by the unfavorable chambersection shape onto axial planes.

Figure 15 shows the methane mass fraction of thebaseline and one of the Pareto front cases at 800 rpm.As shown in this �gure, a considerable improvementof methane distribution near the spark plug can beseen. It is notable that injection timing has a moreimportant e�ect than geometry on the optimizationprocess. Therefore, the piston crown is not signi�cantlychanged during optimization. Flammability near thespark at 800 rpm is a more challenging objective duringoptimization, because, in many cases, it becomes in-�nitesimal. Late injection at 800 rpm for a narrow bowlshape piston would result in an aggregation of CNG inthe piston crown shape, due to the limited momentumof the injected gas. In a large bore shape and at headpiston, CNG is inclined to the lower pressure part in thecombustion chamber, and is accumulated in crevicesand low pressure parts. The vicinity of the sparkplug at ignition time has the highest pressure of thecombustion chamber. Therefore, late injection wouldnot prepare a suitable ammable mixture at the startof combustion near the spark plug. These e�ects ratio-nalize advanced injection for optimized con�guration.It is also notable that a nearly homogeneous mixture isprepared at the start of ignition for 800 rpm, and theconnectivity of objectives to each other can be cleared.

4. Conclusions

A multi-dimensional CFD simulations code with aMOGA is introduced for development of a CNG-DI

engine. Optimization is focused on piston bowl geome-try and injection timing for the development of EF7 toCNG direct injection at low and high speeds. Therewere seven optimization points related to the com-bustion chamber geometry and injection timing underdi�erent engine operating conditions. The ammablemass of the mixture near the spark plug and total in-cylinder ammability were optimized simultaneouslyduring optimization. The ammable mass near thespark is essential for the start of combustion and totalin-cylinder ammability maintains ame propagation.Conclusive remarks are as follows:

� Optimization showed that high-speed operating con-ditions are more sensitive to combustion chambergeometrical design compared with the low-speedconditions. This is due to the limited time of gasdi�usion at high speed in comparison to low speed.

� Total in-cylinder ammability and ammabilitynear the spark at 800 rpm have more connectivity toeach other than those of 2000 rpm. Therefore, theycan be simultaneously improved during optimiza-tion, while the Pareto front at 800 rpm is narrowerthan 2000 rpm. It is due to a nearly homogenousmixture at 800 rpm.

� At 2000 rpm, the shallow type bowl geometry ismore appropriate for the CNG-DI engine becausethis shape prepares the mixture better than thedeep narrow bowl. It is also notable that at lowspeed, geometry does not have an important e�ecton mixture distribution.

� The most important e�ect on mixture distributionat low speed is injection timing, which shows thatan almost retarded injection would result in bettermixture preparation.

� Among the optimization variables, injection timingis very important and advanced injection for highspeed and late injection for low speed would resultin optimized fuel economy.

Nomenclature

A Geometry variable


B Geometry variable

Binj Geometry variable

BTDC Before Top Dead Center

BI Bi-fuelBSFC Brake Speci�c Fuel Consumption

Cp Heat transfer coe�cient (m2s�2K�1)

Cu Constant

CA Crank Angle

CO Carbon monoxide

D Geometry variable

DI Direct Injection

Dkj Binary di�usivity for k � j system(m2t�1)

E Geometry variable

Ee Geometry variable

EF7 Engine Family 7

F Geometry variable

G Geometry variable

Gp Geometry variable

GDI Gasoline Direct Injection

h Enthalpy (kgm2s�2)

Ho Geometry variable

Hp Geometry variable

k Turbulent kinetic energy (m2s�2)

K Thermal conductivity (kg.m.s�2.K�1)

MOGA Multi Objective Genetic Algorithm

NOx Nitric oxide

p pressure (kgm�1s�2)

Pr Prandtl numberR Term in RNG turbulence modelRNG Renormalization group

RAFR Relative Air Fuel RatioSc Schmidt numberSI Spark Ignition

T Temperature (K)

u Axial velocity (ms�1)

x Longitudinal coordinate

Y Mass fractionz Axial coordinate

Greek symbols

� Inverse e�ective Prandtl number

� Energy dissipation rate (m2s�3)

� Dynamic viscosity (kg m1s�1)

� Kinematic viscosity (m2s�1)

� Density (kgm�3)

Subscripts

e� E�ectivet TurbulentT Theraml

References

1. Baumgarten, C., Mixture Formation in Internal Com-bustion Engines, Springer Berlin, Heidelberg (2006).

2. Croissant, K. and Kendlbacher, C. \Requirementsfor the engine management system of gasoline directinjection engines", Direkteinspritzung im OttomotorCongress, Essen, Germany (1997).

3. Zhao, F.Q., Lai, M.C., Harrington, D.L. and En-gineers, S.O.A., A Review of Mixture Preparationand Combustion Control Strategies for Spark-IgnitedDirect-Injection Gasoline Engines, Society of Automo-tive Engineers (1997).

4. Harada, J., Tomita, T., Mizuno, H., Mashiki, Z.and Ito, Y. \Development of direct injection gasolineengine", SAE Technical Paper 970540, Detroit, Michi-gan, United States SAE, pp. (1997).

5. Zeng, K., Huang, Z., Liu, B., Liu, L., Jiang, D.,Ren, Y., et al. \Combustion characteristics of a direct-injection natural gas engine under various fuel injectiontimings", Applied Thermal Engineering, 26, pp. 806-813 (2006).

6. Huang, Z.-H., Liu, L.-X., Jiang, D.-M., Ren, Y., Liu,B., Zeng, K., et al. \Study on cycle-by-cycle variationsof combustion in a natural-gas direct-injection engine",Proceedings of the Institution of Mechanical Engineers,Part D: Journal of Automobile Engineering, 222, pp.1657-1667 (2008).

7. Huang, Z., Shiga, S., Ueda, T., Nakamura, H., Ishima,T., Obokata, T., et al. \Study of cycle-by-cyclevariations of natural gas direct injection combustionusing a rapid compression machine", Proceedings of theInstitution of Mechanical Engineers, Part D: Journalof Automobile Engineering, 217, pp. 53-61 (2003).

8. Huang, Z., Shiga, S., Ueda, T., Nakamura, H., Ishima,T., Obokata, T., et al. \E�ect of fuel injection timingrelative to ignition timing on the natural-gas direct-injection combustion", Journal of Engineering for GasTurbines and Power, 125, pp. 783-790 (2003).

9. Huang, Z., Shiga, S., Ueda, T., Jingu, N., Nakamura,H., Ishima, T., et al. \A basic behavior of CNG


DI combustion in a spark-ignited rapid compressionmachine", JSME International Journal Series B, 45,pp. 891-900 (2002).

10. Shiga, S., Ozone, S., Machacon, H., Karasawa, T.,Nakamura, H., Ueda, T., et al. \A study of the com-bustion and emission characteristics of compressed-natural-gas direct-injection strati�ed combustion us-ing a rapid-compression-machine", Combustion andFlame, 129, pp. 1-10 (2002).

11. Zheng, J., Wang, J., Wang, B. and Huang, Z. \E�ectof the compression ratio on the performance andcombustion of a natural-gas direct-injection engine",Proceedings of the Institution of Mechanical Engineers,Part D: Journal of Automobile Engineering, 223, pp.85-98 (2009).

12. Hassan, M.H., Kalam, M.A., Mahlia, T.I., Aris, I.,Nizam, M.K., Abdullah, S., et al. \Experimentaltest of a new compressed natural gas direct injectionengine", Energy & Fuels, 23, pp. 4981-4987 (2009).

13. Kalam, M. and Masjuki, H. \An experimental inves-tigation of high performance natural gas engine withdirect injection", Energy, 36, pp. 3563-3571 (2011).

14. Sen, A.K., Zheng, J. and Huang, Z. \Dynamicsof cycle-to-cycle variations in a natural gas direct-injection spark-ignition engine", Applied Energy, 88,pp. 2324-2334 (2011).

15. Tadesse, G. and Aziz, A.R.A. \E�ect of boost pressureon engine performance and exhaust emissions in direct-injection compressed natural gas (CNG-DI) spark ig-nition engine", SAE Paper No 2009-32-0135 (2009).

16. Abdullah, S., Kurniawan, W.H. and Shamsudeen, A.\Numerical analysis of the combustion process in acompressed natural gas direct injection engine", Jour-nal of Applied Fluid Mechanics, 1, pp. 65-86 (2008).

17. Agarwal, A. and Assanis, D. \Multi-dimensional mod-eling of ignition, combustion and nitric oxide formationin direct injection natural gas engines", SAE Transac-tions, 109, pp. 1088-1103 (2000).

18. Baratta, M., Catania, A.E., Spessa, E., Herrmann,L. and Roessler, K. \Multi-dimensional modeling ofdirect natural-gas injection and mixture formation ina strati�ed-charge si engine with centrally mountedinjector", SAE International Journal of Engines, 1,pp. 607-626 (2009).

19. Ali, M.F.M., Kodoguchi, Y., Oka, Y. and Kaida, T.\Improvement of combustion of CNG engine usingCNG direct injection and gas-jet ignition method",SAE Technical Paper 2011-01-1994 (2011).

20. Andreassi, L., Cordiner, S., Mulone, V., Reynolds,C. and Evans, R. \A mixed numerical-experimentalanalysis for the development of a partially strati�edcompressed natural gas engine", ICE 2005 7th Inter-national Conference on Engines for Automobile (2005).

21. Andreassi, L., Facci, A. and Ubertini, S. \Multidi-mensional modelling of gaseous injection in moderndirect injection internal combustion engines: Analisysof di�erent fuel injection strategies", ICE 2009 9th

International Conference on Engines for Automobile(2009).

22. Douailler, B., Ravet, F., Delpech, V., Soleri, D.,Reveille, B. and Kumar, R. \Direct injection of CNGon high compression ratio spark ignition engine: Nu-merical and experimental investigation", SAE Tech-nical Paper 2011-01-0923, Detroit, Michigan, UnitedStates (2011).

23. Mohamad, T.I., Yuso�, A., Abdullah, S., Jermy, M.,Harrison, M. and Geok, H.H. \The combustion andperformance of a converted direct injection compressednatural gas engine using spark plug fuel injector", SAETechnical Paper 2010-32-0078, Linz, Austria (2010).

24. Chiodi, M., Berner, H.-J. and Bargene, M. \Inves-tigation on di�erent injection strategies in a direct-injected turbocharged CNG engine", SAE TechnicalPaper 2006-01-3000, Perugia, Italy (2006).

25. Kurniawan, W.H., Abdullah, S., Nopiah, Z.M. andSopian, K. \The development of arti�cial neural net-work for prediction of performance and emissions ina compressed natural gas engine with direct injectionsystem", SAE Technical Paper 2007-01-4101, Chicago,Illinois, United States (2007).

26. Kurniawan, W.H., Abdullah, S., Nopiah, Z.M. andSopian, K. \The application of arti�cial neural networkin predicting and optimizing power and emissions ina compressed natural gas direct injection engine",SAE Technical Paper 2007-01-4264, 2007, Rosemont,Illinois, United States (2007).

27. Kurniawan, W.H., Abdullah, S., Nopiah, Z.M. andSopian, K. \Multi-objective optimization of combus-tion process in a compressed natural gas direct injec-tion engine using coupled code of CFD and genetic al-gorithm", SAE Technical Paper 2007-01-1902 (2007).

28. Amsden, A.A., Los Alamos National Laboratory Report(1999).

29. Han, Z. and Reitz, R.D. \Turbulence modeling ofinternal combustion engines using RNG k-" models",Combustion Science and Technology, 106, pp. 267-295(1995).

30. Ra, Y., Kong, S.C., Reitz, R.D., Rutland, C.J.and Han, Z. \Multidimensional modeling of transientgas jet injection using coarse computational grids",SAE Technical Paper 2005-01-0208, Detroit, Michi-gan, United States (2005).

31. Ouellette, P., Direct Injection of Natural Gas for DieselEngine Fueling, Vancouver, B.C., Canada, Universityof British Columbia (1996).

32. Papageorgakis, G. and Assanis, D.N. \Optimizinggaseous fuel-air mixing in direct injection enginesusingan RNG based k-" model", SAE Paper No 980135:SAE (1998).

33. Kim, G.-H., Kirkpatrick, A. and Mitchell, C. \Com-putational modeling of natural gas injection in a large


bore engine", Journal of Engineering for Gas Turbinesand Power, 126, p. 9 (2004).

34. Park, S.W. \Optimization of combustion chambergeometry for stoichiometric diesel combustion using amicro genetic algorithm", Fuel Processing Technology,91, pp. 1742-1752 (2010).

35. Genzale, C.L., Reitz, R.D. and Wickman, D.D. \Acomputational investigation into the e�ects of spraytargeting, bowl geometry and swirl ratio for low-temperature combustion in a heavy-duty diesel en-gine", SAE Technical Paper 2007-01-0119, Detroit,Michigan, United States (2007).

36. Shi, Y. and Reitz, R.D. \Optimization study of thee�ects of bowl geometry, spray targeting, and swirlratio for a heavy-duty diesel engine operated at low andhigh load", International Journal of Engine Research,9, pp. 325-346 (2008).

37. Kuan, K. and Kou, Y., Principles of Combustion, JohnWiley & Sons (1986).

38. Chitsaz, I., Saidi, M.H., Mozafari, A.A. and Hajialimo-hammadi, A. \Experimental and numerical investiga-tion on the jet characteristics of spark ignition directinjection gaseous injector", Applied Energy, 105, p. 8(2013).

39. Catania, A.E., Misul, D., Spessa, E. and Vassallo,A. \Analysis of combustion parameters and theirrelation to operating variables and exhaust emissionsin an upgraded multivalve bi-fuel CNG SI engine",SAE Technical Paper 2004-01-0983, Detroit, Michi-gan, United States (2004).

40. Bai, Y.-L., Wang, J.-X., Wang, Z. and Shuai, S.-J. \Knocking suppression by strati�ed stoichiometricmixture with two-zone homogeneity in a DISIengine",Journal of Engineering for Gas Turbines and Power,135, pp. 012803-012803 (2012).

Biographies

Iman Chitsaz received his PhD degree from SharifUniversity of Technology (SUT), Tehran, Iran, in 2013.He has been working for six years as an engineerat the Irankhodro Powertrain Co. (IP-CO), and, in2013, he and his colleges founded a research laboratorythere, which investigates injection and combustionphenomena. His research interests include direct injec-tion spark ignition engines, fuel injection systems andoptical diagnostic methods for spray and combustion.He is author or co-author of over 15 national andinternational conference and journal papers.

Mohammad Hassan Saidi is Professor and Chair-man of the School of Mechanical Engineering at SharifUniversity of Technology, Tehran, Iran. His current re-search interests include MEMS, heat transfer enhance-ment in boiling and condensation, modeling of pulsetube refrigeration, vortex tube refrigerators, indoor airquality and cleanroom technology, energy e�ciency inhome appliances, and desiccant cooling systems.

Aliasghar Mozafari received his PhD degree fromthe University of London (QMC), UK, in 1988. He hasbeen faculty member of the Mechanical EngineeringDepartment at Sharif University of Technology (SUT),Tehran, Iran, for 38 years, Head of the department forsix years and Educational Deputy of the department forfour years. He has also been member of the Center ofExcellence in Energy Conversion, as well as Director ofthe International Students' O�ce at SUT. He is authoror co-author of over 90 national and internationalconference and journal papers and has supervised over40 graduate theses. He has also authored a book,published in 1993.

CFD based optimization of the mixture formation in...

Documents

Transcript of CFD based optimization of the mixture formation in...