Vous êtes sur la page 1sur 11

Fuel 105 (2013) 272–282

Contents lists available at SciVerse ScienceDirect

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

A further evaluation of the mixing controlled direct chemistry (MCDC)


combustion model for diesel engine combustion using large eddy simulation
Yuxin Zhang ⇑, Christopher J. Rutland
Engine Research Center, University of Wisconsin-Madison, Madison, WI 53705, USA

h i g h l i g h t s

" A review of MCDC model and Kong-Reitz model is presented.


" We perform a detailed comparison between experiments, MCDC and other models.
" A single cylinder oil test engine is used for model validation over six operating modes.
" MCDC model shows improved prediction cylinder averaged results.
" The improvements are analyzed by evaluation of the time scales from MCDC model.

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

Article history: The local mixing effect at sub-grid scale (SGS) is explored using Large Eddy Simulation (LES) turbulence
Received 8 February 2012 model with a mixing controlled direct chemistry (MCDC) model. The MCDC model is developed using the
Received in revised form 8 July 2012 computation of chemical kinetics, thermo-chemical equilibrium state and SGS mixing effects by combin-
Accepted 18 September 2012
ing a mixing time scale and a kinetic time scale with the integration of direct chemistry solver (DCS) solu-
Available online 8 October 2012
tions. The mixing time scale is based on the SGS scalar variance and SGS scalar dissipation rate. The
kinetic time scale is based on thermal–chemical equilibrium state and the kinetic reaction rate. These
Keywords:
two time scales have been previously evaluated in experiments and computations, showed reasonable
Diesel engine modeling
Scalar dissipation rate
quantifications of local mixing and kinetic progress, which improved accuracy of combustion modeling.
Large eddy simulation In this study, the conventional diffusion-type operating conditions are represented by a Caterpillar
Kinetic time scale 3401™ single cylinder oil test engine (SCOTE). The cylinder-averaged results are compared between
Mixing time scale the DCS model, MCDC model and ‘Kong-Reitz’ model. A grid dependence study for the MCDC model on
a baseline grid and a coarse grid is performed. The mixing time scale and kinetic time scale analysis
for the MCDC model and ‘Kong-Reitz’ model is provided to investigate the differences in predictions.
Compared to the DCS model and ‘Kong-Reitz’ model, the MCDC model shows improved capability in
modeling mixing effects at the SGS level, which is represented by the improved prediction of combustion
phasing and good matching with experimental results.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction compression ignition (HCCI) and conventional engine simulations


[5]. In turbulence modeling, more advanced turbulence models,
Turbulent combustion in diesel engines is a mechanical–chem- such as Large Eddy Simulation (LES), have been applied in engine
ical process which inherently involves a wide range of length and simulations [6,7]. Other numerous progresses in multiphase mod-
time scales [1,2]. To improve the accuracy of engine performance eling and combustion modeling have also contributed to a better
and emission predictions, several efforts have been made in Inter- understanding of the combustion process [8]. These new modeling
nal Combustion (IC) engine CFD simulations. In fuel chemistry, im- approaches present insightful understanding of the engine com-
proved oxidation mechanisms of hydro-carbon fuels have been bustion process and new opportunities for improving the model
developed using experimental and computational efforts [3,4]. Re- accuracy.
duced or skeleton mechanisms, which include major reaction path- Recent trend in turbulent combustion modeling shows that
ways, have been successfully applied in homogeneous charge more efforts will be employed on LES-based approaches [9,10].
For high accuracy applications, LES can offer several advantages
⇑ Corresponding author. Tel.: +1 517 763 6464. over traditional Reynolds Averaged Navier Stokes (RANSs) model-
E-mail addresses: zhangyu@ge.com, zhang.spencer@gmail.com (Y. Zhang). ing approaches, such as more flow structures and time-accurate

0016-2361/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.fuel.2012.09.050
Y. Zhang, C.J. Rutland / Fuel 105 (2013) 272–282 273

solutions [9]. Based on newly formulated time scales, a mixing equilibrium state based on Gibbs’s function continuation method,
controlled direct chemistry (MCDC) model has been recently which is calculated within every computation cell at each time step
developed from the RANS-based ‘Kong-Reitz’ model in the context from the implemented CEQ solver [14]. The difference is divided by
of LES [11] and finite rate direct chemistry solver (DCS). Because of the rate of change in internal energy obtained from the DCS solution,
the high modeling accuracy requirements on diesel auto ignition U_ kinetics . This results in a kinetics based time scale.
process, DCS is one of the major modeling approaches in diesel en-
U  U
gine combustion modeling. A major improvement of the MCDC skinetics;U ¼ _ ð3Þ
model is the new formulation of mixing and kinetic time scales, U kinetics
which accurately represents the importance of sub-grid scale This kinetic time scale can be interpreted as the time for the mix-
(SGS) mixing in combustion based on local scalar variances/dissi- ture to reach chemical equilibrium based on the current reaction
pation rate and chemical equilibrium status. Evaluation of these rate, or the inverse of reaction rate normalized by the specific inter-
time scales in reacting/non-reacting engine flows has shown rea- nal energy difference between the current state and equilibrium
sonable values and behaviors [12]. Preliminary tests of this model state.
in Sandia Cummins™ N-14 engine running at low temperature The mixing time scale smixing is the time needed for the cell to
combustion (LTC) mode have shown improved prediction of engine reach perfectly mixed conditions. The subgrid mixture fraction var-
pressure trace and heat release compared to the DCS approach iance is a good indicator of the inhomogeneity of mixture fraction
[11]. In addition, a preliminary evaluation of these newly formu- at the subgrid scale SGS level. The subgrid scalar dissipation rate
lated mixing time scales in high-powered diesel engine mode tran- vsgs is the rate at which a mixture returns to homogeneity. The
sition simulations is recently performed [13]. The results show that g
mixing time scale is obtained by combining Z 002 and vsgs, similar
these time scales are promising to the understanding of transition
to the definition of the kinetic time scale:
stage of diesel engine combustion between partially premixed
compression ignition (PCCI) and conventional diesel engine operat- g
Z 002
ing conditions. However, a further evaluation of this model in con- smixing ¼ ð4Þ
ventional, diffusion-type of diesel engine is considered to be
vsgs
necessary to obtain more evaluations of MCDC model over a wide In LES, the governing equations are obtained by spatially filtering
operating condition. In this study, a single cylinder oil test engine the basic conservation equations. The filter width is the local grid
(SCOTE) of Caterpillar™ 3401 heavy-duty diesel engine is tested size. A box filter is used since it is suitable for the finite volume
over a wide range of operating conditions. The original RANS-based method that is used in the present computational fluid dynamics
‘Kong-Reitz’ model and the LES-based DCS and MCDC model are (CFD) code. A Favre filtering is applied to account for variable den-
compared with experimental data. The grid dependence study is sity of a fluid and is denoted by the e symbol.
performed for the MCDC model using two sets of computation grid. The mixing time scale requires the SGS mixture fraction vari-
In addition, a detailed analysis of these simulation results is per- g
ance Z 002 , and the SGS scalar dissipation rate vsgs. The subgrid scale
formed based on time scale analyses at different stages of the com-
g
bustion process. mixture fraction variance Z 002 is a major measure of the inhomoge-
neity in a computational cell [15,16]. It is formulated as,
2. Model formulation g fZ
Z 002 ¼ ZZ eZ
e ð5Þ
2.1. Basic formulation g
The transport equation of Z 002 in multi-phase reacting flow with pos-
sible closure is presented in [15]. In this work, an algebraic model is
The ‘Kong-Reitz’ model and the MCDC model share the assump- used based on the formulation used by Ihme et al. [17],
tion that the reaction rate for each species is determined by the ki-
2
netic process as well as the relative magnitude of mixing and g D Cu
Z 002 ¼ vres ð6Þ
reaction effects, which can be characterized by the local Damköh- mt C v C e
ler number based on a turbulent time scale and a kinetic time e 2
Here vres ¼ 2Dð@ Z=@x i Þ ; is the resolved scalar dissipation rate. Tur-
scale. The basic formulation of this type of model is based on
bulent viscosity mt is calculated from the SGS kinetic energy based
DCS solutions and local Damköhler number, which is determined
turbulent viscosity using a uniform Schmidt number assumption.
from a kinetic time scale skinetics,U, and a mixing time scale time
D is the base filter size; Cu, Cv and Ce are constants determined from
scale smixing,
energy spectra of locally homogeneous turbulence [18,19]. As sug-
1 gested in [17], Cu/(CvCe) has a range of 0.02–0.05.
-_ i ¼ -_ kinetics;i ð1Þ
The subgrid scale scalar dissipation rate,vsgs, is defined as,
1 þ Da
!
smixing @Zg@Z @ Ze @Ze
Da ¼ ð2Þ vsgs ¼ 2D  ð7Þ
skinetics @xi @xi @xi @xi
Here Da is the Damköhler number, - _ kinetics;i is the DCS reaction rate Here vsgs is modeled using a similarity type of model based on a
for species i; skinetics is the kinetic time scale and smixing is the mixing Leonard type term [20,21]. It is scaled by the ratio of SGS Reynolds
time scale. These time scales have different formulations in the number Resgs [22,23], which is calculated based on the SGS kinetic
‘Kong-Reitz’ model and the MCDC model. energy ksgs, base filter size D and molecular viscosity. The resulting
term is given by,
2.2. Time scales in the MCDC model 0 1
SðResgs Þ @ ed
Z @ e @c
Z e @c
Z e
Z
vsgs ¼ 2D @  A ð8Þ
In the MCDC model, the kinetic time scale skinetics is defined as SðRests Þ  SðResgs Þ @xi @xi @xi @xi
skinetics,U, which is the time needed for the cell to reach its chemical
equilibrium state under perfectly mixed conditions. The kinetic time Here S(Resgs) is based on the scaling properties from a priori tests on
scale skinetics,U is based on the difference between the specific internal direct numerical simulation results of forced isotropic turbulence
energy at the current state, U, and the equilibrium state, U⁄. The local (Fig. 1)[22,23]. In this term, the peaked over bar, b, denotes a test
274 Y. Zhang, C.J. Rutland / Fuel 105 (2013) 272–282

(complete consumption of fuel and oxygen). Combustion occurs at


1.0
the molecular scale as a result of collisions between reacting mole-
test filter cules. Subsequently, combustion is strongly influenced by turbu-
0.8 lence since the turbulence has significant effects on the molecular
χsgs/χtotal transport properties and on the preparation (mixing) of the reac-
χsgs/χ total

base filter tants. The delay coefficient f changes from 0 to 1 accordingly,


0.6 depending on the local conditions. In other words, the initiation

of combustion relies on laminar chemistry and turbulence starts
χsgs to have an influence only after combustion events have already
0.4 been observed [24].

10 100 1000 2.4. Summary of the MCDC and ‘Kong-Reitz’ model


Resgs
The specific internal energy based kinetic time scale skinetics,U
Fig. 1. vsgs and Lv normalized by total dissipation rate plotted as a function of sub- (Eq. (3)) can be regarded as the time for the mixture to reach chem-
grid Reynolds number using the DNS data [22]. The test is based on a 2563 cube ical equilibrium based on the current reaction rate, or the inverse
taken from a DNS cube of 5123 for forced isotropic turbulence. Dtest/Dbase = 2 [23].
of reaction rate normalized by the specific internal energy differ-
ence between the current state and equilibrium state. Compared
filter, is usually twice the size of base filter. Sub-test filter Reynolds
to the single species based time scale skinetics,1spd (Eq. (9)) in the
number Rests, characterized by the test filter size (two times of base
‘Kong-Reitz’ model, skinetics,U shows benefits in two aspects. First,
filter size), is scaled from Resgs based on scaling properties [22,23].
skinetics,U has an improved estimate of local equilibrium state based
on Gibbs’s function continuation method [25]. In comparison,
2.3. Time scales in ‘Kong-Reitz’ model
skinetics,1spd assumes certain species as depleted at equilibrium
regardless of the current thermo-chemical state. Second, skinetics,U
In the ‘Kong-Reitz’ model, it is assumed that certain species are
is of reference value as long as the species composition in a cell
depleted at equilibrium regardless of the current thermo-chemical
is reactive, regardless of the concentrations of certain species. In
state. Therefore skinetics in Eq. (2) is formulated as skinetics,1spd, which
comparison, the skinetics,1spd in the ‘Kong-Reitz’ model is based on
is regarded as the larger one of fuel based time scale sfuel,kinetics and
a single species. This time scale might show strong species depen-
CO based time scale sCO,kinetics .
dence even though the species is not important in certain reaction
skinetics;1spd ¼ maxðsfuel;kinetics ; sCO;kinetics Þ ð9Þ pathways.
For mixing time scales, the mixing time scale smixing in Eq. (4)
Y g
sfuel;kinetics ¼ _ fuel ð10Þ includes Z 002 and vsgs. These two parameters directly characterize
Y fuel;kinetics the inhomogeneity and mixing rate at SGS level and formulate a
mixing time scale of strong physical sense without the complexity
Y of model constants in the time scale formulation. In comparison,
sCO;kinetics ¼ _ CO ð11Þ
Y CO;kinetics the mixing time scale sturb in Eq. (12) used by the ‘Kong-Reitz’
model is scaled from the eddy turn over time scale k/e, which is
In ‘Kong-Reitz’ model, the equilibrium state at fuel-rich conditions not directly related to the local mixing status. This time scale re-
is ignored. The concentration of fuel species and CO species is as- quires a small model coefficient C and a delay coefficient f with
sumed to be zero at the chemical equilibrium state, the time scales complicated formulations (Eq. (13)).
in Eq. (10) and (11) are formulated as current species concentration
(Yfuel, YCO) divided by the current kinetic consumption of this spe-
cies ðY_ fuel;kinetics ; Y_ CO;kinetics Þ. 3. Validation in conventional diesel engines
In the ‘Kong-Reitz’ model, the mixing time scale smixing in Eq. (2)
is formulated as turbulent time scale sturb. It is proportional to the 3.1. Simulation conditions
eddy turnover time scale seddy = k/e, which is determined by the
turbulence kinetic energy and its dissipation rate, i.e., k and e, are 3.1.1. Experiment setup
calculated from the k  e turbulence models. In this section, six modes over a wide range of engine operating
k conditions for single cylinder oil test engine (SCOTE) of Caterpil-
sturb ¼ C f ð12Þ lar™ 3401 heavy-duty diesel engine are used to test the MCDC
e
model. The experiments are performed by Montgomery and Reitz
Here C = 0.1 if the renormalization group k  e model is used. The [26]. The engine operating conditions ranged from idling to full
delay coefficient f simulates the influence of turbulence on combus- load at different engine speeds (Tables 1 and 2).
tion after ignition has occurred, and is given as:
1  er Table 1
f ¼ ð13Þ
0:632 Specifications of the Caterpillar 3401B™ engine.

In the context where only hydrocarbon fuel is considered, parame- Bore  Stroke (mm) 137.2  165.1
ter r is the ratio of the amount of products to that of the total reac- Connecting rod length (mm) 261.62
Compression ratio 15.1
tive species (i.e., all except N2):
Displacement volume (L) 2.44
Y CO2 þ Y H2 O þ Y CO þ Y H2 Piston shape Mexican hat
r¼ ð14Þ Combustion chamber Quiescent
1  Y N2 Injection system Common rail
Number of nozzles 6
The parameter r indicates the completeness of combustion in a
Hole diameter (mm) 0.259
specific region. Its value varies from 0 (no combustion yet) to 1
Y. Zhang, C.J. Rutland / Fuel 105 (2013) 272–282 275

kinetic energy, ksgs, are initialized with non-reacting simulations.


Table 2
In this flow field initialization process, a non-reacting cycle is sim-
Operating conditions of Caterpillar 3401B™ engine.
ulated using RANS simulation. The flow field is then used as the ini-
Mode1 Mode2 Mode3 Mode4 Mode5 Mode6 tial condition of LES non-reacting simulations. Usually two or more
Engine RPM 750 953 1074 1657 1668 1690 non-reacting cycles are simulated using LES until the in cylinder
Engine load (%) 0 25 75 100 50 25 trapped mass becomes stable. The flow field from the cold flow
Intake temperature (K) 299 302 304 313 305 302
LES cycles is used for the initial condition of LES reacting simula-
Intake pressure (kPa) 100 108 168 239 164 132
Fuel injected (g/cycle) 0.0226 0.0707 0.196 0.204 0.106 0.062 tions. In-cylinder averaged swirl ratio, pressure and temperature
Start of injection (°ATDC) 8 0.5 5.5 7.5 2.0 1 are expected to match the experimental conditions. A uniform sca-
Duration of injection (°) 5.0 8.0 22.0 26.5 16.5 11.5 g
lar field with negligibly small mixture fraction variance Z 002 is as-
sumed. For the same operating conditions, after the trapped
mass variation becomes stable, the LES reacting simulation of die-
As shown in Fig. 2, two computation grids with different grid sel engine combustion results does not appear to be sensitive to
resolutions are used. The finer version or the baseline grid contains the different initial flow field sampled from the 1st, 2nd and 3rd
about 400,000 cells at bottom dead center (BDC), with an averaged cycles. The reason is that spray-induced turbulence is significantly
cell size of about 1.1 mm. It is used for DCS method, MCDC model stronger compared to that induced by the intake air flow at the low
and ‘Kong-Reitz’ model. The baseline grid is considered to have a swirl number condition found on this test engine. This observation
sufficient resolution based on previous studies by Hu et al. [27]. is consistent with the experimental findings, which indicates that
In Fig. 3, the snapshots of the temperature field at 7 CA after injec- negligible cyclic variations were found in the SCOTE test results.
tion and 35 CA after injection using the fine mesh are showed to Therefore, for the steady engine operating conditions evaluated
illustrate the onset and the late stage of combustion for a high in this paper, no consecutive multi-cycle simulation was sampled
speed, low load condition (mode 6) [28]. for the final results.
The coarse grid contains about 220,000 cells at BDC, with an Throughout this work, diesel kinetics is modeled using a re-
averaged cell size of about 1.6 mm. The coarse version of grid is duced n-heptane (C7H16) oxidation mechanism with the physical
used for the grid dependence study of the MCDC model. Previous properties of tetradecane (C14H30). The mechanism is a reduced
analysis showed that the baseline grid can resolve up to 80% of primary reference fuel (PRF) mechanism, which is found to be well
the turbulent kinetic energy. The coarse grid can resolve about suited for diesel engine combustion simulations [5]. The simula-
75% of the total kinetic energy [11]. tions are performed on Dell PowerEdge™ SC1435 systems using
In this study, only closed-cycle calculations, from intake valve AMD Opteron2384™ processors running the CentOS 5 Linux
closure (IVC) to exhaust valve opening (EVO), have been run. Spe- distribution.
cifically, at IVC, the resolved scale velocity field and subgrid-scale
3.1.2. KIVA3V-LES code
Engineering-level LES engine simulations are performed using
the LES version of KIVA-3V Release 2 (KIVA3V-LES) developed by
the Engine Research Center at University of Wisconsin-Madison
[29]. The code uses a quasi second-order upwind scheme for con-
vection terms, a second order central difference for diffusion terms
and a first order splitting scheme for time difference. The compu-
tation models for the turbulence and mixing process include the
LES dynamic structure model for subgrid-scale stress [30], an arti-
ficial turbulent diffusivity model for subgrid scalar flux [31], a one-
equation model for subgrid-scale kinetic energy [32], a similarity
model for the SGS scalar dissipation rate [23], and an algebraic
model for SGS mixture fraction [17]. The Werner–Wengle model
for the unsteady wall shear stress modeling is incorporated into
Fig. 2. Full cylinder computation grid of Caterpillar 3401™ metal engine. (a) this code [33].
Baseline grid with 400,000 cells, average filter size = 1.2 mm@TDC; (b) coarse grid
The code contains a complete set of models for multi-phase
with 220,000 cells, average filter size = 1.7 mm@TDC.
reacting follow applications. Kelvin–Helmholtz and Rayleigh–

Fig. 3. Top view temperature contour of Mode 6 at (a): +6°CA ATDC, (b) +34°CA ATDC [28].
276 Y. Zhang, C.J. Rutland / Fuel 105 (2013) 272–282

Taylor spray breakup models are used for the modeling of primary CA 10 CA 90
and secondary breakup process of the diesel spray [34]. A Lagrang- Experiment
DCS
ian approach is adopted for the spray modeling [35], The spray Mode 6 Kong-Reitz
MCDC-Baseline
break-up model parameters are calibrated against the experimen- MCDC-Coarse
tal data found in previous studies [27]. Other spray models, such as CA 10 CA 90
a comprehensive collision model for multi-dimensional engine Experiment
DCS
spray computations [36], Frössling’s correlation-based evaporation Mode 4 Kong-Reitz
MCDC-Baseline
model [37], a phenomenological nozzle flow model considering MCDC-Coarse
cavitation effects, is also included in this computation code [38]. CA 10 CA 90
Experiment
To account for the momentum transfer from gas to liquid phase, DCS
the acceleration due to the drag on a liquid drop is modeled using Mode 1 Kong-Reitz
MCDC-Baseline
a drag function [35]. In the context of LES, a local de-convolution MCDC-Coarse

model for spray source term in the one-equation subgrid kinetic


-10 0 10 20 30 40 50 60 70
energy model is incorporated to account for the momentum trans- o
fer from liquid to the gas phase [8]. In terms of interaction between Crank Angle [ aTDC]
evaporating droplet and chemistry, on one hand, the heating effect
Fig. 4. Combustion phasing and duration from experimental results and combus-
and ambient fuel vapor partial pressure change from the combus- tion models; the combustion phasing and duration are characterized by CA10 and
tion process is accounted for by the droplet evaporation model CA90 for modes 1, 4 and 6.
[37]. On the other hand, the cooling effect from the droplet
evaporation is considered by solving cell energy equation [40].
Two types of chemistry solver are implemented for the MCDC crank angle corresponding to 10% of the TCHR is abbreviated as
model. For multi-step fuel oxidation computations, an updated CA 10.
version of CHEMIKIN-II is used as DCS solver, providing thermo- To evaluate the combustion phasing at different operating con-
chemical state based on chemical kinetics [39]. For chemical equi- ditions, CA 20 and CA 50 are chosen for the analysis of three oper-
librium state calculation, a CEQ solver based on Gibbs’s function ating conditions: mode 1 (0% of load), mode 4 (100% of load) and
continuation method is implemented [14]. The DCS solver, which mode 6 (25% of load). The resolution of this analysis is at a quarter
takes as much as 60–75% of the total computation cost, is currently of the crank angle. In comparison with the experiments (Fig. 4), the
paralleled using MPI. The computation cost of each of the test cases DCS approach with the baseline grid shows the earliest prediction
using MCDC and ‘Kong-Reitz’ model averages at about 400 CPU h/ among these computations by showing an advancement of CA 10
case and 200 CPU h/case respectively. and CA 90 between 0.0 and 2.5 CA. The MCDC approach with the
In addition, ‘Kong-Reitz’ model simulations are performed using baseline grid shows a difference between 0.25 and 0.25 CA. The
the standard version of KIVA-3V Release 2 (KIVA3VR2) developed MCDC approach with the coarse grid shows a phasing between
by the Engine Research Center at University of Wisconsin-Madison the DCS and MCDC using the baseline grid, with advancements
[29] This version uses renormalized group (RNG) k  e turbulence from 0.25 CA to 1.25 CA. In comparison, the RANS-based ‘Kong-
model. It shares some of the models used by LES version, such as Reitz’ model using the baseline grid shows an early phasing predic-
the spray breakup and droplet evaporation models [32–41]. The tion for mode 4 and a late phasing prediction for modes 1 and 6.
‘Kong-Reitz’ combustion model is implemented as one of the stan- In Fig. 4, the combustion duration is characterized as the crank
dard combustion models in this code [24]. angles between CA10 and CA90. The DCS approach shows slightly
shorter combustion durations (0.25 CA to 1.0 CA) compared to
3.2. Cylinder-averaged results experimental results. The ‘‘Kong-Reitz’’ model shows a 0.75 oCA
longer duration for Mode 1 and a 1.75 CA shorter duration for
For each operating conditions (Table 2), four simulations are Mode 6. The combustion duration from MCDC model on coarse
performed. Three simulations are performed using the LES turbu- grid is slightly smaller than baseline grid. The combustion duration
lence model, including DCS simulation using the baseline grid, of these two cases show good match with the experimental results
the MCDC model simulation using the baseline grid, and the MCDC in spite of slightly differences in combustion phasing.
model simulation using the coarse grid. A simulation using the
‘Kong-Reitz’ model is performed using RANS and the baseline grid. 3.2.2. MCDC compared with DCS method
In this study, the evaluation is based on the cylinder pressure trace As shown in Fig. 5, premature heat release rates and pressure
and the apparent heat release rate (AHRR). rises are observed in DCS simulations. In comparison, the MCDC
model shows good agreements with the experimental results in
3.2.1. Analysis on the combustion phasing and duration terms of pressure trace and AHRR for most of the operating condi-
Based on the cylinder averaged results from these four methods, tions. In comparison with the DCS model results, the phasing and
it is observed that DCS method, ‘Kong-Reitz’ model and MCDC magnitude of the heat release rates and pressure rises are im-
model methods show capability in predicting the general charac- proved in MCDC model results.
teristic of the combustion process. However, these methods show The improvements of the MCDC model vary between different
different characteristics over these six operating conditions. To operating conditions. For mode 1 (idling operating conditions) in
better quantify these differences, cumulative chemical heat release Fig. 5a, the amount of injected fuel is the smallest among the six
(CCHR) and total chemical heat release (TCHR) are used for the conditions. The relatively longer mixture preparation time be-
analysis on combustion phasing. CCHR is the integral of the chem- tween start of injection and CA 10, which is 1.0 ms or 3 CA at
ical heat release in the cylinder from the start of ignition till the 750 rpm, reduces the stratification of fuel vapor and air. It also re-
sampling time. TCHR is the total chemical heat release from the in- duced the peak temperature where the reaction progress tends to
jected fuel in an engine cycle. By normalizing CHRR with TCHR, the be faster and limited by mixing conditions. For this partially pre-
obtained percentage can be regarded as an indicator of the com- mixed operating condition, the mixing effect is minimal and the
bustion progress. The corresponding crank angle for these percent- DCS results shows a good agreement with experimental results,
ages is a good measure of the combustion phasing analysis for presumably because the SGS mixing is not an important factor in
internal combustion engines. For convenience, the CCHR at the the combustion process. With the mixing time scale based on
Y. Zhang, C.J. Rutland / Fuel 105 (2013) 272–282 277

7 700
Experiment Experiment Experiment Experiment
DCS DCS 7 DCS DCS 1000
MCDC-Baseline MCDC-Baseline 600 MCDC-Baseline MCDC-Baseline
6 MCDC-Coarse MCDC-Coarse MCDC-Coarse MCDC-Coarse
Pressure [Mpa] Kong-Reitz Kong-Reitz 6 800

Pressure [Mpa]
500 Kong-Reitz Kong-Reitz

AHRR [J/ CA]

AHRR [J/ CA]


5

o
400 5 600

4 300
4 400
200
3 3 200
100
(a) (b)
2 0 2 0
-15 -10 -5 0 5 10 15 -10 0 10 20 30
o o
Crank Angle [ aTDC] Crank Angle [ aTDC]

1000
Experiment
1200
10 Experiment Experiment Experiment
DCS
14 DCS
DCS DCS
MCDC-Baseline MCDC-Baseline
MCDC-Baseline
800
MCDC-Baseline 1000
MCDC-Coarse MCDC-Coarse 12 MCDC-Coarse MCDC-Coarse
8 Kong-Reitz Kong-Reitz

Pressure [Mpa]
Pressure [Mpa]

Kong-Reitz Kong-Reitz

AHRR [J/ CA]


AHRR [J/ CA]
10 800
600

o
6 o 8
600
400 6
4 400
4
200
2
200
2
(c) (d)
0 0 0
-10 0 10 20 30 40 -10 0 10 20 30 40 50
o o
Crank Angle [ aTDC] Crank Angle [ aTDC]

10 800 8
Experiment Experiment Experiment Experiment
DCS DCS DCS
600
DCS
MCDC-Baseline MCDC-Baseline 700 7 MCDC-Baseline MCDC-Baseline
8 MCDC-Coarse MCDC-Coarse MCDC-Coarse MCDC-Coarse 500
600
Pressure [Mpa]

Kong-Reitz Kong-Reitz Kong-Reitz


Kong-Reitz 6
Pressure [Mpa]

AHRR [J/ CA]


AHRR [J/ CA]

500 400

o
6
o

5
400 300
4 4
300
200
200 3
2 100
(e) 100 2 (f)
0 0 0
-10 0 10 20 30 40 0 10 20 30 40
o
Crank Angle [ aTDC]
o Crank Angle [ aTDC]

Fig. 5. Experimental, direct chemistry solver (DCS), ‘Kong-Reitz’ model, and mixing controlled direct chemistry (MCDC) predicted in-cylinder pressure trace and apparent
heat release rate (AHRR) from the 6-mode test of Caterpillar 3401B Engine: (a) mode 1 (0% load, 750 rpm); (b) mode 2 (25% load, 953 rpm); (c) mode 3 (75% load, 1074 rpm);
(d) mode 4 (100% load, 1657 rpm); (e) mode 5 (50% load, 1668 rpm); (f) mode 6 (25% load, 1690 rpm).

g
mixture fraction variance Z 002 and dissipation vsgs, the MCDC model in terms of pressure trace and AHRR. For idling condition (mode 1)
shows no excessive correction on the DCS solution and the results in Fig. 5a, because of relatively longer residence time and small
are similar to the DCS model results and experimental results amount of fuel, most of the combustion processes are premixed
(Figs. 4 and 5). and happen at relative lower temperature. Therefore the mixing ef-
As the engine load gradually increases, the difference between fect is less important compared to the kinetic process. As a result,
DCS and MCDC results gradually increased. A good example is the MCDC model does not show significant difference compared to
mode 6 (Fig. 5f), which is a 75% load case with an initial cylinder the ‘Kong-Reitz’ model.
averaged pressure of 132 kPa at IVC. The DCS results show a high For higher load operating conditions (modes 2–6) In Fig. 5b–f, it
initial peak of about 400 J/°CA in AHRR at 9.75° aTDC, which is is found that the MCDC model is able to capture the details of pres-
about 25% higher and 1.5 CA earlier than the experimental results. sure traces compared to experimental results. In comparison, the
In comparison, the MCDC model shows a more accurate prediction ‘Kong-Reitz’ model had a lower prediction in pressure trace for
of the AHRR peak with a difference of 6% lower and 0.25 CA earlier modes 2, 3, 5 and 6 (Figs. 5b–c and e–f) and a higher prediction
than the experimental results. in pressure trace for mode 4 (Fig. 5d), resulting in a late (modes
2, 3, 5 and 6) or early (mode 4) prediction of the AHRR peak. For
3.2.3. MCDC model compared with ‘Kong-Reitz’ model the case with highest engine speed (mode 6) in Fig. 5f, an apparent
As shown in Fig. 5, for most operating conditions, MCDC model delayed prediction of AHRR peak (2 °CA) and mismatch of heat
results show improvements compared with the ‘Kong-Reitz’ model pressure trace with peak difference of 0.75 MPa are found
278 Y. Zhang, C.J. Rutland / Fuel 105 (2013) 272–282

compared to experimental results. We believe the improved re- will increase the scaling coefficient for vsgs in Eq. (8) with the same
sults from the MCDC model are mainly resulted from the improved mixture fraction gradients. This indicates that smixing responds to
formulation of mixing and kinetic time scales, which will be dis- the effects of moving and evaporating droplets on the SGS mixing
cussed in Chapter 4. process. In comparison, much smaller values of sturb (Eq. (12)) are
found in the fuel lean region, which is mainly resulted from the
3.2.4. Grid dependence study of MCDC model small value of delay coefficient f (Eq. (13)) in fuel lean regions.
The grid dependence study is considered to be an important Both of the two chemical time scales show most of the small
part for the development of engine combustion models. The con- values around the stoichiometric value of Zstoichiometric = 0.062. The
sistency of model predictions at different grid resolutions can en- samples of skinetics,1spd (Eq. (9)) distribute in two narrow belt-
hance the applicability of the model for different engine designs. shaped envelopes (Fig. 6(b)). The belt on the left is mainly deter-
For the LES based combustion models, different grid resolutions, mined from sfuel,kinetics (Eq. (10)) and the belt on the right is mainly
or base filter sizes, can result in different effects on the filtered from the sCO,kinetics (Eq. (11)). These belt-shaped envelopes show the
reaction rate from the SGS mixing process. The analysis on the dif- strong species dependence of skinetics,1spd. In comparison, skinetics,U
ference can benefit the understanding of the models as well as the (Eq. (3)) distributes in a wider and fuller envelope (104–102 s)
possible improvements of sub-models such as scalar dissipation without multiple belt-like structures.
rate mixture fraction variance models. The two aforementioned In the middle stage of idling conditions (Fig. 6c–d), both times
grids of Caterpillar 3401™ engine are used for the grid resolution cales skinetics,U and skinetics,1spd are smaller compared to the early
sensitivity study. stage of combustion. However, skinetic,1spd shows a more significant
As shown in Fig. 5, the coarse-grid MCDC results are able to cap- decrease compared to skinetics,U. Both chemical time scales
ture part of the mixing effects compared to experimental results in show most of the small values around the stoichiometric value of
terms of cylinder-averaged quantities, such as pressure trace and Zstoichiometric = 0.062. It is further noticed that kinetic time scales
AHRR. However, these predictions show a peak pressure 0.5 to smaller than the CFD time step is plotted as the time step. In
2.5 °CA earlier compared to the baseline grid MCDC results. Based Fig. 6c–d, more values of skinetics,1spd are equal to or smaller than
g the computational time step (5.0  107 s), indicating that fuel or
on the scaling properties of the algebraic model for Z 002 (Eq. (6))
CO species are both nearly depleted in these cells. This indicates
used in this work, the under-resolved flow field is expected to pro-
g
skinetics,1spd, which is based on the larger value of skinetics,fuel and
vide less accurate SGS quantities, such as Z 002 and smixing. Detailed skinetics,CO, is no longer of value for reference in reaction progress
analysis based on the smixing and skinetics,U are presented in Section 4. quantification.
Even with the difference in the behavior of kinetic time scales,
4. Evaluation of time scales the relative magnitude of mixing time scales compared to kinetic
time scales are similar for the MCDC model and the ‘Kong-Reitz’
Qualitative evaluation of the time scales is performed by sam- model. The mixing time scale (smixing or sturb) has a range between
pling computational cells in the reactive region (mixture fraction one order smaller than the skinetics (near Zstoichiometric) and four or-
greater than 105) of the domain. Samples are taken every quarter ders higher than skinetics (fuel-rich or fuel-lean conditions) for both
of crank angle from the fuel injection till the end of combustion. models. Therefore, the relative magnitudes of smixing compared to
For the early stage of combustion, the samples show that the time skinetics are considered similar from these two simulations based
scales sampled between CA10 and CA20 are similar within either on the basic model formulation (Eq. (1)), resulting in a similar pre-
of these two operating conditions (idling and high load). This sim- diction in mixing effects and combustion phasing.
ilarity is also found for the samples from between CA40 and CA60.
Therefore, two representative samples are taken at CA15 and CA50. 4.1.2. High load conditions (mode 3)
In the early stage of combustion (Fig. 7a and b), smixing shows
4.1. Comparison of the MCDC model and the ‘Kong-Reitz’ model dependence on mixture fraction in a similar way as in idling oper-
ating conditions (Fig. 6a and b). The distribution of Z e is between 0
Among the six operating conditions, cylinder averaged results and 0.4, with a significant amount of the domain in fuel-rich con-
show that the MCDC model and the ‘Kong-Reitz’ model have the ditions. The values of skinetics,1spd distribute in a wider range (105–
smallest differences for mode 1 (Fig. 5a) and larger differences 101 s) compared to the early stage for idling conditions (Fig. 6a
for modes 3, 5 and 6 (Fig. 5c, e and f). Here, the time scales sampled and b), with a belt-like or banded structure in the temperature –
from mode 1 and mode 3 are compared to evaluate the impacts of mixture fraction space. The values of skinetics,U falls in a wider
the time scales on the cylinder-averaged predictions. envelope compared to skinetics,1spd, without obvious structures in
the distribution of skinetics,1spd. One major reason for this is that
4.1.1. Idling conditions (mode 1) the species-dependent skinetics,1spd exaggerates the importance of
In the early stage of combustion, the ignition starts at 0.5 CA changes of single species in the whole reaction process and is more
after the end of injection (Fig. 5a). As shown in Fig. 6a and b, the likely to be banded.
brief and small-amount of fuel injection in mode 1 (Table 2) gives In the middle stage of the combustion (Fig. 7c and d), the values
the mixture a good mixing condition with the distribution of Z e be- of smixing continues to show dependence on Z, e with smaller values
tween 0 and 0.18. The mixture with more homogeneity and longer found in fuel-rich conditions, for a similar reason as in the early
residence time results in more significant heat release in partially stage of the combustion. Both kinetic time scales show similar
premixed conditions, which is represented by a narrow and tall behavior compared to early stage of combustion (Fig. 7a and b) ex-
peak in AHRR trace. cept that skinetics,U shows a wider distribution range (105–1 s),
During early stages of idling conditions, smixing (Eq. (4)) shows without obvious structures in the distribution.
small values in fuel rich regions (Fig. 6a and b). A detailed analysis For both early stage and middle stage of the high load engine
by the authors (not shown) showed that these newly generated va- operating conditions, many of the sampled species-based time
por clouds are generally found with the presence of droplets in the scale skinetics,1spd shows smaller values compared to the local time
early stage of combustion. A further analysis showed that these step (5.0  107 s). The small values are only found in the middle
cells usually showed a higher Resgs because of higher ksgs compared stage of mode 1 cases (Fig. 6c and d). This difference indicates that
to other cells. Based on the scalar property in Fig. 1, a higher Resgs the species-based time scale skinetics,1spd loses the value for reaction
Y. Zhang, C.J. Rutland / Fuel 105 (2013) 272–282 279

e (a) Early stage


Fig. 6. Time scales smixing, sturb, skinetics,U and skinetic,1spd sampled from idling operating condition (mode 1) of Caterpillar™ 3401 SCOTE engine plotted against Z.
of combustion from the MCDC model; (b) early stage of combustion from the ‘Kong-Reitz’ model; (c) middle stage of combustion from the MCDC model; (d) middle stage of
combustion from the ‘Kong-Reitz’ model.

quantification for the high temperature regions where significant range is between 0 and 0.29. This difference is considered to be a
release comes from the oxidation of intermediate species rather result of the enlarged base filter size, which increased the volume
than single species such as fuel or CO. In comparison, skinetics,U re- over which the spatial averaging is performed.
tains the kinetic details at these conditions and can better quantify Second, a smaller estimate of the mixing time scales is found for
the exothermic progress by comparing with the local equilibrium the coarse grid near Zstoichiometric in the early stage of combustion
state. Therefore, skinetics,U is expected to provide a better reaction (Fig. 8a and b). Further analysis shows that a smaller estimate of
process quantification compared to skinetics,1spd. Compared to the g
the mixture fraction variance Z 002 is predicted. The model constant
MCDC model (Fig. 7a), the small values of skinetics,1spd result in a
Cu/Ce in Eq. (6) scales with the g/D (g:local base filter size; D:
higher averaged Damköhler number for ‘Kong-Reitz’ model, which
Kolomogrov length scale). However, the value of local g/D is very
showed a delayed combustion phasing compared to MCDC model.
difficult to be quantified in the current study. Therefore, a constant
For the time scales sampled from MCDC model results in early
is used in Eq. (6) as suggested by Ihme [17]. A dynamic model con-
stage and middle stage of combustion (Fig. 7a and c), a significant
stant to capture this dependence is considered to be necessary in
amount of smixing are comparable or larger than skinetics,U, especially
g
for the cells on the rich side near Zstoichiometric. This indicates that future study to improve the estimate of Z 002 on coarse grids.
mixing has become a limiting factor in near stoichiometric condi- At the middle stage of combustion (CA50, Fig. 8(c-d)), the mixing
tions for this high load mode. time scale smixing distributes in a range between 106 s and 102 s for
both baseline and coarse grid. The kinetic time scale skinetics,U dis-
tributes in a range between 105 s and 102 s. But the percentage
4.2. Grid Dependence of time scales in the MCDC model of ‘mixing-controlled’ cells (skinetics,U > smixing) for the baseline grid
is larger (25%) compared to coarse grid (19%), resulting in more
In terms of goal quantities such as pressure trace and AHRR, the kinetically limited combustion compared to the baseline grid.
coarse-grid MCDC results are able to capture part of the mixing ef- Compared to the early stage (Fig. 8b), the kinetic mixing time scale
fects compared to experimental results (Fig. 5). However, these skinetics,U is greatly reduced in the middle stage of combustion. It is
predictions are early compared to the MCDC results on baseline argued that the fast reaction in the early stage of combustion con-
resolution grid. In this analysis, the full load operating condition sumes larger amount of fuel and consequently yield higher tempera-
(mode 4, 100% load) is used, because the highly stratified mixture ture than experimental results. In return, these conditions reduce the
from the strong injection is expected to show the different behav- energy level difference (numerator of Eq. (3), highly cell fuel concen-
iors of time scales from different grid resolutions. At the early stage tration dependent) and enhance the reaction rate (denominator of
of combustion (CA15, Fig. 8a and b), two differences are found be- Eq. (3), highly temperature dependent). Therefore the mean kinetic
tween the baseline grid and coarse grid. time scale is reduced compared to the kinetic time scales from
First, two time scales on the coarse version of grid distribute in a baseline grid at middle stage of combustion (Fig. 8c and d).
smaller mixture fraction range. About 95% of the reactive cells have The differences between the time scales on baseline grid and
a mixture fraction between 0 and 0.24. For the baseline grid, this coarse grid at early stage and middle stage of combustion are
280 Y. Zhang, C.J. Rutland / Fuel 105 (2013) 272–282

e (a) Early
Fig. 7. Time scales smixing, sturb, skinetics,U and skinetic,1spd sampled from full load operating condition (mode 4) of Caterpillar™ 3401 SCOTE engine plotted against Z.
stage of combustion from the MCDC model; (b) early stage of combustion from the ‘Kong-Reitz’ model; (c) middle stage of combustion from the MCDC model; (d) middle
stage of combustion from the ‘Kong-Reitz’ model.

believed to result in the difference in cylinder averaged pressure 2. Compared to the original RANS-based ‘Kong-Reitz’ model, the
and HRR traces. Compared to the experimental results and the LES-based MCDC combustion model shows improved predic-
baseline grid results, the coarse grid results generally show an tions in terms of pressure traces and heat release rates. These
early and premature heat release in the early stage of combustion, improvements are considered to benefit from the mixing time
and a lower heat release rate at middle stage and late stage of scales and kinetic time scales formulated for the MCDC model.
combustion. g
For mixing time scales, the MCDC model includes Z 002 and vsgs.
These two parameters directly characterize the inhomogeneity
5. Conclusions and mixing rate at SGS level and formulate a mixing time scale
of strong physical sense. In comparison, the mixing time scale
An evaluation of the mixing controlled direct chemistry (MCDC) used by the ‘Kong-Reitz’ model is scaled from the turbulent
model has been performed on a single cylinder oil test engine kinetic energy time scale (Eq. (12)), which is not directly related
(SCOTE) of Caterpillar™ 3401 heavy-duty diesel engine over a wide to the local mixing status. This time scale requires small model
range of operating conditions. By comparing with DCS model and coefficients and the complexity of progress variable in the formu-
‘Kong-Reitz’ model, the MCDC model shows good predictions of lation (Eqs. (13) and (14)), which are mainly validated on HCCI
combustion phasing and duration on the baseline grid. Major re- type of combustion without strong mixing processes [24]. The
sults are summarized as follows: specific internal energy based kinetic time scale skinetics,U shows
different behaviors compared to a single species based time scale
1. Over a wide range of operating conditions, the MCDC model (Figs. 6–8). The values of skinetics,U show dependence on the
shows more accurate predictions of the diesel engine combus- mixture fraction with smaller values around the stoichiometric
tion process in terms of in cylinder heat release rate compared mixture fraction Zstoichiometric. As temperature increases, cylinder
to the DCS model. The model includes DCS based kinetics compu- averaged skinetics,U becomes smaller and the distribution
tations and modeling of SGS mixing conditions. The accurate envelope is wider. In comparison with the mixing time scale
chemical kinetics computation is an important part of MCDC smixing, skinetics,U is generally larger except in high-temperature
model. For early stages of combustion, the kinetic computations conditions around Zstoichiometric. At these conditions, the combus-
are important for the accurate prediction of ignition. They are tion could be regarded as controlled by SGS mixing conditions.
also important for later stages of combustion by providing neces- 3. In the grid dependence study of the MCDC model, the coarse
sary species information. For middle stage of combustion, mixing grid shows an early prediction of the pressure trace rise and a
time scale plays an important role in modeling the mixing limit- slightly premature AHRR. The time scale analysis shows that
ing effects in the conditions where chemistry is much faster. The the coarse grid failed to provide SGS level information to quan-
model, which incorporates effects of mixing, performs better tify local mixing and kinetic processes at a good accuracy,
than the alternatives considered in the conditions considered. resulting in premature prediction in heat release at the early
Y. Zhang, C.J. Rutland / Fuel 105 (2013) 272–282 281

e (a) Early
Fig. 8. Time scales smixing, sturb, skinetics,U and skinetic,1spd sampled from 50% load operating condition (mode 5) of Caterpillar™ 3401 SCOTE engine plotted against Z.
stage of combustion from baseline grid; (b) early stage of combustion from coarse grid; (c) middle stage of combustion from baseline grid; (d) middle stage of combustion
from coarse grid.

stage of combustion and an overestimate of the mixing effect at [4] Curran HJ, Gaffuri P, Pitz WJ, Westbrook CK. A comprehensive modeling study
of n-heptane oxidation. Combust Flame 1998;114(1–2):149–77.
the middle stage of combustion. However, a systematic study
[5] Ra Y, Reitz RD. A reduced chemical kinetic model for IC engine
on the grid resolution requirement of MCDC model is still combustion simulations with primary reference fuels. Combust Flame
needed for a fundamental understanding of the model behavior. 2008;155(4):713–38.
The sensitivity of the mixing time scale on grid resolution and [6] Haworth DC, Jansen K. Large-eddy simulation on unstructured deforming
meshes: towards reciprocating IC engines. Comput Fluids
its impact on the overall prediction of combustion process will 2000;29(5):493–524.
be the interest in future studies. [7] Janicka J, Sadiki A. Large eddy simulation of turbulent combustion systems.
4. Further improvements are need for the existing model of sub- Proc Combust Inst 2005;30(1):537–47.
[8] Bharadwaj N, Rutland CJ, Chang S-m. Large eddy simulation modelling of
grid scale mixture fraction variance. In this model, the droplet spray-induced turbulence effects. Int J Engine Res 2009;10(2):97–119.
evaporation process is not considered in the model formulation. [9] Pitsch H. Large eddy simulation of turbulent combustion. Annu Rev Fluid
g Mech 2006;38(1):453–82. http://dx.doi.org/10.1146/annurev.fluid.38.050304.
A transport equation based model for Z 002 is considered to be 092133.
promising to incorporate the effects of droplet evaporation on [10] Rutland CJ. Large-eddy simulations for internal combustion engines-a reveiw.
Int J Engine Res 2011;12(5):421–51.
the mixture fraction field [15].
[11] Zhang Y, Rutland CJ. A mixing controlled direct chemistry (MCDC) model for
diesel engine combustion modelling using large eddy simulation. Combust
Theor Model 2012;16(3):571–88. 0(1080/13647830), 2011, 642819.
[12] Zhang Y, Rutland CJ. LES scalar dissipation rate and combustion modeling for
Acknowledgements diesel engines. In: 7th US national technical meeting of the combustion
institute, atlanta, georgia, USA, 2G06; 2011.
[13] Banerjee S, Rutland CJ. Modeling combustion control for high powered diesel
This study is supported by the US Department of Energy under mode switching using large eddy simulation. ERC Symposium, Madison,
Contract DEFC26-06NT42628. We also heartily thank Joshua Leach Wisconsin, USA; June 2011.
and Dr. Youngchul Ra for computer supports and inspiring discus- [14] Pope SB. CEQ: A Fortran library to compute equilibrium compositions using
Gibbs function continuation, <http://eccentric.mae.cornell.edu/~pope/CEQ>;
sions respectively. 2003 [accessed 03.01.11].
[15] Pera C, Réveillon J, Vervisch L, Domingo P. Modeling subgrid scale
mixture fraction variance in LES of evaporating spray. Combust Flame
References 2006;146(4):635–48.
[16] Wiles MA, Probst DM, Ghandhi JB. Bulk cylinder flowfield effects on mixing in
[1] Dec JE. Advanced compression-ignition engines– understanding the in- DISI engines. SAE technical paper, 2005-01-0096; 2005.
cylinder processes. Proc Combust Inst 2009;32(2):2727–42. [17] Ihme M. Pollutant formation and noise emission in turbulent non-premixed
[2] Dec JE, Hwang W, Sjöberg M. An Investigation of thermal stratification in HCCI flames. Ph.D thesis, Stanford University; 2007. p. 1–203.
engines using chemiluminescence imaging. SAE technical paper, 2006-01- [18] Pope SB. Turbulent flow. United Kingdom, London: Cambridge University
1518; 2006. Press; 2000.
[3] Westbrook CK. Chemical kinetics of hydrocarbon ignition in practical [19] Fox RO. Computational models for turbulent reacting flows. London, United
combustion systems. Proc Combust Inst 2000;28(2):1563–77. Kingdom: Cambridge University Press; 2003.
282 Y. Zhang, C.J. Rutland / Fuel 105 (2013) 272–282

[20] Chumakov SG, Rutland CJ. Dynamic structure subgrid-scale models for large [31] Li YH, Kong SC. Diesel combustion modeling using LES turbulence model with
eddy simulation. Int J Numer Meth Fluids 2005;47(8–9):911–23. detailed chemistry. Combust Theory Model 2008;12(2):205–19.
[21] Leonard A, Frenkiel FN, Munn RE. Energy cascade in large-eddy simulations of [32] Menon S, Yeung PK, Kim WW. Effect of subgrid models on the computed
turbulent fluid flows. In: Frenkiel FN, Munn RE, editors. Advances in interscale energy transfer in isotropic turbulence. Comput Fluids
geophysics. Elsevier; 1975. p. 237–48. 1996;25(2):165–80.
[22] Chumakov SG. Scaling properties of subgrid-scale energy dissipation. Phys [33] Werner H, Wengle H. Large-eddy simulation of turbulent flow over and around
Fluids 2007;19(5). p. 058104. a cube in a plate channel. In: 8th Symp. on Turbulent Shear Flows, Munich,
[23] Zhang Y, Petersen B, Ghandhi JB, Rutland CJ. Large eddy simulation of scalar Germany, Septermper 9–11, 1991. p. 19-4-1–6.
dissipation rate in an internal combustion engine. SAE technical paper, 2010- [34] Reitz RD, Rutland CJ. Development and testing of diesel engine CFD models.
01-0625; 2010. Prog Energy Combust Sci 1995;21:173–96.
[24] Kong SC, Marriott CD, Reitz RD, Christensen M. Modeling and experiments of [35] Amsden AA, Ramshaw JD, O’Rourke PJ, Dukowicz JK, KIVA: A computer
HCCI engine combustion using detailed chemical kinetics with program for two- and three-dimensional fluid flows with chemical reactions
multidimensional CFD. SAE technical paper, 2001-01-1026; 2001. and fuel sprays. Tech. Rep. LA-10245-MS, Los Alamos National Laboratory;
[25] Pope SB. CEQ: A fortran library to compute equilibrium compositions using 1985.
Gibbs function continuation. [accessed: 2003]. [36] Munnannur A, Reitz RD. A comprehensive collision model for multi-
[26] Montgomery DT, Reitz R. Six-mode cycle evaluation of the effect of EGR and dimensional engine spray computations. Atomization Sprays
multiple injections on particulate and NOx emissions from a DI diesel engine. 2009;19(7):597–619. http://dx.doi.org/10.1615/AtomizSpr.v19.i7.10.
SAE technical papers, 960316; 1996. [37] Lefebvre H. Atomization and sprays, combustion: an international series.
[27] Hu B, Rutland CJ, Shethaji TA. A mixed-mode combustion model for large-eddy Taylor & Francis 1989.
simulation of diesel engines. Combust Sci Technol 2010;182(9):1279–320. [38] Sarre CVK, Kong SC, Reitz RD. Modeling the effects of injector nozzle geometry
[28] Banerjee S. Study of low temperature combustion using large eddy on diesel sprays. SAE technical paper, 1999-01-0912; 1999.
simulations. Ph.D thesis, University of Wisconsin-Madison; 2011. [39] Kee RJR, Rupley FM, Miller JA. CHEMKIN-II: a Fortran chemicalkinetics package
[29] Amsden A. KIVA3V: A block-structured KIVA program for engines with for the analysis of gas-phase chemical kinetics, Sandia National Laboratory,
vertical or canted valves. Los Alamos National Laboratory Report, LA-13313- Report, SAND89-8009; 1991.
MS; 1993. [40] Amsden AA, Ramshaw JD, et al. KIVA: A computer program for two and three-
[30] Pomraning E, Rutland CJ. A dynamic one-equation non-viscosity LES model. dimensional fluid flows with chemical reactions and fuel sprays. Tech. Rep. LA-
AIAA J 2002;44:689–701. 10245-MS, Los Alamos National Laboratory; 1985.

Vous aimerez peut-être aussi