Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Direct Numerical Simulation of ignition front propagation in a

constant volume with thermal stratification

Evatt R. Hawkes∗, Ramanan Sankaran, Jacqueline H. Chen,


Combustion Research Facility
Sandia National Laboratories
Livermore, CA, 94551

and Hong G. Im
Department of Mechanical Engineering
University of Michigan, Ann Arbor, MI 48109-2125
Abstract
The influence of thermal stratification on auto-ignition at constant volume and high pressure is studied
by Direct Numerical Simulation (DNS) with complex H2/air chemistry with a view to providing better
understanding of combustion processes in homogeneous charge compression ignition engines. In particular
the dependence of overall ignition progress on initial mixture conditions is determined. The propagation
speed of ignition fronts that emanate from “hot spots” given by a temperature spectrum is monitored
by using the displacement velocity of a scalar that tracks the location of maximum heat release. The
evolution of the front velocity is compared for different initial temperature distributions and the role of
scalar dissipation of heat and mass is identified. It is observed that both deflagrative as well as spontaneous
ignition front propagation occur depending upon the local temperature gradient. It is found that the
ratio of the instantaneous front speed to the deflagrative speed is a good measure of the local mode
of propagation. This is verified by examining the energy and species balances. A parametric study in
the amplitudes of the initial temperature fluctuation is performed and shows that this parameter has a
significant influence on the observed combustion mode. Higher levels of stratification lead to more front-like
structures. Predictions of the multi-zone model are presented and explained using the diagnostics developed.

Introduction geneous charge at constant volume is crucial in the


development of control strategies for HCCI engines.
Homogeneous charge compression ignition (HCCI)
engines are considered a viable concept as an alter-
Specific Objectives
native to diesel engines [1]. At present, the com-
bustion mode in HCCI engines is not well under-
stood, as both volumetric and front-like combustion The overall goals of the present study are to under-
modes can occur [2]. A major challenge posed by stand the influence of temperature inhomogeneities
this method of combustion is to control the heat re- on characteristics of ignition and subsequent com-
lease rate, and in particular, to provide a means to bustion in an HCCI-like environment, and to pro-
spread it out over several crank angle degrees, sup- vide information relevant to the modeling of HCCI
pressing the occurrence of damaging engine knock. combustion. The work is an extension of our pre-
One possible control strategy is to introduce inhomo- vious study Sankaran et al. [3]. The evolution of
geneities in the temperature or mixture composition autoignition from an initial spectrum of “hot spots”
in order to produce the desired heat release rate [1]. at high pressure is simulated using two-dimensional
Moreover, incomplete turbulent mixing and temper- DNS with a detailed hydrogen/air reaction mecha-
ature stratification between the bulk gases and the nism. The two dimensional approximation and the
cylinder wall lead to spatial nonuniformities that also use of simple hydrogen-air kinetics are employed to
contribute to a range of combustion modes that are limit the computational cost to feasible levels.
distinct from homogeneous autoignition. Therefore, One of the primary objectives in this study is to
a better understanding of autoignition of an inhomo- develop systematic and rational methods to identify
∗ Corresponding author: erhawke@ca.sandia.gov and analyze various regimes of ignition and subse-
Associated Web site: http://www.ca.sandia.gov/crf/ quent combustion processes. In particular we are
Proceedings of the European Combustion Meeting 2005 interested in characterizing the role of dissipation of

1
heat and mass. These effects are divided conceptu- very high front speeds, which correspond to low ini-
ally into transport within ignition fronts that em- tial temperature gradients, molecular diffusion is in-
anate from “hot spots” and passive scalar dissipa- significant and the front propagates as a spontaneous
tion, which modifies the statistics of the pre-ignition ignition front. However, as initial temperature gra-
temperature field. The significance of passive scalar dients are increased, the speed of the ignition front
mixing is examined using a mixing timescale based is reduced and there is sufficient time for the trans-
on enthalpy fluctuations. Transport within igni- port of heat and radicals to the unburned reactants,
tion fronts is analyzed by monitoring the propaga- resulting in a greater importance of diffusion. At suf-
tion speed of ignition fronts using the displacement ficiently low speeds, a reaction-diffusion balance oc-
speed of a scalar that tracks the location of maxi- curs as in normal premixed flame. These theoretical
mum heat release. The theoretical basis stems from arguments demonstrate the importance of the front
an earlier study by Zel’dovich [4], who described ig- speed as an indicator of the transition of the com-
nition of a nonuniform mixture in several distinct bustion mode from spontaneous ignition front prop-
regimes of propagation, including deflagration, spon- agation to diffusion controlled deflagration.
taneous ignition front propagation, and detonation. If an ignition delay time is defined for each fluid
He further showed that the spontaneous ignition parcel as the time taken for a scalar φ to reach a
front speed is inversely proportional to the initial critical level φc , then the speed sd of the ignition
temperature gradient, suggesting a characterization front may be identified with the displacement speed
of ignition regimes based on the speed of the igni- [14] of the iso-contour φ = φc . A density-weighted
tion front. The concept has been applied to iden- speed can be defined as s∗d = ρsd /ρ0 , where the ρ0 is
tifying ignition regimes in various one-dimensional a representative density of the fresh charge. Statis-
configurations [5, 6, 7, 8]. The present work extends tics were extracted from the YH2 = 8.5 × 10−4 iso-
the idea into more realistic two-dimensional ignition contour, which corresponds approximately to the lo-
problems, where details of multi-dimensional aspects cation of maximum heat release through most of the
including wrinkling and mutual interaction of fronts simulation. In this study s∗d is compared with the
are fully accounted for. steady laminar flame speed to study the transition
The second main objective is to provide informa- between deflagration and spontaneous ignition front
tion relevant to the modeling of HCCI. In the case of propagation.
purely homogeneous combustion, the modeling prob-
lem reduces to simple homogeneous ignition calcula- Numerical Implementation and Initial Condi-
tions. However, stratification of the charge, whether tions
in terms of temperature or mixture composition, in-
troduces a range of ignition delay times into the The autoignition of a lean hydrogen-air mixture
cylinder, which must be accounted for in a model. in a closed volume in the presence of an inhomo-
At low levels of stratification or long stratification geneous temperature field is simulated. A uniform
length scales, molecular mixing effects can be ne- hydrogen/air mixture with a fuel-air equivalence ra-
glected [4, 9]. Presuming that length scales are not tio of 0.1 is chosen. In all cases presented, the mean
so long so that compressible flow effects are impor- initial temperature is 1070K, and the uniform initial
tant, a reasonable approximation is then that differ- pressure is 41 atm. The DNS code, along with the
ent locations in the cylinder interact only by pres- transport and chemical kinetic approximations em-
sure work. This modeling strategy, referred to as the ployed are described in our previous study [3]. The
multi-zone model, has been advanced by Aceves et turbulent flow field is prescribed by an initial tur-
al. [10, 11, 12]. Using the DNS as a numerical ex- bulent kinetic energy spectrum function [13] and a
periment, and the diagnostics of the front speed and random temperature field is superimposed on a mean
mixing timescales as indicators of the role of dissipa- temperature field. The temperature spectrum, which
tion, we evaluate the performance of the model, and is similar to the kinetic energy spectrum, is used to
provide understanding of its limitations. specify the characteristic scales of the initial exother-
mic spots.
The initial autocorrelation integral scale of the ve-
Front Speed locity fluctuations is 0.34mm, and the most energetic
length scale is 1.0mm. The velocity fluctuation is 0.5
Consider the autoignition of a thermally stratified m/s. These parameters were selected so that the tur-
homogeneous mixture. The mixture initially ignites bulence integral time scale based on the most ener-
at the hottest location, and subsequently a combus- getic length scale is comparable to the homogeneous
tion wave forms and propagates. The speed of the ignition delay of the mixture at the same mean tem-
advancing combustion wave is used in this study as perature and pressure. The homogenous ignition de-
an indicator of the combustion regime and the role lay is τ0 = 2.9ms. The temperature field has an auto-
of molecular transport within ignition fronts, follow- correlation length scale of 0.15mm, and the most en-
ing the theoretical arguments of Zel’dovich [4]. For ergetic length scale is 1.32mm. A parametric study

2
is conducted for the root-mean-square temperature
fluctuation, T0 . Otherwise identical cases are initial-
ized with T0 of 3.75K, 7.5K, 15.0K, and 30.0K. The
cases are referred to by these amplitudes. The case
having T0 =15.0K was used as Case A in our previous
study [3]. The domain is 4.1mm in each of the two
spatial directions. A grid spacing of 4.3 microns was
required to resolve the ignition fronts at this pres-
sure.

Results and Discussion

DNS results - T0 = 15K


In this section, the utility of the ignition front
speed and species budget to delineate between de-
flagrative versus spontaneous ignition front propaga-
tion is assessed using the case having T0 = 15K.
The ignition process proceeds as follows [3]. Af-
ter an initial ignition delay, which is governed by the
competition between reaction and diffusion in sev-
eral candidate kernels, the strongest ignition kernels
ignite and form combustion fronts. The thickness of
Figure 1: Heat release iso-contours (rainbow color scale -
these fronts varies significantly and determines their
red denotes maximum heat release), YH2 surface iso-contour
propagation speed and the importance of diffusive
(white lines: s∗d > 1.1sL , black lines: s∗d ≤ 1.1sL ), and loca-
transport. As fronts propagate, the remaining charge
tion of representative front structures A: spontaneous ignition
is heated by compression, accelerating the ignition
front (B,C): deflagration fronts.
of later-igniting kernels. Near the peak heat release,
fronts begin to merge and annihilate, and eventually
the end-gas is consumed in a more volumetric pro- dient. These data correspond to a spontaneous ig-
cess. nition front in that the effects of diffusion are found
Instantaneous solution fields are analyzed to in- to be much smaller compared to reaction. Cut B,
vestigate the characteristics of the front propaga- on the other hand, shows a front which has a speed
tion. The theoretical arguments of Zel’dovich in- approximately equal to sL . In this case the temper-
dicate that the front speed could be used as a di- ature gradient is higher, and diffusion becomes more
agnostic of the transition between deflagration and significant. Finally, cut C shows an area with a much
ignition front propagation regimes. Therefore, we larger temperature gradient. At this location, where
define a cut-off criterion, s∗d = 1.1sL , to be a reason- s∗d ≈ sL , diffusion nearly counterbalances reaction,
able boundary to distinguish between the two igni- and the front behaves much like a normal deflagra-
tion regimes. Figure 1 shows the heat release (color) tion.
and YH2 = 8.5×10−4 (white/black lines) iso-contours
at 0.83τ0 when ignition kernels are forming and igni- 10
Budget Terms (1/sec)

A B C 5
|∇T| (10 K/mm)

tion fronts have begun to propagate. This instanta-


5 4
neous field is representative of the events during the
interval of major heat release. The figure shows that 0 3
3

this iso-contour selection provides good tracking of 2


-5
the maximum heat release location. Similar results
are obtained from the inception of front propagation -10 1
through the burnout of the end-gas. Along the YH2 0 0.015 0.03 0 0.015 0.03 0 0.015 0.03
iso-contour lines, the black and white segments de- x (cm) x (cm) x (cm)
note the deflagration (s∗d ≤ 1.1sL ) and spontaneous
ignition regime (s∗d > 1.1sL ), respectively, thereby Figure 2: a) Structure for YH2 budget terms and b) temper-
indicating the relative importance of the two regimes ature gradient for locations A, B and C in Figure 1. Propa-
at a given instant in time. gation is to the left. Dashed line: Reaction. Solid line: H2
To underscore the importance of the front speed in Diffusion. Dotted line: |∇T |.
determining the propagation regime, Figure 2 shows
the reaction-diffusion budget for the evolution of H2
mass fraction. Cut A in Fig. 1 represents a spon- DNS results - variation of T0
taneous ignition front, having a speed of approxi- The parametric study examines the influence of
mately 2.2 sL , with a relatively low temperature gra- the amplitude of the temperature fluctuations - all

3
other parameters remain fixed. Figure 3 shows the 1.16
mean heat release as a function of time for the dif- 1.08
ferent cases, referred to by the initial temperature 0.99
fluctuation. In this figure and throughout the paper, 0.91
0.83
the heat release is normalized by the maximum heat 0.74
release of the constant volume ignition at the mean 0.66
temperature, H.R.0 , and time is normalized by the 0.58
homogeneous ignition delay time τ0 , defined as the 0.50
time to the point of maximum heat release. Two 0.41
0.33
points are immediately obvious. First, the larger 0.25
temperature fluctuations spread out the heat release 0.17
and reduce its peak. A less obvious result is that 0.08
the peak heat release is advanced for larger T0 . This 0.00
is primarily because early igniting regions compress
the later igniting regions, accelerating the onset of
ignition. A secondary reason is that effects of molec-
ular transport within propagating fronts also accel-
erate the combustion of later igniting kernels. These
results demonstrate that there is a strong influence
of the initial temperature distribution on the tim-
ing and duration of combustion. For practical ap-
plications, the result also implies that increasing in-
cylinder temperature fluctuations near top dead cen-
ter would be an effective way of smoothing the heat-
release profile [1].
Heat Release / H.R. 0

0.6 T’ = 3.75 K
T’ = 7.50 K
0.4 T’ = 15.0 K
T’ = 30.0 K
0.2

00 0.2 0.4 0.6 0.8 1


Time / τ0

Figure 3: Normalized mean heat release versus time.

Next, the nature of the combustion mode for


the different cases is assessed. Figure 4 shows iso-
contours of heat release rate at the time of maximum
heat release for each case. It is evident that T0 has
a large effect on the character of the combustion.
Lower values of T0 result in a more homogeneous,
volumetric, combustion process, while larger values
tend to promote front-like structures.
The fraction of the front length flagged as defla-
grations using the criteria s∗d /sL < 1.1 is compared
between the different cases in Figure 5. Case T0 =3.75
has no flame elements propagating in a diffusion-
controlled regime - it burns purely by spontaneous ig-
nition front propagation. Comparing Cases T0 =7.5,
T0 =15.0, and T0 =30, it is noted that the amount of Figure 4: Normalized heat release iso-contours for Cases
deflagration observed is strongly dependent on the T0 =3.75K, T0 =7.5K, T0 =15.0K, and T0 =30.0K, shown in this
thermal stratification, and increases with increased order from top to bottom (rainbow color scale - white denotes
thermal stratification. maximum heat release). The size of the computational domain
The other controlling parameter for the impor- is 4.1mm.

4
1 gressively deteriorates as the temperature fluctuation
is increased. This is attributed to the increasing im-
Fraction of Length
portance of deflagrations in the calculations. It is
0.8 noted that the magnitude of the errors is qualita-
tively consistent with the predicted fraction of defla-
gration.
0.6 T’ = 3.75 K
T’ = 7.50 K
T’ = 15.0 K Conclusions
T’ = 30.0 K
0.4 Direct numerical simulations of lean hydrogen-air
ignition at high pressure at constant volume in the
0.2 presence of temperature inhomogeneities have been
conducted to begin to understand the combustion
process in HCCI engines. The influence of the initial
0 thermal stratification on the overall ignition modes
0.4 0.6 0.8 1 and the predictions of the multi-zone model were as-
Time / τ0 sessed.
Two diagnostic indicators to assess the role of
Figure 5: Fraction of front length having s∗d /sL < 1.1. transport of heat and mass were developed. Front
speeds were used to characterize the importance
of transport within ignition fronts, while a mixing
timescale was used to understand passive scalar mix-
tance of molecular diffusion effects is the mixing ing. The front speed was found to be a good mea-
timescale. At the beginning of the simulation, since sure of the local mode of propagation, which was
the initial temperature fluctuation fields differ only verified examining budgets of the species evolution
by a scale factor, the timescales are identical. Later equations.
they diverge due to the different combustion process
A parametric study was performed by varying the
encountered in each case, but always remain compa-
amplitude of the temperature fluctuation. This was
rable. The spread of mixing timescales at the point of
shown to significantly affect the observed combustion
maximum heat release in each case is approximately
mode, with low stratification resulting in distributed
15%. The mixing timescales are initially compara-
autoignitive combustion modes, and higher strati-
ble to the ignition delay but decrease in time due to
fication resulting in combustion in front-like struc-
the effects of turbulent straining to a minimum of
tures. The stratification also affected the perfor-
approximately one fifth of the ignition delay time.
mance of the multi-zone model. Improved predic-
Hence passive scalar dissipation effects are projected
tions were obtained for lower levels of stratifications.
to be significant and nearly equally so in all cases.
The observed predictions were explained using the
However, it was already observed that the impor-
diagnostic indicators developed.
tance of deflagrations differed between the cases.
Having examined the front speeds and mixing 1 Acknowledgments
timescales, we now present the predictions of the
multi-zone model for each case. The multi-zone cal- The work at UM was supported by the Consor-
culations are initialized from the DNS field at the tium on HCCI Engine Research directed by the UM
initial time of the simulation (denoted MZ 0%), and and funded by the Department of Energy, and also
from the time of 10% pressure rise (denoted MZ by Department of Energy, Office of Basic Energy Sci-
10%). Figures 6(a)-(d) show the multi-zone predic- ences, SciDAC Computational Chemistry Program.
tions for the cases with different amplitudes of tem- Sandia National Laboratories (SNL) is a multipro-
perature fluctuations. Overall, the predictions start- gram laboratory operated by Sandia Corporation, a
ing from the initial condition of the DNS are poor Lockheed Martin Company, for the United States
in all cases, worsening for cases with larger temper- Department of Energy under contract DE-AC04-94-
ature gradient. The large discrepancy is a result of AL85000. The work at SNL was supported by the
not accounting for changes in the effective PDF of the Division of Chemical Sciences, Geosciences and Bio-
initial temperature fluctuations due to scalar mixing. sciences, the Office of Basic Energy Sciences, the U.
This is evident because it was found that for Cases S. Department of Energy. Calculations were per-
T0 =3.75K and T0 =7.5K deflagrations are insignifi- formed at SNL and at the National Energy Research
cant. When the calculations are started from 10% Scientific Computing Center, which is supported by
pressure rise, and the most significant scalar mix- the Office of Science of the U.S. Department of En-
ing has already occurred, the predictions improve in ergy under Contract No. DE-AC03-76SF00098. The
all cases. For Case T0 =3.75K, the prediction is very High Performance Computing and Networking de-
good. However it is noted that the prediction pro- partment at SNL provided access to a 256 processor

5
1 DNS T’ = 3.75 K
Infiniband testbed. The authors acknowledge fruit-
MZ-0% T’ = 3.75 K ful discussions with Drs. John Hewson, John Dec
Heat Release / H.R. 0
0.8 MZ-10% T’ = 3.75 K and Magnus Sjöberg.

References
0.6
[1] K. Epping, S. Aceves, R. Bechtold, J. Dec, SAE
0.4 Technical Paper 2002-01-1923.

[2] A. Hultqvist, M. Christenson, B. Johansson, M.


0.2
Richter, J. Nygren, J. Hult, M. Alden, SAE
Technical Paper 2002-01-0424.
0
0.6 0.8 1
a) Time / τ0 [3] R. Sankaran, H.G. Im, E.R. Hawkes, J.H. Chen,
Proc Combust. Inst. 30 (2004) to appear
DNS T’ = 7.50 K
MZ-0% T’ = 7.50 K [4] Ya. B. Zel’dovich, Combust. Flame 39 (1980)
0.6
Heat Release / H.R. 0

MZ-10% T’ = 7.50 K 211-214.

[5] D. Bradley, C. Morley, X.J. Gu, D.R. Emerson,


0.4 SAE Technical Paper 2002-01-2868.

[6] X.J. Gu, D.R. Emerson, D. Bradley, Combust.


0.2 Flame 133 (2003) 63-74

[7] J. H. Chen, S. D. Mason, J. C. Hewson, Pro-


ceedings of the Third Joint Sections Meeting of
0 the U.S. Sections of the Combustion Institute,
0.6 0.8 1
b) Time / τ0 (2003).
0.5 DNS T’ = 15.0 K [8] R. Sankaran, H.G. Im, submitted to Combust.
MZ-0% T’ = 15.0 K Theory and Modelling, (2004).
Heat Release / H.R. 0

0.4 MZ-10% T’ = 15.0 K


[9] A. Liñan, F.A. Williams, Combust. Sci. Tech.
105 (1995) 245-263.
0.3
[10] S.M. Aceves, D.L. Flowers, C.K. Westbrook,
0.2 J.R. Smith, W. Pitz, R. Dibble, M. Christensen,
and B. Johansson, SAE Paper 2000-01-0327,
0.1 (2000).

[11] S.M. Aceves, D.L. Flowers, J. Martinez-Frias,


0 J.R. Smith, C.K. Westbrook, W. Pitz, R. Dib-
0.4 0.6 0.8 1 1.2
c) Time / τ0 ble, J.F. Wright, W.C. Akinyemi, R.P. Hessel,
0.25 SAE Paper 2001-01-1027, (2001).
DNS T’ = 30.0 K
MZ-0% T’ = 30.0 K [12] S.M. Aceves, D.L. Flowers, F. Espinosa-Loza, J.
Heat Release / H.R. 0

0.2 MZ-10% T’ = 30.0 K


Martinez-Frias, R.W. Dibble, M. Christensen,
B. Johansson, R.P. Hessel, SAE Paper 2002-01-
0.15 2869, (2002).

[13] Hinze, J. O., Turbulence, McGraw-Hill, New


0.1
York, (1975).
0.05 [14] T. Echekki, J.H. Chen, Combust. Flame 118
(1999) 308-311.
0
d) 0.2 0.4 0.6 0.8 1 1.2 1.4
Time / τ0

Figure 6: Normalized mean heat release versus time for the


DNS and the multi-zone model started from the DNS initial
condition (MZ 0%) and at 10% pressure rise (MZ 10%): a)
Case T0 =3.75K, b) Case T0 =7.5K, c) Case T0 =15K, d) Case
T0 =30K.

You might also like