Zhang Et Al 2021 - Modeling SD in FSD

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Modeling the displacement speed in the

flame surface density method for turbulent


premixed flames at high pressures
Cite as: Phys. Fluids 33, 045118 (2021); https://doi.org/10.1063/5.0045750
Submitted: 29 January 2021 . Accepted: 25 March 2021 . Published Online: 13 April 2021

Shiming Zhang (张师茗 ), Zhen Lu (卢臻 ), and Yue Yang (杨越 )

COLLECTIONS

Paper published as part of the special topic on In Memory of Edward E. (Ted) O'Brien

This paper was selected as Featured

ARTICLES YOU MAY BE INTERESTED IN

A near-wall predictive model for passive scalars using minimal flow unit
Physics of Fluids 33, 045119 (2021); https://doi.org/10.1063/5.0047472

A scaling improved inner–outer decomposition of near-wall turbulent motions


Physics of Fluids 33, 045120 (2021); https://doi.org/10.1063/5.0046502

The scales of the leading-edge separation bubble


Physics of Fluids 33, 045101 (2021); https://doi.org/10.1063/5.0045204

Phys. Fluids 33, 045118 (2021); https://doi.org/10.1063/5.0045750 33, 045118

© 2021 Author(s).
Physics of Fluids ARTICLE scitation.org/journal/phf

Modeling the displacement speed in the flame


surface density method for turbulent premixed
flames at high pressures
Cite as: Phys. Fluids 33, 045118 (2021); doi: 10.1063/5.0045750
Submitted: 29 January 2021 . Accepted: 25 March 2021 .
Published Online: 13 April 2021

Shiming Zhang (张师茗),1 Zhen Lu (卢臻),1 and Yue Yang (杨越)2,a)

AFFILIATIONS
1
State Key Laboratory for Turbulence and Complex Systems, College of Engineering, Peking University, Beijing 100871, China
2
State Key Laboratory for Turbulence and Complex Systems, College of Engineering, Peking University, Beijing 100871, China and
CAPT and BIC-ESAT, Peking University, Beijing 100871, China

Note: This paper is part of the special topic, In Memory of Edward E. (Ted) O'Brien.
a)
Author to whom correspondence should be addressed: yyg@pku.edu.cn

ABSTRACT
We propose a model of the local displacement speed for the large-eddy simulation (LES) of turbulent premixed combustion with the flame
surface density (FSD) method. This model accounts for flame stretch and curvature effects in order to improve the prediction of LES–FSD
for turbulent premixed flames with the non-unity Lewis number at high pressures. The proposed model is validated by a priori and a posteri-
ori tests, with the direct numerical simulation (DNS) and LES–FSD of statistically planar turbulent premixed flames of lean hydrogen at
1 and 10 atm and a range of turbulence intensities. The a priori test confirms that the proposed model accurately estimates the negative cor-
relation of the displacement speed and the flame curvature. In the a posteriori test, the comparison of DNS and LES–FSD results demon-
strates that the proposed model improves the prediction on the turbulent burning velocity and the turbulent flame thickness, compared with
those from the existing displacement–speed model. The stretch effect in the present model leads to the rise of the turbulent burning velocity
and the suppression of the “bending” phenomenon in LES–FSD, consistent with the DNS result.
Published under license by AIP Publishing. https://doi.org/10.1063/5.0045750

I. INTRODUCTION sd is dominant over curvature and flame stretch effects, so the approxi-
Turbulent premixed flame is one of the fundamental phenomena mation of one-dimensional flamelets is still suitable for modeling
in turbulent combustion. It plays an important role in clean and effi- flames in weak turbulence.5
cient energy utilization, especially for low-emission engine designs.1,2 However, the local displacement speed in the thin reaction zone
The displacement speed sd, i.e., the propagation speed of a flame front regime is strongly influenced by the flame stretch and curvature.5,7
relative to the convective flow of adjacent gas,3,4 is crucial in many Several studies of the direct numerical simulation (DNS) on turbulent
modeling approaches for turbulent premixed flames, e.g., the flame premixed combustion confirmed that the linear correlation between
surface density (FSD) method.5,6 Moreover, sd can be measured in the local displacement speed and the flame stretch is still a good
experiments, so it is widely employed to characterize turbulent pre- approximation.5,8 Additionally, effects of differential diffusion on the
mixed flames, such as counterflow, low-swirl, spherical, and V-shaped local sd can be notable due to the influence of detailed transport.9
flames.1 Abbasi-Atibeh and Bergthorson10,11 found that the local sd increases
In a one-dimensional freely propagating laminar premixed flame, apparently with decreasing of the Lewis number (Le) for Le < 1.
sd is directly related to the laminar flame speed via the density ratio. A number of efforts have been made to quantify the complex
For turbulent premixed flames in the corrugated flamelet regime, it is dependence of sd on the flame stretch and curvature in turbulent pre-
assumed that the laminar flame thickness is smaller than the mixed flames. Based on the theoretical analysis of laminar stretched
Kolmogorov scale. Thus, the entire flame structure is embedded within flames, a linear relation between sd and flame stretch with the
a locally quasi-laminar flow field, and the laminar burning velocity Markstein length can be obtained.12 Considering further experimental
remains well defined. Under such conditions, the propagation effect of and numerical observations on nonlinear dependence of sd on the

Phys. Fluids 33, 045118 (2021); doi: 10.1063/5.0045750 33, 045118-1


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

flame stretch and curvature, various algebraic models have been pro- where dij is the Kronecker delta function and ni is the component of
posed and validated against DNS data.8,13 Additionally, Bray and the unit vector n ¼ rc=jrcj normal to the flame surface. The terms
Cant14 employed a flamelet library to estimate the stretch effects on on the RHS of Eq. (3) represent the subfilter convection, strain-rate
turbulent flame propagation. Then, the tabulation approach consider- effects, flame propagation effects, and flame curvature effects,
ing filter effects was developed for the large-eddy simulation (LES).15 respectively.
On the other hand, most modeling studies of sd are based on a priori Following Ref. 17, these terms are modeled as
analysis and restricted to flames with Le close to unity.

Recently, Lu and Yang16 demonstrated that the local displacement @    @  t @R


 hui iA  u~i R ¼ ; (4)
speed of lean hydrogen turbulent premixed flames with Le ¼ 0:32 is @xi @xi ScR @xi
  pffiffiffi
significantly enhanced due to the flame stretch at high pressures. They   @ui @ u~i /Ck k
characterized this effect by a stretch factor I0 and found I0  3 at dij  ni nj R ¼ ðdij  nij Þ Rþ R; (5)
@xj A @xj D
10 atm, which is much larger than the general choice I0 ¼ 1 for a
majority of fuels and conditions.12 Therefore, the flame stretch effects @   @  
 hsd ni iA R ¼  hsd iA hni iA R ; (6)
should be properly modeled for the accurate estimation of the displace- @xi @xi
ment speed in the modeling of turbulent premixed flames.  
@ni @hni iA aN bs0L R2
In the present study, we propose a model of sd considering the sd R ¼ hsd iA R ; (7)
@xi A @xi 1  ~c
curvature and flame stretch effects. The model performance is assessed
via a priori and a posteriori tests using DNS and LES-FSD calculations where  t is the turbulent viscosity, ScR is the Schmidt number of R, b
of the same statistically planar turbulent flames. The outline of this ¼ 319 and / ¼ 117 are model constants, Ck ¼ 0:75 exp ½1:2=
paper is as follows. The modeling approach is introduced in Sec. II. ðu0D =s0L Þ0:3 ðD=d0L Þ2=3 is an efficiency
ffi 0 function
pffiffiffiffiffiffiffiffiffi
23
with the subgrid
0
Simulation setups and numerical methods are then presented in velocity fluctuation uD ¼ 2k=3, sL and dL are the laminar flame
0

Sec. III. The model assessment is discussed in detail in Sec. IV. Some speed and the thermal thickness of the unstrained laminar flame,
conclusions are drawn in Sec. V. respectively, k is the subgrid turbulent kinetic energy, D is the
filter size, aN ¼ 1  hnk iA hnk iA is an orientation factor,17,24
II. MODELING APPROACH
nij ¼ hni iA hnj iA þ 13 dij ð1  hnk iA hnk iA Þ is a tensor in the modeling
A. Flame surface density method of the strain rate,18 and
The premixed combustion can be generally characterized by the
1 @~c
progress variable c  ðT  Tb Þ=ðTb  Tu Þ,5 where subscripts b and u, hni iA ¼  (8)
respectively, denote the quantities in unburnt reactants and burnt R @xi
products, and T is the temperature. In the context of LES, the FSD is the modeled surface-averaged normal vector.17
method17 has been widely adopted to close the nonlinear source term
in the three-dimensional transport equation for the Favre-averaged B. Modeling of the displacement speed
progress variable as
In Eqs. (2) and (3), the displacement speed sd plays a significant
@ q u~i ~c
q~c @ @ role in the LES-FSD approach. In particular, the propagation and cur-
þ þ  u~i c  u~i ~c Þ ¼ hqsd iA R;
ðq (1)
@t @xi @xi vature terms involving sd are marked in Eq. (3), and they are modeled
where t is the time, xi and ui with i ¼ 1, 2, 3 are the coordinates and in Eqs. (6) and (7), respectively. The definition of sd implies the
velocity components, respectively, q is the density, R  jrcj is the decomposition5
generalized FSD,12 q and ~q denote the regular and Favre filterings of a
n  rðqDn  rT Þ þ x_ T
variable q, respectively, and hqiA  qjrcj=jrcj denotes the average sd ¼  Dj; (9)
over an isosurface of c. qjrTj
The right-hand side (RHS) of Eq. (1) describes that each flame where x_ T is the thermal source term, D is the molecular diffusivity,
surface moves at a local propagation speed sd, where the surface- and j ¼ r  n is the flame curvature. In an unstretched one-
averaged mass flux is typically modeled by17–19 dimensional laminar flame with j ¼ 0, the first term on the RHS of
Eq. (9) equals to qu s0L =q, where qu is the density of unburned mixture,
 hsd iA ;
hqsd iA ¼ q (2) so Eq. (9) becomes
and R can be evaluated by algebraic models18 or by solving its trans- qu s0L
port equation.20,21 Since the hydrodynamic instability at high pres- sd ¼ : (10)
q
sures may not be well characterized by algebraic models,22 we
calculate R by solving the FSD transport equation For turbulent premixed combustion in the regime of corrugated
flamelets, curvature effects are not important compared to the quasi-
  laminar propagation.5,25 Assuming the one-dimensional laminar
@R @ u~i R @      @ui
þ ¼ hui iA  u~i R þ dij  ni nj R flamelet on the turbulent flame front and neglecting the curvature
@t @xi @xi @xj A
term, Eq. (10) is applied to estimate
@   @ni
 hsd ni iA R þ hsd i R; (3)
@xi @xi A ð1Þ qu s0L
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflffl{zfflfflfflfflfflffl} hsd iA ¼ ; (11)
Propagation Curvature 
q

Phys. Fluids 33, 045118 (2021); doi: 10.1063/5.0045750 33, 045118-2


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

where the superscript (1) denotes the first hsd iA model in the present TABLE II. Parameters of unstretched laminar flames in DNS cases.
study. This model has been validated in turbulent premixed flames
with weak stretch and curvature effects26 and widely adopted in FSD p ðatmÞ s0L ðm=sÞ d0L ðlmÞ  ðm2 =sÞ F
Tpeak ðKÞ
modeling17,18 due to its simplicity. On the other hand, the local dis-
1 0.931 365.4 1:94  105 1294.0
placement speed in the thin reaction zone regime is strongly influ-
10 0.348 46.4 1:94  106 1505.4
enced by the flame stretch and curvature. Recent a priori studies13,27
also confirmed that the model in Eq. (11) is insufficient for modeling
premixed combustion in strong turbulence.
From Eq. (9), we propose a new model turbulence (HIT) to assess the hsd iA models in Eqs. (11) and (12)
using a priori and a posteriori tests.
ð2Þ qu s0L Is The present DNS and LES cases have the same parameters for
hsd iA ¼  DhjiA ; (12)

q turbulence and flames. The unburned reactant is the lean hydrogen/air
mixture with the unburned temperature Tu ¼ 300 K and the equiva-
where the superscript (2) denotes the second hsd iA model in this study, lence ratio 0.6. Two pressure conditions 1 atm and 10 atm are consid-
and Is is a factor characterizing the effects of flame straining on the ered. Parameters of the unstretched laminar premixed flames are listed
flame speed. This strain factor is modeled by16 peak
in Table II, including s0L ; d0L , and the temperature TF at the maxi-
sL ðK Þ mum fuel consumption rate, which are obtained by the simulation of
Is ¼ ; (13) one-dimensional freely propagating flames.
s0L For each pressure condition, four cases with different turbulence
where sL ðKÞ is the consumption speed of a stretched laminar flame in intensities are conducted. In these cases, u0 =s0L varies from 5 to 20,
terms of the Karlovitz factor K. This model implicitly assumes that the while lt =d0L is kept as unity. In total, we conducted eight DNS and eight
comsumption speed sc in the subgrid has a delta distribution.12 For LES cases for two pressure conditions. Table III lists the parameters of
given turbulence intensity u0 and integral length scale lt, we estimate the turbulent premixed flames, including u0 =s0L ; lt =d0L , the turbulence
!2 Reynolds number Re0 ¼ ðu0 =s0L Þðlt =d0L Þ, the Damk€ohler number Da

1 3 1
¼ ðlt =u0 Þ=ðd0L =s0L Þ, and the Karlovitz number Ka ¼ ðu0 =s0L Þ2 ðd0L =lt Þ2 .
0:157 p 2 u0
Re2 ;
1
K¼ (14) From the regime diagram of turbulent premixed combustion in 5
Le p0 s0L
Fig. 1, all the present cases are in the thin-reaction-zone regime.
where p is the pressure, p0 ¼ 1 atm is a reference pressure, and
Re ¼ u0 lt = is the turbulent Reynolds number with the kinematic vis- B. Numerical methods
cosity . The model Eq. (14) is extended from that in Bradley28 by In the DNS, the transport equations of mass, momentum, spe-
incorporating the effects of Le and pressure. In the implementation, cies, and temperature are solved in the low-Mach-number limit.29 In
sL ðKÞ is obtained from the calculation of a laminar premixed counter- the LES, the filtered equations of continuity and momentum conserva-
flow flame with a given strain rate at ¼ Ks0L =d0L . tion, together with the Favre-filtered transport equation of ~c in Eq. (1)
ð2Þ
The parameters in the modeling of hsd iA are listed in Table I. and the transport equation of R in Eq. (3), are solved with the dynamic
The large variation of Is with pressure, i.e., Is  1:11 for p ¼ 1 atm and Smagorinsky models for subgrid stresses, turbulent kinetic energy, and
Is  2:1  2:5 for p ¼ 10 atm, implies that Is has an impact on model- scalar fluxes.30
ing of the turbulent burning velocity at high pressures.16 For both DNS and LES, the transport equations are solved on a
staggered grid using the NGA code.31 The momentum equations are
III. SIMULATION OVERVIEW discretized with a second-order, centered, kinetic-energy conservative
A. Case parameters finite difference scheme. The third-order bounded quadratic-upwind
biased interpolative convective (QUICK) scheme32 and the third-
We carry out a series of DNS and LES cases of statistically planar
order weighted essentially nonoscillatory (WENO) scheme33 are
turbulent premixed flames propagating in homogeneous isotropic
employed for treating convection terms in the scalar transport equa-
ð2Þ tions in DNS and LES, respectively. A semi-implicit Crank–Nicolson
TABLE I. Parameters in the modeling of hsd iA .
scheme34 is applied for the time advancement of the transport equa-
tions. The Strang splitting algorithm is employed for transport-
p ðatmÞ u0 =s0L Is sL ðKÞ ðm=sÞ K chemistry coupling in DNS.35,36 The stiff solver DVODE37 is applied
1 5 1.114 1.038 1.351
1 10 1.133 1.056 3.822 TABLE III. Parameters of turbulent premixed flames in DNS and LES cases.
1 15 1.110 1.034 7.021
u0 =s0L lt =d0L Da Ka Re0
1 20 1.077 1.003 10.810
10 5 2.116 0.736 6.228 5 1 0.2 11.18 5
10 10 2.431 0.846 17.616 10 1 0.1 31.63 10
10 15 2.516 0.875 32.364 15 1 0.067 58.09 15
10 20 2.502 0.870 49.827 20 1 0.05 89.44 20

Phys. Fluids 33, 045118 (2021); doi: 10.1063/5.0045750 33, 045118-3


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

domain from 0:05Lx to 0:75Lx to maintain the turbulence intensity.


This linear forcing has been validated in DNS16,41 and LES.42,43
Statistics of each case are calculated over a period of at least 15Te
after reaching the statistically stationary state, where Te ¼ lt =u0
denotes the eddy turnover time. The grid setup and simulation accu-
racy have been extensively validated in the previous studies.16,41
IV. RESULTS AND DISCUSSION
A. A priori test
In the LES-FSD approach, the key models of hsd iA in Eqs. (11)
and (12) are evaluated by a priori tests. For the DNS data, all the quan-
tities are calculated at the flame surface represented by the isothermal
surface of T ¼ Tpeak
F
, which marks the location of the reaction layer of
the flame with the maximum fuel consumption rate.44 This surface
corresponds to isosurfaces of c ¼ 0.648 for p ¼ 1 atm and c ¼ 0.784 for
p ¼ 10 atm. In our implementation, these quantities are first calculated
in the entire computational domain and then interpolated to the flame
surface45 using the marching cubes algorithm for extracting
isosurfaces.
The displacement speed in the DNS with detailed chemistry and
FIG. 1. Parameters (circles) of all cases in the regime diagram of turbulent pre-
mixed combustion. species transport is evaluated as46
X
ns
r  ðkTÞ  cp;m jm  rT þ cp x_ T
to the chemistry substep in DNS with a 9-species, 23-reaction detailed m¼1
mechanism for hydrogen combustion.38 sd ¼ ; (15)
qcp jrTj

C. Setup of computational grids where k is the thermal conductivity, ns is the number of species in the
mixture, cp is the heat capacity of mixture, cp;m is the heat capacity,
In DNS and LES, computational domains are cuboid with inlet
and outlet boundary conditions in the streamwise x-direction. The and jm is the diffusion flux of the m-th species.
In order to examine the dependence of sd on the flame curvature
inlet bulk flow is imposed by the velocity fluctuation from statistically
stationary HIT. Periodic boundary conditions are applied in lateral y- suggested by Eq. (9), we present scatter plots of the normalized
and z-directions. The computational domain size is Lx  Ly  Lz with density-weighted displacement speed qsd =ðqu s0L Þ against the normal-
Ly ¼ Lz ¼ L ¼ 5:3lt ; Lx ¼ nx L, and nx ¼ 12 for DNS and nx ¼ 24 ized curvature jd0L in Fig. 2, at four conditions with two pressures of 1
for LES. The computational domains are discretized on uniform grids atm and 10 atm and two turbulence intensities of u0 =s0L ¼ 5 and 20.
of Nx  Ny  Nz , with Nx ¼ nx N and Ny ¼ Nz ¼ N. The scatter points are color-coded by the joint probability density
The detailed parameters for DNS and LES grids are listed in function (PDF) of qsd and j. For comparison, this figure also plots a
Table IV. For DNS, N is determined by two criteria, kmax g  1:539 linear fit of qsd =ðqu s0L Þ in terms of jd0L from DNS (dashed lines), and
ð1Þ ð2Þ
and at least 24 points in d0L for resolving small-scale turbulence and the model results of hsd iA (dotted lines) in Eq. (11) and hsd iA (solid
flames, respectively. Consequently, N ¼ 128 applies for most cases in lines) in Eq. (12) with D ! 0 for the LES–FSD.
Table III except N ¼ 256 for the case of u0 =s0L ¼ 20 and p ¼ 1 atm. For Although the scattering level of DNS data points grows with the
LES, all the cases are discretized with N ¼ 16. The LES grid resolves at turbulence intensity, the scatterplot indicates a general dependence of
least 90% of the turbulent kinetic energy in the a priori test for non- the displacement speed on the flame curvature, and the negative corre-
reacting HIT with the corresponding DNS. The timesteps are selected lation of sd and j agrees with Eq. (9). Regarding the form of the model
ð1Þ
to ensure the Courant-Friedrichs-Lewy (CFL) numbers less than 0.5 hsd iA in Eq. (11), the modeled displacement speed does not depend
and 0.3 for DNS and LES, respectively. on the flame curvature, resulting in a constant value of q  hsd iA in
ð2Þ
The flow field is initialized by the velocity field obtained from a Fig. 2. On the contrary, the model hsd iA in Eq. (12) implies a linear
separate DNS or LES of nonreacting statistically stationary HIT, and dependence of hsd iA on the curvature. The slope of – D in Eq. (12)
the scalar field is initialized as an unstretched laminar flame. A linear represents the curvature effect and agrees with the linear fit of DNS
forcing method40 is employed for both DNS and LES within the sub- data in Fig. 2.
In addition, the modeled strain factor Is includes the thermal-
TABLE IV. Grid parameters in DNS and LES cases. diffusive effects in the modeling of hsd iA , which is important for turbu-
lent premixed flames with non-unity Le or at high pressures. As shown
N nx CFL in the zoom-in subplot in Fig. 2, the averaged qsd =ðqu s0L Þ at j ¼ 0 is
away from unity for the lean hydrogen flames. For the present model,
DNS 128 or 256 12 0.5 the intercept of lines at j ¼ 0 represents the value of Is. With the aid of
ð2Þ
LES 16 24 0.3 Is modeled by Eq. (13), the estimation of hsd iA (solid line) agrees with
the DNS fit (dashed line) in the subplot.

Phys. Fluids 33, 045118 (2021); doi: 10.1063/5.0045750 33, 045118-4


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 2. Comparisons of the scatter points


with the linear fit (black dashed line)
obtained from the DNS and the values
ð1Þ
calculated by models hsd iA in Eq. (11)
ð2Þ
(blue dotted line) and hsd iA in Eq. (12)
(red solid line) for the LES–FSD.

Based on the above a priori analysis, it is expected that our pro- u0 =s0L ¼ 20, along with the isocontour lines for ~c ¼ 0.1 (purple) and
ð2Þ
posed model hsd iA for the displacement speed can improve the accu- 0.9 (yellow). The turbulent flame thickness, or the thickness of the tur-
racy of LES–FSD on the lean hydrogen turbulent premixed flames. bulent flame brush, is roughly estimated by the averaged distance
between the two iso-lines. For the DNS, the flame surface is strongly
wrinkled, so that the turbulent flame brush spans much wider than d0L .
B. A posteriori test ð2Þ
For the LES–FSD, the turbulent flame thickness with hsd iA is similar
ð1Þ
In the a posteriori test, we compare LES–FSD results with the two to that in the DNS, while the thickness with hsd iA is significantly
displacement–speed models with DNS results to further assess the per- larger than that in the DNS.
formance of model prediction. The instantaneous contours of R from Among the statistics of turbulent premixed combustion, the tur-
DNS and LES are depicted in Fig. 3 for the case with p ¼ 10 atm and bulent burning velocity sT is of vital importance.5 In the DNS, sT with
the consumption-based definition is calculated by the integration of
the fuel reaction rate over the computational domain.16,41 In the
LES–FSD, sT can be evaluated as47
ð
1
sT ¼ hqsd iA RdV; (16)
qu AL X
where AL ¼ Ly  Lz is the planar laminar flame surface and X is the
entire computational domain. Furthermore, the turbulent flame area is
obtained by
ð
AT ¼ RdV: (17)
X
FIG. 3. Instantaneous contours of R and contour lines of ~c ¼ 0:1 (purple) and 0.9 ð1Þ
(yellow) at u0 =s0L ¼ 20 and p ¼ 10 atm in the DNS and the LES–FSD with different For the model hsd iA in Eq. (11), Eqs. (16) and (17) imply the
hsd i models. Damk€ohler's first hypothesis48 sT =s0L ¼ AT =AL . On the other hand,

Phys. Fluids 33, 045118 (2021); doi: 10.1063/5.0045750 33, 045118-5


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

ð1Þ
Lu and Yang16 demonstrated that I0 plays an important role in turbulence intensities in Fig. 4. For the LES–FSD with hsd iA , the
ð1Þ
enhancing sT of the lean hydrogen turbulent premixed flames at high “bending” trends are similar at p ¼ 1 atm and 10 atm, because hsd iA
ð2Þ
pressures. Since the present model hsd iA explicitly incorporates the does not include the dependence of flame characteristics on the pres-
strain factor, it is expected to improve the prediction of sT in the sure. It leads to the notable under-estimation of sT at 10 atm for LES-
ð1Þ ð2Þ
LES–FSD. We remark that the curvature effect is generally negligible FSD with hsd iA . By contrast, the modeled hsd iA includes the
on sT for this configuration, because the curvature distribution is thermal-diffusive effects at different pressures via Is, so the “bending”
nearly symmetric with the zero mean.16,49 at large u0 =s0L is diminished with the growth of Is at high pressures.
ð2Þ
Figure 4 compares sT =s0L and AT =AL obtained from the DNS Thus, the LES–FSD with hsd iA is able to reproduce the suppression
ð1Þ
(dashed lines) and the LES-FSD with models hsd iA (dashed-dotted of the “bending” at elevated pressures.
ð2Þ
lines) and hsd iA (solid lines) at p ¼ 1 and 10 atm and u0 =s0L from 5 to Although the present displacement speed model improves the
20. The (0, 1) point in each plot corresponds to the laminar flame state prediction of the turbulent burning velocity by enhancing Is in the
source term of Eq. (1), the value of sT is still under-estimated in Fig. 4,
with u0 =s0L ¼ 0. For the cases with p ¼ 1 atm; sT =s0L calculated from
ð1Þ ð2Þ which appears to be caused by the under-estimated flame area. We
the DNS and the LES–FSD with hsd iA and hsd iA has a generally good ð2Þ
ð2Þ argue that Is in hsd iA enlarges the source term of Eq. (1), but in the
agreement, whereas AT =AL obtained with hsd iA is slightly smaller than mean time the effective speed ratio u0 =sL ¼ u0 =ðs0L Is Þ is reduced by Is.
ð1Þ
that with hsd iA . These differences are small as Is is close to unity for the As the turbulent flame area increases with u0 =sL ,52 the application of
lean hydrogen mixture at 1 atm. For the cases with p ¼ 10 atm, Is of the Is > 1 leads to the reduction of AT. Specifically, for the case with
lean hydrogen mixture rises, so that the modeling of hsd iA becomes cru- u0 =s0L ¼ 20 at p ¼ 10 atm, we find that u0 =ðs0L Is Þ  10 for the model
ð2Þ
cial. We observe that sT calculated with the model hsd iA is larger than ð2Þ
hsd iA in LES appears to reduce AT almost by half. The reduction of
ð1Þ
that with the model hsd iA and is closer to the DNS result. AT partly cancels out the enhancement of Is in the source term, making
As reported in the previous DNS of lean hydrogen turbulent pre- the prediction of sT still lower than the DNS values, which needs addi-
mixed flames at different pressure conditions,16 the “bending” phe- tional improvement in the future work.
nomenon41,50,51 of sT is suppressed at high pressures. For the present In order to show the consistent model performance and reveal
DNS, sT has a very mild growth with u0 =s0L at strong turbulence for the effects of hsd iA on LES–FSD modeling, we further examine the
p ¼ 1 atm, and sT at p ¼ 10 atm keeps increasing over a range of modeled propagation and curvature terms in Eq. (3) for FSD

FIG. 4. Comparisons of sT =s0L and


AT =AL calculated from the DNS and the
ð1Þ ð2Þ
LES–FSD with models hsd iA and hsd iA
at various turbulence intensities and pres-
sures. Error bars represent the standard
deviation.

Phys. Fluids 33, 045118 (2021); doi: 10.1063/5.0045750 33, 045118-6


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 5. (a) Comparison of hPiyz obtained


in the DNS and the LES–FSD with differ-
ent hsd iA models along the streamwise x
direction at p ¼ 10 atm and u0 =s0L ¼ 10,
as well as (b) the comparison for the cur-
vature contribution to hPiyz .

ð2Þ
transport. The propagation term P  r  ðhsd iA hniA RÞ on the RHS negative in the DNS result. Since hsd iA incorporates the dependence
of Eq. (6) is directly related to the displacement speed. As there is no of hsd iA on the flame curvature, the modeled hCmean iyz qualitatively
other curvature-dependent factor in the model of Eq. (6), the curva- agrees with the DNS result. Moreover, we find that the curvature effect
ture effects on the averaged propagation term hPiyz cancel out each in the modeling of hsd iA has a notable negative contribution to Cmean
other. Here, hqiyz denotes the average of a quantity q over the y–z in Fig. 6(b). The estimation DhjiA Rr  hniA for this part of contri-
ð2Þ
plane with an additional time average of q over a time period of 45Te bution from hsd iA qualitatively agrees with the DNS result.
in the statistically stationary stage. Therefore, Is has a direct influence
on the magnitude of hPiyz . Figure 5(a) plots the profile of hPiyz V. CONCLUSIONS
obtained from the DNS and the LES–FSD with two hsd iA models at We propose a model of the displacement speed for the LES–FSD
p ¼ 10 atm and u0 =s0L ¼ 10. With Is  2:5 larger than unity, the prop- modeling of turbulent premixed flames with non-unity Le at a range
ð2Þ
agation term in LES-FSD is enhanced with the model hsd iA . This also of pressures. Based on the definition of the displacement speed, this
ð2Þ ð1Þ
contributes to a larger sT computed with hsd iA than that with hsd iA model accounts for flame stretch and curvature effects, and the influ-
in Fig. 4. In addition, the two terms on the RHS of Eq. (12) denote the ence of chemistry and transport properties is included via the strain
strain and curvature effects, respectively. In the modeling of P, the cur- factor calculated from a separate laminar flame simulation. The model
vature effect r  ðDhjiA hniA RÞ only shows a moderate oscillation performance is assessed by a priori and a posteriori tests with a series
very close to the flame front in Fig. 5(b), so the strain effect is domi- of DNS and LES-FSD cases of statistically planar turbulent premixed
nant in P at high pressures. flames of lean hydrogen at 1 and 10 atm and various turbulence
Another important term Cmean  hsd iA Rr  hniA in the FSD intensities.
transport equation, the first term on the RHS of Eq. (7), represents the In the a priori test, the DNS result confirms that the displacement
mean-curvature effect and involves the modeled hsd iA . The averaged speed and the flame curvature have a negative statistical correlation in
mean-curvature term hCmean iyz obtained from the DNS and the LES the scatterplot. We demonstrate that this correlation is well approxi-
with different hsd iA models at p ¼ 10 atm and u0 =s0L ¼ 10 is plotted in mated by our proposed model in Eq. (12) via the strain factor and a
ð1Þ
Fig. 6(a). For the model hsd iA ; hsd iA is always positive, so the sign of linear function of the curvature.
the mean-curvature term is the same as r  hniA , leading to positive In the a posteriori test, we compare the important statistics calcu-
and negative hCmean iyz at the leading and trailing edges, respectively. lated from the DNS and the LES-FSD with different displacement-
However, hsd iA and r  hniA have a negative correlation in the a priori speed models of the same flame. The predictions of the turbulent
analysis in Fig. 2. Therefore, the mean-curvature term tends to be flame thickness and the turbulent burning velocity are improved in

FIG. 6. (a) Comparison of hCmean iyz


obtained in the DNS and the LES–FSD
with different hsd iA models along the
streamwise x direction at p ¼ 10 atm and
u0 =s0L ¼ 10, as well as (b) the comparison
for the curvature contribution to hCmean iyz .

Phys. Fluids 33, 045118 (2021); doi: 10.1063/5.0045750 33, 045118-7


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

ð2Þ
the LES–FSD with the proposed model hsd iA , compared with the 10
E. Abbasi-Atibeh and J. M. Bergthorson, “Differential diffusion effects in
ð1Þ
constant model hsd iA . In particular, the persistent growth of the local counter-flow premixed hydrogen-enriched methane and propane flames,”
Proc. Combust. Inst. 37, 2399–2406 (2019).
sd at high pressures is reproduced via the modeled strain factor Is in 11
E. Abbasi-Atibeh and J. M. Bergthorson, “The effects of differential diffusion in
Eq. (13). This leads to the rise of the turbulent burning velocity and counter-flow premixed flames with dilution and hydrogen enrichment,”
the suppression of “bending” in LES–FSD, consistent with the DNS Combust. Flame 209, 337–352 (2019).
12
result. The further comparison for the displacement-speed related T. Poinsot and D. Veynante, Theoretical and Numerical Combustion, 3rd ed.
terms in the FSD transport equation shows that the averaged propaga- (2012).
13
A. Herbert, U. Ahmed, N. Chakraborty, and M. Klein, “Applicability of extrap-
tion and mean-curvatures terms calculated from the LES-FSD with
ð2Þ olation relations for curvature and stretch rate dependences of displacement
hsd iA qualitatively agree with those calculated from the DNS, consis- speed for statistically planar turbulent premixed flames,” Combust. Theory
tent with the improved prediction of the turbulent burning velocity. Modell. 24, 1021–1038 (2020).
From the a priori and a posteriori tests, we demonstrate that the 14
K. N. C. Bray and R. S. Cant, “Some applications of Kolmogorov's turbulence
proposed displacement–speed model is suitable for the LES–FSD of research in the field of combustion,” Proc. R. Soc. London, Ser. A 434, 217–240
(1991).
turbulent premixed combustion at high pressures. The proposed 15
B. Fiorina, R. Vicquelin, P. Auzillon, N. Darabiha, O. Gicquel, and D.
modeling method is expected to be applied to the simulation of practi- Veynante, “A filtered tabulated chemistry model for LES of premixed combus-
cal turbulent premixed flames with various geometries in the future tion,” Combust. Flame 157, 465–475 (2010).
work. Although the effects of the Lewis number are considered via the 16
Z. Lu and Y. Yang, “Modeling pressure effects on the turbulent burning velocity
strain factor in the present model for sd, the differential diffusion and for lean hydrogen/air premixed combustion,” Proc. Combust. Inst. (published
thermal-diffusive effects are not explicitly represented in the transport online, 2021).
17
N. Chakraborty and R. S. Cant, “Direct numerical simulation analysis of the
equation of ~c and R, which requires further interrogation of DNS data- flame surface density transport equation in the context of large eddy simula-
set and investigation of LES-FSD modeling methods. tion,” Proc. Combust. Inst. 32, 1445–1453 (2009).
18
E. R. Hawkes, “Large eddy simulation of premixed turbulent combustion,” Ph.D.
ACKNOWLEDGMENTS thesis (Engineering Department, Cambridge University, Cambridge, UK, 2000).
19
N. Chakraborty and R. S. Cant, “A priori analysis of the curvature and propaga-
We gratefully acknowledge Caltech, the University of tion terms of the flame surface density transport equation for large eddy simu-
Colorado at Boulder and Stanford University for licensing the NGA lation,” Phys. Fluids 19, 105101 (2007).
20
S. B. Pope, “The evolution of surfaces in turbulence,” Int. J. Eng. Sci. 26,
code used in the present work. This work has been supported in 445–469 (1988).
part by the National Natural Science Foundation of China (Grant 21
L. Vervisch, E. Bidaux, K. N. C. Bray, and W. Kollmann, “Surface density func-
Nos. 91841302, 11988102, 11925201, and 91541204) and the Xplore tion in premixed turbulent combustion modeling, similarities between proba-
Prize. bility density function and flame surface approaches,” Phys. Fluids 7,
2496–2503 (1995).
DATA AVAILABILITY 22
M. Klein, H. Nachtigal, M. Hansinger, M. Pfitzner, and N. Chakraborty,
“Flame curvature distribution in high pressure turbulent Bunsen premixed
The data that support the findings of this study are available
flames,” Flow, Turbul. Combust. 101, 1173–1187 (2018).
from the corresponding author upon reasonable request. 23
C. Angelberger, D. Veynante, F. Egolfopoulos, and T. Poinsot, “A flame surface
density model for large eddy simulations of turbulent premixed flames,” in
REFERENCES Proceedings of the Summer Program (Center for Turbulence Research,
Stanford, 1998), pp. 66–82.
1
J. F. Driscoll, “Turbulent premixed combustion: Flamelet structure and its 24
E. R. Hawkes and R. S. Cant, “A flame surface density approach to large-eddy
effect on turbulent burning velocities,” Prog. Energy Combust. Sci. 34, 91–134 simulation of premixed turbulent combustion,” Proc. Combust. Inst. 28, 51–58
(2008). (2000).
2
Z. Chen, S. Ruan, and N. Swaminathan, “Large eddy simulation of flame edge 25
H. Pitsch and L. Duchamp de Lageneste, “Large-eddy simulation of premixed
evolution in a spark-ignited methane-air jet,” Proc. Combust. Inst. 36, turbulent combustion using a level-set approach,” Proc. Combust. Inst. 29,
1645–1652 (2017). 2001–2008 (2002).
3
N. Chakraborty and R. S. Cant, “Unsteady effects of strain rate and curvature 26
M. Boger, D. Veynante, H. Boughanem, and A. Trouve, “Direct numerical sim-
on turbulent premixed flames in an inflow–outflow configuration,” Combust. ulation analysis of flame surface density concept for large eddy simulation of
Flame 137, 129–147 (2004). turbulent premixed combustion,” Proc. Combust. Inst. 27, 917–925 (1998).
4
I. Han and K. Y. Huh, “Roles of displacement speed on evolution of flame sur- 27
V. A. Sabelnikov, A. N. Lipatnikov, N. Chakraborty, S. Nishiki, and T.
face density for different turbulent intensities and Lewis numbers in turbulent Hasegawa, “A balance equation for the mean rate of product creation in pre-
premixed combustion,” Combust. Flame 152, 194–205 (2008). mixed turbulent flames,” Proc. Combust. Inst. 36, 1893–1901 (2017).
5
N. Peters, Turbulent Combustion (Cambridge University Press, 2000). 28
D. Bradley, A. K. C. Lau, and M. Lawes, “Flame stretch rate as a determinant of
6
E. R. Hawkes and R. S. Cant, “Implications of a flame surface density approach turbulent burning velocity,” Philos. Trans. R. Soc., A 338, 359–387 (1992).
to large eddy simulation of premixed turbulent combustion,” Combust. Flame 29
A. J. Aspden, N. Zettervall, and C. Fureby, “An a priori analysis of a DNS data-
126, 1617–1629 (2001). base of turbulent lean premixed methane flames for LES with finite-rate chem-
7
W. Han, H. Wang, G. Kuenne, E. R. Hawkes, J. H. Chen, J. Janicka, and C. istry,” Proc. Combust. Inst. 37, 2601–2609 (2019).
Hasse, “Large eddy simulation/dynamic thickened flame modeling of a high 30
C. D. Pierce and P. Moin, “Progress-variable approach for large-eddy simula-
Karlovitz number turbulent premixed jet flame,” Proc. Combust. Inst. 37, tion of non-premixed turbulent combustion,” J. Fluid Mech. 504, 73–97 (2004).
2555–2563 (2019). 31
O. Desjardins, G. Blanquart, G. Balarac, and H. Pitsch, “High order conserva-
8
J. H. Chen and H. G. Im, “Correlation of flame speed with stretch in turbulent tive finite difference scheme for variable density low Mach number turbulent
premixed methane/air flames,” Proc. Combust. Inst. 27, 819–826 (1998). flows,” J. Comput. Phys. 227, 7125–7159 (2008).
9
C. Dopazo and E. E. O'Brien, “Isochoric turbulent mixing of two rapidly react- 32
M. Herrmann, G. Blanquart, and V. Raman, “Flux corrected finite volume
ing chemical species with chemical heat release,” Phys. Fluids 16, 2075–2081 scheme for preserving scalar boundedness in reacting large-dddy simulations,”
(1973). AIAA J. 44, 2879–2886 (2006).

Phys. Fluids 33, 045118 (2021); doi: 10.1063/5.0045750 33, 045118-8


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

33
X.-D. Liu, S. Osher, and T. Chan, “Weighted essentially non-oscillatory 44
A. J. Aspden, J. B. Bell, M. S. Day, and F. N. Egolfopoulos, “Turbulence-flame
schemes,” J. Comput. Phys. 115, 200–212 (1994). interactions in lean premixed dodecane flames,” Proc. Combust. Inst. 36,
34
C. D. Pierce, “Progress-variable approach for large-eddy simulation of turbulent 2005–2016 (2017).
45
combustion,” Ph.D. thesis (Stanford University, Standford, CA, USA, 2001). M. Day, J. Bell, P.-T. Bremer, V. Pascucci, V. Beckner, and M. Lijewski,
35
Z. Ren and S. B. Pope, “Second-order splitting schemes for a class of reactive “Turbulence effects on cellular burning structures in lean premixed hydrogen
systems,” J. Comput. Phys. 227, 8165–8176 (2008). flames,” Combust. Flame 156, 1035–1045 (2009).
36
Z. Lu, H. Zhou, S. Li, Z. Ren, T. Lu, and C. K. Law, “Analysis of operator split- 46
H. A. Uranakara, S. Chaudhuri, H. L. Dave, P. G. Arias, and H. G. Im, “A flame
ting errors for near-limit flame simulations,” J. Comput. Phys. 335, 578–591 particle tracking analysis of turbulence–chemistry interaction in hydrogen–air
(2017). premixed flames,” Combust. Flame 163, 220–240 (2016).
37
P. N. Brown, G. D. Byrne, and A. C. Hindmarsh, “VODE: A variable-coefficient 47
U. Ahmed, N. Chakraborty, and M. Klein, “Insights into the bending effect in
ODE solver,” SIAM J. Sci. Stat. Comput. 10(198), 1038–1051 (1989). premixed turbulent combustion using the flame surface density transport,”
38
M. P. Burke, M. Chaos, Y. Ju, F. L. Dryer, and S. J. Klippenstein, Combust. Sci. Technol. 191, 898–920 (2019).
“Comprehensive H2/O2 kinetic model for high-pressure combustion,” Int. J. 48
G. Damk€ ohler, “Der einfluss der turbulenz auf die flammengeschwindigkeit in
Chem. Kinet. 44, 444–474 (2012). gasgemischen,” Z. Elektrochem. Angew. Phys. Chem. 46, 601–626 (1940).
39
S. B. Pope, Turbulent Flows (Cambridge University Press, 2000). 49
B. Savard, S. Lapointe, and A. Teodorczyk, “Numerical investigation of the
40
P. L. Carroll and G. Blanquart, “A proposed modification to Lundgren's physi- effect of pressure on heat release rate in iso-octane premixed turbulent flames
cal space velocity forcing method for isotropic turbulence,” Phys. Fluids 25, under conditions relevant to SI engines,” Proc. Combust. Inst. 36, 3543–3549
105114 (2013). (2017).
41
J. You and Y. Yang, “Modelling of the turbulent burning velocity based on 50
J. M. Duclos, D. Veynante, and T. Poinsot, “A comparison of flamelet models
Lagrangian statistics of propagating surfaces,” J. Fluid Mech. 887, A11 (2020). for premixed turbulent combustion,” Combust. Flame 95, 101–117 (1993).
42
B. de Laage de Meux, B. Audebert, R. Manceau, and R. Perrin, “Anisotropic 51
A. Lipatnikov and J. Chomiak, “Turbulent flame speed and thickness:
linear forcing for synthetic turbulence generation in large eddy simulation and Phenomenology, evaluation, and application in multi-dimensional simula-
hybrid RANS/LES modeling,” Phys. Fluids 27, 035115 (2015). tions,” Prog. Energy Combust. Sci. 28, 1–74 (2002).
43
C. Wang and M. Ge, “Applying resolved-scale linearly forced isotropic turbu- 52
T. Zheng, J. You, and Y. Yang, “Principal curvatures and area ratio of propa-
lence in rational subgrid-scale modeling,” Acta Mech. Sin. 35, 486–494 (2019). gating surfaces in isotropic turbulence,” Phys. Rev. Fluids 2, 103201 (2017).

Phys. Fluids 33, 045118 (2021); doi: 10.1063/5.0045750 33, 045118-9


Published under license by AIP Publishing

You might also like