A direct numerical simulation (DNS) database of freely propagating statistically planar turbulent premixed flames with a range of different values of Karlovitz number Ka, turbulent Reynolds number Ret, heat release parameter τ, and global Lewis number Le has been used to assess the models of the tangential strain rate term in the generalised flame surface density (FSD) transport equation in the context of Reynolds averaged Navier Stokes (RANS) simulations. The tangential strain rate term has been split into contributions arising due to dilatation rate TD and flame normal strain rate (-TN). Subsequently, TD and (-TN) were split into their resolved (i.e., TD1 and (-TN1)) and unresolved (TD2 and (-TN2)) components. Detailed physical explanations have been provided for the observed behaviours of the components of the tangential strain rate term. This analysis gave way to the modelling of the unresolved dilatation rate and flame normal strain rate contributions. Models have been identified for TD2 and (-TN2) for RANS simulations, which are shown to perform satisfactorily in all cases considered, accounting for the variations in Ka, Ret, τ and Le. The performance of the newly proposed models for the FSD strain rate term have been found to be either comparable to or better than the existing models.
1. Introduction
Flame surface density (FSD) based reaction rate closure is one of the most well-established methods of turbulent premixed combustion modelling in the context of Reynolds averaged Navier Stokes (RANS) simulations [1–12]. In the FSD based formulation, the closure of reaction rate translates to the modelling of flame surface area to volume ratio [13]. The FSD, in RANS simulations, is either evaluated using an algebraic expression [1] or a modelled transport equation for FSD is solved alongside other modelled Favre averaged conservation equations [3–12]. However, algebraic closures may not be suitable when flame surface area generation and destruction are not in equilibrium and recently Richard et al. [14] demonstrated that algebraic closures may not be adequate for simulating in-cylinder processes in spark ignition piston engines. Therefore, it may be necessary to consider high-fidelity transport equation based closures of FSD for turbulent premixed flames. The flame surface area generation/destruction due to fluid straining plays a key role in the FSD transport and thus the statistical analysis and detailed assessments of the modelling of the tangential strain rate term in the FSD transport equation in the context of RANS simulations have been considered in the current work because the strain rate term remains a leading order contributor to the FSD transport [3–12].
The generalised FSD Σgen is defined as [15, 16]
(1)Σgen=|∇c|¯,
where c is the reaction progress variable and the quantities Q- and Q~=ρQ¯/ρ- denote the Reynolds averaged and Favre averaged values of a general quantity Q. The generalised FSD Σgen is often used in order to close the mean reaction rate ω˙¯ of reaction progress variable in the following manner [15]:
(2)ω˙¯+∇·(ρD∇c)¯=(ρSd)¯sΣgen,
where ρ is the gas density, D is the reaction progress variable diffusivity, Sd=(Dc/Dt)/|∇c| is the displacement speed, and (Q)¯s=Q|∇c|¯/|∇c|¯ indicates a surface averaging operation of a general quantity Q [5, 15]. In the context of RANS simulations ω˙¯≫∇·(ρD∇c)¯ and therefore it is evident from (2) that an accurate prediction of Σgen is required for the closure of the mean reaction rate ω˙¯ provided the statistical behaviour of the surface averaged density-weighted displacement speed (i.e., (ρSd)¯s) is satisfactorily accounted for. The exact transport equation for the generalised FSD (i.e., Σgen=|∇c|¯) takes the following form [9, 11, 12, 17, 18]:
(3)∂Σgen∂t+∂(u~jΣgen)∂xj=-∂∂xi[(ui)¯s-u~i]Σgen+((δij-NiNj)∂ui∂xj)¯sΣgen-∂∂xi[(SdNi)¯sΣgen]+(Sd∂Ni∂xi)¯sΣgen,
where u~i=ρui¯/ρ¯ and ui′′=ui-u~i are the Favre mean and fluctuating velocity components in the ith direction and N→=-∇c/|∇c| is the local flame normal vector (which points towards the reactants). The reaction progress variable c can be defined in terms of a suitable reactant mass fraction YR in such a manner that it increases monotonically from zero in unburned gases to unity in fully burned products. In (3), the first term in the left hand side is the transient term, whereas the second term represents the effects of mean advection. All the terms on the right hand side of (3) are unclosed and, hence, require modelling. The first term on the right hand side of (3) represents the effects of turbulent transport, the second term accounts for the flame area generation due to tangential strain rate (i.e., aT=(δij-NiNj)∂ui/∂xj), and the third term originates due to flame normal propagation and thus is commonly referred to as the propagation term. The final term on the right hand side of (3) (i.e., (Sd∇·N→)¯sΣgen) arises due to curvature κm=∇·N→/2 (according to this convention a flame element, which is convex towards the reactants, has a positive curvature and vice versa) and thus is commonly referred to as the FSD curvature term. In the current study, only the a priori DNS assessment of the models of the tangential strain rate term of the generalised FSD transport equation will be considered.
The tangential strain rate term of the generalised FSD transport equation is often split in the following manner:
(4)(aT)¯sΣgen=((δij-NiNj)∂ui∂xj)¯sΣgen=[δij-(NiNj)¯s]∂u~i∂xjΣgen︸SR+([δij-NiNj]∂ui′′∂xj)¯sΣgen︸SUR.
Cant et al. [2] and Candel et al. [3] proposed models for both the resolved and the unresolved parts of the tangential strain rate contribution (i.e., SR and SUR, resp.), which were subsequently assessed by Duclos et al. [4] and Prasad and Gore [8] based on RANS simulations. Veynante et al. [6] and Veynante et al. [7] assessed the performance of the models of SR and SUR based on experimental data. The aforementioned studies [2–4, 6–8] were carried out in the corrugated flamelets regime, where the flame thickness remains smaller than the Kolmogorov length scale. Recently, Katragadda et al. [10] assessed the performances of the existing SR and SUR models in both the corrugated flamelets and thin reaction zones regimes of premixed combustion based on direct numerical simulation (DNS) data, while accounting for the effects of Lewis number and heat release parameter τ. It was shown by Katragadda et al. [10] that both the dilatation rate ∂ui/∂xi and the relative alignment of ∇c with the fluid-dynamic strain rate eij=0.5(∂ui/∂xj+∂uj/∂xi) significantly affect the behaviour of the strain rate term (aT)¯sΣgen. Global Lewis number Le is known to have significant influences on both the dilatation rate and scalar gradient alignment statistics in turbulent premixed flames [19]. Additionally, it was shown in recent studies [19–21] that the Damköhler number Da significantly affects the relative strength of the dilatation rate in comparison to the turbulent straining and alignment characteristics of ∇c with the local principal strain rates. As the turbulent Reynolds number scales as Ret~Ka2Da2 [22] it is expected that the modelling of the strain rate term will also show some Ret dependence.
The effects of Lewis number Le, turbulent Reynolds number Ret, Damköhler number Da, Karlovitz number Ka, and heat release parameter τ on the statistical behaviour and RANS modelling of the tangential strain rate term in the FSD transport equation are yet to be addressed in detail and the present paper aims to address this gap in the existing literature. In order to investigate the effects of Le, Ret, Da, Ka, and τ on the statistical behaviour and RANS modelling of the FSD tangential strain term, a database of three-dimensional compressible direct numerical simulations (DNS) of turbulent premixed flames has been considered. In this respect, the main objectives of this study are as follows:
to demonstrate and explain the statistical behaviours of the tangential strain rate term of the FSD transport equation and its components for a range of different values of Lewis numbers, turbulent Reynolds numbers, Karlovitz and Damköhler numbers, and heat release parameters spanning both the corrugated flamelets and thin reaction zones regimes of turbulent premixed combustion;
to identify models which satisfactorily capture the statistical behaviour of the tangential strain rate term in the FSD transport equation and its components for different values of Le, τ, and Ret and different combustion regimes in the context of RANS simulations.
The remainder of the paper will take the following form. The next section will discuss the relevant mathematical background of the current study, which will then be followed by a brief discussion on the numerical implementation. Following this, the results will be presented and subsequently discussed. Finally, the main findings will be summarised and conclusions will be drawn.
2. Mathematical Background
Direct numerical simulations (DNS) of turbulent combustion should, ideally, account for both the inherent three-dimensionality of turbulence and the detailed chemical structure of the flame. However, until recently combustion DNS studies were limited to be carried out either in two dimensions with detailed chemistry or in three dimensions with simple chemistry. Three-dimensional DNS of turbulent combustion is now possible but such simulations remain extremely computationally expensive [23] and thus are often not suitable for a detailed parametric analysis as carried out in this work. As such, a three-dimensional DNS database with simple chemistry has been considered here for the sake of computational economy. In the context of simple chemistry, the species field can be represented, uniquely, by a reaction progress variable c which can be defined in terms of a suitable mass fraction (i.e., YR) in the following manner:
(5)c=(YR0-YR)(YR0-YR∝),
where subscripts 0 and ∝ were used to denote the values in the unburned and burned gases, respectively. In globally adiabatic, low Mach number unity Lewis number flames the reaction progress variable c is the same as the nondimensional temperature:
(6)T=(T^-T0)(Tad-T0),
where T^ is the instantaneous dimensional temperature, T0 is the unburned gas temperature, and Tad is the adiabatic flame temperature. However, for nonunity Lewis number flames the nondimensional temperature T may locally assume superadiabatic values (i.e., T>1) [24, 25] even under globally adiabatic conditions, whereas c is always bound within 0 and unity (i.e., 0≤c≤1).
As discussed in Section 1, the strain rate term (aT)¯sΣgen is often modelled by splitting it into the resolved SR and the unresolved SUR components. For the purpose of evaluating SR, the quantity (NiNj)¯s requires modelling. Cant et al. [2] modelled SR as
(7)SR=(1.0-nij)∂u~i∂xjΣgen,
where nij is the model expression for (NiNj)¯s, which takes the following form according to Cant et al. [2]:
(8)nij=(NiNj)¯s=(Ni)¯s(Nj)¯s+δij3[1-(Nk)¯s(Nk)¯s].
Veynante et al. [7] modelled (NiNj)¯s as
(9)(NiNi)¯s=∑k≠i(uk′′uk′′)~4k~,(NiNj≠i)¯s=ui′′uj′′~2k~,
where k~=ui′′ui′′~/2 is the turbulent kinetic energy. Cant et al. [2] modelled for the unresolved part of the strain rate term SUR in the following manner:
(10)SUR=0.28ρ0ε~μ0Σgen,
where ρ0 and μ0 are the unburned gas density and viscosity, respectively, and ε~ is the dissipation rate of k~. In the context of coherent-flamelet modelling (CFM) [4] SUR is modelled as
(11)SUR=α0Γkε~k~Σgen,
where α0 is a model constant of the order of unity (i.e., α0=2.0) and Γk is the efficiency function proposed by Meneveau and Poinsot [26] which is a function of lt·SL/αT0 and (2k~/3)/SL with αT0, SL, and lt being the thermal diffusivity in the unburned gas, unstrained laminar burning velocity, and local integral length scale, respectively.
The strain rate term can alternatively be decomposed in the following manner [10]:
(12)(aT)¯sΣgen=SR+SUR=(∂uj∂xj)¯sΣgen︸TD-(NiNj∂ui∂xj)¯sΣgen︸TN.
The terms TD and (-TN) represent contributions of dilatation rate and flame normal strain rate on the Σgen transport. The dilatation rate term TD can be split into the resolved and unresolved components in the following manner:
(13)TD=[∂u~i∂xi|∇c-|]︸TD1+[∂ui∂xi|∇c|]¯-[∂u~i∂xi|∇c-|]︸TD2.
From the expression above, it is evident that the resolved dilatation rate term TD1 can be closed if a suitable relationship between c- and c~ can be found. The contribution of the normal strain rate term (-TN) can be split as
(14)-TN=-(NiNj)¯s∂u~i∂xjΣgen︸TN1-(NiNj∂ui′′∂xj)¯sΣgen︸TN2.
The quantity (-TN1) can be closed using the models for (NiNj)¯s. This can be carried out as shown in (8) and (9).
It is worth noting that the surface averaged quantities involving velocity components in (4) are modelled in terms of the gradients of unconditional Favre-averaged velocity components ∂u~i/∂xj (see (7), (9), (10), and (11)). In actual RANS simulations the unconditional Favre-averaged velocities are readily obtained so it makes sense to model the surface-averaged terms involving velocity components using the Favre-averaged velocity components [2–8]. The modelling of (-TN2) and TD2 (see (12) and (14)) will be discussed based on a priori DNS analysis using the current database in Section 4, where the surface averaged quantities involving velocity components are also modelled in terms of the gradients of unconditional Favre-averaged velocity components ∂u~i/∂xj (see (15) and (17) later in this paper). The present formulation is based on the generalised FSD Σgen, which is defined as [15, 16]: Σgen=|∇c|¯. This indicates that the generalised FSD Σgen is not dependent upon choice of c isosurface. Thus the difference between conditional and unconditional velocities does not play any role in the current analysis similar to several previous studies [2–12].
3. Numerical Implementation
For the present study, the simulation parameters for the DNS database considered are shown in Table 1, namely, the initial values of root-mean-square turbulent velocity fluctuation normalised by the unstrained laminar burning velocity u′/SL, integral length scale normalised by the thermal flame thickness l/δth, turbulent Reynolds number Ret=ρ0u′l/μ0, heat release parameter τ=(Tad-T0)/T0, Lewis number Le, Damköhler number Da=lSL/u′δth, and Karlovitz number Ka=(u′/SL)3/2(l/δth)-1/2. Standard values are chosen for Zel’dovich number β=Tac(Tad-T0)/Tad2, Prandtl number Pr, and the ratio of specific heats (i.e., β=6.0, Pr=0.7, and γ=CP/CV=1.4), where Tac is the activation temperature. It should be noted that the simulation parameters have been chosen in such a manner that combustion takes place within the corrugated flamelets regime (thin reaction zones regime) [22] for case A (cases B–L). The turbulent flow conditions and heat release parameter are kept unaltered but the global Lewis number Le was modified from 0.34 to 1.2 in cases C–G. The cases B and F are identical in terms of turbulent flow conditions but only τ values are different between these cases. As the turbulent Reynolds number Ret scales as Ret~Da2Ka2 [22], the variation of Ret in cases H–L is brought about by modifying Da and Ka independently of each other. In cases H, J, and L (I, J, and K) Da (Ka) is kept unaltered and Ka (Da) is altered to bring about the variation of Ret. The range of Ret considered here is comparable to that of previous laboratory-scale experiments (e.g., Kobayashi et al. [27]).
Initial values of the simulation parameters corresponding to the DNS database considered here.
Case
u′/SL
1/δth
Ret
τ
Le
Da
Ka
A
1.41
9.64
56.70
2.3
1.0
6.84
0.54
B
7.50
2.45
47.00
3.0
1.0
0.33
13.17
C
7.50
2.45
47.00
4.5
0.34
0.33
13.17
D
7.50
2.45
47.00
4.5
0.6
0.33
13.17
E
7.50
2.45
47.00
4.5
0.8
0.33
13.17
F
7.50
2.45
47.00
4.5
1.0
0.33
13.17
G
7.50
2.45
47.00
4.5
1.2
0.33
13.17
H
5.00
1.67
22.00
4.5
1.0
0.33
8.67
I
6.25
1.44
23.50
4.5
1.0
0.23
13.00
J
7.50
2.50
47.85
4.5
1.0
0.33
13.00
K
9.00
4.31
100.00
4.5
1.0
0.48
13.00
L
11.25
3.75
110.00
4.5
1.0
0.33
19.50
A domain of 105αT0/SL×131.95αT0/SL×131.95αT0/SL is taken for case A, which is discretised by a Cartesian grid of 261×128×128 with uniform grid spacing in each direction [28]. In case A, inlet and outlet boundaries are specified in the mean direction of flame propagation (which is aligned with the negative x1-direction), whereas transverse boundaries are taken to be periodic. In cases B–G, a domain of size 42.18αT0/SL×42.18αT0/SL×42.18αT0/SL is discretised using a uniform grid of 230×230×230. In cases H–L, a domain of size 63.37αT0/SL×42.18αT0/SL×42.18αT0/SL is discretised using a uniform grid of 345×230×230. The domain boundaries in the direction of mean flame propagation in cases B–L are taken to be partially nonreflecting and the transverse boundaries are assumed to be periodic. In case A, a 6th order central-difference scheme has been used for spatial discretisation for the internal grid points in the direction of mean flame propagation, which gradually reduces to a one-sided 4th order scheme near nonperiodic boundaries, whereas a spectral method is used for spatial discretisation in the directions normal to the mean direction of flame propagation [28]. In cases B–L, a 10th central difference scheme is used for the internal grid points and the order of differentiation gradually reduces to a 2nd order one-sided scheme near nonperiodic boundaries. The time advancement for all viscous and diffusive terms in case A is carried out using an implicit solver, whereas the convection terms in case A and all the terms in cases B–L are time advanced with the help of a third order low storage Runge-Kutta method [29]. For all cases, the flame is initialised by a steady unstrained planar laminar flame solution and the turbulent fluctuating velocity field is initialised based on an incompressible homogeneous isotropic velocity distribution, which is generated using a standard pseudospectral method [30]. The grid spacing is determined by the resolution of the flame structure, and about 10 grid points are kept within δth for all cases considered here. In all cases flame-turbulence interaction takes place under decaying turbulence and simulations need to be carried out for tsim≥Max(tf,tc), where tf=l/u′ is the initial eddy turn-over time and tc=δth/SL is the chemical time scale. The simulation in case A was run for about 4tf, whereas in cases B–G it was run for a time equivalent to 3.34tf. The simulation time remains either greater than (case A) or equal to (cases B–L) one chemical time scale. The simulation time remains either comparable to or greater than several previous studies [9, 31–35]. The turbulent kinetic energy and its dissipation rate in the unburned gas ahead of the flame were not varying significantly with time when statistics were extracted for all cases. Interested readers are referred to Chakraborty and Cant [11, 12] and Chakraborty et al. [36, 37] for further information on this database and the conditions under which statistics were extracted.
The relevant quantities in all cases considered here have been Reynolds/Favre-averaged by ensemble averaging the relevant quantities in transverse directions (x2-x3 planes). The statistical convergence of the Reynolds/Favre-averaged quantities is examined by comparing the corresponding values obtained using half of the sample size from the distinct half of the domain in the transverse direction with those obtained based on full sample size. Both the qualitative and quantitative agreements between these sets of values are found to be satisfactory and more information can be obtained from [11]. In the next section, only the results obtained based on full sample size will be presented. For statistically planar flames, c~ remains a unique function of the spatial coordinate in the direction of mean flame propagation (i.e., x1-direction); thus all the statistics in Section 4 will be presented as a function of c~.
4. Results and Discussion
The contours of the reaction progress variable c in the central x1-x2 plane when the statistics are extracted are shown in Figure 1 for cases A–L. In case A the contours of c remain parallel to each other as a consequence of the corrugated flamelets regime combustion. In the corrugated flamelets regime the energetic turbulent eddies are unable to penetrate into the flame and wrinkles develop due to large-scale turbulent motion. However, in cases B–L the contours of c are found to be distorted towards the unburned gas side (i.e., c<0.5) representing the preheat zone. This distortion of the preheat zone in cases B–L is characteristic of the thin reaction zones regime combustion. Additionally the contours of c representing the reaction zone (i.e., 0.7≤c≤0.9) remain parallel to each other for all cases. In cases B–L, where combustion takes place in the thin reaction zone regime, the energetic turbulent eddies penetrate into the flame, as the flame thickness remains greater than the Kolmogorov length scale though the reaction zone retains its quasi-laminar structure due to η>δr where η and δr are the Kolmogorov length scale and the reaction zone thickness, respectively.
Contours of c in x1-x2 midplane for cases ((a)–(l)) A–L when statistics were extracted.
The effects of Lewis number on the flame structure can be seen in Figures 1(c)–1(g), where the extent of the flame wrinkling increases as Le decreases. The reactants diffuse more rapidly into the reaction zone than the rate at which heat is conducted out for the flames with Le<1, which leads to simultaneous presence of high concentration of reactants and high temperature. As a result, the cases with Le<1 exhibit greater burning rates than the unity Lewis number flames with the same turbulent flow conditions in the unburned gas. Just the opposite mechanism in terms of reactant and thermal diffusion takes place for Le>1 flames, which reduces the flame area generation and burning rate in these flames in comparison to the corresponding unity Lewis number flames with the same unburned gas turbulence. This gives rise to increased burning rates and flame area generation with decreasing Le, which can be substantiated from Table 2 where the normalised turbulent flame speed ST/SL and normalised flame surface area AT/AL at the time when the statistics were extracted are presented. The values of ST/SL have been evaluated by volume integrating the reaction rate w˙ using the expression ST=(1/ρ0AP)∫Vw˙dV where AP is the projected area of the flame in the direction of mean flame propagation, while the values of AT/AL have been evaluated by volume integrating |∇c| (i.e., ∫V|∇c|dV) under both turbulent and laminar conditions. The preheat zone (c<0.5) becomes increasingly distorted due the increased scale separation between δth and η as a result of increasing Karlovitz number Ka~Ret1/2/Da~δth2/η2 when Ret increases for a given value of Da. Thus, the preheat zone is more susceptible to be penetrated by energetic turbulent eddies with increasing scale separation between δth and η. It is clear from Figures 1(h)–1(l) that the flame wrinkling increases with increasing u′/SL~Ret1/4Ka1/2~Ret1/2/Da1/2, which gives rise to high values of normalised turbulent flame speed ST/SL and normalised flame surface area AT/AL for high values of u′/SL.
The values of normalised turbulent flame speed ST/SL and normalised flame surface area AT/AL when statistics were extracted.
Case
ST/SL
AT/AL
A
1.80
1.81
B
1.98
1.94
C
13.70
3.93
D
4.58
2.66
E
2.53
2.11
F
1.83
1.84
G
1.50
1.76
H
1.14
1.15
I
1.32
1.33
J
1.87
1.87
K
3.62
3.63
L
3.68
3.70
The variations of (aT)¯sΣgen, SR, SUR, TD, and (-TN) with c~ for all cases are shown in Figures 2(a)–2(l), which show that (aT)¯sΣgen, SR, SUR, and TD assume positive values throughout the flame brush and the contribution of SUR supersedes that of SR in all cases, in accordance with earlier findings [2–4, 6–8, 18, 38, 39]. As ∇·u→ remains principally positive within the flame brush and Σgen is a positive semidefinite quantity (i.e., Σgen≥0), the dilatation rate term TD remains positive throughout the flame brush. It can be seen from Table 2 that the effects of heat release strengthen with decreasing Le so the magnitude of ∇·u→ increases significantly with decreasing Le, which along with the increasing trend of |∇c| with decreasing Le [11] leads to high magnitudes of TD for flames with small values of Le (see Figures 2(c)–2(g)).
Variations of the values of (aT)¯sΣgen (solid line), SR (dash line), SUR (star line), TD (box line), and (-TN) (triangle line) with c~ across the flame brush for cases ((a)–(l)) A–L, where all the terms are normalised with respect to SL/δth2.
It can be seen from Figure 2 that the normal strain rate term (-TN) exhibits marked differences from one case to another. The term (-TN) assumes negative value throughout the flame brush for cases A, C, D, E, H, I, J, and K, whereas this term assumes a positive (negative) value towards the unburned (burned) gas side of the flame brush in cases B, F, G, and L. The behavior of TN=(NiNj∂ui/∂xj|∇c|)¯=(eααcos2α+eββcos2β+eγγcos2γ)|∇c|¯ is determined by the alignment of ∇c with the local principal strain rates where eαα, eββ, and eγγ are the most extensive, intermediate, and compressive principal strain rates, respectively, and α, β, and γ are the angles between |∇c| and eαα, eββ, and eγγ, respectively. For passive scalars, the scalar gradient is known to align with the most compressive principal strain rate eγγ [28, 40, 41] but recent studies [19–21, 42] have shown that ∇c may align with the most extensive principal strain rate eαα, where the strain rate induced by flame normal acceleration overcomes the effects of turbulent straining.
The strain rate induced by flame normal acceleration can be scaled as achem~τf(Ka)SL/δth for Le=1.0 flames, whereas the turbulent straining can be scaled as aturb~u′/l, which gives rise to achem/aturb~f(Ka)τlSL/u′δth~τ·f(Ka)Da [20] with f(Ka) being a function which accounts for the weakening of the effects of flame normal acceleration with increasing Ka as the broken reaction zones regime combustion is approached. Alternatively, the turbulent straining aturb can be scaled as aturb~u′/λ following Tennekes and Lumley [43] where λ is the Taylor microscale, which leads to achem/aturb~τ·f(Ka)Da/Ret1/2. The scaling achem/aturb~f(Ka)τlSL/u′δth~τ·f(Ka)Da (achem/aturb~τ·f(Ka)Da/Ret1/2) indicates that ∇c may align with eγγ to yield a positive (negative) contribution of TN (-TN) for large Da flames. By contrast, in cases B, F, G, and L, Da remains small (i.e., Da<1), so that the effects of aturb overcome the effects of achem, giving rise to the predominant alignment of ∇c with eγγ for the major portion of the flame brush, except in the heat releasing zone where the effects of achem overcome the effects of aturb. This is reflected in the predominantly negative (positive) value of TN (-TN) towards the unburned gas side of the flame brush, and positive (negative) values of TN (-TN) are obtained towards the burned gas side of the flame brush in cases B, F, G, and L. Chakraborty et al. [19] demonstrated that achem strengthens with decreasing Le and this effect is particularly strong for the Le≪1 flames due to high values of chemical heat release. As a result, achem/aturb can be scaled as achem/aturb~τ·f(Le,Ka)·Da (achem/aturb~f(Le,Ka)Da/Ret1/2) for nonunity Lewis number flames, where the function f increases with decreasing Le [19]. This suggests that achem may become sufficiently strong to overcome the effects of aturb for small values of Le, even for Da<1 flames. Strong achem in the Le=0.34, 0.6, and 0.8 flames overcomes the effects of aturb and induces a preferential alignment of ∇c with eαα. This gives rise to positive (negative) values of TN (-TN) for the major portion of the flame brush. As the effects of achem weaken with an increase in Le, the extent of the negative contribution of (-TN) progressively decreases with increasing Le, which can be substantiated from Figures 2(c)–2(g). Figures 2(a)–2(l) show that the magnitude of (-TN) remains smaller than the magnitude of TD for all cases, which yields a positive contribution of (aT)¯sΣgen throughout the flame brush even when the contribution of (-TN) remains negative. In all cases, aturb remains of the same order of magnitude as that of achem (i.e., achem/aturb~f(Le,Ka)·τ·Da~O(1) or achem/aturb~f(Le,Ka)·Da/Ret1/2~O(1)) and thus the effects of aturb are partially nullified by the effects induced by achem, which give rise to a smaller magnitude of the normal strain rate contribution (-TN) than the magnitude of the dilatation rate term TD.
It is worth noting that the variations of some of the strain rate terms with c~ in Figure 2 and subsequent figures show wavy behaviour although the variation of c- with c~ for all cases has been found to be reasonably smooth. However, the variation of c- with c~ exhibits weak undulations for some cases (e.g., cases A, C, K, and L), which originate due to large extent of wrinkling in these flames. These undulations in c- variations with c~ give rise to wavy behaviour of |∇c-| (=|∂c-/∂x1| for statistically planar flames) when a 10th order central difference scheme is used for spatial differentiation. This also contributes to the wavy behaviour of the variations of the strain rate terms with c~. Similar wavy behaviour has been seen in several previous DNS studies [9, 10, 20, 21, 28, 42]. This is not due to a lack of statistical convergence but is a consequence of the high levels of wrinkling in these flames and high order of finite difference scheme used for postprocessing purpose. Moreover, this waviness is prevalent only for case A, which was not simulated by the present authors but taken from a widely used well-respected DNS database (see [28]). This waviness is also observed in several previous analyses [20, 21, 28, 42, 44], where case A was used.
Katragadda et al. [10] modelled the tangential strain rate term based on the separate modelling of the strain rate components TD and (-TN). The term TD1=∂u~i/∂xi|∇c-| can be closed if a suitable relationship can be established between c~ and c-. According to Bray-Moss-Libby (BML) analysis [45], this can be achieved by c-=(1+τ)c~/(1+τc~)+O(γ) for unity Lewis number globally adiabatic flames, where O(γ) accounts for the contribution of the reacting mixture. The contribution of O(γ) remains negligible for high Da flames but it might be nonnegligible for low Da combustion. Katragadda et al. [10] proposed an alternative empirical expression c-=(1+τg1.5Le-0.26)c~/(1+τg1.5Le-0.26c~) where g=c′′2~/c~(1-c~) is the segregation factor and it was shown in [10–12] that (1+τg1.5Le-0.26)c~/(1+τg1.5Le-0.26c~) adequately captures c- for low Da flames considered in this analysis and thus predictions of c-=(1+τg1.5Le-0.26)c~/(1+τg1.5Le-0.26c~) are not shown here for the sake of conciseness. Moreover, (1+τg1.5Le-0.26)c~/(1+τg1.5Le-0.26c~) approaches the BML expression c-=(1+τ)c~/(1+τc~) for unity Lewis number flames for Da≫1 combustion under which g=c′′2~/c~(1-c~) approaches unity. Thus c-=(1+τg1.5Le-0.26)c~/(1+τg1.5Le-0.26c~) enables one to close TD1=(∂u~i/∂xi)|∇c-|. The unresolved part of the dilatation rate term TD2 is taken to model as [10]
(15)TD2=2BT2(1+KaL)nτSL(Σgen-|∇c-|)δth(1-c~)m1Lem2,
where the model parameters BT2=1.8, n=0.35, m1=1.8Le-1.5, and m2=1.845. In (15) dilatation rate is scaled as ∇·u→~τ·SL/δthLem2 as the strength of the dilatation rate increases with decreasing Le, and (1-c~)m1 is used to adequately capture the variation of TD2 with c~ obtained from DNS where the exponent m1 is likely to be a function of Le, as the variation of TD2 with c~ is skewed with a peak towards the unburned gas side for Le=0.34 and 0.6 flames, whereas the peak value of TD2 is attained close to the middle of the flame brush (i.e., c~=0.4) for the flames with Le close to unity. According to (15), the contribution of TD2 vanishes when the flame is fully resolved (i.e., Σgen=|∇c-|) and the net contribution of (TD1+TD2) becomes identical to (∂u~i/∂xi)|∇c|. The contribution of TD2 is similar to the dilatation rate contribution TDN=2(∇·u→)ρD∇c′′·∇c′′¯ in the scalar dissipation rate ε~c=ρD∇c′′·∇c′′¯/ρ¯ transport equation [42]. The term TDN=2(∇·u→)ρD∇c′′·∇c′′¯ is modeled as [42]
(16)TDN=2(∇·u→)ρD∇c′′·∇c′′¯=τ·Aε·ρ-·ε~cSLδth.
In (16), Aε is taken to be Aε=Bε/(1+KaL)1/2 with Bε being a constant of the order of unity and KaL=(ε~δth)1/2/SL3/2 is the local Karlovitz number. The local Karlovitz number KaL dependence of Aε ensures that the strength of the dilatation rate contribution to the scalar dissipation rate ε~c transport diminishes with increasing KaL as combustion starts to exhibit the attributes of the broken reaction zones regime [26]. Following the same procedure, the local Karlovitz number KaL dependence is used in (15) in the form of 1/(1+KaL)n. However, this local Karlovitz number dependence is one of the several possibilities and any other function of KaL, which accurately predicts the quantitative behaviour of TD2 and its diminishing strength with increasing Karlovitz number as the broken reaction zones regime is approached, can in principle be used for parameterising the local KaL dependence of TD2. The numerical values of BT2, n, m1, and m2 are optimised based on a least-squares method.
The predictions of (15) are shown in Figures 3(a)–3(l) along with TD2 obtained from the DNS data for cases A–L, respectively. The model performs satisfactorily for a range of different values of Le, Ret, and τ spanning both the corrugated flamelets and thin reaction zones regimes of premixed turbulent combustion.
Variations of TD2 (solid line) and (15) (dash line) with c~ across the flame brush for cases ((a)–(l)) A–L, where all the terms are normalised with respect to SL/δth2.
The variations of (-TN1) and (-TN2) are shown in Figure 4 with c~ for cases A–L. The predictions for (-TN1) according to the (NiNj)¯s models proposed by Cant et al. [2] and Veynante et al. [7] are also shown in Figure 4 for cases A–L. The predictions of (-TN1) according to Cant et al. [2] and Veynante et al. [7] models (denoted as TN1CPB and TN1V models, resp.) are shown in Figure 4, which shows that the performances of both of the models are comparable, and both satisfactorily predict (-TN1) for all cases considered in the current analysis. However, the prediction of the TN1V model is closer to the DNS data than the TN1CPB model. The model given by (8) overpredicts the unresolved part of (NiNj)¯s (i.e., (NiNj)¯s-(Ni)¯s(Nj)¯s=δij(1-(Nk)¯s(Nk)¯s)) due to the assumption of isotropy of the unresolved flame normal fluctuations. However, experimental data [6, 7] suggested that the assumption of isotropy of the unresolved flame normal fluctuations does not hold, and this result has been verified here by the overprediction of (-TN1) by the TN1CPB model. It should be noted that (NiNj)¯s reverts to NiNj when the flame is fully resolved according to the TN1CPB model, but this condition is not satisfied by the model given by (9).
Variations of (-TN1) (solid line) and (-TN2) (dash line) with c~ across the flame brush for cases ((a)–(l)) A–L along with the predictions of TN1CPB given by (8) (star line), TN1V given by (9) (box line), and the newly proposed models for (-TN2) given by (17) using C2 given by (19i) (triangle line) and C2 given by (19ii) (cross line). All the terms in Figure 4 are normalised with respect to SL/δth2.
It is evident from (-TN)=-(eααcos2α+eββcos2β+eγγcos2γ)|∇c|¯ that a preferential alignment between ∇c and eαα (eγγ) leads to a negative (positive) contribution of (-TN). As the magnitude of (-TN2) remains either greater than or comparable to (-TN1) for all cases, it can be expected that ∇c alignment with local principal strain rates affects not only (-TN) but also (-TN2). The alignment of ∇c with local principal strain rates is determined by the competition between the strain rate due to turbulent motion aturb and the strain rate achem induced by flame normal acceleration. The scalar gradient ∇c aligns with eαα (eγγ) when achem (aturb) dominates over aturb (achem). According to Tennekes and Lumley [43] aturb can be scaled as u′/λ~(u′/l)Ret1/2~(ε~/k~)Ret1/2, whereas achem can be scaled as τf(Le,Ka)SL/δth. Thus, the effects of aturb on the term (-TN2) through the preferential alignment of ∇c with eγγ can be scaled as C1(ε~/k~)Σgen, where the model parameter C1 is expected to be a function of local turbulent Reynolds number ReL=ρ0k~2/μ0ε~. By contrast, the effects of ∇c alignment with eαα on the term (-TN2) due to the action of achem can be scaled as -C2·g(Le)τSLΣgen/δth, where g(Le) is a function of Le which increases with decreasing Le due to strengthening of flame normal acceleration as a result of augmented burning rate, and the model parameter C2 is expected to decrease with increasing KaL because the effects of achem are likely to weaken progressively with increasing KaL. Combining the above scaling estimations, the following model for (-TN2) can be proposed:
(17)(-TN2)=ε~k~(C1-τC2DaL)Σgen,
where DaL=k~SL/ε~δth is the local Damköhler number and C1 is given as
(18)C1=0.4+0.1erf[(ReL+1)12],
while the following two parameterisations are proposed for the model parameter C2:(19i)C2=0.47[1-(N)¯k(N)¯k]c~0.3Le2(1+KaL)bg(Le),(19ii)C2=(0.7-0.55erf[ReL+130])×0.47[1-(N)¯k(N)¯k]c~0.3Le2(1+KaL)0.35g(Le),where b in (19i) is given as
(20)b=0.12+0.7erf[LeReL+160]0.8,
and g(Le) from (19i) and (19ii) is defined as follows:
(21)g(Le)=[exp(-Le-0.945)-1].
The parameterisations of C1 given by (18) include turbulent Reynolds number dependence but this model parameter needs to be turbulent Reynolds number independent for high values of ReL (i.e., ReL→∝). Equation (18) ensures that C1 assumes an asymptotic value for ReL→∝. Moreover, C2 according to (19i) and (19ii) ensures that the effects of achem on the term (-TN2) weaken with increasing KaL, whereas g(Le) according to (21) accounts for strengthening of flame normal acceleration with decreasing Le. However, these local Karlovitz number dependences according to (19i) and (19ii) are only two possibilities and any other function of KaL, which accurately predicts the quantitative behaviour of (-TN2) and the diminishing strength of achem on the term (-TN2) with increasing Karlovitz number KaL, can in principle be used for C2. Similarly, any other function which satisfies the expected behaviour of C1 (g) in response to ReL(Le) can in principle be used to parameterise C1 (g) instead of (18) (see (21)). The presence of [1-(Nk)¯s(Nk)¯s] in (19i) and (19ii) ensures that the second term on the right hand side of (17) vanishes when the flow becomes fully resolved (i.e., (Nk)¯s(Nk)¯s=1.0). The numerical values of C1, C2, b, and g are optimised based on a least-squares method.
In Figure 4, the predictions based on (17) using C2 according to (19i) and (19ii) are compared with (-TN2) obtained from DNS. Figure 4 shows that (17) performs satisfactorily for the range of Ret and Le considered here. Using TN1=-nij(∂u~i/∂xj)Σgen (see (8)) and combining (15) and (17) yield the following expression:
(22)(aT)¯sΣgen=2BT2(1+KaL)nτSL(Σgen-|∇c-|)δth(1-c~)m1Lem2+ε~k~(C1-τC2DaL)Σgen+∂u~i∂xj(δij|∇c-|-nijΣgen).
It should be noted that the involvement of (Σgen-|∇c-|) in the first term on the right hand side of (22) and [1-(Nk)¯s(Nk)¯s] in C2 ensures that (aT)¯sΣgen becomes equal to (δij-NiNj)∂ui/∂xj|∇c| when the flow becomes fully resolved (i.e., (Nk)¯s(Nk)¯s=1.0). The predictions of (aT)¯sΣgen according to (22) based on the parameterisations given by (19i) and (19ii) are shown in Figures 5(a)–5(l) for cases A–L, respectively, along with the predictions of the models proposed by Cant et al. [2] (i.e., CPB model) and CFM methodology [4, 6, 7] where SR and SUR are modeled using (8) and (10) in the context of the CPB model and by (9) and (11) in the context of the CFM model, respectively. The CFM model is found to show a peak at c~≈0.3 which is quantitatively similar to cases E–G and J–L, while in cases H and I the peak value of (aT)¯sΣgen occurs towards the products (i.e., c~≈0.6) and in cases C and D the peak value of (aT)¯sΣgen occurs towards the reactants (i.e., c~≈0.2). Although the CFM model performs adequately for cases F, G, and J, it fails to capture the behavior of (aT)¯sΣgen in cases H, I, K, and L. From Figure 5, it is evident that the model CPB can be seen to predict the correct qualitative behavior for most of the cases but in cases H and I it fails to capture the correct magnitude of (aT)¯sΣgen. The new model performs well for all cases considered here, and the difference between the two parameterisations for C2 can be seen to make very little difference. The newly proposed model given by (22) with C2 given by (19i) and (19ii) also captures the behaviours of (aT)¯sΣgen satisfactorily for a range of different values of Le, Ret, and τ spanning both the corrugated flamelets and thin reaction zones regimes of premixed turbulent combustion.
Variations of (aT)¯sΣgen (solid line) with c~ across the flame brush for cases ((a)–(l)) A–L along with the predictions of CPB (dash line), CFM (star line), and the newly proposed model given by (22) using C2 given by (19i) (square line) and C2 given by (19ii) (triangle line). All the terms in Figure 5 are normalised with respect to SL/δth2.
5. Conclusions
A DNS database of freely propagating statistically planar turbulent premixed flames with a range of different values of turbulent Reynolds number Ret, heat release parameter, and global Lewis number Le spanning both the corrugated flamelets and thin reaction zones regimes of premixed combustion has been used to assess the modelling of the tangential strain rate term (aT)¯sΣgen of the generalised FSD transport equation. In order to propose a model for (aT)¯sΣgen, which will be valid for both the corrugated flamelets and thin reaction zones regimes, the statistical behaviours of the strain rate contributions to the FSD transport due to the dilatation rate and flame normal strain rate have been analysed separately.
It has been shown that the dilatation rate contribution TD to the generalised FSD transport equation strengthens with decreasing Lewis number. The contribution of normal strain rate term (-TN) assumes predominantly negative values for cases with Le≪1.0 and Da≫1 where the effects of straining due to flame normal acceleration achem dominate over turbulent straining aturb, while in the Le≈1.0 cases with Da<1 the normal strain rate term (-TN) assumes positive (negative) values towards the reactants (products) side of the flame brush. This analysis gave rise to the explicit modelling of the unresolved dilatation rate and flame normal strain rate contributions. The predictions of the new model have been compared with the predictions of the existing models (i.e., CPB and CFM models) and the performance of the newly developed model has been found to be either comparable to or better than the existing models for the FSD strain rate term.
The present analysis has been carried out using a simple chemistry based DNS database with moderate range of values of turbulent Reynolds number Ret. Although the model parameters have been found to attain asymptotic values for Ret≥50, further validations will be required based on experimental and three-dimensional detailed chemistry DNS data at higher values of Ret. Finally a posteriori assessment of the new models needs to be carried based on actual RANS simulations in a configuration for which well-documented experimental data is available, which will form the basis of future communications.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgment
The financial support of EPSRC is gratefully acknowledged.
CantR. S.BrayK. N. C.A theoretical model of premixed turbulent combustion in closed vessels1989763-42432632-s2.0-002467892910.1016/0010-2180(89)90109-0CantR. S.PopeS. B.BrayK. N. C.Modelling of flamelet surface-to-volume ratio in turbulent premixed combustion19912318098152-s2.0-5814920698810.1016/S0082-0784(06)80334-6CandelS.VeynanteD.LacasF.MaistretE.DarabhiaN.PoinsotT.LarrouturouB. E.Coherent flamelet model: applications and recent extensions1990SingaporeWorld ScientificDuclosJ. M.VeynanteD.PoinsotT.A comparison of flamelet models for premixed turbulent combustion1993951-21011172-s2.0-002768618310.1016/0010-2180(93)90055-8TrouvéA.PoinsotT. J.The evolution equation for the flame surface density in turbulent premixed combustion19942781312-s2.0-002853434510.1017/S0022112094003599ZBL0825.76899VeynanteD.DuclosJ. M.PianaJ.Experimental analysis of flamelet models for premixed turbulent combustion1994251124912562-s2.0-5814920456810.1016/S0082-0784(06)80765-4VeynanteD.PianaJ.DuclosJ. M.MartelC.Experimental analysis of flame surface density models for premixed turbulent combustion19962614134202-s2.0-003035440710.1016/S0082-0784(96)80243-8PrasadR. O. S.GoreJ. P.An evaluation of flame surface density models for turbulent premixed jet flames19981161-21142-s2.0-000207062310.1016/S0010-2180(98)00046-7HanI.HuhK. Y.Roles of displacement speed on evolution of flame surface density for different turbulent intensities and Lewis numbers in turbulent premixed combustion20081521-21942052-s2.0-4304912786910.1016/j.combustflame.2007.10.003KatragaddaM.MalkesonS. P.ChakrabortyN.Modelling of the tangential strain rate term of the flame surface density transport equation in the context of Reynolds Averaged Navier Stokes Simulation2011331429143710.1016/j.proci.2010.06.129ChakrabortyN.CantR. S.Effects of Lewis number on flame surface density transport in turbulent premixed combustion20111589176817872-s2.0-7996038699710.1016/j.combustflame.2011.01.011ChakrabortyN.CantR. S.Turbulent reynolds number dependence of flame surface density transport in the context of reynolds averaged navier stokes simulations2013341347135610.1016/j.proci.2012.07.071CandelS.PoinsotT.Flame stretch and the balance equation for the flame area19907011510.1080/00102209008951608RichardS.ColinO.VermorelO.BenkenidaA.AngelbergerC.VeynanteD.Towards large eddy simulation of combustion in spark ignition engines200731230593066BogerM.VeynanteD.BoughanemH.TrouvéA.Direct numerical simulation analysis of flame surface density concept for large eddy simulation of turbulent premixed combustion19982719179252-s2.0-0032286617VeynanteD.VervischL.Turbulent combustion modeling20022831932662-s2.0-003648910610.1016/S0360-1285(01)00017-XPopeS. B.The evolution of surfaces in turbulence19882654454692-s2.0-002383047510.1016/0020-7225(88)90004-3ZBL0641.76054ChakrabortyN.CantR. S.A priori analysis of the curvature and propagation terms of the flame surface density transport equation for large eddy simulation200719102-s2.0-3574896142810.1063/1.2772326105101ZBL1182.76128ChakrabortyN.KleinM.SwaminathanN.The effects of Lewis number on the reactive scalar gradient alignment with local strain rate in turbulent premixed flames200931114091417ChakrabortyN.SwaminathanN.Influence of the Damköhler number on turbulence-scalar interaction in premixed flames. I. Physical insight20071942-s2.0-3424820367210.1063/1.2714070045103ZBL1146.76342SwaminathanN.GroutR. W.Interaction of turbulence and scalar fields in premixed flames20061842-s2.0-3374560016010.1063/1.2186590045102ZBL1185.80011PetersN.2000Cambridge, UKCambridge University PressChenJ. H.ChoudharyA.De SupinskiB.DevriesM.HawkesE. R.KlaskyS.LiaoW. K.MaK. L.Mellor-CrummeyJ.PodhorszkiN.SankaranR.ShendeS.YooC. S.Terascale direct numerical simulations of turbulent combustion using S3D2009212-s2.0-6765069566010.1088/1749-4699/2/1/015001015001ChakrabortyN.CantR. S.Influence of Lewis number on curvature effects in turbulent premixed flame propagation in the thin reaction zones regime200517102-s2.0-3364574151610.1063/1.2084231105105ZBL1188.76026RutlandC. J.TrouvéA.Direct simulations of premixed turbulent flames with nonunity Lewis numbers1993941-241572-s2.0-002762730310.1016/0010-2180(93)90018-XMeneveauC.PoinsotT.Stretching and quenching of flamelets in premixed turbulent combustion19918643113322-s2.0-002605654210.1016/0010-2180(91)90126-VKobayashiH.KawahataT.SeyamaK.FujimaraT.KimJ. S.Relationship between the smallest scale of flame wrinkles and turbulence characteristics of high-pressure, high-temperature turbulent premixed flames,200229217921800RutlandC.CantR. S.Turbulent transport in premixed flamesProceedings of the Summer Program1994Centre of Turbulence Research, NASA Ames/Stanford University7594WrayA. A.Minimal storage time advancement schemes for spectral methods1990MS 202 A-1Mountain View, Calif, USANASA Ames Research CenterRogalloR. S.Numerical experiments in homogeneous turbulence198181315Mountain View, Calif, USANASA Ames Research CenterHanI.HuhK. Y.Roles of displacement speed on evolution of flame surface density for different turbulent intensities and Lewis numbers in turbulent premixed combustion20081521-21942052-s2.0-4304912786910.1016/j.combustflame.2007.10.003CharletteF.MeneveauC.VeynanteD.A power-law flame wrinkling model for LES of premixed turbulent combustion part I: non-dynamic formulation and initial tests20021311-21591802-s2.0-003680962610.1016/S0010-2180(02)00400-5HaworthD. C.BlintR. J.CuenotB.PoinsotT. J.Numerical simulation of turbulent propane-air combustion with nonhomogeneous reactants200012133954172-s2.0-003362263210.1016/S0010-2180(99)00148-0JiménezC.CuenotB.PoinsotT.HaworthD.Numerical simulation and modeling for lean stratified propane-air flames20021281-21212-s2.0-003600759910.1016/S0010-2180(01)00328-5GroutW. R.An age extended progress variable for conditioning reaction rates200719102-s2.0-3574897884410.1063/1.2773998105107ZBL1182.76291ChakrabortyN.HartungG.KatragaddaM.KaminskiC. F.A numerical comparison of 2D and 3D density-weighted displacement speed statistics and implications for laser based measurements of flame displacement speed201115813721390ChakrabortyN.KleinM.CantR. S.Effects of turbulent reynolds number on the displacement speed statistics in the thin reaction zones regime of turbulent premixed combustion20112011192-s2.0-8005271402410.1155/2011/473679473679ChakrabortyN.CantR. S.Direct numerical simulation analysis of the flame surface density transport equation in the context of large eddy simulation20093214451453HawkesE. R.CantR. S.Implications of a flame surface density approach to large eddy simulation of premixed turbulent combustion20011263161716292-s2.0-003482338010.1016/S0010-2180(01)00273-5AshurstW. T.KersteinA. R.KerrR. M.GibsonC. H.Alignment of vorticity and scalar gradient with strain rate in simulated Navier-Stokes turbulence1987308234323532-s2.0-002314737510.1063/1.866513BatchelorG. K.The effect of homogeneous turbulence on material lines and surfaces1952231349366ZBL0046.42201ChakrabortyN.RogersonJ. W.SwaminathanN.A priori assessment of closures for scalar dissipation rate transport in turbulent premixed flames using direct numerical simulation20082042-s2.0-4314912335310.1063/1.2903846045106ZBL1182.76126TennekesH.LumleyJ. L.199917thCambridg, UKMITKollaH.RogersonJ. W.ChakrabortyN.SwaminathanN.Scalar dissipation rate modeling and its validation200918135185352-s2.0-6614909339510.1080/00102200802612419BrayK. N. C.LibbyP. A.MossJ. B.Unified modeling approach for premixed turbulent combustion-part I: general formulation1985611871022-s2.0-002210141310.1016/0010-2180(85)90075-6