Vous êtes sur la page 1sur 28

Flow, Turbulence and Combustion 72: 1–28, 2004.

1
© 2004 Kluwer Academic Publishers. Printed in the Netherlands.

Large Eddy Simulation of Premixed Turbulent


Combustion Using  Flame Surface Wrinkling
Model

G. TABOR1 and H.G. WELLER2


1 School of Engineering, Computer Science and Mathematics, Harrison Building, University of
Exeter, North Park Road, Exeter EX4 4QF, U.K.; E-mail: g.r.tabor@ex.ac.uk
2 Nabla Ltd., The Mews, Picketts Lodge, Picketts Lane, Salfords, Surrey RH1 5RG, U.K.

Received 24 April 2002; accepted in revised form 5 May 2003


Abstract. One commonly-used method for deriving the RANS equations for multicomponent flow
is the technique of conditional averaging. In this paper the concept is extended to LES, by introducing
the operations of conditional filtering and surface filtering. Properties of the filtered indicator function
b are investigated mathematically and computationally. These techniques are then used to derive
conditionally filtered versions of the Navier–Stokes equations which are appropriate for simulating
multicomponent flow in LES. Transport equations for the favre-averaged indicator function  b and the
unresolved interface properties (the wrinkling and the surface area per unit volume) are also derived.
Since the paper is directed towards modelling premixed combustion in the flamelet regime, closure
of the equations is achieved by introducing physical models based on the picture of the flame as a
wrinkled surface separating burnt and unburnt components of the fluid. This leads to a set of models
for premixed turbulent combustion of varying complexity. The results of applying one of this set of
models to propagation of a spherical flame in isotropic homogeneous turbulence are analysed.

JEL Codes: D24, L60, 047.

Key words: Large Eddy Simulation, premixed turbulent combustion.

1. Introduction
Premixed Turbulent Combustion is a highly complex process, but one which greatly
affects everyday life. The quest to understand the physical processes better is con-
tinual, and one aspect of it is the search for computational models to describe
the processes involved. Such models must of necessity be less detailed than the
physical processes occurring in the system, but should aim to capture the essence
of these processes. In turn, the models can provide a greater understanding of the
processes involved, and provide us with the ability to predict the behaviour of
specific combustion systems. Thus they are of great importance in the design of
combustion devices such as Internal Combustion (IC) engines and gas turbines.
A working model of turbulent combustion must provide adequate treatments for
the turbulence, the chemical reactions of the combustion (and consequential heat
2 G. TABOR AND H.G. WELLER

release), as well as the mutual interaction of these areas, since the combustion alters
the physical properties of the fluid and drives the flow, whilst the flow moves reac-
tants and products around and thus influences the combustion. About the simplest
possible model combines a Reynolds Averaged Navier–Stokes (RANS) description
of the turbulence with a simplistic model of the combustion which provides a
model for the heat release as a straightforward function of the reactant species
concentration (for example, the Eddy Breakup model of Spalding [31]). Numerous
improvements on these simple models have been investigated over the years, in
particular concentrating on improved methods for characterising the species con-
centration at a point (and thus the prediction of the heat release) by PDF techniques,
or improved flame modelling.
Large Eddy Simulation (LES) of premixed turbulent combustion is an active
area of research. It offers the possibility of significant improvements over RANS,
in terms of accuracy of the solution, the ability to handle counter-gradient diffusion,
and the provision of greater information about the turbulent flow field which ren-
ders irrelevant some of the modelling assumptions necessary in RANS combustion
models. The approaches used in LES are based on various ways of computationally
following the flame front. In premixed combustion the flow consists of regions of
unburnt reactants and regions of combusted products. The extent of combustion
of the gas can be described in various ways in terms of a progress variable taking
values between 0 and 1, with the extreme values indicating the presence of unburnt
or fully burnt phases, and the transition between these values marking the flame
front. This can be linked directly to physical properties of the gas, for instance by
utilising normalised temperature (T ) or product mass fraction (Y ):
T − Tu YP
c= or c = . (1)
Tc − Tu YP ,b
The exact linkage is not important however, and the progress variable can be con-
sidered simply as indexing the ammount of combustion, however defined. In this
paper we will use a progress (technically a regress) variable b = 1 − c with
b ∈ [0, 1], where 0 represents fully burnt gas and 1 unburnt gas.
There is a problem here though. In the flame-sheet regime of high Damköhler
and Reynolds numbers the reaction zone is very thin indeed and the transition in
the progress variable is too sharp to be explicitly resolved on the LES mesh. One
approach sometimes employed is the Thickened Flame (TF) approach [7, 10]. In
the TF method the flame front is artificially thickened by multiplying the thermal
and molecular diffusivities by a factor F and reducing the reaction rate by the same
factor. The result is a thickened flame front with the same laminar flame speed
Sl , which can be resolved on the LES computational mesh, and thus its motion
can be calculated without additional SGS modelling. This has several advantages,
simplifying the chemical reaction modelling and eliminating the need for ad hoc
submodels for ignition and flame-wall interactions. However it does involve alter-
ing the physics of the flame front in a substantial manner. In particular the response
LES OF PREMIXED TURBULENT COMBUSTION 3

of the flame to unsteady phenomena and to strain induced by the velocity field is
modified by the thickening procedure [1, 10].
The most common approach for LES, known as the G-equation method, is based
on a level-set approach [3, 16, 19, 21, 22]. Here a function G is constructed to have
the property that the zero value isosurface represents the combustion interface. G
is not related to the progress variable, so other values of G have no physical sig-
nificance and are merely chosen for computational convenience. A straightforward
transport equation is then solved for G:
∂G
+ U.∇G = ST |∇G|, (2)
∂t
where U is the fluid velocity and ST the turbulent flame speed, i.e. the rate of
propagation of the flame front due to combustion. The challenge in this approach
comes from developing adequate modelling for the turbulent flame speed which is
a well-defined quantity that depends on local flow conditions [24]. There are also
numerical problems with the accurate propagation of G.
The other option for simulation of combustion in this regime is to link the
progression of the flame front to additional physical properties, e.g. geometric
properties of the surface. In the G-equation, surface stretch and curvature effects
are treated by consideration of higher moments of G. An alternative class of mod-
els can be constructed based on solving for variables describing these geometrical
parameters [33]. In RANS, the basic ‘laminar flamelet’ models have been extended
[8, 9, 20, 27]: the flame front propagates locally as a laminar flame but at the
same time is being wrinkled due to interactions with the turbulence. The flame
propagation speed can be modelled in terms of the laminar flame speed (a known
quantity) and the degree of wrinkling of the flame at the point, given by the flame
area per unit volume . The system as a whole is described in terms of transport
equations for the filtered progress variable and for . This approach has also been
investigated for LES [2, 15]. An alternative RANS model proposed by Weller
[34, 35], represents the geometric properties of the flame front in terms of the
density of wrinkling , which is the flame area per unit area resolved in the mean
direction of propagation. This choice of variable makes the modelling somewhat
easier compared with the equivalent equation for , for instance by separating out
a term representing flame annihilation by cusp formation. It also provides for a
spectral analysis of the flame-turbulence interaction [36]. This RANS model was
formulated using the technique of Conditional Averaging [11]. The aim of our
current work is to formulate an LES version of this model. In order to do so we
must introduce an analogous techique, that of Conditional Filtering, to derive the
transport equations for a multicomponent system. This technique is the subject
of this paper. Section 2 introduces the concept of Conditional Filtering in LES,
and discusses the regularity of the flame surface in relation to the surface filter-
ing process introduced as part of the analysis. The effect of filtering a simulated
indicator function appropriate for combustion is investigated in 2-d. In Section 3,
Conditional Filtered versions of the Navier–Stokes Equations (NSE) are presented,
4 G. TABOR AND H.G. WELLER

together with transport equations for properties of the surface, in particular area
per unit volume  and wrinkling . Finally, in Section 4 possible closures of the
equations are discussed. The evolution of a spherical flame is calculated using such
a model and its properties discussed.

2. Conditional and Surface Filtering


The Navier–Stokes Equations (NSE) for a compressible fluid are
∂ρ
+ ∇.ρU = 0,
∂t
∂ρU
+ ∇.(ρU ⊗ U) = −∇p + ∇.S,
∂t
∂ρe
+ ∇.ρeU = −p∇.U + S.D + ∇.κ∇e, (3)
∂t
where
1
S = λ∇.UI + 2µD, D = (∇U + ∇UT ). (4)
2
In Conditional Averaging in RANS, an indicator function is introduced [11] which
takes the value 1 in the unburnt region phase and 0 in the burnt region. The NSE
are multiplied by this function and then ensemble-averaged: the indicator function
projects out one of the components, and so the resulting equation is for that com-
ponent alone. This process introduces additional terms (in addition to the standard
Reynolds Stress term arising from the ensemble averaging process) which can
be written in terms of a surface average operation which represents the effect of
the interface on the dynamics of the phase under consideration. These terms will
commonly require modelling. Transport equations can also be formulated in this
way for the ensemble averaged indicator function, which has the interpretation of
the probability of finding the phase at that point, and for quantities relating to the
small-scale geometry of the interface.
In LES it is assumed that the dependent variables in the NSE can be decomposed
into GS and SGS components, i.e. ψ = ψ + ψ  . The GS component is obtained
by filteringψ, which is a convolution between it and a filter function G with the
properties D G(x) d3 x = 1, lim→0 G(x, ) = δ(x) and G(x, ) ∈ C n (R3 )
with compact support. The decomposition into mean and fluctuating components
is thus analogous to the decomposition in RANS, but with differing interpretations
of the resulting variables. We can adapt LES to include the concept of conditional
averaging by introducing an indicator function I which is a generalised function
(or distribution) such that

1 if (x, t) is in phase A (say the unburnt gas),
I(x, t) = (5)
0 otherwise.
LES OF PREMIXED TURBULENT COMBUSTION 5

Figure 1. Local curvilinear coordinates for the flame surface. Here x⊥ is the direction normal
to the flame surface, and (η, ς) coordinatize the flame surface.

Since I is a generalised function we can take its gradient. This will be zero every-
where but on the interface, where it will have the direction of the outward facing
normal. If we introduce a coordinate system at the interface (x⊥ , x ) where x⊥ is
the direction normal to the interface and x = (η, ς ) are coordinates in the interface
manifold (see Figure 1), then

I(x⊥ , η, ς ) = (x⊥ ),
1 1
∇I = ∂⊥ (x⊥ )n⊥ = δ(x⊥ )n⊥ , (6)
h⊥ h⊥

where is the Heaviside step function. Here h2⊥ is the appropriate element of
the metric tensor for this coordinate system, and thus relates to the curvature (and
hence the wrinkling) of the surface. n⊥ is a unit vector in the x⊥ direction, i.e.
normal to the actual surface.
We can define an interface velocity UI such that

dI ∂I
= + UI .∇I = 0. (7)
dt ∂t
Note that UI is defined here to be the total velocity of the interface irrespective of
whether it is due to advection by the flow or generation (or removal) of phase A.
From this we have the relation
∂I 1
= − UI .n⊥ δ(x⊥ ). (8)
∂t h⊥
6 G. TABOR AND H.G. WELLER

We can now introduce a ‘conditional filtering’ taking the form



ψ = G ∗ (Iψ) = G(x − x )I(x , t)ψ(x , t) d3 x , (9)
D

for a tensor ψ of any rank. ψ is the phase-weighted value of ψ at any point. The
combustion progress variable b can be introduced here as a GS indicator function
by writing ψ = b ψu , where b (x, t) is the probability of the point (x, t) being in
the unburnt gas (phase u). Setting ψ = 1, we find

b (x, t) = G(x − x )I(x , t) d3 x . (10)
D

Since combustion involves compressible flow, we need to deal with density varia-
tions. Typically terms will involve the product ρψ, which we can write

ρψ = b ρψu (11)
u based on the u-phase to split up this
and can define a density-weighted average ψ
second term:
u ,
ρψu = ρu ψ (12)
thus giving
u .
ρψ = b ρu ψ (13)
We also need to be able to deal with spatial and temporal derivatives. If we
assume a constant filter size (of course, issues may arise for non-uniform filters,
see [12]),

∇ ◦ ψ = ∇ ◦ {G ∗ (Iψ)} = G ∗ {∇ ◦ (Iψ)}, (14)


where ∇◦ indicates an appropriate tensor derivative. Also

∂ψ ∂ ∂(Iψ)
= {G ∗ (Iψ)} = G ∗ . (15)
∂t ∂t ∂t
From filtering the identity
I∇ ◦ ψ = ∇ ◦ (ψI) − ψ ◦ ∇I, (16)
using Equation (6), we have

1 3 
∇ ◦ ψ = ∇ ◦ ψ − G(x − x )ψ(x ) ◦ n⊥ δ((x − xI ).n⊥ ) dx
h⊥
D
  
= ∇ ◦ ψ − ψ ◦ n⊥  (17)
LES OF PREMIXED TURBULENT COMBUSTION 7

Here we have defined a surface filtering operation as

 1 1 3 
ψ = G(x − x )ψ(x )δ((x − xI ).n⊥ ) dx (18)
 h⊥
D

Because of the term δ((x − xI ).n⊥ ) this integral is non-zero only in the immediate
vicinity of the interface. If we transform coordinates in this integral to the (x⊥ , x )
set, then we can perform the x⊥ integral, leading to
 
 1
ψ = G⊥ (x⊥ − x⊥ )ψ(x⊥ , x )δ(x⊥ − x⊥,I ) dx⊥

× G (x − x )|J| d2 x

1
= G⊥ (x⊥ − x⊥,I ) G (x − x )ψ(x⊥,I , x )|J| d2 x ,

assuming that the filter G can be split into components perpendicular and parallel to
the interface. |J| is the Jacobian for the (η, ς ) coordinate system. In other words,
the variable ψ is filtered only on the interface manifold, but the influence of the
interface is extended by the filter to its immediate neighbourhood. Writing ψ = 1
in this relation and changing variables as before, gives

 = G⊥ (x⊥ − x⊥,I ) G (x − x )|J| d2 x . (19)

Since |J| d2 x is the area element on the surface,  has the interpretation of being
the ammount of interface within the filter support. In a similar way, the identity
∂ψ ∂ψI ∂I
I = −ψ (20)
∂t ∂t ∂t
generates the commutation relation

∂ ∂ψ   
ψ− = ψUI .n⊥ . (21)
∂t ∂t
Finally, substituting ψ = 1 into Equation (17) gives us
  
∇b = n⊥ , (22)
relating the progress variable b (GS indicator function) to the surface filtered
  
direction vector n⊥ . The identity

∂b   
= −UI .n⊥  (23)
∂t
can be derived in a similar manner from Equation (21), or by filtering Equation (8).
We can decompose the motion of the interface UI into the motion due to advection
8 G. TABOR AND H.G. WELLER

and a term −va n⊥ which is due to the advance of the interface relative to the flow,
and so is due to the generation or destruction of phase A: UI = U + va n⊥ . This
produces the result
∂b    
+ U.n⊥  = − va . (24)
∂t
  
      
The cross term can be split as U.n⊥ = U .n⊥ + v .n ⊥ , where the hashes represent
  

surface fluctuating components. Writing Dc = v .n⊥  gives

∂b  
+ U .∇b + Dc = − va  (25)
∂t

2.1. REGULARITY OF THE SURFACE


In the preceeding section it is implicitly assumed that the surface defined by the
indicator function is well behaved. In this context, the surface is well behaved if
it is a 2-d submanifold of R3 . Over the majority of the surface this will be the
case; however potentially we will need to consider separately a number of points at
which problems arise. Pope [27] distinguishes four cases where the surface regular-
ity breaks down: singularities, internal edges, self-intersections and critical points.
Of these, internal edges do not occur in our case, since our surface delineates a
boundary between distinct areas of space and is at all times closed.
Consider the case of two spherical fronts expanding towards each other un-
til they touch, which is the intersection problem described by Pope. If I is C n -
continuous for the time being, then this corresponds to the formation of a saddle
point in I . At such a point, ∇I = 0, and the surface is no longer parameter-
isable. If the surfaces meet tangentially, the curvature at this point is infinite. If
the spheres merge further, it is conceivable that the point will become a line (ac-
tually a ring) at which one of the radii of curvature is infinite. This could be
described as a negative line cusp. Other forms of propagation can also lead to
the development of other singularities at points or lines. However since it is area-
based properties which are relevant for the front propagation, and these structures
have zero area, this should not affect the modelling. Consider the tip of a cone,
which can be represented as part of a sphere of radius r: it has curvature 1/r 2 , but
an area contribution r 2 sin θ dθ dφ. Hence the contribution of this surface to any
curvature-related properties → 0 as r → 0.
Related problems arise with the calculation of b , and related smoothed vari-
ables. The spatial smoothing removes completely the possibility of infinite surface
curvature, but increases the likelyhood of partial or total overlap. Such overlap will
cause problems if the propagation of the surface is linked to |∇b |, as it is in the
Weller combustion models [34]. In this case the indicator function of the combined
flame will be different from those of the individual flames, leading to an incorrect
LES OF PREMIXED TURBULENT COMBUSTION 9

Figure 2. Filtering of flame front in 2-d. (a) gives the physical flame front as generated by
‘wrinkles’, whilst (b)shows the Delaunay triangulation of this curve. (c) and (d) are the result
of filtering the indicator function using a Gaussian with σ = 5 and σ = 25 (the box dimension
is 100 units). Meshes of these cell sizes are also shown.

value for |∇b |, and an error in the propagation. This situation occurs for example
in the early stages of propagation of a spherical flame front, i.e. as the ignition
kernel propagates outwards. Thus at the start of the simulation appropriate ignition
modelling must be provided. In the case presented here a very simplistic model is
used in which the ignition energy is distributed over a small number of cells at the
centre of the computational domain, but this can probably be improved upon.
10 G. TABOR AND H.G. WELLER

2.2. 2- D SIMULATION OF b
Weller et al. [36] present a model of a combusting flame front using a length-scale
decomposition of turbulence into eddies. The interaction of these individual eddies
with a closed curve can be calculated and the net effect of all eddies generated by
superposition, resulting in a 2-d closed curve with similar geometric properties to
a real turbulent flame: essentially a simulated snapshot of a slice through the flame.
Weller et al. wrote a code (‘wrinkles’) based on this to generate and analyse flame
fronts in 2-d. The original aim was to validate the spectral modelling used in the
Weller RANS model. Here we use the output from this code is to look at the effect
of spatial filtering on a simulated indicator function.
The wrinkles code outputs the flame surface (a line in 2-d) as a series of line
segments. The result of applying it to a circle is shown in Figure 2a, which is a
good approximation to a spherical flame front in 2-d. The first step is to determine
the inside and outside of the closed curve, which is less easy than it sounds. The
points defining the curve were used as the input to a code ‘Triangle’, which is a
2-d mesh generator using constrained quality conforming Delaunay triangulation
[29, 30]. This generates a series of points (and associated triangulation) on the
inside of the curve. The value 1 is then allocated to all these points, thus creating
a representation of the indicator function. To filter the indicator function, the area
was then sampled using a regular 2562 grid using the visualisation code VTK [28].
The data on this regular mesh was then convolved with a 2-d Gaussian function
of width σ : exp (−r 2 /2σ 2 ) and the resulting function contoured. Figures 2c and
2d show the result for σ = 5 and σ = 25 respectively. Meshes with cells of
size σ = 5 and σ = 25 are superimposed for comparison. These show the grid
scale appropriate for a real CFD calculation (see Section 4.2 below). The sampling
plane was of side 100 units, so for the finest mesh each cell contains > 100 points,
providing adequate resolution of the filtered indicator function.
Several points are demonstrated here. The filtered indicator function is resolved
by 2–3 mesh cells. This is slightly sharper than for actual calculations (see Section
4.2) but not much. Meanwhile there is clearly a large degree of wrinkling which is
not resolved and which will be modelled by the wrinkling variable  derived in the
next section. Figure 2c would be appropriate for a well-developed, large spherical
flame, and clearly the overall structure is well resolved. Figure 2d could be taken as
an earlier version of the flame shortly after ignition. With only 6 CFD cells across
the structure, b is inadequately resolved, and the two sides of the flame will interact
in an undesirable manner. This demonstrates that the ignition and very early growth
of the flame will be difficult to simulate, but once the flame has grown sufficiently
the model is adequate for the purpose. Although this is not a perfect analog of the
real situation, not least because the simulated surface is not propagating, it does
indicate the basic idea is sound.
LES OF PREMIXED TURBULENT COMBUSTION 11

3. Conditionally Filtered Equations


3.1. CONTINUITY EQUATION
We can now filter the NSE’s (Equations (3a–3c)) using these relations. Starting
with the continuity equation, we have
∂ρ   
+ ∇.ρU =  ρ(U − UI ).n⊥ . (26)
∂t
Since UI = U + va n⊥ , and using Equation (12), Equation (26) can be written
∂b ρu u = −ρv
 
+ ∇.b ρu U a . (27)
∂t
The definition of b provides a measure of the large-scale geometry of the sur-
face. Substituting n⊥ into Equation (18) and using the same coordinate transfor-
mation:

   1
n⊥ = G⊥ (x⊥ − x⊥,I ) G (x − x )n⊥ (x⊥,I , x )|J| d2 x . (28)

  
Now n⊥ |J| d2 x is a directed area element on the surface, so n⊥ has the interpreta-
tion of being the ammount of directed interface within the filter support. This can
be related to the GS interface direction nf
  nf
n⊥ = (29)

with a wrinkling factor  defined as
1 
= =  . (30)
  
|n⊥ |  G⊥ (x⊥ − x⊥,I ) G (x − x )n⊥ (x⊥,I , x )|J| d2 x 
From this we see that  gives the total subgrid surface area divided by the smoothed
surface area, i.e. the wrinkling. nf is of course a unit vector in the same direction.
Substituting Equation (22) gives

= (31)
|∇b |
with |∇b | the area of the grid scale surface.

3.2. MOMENTUM , ENERGY EQUATIONS


Similarly, conditional filtering the momentum equation gives
∂ u ) + ∇.(b ρu U u )
u ⊗ U
(b ρu U
∂t
 
     
= −∇b pu + ∇. b Su − Bu + (pI − S).n⊥ − ρva U , (32)
12 G. TABOR AND H.G. WELLER

Bu = (ρU ⊗ U)u − ρu U u .
u ⊗ U (33)

The terms in [ ] in (32) represent the effects of the interface on the momentum
balance.
This leaves the energy equation. The l.h.s. of (3c) can be treated straightfor-
wardly:

∂ρe ∂b ρu eu
+ ∇.ρeU = u + ∇. b (ρeU)u − b ρu eu U
+ ∇.b ρu eu U u
∂t ∂t
  
+ ρeva .

Following normal practice, we write b (ρeU)u − b ρu eu Uu = b b. The term on the
third line represents the contribution of the work done by the interface during its
motion to the energy transport. The r.h.s. is more involved. We write

−p∇.U + S.D + ∇.h = −b (p∇.U)u + b (S.D)u + ∇.b hu



u + b ρu πu
= − b ρu ∇.U
   
+ b Su .Du + b ρu εu + ∇.b hu − h.n⊥ 

with
u ,
ρu πu = (p∇.U)u − pu ∇.U
ρu εu = (S.D)u − Su .Du ,

defining the SGS pressure dilatation and dissipation π and ε, respectively. Equating
the two expressions, we have

∂b ρu eu  
u = − b ρu ∇.U
+ ∇.b ρu eu U u + b ρu πu + b Su .Du + b ρu εu
∂t
      
+ ∇.b hu − bu + ρeva − h.n⊥ . (34)

The term in brackets is now the total energy contribution from the interface.

3.3. DERIVATION OF THE  EQUATION


The transport equation for the flame area per unit volume  has been derived for
the RANS case on several occasions [8, 27, 33]. A brief recap is given here using
the conditional filtering methodology. We start by restating the definition of ,
rewritten in (x⊥ , η, ς )-coordinates as

 = G(x − x )δ(x⊥ − xI )|J| dx⊥ d2 x (35)
D
LES OF PREMIXED TURBULENT COMBUSTION 13

Here |J| is the Jacobian for the surface: the Jacobian for the entire transformation
being h⊥ |J|. Candel and Poinsot give the transport equation for a surface element
δA as
dδA
= (−n⊥ .∇UI .n⊥ + ∇.UI )δA. (36)
dt
Our surface element

δA = |J| d2 x

= δ(x⊥ − xI )|J| dx⊥ d2 x

1
dx .
3
= δ(x⊥ − xI )
h⊥
If we multiply (36) by G and integrate over the surface, then the l.h.s. of the
equation becomes

dδA d   
G = + ∇.UI  (37)
dt dt

since the volume δV = dx 3 changes as the surface propagates. Hence (36) be-
comes
d ∂   
= + (UI .∇) = −n⊥ .∇UI .n⊥ . (38)
dt ∂t

3.4. DERIVATION OF THE  EQUATION


A transport equation for the wrinkling factor introduced earlier can now be derived.
Differentiating (31)

∂ 1 ∂  ∂|∇b |
= − . (39)
∂t |∇b | ∂t |∇b | ∂t

Now, by differentiating |∇b | = nf .∇b ,

∂|∇b |  


= −Ut .∇∇b − nf .∇ Ut .∇b − nf .∇Dc ,
∂t

where Ut is the surface-filtered effective velocity of the flame, defined so that

∂b 
+ Ut .∇b = 0. (40)
∂t
14 G. TABOR AND H.G. WELLER

Since
 2 
 
∇ = ∇ ∇b .∇b = 2(∇∇b ).∇b


|∇b | 
= 2 ∇ − |∇b |∇

and so

∂|∇b | Ut  
= − . ∇ − |∇b |∇ − nf .∇ Ut .∇b . (41)
∂t 
Substituting this in Equation (39), and using Equation (38) to eliminate  we have
∂     
+ U .∇ = −n⊥ .∇UI .n⊥ + nf .∇ Ut .nf
∂t
    ∇|∇b |
+  Ut − UI . . (42)
|∇b |
The first term on the r.h.s. here can be further split into surface mean and fluctuating
components
   1   
−n⊥ .∇UI .n⊥ = n⊥ ∇ UI .n⊥ − Q, (43)

where Q involves surface fluctuation terms such as n ⊥ (Section 2) and represents
interaction between the interface and the turbulence leading to generation or re-
  
moval of interface area. The term n⊥ ∇ UI .n⊥ can be grouped conceptually with
the second term in (42), representing the effect of mean stretch and propagation on

. The final term involves the difference between overall propagation velocity Ut
  
and average interface velocity UI , which increases with interface distortion. It also
involves the second spatial derivative term ∇|∇b |/|∇b |. Evaluating this through a
hypothetical Gaussian flame, this term tends to +∞ at the back of the flame and
−∞ at the front. At the front this term forces  → 1, i.e. it has the effect of
smoothing the interface, whilst at the back it has the reverse effect. Thus this term
represents the process of cusp formation in the flame front.

4. Spherical Flame Front


A convenient example is the growth of a spherical flame front in stationary, isotropic
and homogeneous turbulence. Homogeneous isotropic turbulence can be simulated
on a regular cubic grid of side 2n (here grids 64 cells on a side are used, i.e. 643
cells in total) using large scale forcing to generate turbulence [13]. The growth
of a spherical flame front can be simulated in such a case, and comparisons with
LES OF PREMIXED TURBULENT COMBUSTION 15

experimental data have been presented for just such a case [23]. Here a recap is
provided of the detailed combustion modelling [37] (Section 4.1) and the results
compared with theoretical expectations (Section 4.2). The case used is case A from
[23]: an isooctane mixture at 1 atm, initial temperature 358 K, u = 2.36 m/s.
The integral length scale L for this case has been measured at 20 mm, giving
a turbulent Reynolds number based on this of ReT = 1700. The laminar flame
speed has been measured as Su = 0.434 m/s, and the laminar flame thickness
δl = 0.04 mm [5], giving a Damköhler number of 130, and a turbulent Karlovitz
number 0.3, indicating that the case falls easilly into the regime of laminar flamelet
combustion [6].

4.1. DETAILED MODELLING


The final stage in formulating the combustion model is to provide models for those
terms which cannot be explicitly expressed in terms of known quantities, or which
are too difficult to evaluate stabily or accurately. To close these equations, models
have to be developed for the sub-grid scale (SGS) stress tensor, flux vectors and
dissipation as well as for the filtered reaction rate. The SGS stress tensor and flux
vectors are not unique to reacting flows, and hence standard models may be applied.
The principle difficulty in reacting LES is the proper treatment of the reaction zone;
since the characteristic scales for the reaction processes are below the filter width,
mean reaction rate models are required. Within our framework this is represented
by the problem of solving equations for the geometric variables b and .
Here we describe in brief the two-equation formulation of the Weller combus-
tion model. Simpler models substituting algebraic equations for the various trans-
port equations described here lead to alternative one-equation and algebraic for-
mulations, further details being discussed in [37]. Rather than solve for both burnt
and unburnt phases separately we combine them and solve for phase-weighted
quantities such as the total density
ρ = ρu b + ρc (1 − b ). (44)
We can introduce a new regress variable 
b defined so that
ρ
b = ρu b . (45)
Note that as b ∈ [0, 1], 
b ∈ [0, 1].
Combining (31) and (45) with the b equation (27) gives
∂

bρ 
+ ∇.( Uu ) = −ρ u Su |∇b |, (46)
∂t
  
where we have modelled the interface advance term ρva in terms of the laminar
flame speed Su and the unburned gas density ρu . The conditionally filtered unburnt
gas velocity 
Uu may be decomposed into the unconditioned density weighted ve-
Uu = 
locity and the unburnt-burnt gas slip velocity thus  U+(1− b)
Uuc . By analogy
16 G. TABOR AND H.G. WELLER

with the properties of laminar flames, the velocity difference across a turbulent
flame is a function of the turbulent flame speed and the density ratio, in which case
performing a simple one dimensional analysis of the turbulent flame brush there
results
 
 ρu
Uuc ≈ − 1 Su nf . (47)
ρc
However, this does not correctly account for the distribution of ; the assump-
tion being that  is constant through the flame, nor curvature effects or turbulent
fluctuations in the phase velocities. Noting that the required transport form of the
resulting equation must be the same as that of the rest of the filtered transport
equations, a model for the turbulence effect is postulated, for consistency, as a
simple gradient diffusion form, producing the following model for the slip velocity:
 
 ρu ∇ b
Uuc = − 1 Su nf − D , (48)
ρc 
b(1 − b)
 is the sub-grid diffusion coefficient.
where the turbulent diffusion coefficient D
This models two types of effects. The second term is a gradient transport term,
indicating transport proportional to ∇b. The first term however can act in the
opposing direction and can thus represent the counter-gradient transport effects
considered important in most premixed and partially premixed combustion devices
[34].
Combining Equations (46, 48) and rearanging yields

∂ρb
+ ∇.(ρ 
U
b) − ∇.(ρD∇ b)
∂t
 
  ρu
= −∇.b(1 − b)ρ − 1 Su nf − ρ u Su |∇b |. (49)
ρc
Some manipulation on Equations (44, 45) gives the result
 
ρu
ρu − ρ = ρ(1 − b) −1 (50)
ρc
Using this we can break down the first term on the r.h.s. of (49),
 
ρ
∇.b(1 − 
b)ρ u
− 1 Su nf
ρc
= ∇.b (ρu − ρ) Su nf

= ∇.  b − b ρu Su nf
 
= ρu Su nf .∇  b−b +  b − b ∇.ρu Su nf .
LES OF PREMIXED TURBULENT COMBUSTION 17

If we write nf .∇
b |∇
b|, then (49) becomes
∂ρb
+ ∇.(ρ Ub) − ∇.(ρD∇ b)
∂t

= −ρ u Su |∇b| − b − b ∇.ρu Su nf . (51)
In the final term on the r.h.s. here, gradients in ρu are likely to be negligable, and
so we choose to ignore this term and use the equation
∂ρb
+ ∇.(ρ  Ub) − ∇.(ρD∇ b) = −ρ u Su |∇ b| (52)
∂t
as our modelled equation for  b.
We also need to model terms in Equation (42). The first two terms represent the
effects of strain and propagation respectively on the SGS wrinkling . The third
term on the r.h.s. of Equation (42) represents the effect of differential propagation
on the distribution of  through the flame, reducing generation at the front of the
flame and enhanced generation at the back. This term involves high order deriva-
tives which create numerical difficulties for LES. Instead of trying to model each
term in detail we develop einsatz models which represent generation and removal
of wrinkling, modelled by terms G and R( − 1). The problems associated with
these derivatives are thus avoided by including the effect directly into the model
for G, resulting in the following simplified equation for 
∂ 
+ U s .∇ = G − R( − 1) + max[(σs − σt ), 0]. (53)
∂t
The final term here relates to the resolved strain rates
1  T 1       T
σt = ∇ Ut + ∇ Ut  and σs = ∇ UI + ∇ UI .
2 2
The bounding here is introduced to account for the effects of lateral compression
on the flame surface, allowing only removal of wrinkling by resolved extensive
strain.
A spectral approach is applied to the modelling of the turbulence-flame in-
teraction, in which the wrinkling of the flame is decomposed into a length-scale
spectrum [34]. This approach lends itself naturally to LES in that sub-grid flame
properties may be obtained by integrating over the appropriate range of the spec-
trum. However solution of the spectral evolution equations simultaneously with the
transport equation for  is prohibitively expensive and simple algebraic models are
considered more appropriate. The current approach is based on the flame-speed
correlation of Gülder [14], which has proved particularly good by comparison with
full spectral solutions, leading to
eq − 1 0.28 ∗eq
G=R , R= , (54)
eq τη ∗eq − 1
18 G. TABOR AND H.G. WELLER

u
∗eq = 1 + 0.62 Rη , eq = 1 + 2(1 − b )(∗eq − 1), (55)
Su

where τη is the Kolmogorov time-scale, u is the sub-grid turbulence intensity and


Rη is the Kolmogorov Reynolds number. The surface filtered velocity of the flame
  
UI is modelled in a similar manner to the conditionally filtered unburnt gas velocity
(see Equation (48)) as
   ∇.(ρD∇ b)
UI = 
U + (ρ u /ρ − 1)Su nf − nf . (56)

ρ|∇ b|
The resolved strain-rate σt is obtained from 
U and the sub-grid turbulent flame
speed Su  by removing the dilatational component from the strain-rate in the
direction of propagation nf ,
σt = ∇.(
U + Su nf ) − nf .(∇(
U + Su nf )).nf . (57)
The gas expansion due to combustion is assumed to occur in the direction nf ,
thus avoiding the need to accurately model the surface filtered gas velocity. The
surface filtered resolved strain-rate σs is obtained similarly except that the effects
of flow field strain and propagation strain are separated in order that the influence
of flame wrinkling may be modelled appropriately; in the limit of very high wrin-
kling (assuming isotropy) the compressive and extensive effect of the flow field are
assumed to cancel, whereas the effects of propagation strain approaches half of that
for wrinkling aligned with nf . The result is

∇.
U − nf .(∇ 
U).nf ( + 1)[∇.(Su nf ) − nf .(∇(Su nf )).nf ]
σs = + . (58)
 2
The remaining requirement for closure is to provide a formulation for the lam-
inar flame speed Su . This will depend heavilly on the physics of the case under
consideration. For some combustion problems it is reasonable to assume the lam-
inar flame speed is unaffected by strain and curvature, and thus set Su = Su0 , the
unstrained flame speed. However in most cases strain effects cannot be ignored
[26]. For example, in the case of a flame trapped in the shear layer behind a
backward-facing step [25], the strain effects near the step are important in re-
ducing the effective reaction rate and preserving the Kelvin–Helmholtz instability.
Certainly in the case under consideration both strain and curvature effects will be
important [5]. One possible approach is to assume that the laminar flame speed is in
local equilibrium with the local resolved strain rate σs (spectral modelling suggests
that the effects of subgrid strain and curvature largely cancel), and thus the laminar
flame speed can be set to the value attained when in equilibrium Su = Su∞ . Unfor-
tunately, the chemical time scales of lean flames, such as lean propane flames, may
be comparable to the strain and transport time scales, in which case equilibrium
may not be assumed and a full transport equation is required. By analogy with
LES OF PREMIXED TURBULENT COMBUSTION 19

the transport of the flame wrinkling, the filtered laminar flame speed is expected
  
to be transported at the surface filtered velocity of the flame UI . Thus a transport
equation of the form
∂Su  (S 0 − Su )
+ U s .∇Su = −σs Su + σs Su∞ 0u (59)
∂t (Su − Su∞ )
has been implemented. A detailed validation of this model is not possible without
appropriate data on the effect of strain on laminar flames, however the above model
does seem to work well in practice.

4.2. RESULTS
Figure 3 shows the state of the b = 0.5 isosurface at t = 0.0054 s and t = 0.009 s
after ignition. For direct experimental comparisons the exact definition of the ‘edge
of the flame’ will depend heavilly on the experimental technique used to analyse
the flame front. For example, it is possible to realistically simulate the effect of
Schlieren photography by appropriate visualisation of the calculated data [32].
However our purpose here is to be able to analyse the model, not make detailed
comparisons with experiment. The isosurface is coloured by the values of SGS
wrinkling . This enables us to examine the interaction between the grid scale
wrinkling, as evidenced by the shape of the isosurface of  b, and the subgrid scale
wrinkling which is sampled by it.
Figure 4a shows a section through the computational domain at 0.009 s, show-
ing contours of b ∈ [0, 1]. This should be compared with Figure 2b. The computa-
tional mesh is shown overlaid for reference. It will be noticed that the filtered flame
width is slightly broader, covering 3–4 cells rather than the 2–3 of the earlier result.
This is not an enormous difference in breadth, and is attributable to numerical dif-
fusion on a fairly coarse mesh. Convective terms in the b and  b equations have been
treated using the gamma differencing scheme [17]: it may be that the error could
be reduced somewhat using numerical methods designed specifically for interface
tracking [18]. Although conventionally the filter width for LES is linked to the cell
dimension this is not necessary, and the numerical broadening of the flame front
could be interpreted as implying that the implicit filter for the indicator function is
slightly wider than the grid spacing, possibly as a result of backscattering between
the flame and the turbulence. However the extensive investigation of the impact of
the numerics on the SGS modelling is beyond the scope of this paper. Interestingly
the 
b isosurface is slightly narrower, although there is no reason to suspect this is
anything other than a coincidence.
The modelling (e.g. Equation (57)) suggests that there should be a relationship
between  and the strain σs which could be extracted if the data is appropriately
analysed. σs is the flame surface strain field resulting from the surface filtering
process, and  is also a flame surface property (although defined at all points
in the flow). Figure 6a. plots values of  against D = 1/2∇U + ∇U T  for
20 G. TABOR AND H.G. WELLER

(a)

(b)

Figure 3. (a) Flame front as defined by  b = 0.5, 0.009 s after ignition, (b) flame front
0.017 s after ignition. The isosurfaces are shaded using values of the wrinkling  ranging
from  = 1.0 (black) to  = 1.4 (white)
LES OF PREMIXED TURBULENT COMBUSTION 21

(a)

(b)

Figure 4. Flame front 0.009 s after ignition: (a) shows contours of b on a 2-d slice through
the flame front, with the computational mesh superimposed; (b) shows contours of b.
22 G. TABOR AND H.G. WELLER

Figure 5. Distributions of  (left column) and Su (right column) throughout the computa-
tional domain at t = 0.005, 0.01, 0.015 s after ignition.

locations on the flame surface, at times t = 0.005, 0.01, 0.015 s after ignition. A
regression analysis demonstrates that there is a correlation between the two vari-
ables with correlation coefficients around 0.56 for all three times. Extracting this
data proved interesting in its own right. Initially the quantity m = |∇
b| was used
as an indication of the location of the flame: values of this quantity significantly
different from zero (taken as values above the mean m) indicate a slope in  b, in
other words indicate the whole thickness of the flame front. Values of σ and 
from this region were then plotted. However this way only a very weak correlation
LES OF PREMIXED TURBULENT COMBUSTION 23

was found. However the location of the flame can be made much more precise,
which was done by identifying cells where 0.49 <  b < 0.51, and extracting the
data from these. This smaller sample is the one shown here, and produces a much
stronger correlation. A similar effect is noticeable in the correlation between Su and
σ : for the mid-flame data shown in Figure 6b there is a good correlation (coefficient
0.62), but with the dataset expanded to include the probable numerical tail of the
distribution the correlation is much weaker. Thus the modelling of the flame centre
(values  b ∼ 0.5) is physically accurate, but the flame leading and trailing edges are
problematic.
Figure 6c examines the correlation between the SGS and GS flame curvatures.
 represents the SGS curvature, whilst  b represents the GS geometry from which
it should be possible to extract GS curvature information. Since ∇ b gives the GS
surface normal to the interface, a valid measure of its curvature will be ∇.∇ b, so
Figure 6c plots  vs. ∇ 2b for the flame centre, as determined above. The correlation
is rather weak, as can be seen from the diagram. However the correlation between
the SGS turbulent kinetic energy of the fluid, k and , plotted in Figure 6d is much
stronger (correlation coefficient of 0.7). This suggests that  is coupling strongly
to the SGS fluid flow and only weakly to the GS flow.
Figure 7 shows the evolution of various quantities relating to the combustion
model as functions of time. Figure 7a compares the radius against time for several
realisations of the same case with data from the Leeds bomb [4, 5]. A 323 box of
turbulence was used for this calculation, whilst another comparison for this case
using a finer mesh (642 ) has been published elsewhere [23]. To suppliment this we
present profiles of other important parameters. Figure 7b shows the variation of
flame speed Su and surface wrinkling  with time. Both parameters are defined at
all points in the mesh, so the spatial average over the whole domain is presented.
However  in particular has little meaning away from the flame surface. Hence
average values on the flame surface itself are also presented. This is achieved by
filtering these fields using the filter m = ∇ b iff m > m , m = 0 elsewhere (as
discussed above). There is very little difference between the surface value and bulk
value of Su , as one might expect. However there is a significant difference between
the value of  in the vicinity of the flame surface and that elsewhere. Figures 7c and
7d demonstrate the variation of other bulk quantities, i.e. the SGS turbulent kinetic
energy k and the enstrophy ζ = (1/2)|∇×U |, which is a measure of the vortical
nature of the flow. These are split into components averaged over the burnt and
over the unburnt fraction of the gas, together with the total, thus indicating how
the turbulence is being modified by the passage of the flame front. Interestingly,
the total turbulent kinetic energy increases with time, but the enstropy peaks and
then decreases. It seems likely that at about t = 10 ms the symmetry boundary
conditions begin to affect the propagation of the flame, causing the altered trends
generally after t = 10 ms.
24 G. TABOR AND H.G. WELLER

(a)

(b)

(c)

(d)
Figure 6. Correlations between computed properties on the flame surface. (a) SGS wrinkling
 vs. GS strain 12 ∇U + ∇U t . (b) Flame speed Su vs. (1/2)∇U + ∇U t . (c)  vs. GS
curvature ∇ 2
b. (d)  vs. SGS turbulent kinetic energy k.
LES OF PREMIXED TURBULENT COMBUSTION 25

(a)

(b)

(c)

(d)
Figure 7. Variation with time of various quantities in the box: (a) tracks the progress of the
flame surface, comparing Schlieren data from the Leeds bomb with simulated Schlieren radii
evaluated from the computation; (b) shows variations of Su and flame wrinkling parameter 
throughout the box and at the flame surface; (c) shows the variation of the SGS turbulent ki-
netic energy k inside and outside the flame surface, together with the total kinetic energy within
the box; and finally (d) shows the same information for the enstrophy ζ = (1/2)|∇×U |.
26 G. TABOR AND H.G. WELLER

5. Conclusions
This paper presents the mathematical and physical framework of a family of LES
combustion models for premixed and partially premixed combustion in the flamelet
regime. LES holds the prospect of significant improvements over traditional RANS
turbulence modelling, because of its greater accuracy and ability to generate more
information to feed into the combustion modelling. In particular, such models
should be able to better treat countergradient transport, and simulate unsteady
fluctuations around a mean flow. In order to achieve our aim, we have to extend
the concept of conditional averaging to make it appropriate for LES. In RANS,
conditional averaging is a technique allowing the decomposition of any fluid flow
equations into equations appropriate for individual components of the flow. In do-
ing so it provides a formal, rigorous framework for deriving the precise forms of
the surface terms in the conditionally averaged equations. The extension of this to
LES is relatively straightforward, and is in some ways more natural, since the LES
formulation already incorporates the concept of a volume average. By introducing
conditional and surface filtering operations a similar framework can be built up,
with the effect of the interface surface on the bulk flow being dictated by the filter
support. Although the aim of this paper is towards combustion modelling, these
mathematical techniques should prove useful in modelling other multicomponent
flows. Various issues relating to the filtering of an appropriate indicator function
are discussed. In particular it is shown by the simple procedure of filtering a 2-d
step function that the filtered indicator function is spread over 2–3 cells. Thus the
problem of performing LES of the flame front in the high Damköhler regime is
circumvented by simulating the filtered indicator function (representing large scale
geometric information) and modelling the small scale geometric information.
Within this framework filtered versions of the NSE are derived. The resulting
additional interface terms need to be modelled, and from this a family of LES
combustion models of varying complexity and accuracy can be derived. Appro-
priate models based on the flame front wrinkling have been introduced and tested
in previous papers. The modelling is analogous to that used for the RANS variant
of these models, with only the interpretations of the variables being different. If
anything the detailed modelling is more appropriate for the LES case, with the
scale division of the flame front geometry described above. The ability of such
models to simulate complex flows has already been demonstrated [37] for com-
bustion behind a backward facing step. This test case is strongly time-dependent,
and would be a real problem for a RANS-based combustion model. The case used
in this paper is simpler, and the Weller models can be shown to work well for
this case as well, as comparison with experimental cases has shown [23]. Since
the bias of this paper is towards the theoretical aspects of the model, the spherical
flame case is used merely for illustrative purposes, readers being directed to the
earlier paper for details of the comparison. It is seen that the filtered indicator
LES OF PREMIXED TURBULENT COMBUSTION 27

function as calculated is in close qualitative agreement with that generated by


direct filtering of the indicator function. Taken as a whole therefore, the Weller
LES combustion models are both accurate and well grounded in theory.

References
1. Angelberger, C., Veynante, D., Egolfopoulos, F. and Poinsot, T., Large eddy simulation of
combustion instabilities in turbulent premixed flames. In: Proceedings of the Summer Program
(1998) pp. 61–82.
2. Boger, M., Veynante, D., Boughanern, H. and Trouvé, A., Direct numerical simulation analysis
of flame surface density concept for large eddy simulation of turbulent premixed combus-
tion. In: Twenty-Seventh Symposium (International) on Combustion. The Combustion Institute,
Pittsburgh, PA (1988) pp. 917–925.
3. Bourlioux, A., Semi-analytical validation of a dynamic large-eddy simulation procedure for
turbulent premixed flames via the G-equation. Combust. Theory Modelling 4 (2000) 363–389.
4. Bradley, D., Hicks, R.A., Lawes, M. and Sheppard, C.G.W., Final Report, CEC Contract JOU2-
CT-92-0162. Technical report, Combustion Research Group, University of Leeds (1996).
5. Bradley, D., Hicks, R.A., Lawes, M., Sheppard, C.G.W. and Woolley, R., The measurement of
laminar burning velocities and Markstein numbers for iso-octane-air and iso-octane-n-heptane-
air mixtures at elevated temperatures and pressures in an explosion bomb. Combust. Flame 115
(1998) 126–144.
6. Bray, K.N.C., The challenge of turbulent combustion’. In: Twenty-Sixth Symposium (Interna-
tional) on Combustion. The Combustion Institute, Pittsburgh, PA (1996) pp. 1–26.
7. Butler, T.D. and O’Rourke, P.J., A numerical method for two-dimensional unsteady reacting
flows. In: Sixteenth Symposium (International) on Combustion. The Combustion Institute,
Pittsburgh, PA (1977) pp. 1503–1515.
8. Candel, S.M. and Poinsot, T.J., Flame stretch and the balance equation for the flame area.
Combust. Sci. Technol. 70 (1990) 1–15.
9. Cant, R., Pope, S. and Bray, K., Modelling of flamelet surface-to-volume ratio in turbu-
lent premixed combustion. In: Twenty-Third Symposium (International) on Combustion. The
Combustion Institute, Pittsburgh, PA (1990) pp. 809–815.
10. Colin, O., Ducros, F., Veynante, D. and Poinsot, T., A thickened flame model for large eddy
simulations of turbulent premixed combustion. Phys. Fluids 12(7) (2000) 1843–1863.
11. Dopazo, C., On conditional averages for intermittent turbulent flows. J. Fluid Mech. 81 (1977)
433–438.
12. Fureby, C. and Tabor, G., 1997, Mathematical and physical constraints on large eddy
simulations’. Theor. Comput. Fluid Dynam. 9(2) (1997) 75–83.
13. Fureby, C., Tabor, G., Weller, H. and Gosman, A.D., A comparative study of sub grid scale
models in homogeneous isotropic turbulence. Phys. Fluids 9(5) (1997) 1416–1429.
14. Gülder, Ö.L., Turbulent premixed flame propagation models for different combustion regimes.
In: Twenty-Third Symposium (International) on Combustion. The Combustion Institute, Pitts-
burgh, PA (1990) pp. 743–750.
15. Hawkes, E. and Cant, S., A flame surface density approach to large eddy simulation of premixed
turbulent combustion. Proc. Combust. Inst. 28 (2000) 51–58.
16. Im, H.G., Lund, T.S. and Ferziger, J.H., Large Eddy Simulation of turbulent front propagation
with dynamic subgid bodels. Phys. Fluids 9(12) (1997) 3826–3833.
17. Jasak, H., Error analysis and estimation for the finite volume method with applications to fluid
flows. Ph.D. Thesis, Imperial College (1996).
18. Jasak, H. and Weller, H.G., Interface tracking capabilities of the inter-gamma differencing
scheme. Technical report, Imperial College of Science, Technology and Medicine (1995).
28 G. TABOR AND H.G. WELLER

19. Kerstein, A.R., Ashurst, W.T. and Williams, F.A., Field equations for interface propagation in
an unsteady homogeneous flowfield. Phys. Rev. A 37 (1988) 2728–2731.
20. Marble, F.E. and Broadwell, J.E., The coherent flame model of chemical reactions. Technical
Report TRW-9-PU, Project Squib Rep. (1977).
21. Menon, S., Simulation and control of combustion instability in a dump combustor. In: Sixth
International Conference on Numerical Combustion, New Orleans, Louisiana (1996).
22. Menon, S. and Jou, W.H., Large eddy simulations of combustion instability in an axusymmetric
ramjet. Combust. Sci. Technol. 75 (1991) 53.
23. Nwagwe, I.K., Weller, H.G., Tabor, G., Gosman, A.D., Lawes, M., Sheppard, C.G.W. and Woo-
ley, R., Measurements and large eddy simulations of turbulent premixed flame kernel growth.
Proc. Combust. Inst. 28 (2000) 59–66.
24. Peters, N., Wenzel, H. and F. A. Williams, F.A., 2000, ‘Modification of the turbulent burning
velocity by gas expansion. Proc. Combust. Inst. 28 (2000) 235–243.
25. Pitz, R.W. and Daily, J.W., Combustion in a turbulent mixing layer formed at a rearward-facing
step. AIAA J. 21 (1983) 1565–1570.
26. Poinsot, T.J., Veynante, D. and Candel, S., Quenching processes and premixed turbulent
combustion diagrams. J. Fluid Mech. 228 (1991) 561–606.
27. Pope, S.B., The evolution of surfaces in turbulence. Internat. J. Engrg. Sci. 26(5) (1988) 445–
469.
28. Schroeder, W., Martin, K. and Lorensen, W., The Visualisation Toolkit: An Object-Oriented
Approach to 3d Graphics. Prentice Hall, Englewood Cliffs, NJ (1997).
29. Shewchuk, J.R., http://www.cs.cmu.edu/ quake/triangle.html.
30. Shewchuk, J.R., Triangle: Engineering a 2D quality mesh generator and Delaunay triangulator.
In: First Workshop on Applied Computational Geometry, Philadelphia, PA (1996) pp. 124–133.
31. Spalding, D.B., ‘Mixing and chemical reaction in steady confined turbulent flames. In: Thir-
teenth Symposium (International) on Combustion. The Combustion Institute, Pittsburgh, PA
(1970) pp. 649–657.
32. Tabor, G., Nwagwe, I.K., Weller, H. and Gosman, A.D., Visualisation of results from LES of
combustion. J. Vis. 2(2) (1999) 177–184.
33. Vervisch, L., Bidaux, E., Bray, K. and Kollmann, W., Surface density function in premixed
turbulent combustion modelling, similarities between probability density function and flame
surface approach. Phys. Fluids 7(10) (1995) 2496–2503.
34. Weller, H., The development of a new flame area combustion model using conditional aver-
aging. Thermo-Fluids Section Report TF 9307, Imperial College of Science, Technology and
Medicine (1993).
35. Weller, H., Uslu, S., Gosman, A., Maly, R., Herweg, R. and Heel, B., Prediction of combustion
in homogeneous-charge spark-ignition Engines. In: International Symposium COMODIA 94
(1994) pp. 163–169.
36. Weller, H.G., Marooney, C.J. and Gosman, A.D., A new spectral method for calculation of the
time-varying area of a laminar flame in homogeneous turbulence. In: Twenty-Third Symposium
(International) on Combustion. The Combustion Institute, Pittsburgh, PA (1990) pp. 629–636.
37. Weller, H.G., Tabor, G., Gosman, A.D. and Fureby, C., Application of a flame-wrinkling LES
combustion model to a turbulent mixing layer. In: Twenty-Seventh Symposium (International)
on Combustion. The Combustion Institute, Pittsburgh, PA (1998) pp. 899–907.

Vous aimerez peut-être aussi