This action might not be possible to undo. Are you sure you want to continue?

BooksAudiobooksComicsSheet Music### Categories

### Categories

### Categories

Editors' Picks Books

Hand-picked favorites from

our editors

our editors

Editors' Picks Audiobooks

Hand-picked favorites from

our editors

our editors

Editors' Picks Comics

Hand-picked favorites from

our editors

our editors

Editors' Picks Sheet Music

Hand-picked favorites from

our editors

our editors

Top Books

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Audiobooks

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Comics

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Sheet Music

What's trending, bestsellers,

award-winners & more

award-winners & more

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

**www.elsevier.com/locate/combustﬂame
**

A dynamic model for the turbulent burning velocity for

large eddy simulation of premixed combustion

E. Knudsen

∗

, H. Pitsch

Department of Mechanical Engineering, Stanford University, Stanford, CA 94305, USA

Received 2 August 2007; received in revised form 17 April 2008; accepted 4 May 2008

Available online 30 July 2008

Abstract

Turbulent premixed combustion is particularly difﬁcult to describe using large eddy simulation (LES). In LES,

premixed ﬂame structures typically exist on subﬁlter length scales. Consequently, premixed LES models must be

capable of describing how completely unresolved ﬂame structures propagate under the inﬂuence of completely

unresolved eddies. This description is usually accomplished through the implementation of a model for the tur-

bulent burning velocity. Here, a dynamic model for describing the turbulent burning velocity in the context of

LES is presented. This model uses a new surface ﬁltering procedure that is consistent with standard LES ﬁltering.

Additionally, it only uses information that comes directly from the ﬂame front. This latter attribute is important

for two reasons. First, it guarantees that the model can be consistently applied when level set methods, where

arbitrary constraints can be imposed on ﬁeld variables away from fronts, are used to track the ﬂame. Second, it

forces the model to recognize that the physics governing ﬂame front propagation are only valid locally at the front.

Results showing model validation in the context of direct numerical simulation (DNS), and model application in

the context of LES, are presented.

© 2008 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

Keywords: Premixed turbulent combustion; Burning velocity; Large eddy simulation; Level set methods; Dynamic model

1. Introduction

Turbulent premixed ﬂames are particularly difﬁ-

cult to describe in the context of large eddy simulation

(LES). Most industrially relevant premixed ﬂames ex-

ist in either the corrugated ﬂamelets regime or the thin

reactions zones regime [1]. The width of the inner

reaction zone of a ﬂame in these regimes is compa-

rable to, if not smaller than, the Kolmogorov length

scale that describes the size of the smallest turbu-

lent eddies in the ﬂow. Flame preheat zones, which

*

Corresponding author.

E-mail address: ewk@stanford.edu (E. Knudsen).

are typically much broader than reaction zones, may

also, in the corrugated ﬂamelets regime, exist on sub-

Kolmogorov length scales. In LES, by deﬁnition, the

smallest length scales of a ﬂow are ﬁltered out. As a

result, in industrially relevant regimes the transitions

that occur between unburned and burned states occur

mostly on subﬁlter scales.

Premixed combustion models for implicitly ﬁl-

tered LES that use standalone progress variable or

ﬁnite rate chemistry approaches will thus, it seems,

always fail. All models are limited by the accuracy of

the schemes they use for evaluating gradients, and no

scheme is capable of resolving the sharp subgrid tran-

sitions that occur in premixed implicit LES near ﬂame

0010-2180/$ – see front matter © 2008 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

doi:10.1016/j.combustﬂame.2008.05.024

E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760 741

fronts. Premixed implicit LES models that attempt to

resolve ﬂame structure are therefore especially prone

to numerical errors in the most critical regions of the

ﬂowﬁeld.

A variety of methods have been suggested in the

literature in response to the problem of subgrid transi-

tion. Two of the most widely discussed are level set, or

G-equation, methods, and artiﬁcial ﬂame thickening

methods. The dynamically thickened ﬂame model,

for example, uses ﬁnite rate chemistry, but addition-

ally broadens local reaction zones so that they can

be resolved on LES meshes [2]. This broadening is

achieved by increasing molecular diffusivities and,

in a proportionate manner so as to keep the laminar

ﬂame speed constant, spreading out the inﬂuence of

reaction source terms. Thickened ﬂame models there-

fore eliminate the problem of poor resolution. The

thickening procedure, however, has an important con-

sequence. The widened ﬂame attenuates local turbu-

lence and prevents eddies smaller than the thickening

length scale from inﬂuencing the front. This effec-

tively decreases the velocity at which the front propa-

gates and creates the need for a compensating model.

The so-called “efﬁciency function” that is used acts to

ensure that the ﬂame will propagate at appropriately

large speeds in the presence of turbulence [3]. This ef-

ﬁciency function may in one sense be viewed as the

empirical introduction of a model describing the tur-

bulent burning velocity.

In level set methods, ﬂame fronts are character-

ized using isocontours of ﬁeld variables and explicitly

tracked [1,4–7]. At the relevant isocontours, the ﬁeld

variables are governed by equations describing how

the fronts propagate. Away from the relevant isocon-

tours, smooth gradients are prescribed for the ﬁeld

variables to ensure numerical resolution of front dy-

namics. In level set methods, the inner reaction zones

of premixed ﬂames are treated as coherent structures.

The effect of the chemical activity that occurs within

these reaction zones appears in the governing front

equation almost entirely as a front propagation speed.

This speed is approximately equivalent to the lami-

nar burning velocity in the fully resolved case. Due to

the coherent treatment of inner reaction zones, level

set approaches suffer from the drawback of not be-

ing able to inherently consider local ﬂame quenching.

Conversely, they offer the advantage of not inducing

any artiﬁcial interactions between the heat released

by the ﬂame and the ﬂow ﬁeld. Both the laminar un-

stretched burning speed and its dependence on the

local ﬂame stretch rate appear as external parameters

in level set methods. These parameters are typically

well described by both experiments and computa-

tional chemical kinetics studies, and can be used with

conﬁdence in simulations. When level sets are used in

the context of LES, however, ﬁltered descriptions of

the burning velocity that account for subﬁlter turbu-

lence are additionally needed.

In non-premixed combustion, the great advantage

that LES offers is that the scalar mixing process is

reasonably resolved [8,9]. In premixed combustion,

the scalar mixing process is also important, but it re-

mains poorly resolved at the ﬂame front in most LES

situations. The turbulent burning velocity, which is

needed in both of the premixed modeling approaches

discussed here, is a quantity that expresses how this

unresolved mixing process interacts with chemistry.

Traditional burning velocity models rely on a series of

coefﬁcients that have been determined through anal-

yses of both experimental and direct numerical simu-

lation (DNS) data [6,10]. These coefﬁcient-based ap-

proaches have been successfully applied in Reynolds

averaged Navier–Stokes (RANS) simulations, where

level set methods offer an alternative to the problemof

reaction rate closure [6,11]. In LES, however, where

instantaneous ﬂame realizations are available, it is

possible to eliminate the use of constant coefﬁcients

by employing dynamic procedures to determine coef-

ﬁcients automatically.

The general dynamic procedure has been em-

ployed by a variety of researchers in their efforts

to deal with closure problems in reacting ﬂows. Di-

rect reaction rate closure is a particularly challeng-

ing problem because reaction rates strongly depend

on the exact ﬂow and chemical conditions that are

present on the smallest turbulent scales. Dynamic pro-

cedures have therefore been applied to combustion

models that use indirect techniques to account for re-

action rates. Two such approaches are the previously

discussed thickened ﬂame model, and ﬂamelet-type

models, where it is assumed that small scale ﬂame

structures can be precomputed as a function of turbu-

lent parameters. Charlette et al. [12,13], for example,

use a dynamic approach to determine a parameter

in the “wrinkling factor” that appears in the thick-

ened ﬂame model. Knikker et al. [14] dynamically

determine a parameter for this same wrinkling factor

but apply it in the context of the ﬂame surface den-

sity model. Chakraborty and Cant [15] developed a

method of dynamically determining the surface aver-

aged curvature that appears in the ﬂame surface den-

sity model.

Additionally, dynamic procedures have been ap-

plied in the context of level set modeling. Im et al.

[16] and Bourlioux et al. [17], for example, proposed

a dynamic propagation model that treats level set ﬁeld

variables as scalars. Subﬁlter contributions to ﬂame

propagation speed are determined by evaluating a

burning velocity model at two different ﬁlter levels

and comparing the results to differences in the mag-

nitude of the gradient of the level set ﬁeld variable at

those same two levels. Im et al. [16] claim that this

742 E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760

approach can be physically interpreted as enforcing

ﬂame consumption conservation among ﬁlter levels.

They base this claim on work by Kerstein et al. [4],

in which it is demonstrated that a volume integral of

the magnitude of the gradient of a level set ﬁeld vari-

able is equivalent to a measure of the total front area

within the volume. It is important to note, however,

that in Kerstein et al.’s work a periodic and constant

density ﬂow ﬁeld is assumed, and each isocontour of

the level set ﬁeld variable is treated as an equally valid

representation of the ﬂame front. Under that assump-

tion, volume averaging is equivalent to averaging over

multiple front realizations. For variable density ﬂows,

however, this is not the case.

More recent work has stressed that level set gov-

erning equations are only valid at the ﬁeld variable

isocontour that they describe, and that traditional av-

eraging procedures therefore cannot be consistently

used [5,6,18]. Additionally, in real ﬂames where den-

sity and vorticity vary through the ﬂame front, it is

inappropriate to use ﬂow ﬁeld information from away

from the front to describe the propagation of the front

itself. In the present paper, then, a dynamic burning

velocity model is proposed that only considers infor-

mation directly from the 2-D front of interest. Al-

though this model will be developed by taking cues

from the level set approach, and it is therefore par-

ticularly straightforward to implement in that context,

it is not strictly dependent on the use of a level set

method. In general, it is applicable to other premixed

LES techniques. It is important to emphasize at the

outset that some of the quantities that appear in this

model also appear in slightly altered forms in other

premixed models. For example, the variable describ-

ing ﬂame surface area density that is developed for

this model might also be developed using the wrin-

kling factor and the ﬂame surface density that appear

in the ﬂame surface density (FSD) formalism [13,14].

Unlike in the FSD formalism, however, the ﬂame sur-

face area density that is used in this model is evaluated

using only information that comes from the 2-D ﬂame

surface. This information can be generated from sim-

ple geometric knowledge of the surface, and no equa-

tions describing a transported ﬂame surface density

need to be introduced. In general, the strength of this

new model is that it is consistent with the level set

method, and this consistency allows some of the well

known quantities related to the turbulent burning ve-

locity to be viewed and calculated in interesting ways.

The new model requires the use of a unique ﬁlter-

ing procedure that is developed and presented in Sec-

tion 2. In Section 3, the ﬁltering procedure is applied

to develop the dynamic model. Section 4 presents an

evaluation of the model in the context of DNS. The

model is applied in an LES in Section 5. Brief con-

clusions are offered in Section 6.

2. An equation for the dynamics of premixed

ﬂame fronts

Level set equations can be derived by setting the

substantial derivative of a generic ﬁeld variable equal

to zero at a surface of interest. The resulting expres-

sion describes how the ﬁeld variable isocontour asso-

ciated with that surface evolves. In premixed LES, the

derivation of an equation governing ﬂame front be-

havior can be approached in a different way. A ﬂame

front can generally be deﬁned as an isocontour of a

generic progress variable c. This variable might rep-

resent, for example, a non-dimensionalized tempera-

ture. The equation governing the behavior of such a

variable is

(1)

∂c

∂t

+u

j

∂c

∂x

j

=

1

ρ

∂

∂x

j

_

ρD

∂c

∂x

j

_

+

1

ρ

˙ ω

R

,

where u

j

is the local ﬂow velocity in the jth direc-

tion, ρ is the ﬂuid density, D is the diffusivity of the

variable c, and ˙ ω

R

is a source term that describes the

effects of chemical reactions. To derive an equation

describing the evolution of a particular c-isosurface,

information from Eq. (1) needs to be extracted di-

rectly from that isosurface, here arbitrarily deﬁned as

c =c

0

. This extraction operation can be performed by

multiplying Eq. (1) with a delta function, δ(c −c

0

),

(2)

δ(c −c

0

)

_

∂c

∂t

+u

j

∂c

∂x

j

_

=δ(c −c

0

)

_

1

ρ

∂

∂x

j

_

ρD

∂c

∂x

j

_

+

1

ρ

˙ ω

R

_

.

This delta function does not necessarily need to be an

inﬁnitesimally thin Dirac delta. Rather, here δ will be

deﬁned as a normalized Gaussian of ﬁnite width. As

long as this width is small compared with the length

scale of the inner reaction zone of the ﬂame being

considered, multiplication with δ(c − c

0

) will effec-

tively give a null result everywhere except at the ﬂame

front. This ﬁnite width deﬁnition of δ is convenient

because it eliminates the problem of dealing with the

special mathematical properties of the Dirac delta, yet

it remains easy to write.

Just as Dirac delta functions may equivalently be

written as derivatives of Heaviside functions, Gaus-

sians may equivalently be written as derivatives of

error functions. Since the δ function that appears in

Eq. (2) only depends on c, the chain rule may be used

to rewrite the left-hand side of Eq. (2). Remember-

ing that what here will be referred to as the Heavi-

side function H represents an error function of ﬁnite

width, this procedure gives

E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760 743

(3)

δ(c −c

0

)

_

∂c

∂t

+u

j

∂c

∂x

j

_

=

∂[H(c −c

0

)]

∂t

+u

j

∂[H(c −c

0

)]

∂x

j

.

To move the delta function on the right-hand side of

Eq. (2) into the relevant derivatives, the gradient of the

progress variable must ﬁrst be written in terms of the

front normal direction at c

0

, which will be denoted

n

j

,

(4)

∂c

∂x

j

=

∇c

|∇c|

|∇c| =n

j

|∇c|,

where it is assumed that n

j

= ∇c/|∇c| points toward

the burned gases. Use of the product rule on the dif-

fusive term then gives [19,20]

(5)

∂

∂x

j

_

ρD

∂c

∂x

j

_

=ρD|∇c|

∂n

j

∂x

j

+n

j

∂

∂x

j

_

ρD|∇c|

_

.

Finally, the delta function acts on |∇c| as

(6)

δ(c −c

0

)|∇c| =

¸

¸

δ(c −c

0

)∇c

¸

¸

=

¸

¸

∇

_

H(c −c

0

)

_¸

¸

.

Applying the results of these manipulations to the

right-hand side of Eq. (2) produces

(7)

δ(c −c

0

)

_

1

ρ

∂

∂x

j

_

ρD

∂c

∂x

j

_

+

1

ρ

˙ ω

R

_

=Dκ

¸

¸

∇

_

H(c −c

0

)

_¸

¸

+δ(c −c

0

)

1

ρ

×

_

n

j

∂

∂x

j

_

ρD|∇c|

_

+ ˙ ω

R

_

,

where κ is the divergence of the normal vector, or

the curvature. Operating on Eq. (1) with a δ function,

then, gives

(8)

∂[H(c −c

0

)]

∂t

+u

j

∂[H(c −c

0

)]

∂x

j

=Dκ

¸

¸

∇

_

H(c −c

0

)

_¸

¸

+δ(c −c

0

)

1

ρ

×

_

n

j

∂

∂x

j

_

ρD|∇c|

_

+ ˙ ω

R

_

.

Equation (8) governs the behavior of a Heaviside

function that tracks the ﬂame front.

A new variable will now be introduced for the pur-

poses of notational convenience. The variable G will

be deﬁned as

(9) G(x

j

, t ) =H

_

c(x

j

, t ) −c

0

_

.

The G symbol is intentionally used to denote this vari-

able because of its close relation to the standard level

set variable G. However, while both of these vari-

ables are used to describe a front, they differ in an

important way. The level set variable G can be arbi-

trarily deﬁned away from the front it describes, while

the variable G cannot. Since there is nothing arbitrary

about the deﬁnition of a Heaviside function, G can be

directly substituted into Eq. (8), which with Eq. (6)

produces an equation that is well deﬁned at all loca-

tions in space,

(10)

∂G

∂t

+u

j

∂G

∂x

j

=(Dκ)

c

0

|∇G| +s

L,c

0

|∇G|,

where

(11)

s

L,c

0

=

_

1

|∇c|ρ

_

n

j

∂

∂x

j

_

ρD|∇c|

_

+ ˙ ω

R

__

c=c

0

.

Following Peters [6,20], the quantity s

L,c

0

describes

the propagation velocity of the isosurface c = c

0

,

which is a function of the density, the normal diffu-

sion term, and the reaction source term in the progress

variable equation, as expected. In the absence of cur-

vature and strain effects, s

L,c

0

reduces to the un-

stretched 1-D laminar ﬂame speed s

L

. In realistic

settings, however, these effects exist and might be im-

portant [19,21]. Although they cause s

L,c

0

to deviate

from s

L

, s

L

is the best ﬁrst-order approximation that

can be made [6,20]. While it would be possible to

introduce a more elaborate dependency such as a re-

action source term that depends on ﬂame stretch, this

would not signiﬁcantly affect the model development

that is to follow.

Because there is nothing arbitrary about its def-

inition, G can be volumetrically ﬁltered. In the fol-

lowing, the goal will be to develop an equation that

describes the evolution of a ﬁltered G ﬁeld. This equa-

tion will not be numerically tractable for the same rea-

sons that ﬁltered progress variable equations are not

tractable. It will, however, provide the basis for devel-

oping a consistent turbulent burning velocity model

for LES. This will be possible because the ﬁltered

G equation will indicate exactly how to separate re-

solved and subﬁlter propagation components. This

burning velocity model can then be used in a level set

description of the ﬁltered front, or in any other suit-

able front tracking scheme.

Deﬁning F(r

j

; x

j

) to be some appropriately nor-

malized ﬁlter kernel and then applying it to the G ﬁeld

gives

(12) G(x

j

, t ) =

_

V

F(r

j

; x

j

)G(x

j

−r

j

, t ) dr

j

(13)

=

_

V

F(r

j

; x

j

)H

_

c(x

j

−r

j

, t ) −c

0

_

dr

j

.

744 E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760

This ﬁltering procedure is consistent with LES in the

sense that the same ﬁlter kernel F(r

j

; x

j

) that is used

to ﬁlter the Navier–Stokes equations can be used to

ﬁlter a representation of the ﬂame front. It will be as-

sumed here that the ﬁlter kernel F(r

j

; x

j

) commutes

with spatial derivatives.

This same ﬁlter may be applied to Eq. (10). Doing

so produces

(14)

∂G

∂t

+u

j

∂G

∂x

j

=(Dκ)

T ,c

0

|∇G| +s

T ,c

0

|∇G|,

where

(15) s

T ,c

0

|∇G| =s

L,c

0

|∇G|

and

(16) (Dκ)

T ,c

0

|∇G| =(Dκ)

c

0

|∇G|

have been used. Equations (15) and (16) represent the

introduction of deﬁnitions describing the burning ve-

locity associated with a certain ﬁlter level, s

T ,c

0

, and

the ﬁltered curvature, (Dκ)

T ,c

0

.

Since the convective termin Eq. (14) is not known,

it must also be modeled. This term is interesting be-

cause the velocity component under the ﬁlter appears

next to the gradient of G. This gradient acts as a δ

function and the ﬁltered quantity therefore only con-

tains information about velocities that are conditioned

on the density that coincides with the front location,

ρ

c=c

0

. It therefore must be true that

(17) u

j

∂G

∂x

j

=u

j,c=c

0

∂G

∂x

j

.

The conditioning that appears in this relationship

makes it clear that Favre ﬁltered velocities cannot be

directly used to model this term. Conditionally ﬁl-

tered velocities must instead be employed. Speciﬁ-

cally, the resolved and subﬁlter contributions to the

convective term will here be separated by introducing

the variable Γ

c

0

,

(18) u

j,c=c

0

∂G

∂x

j

=u

j,c=c

0

∂G

∂x

j

+Γ

c

0

.

Physically, Γ

c

0

describes how subﬁlter velocity ﬂuc-

tuations convect the ﬁltered front. Equation (18) can

be treated as the deﬁnition of Γ

c

0

.

The problem with Eq. (18) is that conditionally

ﬁltered velocities are usually not available in realis-

tic simulations. It would therefore be convenient to

transform the conditionally ﬁltered quantity into an

unconditional quantity, and it has in fact been shown

that such a transformation can be derived using a

consistency condition describing mass conservation

through a ﬂame front [5]. Modifying the notation used

in that transformation to match current convention,

this conservation property produces, considering con-

ditioning on the unburned gas for example,

(19)

n

j

u

j,u

=n

j

¯u

j

−

_

ρ

u

−ρ

b

ρ

_

_

(Dκ)

T ,u

+s

T ,u

_

p

b

,

where p

b

is the fraction of gas in the ﬁlter volume

that is burned and where the tilde denotes Favre ﬁl-

tering. The insight gained by this manipulation is that

the move from a conditional to an unconditional ve-

locity produces a term that scales with the burning

velocity. If the c = c

0

surface is chosen to represent

the unburned side of the front, Eq. (19) can be used in

Eq. (18),

(20)

u

j,u

∂G

∂x

j

=¯u

j

∂G

∂x

j

−

_

ρ

u

−ρ

b

ρ

_

_

(Dκ)

T ,u

+s

T ,u

_

p

b

|∇G|

+Γ

u

.

Equation (20) can then be substituted into Eq. (14),

where again c = c

0

will denote unburned condition-

ing,

(21)

∂G

∂t

+¯u

j

∂G

∂x

j

+Γ

u

=

__

ρ

u

−ρ

b

ρ

_

p

b

+1

_

×

_

(Dκ)

T ,u

+s

T ,u

_

|∇G|.

The term in the brackets may be further manipulated,

(22)

_

ρ

u

−ρ

b

ρ

_

p

b

+1

=

ρ

u

p

b

−ρ

b

p

b

+ρ

u

(1 −p

b

) +ρ

b

p

b

ρ

=

ρ

u

ρ

,

and the ﬁnal ﬁltered equation may be written as

(23)

∂G

∂t

+¯u

j

∂G

∂x

j

+Γ

u

=

ρ

u

ρ

_

(Dκ)

T ,u

+s

T ,u

_

|∇G|.

Equation (23) will be used below to derive a dynamic

turbulent burning velocity model. It cannot be used

to track a ﬂame front, however, because G is un-

derresolved on implicit LES meshes. Equation (23)

nonetheless informs the use of ﬂame front tracking

methods. Speciﬁcally, the derivations of Eq. (10) and

Eq. (23) show explicitly how ﬁltering affects level set

equations. For example, a level set description of a ﬁl-

tered front can be developed using an isocontour of G,

(24)

¨

G(x

j

, t ) =G

0

∀ G(x

j

, t ) =G

0

,

E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760 745

(25)

¸

¸

∇

¨

G(x

j

, t )

¸

¸

= 1 ∀ G(x

j

, t ),

where the hat on the G variable denotes that it is as-

sociated with a ﬁltered front. A governing equation

for

¨

G that is valid at the front can then be derived by

taking the material derivative of Eq. (24),

(26)

D

¨

G

Dt

=

∂

¨

G

∂t

+u

¨

G,j

∂

¨

G

∂x

j

= 0,

where u

¨

G,j

is the velocity at which the level set

moves. Since the G

0

and G

0

surfaces must propagate

in exactly the same fashion, the velocity u

¨

G,j

can be

developed immediately from the ﬁltered front equa-

tion, Eq. (23). The only intermediate step involves

rewriting this equation using an expanded version

of |∇G|,

(27)

∂G

∂t

+

_

¯u

j

−

ρ

u

ρ

_

(Dκ)

T ,u

+s

T ,u

_

n

j

_

∂G

∂x

j

+Γ

u

= 0,

where n

j

in this new context is n

j

= ∇G/|∇G|. If the

levelset

¨

G = G

0

is to follow the actual ﬁltered front,

then u

¨

G,j

must explicitly contain all of the terms in

the parentheses of Eq. (27). It must additionally ac-

count for the subﬁlter term Γ

u

.

3. Dynamic propagation model

The existence of an appropriate ﬂame surface ﬁl-

tering procedure, such as the one developed in the pre-

vious section, makes it possible to derive a dynamic

turbulent ﬂame speed identity. This identity will relate

turbulent ﬂame speeds associated with different ﬁlter

levels and provide an equation that can be solved for a

model constant. In some respects, this identity is anal-

ogous to Germano’s identity.

When a test-ﬁlter, which here will be denoted by

the hat operator, is applied to Eq. (23), that equation

becomes

(28)

∂G

∂t

+¯u

j

∂G

∂x

j

+(Γ

u

)

=

ρ

u

ρ

(Dκ)

T ,u

|∇G| +

ρ

u

ρ

s

T ,u

|∇G|.

When a ﬁlter and a test-ﬁlter are applied to Eq. (10)

as a single operator, a different result is obtained,

(29)

∂

¨

G

∂t

+

¨

¯u

j

∂

¨

G

∂x

j

+

¨

Γ

u

=

ρ

u

¨

ρ

_

¨

Dκ

_

¨

T ,u

¸

¸

∇

¨

G

¸

¸

+

ρ

u

¨

ρ

s

¨

T ,u

¸

¸

∇

¨

G

¸

¸

.

Equations (28) and (29) form the basis for the de-

velopment of the dynamic model. The development

process will begin with a discussion of the subﬁlter

convective terms that appear in the equations.

Unlike the subﬁlter Reynolds stress term that ap-

pears in the ﬁltered Navier–Stokes equations, the sub-

ﬁlter term Γ

u

that appears here is relatively insigniﬁ-

cant. Speciﬁcally, Γ

u

describes how subﬁlter velocity

ﬂuctuations affect the ﬁltered front. While these ve-

locities tend to wrinkle the front, they act only along

a 2-D surface within the ﬁlter volume. Some of these

velocity ﬂuctuations tend to move the front location

forward, and some tend to move the front location

backward. When they are ﬁltered, therefore, they will

tend to have no effect on the mean front position. The

unique behavior of this subﬁlter term is directly due

to the gradient of G being non-zero only locally at

the front. This insight further demonstrates the impor-

tance of the present ﬁltering approach, since the direct

application of a ﬁlter to a level set equation might not

produce an equivalent result.

The point can be illustrated by considering the

release of a non-propagating front in a ﬂowﬁeld of

isotropic turbulence. In this situation, subﬁlter scale

velocity ﬂuctuations would certainly exist, and the un-

ﬁltered front would become more and more wrinkled.

The mean front position, however, would remain sta-

tionary. Γ

u

should therefore be unable to contribute

to front propagation.

In general, subﬁlter velocity ﬂuctuations can of

course contribute to the movement of a ﬁltered front.

What is important to recognize is that these contribu-

tions are all made in conjunction with another phys-

ical mechanism. For example, velocity ﬂuctuations

might promote an increase in ﬂame surface area that

in turn leads to an increased consumption of unburned

mixture. Or, they might promote the growth of sub-

ﬁlter ﬂame curvature, which would lead to increased

heat diffusion and an adjusted burning velocity. The

difference between these cases and the case of pas-

sive front release is the presence of self-sustaining

front propagation. But, since Γ

u

is uncorrelated with

a propagation speed, the approximations Γ

u

≈ 0 and

¨

Γ

u

≈ 0 must hold. DNS data showing the validity

of these approximations will be presented in Sec-

tion 4.

The broad idea behind deriving a dynamic model

for s

T ,u

is to produce a front speed that ensures the

mean ﬂame front position is independent of the ﬁlter

being used. This can be achieved by forcing equations

(28) and (29), which were created using different se-

quences of ﬁlter application, to produce equivalent

front positions. In other words, since the test-ﬁlter in

Eq. (28) commutes with derivatives, subtracting these

two equations produces a relation that will hold at all

times if ﬁltering is being properly performed,

746 E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760

(30)

¯u

j

∂G

∂x

j

−

_

ρ

u

ρ

(Dκ)

T ,u

+

ρ

u

ρ

s

T ,u

_

|∇G|

=

¨

¯u

j

∂

¨

G

∂x

j

−

_

ρ

u

¨

ρ

_

¨

Dκ

_

¨

T ,u

+

ρ

u

¨

ρ

s

¨

T ,u

_

¸

¸

∇

¨

G

¸

¸

.

Like the identities used to describe subﬁlter stress

models, Eq. (30) does not hold locally and instan-

taneously. In part, this is because the propagation

speeds s

T ,u

and s

¨

T ,u

are only deﬁned at their respec-

tive fronts, which exist at different spatial locations. It

is also in part due to the need for model constants to

be reasonably bounded, which will not occur if satis-

faction of the equation is forced for all time and space.

Rather, Eq. (30) holds statistically for a given number

of realizations. In dynamic subﬁlter stress models, an

appropriately large number of realizations are ensured

by averaging the terms in the dynamic equation over

some volume of the ﬂow. The same technique will be

used here.

Given curvature and burning velocity models, the

terms in the parentheses of Eq. (30) are computable.

The terms that depend on G, however, are not. Again,

this is because jumps in G occur over single ﬁlter

widths, making the G variable computationally in-

tractable and unavailable. A similar problem occurs

with the gradients of G that appear in the convec-

tive terms. Another means of computing these terms

is therefore needed. An alternative approach can be

developed by starting from the deﬁnition of G,

(31)

¸

¸

∇G(x

j

, t )

¸

¸

=

¸

¸

¸

¸

¸

∇

_

V

F(r

j

; x

j

)

×H

_

c(x

j

−r

j

, t ) −c

0

_

dr

j

¸

¸

¸

¸

¸

=

¸

¸

¸

¸

¸

_

V

Fn

j

δ(c −c

0

)|∇c| dr

j

¸

¸

¸

¸

¸

,

where n

j

= ∇c/|∇c|. In work by Maz’ja [22] and

subsequently Kollmann et al. [23], it has been shown

that the integral in Eq. (31) can be rewritten in terms

of the local ﬂame surface area. Speciﬁcally, Kollmann

et al. present the relation

(32)

_

V

f (x

j

)δ

_

Φ(x

j

) −ψ

_

|∇Φ| dx

j

=

_

S,Φ(x

j

)=ψ

f (x

j

) dA(x

j

),

where Φ is a scalar and ψ is the value of that scalar

describing an isocontour of interest. Wenzel and Pe-

ters [24] use this relation to describe the area of ﬂame

fronts in their level set based study of how premixed

ﬂames propagate. They leverage the fact that their

study permits treating multiple level set isocontours

as equally valid representations of ﬂame fronts, and

go on to integrate this expression over all of these

isocontours. They then recover Kerstein et al.’s [4]

expression showing that the magnitude of the volu-

metrically integrated level set gradient represents an

ensemble averaged ﬂame area. In the present study

where the goal is to extract information from only

one isosurface, a different approach will be taken.

Equation (32) will be applied to only the surface of

interest, and no volumetric averaging will be per-

formed.

Applying Eq. (32) to Eq. (31) produces

(33)

¸

¸

¸

¸

¸

_

V

Fn

j

δ(c −c

0

)|∇c| dV

¸

¸

¸

¸

¸

=

¸

¸

¸

¸

¸

_

S,c=c

0

Fn

j

dA

¸

¸

¸

¸

¸

= |N

j

|.

N

j

deﬁnes the front conditioned ﬁltered normal vec-

tor. It is important to note that the presence of the ﬁlter

kernel F makes N

j

a volume weighted quantity, and

that there is no restriction on its magnitude as there is

in the case of a typical normal vector. While Eq. (33)

provides a useful way of looking at |∇G|, N

j

is no

more computationally tractable than |∇G|. N

j

does,

however, suggest a model for |∇G|.

In the well resolved limit in which a front appears

in a ﬁlter volume as a planar surface, the normal vec-

tor n

j

= ∇c/|∇c| in Eq. (33) can be brought outside

the integral. Because |n

j

| = 1, |N

j

| would then re-

duce to a measure of the front area per ﬁlter volume.

As the ﬁlter width in Eq. (33) grows, the possibility

of ﬁnding subﬁlter front wrinkling also grows. When

this wrinkling occurs, front normals within the ﬁlter

volume will point in misaligned directions. In the lim-

iting case of very large ﬁlter widths, a ﬁlter volume

might encompass a series of opposed fronts. Front

normal vectors would then tend to cancel each other

out when integrated, and |N

j

| would approach zero.

In regimes between the completely resolved and the

completely unresolved limits, the value of |N

j

| re-

ﬂects a balance between a tendency to decrease in

magnitude due to misaligned normals and a tendency

to increase in magnitude due to increased surface area

per volume.

These tendencies suggest that a link exists be-

tween |N

j

| and the ﬂame surface density, Σ = |∇c|,

and that models from the ﬂame surface density (FSD)

formalism might be applicable. The presence of δ(c −

c

0

) and n

j

in Eq. (31), however, differentiate Σ

and |N

j

|. The difference due to the δ function can

E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760 747

be overcome by considering the “ﬁne-grained” ﬂame

surface density, Σ

[14,15], but Σ

**does not account
**

for the normal vector n

j

. The “wrinkling factor” Ξ

must additionally be introduced to account for n

j

.

Apparently, |N

j

| is equivalent to the quantity Σ

/Ξ

[14] in the FSD formalism. This implies that FSD

models could be introduced to close Eq. (31). Knikker

et al. [14], for example, model Σ

/Ξ using Σ and a

fractal expression. While this approach is an option,

it requires the further development of equations and

assumptions regarding Σ. Therefore, in an effort to

prevent the introduction of further equations, an ap-

proach stemming from the level set framework will

be introduced.

The regime limits that have been described can be

satisﬁed by modeling |N

j

| using the area of the ﬁl-

tered ﬂame front that passes through the local ﬁlter

volume, A

Δ

, and the ﬁlter volume itself, V

Δ

,

(34) |∇G| = |N

j

| ≈

A

Δ

V

Δ

.

In the fully resolved limit, the ﬁltered ﬂame area cor-

responds to the actual planar ﬂame area within the

ﬁlter volume, in exact agreement with Eq. (33). In

the poorly resolved limit, A

Δ

will still represent the

area of a planar ﬂame, and thus will be much smaller

than the actual ﬂame area in the ﬁlter volume. This re-

duced area accounts for the effects of the misaligned

normals. In areas where the ﬂame front does not exist,

A

Δ

is zero, and Eq. (33) is still satisﬁed. As discussed

above, most dynamic models need to consider multi-

ple realizations of a phenomenon, and a ﬁlter volume

might therefore need to encompass a multi-cell region

of space. In such a case, the area of the entire ﬁl-

tered ﬂame front that exists within that region would

be used to represent A

Δ

.

With a model prescribed for |∇G|, the convec-

tive terms appearing in Eq. (30) are easily dealt with.

The gradients in these terms must have the same

magnitude as |∇G|. Furthermore, it would be incon-

sistent if they pointed in any direction other than

the normal vector associated with the ﬁltered front,

¯ n

j

= ∇G/|∇G|. Since the previously deﬁned level

set variable

¨

G describes the same front as G, it is

acceptable to write n

j

as the computable quantity

¯ n

j

= ∇

¨

G/|∇

¨

G|. Therefore,

(35) ¯u

j

∂G

∂x

j

≈¯u

j

n

j

A

Δ

V

Δ

,

and a tractable form of the convective terms becomes

available.

These models can now be introduced into a ﬁnal

form of the dynamic equation,

(36)

¯u

j

n

j

A

Δ

V

Δ

−

_

ρ

u

ρ

(Dκ)

T ,u

+

ρ

u

ρ

s

T ,u

_

A

Δ

V

Δ

=

¨

¯u

j

¨

n

j

¨

A

Δ

V

Δ

−

_

ρ

u

¨

ρ

_

¨

Dκ

_

¨

T ,u

+

ρ

u

¨

ρ

s

¨

T ,u

_

¨

A

Δ

V

Δ

,

where Δ

**denotes the test-ﬁlter width. This form of
**

the equation is very general and can account for

all premixed regimes except for the broken reaction

zones regime. Nonetheless, it is interesting to con-

sider what happens to the equation in simpliﬁed lim-

iting cases. For example, if convective effects are ne-

glected and if the limit of the corrugated ﬂamelets

regime, where curvature effects are small, is consid-

ered, the dynamic equation reduces to

(37)

ρ

u

ρ

s

T ,u

A

Δ

V

Δ

=

ρ

u

¨

ρ

s

¨

T ,u

¨

A

Δ

V

Δ

.

A

Δ

is zero at all locations in the test-ﬁlter domain that

are farther away from the ﬂame front than the ﬁlter

width Δ. If a box ﬁlter is used and it is assumed that

the ﬁltered ﬂame is homogeneous along the directions

parallel to its surface, the test-ﬁltering operation that

acts on the left-hand side of Eq. (37) simply changes

V

Δ

to V

Δ

, and A

Δ

to A

Δ

. A

Δ

describes the area of

the ﬁltered ﬂame in the test-ﬁlter volume. This leaves

(38)

ρ

u

ρ

s

T ,u

A

Δ

=

ρ

u

¨

ρ

s

¨

T ,u

¨

A

Δ

.

If the differences between the ﬁltered and test-ﬁltered

density are assumed to be small, the expression can

be further simpliﬁed to

(39)

s

T ,u

s

¨

T ,u

=

¨

A

Δ

A

Δ

.

This expression is in agreement with Damköhler’s hy-

pothesis that the turbulent ﬂame speed scales with the

turbulent ﬂame surface area [25]. This dynamic model

may therefore be viewed as an enforcement of a con-

stant rate of ﬂame mass consumption, independent of

the ﬁlter being used. This condition satisﬁes the stated

goal of a having a propagation model that produces

ﬁlter-independent mean front positions.

To apply this procedure to the determination of an

unknown constant, actual burning velocity and diffu-

sive propagation models must be selected. A model

for (Dκ)

T ,u

that considers interactions between re-

solved curvature and resolved mixing, and between

resolved curvature and subﬁlter mixing, has been pro-

posed [5],

748 E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760

(40) (Dκ)

T ,u

=D

u

κ +D

t,Δ,u

κ,

where D

t,Δ,u

is the turbulent diffusivity conditioned

on the unburned gas density. In this formulation, inter-

actions between subﬁlter curvature and subﬁlter mix-

ing are left to be determined by the burning speed

model, s

T ,u

. For the remainder of this paper, this pro-

posed model will be adopted. A model for describ-

ing the burning speed itself will be taken from Peters

[5,6],

(41)

s

T ,u

−s

L,u

u

Δ

= −γ Da

Δ

+

_

(γ Da

Δ

)

2

+γ α Da

Δ

_

1/2

,

where α and γ are RANS-type model constants, and

where Da

Δ

is the Damköhler number associated with

the ﬁlter width. The model constants have, relative to

Peters’ original model, been renamed and regrouped.

This regrouping permits the appearance of the model

constant α in the limits of both large and small Da

Δ

.

These limits correspond to the corrugated ﬂamelets

and thin reaction zones regimes, respectively,

(42) s

T ,u

=s

L,u

+u

Δ

α

2

,

(43) s

T ,u

=s

L,u

_

1 +(γ α)

1/2

_

D

t,Δ,u

D

u

_

1/2

_

.

In the model validation that follows, Eq. (41) will be

used to model s

T ,u

, and α in Eq. (41) will be treated

as a dynamic coefﬁcient. The dynamic identity will

then take the speciﬁc form,

(44)

¯u

j

n

j

A

Δ

V

Δ

−

ρ

u

ρ

(D

u

κ +D

t,Δ,u

κ)

A

Δ

V

Δ

−

ρ

u

ρ

_

s

L,u

+u

Δ

_

−γ Da

Δ

+

_

(γ Da

Δ

)

2

+γ α Da

Δ

_

1/2

__

A

Δ

V

Δ

=

¨

¯u

j

¨

n

j

¨

A

Δ

V

Δ

−

ρ

u

¨

ρ

(D

u

¨

κ +D

t,Δ

,u

¨

κ )

¨

A

Δ

V

Δ

−

ρ

u

¨

ρ

_

s

L,u

+u

Δ

_

−γ Da

Δ

+

_

(γ Da

Δ

)

2

+γ α Da

Δ

_

1/2

__¨

A

Δ

V

Δ

.

Equation (44) is somewhat more complex than a typ-

ical dynamic equation for a Smagorinsky subﬁlter

viscosity coefﬁcient. This is because the dynamic co-

efﬁcient α in Eq. (44) appears in an expression that

is raised to the 1/2 power. This prevents α from be-

ing explicitly solved for in terms of the other vari-

ables in the equation. In a practical simulation, this

means that α must be locally solved for using an

iterative procedure such as Newton’s method. This

procedure requires information from both the local

computational cell and the surrounding cells that are

used in the test-ﬁltering operation. While more ex-

pensive than a simple algebraic evaluation, the full

operation is nonetheless still computationally cheap

and does not signiﬁcantly affect the cost of a simula-

tion.

Additionally, as discussed earlier in this section,

Eq. (44) is not expected to hold instantaneously for

all ﬂow realizations. The value of α that solves this

equation in a practical LES, for example, will occa-

sionally be negative. Negative values of α are incom-

patible with the burning velocity model that is used

here because they can result in a negative value being

raised to the 1/2 power. To prevent the occurrence

of negative values of α, a certain amount of aver-

aging of the terms in Eq. (44) must be performed.

Just as in the computation of Smagorinsky subﬁlter

viscosity coefﬁcients, averaging can be performed lo-

cally, in a homogeneous ﬂow direction [26], or using

more complex techniques [27]. In the remainder of

this work, both the subﬁlter velocity ﬂuctuation u

Δ

and the ﬁltered ﬂame area per unit volume A

Δ

/V

Δ

will be averaged. The ﬁltered terms on the left-hand

side of Eq. (44) will be averaged before they are test-

ﬁltered for use in the model. The test-ﬁltered terms on

the right-hand side of Eq. (44) will be averaged after

the test-ﬁlter is applied. The ﬁnal form of the dynamic

equation, with these averaging procedures included, is

shown in Eq. (45),

(45)

¯u

j

n

j

_

A

Δ

V

Δ

_

−

ρ

u

ρ

(D

u

κ +D

t,Δ,u

κ)

_

A

Δ

V

Δ

_

−

ρ

u

ρ

_

s

L,u

+u

Δ

_

−γ Da

Δ

+

__

γ Da

Δ

_

2

+γ αDa

Δ

_

1/2

__

_

A

Δ

V

Δ

_

=

¨

¯u

j

¨

n

j

_

¨

A

Δ

V

Δ

_

−

ρ

u

¨

ρ

(D

u

¨

κ +D

t,Δ

,u

¨

κ )

_

¨

A

Δ

V

Δ

_

−

ρ

u

¨

ρ

_

s

L,u

+u

Δ

_

−γ Da

Δ

+

__

γ Da

Δ

_

2

+γ αDa

Δ

_

1/2

__

×

_

¨

A

Δ

V

Δ

_

,

where the · operator denotes the averaging proce-

dure.

E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760 749

Table 1

DNS parameters

Simulation constants Turbulence parameters

Mesh size = 512 ×256 ×256 Re

λ

= 48

Cell width, x = 1.0 ×10

−3

m Re

t

= 149

ν = 1.87 ×10

−5

m

2

/s Integral length scale, l

t

= 32.0 ×10

−3

m

ρ = 1.16 kg/m

3

Eddy turnover time, τ = 0.32 s

Burning velocity, s

L

= 0.05, 0.15 m/s Kolmogorov scale, η = 5.0 ×10

−4

m

Forcing coefﬁcient, A= 1.038 Largest eddy size, l = 32.0 ×10

−3

m

4. Validation in DNS

A DNS of a front propagating in forced isotropic

turbulence is performed to validate this model. The

parameters describing the DNS are shown in Table 1.

Turbulence is forced using the linear scheme of Ros-

ales and Meneveau [28]. The simulation is run using

a constant density ﬂow ﬁeld, so gas expansion is not

considered. A uniform Cartesian mesh is used, but in

the direction of front propagation the domain length is

doubled. This is done to allow the ﬂame brush thick-

ness to increase to a width of multiple integral scales.

The forced Navier–Stokes equations,

∂u

j

∂x

j

= 0,

(46)

∂(ρu

i

)

∂t

+

∂(ρu

j

u

i

)

∂x

j

= −

∂P

∂x

i

+μ

∂

2

u

i

∂x

2

j

+Aρu

i

,

are solved to describe the ﬂow physics. A is the forc-

ing coefﬁcient used by Rosales and Meneveau [28].

These equations are solved in the incompressible

limit, and a Poisson equation for pressure is therefore

used to enforce the conservation of mass equation. Pe-

riodic boundary conditions are applied to the velocity

ﬁeld in all three coordinate directions.

The level set equation

(47)

∂G

∂t

+u

j

∂G

∂x

j

=s

L

|∇G|

∀ G=G

0

, |∇G| = 1,

is solved to describe front evolution. Since no curva-

ture terms appear in this equation, the simulation ef-

fectively describes front propagation in the corrugated

ﬂamelets regime. A reinitialization procedure is per-

formed after every three time steps to force the level

set ﬁeld variable away from the front to conform to a

distance function. Reinitialization is accomplished by

using an iterative marker method to estimate the dis-

tance to the front, and then solving a PDE in pseudo-

time to improve the accuracy of the estimate. The

third-order WENO scheme of Peng et al. [29] is used

for the PDE step.

Statistics involving the front are computed using

information from only one isocontour of the level set

ﬁeld variable (G = G

0

). This isocontour is used to

compute the area of the ﬂame, as needed in the model.

The ﬁrst-order δ function scheme of Smereka [30]

is used for the area computation. Neumann bound-

ary conditions are prescribed for the level set at each

end of the domain in the direction of propagation. The

front, however, is never allowed to come so near these

boundaries that their treatment inﬂuences behavior.

When the front approaches the edge of the domain

along the long axis, a data shift is performed which

moves the front and all the ﬁeld variables about a

domain length in the direction opposite of front prop-

agation. Periodic boundary conditions are prescribed

for the level set in the other two directions.

A parallel, structured code that is fourth-order ac-

curate in space and second-order accurate in time is

used to compute the ﬂow. The code is run using an

explicit solver, and the CFL number is limited to 0.8.

Because the linear forcing scheme used here adds en-

ergy to the ﬂow at all wavenumbers, the turbulent

ﬂowﬁeld is initialized within a 256

3

cube and then

copied to an adjacent cube. This prevents the gener-

ation of wavenumbers smaller than the inverse of the

box size. Fig. 1 shows two instantaneous realizations

of the ﬂowﬁeld and front.

Fig. 2 shows the average turbulent kinetic energy

in the domain as a function of time. Initially, the ki-

netic energy rapidly decays due to dissipation. Af-

ter several eddy turnovers, however, the effects of

the forcing scheme become apparent and the tur-

bulent energy begins to increase. Once the forcing

scheme strikes a balance with the viscous dissipa-

tion, the energy in the domain stabilizes in time. In

general, the trends seen in Fig. 2 agree well with

the results reported by the developers of the forc-

ing scheme [28]. The level set is released after sta-

bilization has occurred, at approximately 27.5 eddy

turnover times.

Fig. 3 shows the energy spectrum of the DNS,

computed at the point in time of level set release.

Even though this DNS only captures a very limited

inertial range, the spectrum does indicate that turbu-

lent physics are present. Additionally, the dissipative

750 E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760

Fig. 1. Snapshots from a DNS of front propagation with

u

/s

L,u

≈ 1.0. The level set is the wrinkled surface, and the

cut plane shows vorticity magnitude. The top image shows

an early time in the simulation, and the bottom image shows

the ﬁeld several eddy turnover times later. Fromthis perspec-

tive, the front propagates from the upper left to the lower

right.

falloff occurs smoothly and monotonically, indicating

that the smallest scales of the ﬂow are resolved.

It was argued in Section 3 that Γ

c

0

does not con-

tribute to resolved front propagation. Kollmann’s rela-

tion, Eq. (32), can be used to write this term in a form

that can be evaluated directly from the DNS data. Af-

ter manipulation, the computable expression for Γ

c

0

is

(48)

Γ

c

0

=

_

S,c=c

0

Fu

j,c=c

0

n

j

dA

−u

j,c=c

0

_

S,c=c

0

Fn

j

dA.

Fig. 4 shows Γ

c

0

as a function of time for various

ﬁlter widths. The values of Γ

c

0

that are shown in

this ﬁgure are evaluated at the location in the com-

putational domain where the instantaneous unﬁltered

front intersects a particular vector. This particular

vector lies along the domain centerline that points

Fig. 2. Average kinetic energy, k =

1

2

u

j

u

j

, in the DNS

as a function of time. The initial turbulent kinetic energy

has largely dissipated by 3 eddy turnover times. The forcing

scheme then increases and eventually stabilizes k. The level

set is initialized and released at 27.5 eddy turnover times.

Fig. 3. Energy spectrum from the DNS. This spectrum was

computed at the point in time of level set initialization and

release. DNS data (◦ ◦ ◦), κ

−5/3

scaling (—).

in the direction of mean front propagation. The lo-

cation of the intersection changes with time as the

front moves through the domain, and the evaluation

therefore tracks the front. This tracking scheme is

not Lagrangian because it does not track a particu-

lar section of the front, but it does show how Γ

c

0

evolves at the front over a long period of time. Occa-

sionally, this tracking scheme suffers at smaller ﬁlter

widths when turbulence convects sections of the front

laterally across the domain centerline, and therefore

ahead of other sections of the front. This data sam-

pling scheme then switches to providing information

about the new front section rather than the section be-

hind it. This causes a small discontinuity in the time

signal of Γ

c

0

, but does not affect the conclusions that

can be drawn from the data.

Fig. 4 demonstrates that Γ

c

0

can grow to val-

ues that are comparable in magnitude to the resolved

front convection term, u · ∇G. The quantity Γ

c

0

de-

scribes seemingly small subﬁlter effects, but it can at

times reach relatively large magnitudes because the

E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760 751

Fig. 4. Time evolution of Γ

c

0

(—) and u · ∇G (- - -), evaluated at a position that follows the unﬁltered front. Note the different

vertical axis scales.

volumetrically ﬁltered velocity that the resolved con-

vection term sees differs from the velocity that the

front itself sees. In spite of the fact that Γ

c

0

may

instantaneously be large and positive or large and neg-

ative, Fig. 4 shows that it always ﬂuctuates around

zero, regardless of the ﬁlter resolution. This tendency

to return to zero supports the prior assumption that

Γ

c

0

≈ 0. When considered instantaneously, this ap-

pears to be a poor assumption, but when considered

in the context of describing multiple realizations of

a subﬁlter phenomenon, it is appropriate. Additional

support for this assumption is provided by the fact

that Γ

c

0

decreases in magnitude as the ﬁlter width in-

creases. Physically, this occurs because a larger ﬂame

area is considered and the positive and negative con-

tributions to Γ

c

0

are more likely to cancel with one

another. This behavior suggests that Γ

c

0

is often in-

signiﬁcant relative to the other terms that inﬂuence

front dynamics. The propagation speed of the front,

for example, does not decrease at larger ﬁlter widths,

and neither would resolved convective terms if bulk

ﬂows existed. Finally, Γ

c

0

seemingly bears a resem-

blance to the subﬁlter stress terms in the momentum

transport equations that are typically modeled using a

turbulent viscosity. The tendency of Γ

c

0

to decrease

as the ﬁlter width increases, however, stands in con-

trast to the behavior of the subﬁlter stress. Modeled

turbulent viscosity increases as the ﬁlter width in-

creases because this term is forced to describe more

and more of the turbulent transport. The opposite

trend is observed in Γ

c

0

, which again supports the

assumption that this term does not require modeling.

The procedure that is used to evaluate Γ

c

0

can

also be used to evaluate the convective terms that ap-

pear in the dynamic model (Eq. (30), for example). In

Fig. 5, these convective terms are plotted as a func-

tion of time. Just as in Fig. 4, the terms are evaluated

at locations where the instantaneous position of the

unﬁltered front intersects the vector lying along the

domain centerline that points in the direction of mean

front propagation. This choice is again made because

the terms are only interesting near the front. Fig. 5

shows the convective terms at the front when a ﬁl-

ter and a test-ﬁlter are applied either sequentially or

concurrently. The terms are plotted using two differ-

ent ﬁlter resolutions, and the ﬁgure also shows the

differences in these terms as a function of ﬁlter resolu-

tion. Fig. 5 demonstrates that the differences between

the concurrently and sequentially ﬁltered convective

terms decrease as the ﬁlter width is increased. This is

the same behavior that is observed with Γ

c

0

. Further-

more, the physical argument that is used to claim that

Γ

c

0

is negligible is equally applicable to the convec-

tive terms. The discrepancies that occur between the

ﬁltered and test-ﬁltered velocities at the front location

may be viewed as instantaneous deviations from the

test-ﬁltered velocity ﬁeld. When considered as an ef-

fect on the front over an entire ﬁlter volume, these

deviations should not be capable of moving the ﬁl-

tered front because they should tend to, in the limit of

a large number of realizations, cancel one another out.

752 E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760

Fig. 5. Time evolution of

u · ∇G and

¨

u · ∇

¨

G, evaluated at a position that follows the unﬁltered front. The ﬁrst two plots show

u · ∇G (—) and

¨

u · ∇

¨

G (- - -) at different ﬁlter resolutions. The third plot shows the difference of these terms at Δ/η = 8 (- - -)

and Δ/η = 32 (—). In each plot, the test-ﬁlter width Δ

is Δ

= 1.5Δ.

These arguments and data suggest that it may be pos-

sible to neglect the resolved convective terms in the

dynamic model.

The DNS can also be used to evaluate the model

for |∇G| that was proposed in Eq. (34). Fig. 6 shows

the time evolution of both the model and |∇G| at a va-

riety of ﬁlter resolutions. These quantities are evalu-

ated at the same front locations that are used in Figs. 4

and 5. While not perfect, the model in Eq. (34) is val-

idated by the DNS data. Fig. 6 shows that |∇G| is

accurately represented by A

Δ

/V

Δ

. The model real-

istically describes the magnitude of the exact term at

all ﬁlter resolutions, and additionally captures the ma-

jority of the ﬂuctuations of the exact term, including

how these ﬂuctuations scale with ﬁlter width. Instan-

taneous model errors certainly appear, but these errors

are not cumulative, and are not biased toward either

overprediction or underprediction.

Since a reasonable level of support is established

for the model describing |∇G|, the dynamic proce-

dure itself can now be considered. Up to this point, it

has been implicitly assumed that a dynamic procedure

is necessary, or at the very least helpful, in predict-

ing how propagating surfaces behave. The validity of

this assumption can be checked using the DNS. In

Fig. 7, some probability distribution functions (PDFs)

describing the turbulent burning velocity are plotted.

These PDFs are generated using the full ﬂow and

level set ﬁelds from a single instant in time. The time

instant that is chosen occurs after the front has be-

Fig. 6. Time evolution of |∇G| (—), and of the model for

|∇G| proposed in Eq. (34), A

Δ

/V

Δ

(- - -). From top to bot-

tom, the ﬁlter resolutions used in the plot are Δ/η = 4, 8, 16,

and 32.

come fully wrinkled by the turbulence. The PDFs in

this ﬁgure are parameterized by ﬁlter width. The tur-

bulent burning velocity associated with a given ﬁlter

width and a given point in space is calculated using

the deﬁnition of s

T ,u

in Eq. (15). Fig. 7 illustrates

that s

T ,u

/s

L,u

is unimodal, but spread out. This sug-

gests that, even at a constant ﬁlter width, a model is

needed to accurately describe s

T ,u

/s

L,u

.

Fig. 8 demonstrates that it is in fact a dynamic

model, rather than a static model, that is needed. In

Fig. 8, the turbulent propagation velocity is shown in

a scatter plot as a function of u

Δ

. Since the front prop-

E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760 753

Fig. 7. PDFs of s

T ,u

/s

L,u

, evaluated using an instantaneous

data ﬁeld from the DNS. In this data set, u

/s

L,u

≈ 2.5 and

the test-ﬁlter resolutions are, from the curve with the high-

est peak to the curve with the lowest, Δ/η = 4, 8, 16, 32,

and 64.

agation in this DNS occurs in the corrugated ﬂamelets

regime, the model for the turbulent propagation speed

in Eq. (41) is dependent only on u

Δ

. The scatter plot,

however, shows that a wide range of turbulent prop-

agation speeds exist for any given u

Δ

. In this ﬁgure,

u

Δ

is computed by determining the amount of kinetic

energy that is removed by the ﬁltering procedure,

(49)

_

u

Δ

_

2

=

2

3

_

k −k

Δ

_

,

where k is the kinetic energy of the unﬁltered veloc-

ity ﬁeld and k

Δ

is the kinetic energy of the ﬁltered

velocity ﬁeld. Note that the averaging procedure dis-

cussed at the end of Section 3 is applied to the ki-

netic energies, and that the averaging occurs over a

ﬁlter width in each direction. Because s

T ,u

/s

L,u

does

not unconditionally and monotonically increase with

u

Δ

, the value of the coefﬁcient α cannot be constant.

This result is veriﬁed by the second plot in Fig. 8,

which shows PDFs of the dynamic coefﬁcient α at

various ﬁlter widths. Interestingly, the mean value of

α approaches zero as the ﬁlter width is decreased.

This implies that the turbulent propagation speed ap-

proaches the laminar propagation speed faster than

u

Δ

approaches zero. This result is consistent with the

idea of the Gibson scale [6], which states that ed-

dies with velocities smaller than s

L,u

cannot affect

the ﬂame front. While the static turbulent burning ve-

locity model in Eq. (41) does not capture this effect,

the dynamic model does.

The turbulent burning velocity itself is shown as

a function of time and ﬁlter width in Fig. 9. s

T ,u

is

evaluated directly from its deﬁnition, Eq. (15), and

at the same time-dependent front locations used in

Figs. 4 and 5. Fig. 9 shows that, as expected, the use

of a larger ﬁlter tends to result in a larger turbulent

propagation speed. This ﬁgure also emphasizes that

the turbulent propagation speed is a strongly varying

Fig. 8. Distributions describing the turbulent burning veloc-

ity when u

/s

L,u

≈ 2.5. Top: s

T ,u

/s

L,u

vs u

Δ

when the

ﬁlter width is Δ/η = 32. Bottom: PDFs of the dynamic

coefﬁcient α in Eq. (36), parameterized by the test-ﬁlter res-

olution. The ﬁlter widths are, from the curve with the high-

est peak to the curve with the lowest, Δ/η = 4, 8, 16, 32,

and 64.

Fig. 9. Time evolution of s

T ,u

/s

L,u

at a particular location

on the propagating front. The ﬁlters used are: Δ/η = 4 (—),

16 (- - -), and 32 (- · · -).

quantity. When a large ﬁlter is used, some sections of

the ﬁltered front might be found propagating at the

laminar burning speed, while others might propagate

at four times this speed.

Dynamic models are designed to capture these

kinds of ﬂuctuations, but it is well known that dy-

namic approaches do not produce desirable results

when used instantaneously and without averaging.

754 E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760

Fig. 10. Time evolution of the unaveraged dynamic model at a particular location on the propagating front. The data in the second

and third columns are calculated both directly from the DNS (—), and from the dynamic model (- - -).

Fig. 10 shows this to be true for the proposed dy-

namic propagation model. In this ﬁgure, the dynamic

model is applied to the DNS data, but averaging is

only used to help describe the turbulent velocity ﬂuc-

tuations, u

Δ

. All other quantities are left unaveraged.

The difference between the concurrently and sequen-

tially ﬁltered areas appearing in the dynamic equa-

tion is shown in the left part of Fig. 10. This differ-

ence should, in an average sense, be positive. Locally,

however, it oscillates between positive and negative

values. Furthermore, one of the ﬁltered areas that

comprises this difference occasionally decreases in

magnitude and becomes relatively small compared to

the other. When this happens the value of the com-

puted dynamic coefﬁcient, α, strays outside reason-

able bounding limits. This behavior is shown in the

middle set of plots in the ﬁgure. While the values of

α that are computed directly from the DNS data all

fall within an expected range, those computed from

the dynamic model do not. In an effort to compensate

for these deviations, α is artiﬁcially bounded between

−1 and 5 in the ﬁgure. In spite of this bounding of the

coefﬁcient, the resulting dynamic turbulent propaga-

tion speed predictions shown on the right in Fig. 10

still do not agree with the exact propagation speeds

calculated from the DNS. Although this problem is,

as discussed above, present in all dynamic models,

the failure that occurs in the model proposed here is

worth discussing. The particular mechanism of fail-

ure in this case is inﬂuenced by the fact, shown in

Fig. 8, that the turbulent burning velocity does not in-

stantaneously correlate with the ﬂuctuating turbulent

velocity magnitude, u

Δ

. It only correlates with this

quantity in a very averaged sense, and burning speed

models that employ this quantity can therefore only

be used dynamically in a similarly averaged fashion.

Finally, in further support of this point, even the val-

ues of α that are calculated from the DNS data are no

longer well bounded when the velocity ﬂuctuations

used in their computation are not averaged.

In Fig. 11, the implementation of the dynamic

model that employs the full averaging procedure sug-

gested in Eq. (45) is tested and compared to the DNS

data. The model is applied in a postprocessing step us-

ing three sets of ﬁlters and test-ﬁlters, and the terms

in the dynamic equation are averaged over one quar-

ter of the front. The plots in the ﬁgure show the mean

front position, the mean front speed, and the dynamic

model coefﬁcient as a function of time. Note that the

“mean front” denotes the completely ﬁltered, or pla-

nar, realization of the front. When a ﬁlter smaller than

the domain width is used, the mean front speed is cal-

culated by multiplying the ﬁltered front speed by the

ratio of the ﬁltered and planar front areas, thus en-

suring a front consumption rate that is independent of

ﬁlter width. In each plot, the dynamic model is com-

pared with the static turbulent burning velocity model

from Eq. (41), the laminar burning velocity, and the

DNS data.

In the ﬁrst combination of ﬁlters shown in Fig. 11,

the ﬁlter width corresponds to the mesh cell width,

Δ/η = 2, and the ﬁltered data is therefore fully

E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760 755

Fig. 11. Modeled time evolution of the DNS front. The propagation is computed using the: DNS data (◦ ◦ ◦), laminar burning

speed s

L,u

( ), static s

T ,u

model ( ), dynamic s

T ,u

model with: Δ/η = 2 and Δ

/η = 512 (—), Δ/η = 3 and Δ

/η = 6

(- - -), Δ/η = 6 and Δ

/η = 12 (- · · -).

resolved. The test-ﬁlter width, Δ

/η = 512, corre-

sponds to the domain size, and the test-ﬁltered data

is fully unresolved. This is admittedly a purely aca-

demic ﬁltering scheme, but when used with this ﬁlter

combination, the dynamic model produces excellent

results, as expected. The model captures the average

magnitude of the propagation velocity, as well as the

instantaneous propagation velocity ﬂuctuations. The

front position is calculated by integrating the mean

propagation speed, and is accurately, although not

perfectly, predicted. The accuracy of these results is

directly attributable to the dynamic model’s access to

the area of the fully resolved front. With these ﬁlters,

the left-hand side of Eq. (36) reproduces the exact

area per ﬁlter volume of the front, and the ﬁltered

propagation speed, s

T ,u

, is simply the laminar burn-

ing velocity. Solving the equation therefore forces the

dynamic coefﬁcient α to guarantee that the mean front

consumes exactly the same volume of space that the

fully resolved front does. While this result does not

prove anything about the model’s capabilities in re-

alistic LES settings, it does suggest that the use of

surface area as a dynamic scaling parameter is sound.

The accuracy of the model suffers mildly when the

second and third combinations of ﬁlters are used. In

the second case, the ﬁlter width is Δ/η = 3, and the

test-ﬁlter width is twice the ﬁlter width, Δ

/η = 6. In

the third case, the ﬁlter width is Δ/η = 6, and the test-

ﬁlter width is again twice the ﬁlter width, Δ

/η = 12.

These ﬁlters correspond to realistic LES ﬁlters and

produce results that, while imperfect, are more accu-

rate than the corresponding static model results. The

ﬁnal front displacement is somewhat overpredicted,

but the dynamic model for the most part reasonably

reproduces the DNS data. It especially improves pre-

dictions, as compared to the static model, during the

ﬁrst few eddy turnover times. In this early devel-

opment period, the ﬂame front exists in a transient

regime and has not yet fully equilibrated with the tur-

bulence. The dynamic model efﬁciently captures this

behavior. The overprediction errors that do occur ap-

pear to depend strongly on how u

Δ

and α interact.

Speciﬁcally, when α is allowed to change in order to

conserve front volume consumption among different

ﬁlter levels, it can drift away from values that pro-

duce ideal propagation speed estimates. For example,

for the second and third ﬁlter combinations applied in

Fig. 11, the dynamic model does ensure that the test-

ﬁltered front consumes just as much volume as the

ﬁltered front. The value of α that guarantees this prop-

erty, however, produces a somewhat underestimated

front speed when it is returned to the burning velocity

model. Unfortunately, the burning velocity model that

is employed here is susceptible to this issue. This be-

havior occurs because, as discussed above, s

T ,u

does

not appear to locally correlate well with u

Δ

. Addi-

tionally, s

T ,u

strongly depends on the product of α

and u

Δ

, especially in the corrugated ﬂamelets regime.

Therefore, while a dynamically induced change in the

value of α might be correctly describing a trend based

756 E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760

on a changing value of u

Δ

, it is possible that the

changing α value is incorrectly describing the magni-

tude of that trend. Put another way, α might become

either too large or too small when modeled veloc-

ity ﬂuctuation estimates either over- or underpredict

the local inﬂuence of turbulence on the propagation

speed.

In general, these DNS results validate the basic

premise of the dynamic model. In regimes where

subﬁlter velocity ﬂuctuations can be accurately esti-

mated, the application of a dynamic procedure can

improve front speed predictions. In certain cases, such

as when a ﬂame’s brush thickness is undergoing a

transient phase, it can dramatically improve them.

While the procedure is thus justiﬁed, the particular

burning velocity model that has been applied presents

some challenges to the dynamic procedure. Speciﬁ-

cally, when Eq. (41) is used in regions where velocity

ﬂuctuations do not correlate well with the propagation

speed, the accuracy of the dynamic model suffers.

Similar issues regarding turbulent velocity ﬂuctu-

ations have been noted by other researchers. In their

formulation, for example, Im et al. [16] observed and

commented on a sensitivity to the methods they used

to model these velocity ﬂuctuations. Charlette et al.

[12] showed that in the limit where s

T ,u

scales en-

tirely with u

Δ

, the dynamic equation, Eq. (36), is

an ill posed problem for determining α. Charlette et

al. subsequently suggested a burning velocity model

based on a power law in an effort to deal with this

problem. While the issue of ill-posedness offers in-

sight into the results presented in this section, it

should not, by itself, rule out a continued considera-

tion of Eq. (41). Primarily, this is because the limiting

case where the problem becomes ill-posed seems to

occur in only a small subset of industrially relevant

ﬂows. If, for example, the burning velocity model in

Eq. (41) is to become a function of only u

Δ

, two con-

ditions must hold. The ﬁrst condition is that the func-

tional dependency on the Damköhler number must

disappear. If it did not, the burning velocity would still

scale with the ﬁlter width. This dependency only dis-

appears for Damköhler numbers above approximately

20 [6]. The second condition is that u

Δ

must be sig-

niﬁcantly larger than s

L,u

. If it is not, then the laminar

burning velocity itself still plays an important role in

turbulent propagation. These two conditions lead to a

constraint on the ratio of the ﬁlter width and the ﬂame

thickness. If they are to be satisﬁed only weakly, so

that Da = 20 and u

Δ

/s

L,u

= 5, then the equation for

the Damköhler number shows that Δ/

f

= 100. In

most industrially relevant LES computations, the ra-

tio of Δ and

f

is signiﬁcantly smaller than 100,

and the ill-posed limit is not approached. The DNS

that is considered here, however, does seem to reach

these conditions when the ﬁlter width is large. This is

because the Damköhler number in the DNS is effec-

tively inﬁnity.

Apparently, at least two issues regarding the dy-

namic model proposed here require further attention.

The ﬁrst is the need for an improved method of de-

scribing propagation in the limit where s

T ,u

∼ u

Δ

.

In this limit, all propagation occurs on the subﬁlter

scale, and the dynamic approach of examining how

information transfers between the resolved and subﬁl-

ter scales becomes irrelevant. The second is the need

for improved methods of describing subﬁlter veloc-

ity ﬂuctuations, and of describing precisely how front

propagation depends on them. Addressing these is-

sues would improve the precision with which a dy-

namic burning velocity model could operate.

5. Application in LES

To demonstrate the dynamic model’s applicability

in LES, it is tested in a simulation of the F3 turbu-

lent premixed Bunsen ﬂame studied experimentally

by Chen et al. [31]. The F3 ﬂame is fed by a stoi-

chiometric (φ = 1.0) methane–air mixture that exits a

jet nozzle at a Reynolds number of ∼23,000. A pilot

ﬂame of equivalent composition is used for stabiliza-

tion. The nozzle diameter of the Bunsen jet is 12 mm,

and the gas exit velocity is 30 m/s. In the context of

the premixed regime diagram, the F3 ﬂame exists in

the thin reaction zone regime.

This ﬂame is experimentally well-characterized,

and therefore, due to the scarcity of experimental tur-

bulent premixed data, has been the subject of a num-

ber of numerical studies [1,32,33]. The simulation

performed here closely follows prior work in which a

level set based premixed combustion model was ap-

plied in the context of LES [1]. More speciﬁcally,

the code used in the present work is a variant of the

code described in that earlier work. The differences

that do exist, such as the present use of Lagrangian

rather than Germano averaging in the subgrid viscos-

ity model [27], and the use of a parallel fast marching

method for level set reinitialization, do not qualita-

tively affect the computational approach.

A cylindrical coordinate mesh consisting of 323×

117 × 64 nodes in the axial, radial, and azimuthal di-

rections, respectively, is used for this LES. This ∼2.4

million node mesh is run in parallel on 32 processors.

The computation of a ﬂow through time takes roughly

35 wall clock hours. The computational domain of the

simulation reaches from a half jet diameter upstream

of the nozzle to 30 jet diameters downstream of the

nozzle, and from the nozzle centerline to a distance

of 6 jet diameters in the radial direction.

The dynamic burning velocity model developed

above is used to describe the propagation speed of the

E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760 757

ﬂame front. At each time step and at each mesh point,

a slightly modiﬁed version of the dynamic equation,

Eq. (44), in which the convective terms have been

neglected is solved. There are two reasons for neglect-

ing the dynamic convective terms in this LES. First,

it was argued in Section 4 that these terms should

not contribute to front propagation when considered

in a statistically averaged sense. Second, in this par-

ticular ﬂame the jet velocity is signiﬁcantly larger

than the turbulent burning velocity. Therefore, in sit-

uations where these terms are not considered on a

sufﬁciently averaged basis, they will not appropriately

cancel out and will, due to their magnitude, inappro-

priately dominate the dynamic equation. All curvature

terms, which are known to be important in the thin re-

action zone regime, are dynamically considered.

In practice, the process of solving the dynamic

equation is initiated by evaluating a Heaviside func-

tion everywhere in the domain using the value of

the local level set ﬁeld variable. The introduction of

this Heaviside function allows the ﬂame front to be

test-ﬁltered in a mathematically consistent way, as de-

scribed earlier in this paper. The test-ﬁltering of the

Heaviside function is performed using the standard

LES test-ﬁlter prescribed in the code. Similarly, the

local value of the kinetic energy is computed every-

where both before and after application of the test-

ﬁlter to the velocity ﬁeld. These kinetic energy esti-

mates will be needed in evaluating the velocity ﬂuc-

tuations required by the model. Next, again using the

scheme of Smereka [30], the areas of the ﬁltered and

test-ﬁltered fronts are computed, each on a per-ﬁlter-

volume basis. Finally, to ensure that a statistically

appropriate number of realizations are used in the

dynamic calculations, all of these quantities are av-

eraged in space. In this LES, averaging is performed

in the homogeneous (azimuthal) ﬂow direction. Since

the burning velocity is only needed in the vicinity of

the ﬂame front, this averaging procedure neglects all

mesh points that are not within a distance of several

cells from the actual ﬂame front.

Estimates of the subﬁlter velocity ﬂuctuations

needed in the burning velocity model are then made.

Velocity ﬂuctuations at the ﬁlter level are estimated

by dividing the subﬁlter turbulent viscosity by the

mesh cell size, u

Δ

= ν

t,Δ

/x. While it might seem

reasonable to then estimate the test-ﬁltered velocity

ﬂuctuations, u

Δ

**, by computing a test-ﬁltered tur-
**

bulent viscosity, this approach does not work well

in conjunction with Peters’ turbulent burning veloc-

ity model. Speciﬁcally, nothing guarantees that test-

ﬁltered turbulent viscosities will produce test-ﬁltered

velocity ﬂuctuations that are larger than their ﬁltered

counterparts. If this implementation of the dynamic

model were to encounter a situation where u

Δ

<u

Δ

,

it would fail. To avoid this situation, the test-ﬁltered

velocity ﬂuctuations are computed by taking the dif-

ference of the computed ﬁltered and test-ﬁltered ki-

netic energies, backing out the associated ﬂuctuation

magnitude, and adding it to the u

Δ

estimate,

(50)

_

u

Δ

_

2

=

_

u

Δ

_

2

+

2

3

_

k

Δ

−k

Δ

_

.

This choice guarantees that the test-ﬁlter scale will

contain more energy than the ﬁlter scale. The dy-

namic coefﬁcient α is ﬁnally solved for using a simple

Newton–Raphson technique.

Fig. 12 shows a single snapshot from the LES. The

local inﬂuence of the dynamic model can be seen in

the coloring of the ﬂame front. This coloring shows

the ratio of the ﬁltered and test-ﬁltered ﬂame areas,

on a per-test-ﬁlter-volume basis. In the ﬂat, planar

regions of the ﬂame, this ratio is nearly one and is in-

dicated by the blue portion of the color spectrum. In

regions of the ﬂame where turbulence has produced a

large amount of wrinking, the ratio increases to 1.15.

The red end of the color spectrum represents these

larger values.

Two subsequent LES runs were performed for the

purposes of comparison. In the ﬁrst of these, the tur-

bulent burning velocity is taken to simply be equiv-

alent to the laminar burning velocity. In the second,

the turbulent burning velocity is computed using the

full model from Eq. (41), but with the value of the

coefﬁcient α left constant. In Fig. 13, instantaneous

data from these simulations is used to plot the turbu-

lent burning velocity for both the dynamic and static

model cases, as a function of the turbulent veloc-

ity ﬂuctuation magnitude. This ﬁgure shows that the

dynamic model signiﬁcantly impacts the computed

ﬂame propagation speed. Speciﬁcally, the dynami-

cally computed ﬂame speeds are approximately 50%

smaller than the statically computed ﬂame speeds

over a wide range of u

Δ

. This suggests either that the

ﬂame front itself is fairly well resolved in this LES,

or that the ﬂame brush thickness does not often equi-

librate with the turbulence in this ﬂame. The former

of these explanations seems to be most plausible.

Fig. 14 shows time averaged statistics from the

LES simulations. The axial velocity, temperature, and

water mass fraction plots each compare favorably,

although not perfectly, with experiments. The most

interesting feature of these statistics is their very

weak dependence on the turbulent propagation speed.

Speciﬁcally, the dynamic propagation model results

and the results computed using simply the laminar

burning velocity virtually overlap. While this behav-

ior is supported by Fig. 13, it would be expected

that the static model results, which predict 50% faster

burning rates, would be signiﬁcantly different. Fig. 14

shows that this is not the case, and that the static

model results differ from the dynamic results by a few

758 E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760

Fig. 12. Instantaneous snapshot from an LES of the F3 ﬂame. The cut plane shows a contour plot of temperature, which ranges

from 300 K (black) to 2200 K (bright yellow). The premixed ﬂame, represented by the

¨

G=G

0

isosurface, is colored according

to the ratio of A

Δ

/V

Δ

and

¨

A

Δ

/V

Δ

.

Fig. 13. Instantaneous scatter plot of the turbulent burning velocity as a function of the local turbulent velocity ﬂuctuation

magnitude. Left: s

T ,u

is computed locally in the LES using a static value of α. Right: s

T ,u

is computed locally in the LES using

a dynamically determined value of α.

percent at most. These small differences can be seen

most clearly at the x/D

o

= 12.5 station. No experi-

mental data is available at this location, but the faster

ﬂame speed predicted by the static model is evident

in the slightly higher temperatures and water mass

fractions, as compared to the dynamic model, that are

produced.

A comparison of the convection and ﬂame propa-

gation time scales in this burner helps to explain these

time averaged results. A convection time scale can be

formed by considering how long it takes a particle

injected at the nozzle to propagate one burner diame-

ter downstream, t

c

= D

o

/U

o

∼ 4 × 10

−4

s. A ﬂame

propagation time scale can be formed by considering

howlong it takes a laminar CH

4

/air premixed ﬂame to

propagate from the burner nozzle to the centerline of

the burner, t

f

= (D

o

/2)/s

L,u

∼ 1.6 × 10

−2

s. Com-

paring these time scales reveals t

f

= 40t

c

, which in

turn implies that the premixed front will be convected

far downstream before ﬂame propagation can close it

off at the centerline. This time scale disparity necessi-

tates the use of a pilot ﬂame in this burner. Addition-

ally, since this ﬂame in reality closes off 12 to 15 di-

ameters downstreamof the nozzle [31], this time scale

analysis supports the idea that turbulence, and not

propagation, plays the dominant role in determining

the ﬂame height. As a result, the ﬂame displays only a

weak dependency on the modeled propagation speed.

These results show that the proposed dynamic

burning velocity model can be effectively used in

large eddy simulations, and furthermore that the dy-

namic model predicts signiﬁcantly different propaga-

tion speeds than the corresponding static model. Due

to the F3 burner’s lack of sensitivity to ﬂame speed,

however, these results do not yet suggest that the dy-

namic model is superior to the static model. Further

testing is needed to demonstrate this in a comparative

sense.

6. Conclusions

In consistent LES procedures, subﬁlter models

should act to ensure that all mean predicted quantities

E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760 759

Fig. 14. Time averaged statistics from LES of the F3 ﬂame. Left to right: axial velocity, temperature, and H

2

O mass fraction

at sequential downstream locations. Experimentally measured values (◦ ◦ ◦) are plotted along with LES results computed using

propagation speeds of s

L,u

(—), a dynamically determined s

T ,u

(- - -), and a statically determined s

T ,u

(- · · -). The s

L,u

and

dynamic s

T ,u

results virtually overlap.

are, within certain ranges, independent of the ﬁlters

being used. In this paper, a dynamic model for cal-

culating the propagation speed of ﬂame fronts was

presented. This model was derived in a way that en-

sures this ﬁlter independence criteria is met. In the

derivation of the model, a new approach to writing

a ﬁltered ﬂame front equation was developed. This

approach was useful because it both worked in con-

junction with standard LES ﬁltering techniques and

because it used information from only a 2-D sur-

face. Using this approach, the terms that describe

subﬁlter inﬂuences on the turbulent burning velocity

were explicitly determined. Furthermore, these terms

were used to derive a dynamic identity for the burn-

ing velocity. When enforced, this identity ensures

that evolving a ﬂame front and then ﬁltering the re-

sult yields the same answer that evolving a ﬁltered

front does. A DNS was performed to validate the

proposed dynamic model. Results showed that the

model predicted the speed of a propagating turbu-

lent front with more accuracy than a static turbulent

burning velocity model. Applying the model in an

LES of a turbulent Bunsen ﬂame produced results

that were in agreement with experimental measure-

ments.

Acknowledgments

Support from the United States Department of De-

fense NDSEG Program, the United States Air Force

Ofﬁce of Scientiﬁc Research (AFOSR), and from

NASA is gratefully acknowledged.

References

[1] H. Pitsch, L. Duchamp De Lageneste, Proc. Combust.

Inst. 29 (2002) 2001–2008.

[2] J.P. Legier, T. Poinsot, D. Veynante, in: Proc. of the

CTR Summer Program, Stanford University, 2000, pp.

157–168.

[3] O. Colin, F. Ducros, D. Veynante, T. Poinsot, Phys. Flu-

ids 12 (7) (2000) 1843–1862.

[4] A.R. Kerstein, W.T. Ashurst, F.A. Williams, Phys. Rev.

A 37 (7) (1988) 2728–2731.

[5] H. Pitsch, Combust. Flame 143 (2005) 587–598.

[6] N. Peters, Turbulent Combustion, Cambridge Univ.

Press, Cambridge, UK, 2000.

[7] F.A. Williams, in: J.D. Buckmaster (Ed.), The Mathe-

matics of Combustion, SIAM, 1985, p. 97.

[8] H. Pitsch, Proc. Combust. Inst. 29 (2002) 1971–1978.

[9] H. Pitsch, H. Steiner, Phys. Fluids 12 (10) (2000) 2541–

2554.

760 E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760

[10] R.G. Abdel-Gayed, D. Bradley, Philos. Trans. R. Soc.

London A 301 (1457) (1981) 1–25.

[11] M. Herrmann, Ph.D. thesis, RWTH Aachen, 2000.

[12] F. Charlette, C. Meneveau, D. Veynante, Combust.

Flame 131 (2002) 159–180.

[13] F. Charlette, C. Meneveau, D. Veynante, Combust.

Flame 131 (2002) 181–197.

[14] R. Knikker, D. Veynante, C. Meneveau, Phys. Flu-

ids 16 (11) (2004) L91–L94.

[15] N. Chakraborty, R.S. Cant, Phys. Fluids 19 (10) (2007)

5101–5122.

[16] H.G. Im, T.S. Lund, J.H. Ferziger, Phys. Fluids 9 (12)

(1997) 3826–3833.

[17] A. Bourlioux, H.G. Im, J.H. Ferziger, in: Proc. of the

CTR Summer Program, Stanford University, 1996, pp.

137–148.

[18] M. Oberlack, H. Wenzel, N. Peters, Combust. Theory

Modelling 5 (2001) 363–383.

[19] T. Echekki, J.H. Chen, Combust. Flame 118 (1999)

308–311.

[20] N. Peters, J. Fluid Mech. 384 (1999) 107–132.

[21] M. Matalon, J. Matkowsky, J. Fluid Mech. 124 (1982)

239–259.

[22] V.G. Maz’ja, Sobolev Spaces, Springer-Verlag, Berlin,

1985.

[23] W. Kollmann, J.H. Chen, Proc. Combust. Inst. 25

(1994) 1091–1098.

[24] H. Wenzel, N. Peters, Combust. Sci. Technol. 158

(2000) 273–297.

[25] G. Damköhler, Z. Elektrochem. 46 (1941) 601–652.

[26] M. Germano, U. Piomelli, P. Moin, W.H. Cabot, Phys.

Fluids A 3 (7) (1991) 1760–1765.

[27] C. Meneveau, T.S. Lund, W.H. Cabot, J. Fluid

Mech. 319 (1996) 353–385.

[28] C. Rosales, C. Meneveau, Phys. Fluids 17 (9) (2005)

1–8.

[29] D. Peng, B. Merriman, S. Osher, H. Zhao, M. Kang, J.

Comput. Phys. 155 (2) (1999) 410–438.

[30] P. Smereka, J. Comput. Phys. 211 (1) (2006) 77–90.

[31] Y.-C. Chen, N. Peters, G.A. Schneemann, N. Wruck,

U. Renz, M.S. Mansour, Combust. Flame 107 (1996)

223–244.

[32] M. Herrmann, Combust. Flame 145 (2006) 357–375.

[33] R.P. Lindstedt, E.M. Vaos, Combust. Flame 145 (2006)

495–511.

E. Knudsen, H. Pitsch / Combustion and Flame 154 (2008) 740–760

741

fronts. Premixed implicit LES models that attempt to resolve ﬂame structure are therefore especially prone to numerical errors in the most critical regions of the ﬂowﬁeld. A variety of methods have been suggested in the literature in response to the problem of subgrid transition. Two of the most widely discussed are level set, or G-equation, methods, and artiﬁcial ﬂame thickening methods. The dynamically thickened ﬂame model, for example, uses ﬁnite rate chemistry, but additionally broadens local reaction zones so that they can be resolved on LES meshes [2]. This broadening is achieved by increasing molecular diffusivities and, in a proportionate manner so as to keep the laminar ﬂame speed constant, spreading out the inﬂuence of reaction source terms. Thickened ﬂame models therefore eliminate the problem of poor resolution. The thickening procedure, however, has an important consequence. The widened ﬂame attenuates local turbulence and prevents eddies smaller than the thickening length scale from inﬂuencing the front. This effectively decreases the velocity at which the front propagates and creates the need for a compensating model. The so-called “efﬁciency function” that is used acts to ensure that the ﬂame will propagate at appropriately large speeds in the presence of turbulence [3]. This efﬁciency function may in one sense be viewed as the empirical introduction of a model describing the turbulent burning velocity. In level set methods, ﬂame fronts are characterized using isocontours of ﬁeld variables and explicitly tracked [1,4–7]. At the relevant isocontours, the ﬁeld variables are governed by equations describing how the fronts propagate. Away from the relevant isocontours, smooth gradients are prescribed for the ﬁeld variables to ensure numerical resolution of front dynamics. In level set methods, the inner reaction zones of premixed ﬂames are treated as coherent structures. The effect of the chemical activity that occurs within these reaction zones appears in the governing front equation almost entirely as a front propagation speed. This speed is approximately equivalent to the laminar burning velocity in the fully resolved case. Due to the coherent treatment of inner reaction zones, level set approaches suffer from the drawback of not being able to inherently consider local ﬂame quenching. Conversely, they offer the advantage of not inducing any artiﬁcial interactions between the heat released by the ﬂame and the ﬂow ﬁeld. Both the laminar unstretched burning speed and its dependence on the local ﬂame stretch rate appear as external parameters in level set methods. These parameters are typically well described by both experiments and computational chemical kinetics studies, and can be used with conﬁdence in simulations. When level sets are used in the context of LES, however, ﬁltered descriptions of

the burning velocity that account for subﬁlter turbulence are additionally needed. In non-premixed combustion, the great advantage that LES offers is that the scalar mixing process is reasonably resolved [8,9]. In premixed combustion, the scalar mixing process is also important, but it remains poorly resolved at the ﬂame front in most LES situations. The turbulent burning velocity, which is needed in both of the premixed modeling approaches discussed here, is a quantity that expresses how this unresolved mixing process interacts with chemistry. Traditional burning velocity models rely on a series of coefﬁcients that have been determined through analyses of both experimental and direct numerical simulation (DNS) data [6,10]. These coefﬁcient-based approaches have been successfully applied in Reynolds averaged Navier–Stokes (RANS) simulations, where level set methods offer an alternative to the problem of reaction rate closure [6,11]. In LES, however, where instantaneous ﬂame realizations are available, it is possible to eliminate the use of constant coefﬁcients by employing dynamic procedures to determine coefﬁcients automatically. The general dynamic procedure has been employed by a variety of researchers in their efforts to deal with closure problems in reacting ﬂows. Direct reaction rate closure is a particularly challenging problem because reaction rates strongly depend on the exact ﬂow and chemical conditions that are present on the smallest turbulent scales. Dynamic procedures have therefore been applied to combustion models that use indirect techniques to account for reaction rates. Two such approaches are the previously discussed thickened ﬂame model, and ﬂamelet-type models, where it is assumed that small scale ﬂame structures can be precomputed as a function of turbulent parameters. Charlette et al. [12,13], for example, use a dynamic approach to determine a parameter in the “wrinkling factor” that appears in the thickened ﬂame model. Knikker et al. [14] dynamically determine a parameter for this same wrinkling factor but apply it in the context of the ﬂame surface density model. Chakraborty and Cant [15] developed a method of dynamically determining the surface averaged curvature that appears in the ﬂame surface density model. Additionally, dynamic procedures have been applied in the context of level set modeling. Im et al. [16] and Bourlioux et al. [17], for example, proposed a dynamic propagation model that treats level set ﬁeld variables as scalars. Subﬁlter contributions to ﬂame propagation speed are determined by evaluating a burning velocity model at two different ﬁlter levels and comparing the results to differences in the magnitude of the gradient of the level set ﬁeld variable at those same two levels. Im et al. [16] claim that this

this is not the case. The resulting expression describes how the ﬁeld variable isocontour associated with that surface evolves. and it is therefore particularly straightforward to implement in that context. Rather. ˙ ρ (1) where uj is the local ﬂow velocity in the j th direction. (1) needs to be extracted directly from that isosurface. This information can be generated from simple geometric knowledge of the surface. Although this model will be developed by taking cues from the level set approach. Brief conclusions are offered in Section 6. It is important to note. ρ is the ﬂuid density. in real ﬂames where density and vorticity vary through the ﬂame front. Unlike in the FSD formalism. (2). Additionally.6. Pitsch / Combustion and Flame 154 (2008) 740–760 approach can be physically interpreted as enforcing ﬂame consumption conservation among ﬁlter levels. however. and ωR is a source term that describes the ˙ effects of chemical reactions. and each isocontour of the level set ﬁeld variable is treated as an equally valid representation of the ﬂame front. the variable describing ﬂame surface area density that is developed for this model might also be developed using the wrinkling factor and the ﬂame surface density that appear in the ﬂame surface density (FSD) formalism [13. H. (1) with a delta function. The model is applied in an LES in Section 5. the chain rule may be used to rewrite the left-hand side of Eq. then. δ(c − c0 ). that in Kerstein et al. however. Remembering that what here will be referred to as the Heaviside function H represents an error function of ﬁnite width. Section 4 presents an evaluation of the model in the context of DNS. however. and no equations describing a transported ﬂame surface density need to be introduced. it is inappropriate to use ﬂow ﬁeld information from away from the front to describe the propagation of the front itself. the strength of this new model is that it is consistent with the level set method. As long as this width is small compared with the length scale of the inner reaction zone of the ﬂame being considered. More recent work has stressed that level set governing equations are only valid at the ﬁeld variable isocontour that they describe.’s work a periodic and constant density ﬂow ﬁeld is assumed. ˙ ρ (2) = δ(c − c0 ) This delta function does not necessarily need to be an inﬁnitesimally thin Dirac delta. The equation governing the behavior of such a variable is ∂c 1 ∂ ∂c ∂c + uj ρD = ∂t ∂xj ρ ∂xj ∂xj + 1 ωR . In Section 3. Gaussians may equivalently be written as derivatives of error functions. A ﬂame front can generally be deﬁned as an isocontour of a generic progress variable c. yet it remains easy to write. it is applicable to other premixed LES techniques. D is the diffusivity of the variable c. (2) only depends on c. For variable density ﬂows. and that traditional averaging procedures therefore cannot be consistently used [5. a non-dimensionalized temperature. this procedure gives . Since the δ function that appears in Eq. the ﬁltering procedure is applied to develop the dynamic model. the derivation of an equation governing ﬂame front behavior can be approached in a different way. For example. 2. Just as Dirac delta functions may equivalently be written as derivatives of Heaviside functions. a dynamic burning velocity model is proposed that only considers information directly from the 2-D front of interest. In premixed LES. This variable might represent. and this consistency allows some of the well known quantities related to the turbulent burning velocity to be viewed and calculated in interesting ways. the ﬂame surface area density that is used in this model is evaluated using only information that comes from the 2-D ﬂame surface. δ(c − c0 ) ∂c ∂c + uj ∂t ∂xj ∂c 1 ∂ ρD ρ ∂xj ∂xj + 1 ωR .14]. In the present paper. Under that assumption. it is not strictly dependent on the use of a level set method. They base this claim on work by Kerstein et al. for example. Knudsen.18]. This ﬁnite width deﬁnition of δ is convenient because it eliminates the problem of dealing with the special mathematical properties of the Dirac delta. information from Eq. [4]. multiplication with δ(c − c0 ) will effectively give a null result everywhere except at the ﬂame front. in which it is demonstrated that a volume integral of the magnitude of the gradient of a level set ﬁeld variable is equivalent to a measure of the total front area within the volume.742 E. volume averaging is equivalent to averaging over multiple front realizations. It is important to emphasize at the outset that some of the quantities that appear in this model also appear in slightly altered forms in other premixed models. In general. This extraction operation can be performed by multiplying Eq. To derive an equation describing the evolution of a particular c-isosurface. here arbitrarily deﬁned as c = c0 . In general. An equation for the dynamics of premixed ﬂame fronts Level set equations can be derived by setting the substantial derivative of a generic ﬁeld variable equal to zero at a surface of interest. here δ will be deﬁned as a normalized Gaussian of ﬁnite width. The new model requires the use of a unique ﬁltering procedure that is developed and presented in Section 2.

however.c0 |∇G|. ∂t ∂xj where sL. In realistic settings. Since there is nothing arbitrary about the deﬁnition of a Heaviside function. ˙ ∂xj 1 ρ (8) Equation (8) governs the behavior of a Heaviside function that tracks the ﬂame front. ∂t ∂xj (3) To move the delta function on the right-hand side of Eq. (9) F (rj . G can be directly substituted into Eq. or the curvature.20]. The level set variable G can be arbitrarily deﬁned away from the front it describes. ∇c ∂c |∇c| = nj |∇c|. (2) into the relevant derivatives. Knudsen. It will. then. which will be denoted nj . sL. or in any other suitable front tracking scheme. The variable G will be deﬁned as G(xj . c=c0 (10) where it is assumed that nj = ∇c/|∇c| points toward the burned gases.20] ∂c ∂ ρD ∂xj ∂xj = ρD|∇c| ∂nj ∂ + nj ρD|∇c| . (1) with a δ function. t) − c0 . which is a function of the density. In the absence of curvature and strain effects. while the variable G cannot. This burning velocity model can then be used in a level set description of the ﬁltered front.c0 describes the propagation velocity of the isosurface c = c0 . however.c0 = ∂ 1 nj ρD|∇c| + ωR ˙ |∇c|ρ ∂xj . the gradient of the progress variable must ﬁrst be written in terms of the front normal direction at c0 . Use of the product rule on the diffusive term then gives [19. t) − c0 drj . xj )G(xj − rj .c0 to deviate from sL . Because there is nothing arbitrary about its definition. H.c0 reduces to the unstretched 1-D laminar ﬂame speed sL . the normal diffusion term. This equation will not be numerically tractable for the same reasons that ﬁltered progress variable equations are not tractable. which with Eq. While it would be possible to introduce a more elaborate dependency such as a reaction source term that depends on ﬂame stretch.21]. = ∂xj |∇c| (4) set variable G. while both of these variables are used to describe a front. Deﬁning F (rj . (8). they differ in an important way. A new variable will now be introduced for the purposes of notational convenience. provide the basis for developing a consistent turbulent burning velocity model for LES. these effects exist and might be important [19. as expected. t) = V Finally. this would not signiﬁcantly affect the model development that is to follow. the quantity sL. sL is the best ﬁrst-order approximation that can be made [6. and the reaction source term in the progress variable equation. (13) . ˙ ∂xj where κ is the divergence of the normal vector. Although they cause sL. (2) produces δ(c − c0 ) ∂c 1 ∂ ρD ρ ∂xj ∂xj + 1 ωR ˙ ρ 1 ρ (7) = Dκ ∇ H (c − c0 ) + δ(c − c0 ) × nj ∂ ρD|∇c| + ωR . (6) produces an equation that is well deﬁned at all locations in space. gives ∂[H (c − c0 )] ∂[H (c − c0 )] + uj ∂t ∂xj = Dκ ∇ H (c − c0 ) + δ(c − c0 ) × nj ∂ ρD|∇c| + ωR . t) = H c(xj . This will be possible because the ﬁltered G equation will indicate exactly how to separate resolved and subﬁlter propagation components. xj )H c(xj − rj . G can be volumetrically ﬁltered. the delta function acts on |∇c| as δ(c − c0 )|∇c| = δ(c − c0 )∇c = ∇ H (c − c0 ) . (6) Applying the results of these manipulations to the right-hand side of Eq. Operating on Eq. xj ) to be some appropriately normalized ﬁlter kernel and then applying it to the G ﬁeld gives G(xj .20].E. t) drj (12) The G symbol is intentionally used to denote this variable because of its close relation to the standard level = V F (rj . In the following. ∂G ∂G + uj = (Dκ)c0 |∇G| + sL. ∂xj ∂xj (5) (11) Following Peters [6. the goal will be to develop an equation that describes the evolution of a ﬁltered G ﬁeld. Pitsch / Combustion and Flame 154 (2008) 740–760 743 δ(c − c0 ) ∂c ∂c + uj ∂t ∂xj ∂[H (c − c0 )] ∂[H (c − c0 )] + uj = . However.

Equation (23) nonetheless informs the use of ﬂame front tracking methods. Γ c0 describes how subﬁlter velocity ﬂuctuations convect the ﬁltered front. (23) show explicitly how ﬁltering affects level set equations. the ﬁltered curvature. This term is interesting because the velocity component under the ﬁlter appears next to the gradient of G. nj uj. a level set description of a ﬁltered front can be developed using an isocontour of G. (19) can be used in Eq. = ρ and the ﬁnal ﬁltered equation may be written as ∂G ∂G + uj +Γu ∂t ∂xj ρu (Dκ)T .744 E.c |∇G| = (Dκ)c0 |∇G| 0 (16) have been used. xj ) that is used to ﬁlter the Navier–Stokes equations can be used to ﬁlter a representation of the ﬂame front. Knudsen. ∂G ∂G + uj +Γu ∂t ∂xj = ρu − ρb pb + 1 ρ (21) × (Dκ)T . the resolved and subﬁlter contributions to the convective term will here be separated by introducing the variable Γ c0 . however.u + sT .u = nj uj − ρu − ρb ρ (Dκ)T . uj. Equations (15) and (16) represent the introduction of deﬁnitions describing the burning velocity associated with a certain ﬁlter level.c . the derivations of Eq.c |∇G| = sL. sT .u |∇G|. 0 Since the convective term in Eq.u pb . G(xj . Modifying the notation used in that transformation to match current convention. For example. It cannot be used to track a ﬂame front. ∂xj ∂xj (18) (22) Physically.c0 |∇G| 0 this conservation property produces. it must also be modeled. (23) Equation (23) will be used below to derive a dynamic turbulent burning velocity model. (19) (14) (15) where pb is the fraction of gas in the ﬁlter volume that is burned and where the tilde denotes Favre ﬁltering. ρc=c0 . H. and it has in fact been shown that such a transformation can be derived using a consistency condition describing mass conservation through a ﬂame front [5].c=c0 . (Dκ)T .c |∇G| + sT . The insight gained by this manipulation is that the move from a conditional to an unconditional velocity produces a term that scales with the burning velocity. Eq. = ρ The conditioning that appears in this relationship makes it clear that Favre ﬁltered velocities cannot be directly used to model this term. Pitsch / Combustion and Flame 154 (2008) 740–760 This ﬁltering procedure is consistent with LES in the sense that the same ﬁlter kernel F (rj . If the c = c0 surface is chosen to represent the unburned side of the front. t) = G0 ∀ G(xj . 0 0 ∂t ∂xj where sT . The term in the brackets may be further manipulated.u + sT .c=c0 ∂G ∂G = uj.c . (10). It will be assumed here that the ﬁlter kernel F (rj . Speciﬁcally. This same ﬁlter may be applied to Eq. The problem with Eq. This gradient acts as a δ function and the ﬁltered quantity therefore only contains information about velocities that are conditioned on the density that coincides with the front location. uj. (14). because G is underresolved on implicit LES meshes.c=c0 + Γ c0 . (24) . t) = G0 .u + sT . Speciﬁcally. and 0 − + Γ u. ρu − ρb pb + 1 ρ ρu pb − ρb pb + ρu (1 − pb ) + ρb pb = ρ ρu . Conditionally ﬁltered velocities must instead be employed. It would therefore be convenient to transform the conditionally ﬁltered quantity into an unconditional quantity. (14) is not known.u + sT . Doing so produces ∂G ∂G + uj = (Dκ)T .u |∇G|. Equation (18) can be treated as the deﬁnition of Γ c0 .u ∂G ∂G = uj ∂xj ∂xj ρu − ρb ρ (Dκ)T . where again c = c0 will denote unburned conditioning. It therefore must be true that uj ∂G ∂G = uj. ∂xj ∂xj (17) Equation (20) can then be substituted into Eq. (18). xj ) commutes with spatial derivatives. (10) and Eq. (18) is that conditionally ﬁltered velocities are usually not available in realistic simulations. considering conditioning on the unburned gas for example.u pb |∇G| (20) and (Dκ)T .c |∇G|.

Pitsch / Combustion and Flame 154 (2008) 740–760 745 ∇ G(xj . would remain stationary. The point can be illustrated by considering the release of a non-propagating front in a ﬂowﬁeld of isotropic turbulence. In other words.u + sT . (23). Dt ∂t ∂xj (26) where uG.E. Knudsen.u (28) When a ﬁlter and a test-ﬁlter are applied to Eq. What is important to recognize is that these contributions are all made in conjunction with another physical mechanism. they will tend to have no effect on the mean front position. ρ ρ T . Γ u describes how subﬁlter velocity ﬂuctuations affect the ﬁltered front. The mean front position. they act only along a 2-D surface within the ﬁlter volume. Since the G0 and G0 surfaces must propagate in exactly the same fashion. and the unﬁltered front would become more and more wrinkled. this identity is analogous to Germano’s identity. (27). since the direct application of a ﬁlter to a level set equation might not produce an equivalent result. This insight further demonstrates the importance of the present ﬁltering approach.u is to produce a front speed that ensures the mean ﬂame front position is independent of the ﬁlter being used. But. In general. In some respects. For example. Eq. When a test-ﬁlter. If the levelset G = G0 is to follow the actual ﬁltered front. When they are ﬁltered. s = Dκ T . the approximations Γ u ≈ 0 and Γ u ≈ 0 must hold. which here will be denoted by the hat operator. This can be achieved by forcing equations (28) and (29). the velocity uG. .j can be developed immediately from the ﬁltered front equation. to produce equivalent front positions. Unlike the subﬁlter Reynolds stress term that appears in the ﬁltered Navier–Stokes equations. and some tend to move the front location backward. therefore. The difference between these cases and the case of passive front release is the presence of self-sustaining front propagation.u ρ ρ T . A governing equation for G that is valid at the front can then be derived by taking the material derivative of Eq. subﬁlter scale velocity ﬂuctuations would certainly exist. (24). since the test-ﬁlter in Eq. Some of these velocity ﬂuctuations tend to move the front location forward. that equation becomes ∂G ∂G + uj + (Γ u ) ∂t ∂xj = ρu ρu (Dκ)T . (10) as a single operator. The broad idea behind deriving a dynamic model for sT . ρu ∂G ∂G + uj − (Dκ)T . Or. While these velocities tend to wrinkle the front.u (29) Equations (28) and (29) form the basis for the development of the dynamic model. ∂G ∂G + uj +Γu ∂t ∂xj ρu ρu ∇G + ∇G . It must additionally account for the subﬁlter term Γ u . (28) commutes with derivatives. ∂G DG ∂ G = + uG. Dynamic propagation model The existence of an appropriate ﬂame surface ﬁltering procedure.u nj ∂t ρ ∂xj + Γ u = 0. (27) where nj in this new context is nj = ∇G/|∇G|. DNS data showing the validity of these approximations will be presented in Section 4. (23). t). which were created using different sequences of ﬁlter application. a different result is obtained. then uG. such as the one developed in the previous section.u |∇G| + s |∇G|. H. The development process will begin with a discussion of the subﬁlter convective terms that appear in the equations. (25) where the hat on the G variable denotes that it is associated with a ﬁltered front. In this situation. makes it possible to derive a dynamic turbulent ﬂame speed identity. however. The unique behavior of this subﬁlter term is directly due to the gradient of G being non-zero only locally at the front. subtracting these two equations produces a relation that will hold at all times if ﬁltering is being properly performed. is applied to Eq. t) = 1 ∀ G(xj . which would lead to increased heat diffusion and an adjusted burning velocity. 3. Γ u should therefore be unable to contribute to front propagation. subﬁlter velocity ﬂuctuations can of course contribute to the movement of a ﬁltered front. velocity ﬂuctuations might promote an increase in ﬂame surface area that in turn leads to an increased consumption of unburned mixture. the subﬁlter term Γ u that appears here is relatively insigniﬁcant. they might promote the growth of subﬁlter ﬂame curvature.j must explicitly contain all of the terms in the parentheses of Eq.j = 0.j is the velocity at which the level set moves. This identity will relate turbulent ﬂame speeds associated with different ﬁlter levels and provide an equation that can be solved for a model constant. Speciﬁcally. since Γ u is uncorrelated with a propagation speed. The only intermediate step involves rewriting this equation using an expanded version of |∇G|.

Applying Eq. [23]. Eq. xj ) × H c(xj − rj . When this wrinkling occurs.u ρu ρu s Dκ + T . Another means of computing these terms is therefore needed. Because |nj | = 1. In the limiting case of very large ﬁlter widths. A similar problem occurs with the gradients of G that appear in the convective terms. and |N j | would approach zero. Kollmann et al. and that there is no restriction on its magnitude as there is in the case of a typical normal vector. it has been shown that the integral in Eq. N j is no more computationally tractable than |∇G|. this is because jumps in G occur over single ﬁlter widths. ∇G(xj . They then recover Kerstein et al. (32) where Φ is a scalar and ψ is the value of that scalar describing an isocontour of interest. however. differentiate Σ and |N j |. (30) are computable. (31) where nj = ∇c/|∇c|. Wenzel and Peters [24] use this relation to describe the area of ﬂame N j deﬁnes the front conditioned ﬁltered normal vector. making the G variable computationally intractable and unavailable. Pitsch / Combustion and Flame 154 (2008) 740–760 uj ∂G − ∂xj = uj ρu ρu (Dκ)T . this is because the propagation are only deﬁned at their respecspeeds sT . are not. the terms in the parentheses of Eq. (33) can be brought outside the integral. They leverage the fact that their study permits treating multiple level set isocontours as equally valid representations of ﬂame fronts. suggest a model for |∇G|. Eq. Σ = |∇c|. which will not occur if satisfaction of the equation is forced for all time and space.u ∇ G .u ρ ρ T . These tendencies suggest that a link exists between |N j | and the ﬂame surface density. An alternative approach can be developed by starting from the deﬁnition of G. The same technique will be used here. (33) F (rj . In part. In regimes between the completely resolved and the completely unresolved limits. The terms that depend on G. front normals within the ﬁlter volume will point in misaligned directions. (30) ∂G − ∂xj Like the identities used to describe subﬁlter stress models.746 E. In the well resolved limit in which a front appears in a ﬁlter volume as a planar surface.u + s |∇G| ρ ρ T .Φ(xj )=ψ f (xj ) dA(xj ). (30) holds statistically for a given number of realizations. (32) to Eq. (31). The presence of δ(c − c0 ) and nj in Eq. Knudsen. H. and no volumetric averaging will be performed. a different approach will be taken. the possibility of ﬁnding subﬁlter front wrinkling also grows.u tive fronts. (31) can be rewritten in terms of the local ﬂame surface area. It is important to note that the presence of the ﬁlter kernel F makes N j a volume weighted quantity. The difference due to the δ function can . While Eq. the normal vector nj = ∇c/|∇c| in Eq.c=c0 F nj dA = |N j |. the value of |N j | reﬂects a balance between a tendency to decrease in magnitude due to misaligned normals and a tendency to increase in magnitude due to increased surface area per volume. In dynamic subﬁlter stress models. an appropriately large number of realizations are ensured by averaging the terms in the dynamic equation over some volume of the ﬂow. and that models from the ﬂame surface density (FSD) formalism might be applicable. |N j | would then reduce to a measure of the front area per ﬁlter volume. (33) grows. As the ﬁlter width in Eq. and go on to integrate this expression over all of these isocontours. Speciﬁcally. t) − c0 drj = V F nj δ(c − c0 )|∇c| drj . (30) does not hold locally and instantaneously. N j does. however. present the relation f (xj )δ Φ(xj ) − ψ |∇Φ| dxj V = S. It is also in part due to the need for model constants to be reasonably bounded. Equation (32) will be applied to only the surface of interest. In work by Maz’ja [22] and subsequently Kollmann et al. Front normal vectors would then tend to cancel each other out when integrated. Rather.u and s T . In the present study where the goal is to extract information from only one isosurface. (33) provides a useful way of looking at |∇G|. which exist at different spatial locations.’s [4] expression showing that the magnitude of the volumetrically integrated level set gradient represents an ensemble averaged ﬂame area. a ﬁlter volume might encompass a series of opposed fronts. t) = ∇ V fronts in their level set based study of how premixed ﬂames propagate. however. (31) produces F nj δ(c − c0 )|∇c| dV V = S. Given curvature and burning velocity models. Again.

15]. in an effort to prevent the introduction of further equations. VΔ (34) uj nj AΔ − VΔ ρu AΔ ρu (Dκ)T .u that considers interactions between resolved curvature and resolved mixing. and thus will be much smaller than the actual ﬂame area in the ﬁlter volume. This dynamic model may therefore be viewed as an enforcement of a constant rate of ﬂame mass consumption. in exact agreement with Eq. (37) simply changes VΔ to VΔ . Therefore. and AΔ to AΔ . With a model prescribed for |∇G|. Knikker et al. This condition satisﬁes the stated goal of a having a propagation model that produces ﬁlter-independent mean front positions. The gradients in these terms must have the same magnitude as |∇G|. it requires the further development of equations and assumptions regarding Σ . it is acceptable to write nj as the computable quantity nj = ∇ G/|∇ G|. . Knudsen.u s T . (33). Furthermore. for example. This implies that FSD models could be introduced to close Eq. if convective effects are neglected and if the limit of the corrugated ﬂamelets regime. has been proposed [5]. Pitsch / Combustion and Flame 154 (2008) 740–760 747 be overcome by considering the “ﬁne-grained” ﬂame surface density. an approach stemming from the level set framework will be introduced. Apparently. These models can now be introduced into a ﬁnal form of the dynamic equation. |N j | is equivalent to the quantity Σ /Ξ [14] in the FSD formalism. and between resolved curvature and subﬁlter mixing.u = AΔ AΔ . VΔ . In the poorly resolved limit. it would be inconsistent if they pointed in any direction other than the normal vector associated with the ﬁltered front. (31). the area of the entire ﬁltered ﬂame front that exists within that region would be used to represent AΔ . the test-ﬁltering operation that acts on the left-hand side of Eq. The “wrinkling factor” Ξ must additionally be introduced to account for nj .u VΔ (37) In the fully resolved limit. independent of the ﬁlter being used. the ﬁltered ﬂame area corresponds to the actual planar ﬂame area within the ﬁlter volume. AΔ will still represent the area of a planar ﬂame. |∇G| = |N j | ≈ AΔ . A model for (Dκ)T .u VΔ ρ ρ where Δ denotes the test-ﬁlter width. AΔ describes the area of the ﬁltered ﬂame in the test-ﬁlter volume. the expression can be further simpliﬁed to sT . In such a case. Since the previously deﬁned level ¯ set variable G describes the same front as G. the dynamic equation reduces to ρu ρu AΔ AΔ s s = . (33) is still satisﬁed. H. This reduced area accounts for the effects of the misaligned normals. where curvature effects are small. ¯ uj ∂G AΔ ≈ uj nj . (30) are easily dealt with. and the ﬁlter volume itself. This form of the equation is very general and can account for all premixed regimes except for the broken reaction zones regime. For example. Σ [14. (39) and a tractable form of the convective terms becomes available. Nonetheless.u (38) If the differences between the ﬁltered and test-ﬁltered density are assumed to be small. the convective terms appearing in Eq. it is interesting to consider what happens to the equation in simpliﬁed limiting cases.u Δ ρ T . model Σ /Ξ using Σ and a fractal expression. is considered.u T . In areas where the ﬂame front does not exist.E. This leaves ρu ρu s A = s AΔ . T . While this approach is an option.u VΔ ρ T . ρ T . As discussed above. and a ﬁlter volume might therefore need to encompass a multi-cell region of space. Therefore. If a box ﬁlter is used and it is assumed that the ﬁltered ﬂame is homogeneous along the directions parallel to its surface. most dynamic models need to consider multiple realizations of a phenomenon.u + s ρ ρ T . ∂xj VΔ (35) AΔ is zero at all locations in the test-ﬁlter domain that are farther away from the ﬂame front than the ﬁlter width Δ. AΔ is zero. nj = ∇G/|∇G|.u VΔ = uj nj − AΔ VΔ (36) ρu AΔ ρu s Dκ + . To apply this procedure to the determination of an unknown constant. [14]. and Eq. but Σ does not account for the normal vector nj . actual burning velocity and diffusive propagation models must be selected. This expression is in agreement with Damköhler’s hypothesis that the turbulent ﬂame speed scales with the turbulent ﬂame surface area [25]. ρ T . AΔ . The regime limits that have been described can be satisﬁed by modeling |N j | using the area of the ﬁltered ﬂame front that passes through the local ﬁlter volume.

(41) will be used to model sT . (44) must be performed. Knudsen.Δ . (44) is not expected to hold instantaneously for all ﬂow realizations. This is because the dynamic coefﬁcient α in Eq.u κ ) Δ VΔ ρ −γ DaΔ 1/2 + (γ DaΔ )2 + γ α DaΔ ρu sL. (44) will be averaged before they are testﬁltered for use in the model.u = sL.u + uΔ .u + uΔ ρ γ DaΔ AΔ .Δ .u + uΔ −γ DaΔ ρ 1/2 AΔ ρu AΔ (Du κ + Dt. (44) will be averaged after the test-ﬁlter is applied. To prevent the occurrence of negative values of α.u κ) − VΔ ρ VΔ + (γ DaΔ )2 + γ α DaΔ = uj nj − AΔ VΔ − + ρu sL. The dynamic identity will then take the speciﬁc form.Δ. respectively. (42) 2 Dt. uj nj AΔ ρu AΔ (Du κ + Dt. or using more complex techniques [27].u κ ) Δ − VΔ VΔ ρ + γ α DaΔ 1/2 AΔ VΔ ρu sL. sT .u .Δ. The model constants have. The test-ﬁltered terms on the right-hand side of Eq. Eq. While more expensive than a simple algebraic evaluation. VΔ (44) − + × − ρu A (Du κ + Dt. In the remainder of this work. means that α must be locally solved for using an iterative procedure such as Newton’s method. and where DaΔ is the Damköhler number associated with the ﬁlter width. relative to Peters’ original model. α sT .u = −γ DaΔ uΔ + (γ DaΔ )2 + γ α DaΔ 1/2 . This prevents α from being explicitly solved for in terms of the other variables in the equation. averaging can be performed locally. In a practical simulation. . a certain amount of averaging of the terms in Eq. Pitsch / Combustion and Flame 154 (2008) 740–760 (Dκ)T . this proposed model will be adopted. (41) where α and γ are RANS-type model constants. VΔ 2 Equation (44) is somewhat more complex than a typical dynamic equation for a Smagorinsky subﬁlter viscosity coefﬁcient. with these averaging procedures included. The ﬁltered terms on the left-hand side of Eq.u . Additionally. This procedure requires information from both the local computational cell and the surrounding cells that are used in the test-ﬁltering operation. H. this + γ α DaΔ (45) where the · operator denotes the averaging procedure. for example. the full operation is nonetheless still computationally cheap and does not signiﬁcantly affect the cost of a simulation.u = Du κ + Dt.Δ. (41) will be treated as a dynamic coefﬁcient. been renamed and regrouped.u − sL.u = sL. interactions between subﬁlter curvature and subﬁlter mixing are left to be determined by the burning speed model.u + uΔ −γ DaΔ ρ γ DaΔ AΔ VΔ 2 ρu AΔ A (Du κ + Dt.Δ. (40) where Dt. as discussed earlier in this section. These limits correspond to the corrugated ﬂamelets and thin reaction zones regimes. is shown in Eq. in a homogeneous ﬂow direction [26]. Eq. In this formulation.u κ.Δ.u 1/2 . A model for describing the burning speed itself will be taken from Peters [5. This regrouping permits the appearance of the model constant α in the limits of both large and small DaΔ . The ﬁnal form of the dynamic equation. (44) appears in an expression that is raised to the 1/2 power.748 E.6]. Negative values of α are incompatible with the burning velocity model that is used here because they can result in a negative value being raised to the 1/2 power. (45). Just as in the computation of Smagorinsky subﬁlter viscosity coefﬁcients. For the remainder of this paper. will occasionally be negative.u 1 + (γ α)1/2 (43) Du In the model validation that follows. sT .u is the turbulent diffusivity conditioned on the unburned gas density. both the subﬁlter velocity ﬂuctuation uΔ and the ﬁltered ﬂame area per unit volume AΔ /VΔ will be averaged.u κ) − VΔ ρ VΔ uj nj − ρu sL. and α in Eq. The value of α that solves this equation in a practical LES.u + uΔ −γ DaΔ ρ 1/2 = uj nj AΔ . sT .

A is the forcing coefﬁcient used by Rosales and Meneveau [28]. (47) is solved to describe front evolution. The ﬁrst-order δ function scheme of Smereka [30] is used for the area computation. the trends seen in Fig. A parallel. This is done to allow the ﬂame brush thickness to increase to a width of multiple integral scales. but in the direction of front propagation the domain length is doubled. The level set equation ∂G ∂G + uj = sL |∇G| ∂t ∂xj ∀ G = G0 . Pitsch / Combustion and Flame 154 (2008) 740–760 Table 1 DNS parameters Simulation constants Mesh size = 512 × 256 × 256 Cell width. Because the linear forcing scheme used here adds energy to the ﬂow at all wavenumbers. The front. is never allowed to come so near these boundaries that their treatment inﬂuences behavior. x = 1. The parameters describing the DNS are shown in Table 1. Additionally. Knudsen.15 m/s Forcing coefﬁcient.32 s Kolmogorov scale. computed at the point in time of level set release. A uniform Cartesian mesh is used.87 × 10−5 m2 /s ρ = 1. Once the forcing scheme strikes a balance with the viscous dissipation. [29] is used for the PDE step. The simulation is run using a constant density ﬂow ﬁeld.E. sL = 0.16 kg/m3 Burning velocity. and the CFL number is limited to 0. 1 shows two instantaneous realizations of the ﬂowﬁeld and front. η = 5. the turbulent ﬂowﬁeld is initialized within a 2563 cube and then copied to an adjacent cube. and then solving a PDE in pseudotime to improve the accuracy of the estimate. |∇G| = 1.0 × 10−3 m 4. A = 1. Fig. H. structured code that is fourth-order accurate in space and second-order accurate in time is used to compute the ﬂow. Neumann boundary conditions are prescribed for the level set at each end of the domain in the direction of propagation. the spectrum does indicate that turbulent physics are present. In general. lt = 32. the simulation effectively describes front propagation in the corrugated ﬂamelets regime. The third-order WENO scheme of Peng et al. These equations are solved in the incompressible limit. so gas expansion is not considered. This prevents the generation of wavenumbers smaller than the inverse of the box size. When the front approaches the edge of the domain along the long axis. Reinitialization is accomplished by using an iterative marker method to estimate the distance to the front. 2 shows the average turbulent kinetic energy in the domain as a function of time.05.0 × 10−3 m ν = 1. as needed in the model. and a Poisson equation for pressure is therefore used to enforce the conservation of mass equation. the energy in the domain stabilizes in time. 3 shows the energy spectrum of the DNS. however.0 × 10−3 m Eddy turnover time. Fig. A reinitialization procedure is performed after every three time steps to force the level set ﬁeld variable away from the front to conform to a distance function. Initially. however. ∂uj = 0. 2 agree well with the results reported by the developers of the forcing scheme [28]. ∂xj ∂(ρui ) ∂(ρuj ui ) + ∂t ∂xj =− ∂P ∂ 2 ui + μ 2 + Aρui . ∂xi ∂x j (46) are solved to describe the ﬂow physics. a data shift is performed which moves the front and all the ﬁeld variables about a domain length in the direction opposite of front propagation.0 × 10−4 m Largest eddy size. Fig. The code is run using an explicit solver. Since no curvature terms appear in this equation. the dissipative . τ = 0. the kinetic energy rapidly decays due to dissipation. Periodic boundary conditions are applied to the velocity ﬁeld in all three coordinate directions. Turbulence is forced using the linear scheme of Rosales and Meneveau [28].5 eddy turnover times.038 Turbulence parameters 749 Reλ = 48 Ret = 149 Integral length scale. The forced Navier–Stokes equations. Statistics involving the front are computed using information from only one isocontour of the level set ﬁeld variable (G = G0 ). This isocontour is used to compute the area of the ﬂame. Even though this DNS only captures a very limited inertial range. Validation in DNS A DNS of a front propagating in forced isotropic turbulence is performed to validate this model. l = 32. 0. at approximately 27.8. Periodic boundary conditions are prescribed for the level set in the other two directions. the effects of the forcing scheme become apparent and the turbulent energy begins to increase. The level set is released after stabilization has occurred. After several eddy turnovers.

4 demonstrates that Γ c0 can grow to values that are comparable in magnitude to the resolved front convection term.0. in the DNS 2 as a function of time. This data sampling scheme then switches to providing information about the new front section rather than the section behind it.c=c0 S. The level set is the wrinkled surface. DNS data (◦ ◦ ◦). Kollmann’s relation. u · ∇G. Occasionally. H. Fig. The top image shows an early time in the simulation. The values of Γ c0 that are shown in this ﬁgure are evaluated at the location in the computational domain where the instantaneous unﬁltered front intersects a particular vector. and therefore ahead of other sections of the front. The location of the intersection changes with time as the front moves through the domain. The forcing scheme then increases and eventually stabilizes k. but it can at times reach relatively large magnitudes because the . Eq. 4 shows Γ c0 as a function of time for various ﬁlter widths. After manipulation. 1. indicating that the smallest scales of the ﬂow are resolved. this tracking scheme suffers at smaller ﬁlter widths when turbulence convects sections of the front laterally across the domain centerline.c=c0 nj dA − uj. The quantity Γ c0 describes seemingly small subﬁlter effects. It was argued in Section 3 that Γ c0 does not contribute to resolved front propagation. (32). and the evaluation therefore tracks the front.u ≈ 1. k = 1 uj uj . This tracking scheme is not Lagrangian because it does not track a particular section of the front. 3.c=c0 F uj. the computable expression for Γ c0 is Γ c0 = S. Average kinetic energy. The initial turbulent kinetic energy has largely dissipated by 3 eddy turnover times. (48) Fig. κ −5/3 scaling (—). Energy spectrum from the DNS. but it does show how Γ c0 evolves at the front over a long period of time. Fig.c=c0 F nj dA.5 eddy turnover times. and the bottom image shows the ﬁeld several eddy turnover times later. This particular vector lies along the domain centerline that points in the direction of mean front propagation. The level set is initialized and released at 27.750 E. Knudsen. falloff occurs smoothly and monotonically. the front propagates from the upper left to the lower right. and the cut plane shows vorticity magnitude. can be used to write this term in a form that can be evaluated directly from the DNS data. From this perspective. but does not affect the conclusions that can be drawn from the data. Fig. Snapshots from a DNS of front propagation with u /sL. Pitsch / Combustion and Flame 154 (2008) 740–760 Fig. This spectrum was computed at the point in time of level set initialization and release. This causes a small discontinuity in the time signal of Γ c0 . 2.

but when considered in the context of describing multiple realizations of a subﬁlter phenomenon. When considered as an effect on the front over an entire ﬁlter volume. does not decrease at larger ﬁlter widths. Physically. Fig. it is appropriate. Just as in Fig. In Fig. Fig. Additional support for this assumption is provided by the fact that Γ c0 decreases in magnitude as the ﬁlter width increases. The opposite trend is observed in Γ c0 . 5. in the limit of a large number of realizations. The procedure that is used to evaluate Γ c0 can also be used to evaluate the convective terms that appear in the dynamic model (Eq. stands in contrast to the behavior of the subﬁlter stress. volumetrically ﬁltered velocity that the resolved convection term sees differs from the velocity that the front itself sees. cancel one another out. these convective terms are plotted as a function of time. In spite of the fact that Γ c0 may instantaneously be large and positive or large and negative. Modeled turbulent viscosity increases as the ﬁlter width increases because this term is forced to describe more and more of the turbulent transport.E. this occurs because a larger ﬂame area is considered and the positive and negative contributions to Γ c0 are more likely to cancel with one another. Γ c0 seemingly bears a resemblance to the subﬁlter stress terms in the momentum transport equations that are typically modeled using a turbulent viscosity. however. these deviations should not be capable of moving the ﬁltered front because they should tend to. The terms are plotted using two different ﬁlter resolutions.. The tendency of Γ c0 to decrease as the ﬁlter width increases. When considered instantaneously. (30). Furthermore. Time evolution of Γ c0 (—) and u · ∇G (. Fig. 4. This choice is again made because the terms are only interesting near the front. This behavior suggests that Γ c0 is often insigniﬁcant relative to the other terms that inﬂuence front dynamics. This tendency to return to zero supports the prior assumption that Γ c0 ≈ 0. The discrepancies that occur between the ﬁltered and test-ﬁltered velocities at the front location may be viewed as instantaneous deviations from the test-ﬁltered velocity ﬁeld. and the ﬁgure also shows the differences in these terms as a function of ﬁlter resolution. the terms are evaluated at locations where the instantaneous position of the unﬁltered front intersects the vector lying along the domain centerline that points in the direction of mean front propagation. The propagation speed of the front. the physical argument that is used to claim that Γ c0 is negligible is equally applicable to the convective terms. evaluated at a position that follows the unﬁltered front. for example. regardless of the ﬁlter resolution. Finally. 4. H. 5 demonstrates that the differences between the concurrently and sequentially ﬁltered convective terms decrease as the ﬁlter width is increased. this appears to be a poor assumption. and neither would resolved convective terms if bulk ﬂows existed. 4 shows that it always ﬂuctuates around zero. for example).-). This is the same behavior that is observed with Γ c0 . . 5 shows the convective terms at the front when a ﬁlter and a test-ﬁlter are applied either sequentially or concurrently. Note the different vertical axis scales. which again supports the assumption that this term does not require modeling. Knudsen. Pitsch / Combustion and Flame 154 (2008) 740–760 751 Fig.

These quantities are evaluated at the same front locations that are used in Figs. 6 shows that |∇G| is accurately represented by AΔ /VΔ . 8 demonstrates that it is in fact a dynamic model. in predicting how propagating surfaces behave. 4 and 5. (34). but these errors are not cumulative. 7 illustrates that sT . The validity of this assumption can be checked using the DNS. H. In each plot. and 32. Fig.u . 7.-) and Δ/η = 32 (—). and additionally captures the majority of the ﬂuctuations of the exact term. 8. even at a constant ﬁlter width.. but spread out. Fig. AΔ /VΔ (. or at the very least helpful. the model in Eq. In Fig. (15). These PDFs are generated using the full ﬂow and level set ﬁelds from a single instant in time. The time instant that is chosen occurs after the front has be- Fig. Since the front prop- .u in Eq. From top to bottom. The turbulent burning velocity associated with a given ﬁlter width and a given point in space is calculated using the deﬁnition of sT .752 E. the test-ﬁlter width Δ is Δ = 1. evaluated at a position that follows the unﬁltered front. come fully wrinkled by the turbulence. 6 shows the time evolution of both the model and |∇G| at a variety of ﬁlter resolutions.5Δ. it has been implicitly assumed that a dynamic procedure is necessary. some probability distribution functions (PDFs) describing the turbulent burning velocity are plotted. 8. Time evolution of |∇G| (—). Fig. the dynamic procedure itself can now be considered.. including how these ﬂuctuations scale with ﬁlter width. The DNS can also be used to evaluate the model for |∇G| that was proposed in Eq. Fig. While not perfect. This suggests that. that is needed. These arguments and data suggest that it may be possible to neglect the resolved convective terms in the dynamic model. and of the model for |∇G| proposed in Eq. Time evolution of u · ∇G and u · ∇ G. (34). the ﬁlter resolutions used in the plot are Δ/η = 4. Pitsch / Combustion and Flame 154 (2008) 740–760 Fig. 16. rather than a static model. and are not biased toward either overprediction or underprediction. Knudsen. 6.u /sL. the turbulent propagation velocity is shown in a scatter plot as a function of uΔ . The PDFs in this ﬁgure are parameterized by ﬁlter width. Instantaneous model errors certainly appear. The model realistically describes the magnitude of the exact term at all ﬁlter resolutions.-) at different ﬁlter resolutions. In Fig.u is unimodal. 5. Since a reasonable level of support is established for the model describing |∇G|.-). Up to this point. (34) is validated by the DNS data.. The third plot shows the difference of these terms at Δ/η = 8 (.u /sL. The ﬁrst two plots show u · ∇G (—) and u · ∇ G (. a model is needed to accurately describe sT .

and 64.u at a particular location on the propagating front. 32. 9. Because sT . which shows PDFs of the dynamic coefﬁcient α at various ﬁlter widths. (15). and 64.u does not unconditionally and monotonically increase with uΔ . and at the same time-dependent front locations used in Figs.5. 8. u /sL.u is evaluated directly from its deﬁnition. Knudsen. Top: sT . the model for the turbulent propagation speed in Eq. (36). 16. 32. 4 and 5. 9 shows that. and 32 (.5 and the test-ﬁlter resolutions are. 7. 16 (.u vs uΔ when the ﬁlter width is Δ/η = 32. but it is well known that dynamic approaches do not produce desirable results when used instantaneously and without averaging. the value of the coefﬁcient α cannot be constant. Note that the averaging procedure discussed at the end of Section 3 is applied to the kinetic energies. 8. This result is veriﬁed by the second plot in Fig. The turbulent burning velocity itself is shown as a function of time and ﬁlter width in Fig. This result is consistent with the idea of the Gibson scale [6]. while others might propagate at four times this speed.-). Interestingly. . some sections of the ﬁltered front might be found propagating at the laminar burning speed. When a large ﬁlter is used. PDFs of sT . H.u /sL. (49) 3 where k is the kinetic energy of the unﬁltered velocity ﬁeld and kΔ is the kinetic energy of the ﬁltered velocity ﬁeld. Fig. and that the averaging occurs over a ﬁlter width in each direction. 8. from the curve with the highest peak to the curve with the lowest.u cannot affect the ﬂame front. the use of a larger ﬁlter tends to result in a larger turbulent propagation speed. Pitsch / Combustion and Flame 154 (2008) 740–760 753 Fig.· · -). evaluated using an instantaneous data ﬁeld from the DNS. however. the mean value of α approaches zero as the ﬁlter width is decreased. Δ/η = 4. as expected. Time evolution of sT . uΔ is computed by determining the amount of kinetic energy that is removed by the ﬁltering procedure..u /sL. This ﬁgure also emphasizes that the turbulent propagation speed is a strongly varying uΔ 2 = Fig. Fig. The ﬁlters used are: Δ/η = 4 (—).u . sT . 9. In this ﬁgure. Dynamic models are designed to capture these kinds of ﬂuctuations. shows that a wide range of turbulent propagation speeds exist for any given uΔ .E.u /sL. (41) is dependent only on uΔ . Eq. 8. (41) does not capture this effect. The scatter plot. This implies that the turbulent propagation speed approaches the laminar propagation speed faster than uΔ approaches zero. which states that eddies with velocities smaller than sL. 2 k − kΔ . Distributions describing the turbulent burning velocity when u /sL.u /sL. While the static turbulent burning velocity model in Eq. In this data set.u ≈ 2. Δ/η = 4. quantity.u ≈ 2. the dynamic model does. 16. Bottom: PDFs of the dynamic coefﬁcient α in Eq. from the curve with the highest peak to the curve with the lowest. parameterized by the test-ﬁlter resolution. The ﬁlter widths are. agation in this DNS occurs in the corrugated ﬂamelets regime.

the resulting dynamic turbulent propagation speed predictions shown on the right in Fig. The difference between the concurrently and sequentially ﬁltered areas appearing in the dynamic equation is shown in the left part of Fig. in an average sense. 10 shows this to be true for the proposed dynamic propagation model. but averaging is only used to help describe the turbulent velocity ﬂuctuations. the failure that occurs in the model proposed here is worth discussing. and burning speed models that employ this quantity can therefore only be used dynamically in a similarly averaged fashion. even the values of α that are calculated from the DNS data are no longer well bounded when the velocity ﬂuctuations used in their computation are not averaged. realization of the front. 10. one of the ﬁltered areas that comprises this difference occasionally decreases in magnitude and becomes relatively small compared to the other. While the values of α that are computed directly from the DNS data all fall within an expected range. the implementation of the dynamic model that employs the full averaging procedure suggested in Eq. In Fig. however. In an effort to compensate for these deviations. 10. When a ﬁlter smaller than the domain width is used. 11. In the ﬁrst combination of ﬁlters shown in Fig. The particular mechanism of failure in this case is inﬂuenced by the fact. This difference should.. Although this problem is. the dynamic model is compared with the static turbulent burning velocity model from Eq. present in all dynamic models. uΔ . This behavior is shown in the middle set of plots in the ﬁgure. shown in Fig. thus ensuring a front consumption rate that is independent of ﬁlter width. Δ/η = 2. Finally. or planar. those computed from the dynamic model do not. the mean front speed. Locally. in further support of this point. 11. strays outside reasonable bounding limits. that the turbulent burning velocity does not in- stantaneously correlate with the ﬂuctuating turbulent velocity magnitude. and the dynamic model coefﬁcient as a function of time. and the terms in the dynamic equation are averaged over one quarter of the front. H. uΔ . Knudsen. Furthermore. it oscillates between positive and negative values. as discussed above. In each plot. The model is applied in a postprocessing step using three sets of ﬁlters and test-ﬁlters. Pitsch / Combustion and Flame 154 (2008) 740–760 Fig. be positive.754 E. Fig. In spite of this bounding of the coefﬁcient.-). the mean front speed is calculated by multiplying the ﬁltered front speed by the ratio of the ﬁltered and planar front areas. the laminar burning velocity. (41). α. Time evolution of the unaveraged dynamic model at a particular location on the propagating front. the ﬁlter width corresponds to the mesh cell width. Note that the “mean front” denotes the completely ﬁltered. In this ﬁgure. The data in the second and third columns are calculated both directly from the DNS (—). and the ﬁltered data is therefore fully . the dynamic model is applied to the DNS data. When this happens the value of the computed dynamic coefﬁcient. and the DNS data. and from the dynamic model (. The plots in the ﬁgure show the mean front position. It only correlates with this quantity in a very averaged sense. All other quantities are left unaveraged. 8. 10 still do not agree with the exact propagation speeds calculated from the DNS. (45) is tested and compared to the DNS data. α is artiﬁcially bounded between −1 and 5 in the ﬁgure.

static sT . Knudsen. The test-ﬁlter width. The dynamic model efﬁciently captures this behavior. it does suggest that the use of surface area as a dynamic scaling parameter is sound. Solving the equation therefore forces the dynamic coefﬁcient α to guarantee that the mean front consumes exactly the same volume of space that the fully resolved front does. is simply the laminar burning velocity.E. While this result does not prove anything about the model’s capabilities in realistic LES settings. and the test-ﬁltered data is fully unresolved. In the third case. predicted. The propagation is computed using the: DNS data (◦ ◦ ◦).u model with: Δ/η = 2 and Δ /η = 512 (—). The ﬁnal front displacement is somewhat overpredicted. dynamic sT .. Unfortunately. however.u does not appear to locally correlate well with uΔ . the dynamic model produces excellent results. laminar burning ). In the second case. Δ /η = 12. The accuracy of these results is directly attributable to the dynamic model’s access to the area of the fully resolved front. as well as the instantaneous propagation velocity ﬂuctuations. The accuracy of the model suffers mildly when the second and third combinations of ﬁlters are used. In this early development period. 11. The model captures the average magnitude of the propagation velocity.· · -). the ﬂame front exists in a transient regime and has not yet fully equilibrated with the turbulence. are more accurate than the corresponding static model results. Δ/η = 3 and Δ /η = 6 speed sL. These ﬁlters correspond to realistic LES ﬁlters and produce results that. the dynamic model does ensure that the testﬁltered front consumes just as much volume as the ﬁltered front. and the ﬁltered propagation speed. it can drift away from values that produce ideal propagation speed estimates. for the second and third ﬁlter combinations applied in Fig.u strongly depends on the product of α and uΔ . sT . resolved. Additionally.u model ( ). It especially improves predictions. Therefore. the ﬁlter width is Δ/η = 3. The value of α that guarantees this property.u ( (. produces a somewhat underestimated front speed when it is returned to the burning velocity model. H. and the testﬁlter width is again twice the ﬁlter width. For example.-).u . The overprediction errors that do occur appear to depend strongly on how uΔ and α interact. corresponds to the domain size. the burning velocity model that is employed here is susceptible to this issue. especially in the corrugated ﬂamelets regime. as discussed above. The front position is calculated by integrating the mean propagation speed. and the test-ﬁlter width is twice the ﬁlter width. This is admittedly a purely academic ﬁltering scheme. Pitsch / Combustion and Flame 154 (2008) 740–760 755 Fig. Modeled time evolution of the DNS front. but when used with this ﬁlter combination. the ﬁlter width is Δ/η = 6. Δ /η = 6. With these ﬁlters. 11. during the ﬁrst few eddy turnover times. Δ /η = 512. This behavior occurs because. while imperfect. although not perfectly. sT . Speciﬁcally. as compared to the static model. when α is allowed to change in order to conserve front volume consumption among different ﬁlter levels. (36) reproduces the exact area per ﬁlter volume of the front. as expected. Δ/η = 6 and Δ /η = 12 (. the left-hand side of Eq. and is accurately. while a dynamically induced change in the value of α might be correctly describing a trend based . sT . but the dynamic model for the most part reasonably reproduces the DNS data.

and the gas exit velocity is 30 m/s. In their formulation. the burning velocity model in Eq. Pitsch / Combustion and Flame 154 (2008) 740–760 on a changing value of uΔ . Apparently. In most industrially relevant LES computations. has been the subject of a number of numerical studies [1. due to the scarcity of experimental turbulent premixed data. This is because the Damköhler number in the DNS is effectively inﬁnity. (41).0) methane–air mixture that exits a jet nozzle at a Reynolds number of ∼23. The second is the need for improved methods of describing subﬁlter velocity ﬂuctuations. such as the present use of Lagrangian rather than Germano averaging in the subgrid viscosity model [27]. Put another way. In this limit.or underpredict the local inﬂuence of turbulence on the propagation speed. If it is not. (36). The computational domain of the simulation reaches from a half jet diameter upstream of the nozzle to 30 jet diameters downstream of the nozzle. the burning velocity would still scale with the ﬁlter width. [31]. the code used in the present work is a variant of the code described in that earlier work. Charlette et al. and therefore. The simulation performed here closely follows prior work in which a level set based premixed combustion model was applied in the context of LES [1]. such as when a ﬂame’s brush thickness is undergoing a transient phase. (41) is to become a function of only uΔ . then the laminar burning velocity itself still plays an important role in turbulent propagation. A pilot ﬂame of equivalent composition is used for stabilization. these DNS results validate the basic premise of the dynamic model. The DNS that is considered here. The dynamic burning velocity model developed above is used to describe the propagation speed of the .u ∼ uΔ . If it did not. the application of a dynamic procedure can improve front speed predictions. is used for this LES. H. Application in LES To demonstrate the dynamic model’s applicability in LES. Knudsen. radial. when Eq. it can dramatically improve them. for example. In general. do not qualitatively affect the computational approach. More speciﬁcally. The nozzle diameter of the Bunsen jet is 12 mm. and the ill-posed limit is not approached. While the procedure is thus justiﬁed. the particular burning velocity model that has been applied presents some challenges to the dynamic procedure. The F3 ﬂame is fed by a stoichiometric (φ = 1. the ratio of Δ and f is signiﬁcantly smaller than 100. by itself. and the dynamic approach of examining how information transfers between the resolved and subﬁlter scales becomes irrelevant. In the context of the premixed regime diagram.756 E. Similar issues regarding turbulent velocity ﬂuctuations have been noted by other researchers. A cylindrical coordinate mesh consisting of 323 × 117 × 64 nodes in the axial. at least two issues regarding the dynamic model proposed here require further attention. and azimuthal directions. this is because the limiting case where the problem becomes ill-posed seems to occur in only a small subset of industrially relevant ﬂows. Primarily. The ﬁrst is the need for an improved method of describing propagation in the limit where sT . subsequently suggested a burning velocity model based on a power law in an effort to deal with this problem. two conditions must hold. In certain cases. for example. If they are to be satisﬁed only weakly. however. it is tested in a simulation of the F3 turbulent premixed Bunsen ﬂame studied experimentally by Chen et al. (41) is used in regions where velocity ﬂuctuations do not correlate well with the propagation speed.u . [16] observed and commented on a sensitivity to the methods they used to model these velocity ﬂuctuations.32. then the equation for the Damköhler number shows that Δ/ f = 100. Charlette et al. the accuracy of the dynamic model suffers. This dependency only disappears for Damköhler numbers above approximately 20 [6]. the dynamic equation. If. all propagation occurs on the subﬁlter scale. In regimes where subﬁlter velocity ﬂuctuations can be accurately estimated. and from the nozzle centerline to a distance of 6 jet diameters in the radial direction. respectively. Speciﬁcally.u = 5.4 million node mesh is run in parallel on 32 processors. so that Da = 20 and uΔ /sL. Im et al. Eq. it is possible that the changing α value is incorrectly describing the magnitude of that trend. [12] showed that in the limit where sT . does seem to reach these conditions when the ﬁlter width is large. The differences that do exist.u scales entirely with uΔ . While the issue of ill-posedness offers insight into the results presented in this section. These two conditions lead to a constraint on the ratio of the ﬁlter width and the ﬂame thickness. rule out a continued consideration of Eq. Addressing these issues would improve the precision with which a dynamic burning velocity model could operate. α might become either too large or too small when modeled velocity ﬂuctuation estimates either over. The ﬁrst condition is that the functional dependency on the Damköhler number must disappear. and the use of a parallel fast marching method for level set reinitialization. is an ill posed problem for determining α. it should not. The second condition is that uΔ must be signiﬁcantly larger than sL. The computation of a ﬂow through time takes roughly 35 wall clock hours.000. and of describing precisely how front propagation depends on them. This ∼2. This ﬂame is experimentally well-characterized. 5. the F3 ﬂame exists in the thin reaction zone regime.33].

uΔ . this approach does not work well in conjunction with Peters’ turbulent burning velocity model. would be signiﬁcantly different. the ratio increases to 1. Speciﬁcally. instantaneous data from these simulations is used to plot the turbulent burning velocity for both the dynamic and static model cases. Speciﬁcally. backing out the associated ﬂuctuation magnitude. In this LES. the turbulent burning velocity is computed using the full model from Eq. (41). This ﬁgure shows that the dynamic model signiﬁcantly impacts the computed ﬂame propagation speed. again using the scheme of Smereka [30]. This coloring shows the ratio of the ﬁltered and test-ﬁltered ﬂame areas. temperature. Fig. the test-ﬁltered velocity ﬂuctuations are computed by taking the difference of the computed ﬁltered and test-ﬁltered kinetic energies. in this particular ﬂame the jet velocity is signiﬁcantly larger than the turbulent burning velocity. (50) 3 This choice guarantees that the test-ﬁlter scale will contain more energy than the ﬁlter scale. and that the static model results differ from the dynamic results by a few uΔ 2 = uΔ 2 + . in which the convective terms have been neglected is solved. The red end of the color spectrum represents these larger values. each on a per-ﬁltervolume basis. H. This suggests either that the ﬂame front itself is fairly well resolved in this LES. In Fig. the areas of the ﬁltered and test-ﬁltered fronts are computed. Knudsen. as a function of the turbulent velocity ﬂuctuation magnitude. In the ﬂat. Two subsequent LES runs were performed for the purposes of comparison.Δ / x. it would fail. Second. To avoid this situation. Estimates of the subﬁlter velocity ﬂuctuations needed in the burning velocity model are then made. which predict 50% faster burning rates. which are known to be important in the thin reaction zone regime. Fig. The local inﬂuence of the dynamic model can be seen in the coloring of the ﬂame front. uΔ = νt. the local value of the kinetic energy is computed everywhere both before and after application of the testﬁlter to the velocity ﬁeld. it was argued in Section 4 that these terms should not contribute to front propagation when considered in a statistically averaged sense. this averaging procedure neglects all mesh points that are not within a distance of several cells from the actual ﬂame front. 2 kΔ − kΔ . this ratio is nearly one and is indicated by the blue portion of the color spectrum. Next. as described earlier in this paper. (44). the turbulent burning velocity is taken to simply be equivalent to the laminar burning velocity. The axial velocity. nothing guarantees that testﬁltered turbulent viscosities will produce test-ﬁltered velocity ﬂuctuations that are larger than their ﬁltered counterparts. 13. a slightly modiﬁed version of the dynamic equation. 14 shows that this is not the case. Finally.15. inappropriately dominate the dynamic equation. are dynamically considered. First. The introduction of this Heaviside function allows the ﬂame front to be test-ﬁltered in a mathematically consistent way. due to their magnitude. and water mass fraction plots each compare favorably. In regions of the ﬂame where turbulence has produced a large amount of wrinking.E. Velocity ﬂuctuations at the ﬁlter level are estimated by dividing the subﬁlter turbulent viscosity by the mesh cell size. on a per-test-ﬁlter-volume basis. Speciﬁcally. although not perfectly. the process of solving the dynamic equation is initiated by evaluating a Heaviside function everywhere in the domain using the value of the local level set ﬁeld variable. Similarly. The former of these explanations seems to be most plausible. 14 shows time averaged statistics from the LES simulations. In the second. 12 shows a single snapshot from the LES. At each time step and at each mesh point. or that the ﬂame brush thickness does not often equilibrate with the turbulence in this ﬂame. Therefore. planar regions of the ﬂame. Fig. all of these quantities are averaged in space. The most interesting feature of these statistics is their very weak dependence on the turbulent propagation speed. 13. Since the burning velocity is only needed in the vicinity of the ﬂame front. The dynamic coefﬁcient α is ﬁnally solved for using a simple Newton–Raphson technique. In the ﬁrst of these. but with the value of the coefﬁcient α left constant. the dynamically computed ﬂame speeds are approximately 50% smaller than the statically computed ﬂame speeds over a wide range of uΔ . and adding it to the uΔ estimate. These kinetic energy estimates will be needed in evaluating the velocity ﬂuctuations required by the model. the dynamic propagation model results and the results computed using simply the laminar burning velocity virtually overlap. Pitsch / Combustion and Flame 154 (2008) 740–760 757 ﬂame front. The test-ﬁltering of the Heaviside function is performed using the standard LES test-ﬁlter prescribed in the code. All curvature terms. There are two reasons for neglecting the dynamic convective terms in this LES. in situations where these terms are not considered on a sufﬁciently averaged basis. they will not appropriately cancel out and will. to ensure that a statistically appropriate number of realizations are used in the dynamic calculations. with experiments. If this implementation of the dynamic model were to encounter a situation where uΔ < uΔ . averaging is performed in the homogeneous (azimuthal) ﬂow direction. Eq. In practice. it would be expected that the static model results. While this behavior is supported by Fig. by computing a test-ﬁltered turbulent viscosity. While it might seem reasonable to then estimate the test-ﬁltered velocity ﬂuctuations.

these results do not yet suggest that the dynamic model is superior to the static model. plays the dominant role in determining the ﬂame height.5 station. These small differences can be seen most clearly at the x/Do = 12.758 E. which in turn implies that the premixed front will be convected far downstream before ﬂame propagation can close it off at the centerline. percent at most. Pitsch / Combustion and Flame 154 (2008) 740–760 Fig. Instantaneous snapshot from an LES of the F3 ﬂame. No experimental data is available at this location. Knudsen. since this ﬂame in reality closes off 12 to 15 diameters downstream of the nozzle [31]. A ﬂame propagation time scale can be formed by considering how long it takes a laminar CH4 /air premixed ﬂame to propagate from the burner nozzle to the centerline of the burner. which ranges from 300 K (black) to 2200 K (bright yellow). however.u ∼ 1. Fig. that are produced.u is computed locally in the LES using a static value of α. Right: sT . These results show that the proposed dynamic burning velocity model can be effectively used in large eddy simulations. and furthermore that the dynamic model predicts signiﬁcantly different propagation speeds than the corresponding static model. tf = (Do /2)/sL. the ﬂame displays only a weak dependency on the modeled propagation speed. The cut plane shows a contour plot of temperature.6 × 10−2 s. Comparing these time scales reveals tf = 40tc . 13. is colored according to the ratio of AΔ /VΔ and AΔ /VΔ . Due to the F3 burner’s lack of sensitivity to ﬂame speed. Left: sT . represented by the G = G0 isosurface. tc = Do /Uo ∼ 4 × 10−4 s. The premixed ﬂame. H. as compared to the dynamic model. subﬁlter models should act to ensure that all mean predicted quantities . this time scale analysis supports the idea that turbulence. Addition- ally. A convection time scale can be formed by considering how long it takes a particle injected at the nozzle to propagate one burner diameter downstream. Further testing is needed to demonstrate this in a comparative sense. but the faster ﬂame speed predicted by the static model is evident in the slightly higher temperatures and water mass fractions. This time scale disparity necessitates the use of a pilot ﬂame in this burner. 12. A comparison of the convection and ﬂame propagation time scales in this burner helps to explain these time averaged results. Instantaneous scatter plot of the turbulent burning velocity as a function of the local turbulent velocity ﬂuctuation magnitude.u is computed locally in the LES using a dynamically determined value of α. Conclusions In consistent LES procedures. As a result. 6. and not propagation.

[7] F. this identity ensures that evolving a ﬂame front and then ﬁltering the result yields the same answer that evolving a ﬁltered front does. [2] J. [6] N. Acknowledgments Support from the United States Department of Defense NDSEG Program. a dynamically determined sT . Colin. H. [5] H. [4] A. SIAM. Inst. Ducros. The Mathematics of Combustion.P. the United States Air Force Ofﬁce of Scientiﬁc Research (AFOSR). Results showed that the model predicted the speed of a propagating turbulent front with more accuracy than a static turbulent burning velocity model. A DNS was performed to validate the proposed dynamic model. Peters. Cambridge Univ.u and dynamic sT . Pitsch. Buckmaster (Ed. and H2 O mass fraction at sequential downstream locations. A 37 (7) (1988) 2728–2731. a dynamic model for calculating the propagation speed of ﬂame fronts was presented. Pitsch / Combustion and Flame 154 (2008) 740–760 759 Fig. Poinsot. Kerstein. Duchamp De Lageneste. Fluids 12 (10) (2000) 2541– 2554. 2000.T. Flame 143 (2005) 587–598. Phys. Cambridge. Ashurst. these terms were used to derive a dynamic identity for the burning velocity. Stanford University. 14. W. 157–168. Poinsot. F. L. Pitsch. Combust. Using this approach.u (. 29 (2002) 1971–1978. Proc. When enforced. within certain ranges. In the derivation of the model.A. Applying the model in an LES of a turbulent Bunsen ﬂame produced results that were in agreement with experimental measurements. and a statically determined sT .. Williams. References [1] H. Time averaged statistics from LES of the F3 ﬂame. This approach was useful because it both worked in conjunction with standard LES ﬁltering techniques and because it used information from only a 2-D surface. The sL. in: J. T. are.R.-). pp. Phys. Experimentally measured values (◦ ◦ ◦) are plotted along with LES results computed using propagation speeds of sL. Rev. Pitsch. [3] O. F. Williams. In this paper. This model was derived in a way that ensures this ﬁlter independence criteria is met. Knudsen. temperature. UK. H. Pitsch. of the CTR Summer Program. 2000. Phys.u (. T. the terms that describe subﬁlter inﬂuences on the turbulent burning velocity were explicitly determined. independent of the ﬁlters being used. Furthermore. Fluids 12 (7) (2000) 1843–1862. Press. a new approach to writing a ﬁltered ﬂame front equation was developed. Legier. 97.D. Inst. Veynante.E.· · -). 1985.A.u results virtually overlap. Combust. Turbulent Combustion. [9] H. [8] H. Veynante. Proc.). and from NASA is gratefully acknowledged. Left to right: axial velocity. D.u (—). 29 (2002) 2001–2008. Steiner. Combust. p. . D. in: Proc.

Lindstedt. Fluids 19 (10) (2007) 5101–5122. C. [26] M. Moin.H. J. T. Meneveau.H. M. N. D. 137–148. Ferziger. E. R. Trans. J. N. Pitsch / Combustion and Flame 154 (2008) 740–760 [22] V. C. Combust. Meneveau. Comput.S. J. Technol. W.S. Phys. Fluid Mech. Rosales.H. in: Proc. Flame 107 (1996) 223–244. Charlette. C. Soc.H. Fluid Mech. Peng. Flame 118 (1999) 308–311.G. Veynante. [16] H. Sobolev Spaces. Flame 131 (2002) 181–197. 46 (1941) 601–652. Zhao. Proc. S. H. Phys. Knudsen. Damköhler. [30] P. Osher. [21] M.M. N. Schneemann. Peters. Meneveau. Phys. J. Combust. [20] N. B.P. 1996. Combust. [19] T. D.-C. [28] C. Mansour. Comput.G. Flame 131 (2002) 159–180. Veynante. [18] M. [33] R. Wruck. Peters. P. H. Combust. Chen. [12] F. Phys. Vaos. Fluids 16 (11) (2004) L91–L94. Meneveau. 25 (1994) 1091–1098. J. Matkowsky. Combust.A. Kollmann. Z. Fluids 17 (9) (2005) 1–8. Cant. Matalon. J. Phys. Combust. Chen. [31] Y. Bradley. Flame 145 (2006) 495–511. 2000. Wenzel. [15] N. Oberlack.H. Fluids 9 (12) (1997) 3826–3833. Meneveau. Cabot.G. Fluid Mech.S. J. Herrmann. RWTH Aachen. 158 (2000) 273–297. J. [13] F. Flame 145 (2006) 357–375. [25] G. Chen. H. D. C. Renz. Knikker. Phys. 211 (1) (2006) 77–90. Stanford University. Cabot. pp. Lund. Inst. Im. Lund. M. Charlette.G.S. D. [27] C. Veynante. [11] M. Peters. Combust. G. thesis. Combust. 384 (1999) 107–132. 155 (2) (1999) 410–438.D. Merriman. [23] W. R. W. J. Philos. 319 (1996) 353–385. Herrmann. Echekki. Kang. Smereka. [24] H. Peters. . Bourlioux. Sci. Fluids A 3 (7) (1991) 1760–1765. London A 301 (1457) (1981) 1–25. Abdel-Gayed. Wenzel. [17] A. [32] M. Elektrochem. [14] R. Im. [10] R. Maz’ja. Ferziger. Combust. Germano. 1985. U. T. H. U. N. 124 (1982) 239–259. Theory Modelling 5 (2001) 363–383. Phys. J. of the CTR Summer Program. Springer-Verlag. Piomelli. Ph. Berlin. Chakraborty. [29] D.760 E.H.

by jessicaliudd

Uncertainty analysis for filter radiometry based on the uncertainty associated with integrated quantities

by National Physical Laboratory

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd