Large Eddy Simulation: Inert Flows

W P Jones
Department of Mechanical Engineering Combustion Summer School
Imperial College London Cambridge
Exhibition Road July 2019
London SW7 2AZ
 Turbulent Flows: DNS

Inert Flows Resolve the smallest scales

9
Number of mesh points, N xyx Re 4

3
Time step t Re 4

CPU time Re3

DNS is restricted to simple canonical flows at relatively low Reynolds

Numbers

Page 2 © Imperial College London

 Modelling Approaches

Reynold/Favre Averaged Approaches (RANS)

• Quasi-steady and quasi-homogeneous assumptions

• 40-50 years research with little improvement.

Large Eddy Simulation (LES)

• Assumptions: scale separation, i.e. high turbulence Reynolds numbers

• Resolve large scale energy containing motions responsible for
transport.
• Model fine scale dissipative motions / replace physical viscosity with
sub-grid scale (sgs) viscosity.
• sgs viscosity provides mechanism for dissipation.
• Mean profiles, Reynolds stresses, turbulent kinetic energy etc.
insensitive to sgs viscosity.
Page 3 © Imperial College London
 Large Eddy Simulation

Introduce a spatial filter:

x, t x ,t G x x , d 3x

where G x x, d 3x 1; G( x x , ) 0

Generalised moments:
2 2
, , ui u j ui u j etc.

The filter width: Selected by analyst

13
usually linked to mesh spacing, e.g. x y z
Page 4 © Imperial College London
 Filtered Equations
Continuity
ui
0
t xi
Momentum
ui ui p ui uj
uj ui u j ui u j
t xj xj xj xj xi x
j
Scalar Quantities Sub-grid Stress

uj uj uj S
t xj xj xj x
j
Sub-grid Flux
Unknowns:-
the sub-grid stress & scalar flux: ij ui u j ui u j ; j uj uj
the filtered formation rate: S
Page 5 © Imperial College London
 LES: Justification

Turbulent Flow
• Large scale energy containing motions responsible for turbulent
transport.
• Energy transfer from large to small scales.
• Energy Dissipation via viscosity at the smallest scales.
• Energy dissipation rate and hence turbulent transport independent of
viscosity and indeed the precise dissipation mechanism, (Kolmogorov).
Large Eddy Simulation
• Resolve large scale energy containing motions responsible for
transport.
• Model fine scale dissipative motions / replace physical viscosity with
sub-grid scale (sgs) viscosity.
• sgs viscosity provides mechanism for dissipation.
• Mean profiles, Reynolds stresses, turbulent kinetic energy etc
insensitive to sgs viscosity
 Sub-Grid Models: I
Eddy Viscosity Sub-grid scale models
uk ij ui uj sgs
2 ; j
ij kk sgs
xk 3
sgs
xj xi sgs xj
where sgs 0.7. Note: only the anisotropic part of the stress is
required.

Smagorinsky Model

2
sgs
Cs eij
whe re is the Frobenus norm of the filtered rate of; strain and Cs 0.1

Page 7 © Imperial College London

 Sub-Grid Models: II

Dynamic Model
Introduce a test Filter (with filter width T larger than )

Tij ui u j ui u j ij Tij ij ui uj ui uj

with
Tija
2
2 sgs e ij and sgs
Cs T e ij
2
n 1 a
2C s eij eij ij e ij
2
n
C s 3
e ij
Page 8 © Imperial College London
Sub Grid Models: III

One Equation Models

3
ksgs ksgs sgs
ksgs ksgs2
uj 2 sgs
eij eij C
t xj xj sgs xj

with sgs
C ksgs and C 1.0; C 0.09

Note: ksgs must be initialised to a non-zero value at the start of any

computation!
 Numerical Requirements

Discretisation:-
• Spatial Derivatives
Convection terms: the use of ‘dissipation free’ schemes is desirable if
excessive CPU times/memory are to be avoided.
at least second order accurate central differences. Asymmetric
approximation such as QUICK and upwind schemes are too diffusive!
Diffusion and Pressure gradient Terms: Second order accurate central
approximations yield reasonable results.
• Time
At least second order accurate: Crank-Nicholson, Adams-Bashforth
3-step Runge-Kutta and three-point backward difference
approximations have all been used to good effect.

Solution Methods
• If ∆ is linked to the mesh spacing then the solution will not be ‘smooth’ on
the mesh high frequency ‘noise’
 Numerical Accuracy

Grid independence Energy Spectrum

• If ∆ is linked to the mesh spacing then the
only truly grid independent solution will be a
DNS.
• Resolve 80-90% of turbulence energy
mean quantities, Reynolds stresses etc
independent of mesh size.

Mesh Quality Indicators??

• There is no reliable single mesh accuracy

indicator

Mesh size – ‘a rule of thumb’

• Equal mesh spacing, i.e. ∆x≈∆y≈∆z,
expansion ratios close to unity; sufficiently
fine to resolve mean motion
 LES of an isothermal Jet

Page 6 © Imperial College London

LES OF A JET IN A CROSS-FLOW
MEAN VELOCITY AND RMS INTENSITIES

Measurements, Crab (1979)

Fine grid 154 72 90 (997,920)
·- - - - - Coarse grid 67 33 40 (88,440)
 COMPARISON OF SUB-GRID CLOSURES I
Profiles at x/D = 8
 COMPARISON OF SUB-GRID CLOSURES II
Profiles at z/D = 0 and z/D=0.5
NEAR WALL FLOWS: THE VISCOUS SUBLAYER
The viscous sublayer: the ‘large’ eddy structure scales on the
sublayer thickness.
7 21
Thickness: sl Re 8
numerical resolution, N Re 8

‘Resolve’ the sublayer? - Only feasible for Reynolds numbers up

to about 50,000

Approximate Wall Boundary Conditions are needed for higher

Reynolds numbers, i.e. wall function approach.

Page 17 © Imperial College London

APPROXIMATE WALL BOUNDARY CONDITIONS

Page 18 © Imperial College Londonz

PLANE CHANNEL FLOW

Page 19 © Imperial College London

RESOLVED VISCOUS SUBLAYER RESULTS

Page 20 © Imperial College London

WALL FUNCTIONS: MEAN PROFILES

Page 21 © Imperial College London

WALL FUNCTIONS: TURBULENCE INTENSITIES

Page 22 © Imperial College London

 Near Wall Models: Recent developments

Combine RANS for near wall region

with LES for the outer region.

But LES involve spatial filtering and

RANS time or ensemble averaging.
Matching?

Success to date has been limited, but

see references.
 Conclusions
Velocity
Large Eddy Simulation is capable of reproducing the velocity field in many
inert flows.
Providing that the Reynolds number, Re u L sgs is large and the flow is
adequately resolved then results are relatively insensitive to the sub-grid
stress and scalar-flux models.
Approximate wall boundary conditions are needed for practical high
Reynolds number flows.
For statistically stationary, i.e. ‘steady’, flows the approaches based on a log
law yield realistic wall shear stresses and turbulence intensity profiles.
More research is needed for non-stationary near wall flows.

Page 24 © Imperial College London

References: Wall Modelling

1. U. Piomelli, E. Balaras,Wall layer models for large eddy simulation, Annual Review of Fluid Mechanics 34 (2002) 349{374.
2. H. Choi and P. Moin, “Grid-point requirements for large eddy simulation: Chapman’s estimates revisited,” Phys. Fluids 24, 011702 (2012).
3. U. Piomelli and W. Balas, “Wall-layer models for large-eddy simulation,” Prog. Aero. Sci. 34,349–374 (2008).
4. J. Larsson, S. Kawai, J. Bodart, and I. Bermejo-Moreno, “Large eddy simulation with modelled wall-stress: recent progress and future
directions,” Mech. Eng. Rev. 3, 15–00418 (2016).
5. S. T. Bose and G. I. Park, “Wall-modelled large-eddy simulation for complex turbulent flows,” Ann. Rev. Fluid Mech. 50, 535–561 (2018).
6. W. Cabot and P. Moin, “Approximate wall boundary condition in the large-eddy simulation of high Reynolds number flow,” Flow Turb.
Combust. 63, 269–291 (1999).
7. M. Wang and P. Moin, “Dynamic wall modelling for large-eddy simulation of complex turbulent flows,” Phys. Fluids 14, 2043–51 (2002).
8. S. Kawai and J. Larsson, “Wall-modelling in large eddy simulation: Length scales, grid resolution, and accuracy,” Phys. Fluids 24, 015105
(2012).
9. G. Hoffmann and C. Benocci, “Approximate wall boundary conditions for large eddy simulations,” in Advances in Turbulence V, edited by R.
Benzi (Springer, Berlin, 1995) pp. 222–228.
10. S. Kawai and J. Larsson, “Dynamic non-equilibrium wall-modelling for large eddy simulation at high Reynolds numbers,” Phys. Fluids 25,
015105 (2013).
11. G. I. Park and P. Moin, “An improved dynamic non-equilibrium wall-model for large eddy simulation,” Phys. Fluids 26, 015108 (2014).
12. E. Lévêque, F. Toschi, L. Shao, and J.-P. Bertoglio, “Shear-improved Smagorinsky model for large-eddy simulation of wall-bounded turbulent
flows,” J. Fluid Mech 570, 491 (2007).
13. J. A. Templeton, G. Medic, and G. Kalitzin, “An eddy-viscosity based near-wall treatment for coarse grid large-eddy simulation,” Phys. Fluids
17, 105101 (2005).
14. X. I. A. Yang, G. I. Park, and P. Moin, “Log-layer mismatch and modelling of the fluctuating wall stress in wall-modelled large-eddy
simulations,” Phys. Rev. Fluids 2, 104601 (2017).
15. X. I. A. Yang, J. Sadique, R. Mittal, and C. Meneveau, “Integral wall model for large eddy simulations of wall-bounded turbulent flows,”
Phys. Fluids 27, 025112 (2015).
16. H. J. Bae, A. Lozano-Durán, S. T. Bose, and P. Moin, “Dynamic slip wall model for large-eddy simulation,” J. Fluid Mech. 859, 400–432
(2019).

Page 25 © Imperial College London

LES Approaches to combustion
LES Approaches to combustion

LES: Approaches to Combustion

W P Jones
Department of Mechanical Engineering Combustion Summer School
Imperial College London Cambridge
July 2019
Exhibition Road
London SW7 2AZ
 Turbulent Flows: DNS

Inert Flows Resolve the smallest scales

9
Number of mesh points, N xyx Re 4

3
Time step t Re 4

CPU time Re3

Combusting Flows
Many reactions of type A B C
E YY
rA aT n exp( ) A B
RT WB
Pressure 1 bar, r 0.1 mm
Pressure 40 bar, r 0.01 mm
 Large Eddy Simulation

Introduce a spatial filter:

x, t x ,t G x x , d 3x
x ,t
x, t x ,t G x x , d 3x
x, t

where G x x, d 3x 1; G( x x , ) 0

Generalised moments: , 2 2
, ui u j ui u j etc.

13
The filter width is given by: x y z
 Filtered Equations
Continuity
𝜕 𝜌 𝜕 𝜌 𝑢𝑖
+ =0
𝜕𝑡 𝜕𝑥𝑖
Momentum
𝑠𝑔𝑠
𝜕 𝜌 𝑢𝑖 𝜕 𝜌 𝑢 𝑢
𝑖 𝑗 𝜕𝑝 𝜕 𝜕 𝑢𝑖 𝜕 𝑢𝑗 2 𝜕 𝑢𝑘 𝜕𝜏 𝑖𝑗
+ =− + µ + − 𝛿𝑖𝑗 − + 𝜌𝑔𝑖
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑖 3 𝜕𝑥𝑘 𝜕𝑥𝑗
Scalars
𝑠𝑔𝑠
𝜕𝜌𝜙𝛼 𝜕 𝜌 𝑢 𝜙
𝑗 𝛼 𝜕 𝜕𝜙𝛼 𝜕𝐽𝛼,𝑗
+ = 𝜌𝐷 + 𝜌𝜔𝛼 𝝓, 𝑇 −
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗
𝑠𝑔𝑠 𝑠𝑔𝑠
Closures are required for the sub-grid stress 𝜏𝑖𝑗 , sub-grid flux 𝐽𝛼,𝑗 and
the filtered rate of formation 𝜌𝜔𝛼 𝝓, 𝑇 terms.

with CS determined dynamically i.e. CS= CS(x,t)

 Combustion LES: Justification

• The arguments concerning turbulent transport in inert flows are

likely to apply in equally in combusting flows, i.e. the mean
profiles, Reynolds stresses, scalar fluxes etc. are likely to be
insensitive to sgs viscosity.

• Heat releasing reactions invariably occur within the unresolved

sub-grid scales so that the modelling of spatially filtered chemical
source terms is potentially much more important.

• Sub-grid scale fluctuations are much smaller than those occurring

in RANS so chemical source modelling is unlikely to exert the
dominating rate controlling effect that occurs in RANS
approaches.
•
 Combustion LES: Approaches
Turbulent transport, e.g. Reynolds stresses, turbulence kinetic energy, scalar
fluxes etc., is likely to remain insensitive to the sgs eddy viscosity.
Combustion will take place predominantly within the sub-grid scales. Thus
sgs combustion models are likely to exert a more important influence in
the LES of turbulent flames.
Approaches:
One/two Scalar Descriptions
Non-premixed Flames
• Conserved scalar/unstrained flamelets/CMC.
Premixed Flames
• Thin flame approach (flame approximated as a propagating
‘material’ surface). Models: flame surface density, level set
methods (‘G’ equation), scalar dissipation.
Partially Premixed/Stratified Flames
• Thickened Flame Model, FGM, FSD, Scalar Dissipation,CMC,
unsteady flamelets
• Sub-grid scale or Filtered joint scalar pdf transport equation method.
Non-Premixed Combustion
Conserved Scalar Formalism

Assumptions:
• High Reynolds Number
• Adiabatic Flame
• ‘Fast’ Reaction

Mixture Fraction
• Normalized element mass fraction
z z ,1
air stream : ξ 0 , fuel stream : 1
z , 2 z ,1

• is a strictly conserved quantity

sgs
uj
t xj xj sgs xj
Laminar Flamelet Computations
Laminar Flamelets
• Flame Thickness < Kolmogorov length Scale
• Instantaneous Dependence of Composition,
Density and Temperature on mixture fraction
and some measure of Flame Stretch is the same
as that prevailing in a laminar flame

T T ,s ; Y Y ,s ; ,s

A simple model: select a single flamelet at a

specified value of sR
e.g. T T , sR T etc, where sR 15s 1
(unstraine d)
Turbulence-Chemistry Interactions

Probability Density Function

1
~
x P ( , x )d
0
Beta PDF
r 1 s 1 r 1 s 1
P ; x, t 1 / 1 d

1 1
where r 1 ; s r
2

2
2 2
C
xi
Sandia Flame D – LES-BOFFIN

COMPUTATIONAL MESH

SETUP INFORMATION

1
0
Flamelet Model

U U

U 'U ' U 'U '

1
1
Flamelet Model
INSTANTANEOUS MIXTURE
FRACTION FIELD

Z'Z'

1
2
Flamelet Model
INSTANTANEOUS TEMPERATURE
FIELD (K)

T 'T '

1
3
 RADIAL PROFILES OF CH4 MASS FRACTION

YCH4 YCH4

YCH4 ,rms YCH4 ,rms

14
RADIAL PROFILES OF CO MASS FRACTION

YCO YCO

YCO ,rms YCO ,rms

1
5
RADIAL PROFILES OF CO2 MASS FRACTION

YCO2 YCO2

YCO2 ,rms YCO2 ,rms

1
6
Local Extinction

• In RANs approaches the scalar dissipation,χ is often used to

provide a measure of flame stretch.
• In LES the local value of χ and thus P(ξ,χ) is insufficient to
describe extinction.

Instantaneous mixture fraction Instantaneous mixture fraction

contours 20mm dissipation rate 20mm
 ;

Premixed Flames: Parameters

Length Time
;
Energy containing t U
motions: 1
1 3 4
2
Kolmogorov scales: k

Chemical reaction scales: L

L L
SL SL
where S L is the burning velocity of an unstrained
laminar flame
Reynolds number: Re U

Damköhler t SL
DA
Number: L U L
1
L
KA Re 2 / D A
Karlovitz Number:
k
 Premixed Flames

YH 2O x, t
Define a reaction progress variable, e.g. c
YH 2O ,burnt
unburnt gas c=0
burnt gas c =1

c c t c
Ui x, t
t xi xi t xi

View Flame as a propagating material surface – propagating

at a speed equal to the local burning velocity, (Bray, Peters,
Candel et al, Bradley et al and others)

x,t is obtained from either FSD, G-equation, Scalar

Dissipation for which an additional equation is solved.
Partially Premixed and Stratified Flames

At a minimum both the mixture fraction and reaction

progress variable is required.

Reaction Progress Variable becomes:

YH 2O (x, t )
c with YH 2O ,b (x, t ) YH 2O ,b ( ( x, t ))
YH 2O ,b (x, t )
and dc c dY cd
dt Y dt dt
Resulting equation complex, Domingo et al (2002);
additional terms sometimes neglected.

A solution: solve equations for Y and c(x,t)

H O
2
Reaction rate obtained as for premixed flames with
additional assumptions, e.g. P (c, ) P (c ) P ( )
Finite Rate Chemistry Effects

• Fully coupled flow and chemical reaction required

• Detailed but reduced chemical mechanism

required

• Currently available approaches:

• Thickened Flame Model
• CMC – with at least double conditioning
• PDF transport equation methods
• Lagrangian stochastic particles
• Eulerian Stochastic fields
 Combustion: Sub-Grid Pdf Equation Method
Ns
Fine grained pdf F ; x, t x, t
1

Sub-grid Pdf Psgs ; x, t x ', t F ; x ', t G( x x '; ) dx '

The modelled sub-grid Pdf Equation

Psgs Psgs Psgs N Psgs

uj
t xj xj xj 1

2
N N P sgs
Psgs
1 1 xi xi xi sgs xi
N
Cd
x, t P
sgs 1

Gao & O’Brien (1993), Dopazo(1975,1979), Brauner et al (2016)

Stochastic Field Solution Method
Represent PDF by N stochastic fields
Configuration field methods, M. Laso and H.C. Öttinger (1993); Stochastic Fields,
Valino (1998), Sabel’nikov (2005)

Ito formulation
n
x, t is advanced from t to t dt according to:
n n
n sgs
d ui dt dt
xi xi sgs xi
12
n
sgs
2 dWi n t 0.5Cd 1
sgs
n n
dt n n
dt
sgs xi
where 1 n N , dWi n i
n
dt
n
and i is a -1,+1 dichotomic vector
 Pdf Equation/Stochastic Fields: Applications

Simulated Flames
• Auto-ignition – Hydrogen and n-heptane
• Lifted flames : Cabre - Hydrogen & methane
• Forced ignition: methane,
• Local extinction – Sandia Flames D, E & F
• Premixed swirl burner (Darmstadt)
• Darmstadt stratified flame
• Lean burn (natural gas) industrial combustor
• Premixed baffle stabilised flame
• Cambridge/Sandia stratified flames
• Ethanol, Methanol and n-heptane Spray flames
• Axisymmetric swirl combustor
• FAUGA Combustor,
Kerosene spray
• Sector combustor
• Genrig combustor
SGT100 Combustor

• Objectives: Operability studies, Bulat et al (2014)

• Snapshot of a Combustor simulation
SGT100 Combustor: Snapshot of Heat Release Rate
 Cambridge Stratified Swirl Burner

Swirl Flow Ui Uo Uco Ʌ Re_i Re_o φ_g

Flame SR = φ2 / φ1
Ratio (m/s) (m/s) (m/s) (mm)
SwB1 0.75/0.75 = 1.0 0.00 8.7 18.7 0.4 1.9 6000 11500 0.75

SwB2 0.75/0.75 = 1.0 0.25

SwB3 0.75/0.75 = 1.0 0.33
SwB5 1.00/0.50 = 2.0 0.00
SwB6 1.00/0.50 = 2.0 0.25
SwB7 1.00/0.50 = 2.0 0.33
SwB9 1.13/0.38 = 3.0 0.00
SwB10 1.13/0.38 = 3.0 0.25
SwB11 1.13/0.38 = 3.0 0.33
 Computational Setup

• BOFFIN-LES
• 13.5M cells & 610 blocks
3.7M cells & 163 blocks
0.8M cells & 27 blocks
• Synthetic turbulence generation
• Dynamic Smagorinsky model
• Constant Prandtl/Schmidt number = 0.7
• Reduced GRI 3.0 with 19 species and
15 reactions
• 8 stochastic fields

Brauner et al (2016)

~ measurement locations
 Results – SwB5

Velocity Magnitude Methane Temperature

Results – SwB5 Velocities

Mean Velocities RMS Velocities

 Results – SwB5 Scalars - Mean

Temperature CH4 O2 CO2 CO H2

10 mm

30 mm

50 mm

70 mm
 Results – SwB5 Scalars - RMS

Temperature CH4 O2 CO2 CO H2

10 mm

30 mm

50 mm

70 mm
 Results - SwB7

Velocity Magnitude Methane Temperature

Results – SwB7 Velocities

Mean Velocities RMS Velocities

 Results – SwB7 Scalars - Mean
Temperature CH4 O2 CO2 CO H2

10 mm

30 mm

50 mm
 Results – SwB7 Scalars - RMS
Temperature CH4 O2 CO2 CO H2

10 mm

30 mm

50 mm
 The Preccinsta Combustor

Iso-surfaces of the instantaneous CH4 mass fraction (left) and heat release
rate (right) coloured by, respectively, the velocity magnitude and mixture
fraction - = 0.7, Fredrich et al (2019).
LES snapshots of the instantaneous (top) and mean Radial profiles of the mean axial (left) and
(bottom) axial velocity, temperature and mixture radial (right) velocity from the experiment and
fraction - LES. The dotted line indicates the zero
velocity.
Experimental (left) and LES (right) scatter plots Instantaneous temperature-mixture fraction
of the instantaneous temperature versus mixture correlation from the experiment (top) and LES
fraction. The solid line represents the adiabatic (bottom) at the first downstream location
flame temperature. including marginal histograms.
Qualitative comparison of experimental OH PLIF Radial profiles of the mean (left) and RMS
intensities to an LES with fully adiabatic walls (right) temperature from the experiment
(left) and the LES accounting for wall heat and LES with fully adiabatic walls and the
transfer (right) - = 0.83. The LES colour scale LES accounting for wall heat transfer
was approximated to match the experimental
one.
 Instantaneous snapshot of the simulated
flame index (left) and line-of-sight integration
of the LES mean HRR compared to an
Radial profiles of the mean (left) and RMS (right) experimental mean OH chemiluminescence
CO2 and CH4 mass fractions from the experiment image (right) - = 0.83. The experimental
and LES image was converted into greyscale.
 Conclusions

Combustion

Thin flame LES combustion models give good results when applied appropriately.
The LES Stochastic field pdf method together with detailed but reduced chemistry
has been applied to a wide range of flames – non-premixed, partially premixed,
premixed and spray flames - to good effect.
Practical liquid fuel systems limited by lack of reduced chemical reaction
mechanisms.
 References

Pascale Domingo, Luc Vervisch and Ken Bray, Partially premixed flamelets in LES of nonpremixed turbulent combustion, Combustion
Theory and Modelling, 6, 529-551 (2002)

F. Gao, E. O’Brien, A large eddy simulation scheme for turbulent reacting flows, Physics of Fluids A 5 (1993) 1282–1284.

M. Laso and H.C. Öttinger,

Calculation of viscoelastic flow using molecular models: the connffessit approach, Journal of Non-Newtonian Fluid Mechanics, 47, 1-
20 (1993)

V. Sabel’nikov, O. Soulard, Rapidly decorrelating velocity-field model as a tool for solving one-point Fokker-Planck equations for
probability density functions of turbulent reactive scalars, Physical Review E 72 (2005) 16301–163022.

L. Valiño, A field Monte Carlo formulation for calculating the probability density function of a single scalar in a turbulent flow, Flow,
Turbulence and Combustion 60 (1998) 157–172

C. Dopazo, Probability density function approach for an axisymmetric heated jet: centerline evolution, Physics of Fluids. 18 (1975)
397.

C. Dopazo, Relaxation of initial probability density functiond in turbulent convection of scalar fields, Physics of Fluids 22 (1) (1979)
20–30.

Brauner, T., Jones, W.P., and Marquis, A.J., LES of a Stratified Swirl Burner using a Sub-Grid PDF Approach, Flow, Turbulence and
Combustion. 96, 965-985 (2016)

Bulat, G, Jones, W P and Marquis A J, NO and CO formation in an industrial gas-turbine combustion chamber using LES with the
Eulerian sub-grid PDF method., Combustion and Flame, 161(7) 1804-1825 (2014)