Available online at www.sciencedirect.comProceedings

ScienceDirect

Proceedings of the Combustion Institute 35 (2015) 3337–3345

www.elsevier.com/locate/proci

of the

CombustionInstitute

Modeling heat loss effects in the large eddy simulationof a model gas turbine combustor withpremixed flamelet generated manifolds

F. Proch a,⇑, A.M. Kempf a,b

a Institute for Combustion and Gasdynamics (IVG), Chair for Fluid Dynamics, University of Duisburg-Essen,

47048 Duisburg, Germanyb Center for Computational Sciences and Simulation (CCSS), University of Duisburg-Essen, 47048 Duisburg, Germany

Available online 12 August 2014

Abstract

Large eddy simulation results are presented for a model gas turbine combustion chamber, which is oper-ated with a premixed and preheated methane/air mixture. The off-center position of the high axial momen-tum confined jet burner causes a strong outer recirculation, which stabilizes the flame. Turbulentcombustion is modeled by the premixed flamelet generated manifolds (PFGM) technique, which is com-bined with the artificial thickened flame (ATF) approach. The influence of different heat loss modelingstrategies on flame propagation and structure is investigated. Besides the established method of using bur-ner-stabilized flames as basis for the non-adiabatic tabulation, an alternative approach based on freelypropagating flames with heat loss inclusion by scaling of the energy equation source term is presented. Dif-ferent grid resolutions are applied to study the impact of cell size and filter width on the results, the effect ofsubfilter modeling is also examined. The simulation setup and the modeling approach are validated bycomparison of computed statistics against measurements. A good overall agreement between simulationand experiment is observed. However, the length of the flame was slightly under-predicted; it is shown thata simple method for consideration of strain effects on the flame has the potential to improve the predictionshere. The effect of heat loss on the combustion process is then characterized further based on probabilitydensity functions obtained from the simulation results.� 2014 Published by Elsevier Inc. on behalf of The Combustion Institute.

Keywords: Large eddy simulation; Confined jet flame; Turbulent premixed combustion; Tabulated chemistry; Heat losses

1. Introduction

Concepts for modern gas turbine combustorsoften feature regions with (partially) premixed

http://dx.doi.org/10.1016/j.proci.2014.07.0361540-7489/� 2014 Published by Elsevier Inc. on behalf of The

⇑ Corresponding author. Address: Carl-Benz Straße199, 47057 Duisburg, Germany. Fax: +49 203 379 8102.

E-mail address: fabian.proch@uni-due.de (F. Proch).

combustion, which are exposed to significant heatloss. Large eddy simulation (LES) has proven inthe past to be a very capable approach for thenumerical simulation of such devices.

Different methods have been established forthe modeling of premixed combustion withinLES, for example based on tracking of the flamefront by the G-equation model [1–3] or modelingof the flame surface density (FSD) [4,5]. Another

Combustion Institute.

3338 F. Proch, A.M. Kempf / Proceedings of the Combustion Institute 35 (2015) 3337–3345

group of models is based on widening of the flamein normal direction by means of either filtering[6–8] or artificial thickening (ATF) [9,10]; thelatter approach is applied here. The combustionprogress and flame propagation are describedthrough the premixed flamelet generatedmanifolds (PFGM) technique [11–13]. Theseapproaches are all derived for adiabatic condi-tions in their basic formulation, and need to beextended by heat loss effects for the proper com-putation of confined geometries. Heat loss dueto radiation has been considered for example byMarracino and Lentini [14], by Ihme and Pitsch[15], by Franchetti et al. [16] and by Schmittet al. [17]. Fiorina et al. [18] suggested the compu-tation of burner stabilized one-dimensional flamesand to use the tabulated results for correcting thesource term and the flame speed. This approachhas been applied within the RANS [19] and theLES [20,21] context.

The present work aims to model and investi-gate the influence of the heat loss on the LESresults in the lean premixed combustion regime.Three different models are compared to the adia-batic reference solution: First a very simple modelwhere the heat loss only affects the temperatureand the transport coefficients, secondly the modelby Fiorina et al. [18], and thirdly an approachwith non-constant heat loss inside the flame.

To properly judge the model behavior, a testcase with a significant amount of heat loss wasrequired, which was found in a confined labora-tory scale burner investigated at DLR Stuttgartby Lammel and coworkers [22]. Its jet-nozzle exitis arranged in an off-center position, resulting in astrong recirculation of the hot combustion prod-ucts, which stabilizes the flame and causes strongheat loss to the burner walls. We consider a con-figuration with values of 90 m/s, 0.71 and 573 Kfor the bulk inflow velocity, the equivalence ratioof the premixed methane/air mixture and the tem-perature at the inlet, respectively. This burner hasalso been investigated with RANS by Donini et al.[19], and with hybrid RANS/LES by Di Domenic-o et al. [23].

2. Modeling approach

In the PFGM approach [11,12], one dimen-sional freely propagating flames are computedwith a detailed chemical mechanism. The resultsare mapped over a small subset of control vari-ables and subsequently stored in a low dimen-sional lookup-table, which is accessed by theCFD solver. In the present work, the one dimen-sional flame computations are carried out withthe software library Cantera [24] for the GRI-3.0[25] mechanism. A unity Lewis number assump-tion for all species is used, which required theimplementation of an additional transport model

into Cantera. The reaction progress is describedby the species mass fraction sum Y C ¼Y CO2

þ Y CO þ Y H2O þ Y H2. We found that this

progress variable definition works well for meth-ane/air over the whole flammability range,although both simpler and more complexformulations exist.

2.1. Inclusion of heat loss in PFGM

To include the heat loss into the PFGM, thesum of sensible and chemical enthalpyh ¼ hs þ

PNk¼1Dh0

f ;kY k is used as second progressvariable. As mentioned above, different methodsare used to generate the non-adiabatic PFGMtable:

For the first method (M1) only the adiabaticfree flame without heat loss is computed. After-wards the gas temperature is successively reducedfrom the adiabatic to the ambient temperature.The gas composition, the laminar flame speed,the laminar flame thickness and the reaction rateare kept constant. The heat loss influences thesolution by reducing the temperature which altersthe density and the transport coefficients.

The second method (M2) relies on the compu-tation of burner stabilized flames [18]. At the inletof the domain, constant values are prescribed forthe mass flow and the temperature. By setting alower mass flow, a higher level of heat loss overthe entire flame is induced. Although the underly-ing assumption of temperature independence ofthe heat loss is unlikely to hold entirely in a realflame, this approach has been used with good suc-cess for the prediction of the flame behavior aswell as the inner flame structure by Fiorina et al.[18], Cecere et al. [20], Ketelheun et al. [21] andDonini et al. [19]. Outside the flammability region,the temperature is reduced corresponding to M1,this time not starting from the adiabatic flamebut from the burner-stabilized flamelet with themaximum heat loss. As it was found that theresults of the CFD simulations performed withinthis work were insensitive to the exact value ofthe flammability limit, it was assumed that it isreached when the flame speed falls below0.05 m/s.

The third method (M3) is based on introducingheat loss into a freely propagating flame. This isachieved by scaling the energy equation sourceterm due to chemical reaction by a constant factorð1� f LÞ over the whole flame. The modifiedenergy equation that was implemented in Canterareads:

ð1Þ

Fig. 1. Comparison of global flame properties andresulting flame structure for M2 and M3, the adiabaticsolution AD is shown as reference.

F. Proch, A.M. Kempf / Proceedings of the Combustion Institute 35 (2015) 3337–3345 3339

In Eq. (1), _m; cp; k; _xk ;W k and jk;z denote the massflow, heat capacity, thermal conductivity, reactionrate, molecular weight and the diffusive mass fluxof species k in z-direction, respectively. The sec-ond term on the RHS of Eq. (1) represents theenergy changes due to differential diffusion, thusit cancels out for the applied unity Lewis numberapproach. Starting with the adiabatic flame(f L ¼ 0), different flames with successively raisedvalues of f L are computed to cover the wholerange of enthalpy defects until the flame speedvalue of 0.05 m/s used as the flammability limitis reached at a value of f L ¼ 0:5. For higherenthalpy defects, the temperature is reduced asfor the first two methods, starting from the flam-elet at the flammability limit (f L ¼ 0:5). Thismodel results in a nearly linear relationshipbetween the heat loss and the temperature overthe flame. This promises to be a more suitableapproximation of the real physical behavior, sincethe strength of sources of heat loss as conductionalong the flame front and radiation is dependingon the temperature. Furthermore, the assumptionof a heat sink at the unburned side of the flame isavoided, which is somewhat questionable in anaerodynamically stabilized flame. Consequently,the flame trajectories no longer describe verticallines in the h-Y C diagram, which results in a mod-ified modeling strategy than for M1 and M2. Theterm f L

PKk1

hk _xkW k needs to be included as sinkterm in the energy equation, it is stored as anadditional quantity inside the PFGM table. Toimprove the resolution and simplify the lookupwithin the CFD code, the two control variablesh and Y C are normalized:

hN ¼h� hminðY CÞ

hmaxðY CÞ � hminðY CÞð2Þ

C ¼ Y C � Y minC ðhÞ

Y maxC ðhÞ � Y min

C ðhÞð3Þ

The resulting manifold is then mapped onto anequidistant hN -C grid by bilinear interpolation,the minimum and maximum values of hðY CÞ andY CðhN Þ are also mapped in the same way andstored for usage inside the CFD simulation.

A comparison of the laminar flame speed andflame thickness as a function of the heat loss forM2 and M3 is shown in Fig. 1, the adiabatic solu-tion (AD) is shown as reference. Also included is acomparison of the resulting flame structure for therespective flame with maximum heat loss. Thelaminar flame speed s0

l is reduced more by the heatloss for M2 than for M3, therefore the necessaryheat loss to reach the minimum flame speed isabout 10% larger for M3. The laminar flamethickness d0

l is growing stronger with heat lossfor M3 than for M2, most obvious towards theextinction limit. As already mentioned, the heatloss is increasing almost linearly with the temper-ature for M3, whereas it stays constant for M2.

The flame structure of the respective flame withmaximum heat loss for M2 and M3 is compared tothe adiabatic case AD in the plots on the right sideof Fig. 1. The temperature dependent heat loss forM3 results in a smooth reduction of the enthalpyover the flame. The temperature profile for M3has longer preheating and oxidation zones and asignificantly lower peak temperature compared toAD, nevertheless the general flame positionmarked by the point of inflection is maintained.For M2, the flame structure differs significantlyfrom AD and M3, the temperature increases toits final value that is comparable to M3 rapidlyat the beginning of the domain and stays constantafterwards. The behavior in temperature is basi-cally also mirrored by the mass fraction of CO2,with the exception that the final value is notaffected by the heat loss and is identical for allthree cases. On the unburned side of the flame, atambient temperature, M2 predicts a significantmass fraction of CO2 in contrast to M3 and AD.

2.2. ATF model

The typical thickness of a premixed flame (0.1–0.5 mm) is not properly resolved on practicallyaffordable LES grids with cell sizes bigger than0.5 mm. To tackle this problem, the ATFapproach artificially thickens the flame by intro-ducing a thickening factor F into the transportequations for enthalpy and progress variable:

@�q~h@tþ @

@xi�q~ui

~h� �

¼ @

@xiFE

kcpþ 1�Xð Þ lt

Sct

� �@~h@xi

!þE

F_xh ð4Þ

@�q~Y C

@tþ @

@xi�q~ui

~Y C

� �¼ @

@xiFE�qDCþ 1�Xð Þ lt

Sct

� �@~Y C

@xi

� �þE

F_xC ð5Þ

In Eqs. (4) and (5), q; ui; k; cp;Dc; _xC; lt and Sct

represent the fluid density, flow velocity, thermal

Fig. 2. Geometrical burner setup and boundary condi-tions, the computational domain is shown in Fig. 3.

3340 F. Proch, A.M. Kempf / Proceedings of the Combustion Institute 35 (2015) 3337–3345

conductivity, heat capacity, progress variable dif-fusion coefficient, progress variable reaction rate,turbulent viscosity and turbulent Schmidt num-ber, respectively. The source term for the enthalpy_xh represents the heat loss for M3 as describedabove, and becomes zero for M1 respectivelyM2. To avoid unphysical effects of the thickeningprocedure in regions without combustion, F isonly applied inside the flame region characterizedby high gradients of the progress variable [26].The flame region is detected with the flame sensorX, which is computed from the dimensionless pro-gress variable gradient of the one-dimensionalCantera flame computations [27]:

XðC; hÞ ¼dY C ðxÞ

dx

max dY CðxÞdx

� 24 35

1�D

ð6Þ

The actual value of the thickening factor is thencomputed from:

F ¼ 1þ X F max � 1ð Þ ð7Þwith

F max ¼ maxnDmesh

d0l

; 1

!ð8Þ

In Eq. (8), Dmesh represents the mesh cell size and nthe number of grid points on which the flamethickness is resolved, it is set to a value of 5 fol-lowing a suggestion by Charlette et al. [10]. Theflame thickness is computed based on the maxi-mum gradient of the temperature profile fromthe one-dimensional computations, T b and T u

denote the temperature on the burned andunburned side, respectively:

d0l ¼

T b � T u

max dTdx

� � ð9Þ

The effect of the velocity fluctuations on the subfil-ter level is modeled with the efficiency function E,which is evaluated from the analytical formulationof Charlette et al. [10], which is used with the cor-rections suggested by Wang et al. [28]. The respec-tive modeling constant was set to the commonlyused value of b ¼ 0:5 [10]. The detailed formula-tion of the model has been omitted for brevity, itcan be found in the available literature [10,27,28].

The influence of radiation was neglected, assimulations with a simple radiation model showedthat the related effects are one order of magnitudesmaller than those of the convective heat transfer.

3. Experimental and numerical configuration

3.1. Experiment

The high-velocity preheated and premixedcombustor has a rectangular cross section with a

width of 5 and a depth of 4 nozzle diameters,the jet nozzle is mounted in an off-center position.The walls are made from Quartz to provide opti-cal access. The geometrical setup and the bound-ary conditions are shown in Fig. 2. Velocityfields have been obtained from PIV measure-ments; profiles of species mole fractions and tem-perature were determined from Raman scatteringmeasurements by Lammel et al. [22].

The wall temperatures have not been measuredduring the experiment. However, according to O.Lammel the quartz glass surface would melt atapproximately 1000 �C, crystallization of the glassstarts at around 650 �C. Based on the agingbehavior of the glass, the wall temperature wasestimated in-between 650 �C and 800 �C (O. Lam-mel, personal communication, November 2013),so we have rounded the temperature to 1000 K.

3.2. Numerical setup and CFD-solver

Computations are carried out with the in-house LES-solver ’PsiPhi’ [16,27,29,30]. Solvedare the Favre-filtered governing equations formass, momentum, progress variable and enthalpyin low-Mach number formulation. Time integra-tion is performed with a low-storage third orderRunge–Kutta scheme, the parallelization is car-ried out using a distributed-memory messagepassing interface (MPI) domain decomposition.The finite volume method (FVM) is used to dis-cretize the equations on an equidistant Cartesiangrid. Convective fluxes are interpolated with acentral difference scheme for momentum, forscalar quantities and density a total variationdiminishing (TVD) scheme with the non-linearCHARM limiter [31] is applied. The geometry isdescribed by immersed boundaries.

The effect of subfilter transport on momentumand scalar fields is considered with the eddy-vis-cosity and eddy-diffusivity approach, respectively.The turbulent Schmidt number is adjusted to 0:7,the turbulent viscosity is computed with the r-model by Nicoud et al. [32], the respective model-ing constant is set to Cm ¼ Cr ¼ 1:5. This modelgives the correct decrease of the turbulent viscos-ity within the near wall region, avoiding the needfor a special wall treatment as required with theclassical Smagorinsky model. A turbulent velocityprofile with a mean value according to Fig. 2 is setat the inflow. Pseudo-turbulent fluctuations with a

Fig. 3. Instantaneous and mean contour plots of axialvelocity (left) and enthalpy loss (right) in a burner crosssection obtained on the fine grid (M3F). In the meanplots, white isolines denote zero velocity (left) respec-tively the flame sensor which marks the combustionregion (right).

F. Proch, A.M. Kempf / Proceedings of the Combustion Institute 35 (2015) 3337–3345 3341

length scale of 0.1 d and an intensity of 9 m/s aresuperimposed, they are generated using a versionof the filtering method of Klein et al. [33,34].The influence of the magnitude and the lengthscale of the fluctuations on the results was studiedand found to be very small. Zero gradient bound-ary conditions are adjusted at the outlet for allquantities, where clipping avoids entrainment dur-ing start-up. The temperatures of the walls arefixed at 1000 K. The sensitivity of the resultsagainst a variation of this temperature was foundto be negligible; flame position and length werenot affected. Based on the conditions at the inlet,the combustion process falls into the thin reactionzones regime of the modified Borghi diagram.

The computational domain has a length of24 d, the cell size is 0.1 d (nozzle diameterresolved by 10 cells) on the coarse and 0.025 d(nozzle diameter resolved by 40 cells) on the finegrid. This results in a total domain size of 38(0.6) million cells and a computational time of95,000 (280) CPU hours on the fine (coarse) grid.The sampling for the flow statistics was startedafter 10 flow-through times based on the bulkvelocity and performed for another 20 flow-through times to consider the slow but importantrecirculating zone on the “right” side of the jet.Tests on the coarse grid have shown that thisamount of sampling time is required and suffi-cient for accurate statistics.

Fig. 4. Radial profiles of the mean and rms of axialvelocity at different downstream locations.

4. Results and discussion

A first impression of the resulting flow andheat loss fields is given by Fig. 3, which showsinstantaneous and mean contour plots of the axialvelocity and the enthalpy loss for M3 on the finegrid (M3F). The structure of the axial velocityfield is dominated by a large recirculation zonedeveloping on the “right” side of the domain.Smaller recirculation zones develop at the “lowerleft” part of the domain and “in front of” as wellas “behind” the jet. The main jet starts breakingup after half a nozzle diameter, which induces tur-bulent fluctuations that are dissipated furtherdownstream. The mean jet is bend to the “right”near the end of the domain and feeds therecirculation zone.

The enthalpy defect is strongest in the lowerpart of the domain, where recirculated burnedgases have been cooled down by around 700 Kin comparison to the adiabatic solution MA dueto heat exchange with the chamber walls. Thereduction of the flame speed by the effect of heatloss results in local extinction near the burner exit.

Radial profiles of velocity and temperature [22]are presented in Figs. 4 and 5, comparing theresults for the adiabatic reference solution andthe models described above on the coarse (C)and fine (F) grid.

The axial velocity statistics in Fig. 4 are ingood agreement with the measurements for M2and M3, whereas AD and M1 struggle to predictthe velocity and the correct position of the recircu-lation zone. This trend is mirrored in the fluctua-tion profiles, which are initially under-predictedby AD and M1 and then dissipate too late. Nearthe nozzle exit, no significant differences betweenthe velocity predictions of the individual modelscan be observed. The velocity predictions donot improve significantly with grid refinement,implying sufficient grid resolution.

Fig. 5. Radial profiles of the mean and rms of temper-ature at different downstream locations.

Fig. 6. Strain correction factor for different amounts ofheat loss from Cantera premixed counterflow flames(symbols) and the respective exponential fit (line).

3342 F. Proch, A.M. Kempf / Proceedings of the Combustion Institute 35 (2015) 3337–3345

As a result of the significant amount of heatloss visible in Fig. 3, the peak temperatures forthe adiabatic simulation AD in Fig. 5 exceed themeasured ones by several hundred Kelvin. In con-trast to that, M1 matches the peak temperaturesof the experiment, but under-predicts the thick-ness of the flame. M2 and M3 basically matchthe mean temperature measurements, some devia-tions occur towards the “left” burner wall. Thepredictions here improve with M3 compared toM2 and with grid refinement. The latter can beexplained by the fact that the higher temperaturegradients in this region can be resolved more ade-quately on the fine grid. Except for AD, all othermethods are able to reproduce the temperaturefluctuations within the burned gas qualitatively.However, the strength of the fluctuations isunder-predicted near the burner exit. Physicallyit seems unlikely that the temperature fluctuationswithin the recirculation zone, which mainly con-sists of burned products, should be as high as inthe flame region.

4.1. Strain correction

Although the measured temperature field iscaptured well by the simulation, it turned out thatall methods under-predicted the length of theflame to a certain degree. To evaluate if this isrelated to the relatively high axial velocity magni-tude, the tabulation method M3 is extended by asimple strain correction (M3S) based on Canterapremixed counterflow flames [35]. One stream rep-resents the hot exhaust gas and the other one thecold unburned mixture from the respective freeflame computations. These computations havebeen carried out for different amounts of heat loss

and varying compressive strain rates, where theenergy equation was computed according to Eq.(1) and the flame speed was evaluated from:

s0L ¼

1

quY CO2 ;b

Z 1

�1_xCO2

dx ð10Þ

In Eq. (10), qu; Y CO2 ;b and _xCO2denote the density

of the unburned mixture, mass fraction of CO2 inthe exhaust gas and reaction rate of CO2, respec-tively. Figure 6 shows the resulting flame speed asa function of the maximum strain rate, the plothas been normalized by the flame speed of therespective free flame. The Favre-filtered strain inthe CFD computation is computed according to[1], with the flame normal vector ~ni pointing intothe fresh mixture:

S ¼ �~ni@ ~ui

@xj~nj with ~ni ¼ �

@eY C@xi

@eY C@xi

ð11Þ

The obtained strain value is then normalized bythe flame speed of the respective free flame, subse-quently the correction factor is evaluated from theexponential fit given in Fig. 6 and applied to theprogress variable reaction ratio, the flame speedand the flame sensor. As the basis of the fit arethe maximum strain rates from the counterflowcomputations and only the resolved strain rate istaken into account, the described procedure repre-sents a lower limit of the influence of the strain onthe flame propagation.

Figure 7 shows the comparison of mean valueand fluctuations of temperature and carbondioxide at the most downstream measurementlocation. The strain correction improves the pre-dictions notably, most visible towards the “left”side of the domain, the resulting flame length isin agreement with the measurements. However,the impact on the temperature profile seems tobe a little bit too high towards the wall, whichlikely indicates that the correction should not beapplied near the wall inside the boundary layer,where high strain rates may occur even in theabsence of burned (and cooled down) products.Although the use of the correction improved theresults significantly, it must be stressed that thedevelopment of this method should not be seenas finalized, and that further investigation on the

Fig. 7. Radial profiles of the mean and rms of temper-ature and carbon dioxide molar fraction at the lastmeasurement position.

F. Proch, A.M. Kempf / Proceedings of the Combustion Institute 35 (2015) 3337–3345 3343

effects of strain in such flames is required – ideallyby DNS. It also remains to be investigated if thestrain itself is the main physical reason for thereduction of the flame speed, or rather someclosely related phenomenon like internal exhaustgas recirculation, as suggested by Di Domenicoet al. [23]. Di Domenico et al. [23] also found noevidence for a relevant contribution of auto igni-tion on the flame stabilization – even for a fasterjet (bulk velocity of 150 m/s).

4.2. Analysis of the heat loss

Figure 8 compares the normalized PDF of thereaction source term conditional Favre-filteredenthalpy defect for the different models and gridresolutions, AD and M1C are skipped for brevity.In all cases combustion is found over a wideenthalpy range which starts at an enthalpy defectof approximately �0.2 MJ/kg on the fine and�0.4 MJ/kg on the coarse grids, respectively.The distributions show a negative skewness witha long tail towards low enthalpies. This can beattributed to the fact that the flame gets more

Fig. 8. Normalized PDF of the Favre-filtered reactionsource term conditional on Favre-filtered enthalpydefect over the entire domain for the presented models.Results are also shown for M3C with an additional top-hat (TH) subfilter model in enthalpy and progressvariable, respectively. Shown are the total and theresolved reaction source term, the latter is obtained withan efficiency function of unity (E ¼ 1 in Eq. (5)). Theratio of the integrated resolved source term and theintegrated total source term is also given. For M3SF thetotal source term after the strain correction is alsoincluded.

sensitive to small disturbances when approachingthe extinction limit, resulting in a stronger influ-ence of the turbulent velocity fluctuations on theflame behavior. To study the influence of the sub-filter contribution, two additional simulationshave been carried out for M3C with an additionaltop-hat FDF closure in enthalpy respectively pro-gress variable [27,36,37]. The resulting impact onthe source term PDFs (and also on the flow statis-tics) is very small, the amount of resolved sourceterm is approximately identical. M2C predicts asmaller amount of heat release at the decreasingside of the distribution compared to M3C, whichresults in a lower amount of resolved source term.In the fine grid simulation M3F around 95% ofthe flame is resolved, the peak value of heatrelease is shifted towards lower enthalpies com-pared to the coarse grid simulations. The fine gridpredicts a larger amount of combustion towardsthe extinction limit than the coarse one. An expla-nation for this behavior is that most of the flamesnear the extinction limit are found in the near-wallregion at the “left” side of the burner, which isresolved better on the fine grid. The efficiencyfunction E reduces to unity at low enthalpy valuesdue to the increase of flame thickness shown inFig. 1, indicated by matching of the resolvedand total source term. The strain correction(M3SF) reduces the skewness and width of thedistribution noticeably, which can be explainedby flame blow off near the flammability limit bythe effect of mean flow strain rate. Thus the prob-ability for re-ignition of these flames is reduceddrastically, causing a shift of the combustion pro-cess to more stable regions with higher enthalpyvalues.

5. Conclusions

A high velocity confined jet burner has beeninvestigated with different methods for the inclu-sion of heat loss in PFGM. Considering the heatloss was necessary as the adiabatic reference sim-ulation (AD) over-predicted the temperature byseveral hundred Kelvin and failed to predict theflow field correctly. The combination of an adia-batic combustion model with the solution of theenergy equation (M1) was able to predict the cor-rect peak values of the temperature, but neitherthe correct flame shape and length nor the correctvelocity field.

To capture the shape of the flame and the cor-rect velocity field, it was necessary to consider theeffect of the heat loss on the flame structure andpropagation velocity. Two methods based onCantera computations with detailed chemistryhave been compared, an established one basedon burner stabilized flames (M2) and one basedon scaling of the energy equation source term(M3). Both methods performed well and lead to

3344 F. Proch, A.M. Kempf / Proceedings of the Combustion Institute 35 (2015) 3337–3345

very comparable results. The new method yieldedslight improvements in the predictions oftemperature towards the “left” burner wall andof velocity further downstream. Only a smallimprovement of the results with grid refinementwas found (M3F), mostly visible for the tempera-ture due to better resolution of the wall boundarylayer. The predictions on the coarse grid showed avery satisfactory agreement with the measure-ments for M2 and M3.

Even though the last two models were able topredict the flame structure and the flow field, theystill under-predicted the flame length by somedegree. It was shown that inclusion of a simplestrain correction method (M3SF) improves theprediction of the flame length. However, the exactformulation of this model and the details of theunderlying physical mechanism require furtherinvestigation.

By analyzing and comparing the PDF of thesource term conditional on the heat loss for thedifferent models, it was shown that the combus-tion process takes place over a wide range ofenthalpy defect. The probability of combustiontowards the extinction limit is increased on thefine grid, implying that the respective parts ofthe flame are mainly located near to the “left” wallof the burner, where the effect of the gridrefinement is strongest.

Acknowledgments

The authors gratefully acknowledge fundingfrom the state of Nordrhein-Westfalen, Germany.Computations have been carried out using theDFG supported HPC resources of the Centerfor Computational Sciences and Simulation(CCSS) of the University of Duisburg-Essen,Germany. We would like to thank DLR Stuttgartand Siemens Energy for providing the experimen-tal data. We further thank Dr. Wolfgang Meier,Dr. Oliver Lammel and Stefan Dederichs formany helpful discussions.

References

[1] N. Peters, Turbulent Combustion, Cambridge Uni-versity Press, 2000.

[2] H. Pitsch, Combust. Flame 143 (2005) 587–598.[3] M. Dusing, A.M. Kempf, F. Flemming, A. Sadiki,

J. Janicka, Prog. Comput. Fluid Dyn. 5 (2005)363–374.

[4] C. Angelberger, D. Veynante, F. Egolfopoulos, T.Poinsot, Proceedings of the Summer Programm,Center for Turbulence Research, Stanford, 1998,pp. 61–82.

[5] T. Ma, O. Stein, N. Chakraborty, A.M. Kempf,Combust. Theory Model. 17 (2013) 431–482.

[6] M. Boger, D. Veynante, H. Boughanem, A. Trouve,Proc. Combust. Inst. 27 (1998) 917–925.

[7] J. Galpin, A. Naudin, L. Vervisch, C. Angelberger,O. Colin, P. Domingo, Combust. Flame 155 (2008)247–266.

[8] B. Fiorina, R. Vicquelin, P. Auzillon, N. Darabiha,O. Gicquel, D. Veynante, Combust. Flame 157(2010) 465–475.

[9] O. Colin, F. Ducros, D. Veynante, T. Poinsot,Phys. Fluids 12 (2000) 1843–1863.

[10] F. Charlette, C. Meneveau, D. Veynante, Combust.Flame 131 (2002) 159–180.

[11] J.A. van Oijen, L.P.H. de Goey, Combust. Sci.Technol. 161 (2000) 113–137.

[12] J.A. van Oijen, R.J.M. Bastiaans, L.P.H. de Goey,Proc. Combust. Inst. 31 (2007) 1377–1384.

[13] G. Kuenne, A. Ketelheun, J. Janicka, Combust.Flame 158 (2011) 1750–1767.

[14] B. Marracino, D. Lentini, Combust. Sci. Technol.128 (1997) 23–48.

[15] M. Ihme, H. Pitsch, Phys. Fluids 20 (2008) 055110.[16] B.M. Franchetti, F.C. Marincola, S. Navarro-

Martinez, A.M. Kempf, Proc. Combust. Inst. 34(2013) 2419–2426.

[17] P. Schmitt, T. Poinsot, B. Schuermans, K. Geigle,J. Fluid Mech. 570 (2007) 17–46.

[18] B. Fiorina, R. Baron, O. Gicquel, D. Thevenin, S.Carpentier, N. Darabiha, Combust. Theory Model.7 (2003) 449–470.

[19] A. Donini, S.M. Martin, R.J.M. Bastiaans, J.A. vanOijen, L.P.H. de Goey, Proceedings of the ASMETurbo Expo 2013.

[20] D. Cecere, E. Giacomazzi, F. Picchia, N. Arcidiac-ono, F. Donato, R. Verzicco, Flow Turbul. Com-bust. 86 (2011) 667–688.

[21] A. Ketelheun, G. Kuenne, J. Janicka, Flow Turbul.Combust. 91 (2013) 867–893.

[22] O. Lammel, M. Stohr, P. Kutne, C. Dem, W.Meier, M. Aigner, J. Eng. Gas Turbul. Power 134(2012), 041506-041506.

[23] M. Di Domenico, P. Gerlinger, B. Noll, Proceed-ings of the ASME Turbo Expo 2011.

[24] D.G. Goodwin, Cantera: an object-oriented soft-ware toolkit for chemical kinetics, thermodynamics,and transport processes, <http://code.google.com/p/cantera>, 2009.

[25] G.P. Smith, D.M. Golden, M. Frenklach et al.,http://www.me.berkeley.edu/gri_mech, 2000.

[26] J.P. Legier, T. Poinsot, D. Veynante, Proceedings ofthe Summer Programm, Center for TurbulenceResearch, Stanford, 2000, pp. 157–168.

[27] F. Proch, A.M. Kempf, Combust. Flame (2014)<http://dx.doi.org/10.1016/j.combustflame.2014.04.010>.

[28] G. Wang, M. Boileau, D. Veynante, Combust.Flame 158 (2011) 2199–2213.

[29] F.C. Marincola, T. Ma, A.M. Kempf, Proc. Com-bust. Inst. 34 (2013) 1307–1315.

[30] M. Pettit, B. Coriton, A. Gomez, A.M. Kempf,Proc. Combust. Inst. 33 (2011) 1391–1399.

[31] G. Zhou, Numerical Simulations of Physical Dis-continuities in Single and Multi-fluid Flows forArbitrary Mach Numbers, Ph.D. Thesis, ChalmersUniversity of Technology, Goteborg, Sweden, 1995.

[32] F. Nicoud, H.B. Toda, O. Cabrit, S. Bose, J. Lee,Phys. Fluids 23 (2011) 085106.

F. Proch, A.M. Kempf / Proceedings of the Combustion Institute 35 (2015) 3337–3345 3345

[33] M. Klein, A. Sadiki, J. Janicka, J. Comput. Phys.186 (2003) 652–665.

[34] A.M. Kempf, S. Wysocki, M. Pettit, Comput. Fluids60 (2012) 58–60.

[35] L. Tay Wo Chong, T. Komarek, M. Zellhuber, J.Lenz, C. Hirsch, W. Polifke, Proceedings of theEuropean Combustion Meeting 4 (2009).

[36] J. Floyd, A.M. Kempf, A. Kronenburg, R.H. Ram,Combust. Theory Model. 13 (2009) 559–588.

[37] C. Olbricht, O.T. Stein, J. Janicka, J.A. van Oijen,S. Wysocki, A.M. Kempf, Fuel 96 (2012) 100–107.