A study of three-dimensional LES of turbulent combustion with radiative heat transfer

Abstract

Combustion, one of the main forms of energy production nowadays, is a complex phenomenon in which turbulent flow, chemical reactions, different phases and heat transfer phenomena can interact. A better understanding of these interactions is essential to improve existing combustion systems and develop new ones. The aim of this work was to study the interaction of turbulent combustion with thermal radiation with the aid of three-dimensional numerical simulation. Using the CORBA computational tool, a code for Large Eddy Simulation (LES) of combustion was coupled with a radiative heat transfer code. This technique allows information to be exchanged between the two codes without major changes in their structure, so that advantage can be taken of the different characteristic time of each phenomenon in a high-performance parallel-computing environment. This study is the continuation of an earlier two-dimensional study. The results show a change in the temperature and species fields as well as in the flame dynamics when thermal radiation is taken into account.

Introduction

Air pollution is one of the main problems facing society nowadays. It is responsible not only for local effects, such as the damage it causes to people’s health in large cities, but also for global changes, including global warming (caused by greenhouse gases). Combustion processes are the main contributors to this problem, as around 90 % of the world’s energy is produced by these processes, which are responsible for emissions of CO, NO x , soots, polycyclic aromatic hydrocarbon (PAH) and CO2, the gases associated with global warming.

To face this challenge, new standards or agreements, such as the Kyoto protocol, which was ratified by more than 170 countries, seek to achieve the “stabilization of greenhouse gas concentrations in the atmosphere at a level that would prevent dangerous anthropogenic interference with the climate system” by a global reduction in greenhouse gas emissions, especially in developed countries. Cleaner combustion processes are therefore required in a wide range of industries. To achieve this, a detailed knowledge of combustion processes and new tools to help design and make improvements to combustion systems are essential.

Advances in combustion science require a better understanding of all the physical processes that control a flame: mixing processes, turbulence, chemistry, radiative heat transfer, etc.

This work focuses on radiative heat transfer in combustion, a phenomenon is of crucial importance, as it:

  • controls charge heating in furnaces;

  • is a key mechanism in thermal heat losses and wall heat fluxes;

  • causes minor pollutant species to form;

  • interacts with soot;

  • controls the propagation of large-scale fires, such as pool fires and forest fires.

Combustion and radiation are very different phenomena. Combustion is controlled by local exchanges and is numerically described through balance over small volumes (finite-volume framework), whereas radiative heat transfer involves long-distance interaction. Hence, taking radiation into account in numerical simulations of combustion leads to various difficulties, such as the very expensive CPU-time requirements of radiation solvers or the need to know spatially and time-resolved information about the reacting flow field, which is usually not available in Reynolds-averaged Navier–Stokes (RANS) approaches. Notwithstanding all these difficulties, the interaction between turbulence and radiation exists and can have a significant effect on flames [7].

Several authors have already attempted to perform numerical simulations of combustion while taking into account radiative heat transfer, but because of limited computational resources were unable to integrate highly accurate physical models of the aerodynamics, combustion chemistry and radiation [36, 11, 13, 27, 41, 47]. For this reason radiation has long been neglected or treated with simple models in numerical simulations of turbulent reacting flows.

Many studies focus on the interaction between turbulence and radiation in combustion, and the SANDIA methane / air turbulent jet diffusion flame, Flame D [1], has been used to validate a range of studies [5, 6, 11, 27]. The simulations were all two-dimensional but used different approaches both for the turbulence / chemistry interactions (presumed [11] and transported [27] probability density functions and the Reynolds stress method [5]) and the radiation model (the discrete ordinates method (DOM) [5, 6], ray tracing [5] and the discrete transfer method (DTM) [11]). Three-dimensional simulations can also be performed, but in general, because of limited computer resources, a standard turbulence model is combined with a P1-Gray or P1-FSK model for radiation as in [47], where a three-dimensional simulation of an oxygen-enriched turbulent non-premixed jet flame is described.

Silva et al. [41] performed a two-dimensional numerical RANS simulation of turbulent non-premixed combustion of methane and air in a cylindrical chamber where thermal radiation exchanges in the combustion chamber were computed by the zonal method, and the gas absorption coefficient dependence on wavelength was resolved using the weighted sum of gray-gases model. The results show that the greatest effect exerted by thermal radiation was on heat transfer and the temperature field. In [4] the same case was studied taking into account the radiation effect of non-gray gases with two different WSGG models, and DOM was used to solve the radiative transfer equation. The authors showed that modeling radiative heat transfer and radiative properties is very important in combustion predictions. In [3] the same authors included turbulence radiation interactions (TRI) in their analysis of the same combustion chamber and concluded that the inclusion of TRI yielded results even closer to the experimental data, although the effect of including radiation (even without TRI) was more important. Hartick et al. [13] used a 20 kW axisymmetric confined swirling combustion system to study the complex phenomenon of TRI. They proposed a new approach to confined diffusion flames and concluded that TRI can be simulated efficiently without the use of tedious stochastic approaches.

Compared with RANS, unsteady simulations, such as direct numerical simulations (DNS) or large eddy simulations (LES), give access to the instantaneous spatial distribution of fresh and burnt gases, key information for radiation models [37]. Direct numerical simulations were combined with DOM [50] and DTM [48] to investigate two-dimensional sooting flames in fires. Wu et al. [49] implemented a photon Monte Carlo method for solving the radiative transfer equation in a turbulent-combustion DNS code to study the turbulence/radiation interaction, and Zhang et al. [52] used DNS and the Monte Carlo transfer method to study the effects of radiation on the boundary layer structure in a turbulent channel flow. The combination of LES and DOM was first used by Desjardin and Frankel [9] to simulate a two-dimensional non-premixed acetylene-air planar jet flame including soot formation and radiation. The same approach was used for three-dimensional simulation of a gas turbine [18] and a combustion chamber [31, 32].

In the present work, the new methodology initially described in the thesis of Lecanu [25] and used in [38] to perform LES of turbulent-combustion flows with radiative heat transfer is used to couple a three-dimensional LES combustion code with a three-dimensional radiative heat transfer code and produce results in a complex geometry with a mesh containing more than 4.7 million cells.

Our approach is based on two dedicated, independent solvers linked through a specialized framework known as CORBA [14] and takes advantage of the different characteristic time of each phenomenon. Highly efficient solvers can then be used for each phenomenon to take advantage of new massively parallel supercomputers. In this study, we used the AVBP LES solver developed by CERFACS and IFP (see, for example, [40]) and the radiation solver based on the parallel three-dimensional DOM code known as DOMASIUM [17, 1921] developed by the RAPSODEE research center at the École des Mines d’Albi-Carmaux.

The goal here was to study the interactions in turbulent combustion with and without thermal radiation by three-dimensional numerical simulation using detailed models for combustion and thermal radiation and to compare the results with experimental findings.

The subgrid scales for turbulence radiation interactions were not used. This subject has been studied by various authors, and more detail can be found in [35, 42]. The experimental setup is the same as that used by [31, 32, 51] and also by the present authors in a previous study [38] in which two-dimensional simulation was used.

First, the combustion and radiation code are described in Sect. 2. Then the details of the numerical modeling (mesh, boundary conditions and combustion and radiation models) are explained in Sects. 3 and 4. This is followed by Sect. 5, in which comparisons are made between coupled cases, non-coupled cases and experimental data. Finally, some conclusions are drawn in Sect. 6.

Combustion and radiative heat transfer solvers

Combustion solver: AVBP

The AVBP LES solver developed by CERFACS and IFP is used here to solve the three-dimensional compressible equations for momentum, mass-fraction species and energy on hybrid (structured and unstructured) meshes [40].

The flow solver is based on the finite-volume and finite-element cell-vertex methods, with an artificial viscosity sensor and explicit time integration.

The Lax–Wendroff scheme (a finite-volume second-order Runge–Kutta time-integration and central second-order spatial-discretization scheme) [15] was used for the simulation in this work, since the aim was to investigate the feasibility of LES of reacting flows including thermal radiation.

The discretization schemes are combined with Navier–Stokes characteristic boundary conditions (NSCBC) [30] and wall-law boundary conditions [39]. The code is parallelized by domain splitting using the MPI library, a convenient approach when solving balances over small volumes. AVBP has been proved to be very efficient on massively parallel machines, reaching a speedup factor of 4,078 (i.e., running 4,078 times faster than on one processor) when using 4,096 processors in an IBM BlueGene/L machine [44].

The time step is determined by the minimum value between the convective and diffusive time step when using the thickening flame model (the chemical time step is only considered for flamelet models, which is not our case) [33]. The convective time step is calculated using the Courant–Friedrichs–Lewy condition (CFL), a stability criterion related to the time required for an acoustic wave to go from one node to the following one at a speed \(| \ \overrightarrow{u} \ | + c\) (\(c\) is the sound speed). The diffusive time step is calculated using a Fourier criterion and is proportional to the diffusion time required to cross a cell.

Radiative heat transfer solver: DOMASIUM code

DOMASIUM (Discrete Ordinates Method Applied with Spectral Integration on Unstructured Meshes) [17, 1921], the radiation solver used in this work, was developed by the RAPSODEE research center at Ecole des Mines d’Albi-Carmaux to solve the radiative transfer equation on complex three-dimensional meshes using DOM. Much effort was spent optimizing the performance of the code, particularly in terms of computational time and memory usage, as these were identified as critical issues at the beginning of the study.

The radiative properties of the gas are described by SNB-CK (statistical narrow-band cumulative-k) or CK [43] models, and the DOMASIUM code is parallelized based on the spectral band and direction of integration. The technique of band number reduction was used in the coupled calculations and yielded a significant reduction in computational time.

As part of this study, tools were specially developed so that the same grid files used with the AVBP LES code could be used with DOMASIUM, which was originally written for unstructured grids using tetrahedral cells. Temperature and mass-fraction fields were taken directly from an AVBP results file. A new option was added to DOMASIUM to enable it to receive the data field from a given AVBP mesh and, with the use of a connectivity table, perform the radiation heat transfer on a coarser mesh and project the final result onto the original mesh.

Different kinds of angular quadrature can be used, especially \(S_N\) [2], \(T_N\) [46] and [28]. The \(S_4\) quadrature scheme, which has 24 directions and is one of the most popular schemes because of its computational efficiency, was chosen here. Specially for this work, the DOMASIUM code was also parallelized based on the direction of integration, which is the best choice for codes that use DOM according to Gonçalves and Coelho [12].

Special care was taken with the ordering of data for each direction resulting from the optimization of the sweep order. Use of the reordered data as a function of the pathway leads to improved code performance as a result of optimal use of the machine’s cache memory. The main increase in performance is in the loops over all the mesh cells. The sweeping order forces the machine to look for the information outside the memory cache for each step in the loop. By reordering the data set, the machine searches for the information in a high-order memory cache, which can be more than eight times faster than searching for the same information in RAM memory, for example, thus optimizing the calculation time. The same kind of strategy was used by, among others, Elias and Coutinho [10].

The radiative transfer equation was discretized using a finite-volume approach, and three different spatial differencing schemes were considered: “exponential” [36], “step” [28] and “diamond mean flux” [45]. The “diamond mean flux” method was chosen for the simulations.

The parallelization of the code was done using MPI, and very good scalability was achieved. We managed to use more than 300 processors in the coupled simulations without degrading the performance because of the exchange of information between processors. Memory optimization allows the use of meshes with more than 3 million cells in machines such as the Zahir IBM SP4 supercomputer at IDRIS (France) or the Linux-based cluster at the EM2C laboratory. These mesh sizes are compatible with meshes used in LES of combustion.

Experimental setup and configuration used in the numerical simulations

The geometry used in the numerical simulations is an experimental configuration investigated by Knikker et al. [2224] and is shown in Figs. 1 and 2. This is the same experimental setup used in previous two-dimensional simulations [38]. A premixed propane/air flow is injected into a rectangular combustor. This simple geometry is a first step toward modeling and understanding, for example, the combustion inside a small helicopter gas turbine. The height, depth and length of the combustion chamber are 50, 80 and 300 mm, respectively. The lateral walls are transparent artificial quartz windows that allow the inside of the whole chamber to be viewed. The upper and lower walls are made of thick ceramic material for thermal isolation and include two narrow windows used to introduce laser sheets. A stainless steel triangular flame holder (height 25 mm), corresponding to a 50 % blockage ratio, is embedded in the lateral windows. A V-shaped turbulent flame stabilized by the hot gases recirculating behind the flame holder is investigated. Among the various cases available in [23, 24], the case of a fixed propane/air flow rate of 20 g s−1, corresponding to an upstream mean velocity of about 5 m s−1 (turbulence level about 5 %) and an equivalence ratio \(\phi\) = 1.0, was chosen for simulation.

Fig. 1
figure1

Experimental setup used for numerical simulations. A turbulent premixed propane/air flame is stabilized downstream of a triangular-shaped flame holder

Fig. 2
figure2

Picture of the experimental setup shown schematically in Fig. 1

The computational domain (Fig. 3) starts 10 cm upstream of the flame holder and continues until 60 cm downstream. A coflow with a coarse grid was added after the combustion chamber to reduce the influence of the outlet boundary condition on the combustion results.

Fig. 3
figure3

Three-dimensional mesh with 4.7 million tetrahedrons used by AVBP

The LES mesh contains about 4.7 million cells (tetrahedrons), and the grid size downstream of the flame holder is \(\Delta\) = 1.0 mm until 15 cm downstream (Fig. 4), when it increases. In the following simulations, the AVBP time step is about \(\Delta t_{\rm {LES}} \approx\) 0.19 μs. The radiation code uses a different mesh to reduce the computational time and machine memory required. Using a connectivity table, the physical properties (CO2 and H2O mass fractions and temperature field from AVBP, and radiation power from DOMASIUM) can be transferred from one mesh to another in both directions. The radiation mesh contains about 3.3 million cells (tetrahedrons). The grid size of both meshes downstream of the flame holder is the same until 12 cm downstream, where the recirculation zone is. This should be well described because it is responsible for flame stabilization. Upstream of the flame holder and in the coflow region, a coarser grid can be used as the power radiation is small with no large variations.

Fig. 4
figure4

Detail of the three-dimensional mesh used by AVBP near the flame holder

The NSCBC were used in the combustion code to impose both the static pressure at the outlet and the velocity components, static temperature and species mass fractions at the inlets smoothly. Consequently, the ingoing waves are calculated as being proportional to the difference between the actual state at the boundary nodes and the set reference. There are two inlets to the domain; the first is the inlet for the premixed propane/air mixture upstream of the flame holder (equivalent ratio \(\phi\) = 1.0; velocity = 5 m s−1 and T 2 = 300 K), and the second is the coflow inlet, where nitrogen is injected at 22 m s−1 and 1,900 K.

A boundary condition defined as a wall with heat loss using a wall-function approach with zero normal velocity (slip) is imposed in the combustion code; the thermal resistance is estimated to be \(R_{\rm th}\) = 0.096 Km2 W−1 in the upper and lower ceramic walls and \(R_{\rm th}\) = 0.086 Km2 W−1 in the lateral artificial quartz walls. A thermal resistance equal to \(R_{\rm th}\) = 120 Km2 W−1 is used for the triangular-shaped aluminum flame holder. All the other walls are assumed to be adiabatic slip walls.

In the radiation code, an emissivity \(\varepsilon\) is specified for these walls. This is assumed to be constant for the ceramic walls (\(\varepsilon = 0.91\)), while the real emissivity of the material as a function of wavelength is used for the quartz walls (semitransparent at short wavelengths becoming opaque at longer wavelengths, Fig. 5). According to [16], typical stainless steels lightly oxidized at 1,000 K have an emissivity of \(\varepsilon = 0.4\). This figure is used for the triangular-shaped flame holder, while the coflow walls are assumed to have an emissivity of \(\varepsilon = 0.7\). The other walls are assumed to be cold black bodies with \(\varepsilon =1\), i.e., all rays reaching the walls are absorbed (no reflection and no transmission). The radiation fluxes do not heat the walls and transmission through the walls is neglected.

Fig. 5
figure5

Emissivity of clear GE 124 fused quartz as a function of wavelength

Combustion LES and radiation modeling

Unresolved turbulent fluxes are modeled here using the classical Smagorinsky model based on a subgrid-scale viscosity [29]. Turbulent premixed combustion is described by the dynamically thickened flame model [8, 26] (DTFLES). In this formulation, the thickening factor \(F\) is not a constant but increases to \(F_{\rm {max}}\) in flame zones and decreases to unity in non-reactive zones. This model is more suitable for calculating the radiative power than the thickened flame model (TFLES), since it does not change the species or thermal diffusion outside the flame region.

The chemical reaction between propane and air is represented by a one-step global chemical mechanism [40]:

$$\begin{aligned} {\rm C_3H_8 + 5 O_2} \Longrightarrow {\rm 3CO_2 + 4H_2O} \end{aligned}$$
(1)

The reaction rate is given by:

$$\begin{aligned} {\dot{\omega }}_1 = {A [{\rm C_3H_8}]}^{\alpha }[{\rm O}_2]^{\beta } \exp (-E_{a}/{{\rm RT}}) \end{aligned}$$
(2)

The chemical parameters used are summarized in Table 1. These values lead to a laminar flame speed 48 m s−1 for an equivalent ratio \(\phi = 1.0\). Again, this combustion model is quite simple but is appropriate here since only perfectly premixed combustion is considered.

Table 1 Chemical parameter for reaction rates

In the version of AVBP used here, the chemical scheme that describes the methane-air or propane-air chemistry uses constant but different Schmidt numbers for each species so that the laminar flame speed can be reproduced as accurately as possible. These different Schmidt numbers introduce preferential diffusion in the flame front, potentially leading to large variations in the CO2 to H2O ratio, which should be almost constant (Fig. 6). This effect is greatly amplified when the Thickened Flame Model is used. Figure 7 shows the CO2 to H2O ratio for the dihedral flame when different Schmidt numbers are used. While in the burned gases the ratio is accurately reproduced, in the front behind the flame holder the ratio is about 1.4, which is unrealistic. While this effect may be negligible when focusing on flame dynamics, it leads to a large error when dealing with radiation, as CO2 and H2O are the two major absorbing species in the flame and the ratio between them is crucial when estimating the radiative power of the flame accurately.

Fig. 6
figure6

CO2 mass fraction vs H2O mass fraction in the dihedral flame using different Schmidt numbers (top) and the same Schmidt numbers (bottom)

Fig. 7
figure7

Ratio of CO2 mass fraction to H2O mass fraction in the dihedral flame using the different Schmidt numbers

Hence, when dealing with radiative heat transfer, the Schmidt numbers must be set to the same value for all the species, as was done here.

In the radiation code the CK distribution method was used with a reduced number of spectral bands (28 for H2O and 8 for CO2) corresponding to the wavelengths that make the most important contributions to the global radiative power. As mentioned before, the Diamond Mean Flux Scheme (DMFS) and a fourth-order quadrature were adopted, resulting in 24 spectral directions [17, 21]. According to [19], these models are the best compromise between calculation time and accuracy. Radiation heat transfer is assumed to be controlled by resolved-scale motions only (and the burnt gas zones), and no subgrid scale radiation model is taken into account.

Results and discussion

First the results with and without radiative heat transfer are compared, and these are then compared with the experimental data.

Calculations without radiative heat transfer using only the AVBP code were carried out to provide a reference case (identified as NR in the following discussion). A previous two-dimensional study [38] showed that updating the radiative source term every 100 interactions is a good compromise between accuracy and computational time. This coupling frequency was also used for the three-dimensional simulations. Correct load balancing is then achieved by devoting 50 % of the processors to the AVBP LES code and 50 % to the DOMASIUM radiative heat transfer code. Typical runs were conducted using 192 processors (96 for each code).

Coupled simulations were started at \(t_0\) = 0.152 s of a computation without radiative heat transfer, after which simulations with and without radiation were conducted. This procedure allows both flow fields to be compared at the same instant. The time-averaged results used specially for comparison with the experimental data were calculated based on 760,000 interactions (\(\sim\)0.144 s) or about 8.6 characteristic periods of the flow.

The influence of radiation in the three-dimensional combustion simulation

In this section, we will present the coupled combustion and radiation results for the three-dimensional simulations and compare the cases without and with radiative heat transfer. Figure 8 shows instantaneous isosurfaces of temperature without and with radiative heat transfer at the same physical time \(t_2\) = 0.28 s. With radiation, the flow structures are modified and the turbulence structures appear to increase several centimeters downstream of the flame holder. This can also be observed in Figs. 9 and 10, which show temperature cuts in the center plane (z = 0) and symmetrical side planes (z = −20 mm and z = 20 mm), respectively. The temperature distribution changes if radiation is taken into account, when maximum values are lower. The highest temperature peak is found in the central plane of the burner.

Fig. 8
figure8

Instantaneous resolved isosurfaces of temperature (900 K) without (top)/with (bottom) radiative heat transfer colored according to transverse velocity v at \(t_2\) = 0.28 s

Fig. 9
figure9

Cuts in the center plane at z = 0 of instantaneous resolved temperature field without (NR, top) and with (R100, bottom) radiative heat transfer at \(t_2\) = 0.28 s

Fig. 10
figure10

Cuts in planes at z = −20 mm (left) and z = 20 mm (right) of instantaneous resolved temperature field without (NR, top) and with (R100, bottom) radiative heat transfer at \(t_2\) = 0.28 s

Figures 11 and 12 compare the mean resolved temperature in the center plane (z = 0) and symmetrical side planes (z = −20 mm and z = 20 mm) without and with radiative heat transfer, respectively. As expected, the maximum mean temperature is about 80 K lower when radiative heat transfer is taken into account because of the increase in energy transfers from hot zones to cold zones and walls as a result of radiation. Differences in temperature begin just a few centimeters downstream of the flame holder, as illustrated in Figs. 13 and 14, where the profile of the mean resolved temperature along the \(x\) axis in the center plane is shown from the flame holder until the end of the combustion chamber for three different values of \(y\).

Fig. 11
figure11

Cuts in plane at z = 0 of mean resolved temperature field without (NR, top) and with (R100, bottom) radiative heat transfer

Fig. 12
figure12

Cuts in planes at z = −20 mm (left) and z = 20 mm (right) of mean resolved temperature field without (NR, top) and with (R100, bottom) radiative heat transfer

Fig. 13
figure13

Profiles of mean resolved temperature field without (NR) and with (R100) radiative heat transfer along the burner axis at \(y\) = 0 and \(z\) = 0

Fig. 14
figure14

Profiles of mean resolved temperature field without (NR) and with (R100) radiative heat transfer at \(y\) = −0.01 m and \(z\) = 0 (top) and \(y\) = 0.01 m and \(z\) = 0 (bottom)

These temperature differences directly impact the reaction products. For example, the mean CO2 mass fraction (Fig. 15) decreased slightly when radiation was included, in spite of the simple one-step global chemical mechanism used here. These observations are confirmed by Fig. 16, which shows longitudinal cuts of the CO2 mass fractions at \(y\) = −0.01 m and \(y\) = 0.01 m for \(z\) = 0. Radiative heat transfer can be expected to have a more significant impact when using a more detailed chemical mechanism that includes pollutants such as carbon monoxide (CO) or nitric oxides (NO x ), whose formation is known to be very sensitive to local temperature.

Fig. 15
figure15

Cuts at z = 0 of mean resolved CO2 mass-fraction field without (NR, top) and with (R100, bottom) radiative heat transfer

Fig. 16
figure16

Mean CO2 mass-fraction profile at \(y\)  = −0.01 (top) and \(y\) = 0.01 (bottom), \(z\) = 0 without (NR) and with (R100) radiative heat transfer

Figure 17 shows the mean heat release rate field in the central plane (z = 0) without and with radiative heat transfer. The flow and temperature modifications when radiation is taken into account directly impact the reaction rate, corroborating the observation above about CO2 mean mass fractions. Lower values of heat release are found in the first quarter of the combustion chamber with radiative heat transfer.

Fig. 17
figure17

Cuts at z = 0 of mean heat released field without (NR, top) and with (R100, bottom) radiative heat transfer

To take a deeper look at the temperature modifications caused by radiative heat transfer, we plotted in Fig. 18 the difference between the mean resolved temperature without and with radiative heat transfer, \(\bigtriangleup \langle \widetilde{T} \rangle = \langle \widetilde{T}_{NR} \rangle - \langle \widetilde{T}_{R100} \rangle\). The small values of \(\bigtriangleup \langle \widetilde{T} \rangle\) (less than 25 K) just downstream of the flame holder can be explained in part by the high wall temperatures found (about 1,900 K), which reduces heat exchange by radiation between the obstacle wall and the hot burnt gas. Near the end of the combustion zone, the flame interacts with the combustion chamber walls, resulting in a temperature difference of more than 150 K. It should be remembered that radiative heat flux is not taken into account in the wall boundary condition used in the combustion code. The validity of this approximation should be carefully investigated in zones where the flame directly interacts with the wall.

Fig. 18
figure18

Cut in the center plane z = 0 of the difference between mean resolved temperature (\(\bigtriangleup \langle \widetilde{T} \rangle = \langle \widetilde{T}_{NR} \rangle - \langle \widetilde{T}_{R100} \rangle\)) without and with radiative heat transfer

The length of the recirculation zone is about 7 cm and is not greatly affected by radiative heat transfer, as shown in Fig. 19. However, the mean temperature field changes when radiative heat transfer is included, especially in the recirculation zone downstream of the flame holder (Fig. 13), which plays a key role in the flame stabilization mechanism. Changes in the temperature field in this region may induce major changes in the behavior of the flame and its interaction with turbulence motions

Fig. 19
figure19

Profiles of mean velocity u along the burner axis for \(y\) = 0 and \(z\) = 0

.

The filtered RMS (root mean square) temperature illustrates more quantitatively how the flame dynamics change when radiative heat transfer is taken into account, as shown in Fig. 20, which compares the filtered RMS temperature profiles in the central plane of the burner without and with radiation. In spite of the similarity between the profiles, the values of RMS temperature are higher downstream of the recirculation zone behind the obstacle when radiative heat transfer is taken into account, as confirmed by Figs. 21 and 22, where profiles of RMS temperature are shown along the axis \(x\) at z = 0 for different values of \(y\). Differences of more than 80 K can be observed in some regions, indicating stronger flame movements. These findings confirm the influence of radiative heat transfer despite its limited contribution to energy transfer (less than 2 % of the overall heat released).

Fig. 20
figure20

RMS of the resolved temperature in the center plane (z = 0) extracted from the large eddy simulation without (NR, top) and with (R100, bottom) radiative heat transfer

Fig. 21
figure21

RMS temperature profile without (NR) and with (R100) radiative heat transfer along the burner axis (\(y\) = 0 and z = 0)

Fig. 22
figure22

RMS temperature profile along the \(x\) axis without (NR) and with (R100) radiative heat transfer for \(z\) = 0 and y = −0.01 (top) and \(y\)  = 0.01 (bottom)

To aid analysis of the mean resolved temperature changes and corresponding RMS temperature, Fig. 23 shows the transverse profiles of these two quantities at \(x\) = 0.11 m and z = 0 without and with radiative heat transfer in a zone about 0.04 m downstream of the recirculation zone. Mean temperature is always lower with radiation, and peak temperature is about 50 K lower in the central region. The RMS temperature profiles have higher values in the central and peak regions in the simulation with radiative heat transfer.

Fig. 23
figure23

Mean temperature profile (top) and RMS temperature profile (bottom) for the LES without (NR) or with (R100) radiative heat transfer at \(x\) = 110 mm and \(z\) = 0

In Figs. 24 and 25 a different plane of cut has been used, this time parallel to the top and bottom combustion chamber wall in the central zone (\(y\) = 0). The profiles are extracted at \(x\) = 0.13 m, about 0.06 m downstream of the recirculation zone. The mean resolved temperatures are always lower with radiative heat transfer. A smaller difference is found at the boundaries, and the maximum mean temperature is about 50 K lower with radiation. The RMS temperatures in Fig. 25 are generally higher when radiative heat transfer is taken into account, and the maximum RMS value is increased by about 40 K. However, a zone where the RMS temperature is higher without radiation can be observed between the peaks and the axis of the chamber and should be investigated in a future study.

Fig. 24
figure24

Mean resolved temperature profile for the LES without (NR) and with (R100) radiative heat transfer at \(x\) = 130 mm and \(y\) = 0

Fig. 25
figure25

Resolved RMS temperature profile for the LES without (NR) and with (R100) radiative heat transfer at \(x\) = 130 mm and \(y\) = 0

The same explanation put forward in the two-dimensional study [38] can be used here to explain the increase in the resolved RMS temperature: the lower temperature of the burnt gas as a result of radiation heat losses leads to reduced thermal expansion and a decreased flame speed \(S_L^0\), in turn leading to a higher value of \(N_{\rm B} = u' / (\tau S_L^0)\) [\(u'\) is the turbulence intensity, \(\tau = (T_{\rm b} / T_{\rm u}) - 1\) is the heat release factor, \(T_{\rm u}\) and \(T_{\rm b}\) are the fresh and burnt gas temperatures, respectively, and is the laminar flame speed] and increasing the flame’s sensitivity to turbulent motions (larger flame wrinkling).

Figure 26 shows the instantaneous radiative power field. The large positive central zone corresponds to the radiation power emitted by the hot burnt gases. A small absorption zone is visible upstream of the flame front. These findings are confirmed by Fig. 27, which compares profiles of radiative power (in W m−3) at z = 0 and different values of \(x\) downstream of the flame holder. Figure 28 shows mean radiative power along the central axis of the combustion chamber (\(y\) = 0 and z = 0); the maximum occurs at about \(x\) = 0.12 m. The global radiative power calculated with DOMASIUM for this flame is about 1 kW, which represents less than 2 % of the total flame power. Nevertheless, radiative heat transfer has a major effect on the flame dynamics.

Fig. 26
figure26

Instantaneous radiative power field (in W m−3) at \(t_2\) = 0.28 s. Positive values indicate radiative power lost by the gases, and negative values power absorbed

Fig. 27
figure27

Mean radiative power profile (in KW m−3) at z = 0 and different values of \(x\) downstream of the flame holder

Fig. 28
figure28

Mean radiative power profile (in KW m−3) at \(y\) = 0 and z = 0

Comparison with experimental data

As discussed in the previous two-dimensional study [38], comparison of numerical simulations and experimental data is challenging as, to our knowledge, few experiments that are well characterized both in terms of reacting flow fields and radiative heat transfer are available in the literature. One such experiment, however, is described in a recent study by Rankin et al. [34]. Here, we again attempt to compare numerical simulations and experimental data using the experimental data collected in [23, 24]. Planar laser-induced fluorescence (PLIF) of OH radicals was performed in the center plane of the burner shown in Fig. 1. The progress variable \(c\), which follows the temperature (\(c\) = 0 in fresh gases and 1 in fully burnt gases), is extracted from binarized images assuming an infinitely thin flame front. Instantaneous mass-filtered \(\widetilde{c}\) progress variables are then calculated by applying a Gaussian filter of size \(\Delta = F \delta _L^0 = 6.2\) mm, corresponding to the thickness of the resolved three-dimensional flame front. Ensemble averages are extracted by processing 200 images.

The numerical time-averaged filtered progress variable, \(\langle \widetilde{c} \rangle\), and the corresponding resolved variance, \(\langle \widetilde{c}^2\rangle - {\langle \widetilde{c} \rangle }^2\), are reconstructed from the instantaneous filtered temperature fields obtained in the simulation (the operator \(\langle \rangle\) denotes time averages) to extract variables that can be compared with experimental data. These variables are reconstructed from experimental fields under three assumptions: (1) infinitely thin instantaneous flame fronts, (2) two-dimensional images and (3) a Gaussian filter of size \(\Delta = F \delta _L^0.\)

Figure 29 compares the transverse profiles of the mean progress variable \(\langle \widetilde{c} \rangle\) and corresponding resolved variance \(\langle (\widetilde{c})^2 \rangle - {\langle \widetilde{c} \rangle }^2\) extracted from the experiment and the numerical simulations without (NR) and with (R) radiative heat transfer for three downstream locations. The mean progress variable profiles are quite similar with and without radiative heat transfer. However, the top of the curve is wider in the experimental data, probably because the parameters used in the simulation result in a slower reaction rate than in the experimental study or because of the three-dimensional effects neglected when the experimental data are processed. The resolved variance profiles display better agreement with the experimental values when radiative heat transfer is included, especially 11 cm downstream of the flame holder. Upstream, the numerical simulations tend to overestimate the flame brush thickness and resolved variance both without and with radiation (see profiles at \(x\) = 7 cm in Fig. 29). The downstream location \(x\) = 7 cm corresponds roughly to the end of the recirculation zone downstream of the flame holder. In the recirculation zone, the resolved flame thickness \(F \delta _L^0 = 6.2\)  mm is of the order of half the effective flow section and is too large to yield reliable results.

Fig. 29
figure29

Transverse profiles of the mean progress variable \(\langle \widetilde{c} \rangle\) (left) and the corresponding resolved variance \(\langle (\widetilde{c})^2 \rangle - {\langle \widetilde{c} \rangle }^2\) (right) extracted from the experiment (dashed lines) and the numerical simulations without (NR, lines) and with (R, bold lines) radiative heat transfer for three downstream locations 7, 9 and 11 cm from the flame holder, respectively

Conclusion

Three-dimensional simulations were performed in which an LES combustion code was coupled with a radiation solver. A turbulent premixed propane/air flame stabilized downstream of a triangular obstacle, which had previously been studied in two-dimensional simulations, was used as a test case. The results show the three-dimensional character of the flame, the importance of the boundary conditions and that when radiative heat transfer is taken into account there is a change in the flame dynamics and better agreement with experimental data.

References

  1. 1.

    Barlow RS, Frank JH (1998) Effects of turbulence on species mass fractions in methane/air jet flames. In: Twenty-Seventh Symposium (Int) on Combustion, pp 1087–1095

  2. 2.

    Carlson B (1971) Tables of equal weight quadrature eqn over the unit sphere. Tech. Rep. LA-4734, Los Alamos Scientific Laboratory Report

  3. 3.

    Centeno FR, da Silva CV, França FH (2014) The influence of gas radiation on the thermal behavior of a 2d axisymmetric turbulent non-premixed methane-air flame. Energy Convers Manag 79:405–414

    Article  Google Scholar 

  4. 4.

    Centeno FR, Cassol F, Vielmo HA, da Silva CV (2013) Comparison of different WSGG correlations in the computation of thermal radiation in a 2D axisymmetric turbulent non-premixed methane-air flame. J Braz Soc Mech Sci Eng 35(4):419–430

    Article  Google Scholar 

  5. 5.

    Coelho P (2004) Detailed numerical simulation of radiative transfer in a nonluminous turbulent jet diffusion flame. Combust Flame 136(4):481–492

    MathSciNet  Article  Google Scholar 

  6. 6.

    Coelho P, Terrling O, Roekaerts D (2003) Spectral radiative effects and turbulence/radiation interaction in a non-luminous turbulent jet diffusion flame. Combust Flame 133(1–2):75–91

    Article  Google Scholar 

  7. 7.

    Coelho PJ (2013) A theoretical analysis of the influence of turbulence on radiative emission in turbulent diffusion flames of methane. Combust Flame 160(3):610–617

    Article  Google Scholar 

  8. 8.

    Colin O, Ducros F, Veynante D, Poinsot T (2000) A thickened flame model for large eddy simulations of turbulent premixed combustion. Phys Fluids A 12(7):1843–1863

    Article  Google Scholar 

  9. 9.

    Desjardin P, Frankel S (1999) Two-dimensional large eddy simulation of soot formation in the near-field of a strongly radiating nonpremixed acetylene-air turbulent jet flame. Combust Flame 119(1–2):121–132

    Article  Google Scholar 

  10. 10.

    Elias RN, Coutinho ALGA (2007) Stabilized edge-based finite element simulation of free-surface flows. Int J Numer Methods Fluids 54(6–8):965–993. doi:10.1002/fld.1475

    MathSciNet  Article  MATH  Google Scholar 

  11. 11.

    Giordano P, Lentini D (2001) Combustion-radiation-turbulence interaction modeling in absorbing/emitting nonpremixed flames. Combust Sci Technol 172:1–22

    Article  Google Scholar 

  12. 12.

    Gonçalves J, Coelho P (1997) Parallelization of the discrete ordinates method. Numer Heat Transf Part B 32:151–173

    Article  Google Scholar 

  13. 13.

    Hartick J, Tacke M, Fruchtel G, Hassel E, Janicka J (1996) Interaction of turbulence and radiation in confined diffusion flames. Proc Combust Inst 26:75–82

    Article  Google Scholar 

  14. 14.

    Henning M, Vinoski S (1999) Advanced CORBA programming with C++. Addison-Wesley

  15. 15.

    Hirsch C (1989) Numerical computation of internal and external flows. Wiley, New York

    Google Scholar 

  16. 16.

    Incropera FP, DeWitt DP (1996) Fundamentals of heat and mass transfer, 4th edn. Willey, New York

    Google Scholar 

  17. 17.

    Jensen KA, Ripoll JF, Wray AA, Joseph D, El Hafi M (2007) On various modeling approaches to radiative heat transfer in pool fires. Combust Flame 148(4):263–279

    Article  Google Scholar 

  18. 18.

    Jones W, Paul M (2005) Combination of DOM with LES in a gas turbine combustor. Int J Eng Sci 43(5–6):379–397

    Article  MATH  Google Scholar 

  19. 19.

    Joseph D (2004) Modélisation des transferts radiatifs en combustion par la méthode des ordonnées discrètes sur des maillages non-structurés tridimensionnels. PhD thesis, Institut National Polytechnique de Toulouse

  20. 20.

    Joseph D, Coelho P, Cuenot B, El Hafi M (2003) Application of the discrete ordinates method to grey media in complex geometries using unstructured meshes. In: Eurotherm73 on Computational Thermal Radiation in Participating Media, Eurotherm Series, Mons (Belgium), vol 11, pp 97–106

  21. 21.

    Joseph D, El Hafi M, Fournier R, Cuenot B (2005) Comparison of the discrete ordinates method spatial differencing schemes in complex geometries using 3-dimensional unstructured meshes. Int J Therm Sci 44(9):851–864

    Article  Google Scholar 

  22. 22.

    Knikker R, Veynante D, Rolon J, Meneveau C (2000) Planar laser-induced fluorescence in a turbulent premixed flame to analyze large eddy simulation models. In: 10th International Symposium on Applications of Laser techniques to Fluid Mechanics, Lisbon, Portugal

  23. 23.

    Knikker R, Veynante D, Meneveau C (2002) A priori testing of a similarity model for large eddy simulations of turbulent premixed combustion. Proc Combust Inst 29:2105–2111

    Article  Google Scholar 

  24. 24.

    Knikker R, Veynante D, Meneveau C (2004) A dynamic flame surface density model for large eddy simulation of turbulent premixed combustion. Phys Fluids 16(11):L91–L94. doi:10.1063/1.1780549

    Article  Google Scholar 

  25. 25.

    Lecanu M (2005) Couplage muti-physique combustion turbulente—rayonnement—cinétique chimique. Phd thesis, Ecole Centrale Paris

  26. 26.

    Légier J, Veynante D, Poinsot T (2000) Dynamically thickened flame LES model for premixed and non-premixed turbulent combustion. In: Proceedings of the Summer Program, Center for Turbulence Research

  27. 27.

    Li G, Modest M (2003) Importance of turbulence radiation interactions in turbulent diffusion jet flames. J Heat Transf Trans ASME 125(5):831–838

    Article  Google Scholar 

  28. 28.

    Liu J, Shang HM, Chen YS (2000) Development of an unstructured radiation model applicable for two-dimensional planar, axisymmetric, and three-dimensional geometries. J Quant Spectrosc Radiat Transf 66(1):17–33

    Article  Google Scholar 

  29. 29.

    Poinsot T, Veynante D (2005) Theoretical and Numerical Combustion, 2nd edn. Edwards Inc., Philadelphia

    Google Scholar 

  30. 30.

    Poinsot TJ, Lele SK (1992) Boundary conditions for direct simulations of compressible viscous flows. J Comput Phys 101(1):104–129

    MathSciNet  Article  MATH  Google Scholar 

  31. 31.

    Poitou D, El Hafi M, Cuenot B (2011) Analysis of radiation modeling for turbulent combustion: development of a methodology to couple turbulent combustion and radiative heat transfer in LES. J Heat Transf 133(6):062701

    Article  Google Scholar 

  32. 32.

    Poitou D, Amaya J, El Hafi M, Cuénot B (2012) Analysis of the interaction between turbulent combustion and thermal radiation using unsteady coupled LES/DOM simulations. Combust Flame 159(4):1605–1618

    Article  Google Scholar 

  33. 33.

    Prière C (2005) Simulations aux grandes échelles: application au jet transverse. PhD thesis, Institut National Polytechnique de Toulouse

  34. 34.

    Rankin BA, Magnotti G, Barlow RS, Gore JP (2014) Radiation intensity imaging measurements of methane and dimethyl ether turbulent nonpremixed and partially premixed jet flames. Combustion and Flame in Press

  35. 35.

    Roger M, Coelho P, da Silva C (2011) Relevance of the subgrid-scales for large eddy simulations of turbulence-radiation interactions in a turbulent plane jet. In: Sixth International Symposium on Radiative Transfer Journal of Quantitative Spectroscopy and Radiative Transfer vol 112(7), pp 1250–1256

  36. 36.

    Sakami M, Charette A, Le Dez V (1998) Radiative heat transfer in three-dimensional enclosures of complex geometry by using the discrete-ordinates method. J Quant Spectrosc Radiat Transf 59(1–2):117–136

    Article  Google Scholar 

  37. 37.

    dos Santos ED, Isoldi LA, Petry AP, Franca FHR (2014) A numerical study of combined convective and radiative heat transfer in non-reactive turbulent channel flows with several optical thicknesses: a comparison between LES and RANS. J Braz Soc Mech Sci Eng 36(1):207–219. doi:10.1007/s40430-013-0075-1

    Article  Google Scholar 

  38. 38.

    Goncalves dos Santos R, Lecanu M, Ducruix S, Gicquel O, Iacona E, Veynante D (2008) Coupled large eddy simulations of turbulent combustion and radiative heat transfer. Combust Flame 152(3):387–400

    Article  Google Scholar 

  39. 39.

    Schmitt P (2005) Simulation aux grandes échelles de la combustion étagée dans les turbines à gaz et son interaction stabilité—polluants—thermique. PhD thesis, Institut National Polytechnique de Toulouse

  40. 40.

    Selle L, Lartigue G, Poinsot T, Koch R, Schildmacher K, Krebs W, Prade B, Kaufmann P, Veynante D (2004) Compressible large eddy simulation of turbulent combustion in complex geometry on unstructured meshes. Combust Flame 137:489–505

    Article  Google Scholar 

  41. 41.

    Silva CV, França FHR, Vielmo HA (2007) Analysis of the turbulent, non-premixed combustion of natural gas in a cylindrical chamber with and without thermal radiation. Combust Sci Technol 179(8):1605–1630

    Article  Google Scholar 

  42. 42.

    Soucasse L, Rivière P, Soufiani A (2014) Subgrid-scale model for radiative transfer in turbulent participating media. J Comput Phys Part A 257:442–459

    Article  Google Scholar 

  43. 43.

    Soufiani A, Taine J (1997) High temperature gaz radiative property parameters of statistical narrow-band model for \({{\rm H}}_{2}{{\rm O}}, {{\rm CO}}_{2}\) and CO, and correlated-k model for \({{\rm H}}_{2}{{\rm O}}, {{\rm CO}}_{2}\). Int J Heat Mass Transf 40(4):987–991

    Article  Google Scholar 

  44. 44.

    Staffelbach G, Gicquel L, Poinsot T (2005) Highly parallel large eddy simulations of multiburner configurations in industrial gas turbines. In: The Cyprus International Symposium on Complex Effects in Large Eddy Simulations

  45. 45.

    Ströhle J, Schnell U, Hein K (2001) A mean flux discrete ordinates interpolation scheme for general coordinates. In: 3rd International Conference on Heat Transfer, Antalya

  46. 46.

    Thurgood C, Pollard A, Becker H (1995) The tn quadrature set for the discrete ordinates method. J Heat Transf 4(117):1068–1070

    Article  Google Scholar 

  47. 47.

    Wang P, Bai X (2005) Large eddy simulation of turbulent premixed flames using level-set g-equation. Proc Combust Inst 30:583–591

    Article  Google Scholar 

  48. 48.

    Wang Y (2005) Direct numerical simulation of non-premixed combustion with soot and thermal radiation. PhD thesis, University of Maryland

  49. 49.

    Wu Y, Haworth D, Modest M, Cuenot B (2005) Direct numerical simulation of turbulence/radiation interaction in premixed combustion systems. Proc Combust Inst 30:639–646

    Article  Google Scholar 

  50. 50.

    Yoo CS, Im HG, Wang Y, Trouvé A (2005) Interaction of turbulence, chemistry, and radiation in strained nonpremixed flames. J Phys Conf Ser 16:91–100

    Article  Google Scholar 

  51. 51.

    Zhang J, Gicquel O, Veynante D, Taine J (2009) Monte carlo method of radiative transfer applied to a turbulent flame modeling with LES. C R Mécanique 337(6–7):539–549

    Article  Google Scholar 

  52. 52.

    Zhang Y, Vicquelin R, Gicquel O, Taine J (2013) Physical study of radiation effects on the boundary layer structure in a turbulent channel flow. Int J Heat Mass Transf 61:654–666

    Article  Google Scholar 

Download references

Acknowledgments

Rogério Gonçalves dos Santos was the recipient of a scholarship from CNPq (National Council for Scientific and Technological Development, Brazil). The project was integrated in the European Union program DEISA (Distributed European Infrastructure for Supercomputing Applications) in the framework of the 6th PCRD (FOCUS project) and was supported by the French Agence Nationale de la Recherche (ANR) through the CORAYL project (ANR-05-CICG012). Simulations were carried out on the IBM SP4 zahir supercomputer provided by IDRIS (Institut de Développement et de Ressources en Informatique Scientifique) with the assistance of IDRIS (especially J.-M. Dupays, G. Grasseau and D. Girou) for CORBA coupling and CERFACS (Y. Sommerer) for AVBP. Experimental data used for comparison with numerical simulations were processed from the laser-induced fluorescence images collected by Dr. R. Knikker during his PhD.

Author information

Affiliations

Authors

Corresponding author

Correspondence to R. Gonçalves dos Santos.

Additional information

Technical Editor: Francisco Ricardo Cunha.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

dos Santos, R.G., Ducruix, S., Gicquel, O. et al. A study of three-dimensional LES of turbulent combustion with radiative heat transfer. J Braz. Soc. Mech. Sci. Eng. 38, 33–48 (2016). https://doi.org/10.1007/s40430-015-0322-8

Download citation

Keywords

  • Turbulent combustion
  • Radiative heat transfer
  • Large eddy simulation
  • Code coupling