Predicting radiative heat transfer in parallel computations of combustion

A software component was developed to enable predictions of radiative heat transfer in parallel computations of combustion. The discrete ordinates method and the P-1 radiation model were spatially decomposed to solve the radiative transport equation (RTE) on parallel computers. Mathematical libraries developed by third parties were used to solve the matrices that resulted during the solution procedure. Timing studies were performed to compare the performances of the iterative methods in the mathematical libraries. GMRES, BiCGSTAB iterative methods with block jacobi preconditioning emerged as the fastest solver options for the discrete ordinates method. Multigrid preconditioning accelerated the convergence of the iterative methods while solving the P-1 model equations. The parallel performance of the component did not depend strongly on the radiative properties of the medium or the boundary conditions. The global nature of radiative transfer caused degradations in the parallel efficiencies with increase in the number of processors. However, useful speed-ups were obtained in massively parallel simulations that employed large computational grids.; The predictions from coupling different property models with the RTE solution procedures were compared against benchmarks for model problems. The radiative properties were determined by the weighted-sum-of-gray gases model (WSGGM), mean absorption coefficients extracted from a narrow band model (RADCAL) or absorption coefficients extracted from total emissivity data. The WSGGM and Patch mean absorption coefficients extracted from RADCAL gave the most accurate results. However, the shortcomings of employing these property models in a combustion calculation were recognized.; Decoupled radiation calculations with experimental data as input as well as Large Eddy Simulations (LES) of a 38 cm diameter methane pool fire were performed to compare the different approaches to modeling radiative heat transfer. The predictions were compared with experimental data. The different approaches were able to predict the radiative loss fractions with only a moderate loss of accuracy. However, only the discrete ordinates method was able to qualitatively predict the distributions of the heat flux vectors. Results obtained from the calculations performed with the total emissivity model were very close to the non-gray calculations performed using WSGGM or RADCAL, while being significantly faster.