Statistical modeling of thermal radiation transfer in buoyant turbulent diffusion flames

A statistical (Monte Carlo) method for radiative heat transfer has been incorporated in CFD modeling of buoyant turbulent diffusion flames in stagnant air and in a cross-wind. The model and the computational tool have been developed and applied to simulate both burner flames with controlled fuel supply rate and in self-sustained pool fires with burning rates coupled with flame radiation. The gas–soot mixture was treated either as gray (using the effective absorption coefficient derived from total emissivity data or the Planck mean absorption coefficient) or as non-gray (using the weighed sum of gray gases model). The comparison of predicted radiative heat fluxes indicates applicability of the gray media assumption in modeling of thermal radiation in case of high soot content. The effect of turbulence-radiation interaction is approximately taken into account in calculation of radiation emission, which is corrected to allow for temperature self-correlation and absorption-temperature correlation. In modeling buoyant propane flames in still air above 0.3 m diameter burner, extensive comparison is presented of the predictions with the measurements of gas species concentrations, temperature, velocity and their turbulent fluctuations, and radiative heat fluxes obtained in flames with different heat release rates. Similar to previously published experimental data, the predicted burning rate of flames above the acetone pools exposed to flame radiation increases with the pool diameter and approaches a constant level for large pool sizes. The magnitude of predicted burning rates is shown to be in agreement with the reported measurements. Augmentation of burning rate of the pool fire in a cross-wind because of increased net radiative heat flux received by the fuel surface and non-monotonic dependence of burning rate on cross-wind velocity, subject to the pool diameter, is predicted. The statistical treatment of thermal radiation transfer has been found to be robust and computationally efficient.