Assessment of radiative heat transfer characteristics of a combustion mixture in a three-dimensional enclosure using RAD-NETT (with application to a fire resistance test furnace)

Keywords: RAD-NETT Radiation heat transfer Neural network Fire resistance test furnace a b s t r a c t Using RAD-NETT, a neural network correlation of the non-gray radiative absorption properties of combustion gases (CO 2 and H 2 O) and soot, the emissivity and hemispherical absorptivity of a combustion mixture to a boundary in a rectangular enclosure is determined. Results show that the both the emissivity and hemispherical absorptivity have a strong dependence on the mixture properties, as well as the medium temperature and wall temperature. The gray assumption with emissivity equal to absorptivity is generally inaccurate. The numerical model is used to analyze temperature and heat transfer data generated from a fire resistance test furnace. Results show that emission and reflection from the wall boundaries have a major effect of the radiative heat flux measurement in a test sample in a fire resistance test. Numerical results also demonstrate that the furnace was operating essentially in an isothermal condition. From the perspective of a compartment fire, numerical data show that soot emission and emission from the wall are essential in the initiation of flashover in a compartment fire. In fire safety and other analyses of heat transfer in an absorbing and emitting combustion medium, the use of a constant emissivity and absorptivity is a common engineering approach because of the mathematical complexity to account for the non-gray and three-dimensional characteristics of radiative heat transfer. Over the years, a great deal of works have been reported in dealing with the non-gray multi-dimensional aspect of radiative heat transfer, particularly in a combustion media [1–4]. However, the proposed computational approaches are still not sufficiently efficient to eliminate the need to make this simplifying assumption in practical engineering system. Until now, there has been few direct assessment of this important engineering approximation in an actual three-dimensional combustion environment. In a recent work [5], a computationally efficient approach was developed for the evaluation of the total absorptivity of a one-dimensional slab of combustion mixture (CO 2 , H 2 O and soot) using narrow-band absorption data [6]. Specifically, numerical data for the total absorptivity, generated by a direct integration of the spectral data, are correlated by a neural network, RAD-NETT, as a function of the relevant physical parameters such as source temperature, absorption gas temperature, partial pressures and soot volume fraction. Based on RAD-NETT, the analysis of radiative heat transfer in a three-dimensional structure …