Artificial Neural Networks in Radiation Heat Transfer Analysis

Abstract

In the Monte Carlo ray-trace (MCRT) method, millions of rays are emitted and traced throughout an enclosure following the laws of geometrical optics. Each ray represents the path of a discrete quantum of energy emitted from surface element i and eventually absorbed by surface element j. The distribution of rays absorbed by the n surface elements making up the enclosure is interpreted in terms of a radiation distribution factor matrix whose elements represent the probability that energy emitted by element i will be absorbed by element j. Once obtained, the distribution factor matrix may be used to compute the net heat flux distribution on the walls of an enclosure corresponding to a specified surface temperature distribution. It is computationally very expensive to obtain high accuracy in the heat transfer calculation when high spatial resolution is required. This is especially true if a manifold of emissivities is to be considered in a parametric study in which each value of surface emissivity requires a new ray-trace to determine the corresponding distribution factor matrix. Artificial neural networks (ANNs) offer an alternative approach whose computational cost is greatly inferior to that of the traditional MCRT method. Significant computational efficiency is realized by eliminating the need to perform a new ray trace for each value of emissivity. The current contribution introduces and demonstrates through case studies estimation of radiation distribution factor matrices using ANNs and their subsequent use in radiation heat transfer calculations.