Engineered Features and Artificial Neural Networks for the Identification of Temperature-Dependent Radiative Characteristics in Porous Media

The accurate and efficient prediction of radiative properties of materials plays a crucial role in numerous industrial and engineering applications. However, experimental measurement can be cumbersome, particularly for complex materials. As a result, numerical techniques such as Monte Carlo ray tracing (MCRT) have been developed to assess radiative properties. MCRT is a well-known tool for solving radiation heat transport (RHT). Despite its advantages, MCRT requires tracing a large bundle of light beams for convergency and veracity of outcomes, which necessitates further advancements in its employment. To this end, supervised learning algorithms can be used in conjunction with MCRT to provide a quick and accurate substitution for typical MCRT simulations. In this study, machine learning algorithms employing expertly designed features are used to evaluate the radiative properties of heterogeneous porous media, while considering the effects of conductive heat transfer. Directional features have been introduced into the sets of the features implemented in our previous study for the enhancement of predictions. The results show that the applied method can improve prediction accuracy compared to prior studies. Moreover, this approach is generalizable to other radiative heat transfer problems by extracting relevant characteristics in accordance with the involved physics.