ACCURATE ESTIMATION OF RADIATION HEAT TRANSFER IN HETEROGENEOUS PACKED BEDS USING DATA-DRIVEN MODELING

Precise characterization of radiation heat transfer in porous media is computationally challenging, due to various forms of nonlinearity in the radiation transport and stochastic behavior of the complex medium. Any improvement in accurate and fast modeling is immensely appealing due to the broad range of engineering applications of porous media, particularly in inverse design such as in solar cells. This work presents a data-driven surrogate model to predict the light-matter interaction in fully heterogeneous 2D porous media consisting of elliptical particles. We develop a fine-grained pixel-based discrete tomographic imaging simulation of geometric optics and use Monte Carlo ray tracing (MCRT) method to generate supervised labeling data for random topologies in a packed bed. Furthermore, we study the effects of the variation of refractive indices and incoming light wavelength on the macro radiative properties in porous packed beds with elliptical particles. We also develop a highly generic learning model based on the final configurations, topologies and simulated ground truth. Our learning model can potentially overcome the computational limitations of MCRT in addition to improving computational efficiency.