Predetermined thermal conductivity porous medium generated by conditional generation adversarial network

Abstract

Porous media are extensively used in the engineering field. The effective thermal conductivity and porosity are very important properties of porous medium materials. It is of great significance to obtain a porous medium material that meets the needs of effective thermal conductivity and porosity. In this paper, a four-parameter random generation method is used to produce a training data set, a conditional generation adversarial network (CGAN) is built, and a predetermined effective thermal conductivity and porosity are used as inputs to generate a porous medium structure that meets the input conditions. In particular, since the pore structure distribution of porous medium has a great influence on the effective thermal conductivity of the material, a local structure loss function is proposed to participate in the network training, so that the network can better learn the relationship between the pore distribution and the thermal conductivity. By using the lattice Boltzmann method to verify the effective thermal conductivity of the porous medium structure generated by the neural network, the results show that the method can quickly and accurately generate the porous medium structure with predetermined parameters.