Combined analysis of thermofluids and electromagnetism using physics-informed neural networks

Abstract

A physics-informed neural network was developed for estimating a solution to a multi-physics problem involving electromagnetism, fluid dynamics, and heat transfer. The multi-physical phenomenon was modeled on a cylindrical conductor with electrical and magnetic field, as well as heat transfer between the conductor and the surrounding. For improved performance, the physics-informed neural network was divided into seven interconnected neural networks. Domain decomposition and variable separation maximization was achieved by optimizing each neural network and the transfer of data between them. Results generated by the proposed physics-informed neural network showed less than 2% errors when compared to those of analytical analyses and traditional numerical methods.