Method for predicting conductive heat transfer topologies based on Fourier neural operator

Abstract

This paper presents an iterative topology optimizer for conductive heat transfer structures based on the Fourier neural operator (FNO). A data-driven model based on FNO is trained to predict the temperature under different material distributions, different boundary conditions, and different thermal loads. A new method is used to generate data, which makes the modeling process of temperature predictor completely independent of the traditional optimization method. Then by coupling the trained temperature predictor with the solid isotropic material with penalization (SIMP) method, a new iterative topology optimizer is formed. Numerical experiments demonstrate that the proposed method can generate heat transfer structures with good performance, and can apply the model trained on low-resolution data to the structural topology optimization with high resolution, which greatly improves the optimization efficiency. In addition to the heat conduction structure optimization problem, the method developed in this paper is expected to be applied to other optimization problems or coupled with other conventional optimization methods