PINN and KAN temperature prediction of carbon Fiber/Epoxy composite materials irradiated by nuclear environment simulated intense heat fluxes

Abstract

The carbon fiber/epoxy composite (CFEC) materials in nuclear environments must endure extreme temperature conditions, making the understanding of their thermal response characteristics crucial for material design and safety assessment. Here, the physical information neural network (PINN) was used to develop a fast solver, while the Kolmogorov Arnold network (KAN) was employed to analyze and predict the correlation between multiple experimental parameters and the material’s surface temperature. An ensemble learning method was adopted to integrate six machine learning models and enhance the robustness of prediction. The surface temperature characteristics of CFEC are significantly influenced by radiation and cooling processes in high- temperature environments. The KAN established an explicit functional relationship between T_max, t_max and input parameters, providing a theoretical basis for optimizing material design. This research contributes to the developing of standards for CFEC performance evaluation under intense heat fluxes, ensuring product quality and safety.