PINN and KAN Temperature Prediction of Carbon Fiber/Epoxy Composite Materials Irradiated by Nuclear Explosion Simulated Light Source

摘要: Carbon fiber/epoxy composite materials are extensively utilized in military, aerospace, and medical sectors due to their numerous advantages, including light weight, high toughness, and corrosion resistance. The investigation of their thermal effects is crucial for enhancing their resilience to extreme environments. Based on experimental data obtained from nuclear explosion simulated light irradiation experiments, a fast solver based on the physical information neural network (PINN) was developed and implemented. The Kolmogorov Arnold network (KAN) was employed to analyze and predict the correlation between multiple experimental parameters and the material’s surface temperature. A heat conduction model encompassing the entire process from irradiation to cooling was proposed. The PINN method was utilized to analyze the material temperature changes, resulting in a fast solution model for the spatiotemporal evolution of material temperature. An in-depth analysis was conducted on the influence of various experimental parameters on the maximum surface temperature and its occurrence time. By constructing a dataset of each parameter and the target temperature, an efficient KAN prediction model was achieved, capable of establishing explicit functional relationships between input parameters and Tmax, tmax. The KAN prediction model exhibits high interpretability. An ensemble learning strategy was adopted to integrate the six deep learning models for prediction, thereby enhancing the robustness of the prediction results. This study presents a PINN and KAN temperature prediction model of carbon fiber/epoxy composite materials irradiated by nuclear explosion simulated light source, offering thermal protection solutions for equipment based on carbon fiber/epoxy composite materials in nuclear explosion scenarios.