Data-Driven Model for Elastomers under Simultaneous Thermal and Radiation Exposure

Abstract

We present a physics-informed neural network framework for predicting the mechanical performance of elastomers exposed to concurrent thermal and gamma-radiation exposure, such as elastomers in nuclear cables or space electronics. Our demonstrated approach integrates the dual-network hypothesis with the microsphere concept to represent soft and brittle sub-networks, while embedding physical laws directly into the machine learning process. Hard constraints, e.g., incompressibility, bounded network fractions are enforced through network architecture, and soft constraints e.g., monotonicity, polyconvexity, and fading effects are imposed through the loss function. This integration reduces the effective search space, guiding the optimization toward physically admissible solutions and enhancing robustness under sparse data. Validation against published datasets on silicone rubber, ethylene propylene diene monomer, and silica-reinforced silicone foam shows accurate predictions of stress-strain behavior and elongation-at-break at exposure times not used for training. Results confirm that physics-informed constraints improve extrapolation, capture synergistic thermal-radiation effects, and provide a reliable tool for lifetime assessment of nuclear cable insulation and other radiation-exposed elastomers.