Hybrid Statistical–AI approach for RUL prediction of reactor protection system components

Accurate prediction of the Remaining Useful Life (RUL) of electronic components in nuclear Instrumentation and Control (I&C) systems is critical for ensuring the long-term reliability and safety of nuclear power plants (NPPs). This study presents an integrated experimental and computational framework for RUL prediction, focusing on photo-couplers used in the Reactor Protection System (RPS). Accelerated aging tests were conducted under combined thermal and electrical stress conditions to obtain high-fidelity degradation data. Using the resulting data, both statistical models (Weibull analysis and Arrhenius-based regression) and an AI model—a feedforward neural network with polynomial feature expansion—were developed to predict component lifespan. A composite loss function was incorporated into the AI model to ensure monotonic behavior and represent exponential degradation trends. Comparative analysis confirmed strong agreement between the statistical and AI-based predictions, supporting their use as a complementary dual-validation approach. Additionally, a user-friendly graphical software tool was implemented to enable field engineers to perform real-time RUL estimation from sensor data. The results demonstrate the feasibility of combining physics-informed AI with conventional methods, offering a validated and practical solution for predictive maintenance of nuclear I&C components.