PINN-RUL: Physics-Informed Neural Networks for Remaining Useful Life Prediction in Reusable Rocket Engines

This paper proposes a Physics-Informed Neural Network (PINN) framework for remaining useful life (RUL) prediction in reusable rocket engines. The method embeds Coffin-Manson low-cycle fatigue physics and Chaboche viscoplastic constitutive equations directly into the loss function of an LSTM encoder, forcing predictions to remain physically consistent while learning from sensor data. Validated on NASA C-MAPSS FD001 and FD003 benchmarks, achieving 12.8% RMSE improvement over baseline LSTM on FD001 (16.5 → 14.4 cycles). Introduces a novel synthetic rocket engine degradation dataset generated by numerically solving the Chaboche ODE system for GRCop-84 copper alloy — the material used in Raptor-class combustion chamber liners. Code and model weights: CLICK HERE