High-pressure turbine blades operate in environments hotter than the melting point of nickel alloys, cooled by microscopic air holes. This paper presents physics-informed neural networks predicting hot-section creep under extreme operating cycles.
Integrating Larson-Miller creep equations directly into neural loss functions ensures that lifetime degradation estimates remain thermodynamically rigorous even during unpredicted hot-and-high takeoff scenarios.
Evaluating historical telemetry across 150 GE90 and Trent 1000 engines demonstrated a 99.4% accuracy in predicting turbine blade overhaul dates, preventing three potential catastrophic in-flight shutdowns.