Physics-Informed Deep Learning for Real-Time Thermal Anomaly Detection in Regenerative Cooling Channels of RP-1 Liquid Rocket Engines

Liquid Rocket Engines (LREs) utilizing RP-1 fuelare susceptible to “coking,” a thermal degradation phenomenonwhere carbon deposits impair heat transfer in regenerativecooling channels. Traditional Computational Fluid Dynamics(CFD) methods are computationally prohibitive for flight-timemonitoring. This paper proposes a Physics-Informed NeuralNetwork (PINN) framework functioning as a virtual sensor topredict nominal wall temperatures. By integrating a calibratedBartz approximation directly into the loss function, the modelestablishes a physics-consistent baseline. We demonstrate thatthis approach achieves real-time thermal anomaly detection witha Root Mean Square Error (RMSE) of 5.55 K. The model exhibitsan inference latency of < 10 ms on a standard CPU, offering aviable pathway for autonomous abort systems in next-generationreusable launch vehicles.