Physical-Based Neural Network with Degradation Model for Remaining Useful Life Prediction of Aero-engine

Remaining useful life (RUL) prediction is crucial for ensuring the stability and reliability of aero-engine systems. However, RUL prediction methods based on deep learning frameworks often overlook the valuable guidance provided by empirical degradation models, which limits the algorithm performance, particularly in complex systems such as aero-engines. To better incorporate physical prior knowledge into intricate systems and further improve prediction accuracy under varying operating conditions, a novel deep learning network that integrates a degradation model is proposed. This algorithm leverages system energy flow to inform the construction of an efficiency-based feature extraction network, aiming to capture features that more accurately reflect failure mechanisms. A Physics-Informed Neural Network (PINN) that incorporates empirical degradation model is employed to achieve RUL prediction, improving the adaptability across diverse operating conditions. The proposed algorithm is validated using the publicly available C-MAPSS dataset from NASA. The results demonstrate that a significant improvement of 14.48% in prediction accuracy compared to other advanced.