Application of Feature Fusion-Driven Physics-Informed Neural Network in Remaining Useful Life Prediction

Abstract

Addressing the issues of traditional data-driven methods lacking interpretability and physics-informed neural network (PINN) being susceptible to noise due to statistical feature inputs, this paper proposes a feature fusion-based physics-informed neural network method for predicting the remaining useful life (RUL) of bearings. The method first extracts time-domain and frequency-domain features from the original bearing vibration signals; Then, it uses principal component analysis (PCA) to fuse the extracted time-domain and frequency-domain features, constructing a bearing health indicator (HI) to mitigate noise effects and enhance dataset features effectively; Finally, it uses the bearing health indicator and corresponding time $\boldsymbol{t}$ during the degradation stage as inputs for the PINN model to predict the bearing's RUL. Through testing on the Xi'an Jiaotong university bearing dataset, the experimental results demonstrate that the PINN model achieves highly interpretable predictions by fitting the physical relationship between bearing HI and RUL. Additionally, the experiment compares the model with feature fusion input to the model with direct statistical feature input, and the results show that the PINN model with feature fusion input has higher computational efficiency and better accuracy, effectively achieving accurate prediction of bearing remaining useful life, providing strong support for equipment fault prediction and health management.