Improved CNN-LSTM Bearing Remaining Useful Life Prediction Based on the Weibull Loss Function

Abstract

The prediction of the remaining useful life (RUL) of rolling bearings is a critical task in predictive maintenance. This paper presents a deep learning model named knowledge-driven convolutional neural network–long short-term memory (KCNN-LSTM), enhanced by the Weibull-based loss function tailored with historical bearing failure data. By incorporating a probabilistic Weibull modeling mechanism, the model can accurately capture the uncertainty and accelerated degradation trend of bearing failure over time. The prognostics and health management (PHM) 2012 and XJTU-SY bearing datasets are utilized to evaluate the proposed KCNN-LSTM model. The results indicate that the proposed KCNN-LSTM achieves superior performance compared with the conventional CNN-LSTM model, leading to a 10.2% improvement in the score metric and a notable reduction in prediction error. The proposed model offers a practical and effective approach for enhancing predictive maintenance in high-reliability industrial systems.