Graph Structure and Temporal Data Driven Remaining Useful Life Prediction Method for Machinery

With the digital and intelligent transformation in manufacturing industry, the data-driven intelligent maintenance for machinery has been attracted more and more attention from both the academic and industrial society in recent years. However, the existing remaining life prediction methods have some limitations, such as the poor performance for extracting useful time sequence information and for establishing accurate mapping between monitoring data and real degradation trend. To solve the above problems, a graph structure and temporal data driven remaining useful life prediction method is proposed for machinery. First, the remaining life prediction model is constructed by combining the graph convolution neural network(GCN) and the long short-term memory (LSTM). Second, the original data are used to construct the fixed time-step temporal data, and are also used to construct the graph structure data by using the feature correlation algorithm. The graph data and temporal data is used as the input of GCN and LSTM module, respectively. Furthermore, the life prediction model can be trained and optimized by minimizing the prediction loss. Finally, the performance of the proposed method is verified based on an accelerated bearing degradation dataset. The RMSE of the proposed method is 0.107 in a single working condition and 0.099 in a comprehensive working condition. The experimental results show that the proposed method achieves a better performance than that of the other state-of-the-art methods and provides a promising solution for prognostic and health management of machinery in practical industrial application. © 2023 Editorial Office of Chinese Journal of Mechanical Engineering. All rights reserved.