Privacy Protection based on Remaining Useful Life Prediction of Tool

As one of the important components of industrial equipment, the health of tool is of great significance to the smoothness and stability of industrial production. However, the empirically tool change strategy is not conducive to reduce production cost and improve product efficiency. Due to technical protection, manufacturers often do not want to share their production data of tool, which limits the remaining useful life (RUL) prediction effectiveness of tool with deep learning methods that rely on data quality and quantity. To realize RUL prediction with high accuracy and data privacy protection, this paper proposes a new method based on federated learning (FL) and ordered neurons long short-term memory (ON-LSTM) network. Firstly, median filtering and third quartile method are adopted to process outliers and invalid values. Secondly, XGBoost and Pearson correlation coefficient are employed to select and extract features with strong correlation in time domain and frequency domain. Finally, a federated learning prediction framework is designed to realize RUL prediction on the basis of protecting user data privacy. The experimental results show that the proposed method not only protects data privacy, but also has higher prediction accuracy.