Bayesian Long Short Term Memory Network for Remaining Useful Life Prediction of Rolling Bearings International Symposium on Symbiotic Nuclear Power (ISSNP2024)

报告详情

Bayesian Long Short Term Memory Network for Remaining Useful Life Prediction of Rolling Bearings

编号：108 访问权限：公开更新：2024-09-22 19:50:56 浏览：100次口头报告

报告开始：暂无开始时间（Asia/Shanghai）

报告时间：暂无持续时间

所在会场：[暂无会议] [暂无会议段]

演示文件

摘要

With the rapid development of the global nuclear power industry, increasing attention has been paid to the safety issues surrounding nuclear power equipment. In the event of equipment accidents, significant losses and hazards could potentially be incurred to personnel and the environment. As nuclear power plants operate for extended periods, roller bearings, as the key components of the pump transmission core, face potential safety hazards caused by performance degradation. Therefore, effectively predicting the remaining useful life (RUL) of the roller bearings of the circulating water pumps is of great practical significance. This study extracts multiple indicators from vibration signals, including time-domain features, JS divergence, and Gini coefficient, to comprehensively reflect the equipment's failure characteristics and compensate for the one-sidedness of single degradation indicators. Furthermore, it combines Long Short-Term Memory (LSTM) network analysis with Bayesian thinking to quantitatively assess the uncertainty range of prediction outcomes. By analyzing the time-series information contained in the features, it constructs equipment health degradation curves that reflect the sequential changes in the health status of rotating machinery. Finally, it optimizes the RUL prediction effect of the LSTM model by combining Bayesian thinking to quantify the uncertainty range of prediction results. When compared with the pre-optimization model, this method improves the generalization and accuracy of RUL prediction methods.