关键词: 滚动轴承, 故障始发时刻, 剩余寿命预测, 改进图注意力双向门控循环单元

Abstract: Due to the influences of working conditions and operating conditions, the collected status monitoring data was interfered with strong noise in full life cycle of rolling bearings, and the bearing operating life degradation was nonlinear, which seriously affected the accuracy of residual life prediction. So, a bearing residual life prediction method was proposed based on a joint high-precision FPT degradation state evaluation and an IGAT-BiGRU network, and the XJTU-SY full life cycle bearing dataset was used to verify the effectiveness of the proposed method. The results show that the proposed prediction method may effectively capture the deep spatiotemporal features that characterize the bearing degradation states, and significantly improve the residual life prediction accuracy, compared with methods such as CNN-LSTM.

Key words: rolling bearing, first predicting time(FPT), residual life prediction, improved graph attention bidirectional gate recurrent unit(IGAT-BiGRU)