Abstract:  Deep neural networks have been widely used in the field of remaining useful life prediction (RUL). The traditional rolling bearing life prediction model has low prediction accuracy and poor robustness. In order to further improve the accuracy and robustness of the prediction model, a rolling bearing residual life prediction model was proposed by integrating three modules, namely, deep convolutional neural network (DCNN), bidirectional gated recurrent unit (BiGRU), and self-attention mechanism (Self-Attention). Firstly, the timedomain features and frequency-domain features of the original vibration signals were extracted using the DCNN network. Then, the extracted features were evaluated and screened using uncertainty quantification, and the screened features were used to construct a new alternative feature set. Finally, the remaining service life of the bearing was predicted using Self-Attention-BiGRU network. The proposed method was validated on the IEEE PHM2012 dataset. The experimental results show that the DCNN network and Self Attention-BiGRU method provides the optimal prediction results ,with the lowest two error values: mean absolute error (MAE), root mean squared error(RMSE) compared to the prediction results of the three models BiGRU, GRU and BiLSTM. Among them, the MAE value predicted by the RUL of the No. 1 bearing in Case I is respectively decreased by 7.0%, 7.4% and 6.5% compared to the BiGRU, GRU, and BiLSTM networks. The RMSE values is respectively decreased by 7.6%, 8.4% and 6.9% compared to the three other models, and the highest Score value is predicted with a score of 0. 985.The robustness of the proposed method for bearing RUL prediction is demonstrated by the partitioning of different datasets. The experimental results validate the effectiveness of the DCNN network and Self-Attention-BiGRU based model in bearing remaining useful life prediction.
Key words:  rolling bearing; remaining useful life (RUL); bidirectional gated recurrent unit (BiGRU); uncertainty quantization; self-attention mechanism; deep convolutional neural network (DCNN); prognostic and health management (PHM)