Abstract: Deep neural networks have been widely used in remaining useful life (RUL) prediction of rolling element bearings, in order to further optimize the prediction model and fully extract the time series information of the data and improve the prediction accuracy, an RUL prediction algorithm combining deep residual network (DRN) and bidirectional gated recurrent unit (BiGRU) was proposed. Firstly, the sliding window method was used to resample the original data and expand the dataset.Then, a DRN-BiGRU model was designed, the spatial features of the input data were extracted by the DRN part,and BiGRU could capture the relevant features of the past and future directions contained in the time-domain data, fully obtain the timeseries degradation information of the input data, and further improve the feature extraction effect of the model.Finally,the effectiveness of the model was validated using the published PHM2012 dataset, and the prediction results were compared with those obtained using DRN, DRNGRU and full convolutional neural network (FCNN) models. The results show that, comparing with other deep neural networks considered in this study, the proposed algorithm has the smallest prediction error in the RUL prediction of bearings and also has the highest prediction score of 0.985, which verifies the accuracy and validity of this model in the application of rolling element bearings remaining useful life prediction.
Key words: prognostic and health management(PHM); data driven forecasting method; remaining useful life (RUL)prediction model; deep residual network (DRN); bidirectional gated recurrent unit (BiGRU); bearing accelerated degradation dataset