Prediction of Remaining Useful Life of Rolling Bearings Based on Multiscale Efficient Channel Attention CNN and Bidirectional GRU

To effectively capture both local and global features while retaining temporal dependencies in time-series data, and to improve the accuracy of remaining useful life (RUL) prediction of rolling bearings, this paper proposes a hybrid architecture based on a multi-scale efficient channel attention convolutional neural network and bidirectional gated recurrent unit networks. The method based on based on multi-scale efficient channel attention CNN and bidirectional GRU, which is abbreviated MSECNN-BIGRU. The multi-scale efficient channel attention CNN (MSECNN) module can use both local and global features by incorporating multi-scale features and the efficient channel attention mechanism. Considering the superiority of a CNN in processing image data, the Gram angle field theory was applied to translate the one-dimensional vibration signal into Gram’s angle difference field image as the input for the MSECNN model. During the subsequent prediction process, bidirectional GRU (BIGRU) networks were proposed to avoid the one-way GRU model ignoring the influence of the next time series. In the BIGRU, the GRU was applied in both forward and backward directions to fully extract relevant information from the front and back of the sequence data, thereby improving the prediction performance of the model. By combining these modules, the MSECNN-BIGRU model could accurately predict the RUL of rolling bearings. The experimental results showed that the MSECNN-BIGRU model outperformed other classical models, making it a reliable model for predicting the RUL of rolling bearings.