Prediction of remaining useful life for rolling bearings based on DCNN-BiLSTM-CBAM with multi-scale feature extraction

Abstract To address the limitations of single-scale feature extraction and inadequate capture of degradation information in Remai ning Useful Life (RUL) prediction, this paper proposes a rolling bearing RUL prediction method based on multi-scale feature extraction. Firstly, the Inception module is employed to optimise the convolutional neural network, enabling parallel extraction and fusion of multi-scale degradation features. Subsequently, the Convolutional Block Attention Module (CBAM) is integrated to adaptively weight features, enhancing the ability to focus on critical degradation information. Finally, a Bidirectional Long Short-Term Memory (BiLSTM) network is employed for temporal modelling to fully extract the temporal evolution patterns within vibration signals. The proposed model (DCNN-BiLSTM-CBAM) was validated on the PHM2012 dataset. Results demonstrated superior performance over comparison models in terms of Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE), with respective mean values of 0.0087, 0.085, and 0.057 respectively. These values represent reductions of 8.42%, 9.57%, and 12.3% compared to the CNN-LSTM-CBAM model, and reductions of 77.69%, 51.97%, and 25.97% compared to the SACGNet model.