Research on Remaining Useful Life Prediction of Rolling Bearings Based on CDAE-1DCNN-BiLSTM

Rolling bearings are critical components in rotating machinery, and accurate prediction of their remaining useful life (RUL) is essential for maintenance management and fault prevention. To improve the accuracy of RUL prediction for rolling bearings, this study proposes a deep learning model that integrates a convolutional denoising autoencoder (CDAE), a one-dimensional convolutional neural network (1D-CNN), and a bidirectional long short-term memory network (BiLSTM). The CDAE module enhances data quality by reducing noise in vibration signals, 1D-CNN captures local temporal features, and BiLSTM effectively models long-term dependencies within the time series. Additionally, a customized weighted mean squared error loss function is introduced to emphasize critical phases of the time series, assigning higher weights to both early degradation and late failure stages. This configuration improves the model's accuracy in predicting RUL, particularly in critical degradation phases. The model was validated on the XJTU-SY bearing dataset, achieving a mean absolute error of 0.0801 and a root mean squared error of 0.0942, outperforming existing methods in accuracy. This study provides a practical and robust solution for the RUL prediction of rolling bearings, with potential applications in industrial maintenance.