Research on rolling bearing fault diagnosis based on CNN-BiLSTM-Attention optimized by GWO

To mitigate the challenges of weak fault signatures in rolling bearings and their susceptibility to misdiagnosis or missed detection due to interference coupling, an intelligent diagnostic method for rotating machinery based on CNN-BiLSTM-Attention with global optimization via the grey wolf optimizer (GWO) is proposed. Firstly, Convolutional neural network (CNN) is employed to extract local impulse signatures from the raw vibration signal. The convolutional kernels selectively preserve shock-related features while suppressing noise. Secondly, Bidirectional long short-term memory network (BiLSTM) is utilized to capture long-range temporal dependencies. The hidden layer of BiLSTM selectively accumulates important feature representations through repetitive operations, and retains all the useful feature representations captured in the previous step at the end. Meanwhile, attention mechanism (AM) further reinforces the key fault information by adaptively re-weighting the feature maps, highlighting discriminative signatures and attenuating background disturbances so that the network focuses on the correct data and boosts expressive power. Finally, the GWO automatically searches for the optimal hyper-parameters and architecture, namely the number of CNN filters, BiLSTM hidden neurons, AM dropout rate, and fully-connected layer size,thereby avoiding the local optima that plague manual tuning, accelerating convergence, and enhancing generalization. Comparative experiments on the public bearing fault diagnosis dataset of Case Western Reserve University (CWRU) verify that the proposed model can effectively realize rolling bearing fault diagnosis and maintain high recognition accuracy. The research conclusions are of great significance for improving the efficiency and reliability of rolling bearing fault diagnosis, and help promote the development of intelligent operation and maintenance of rotating machinery as well as predictive maintenance technology for rolling bearing fault.