STBCA-Net: A Novel Hybrid Model for Bearing Fault Diagnosis Under Complex Working Conditions

Abstract Bearings are key components of rotating machinery, and their operational status directly affects the reliability and safety of the entire mechanical system. High-precision and efficient fault diagnosis is of great significance for the predictive maintenance of bearing systems and the reduction of operational risks. Aiming at the problems of low accuracy, poor generalization ability, and difficulty of model hyperparameters selection in current methods under complex working conditions, a hybrid model of STBCA integrating Swin Transformer, Bidirectional Mogrifier-Gated Recurrent Unit(BiMGRU), and Cross-attention mechanism is firstly proposed, where Swin Transformer is used to extract the local and global spatial features, BiMGRU is adopted to capture the temporal sequence features, and crossattention is utilized to realize the effective fusion of spatial-temporal features to enhance the discriminability of fault features. Furthermore, an improved crow search algorithm (ICSA) is developed in this study to realize the hyperparameters automatic optimization for the proposed model in the training phase, which gets rid of the drawbacks of inefficiency brought by manual trial-and-error parameters adjustment. Experimental results reveal that the presented method achieves better performance in accuracy, anti-interference and generalization capacity compared with existing mainstream schemes. The findings confirm its rationality and competitive advantages, and offer a feasible technical reference for intelligent fault identification of bearings.