Intelligent Fault Diagnosis of Rolling Bearings Using Adaptive WBKA-Optimized Hybrid CNN-SE-BiLSTM Network

To address the limitations of traditional deep learning models, which rely on empirical hyperparameter tuning and suffer from limited diagnostic accuracy due to insufficient feature extraction capability in rolling bearing fault diagnosis, this paper proposes an intelligent diagnosis method. The proposed approach is based on an adaptive weighted Black-Winged Kite Algorithm (WBKA)-optimized Convolutional Neural Network with Squeeze-and-Excitation Attention-Bidirectional and Long Short-Term Memory (CNN-SE-BiLSTM) network. Firstly, raw vibration signals are preprocessed using z-score normalization to eliminate dimensional inconsistencies and enhance the stability of model training. Secondly, to mitigate the issues of premature convergence and local optima entrapment of the WBKA, an adaptive weighting strategy and dynamic mutation mechanism are designed to improve its global search capability and convergence efficiency. Finally, an enhanced WBKA-optimized hybrid diagnosis model is developed by embedding the Squeeze-and-Excitation (SE) attention module into the CNN-BiLSTM framework. The SE module dynamically recalibrates the importance of channel-wise features, the CNN captures spatial fault patterns, and the BiLSTM models bidirectional temporal dependencies, thereby establishing a collaborative spatiotemporal diagnostic mechanism. Experimental validation on the datasets from Case Western Reserve University and Jiangnan University demonstrates that the optimized CNN-SE-BiLSTM model achieves superior diagnostic accuracy (99.62%-99.98%), with an F1-score improvement of 0.9% over the CNN-BiLSTM baseline, highlighting the efficacy of the SE-guided attention mechanism. Furthermore, the improved WBKA exhibits 60% fewer iterations than PSO (46 vs 115) and stronger robustness, which enables high accuracy and efficiency in intelligent fault diagnosis.