A Feature Selection Approach Based on Hybrid Binary Particle Swarm and Whale Optimisation for Bearing Fault Diagnosis

This study proposes a hybrid feature selection method (HBPSWOA) based on binary particle swarm optimisation and whale optimisation algorithm for bearing fault diagnosis. To validate the proposed method, a fault diagnosis framework integrating feature extraction, feature selection and classification is constructed. In the feature extraction stage, variational mode decomposition and fast Fourier transform are combined to capture both time-domain and frequency-domain features. In the feature selection stage, HBPSWOA integrates the local search efficiency of BPSO with the global exploration ability of WOA. This integration is further enhanced by introducing a Hamming distance-based position update mechanism and a roulette wheel selection strategy, improving solution diversity and robustness. Selected features are evaluated using k-nearest neighbours and support vector machine. The method is validated across three datasets, and its robustness is demonstrated through experiments with Gaussian white noise of varying intensities. Compared to four traditional feature selection methods (BPSO, BWOA, GA and BGWO), the proposed method achieves higher classification accuracy while selecting fewer and more informative features. This optimisation not only enhances classification accuracy but also improves computational efficiency, including under varying noise conditions. The highest accuracy of 99.51% was achieved on the CWRU benchmark dataset using an SVM classifier. Future research directions include exploring its scalability on larger datasets and leveraging deep learning classifiers to fully exploit the potential of the selected features, further enhancing diagnostic performance.