Hybrid Genetic and Binary State Transition Algorithm With Memory Functions for Machine Learning Applications in Diagnosing Bearing Faults

Abstract

In the field of bearing fault diagnosis, effectively extracting critical information from raw motor signals while ensuring high accuracy and minimising computational resources remains a significant challenge. This study proposes a novel bearing fault diagnosis model consisting of three main stages: feature extraction, feature selection, and classification. In the feature extraction stage, empirical mode decomposition (EMD), Hilbert–Huang transform (HHT) and fast fourier transform (FFT) are utilised to extract features from raw motor signals. In the feature selection stage, a novel hybrid feature selection method combining genetic algorithm (GA) and binary state transition algorithm (BSTA) is proposed enhancing the model's performance. This research has also added a new memory function to the algorithm to avoid unnecessary computational waste. In the classification stage, k-nearest neighbours (k-NN) and support vector machine (SVM) are employed to evaluate the classification accuracy after feature selection. To validate the performance of the proposed model, experiments were conducted on four bearing fault datasets, including the University of California Irvine (UCI) benchmark dataset, Motor Bearing Fault Current Signal Dataset, Case Western Reserve University (CWRU) benchmark dataset and Mechanical Fault Prevention Technology (MFPT) benchmark dataset. In case study 1, using the UCI dataset for testing, GBSTA-M reduced computation time by up to 94% compared with traditional algorithms. In case study 3, GBSTA-M combined with SVM achieved an accuracy of 98.7% on the MFPT dataset. Experimental results demonstrate that, compared to conventional methods, the proposed model not only achieves higher fault diagnosis accuracy but also significantly reduces computational resource requirements in specific scenarios while exhibiting excellent robustness.