Bearing fault diagnosis by machine learning and deep learning-based models: A comparative study applying for HUST bearing dataset

Abstract

Diagnosing bearing faults is essential for ensuring the reliability and operational safety of mechanical and electronic systems. This paper presents a comparative analysis of different machine learning-based models for classifying bearing fault conditions, including Support Vector Machines (SVM), Long Short-Term Memory (LSTM) networks, One-Dimensional Convolutional Neural Networks (1D-CNN), Two-Dimensional Convolutional Neural Networks (2D-CNN), and Transformer model. These models are applied to the HUST bearing dataset and evaluated based on their ability to accurately classify defects from vibration signal data. The results indicate that 1D-CNN, 2D-CNN, and Transformer model exhibit superior performance in bearing fault diagnosis. 1D-CNN attained 99.8% accuracy on the training set and 99.83% on the test set, followed by 2D-CNN with 99.1% and 99.3%, respectively. The Transformer model also performed well, reaching 99.7% accuracy within 1 hour of training, similar to 1D-CNN (1 hour) and 2D-CNN (0.8 hours). In contrast, LSTM and SVM exhibited lower accuracy and significantly longer training times, with LSTM requiring 11.5 hours and SVM 8 hours. These findings suggest that 1D-CNN, 2D-CNN, and the Transformer model are highly effective approaches for bearing fault diagnosis, with the Transformer model achieving performance and training efficiency comparable to CNN-based models.