Comparative Study of Classical and Deep Learning Methods for Bearing Fault Diagnosis of Electrical Machines

Fault diagnosis in industrial systems using vibration signal data has attracted growing attention due to the complexity of fault classification. While classical machine learning (ML) methods have been widely used for electrical machine fault detection, deep learning (DL) models, particularly Transformers, have recently demonstrated superior performance. This study presents a comparative evaluation of classical ML approaches, including Artificial Neural Networks, Support Vector Machines, K-Nearest Neighbors, Decision Trees, Random Forest, and Logistic Regression, against DL models such as Convolutional Neural Networks (CNNs) and Transformers. Using the Case Western Reserve University bearing dataset, models were assessed using accuracy, precision, recall, and F1-score. Results show that DL models significantly outperform classical methods, achieving 99% across all metrics, whereas classical models yielded notably lower accuracy (52-63%). The findings demonstrate the effectiveness of CNNs and Transformers for high-accuracy, generalizable fault diagnosis in electrical machines.