Bearing Fault Diagnosis Based on Artificial Intelligence Methods: Machine Learning and Deep Learning

This paper presents a comprehensive study on the application of Artificial Intelligence (AI) methods, specifically machine learning and deep learning, for the diagnosis of bearing faults. The study explores both data preprocessing-dependent methods (Support Vector Machine, Nearest Neighbor, and Decision Tree) and a preprocessing-independent method (1D Convolutional Neural Network). The experiment setup utilizes the Case Western Reserve University dataset for signal acquisition. A detailed strategy for data processing is developed, encompassing initialization, data loading, signal filtration, decomposition, feature extraction in both time- and frequency-domains, and feature selection. Indeed, the study involves working with four datasets, selected based on the distribution curves of the indicators as a function of the number of observations. The results demonstrate remarkable performance of the AI methods in bearing fault diagnosis. The 1D-CNN model, in particular, shows high robustness and accuracy, even in the presence of load variations. The findings of this study shed light on the significant potential of AI methods in improving the accuracy and efficiency of bearing fault diagnosis.