Bearing Fault Detection and Classification Using Machine Learning Neural Networks – Application to Rotating Machines –

Abstract

This thesis focuses on the application of artificial intelligence, particularly Multilayer Perceptron (MLP) neural networks, for the detection and diagnosis of bearing faults in rotating machinery. Bearings, as key components of industrial systems, require early fault detection to prevent costly failures and production downtime. Although traditional methods are effective, they remain limited by their reliance on human intervention and predefined failure modes. This research proposes an automated approach based on machine learning using vibration signals to detect and classify faults. Inputs to the MLP networks are extracted through statistical analysis of the signals, using parameters such as maximum, minimum, and median. Several network architectures are tested on datasets, including the Case Western Reserve University dataset. The results demonstrate that these methods improve diagnostic accuracy, opening promising prospects for predictive maintenance. Finally, the study examines the implications of the findings and suggests ways to enhance the performance and robustness of fault detection systems.