An Artificial Neural Network with Bayes' Theorem (Hybrid) for Automated Bearing Faults Diagnosis

Abstract

Bearing fault diagnosis plays a pivotal role in the realm of condition-based maintenance, with vibration spectra analysis standing out as a highly effective method for discerning issues in rotating machinery. Various signal processing tools, including wavelet analysis, empirical mode decomposition, and Hilbert-Huang transform, have been employed to scrutinize vibration spectra. However, these methodologies often necessitate human expertise to ensure optimal outcomes. In the pursuit of automated fault diagnosis, machine learning tools such as Artificial Neural Networks (ANN) and support vector machines (SVM) have emerged as viable alternatives.  Over recent decades, considerable research efforts have been devoted to exploring the viability of employing Artificial Neural Networks (ANN) for automatic fault diagnosis, resulting in predominantly positive findings. Nevertheless, the accuracy of Artificial Neural Networks (ANN) is intricately linked to factors such as the neural network structure, encompassing considerations like the number of nodes, hidden layers, and the choice of activation function. Addressing this challenge, the present study introduces an innovative hybrid algorithm for automated bearing fault diagnosis, integrating Artificial Neural Networks (ANN) with the Bayes’ Theorem (BT) theory.  The hybrid algorithm exploits Bayes’ Theorem (BT) theory to augment fault diagnosis results derived from Artificial Neural Networks (ANN), specifically by alleviating conflicting outcomes generated within the neural network. The study focuses on characterizing four conditions of bearings, namely a healthy state and three distinct fault types: rolling element, inner race, and outer race faults. Through the proposed hybrid algorithm, working in tandem with artificial neural networks, a demonstrated superiority is established, surpassing the results obtained by Artificial Neural Networks (ANN) in isolation. This pioneering approach not only underscores the potential for heightened accuracy but also underscores the enhanced reliability achievable in automated bearing fault diagnosis.