Extreme Learning Machine Models for Fault Severity Classification and Anomaly Detection in a Spur Gearbox

Research on fault severity in gearboxes is a crucial area of study for the successful implementation of predictive maintenance and the optimization of maintenance strategies. In this context, our research compares several extreme learning machines (ELM) for classifying fault severity in two types of faults: pitting and broken teeth. We have included other classical machine learning models in this comparison and also evaluated their capability for anomaly detection by comparing them to classical algorithms such as isolation forest (iForest), one-class SVM (OC-SVM), robust random cut forest (rrcForest), and local outlier factor (LOF). Our study, which utilized nonlinear information entropy (InfoNLEntropy) features and statistical features extracted from vibration signals, demonstrated that ELM models are accurate and highly efficient for classifying fault severity and detecting anomalies in the early stages of fault progress.