Incipient Bearing Fault Diagnosis in Induction Motors Under Variable Loads Using Vibration Signal-Based Machine Learning

Rotating machinery forms a critical component across numerous industrial sectors. Among these, induction motors are widely utilized in various applications, with bearings being one of their most essential and failureprone components. Bearing faults represent a significant proportion of induction motor failures. Existing research in this domain predominantly focuses on well-defined fault categories such as inner race, outer race, ball, and cage faults. However, in real-world industrial environments, bearings often degrade gradually due to early-stage surface anomalies, including rust, corrosion, surface roughness, and waviness, which are frequently overlooked in conventional studies. This study addresses the important issue of early-stage bearing fault diagnosis under varying mechanical loading conditions: no load, 50% load, and full load, 100%. It employs tri-axial (X, Y, Z) vibration signal analysis, from which statistical and derived features are extracted. Feature selection is performed using recursive feature elimination to identify the most relevant parameters. The selected features, representing six fault classes, are then fed into multiple diagnostic machine learning algorithms, including Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), as well as Deep Neural Networks (DNN), for classification and performance evaluation. Experimental results demonstrate that both DNN and RF outperform XGBoost and SVM, with marginal reductions in accuracy. The proposed approach provides an effective solution for the automated detection of incipient bearing faults in induction motors, thereby helping to prevent unexpected breakdowns and major system failures resulting from bearing degradation.