Modeling and Simulation of Mechanical Failures in Three-Phase Induction Motors Using Support Vector Machines

Espinoza C.; Aranda J.; Albatrino R. · Proceedings of the 2025 IEEE 32nd International Conference on Electronics, Electrical Engineering and Computing, INTERCON 2025 (2025) · DOI: 10.1109/intercon67304.2025.11244702

Resumen

The early detection of mechanical faults in three-phase induction motors is essential for enhancing operational reliability and reducing maintenance costs. This study presents a simulation-based methodology for fault diagnosis using supervised machine learning, specifically Support Vector Machines (SVMs). A 3 HP, 220 V, 60 Hz induction motor was modeled in Simulink, including detailed electrical and mechanical dynamics. Two types of mechanical failures - rotor imbalance and bearing defects - were simulated through controlled perturbations in torque and stiffness. From the resulting vibration and temperature signals, a total of 15 time, frequency, and thermal domain features were extracted in segmented windows. The classification model was trained on a synthetic dataset comprising 300 labeled samples representing three motor conditions: healthy, imbalance, and bearing failure. Using a radial basis function (RBF) kernel, the SVM classifier was optimized via cross-validation and Bayesian search. The model achieved an overall accuracy of 96.7% on the test set, with an area under the ROC curve (AUC) of 1.00 for all classes. The results demonstrate high diagnostic performance, even with a limited and entirely simulated dataset. This work validates the feasibility of using SVMs for accurate fault classification in induction motors based solely on vibration and temperature features. Although tested in a simulated environment, the method shows promising potential for future real-time implementation in predictive maintenance systems. Future work will focus on experimental validation and integration of additional signal types, such as motor current, to improve robustness and extend applicability to real industrial scenarios.

Instituciones: Universidad Peruana de Ciencias Aplicadas