This paper presents an enhanced broken-bar fault diagnosis approach for Asynchronous Motors, based on a simulation-driven Support Vector Machines (SVM) framework. To circumvent the lack of experimental fault data, a high-fidelity dynamic model is developed: the healthy multi-winding Asynchronous Motors is extended with fault-specific resistances to emulate 0, 1, 2, and 3 broken rotor bars under varying operating conditions. Electromechanical and electrical signals signatures (stator currents, speed, torque) are simulated and transformed into discriminative feature vectors, forming a labeled training dataset. A supervised SVM classifier is then implemented in two successive stages, first a binary classification to detect healthy versus faulty states and second a multi-class classification (one-vs-one) to identify the exact number of broken bars. Results show high accuracy when torque and speed features are combined, highlighting the method’s robustness and data efficiency. By integrating physical modeling with interpretable machine learning, the proposed approach offers a reproducible and scalable solution for predictive maintenance in industrial motor systems.
Amrane et al. (Tue,) studied this question.