Hybrid Fault Detection and Diagnosis Approach of Power Connections for Induction Motors

The increasing demand for reliable electric mobility solutions highlights the need for advanced fault detection and diagnosis (FDD) strategies to ensure system reliability and safety. This paper presents a hybrid approach that integrates physics-based models with data-driven techniques for FDD in induction motors (IMs). A Hardware-in-the-Loop (HiL) platform is used to generate synthetic data that replicates real-life operating conditions. In this study, a progressive classification strategy based on a One-Class Support Vector Machine (OCSVM) is trained exclusively on healthy operation data and tested with HiL-generated faulty responses to evaluate its anomaly detection capabilities. Focusing on railway traction systems, the research simulates common IM faults, including opposite-phase wiring, high-resistive connections, and open-phase failures. By extending the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology with physics-based model validation and synthetic data generation, the proposed hybrid strategy enhances scalability, effectively addressing challenges such as limited faulty data and inadequate real-time monitoring. This approach demonstrates significant potential for improving fault detection in electric traction applications.