Envelope-Based Fault Detection and Identification in Three-Phase Motor Drive Systems

The three-phase inverter-based induction motor drive systems are critical to modern electric vehicle powertrains. Reliable fault detection, isolation and identification are vital to reduce the safety risks associated with sudden failures within this system. This article introduces an online fault detection, isolation and identification method for three-phase induction motor drive systems, combining physics-based insights with datadriven approaches. Specifically, we employ partial least squares (PLS) regression, trained under healthy conditions, to predict powertrain signals using the salient input/output relationships within the system. We compare predictions to observed values, and the discrepancies inform anomaly detection through a cumulative sum statistical test. Using well-established models and the physical structure of the system, this approach enables precise real-time identification of both the type of fault and its location within the system. We validate the proposed framework with experimental data collected from a power control unit and motor on a dynamometer test bench under the US06 driving schedule, where we increased the cooling loop temperature over tests until the system failed. The experiments verify the effectiveness of the proposed fault detection, isolation and identification method.