Advanced Fault Diagnosis in Power Electronics: Switch Open Faults in DC-Link Shunt Sensor-Less Drives

Abstract

This paper presents a robust fault diagnosis method employing a Temporal Convolutional Network (TCN) for diagnosing switch open faults in single-shunt sensor-less drives. Traditional diagnostic approaches relying on three-phase output currents and motor speed data from sensors face challenges due to the growing use of single shunt resistors and sensor-less algorithms. These changes make diagnosing switch open faults difficult, leading to unreliable load current reconstruction and distorted inverter switching patterns. Moreover, the accuracy of phase reconstruction is crucial for precise angle and speed estimation, affecting the regularity of post-fault current waveforms. To address these challenges, this study proposes a deep neural network-based diagnostic strategy centered around the TCN structure’s capacity to capture long-term dependencies in DC current waveforms. This approach overcomes the limitations posed by the absence of traditional current and speed sensors, providing a comprehensive solution for accurately identifying post-fault behavior in single-shunt sensor-less drives. Test results demonstrate the feasibility and diagnostic performance of the proposed TCN-based neural network, achieving high accuracy and F1 scores compared to the multilayer perceptron architecture. This research addresses the evolving challenges in fault diagnosis and contributes to advancing the field of power electronics by providing a reliable diagnostic solution for modern systems.