Multi-Open-Circuit Fault Diagnosis in Three-Level Neutral-Point-Clamped Inverters Using Single-Waveform Augmentation via Edge Computing

This study proposes a data-driven fault diagnosis (FD) approach for multi-open-circuit faults (OCFs) in three-level neutral-point-clamped inverters. The proposed method computes a single transformed current value obtained from samplewise normalizing, scaling, and summing three-phase currents. A total variation loss-regularized atrous convolution neural network is trained using feature-augmented datasets, involving data masking and elastic transformations, which are partly composed of transient waveforms. The inference process uses prioritization, multiprocessing, synchronization, and buffer locking to handle data in a circular buffer, and it is deployed on a high-performance and economic neural network inferencing accelerator, Hailo 8. The simulation and experimental results demonstrate that the proposed method achieved high speed and accuracy of 99.5% and 99.47%, respectively, for RL and induction machine. The OCF detection diagnosis time is measured to be 6.31 ms. Finally, comparative analyses are performed to highlight the key advantages of the proposed method in terms of FD speed, number of faults detected, input requirements, and inference platform.