A Data-Driven Open-Circuit Fault Diagnosis Method for Three-Phase T-type Three-level Inverters

This paper presents a data-driven diagnosis method for single-switch open-circuit faults in T-type three-level inverters based on current vector trajectory and convolutional neural network (CNN). First, a simulation model based on T-type topology is built and the three-phase current waveforms under different model parameters are recorded as the dataset. The current vector trajectories are extracted through Clarke transform as the feature (60 × 60). The trajectories are then classified using a convolutional neural network to identify the fault. The method achieved an average validation accuracy rate higher than 99% and is validated through simulation and experiment. It is robust to white noise and changes in model parameters, dc-link voltage, modulation rate, and frequency.