ABSTRACT In this paper, a fault diagnosis method based on a spatial trajectory diagnosis network model is proposed for T‐type three‐level inverters. By providing a rigorous theoretical derivation of the phase current distortion under switching constraints and neutral‐point voltage shifts, the inherent mapping from time‐domain waveform clipping to 3D trajectory deformation is revealed. A classification network model is employed to process these spatial point sets using symmetric functions for robust fault recognition. The proposed method is validated on a dSPACE experimental platform, showing that diagnostic accuracy remains above 92.5% even under 10% white‐noise interference. Furthermore, a comparative study with existing methods demonstrates that the proposed approach achieves an optimal balance between hardware cost and diagnostic depth, identifying both single‐ and multi‐switch faults within 16.2 ms without requiring additional voltage sensors.
Zhang et al. (Thu,) studied this question.