IGBT Open-Circuit Fault Diagnosis for MMC Based on OCDWPT

Considering the issue of high sensor demand and susceptibility to load changes in existing IGBT fault diagnosis for MMC sub-module (SM), a fault diagnosis method based on over-complete discrete wavelet packet decomposition (OCDWPT) is proposed in this paper. Firstly, based on the research on the fault mechanism of MMC, the bridge arm voltage is used as the diagnostic parameter, eliminating the need for collecting capacitor voltage of each SM, thereby reducing the dimension of the fault diagnostic signal. Secondly, the diagnostic parameter is decomposed by OCDWPT and converted to time-frequency domain information. Additionally, the iterations of down-sampling are determined based on practical requirements to minimize computational load and storage space while preserving signal phase characteristics as much as possible. Finally, fault diagnosis is performed using an optimized deep neural network.