Online Open-Circuit Fault Diagnosis for ANPC Inverters Using Edge-Based Lightweight Two-Dimensional CNN
Chunxing Yao, Shuai Xu, Guanzhou Ren, Sijia Wu, Guohua Li, Zhenyao Sun, Guangtong Ma
Abstract
Conventional neural network (CNN) has been extensively applied in the field of fault diagnosis for multilevel inverter. However, most CNN based diagnostic strategies are typically implemented offline. To accomplish precise and online diagnosis for the open-circuit (OC) fault of three-level active neutral-point-clamped (3L-ANPC) inverter, the trained CNN is deployed into an edge computation board. Furthermore, this letter utilizes TensorRT to facilitate the lightweight design of the trained CNN, thereby accelerating the diagnosis speed. In order to simplify the offline training, a specific optimization framework is employed to achieve the automatic adjustment of hyperparameters. Comparative evaluations are performed to highlight the training performance and the generalization ability of the proposed CNN. Finally, the proposed online diagnosis is experimentally validated through a 3L-ANPC inverter fed permanent magnet synchronous motor (PMSM) drives.