Fault Diagnosis Method for Open-Circuit Faults in NPC Three-Level Inverter Based on WKCNN

Abstract

With the increasing demand for high reliability and availability in power conversion equipment within power electronics systems, the fault diagnosis of neutral-point-clamped (NPC) three-level inverters has garnered widespread attention. To address the challenges of fault feature extraction, this article proposes an end-to-end diagnostic approach based on a wavelet kernel convolutional neural network (WKCNN), capable of extracting multi-scale features from current signals to significantly enhance diagnostic accuracy. This method directly uses raw three-phase current signals as input, applying wavelet kernel convolution to automatically capture frequency-domain fault features, combined with a Softmax classifier optimized by the Adam algorithm to achieve fault diagnosis for NPC three-level inverters. Experimental results under various operating conditions demonstrate that this approach maintains robust diagnostic accuracy across multiple fault scenarios, with comparative analysis further confirming its advantages in diagnostic efficiency and performance over traditional machine learning and other deep learning methods.