Multilevel inverters (MLIs) have received a significant attention in literature due to advent of various topologies utilising power semiconductor switches with reduced voltage stress. However, the increased component count and susceptible nature of semiconductor components increase the chance of failure in inverter. The fault in power switches, either open circuit or short circuit, if persist for long time may result in complete system shutdown. This damage can be avoided by detecting and diagnosing the fault in less period of time to continue the healthy operation. The present paper studies an open-circuit fault detection method on conventional five-level NPC inverter topology with output voltage waveform analysis by discrete wavelet packet transform (DWPT) and classifying the fault location by artificial neural network (ANN). The strategy to tolerate both abovementioned type of failure on any location after detection is proposed by implementing a novel redundant architecture in conventional five-level inverter. This architecture has an advantage of utilising less device count to tolerate fault on single as well as multiple switch combinations. The proposed work is studied in MATLAB/Simulink and validated experimentally.

由于采用了具有降低的电压应力的功率半导体开关的各种拓扑的出现，多电平逆变器（MLI）受到了文献的极大关注。但是，增加的组件数量和半导体组件的易感性增加了逆变器发生故障的机会。电源开关的故障（开路或短路）如果持续很长时间，可能会导致系统完全关闭。可以通过在更短的时间内检测和诊断故障以继续正常运行来避免这种损坏。本文研究了一种传统的五级NPC逆变器拓扑开路故障检测方法，该方法通过离散小波包变换（DWPT）分析输出电压波形，并通过人工神经网络（ANN）对故障位置进行分类。通过在常规的五电平逆变器中实现一种新颖的冗余架构，提出了在检测后在任何位置都可以容忍上述两种类型的故障的策略。这种架构的优势是利用更少的设备数量来容忍单个或多个交换机组合上的故障。拟议的工作在MATLAB / Simulink中进行了研究，并进行了实验验证。