Backpropagation Neural Network-Based Open-Circuit Fault Diagnosis for a Five-Level NPC Inverter

This paper introduces a technique for diagnosing open-circuit (OC) faults in a five-level Neutral Point Clamped (NPC) inverter, employing a Back-Propagation Neural Network (BP NN). This inverter is used for energy conversion in an autonomous microgrid with a single Distributed Generator (DG). The five-level NPC inverter has many advantages compared to the two-level inverter. However, it is more likely that one of its insulated gate bipolar transistors (IGBTs) will be affected by an OC fault. The fault detection technique proposed in this study uses normalized diagnostic errors to extract the features. Then it classifies them as inputs to the BP NN that recognizes the OC fault. Normalized diagnostic errors are calculated from the expected and measured line-to-line pole voltages. The accuracy of detecting the OC fault in all the IGBTs in a short time is demonstrated by the simulation results performed by MATLAB.