Mixed-Type Open-Circuit Fault Diagnosis for NPC Inverters Using a Dual-Input CNN

Abstract

The different open-circuit fault causes (i.e. terminal block damage, bonding wire fracture, driver circuit failure, et al.) lead to both one-way open-circuit and two-way open-circuit in neutral-point-clamped (NPC) inverters, namely the mixed-type open-circuit fault. In order to achieve a high diagnosis accuracy, the two types of open-circuit manifestations are simulated with two simulation ways to acquire mixed monitoring signal samples, on this basis, a diagnosis method using a dual-input convolutional neural network (CNN) is proposed. The method uses the three-phase currents as the main monitoring signals, and the six midpoint voltages of the upper and down half bridge arms as the auxiliary monitoring signals for the open-circuit fault diagnosis. The effectiveness of the method is validated by simulation, and the effects of different sample durations and training sample numbers on the network training and diagnosis results are analyzed quantitatively.