Four-state active fault diagnosis method for cascaded H-bridge multilevel inverter

Abstract

The cascaded multilevel inverters have many semiconductor switches with high failure rate, which require timely diagnosis techniques. It is a difficult point that the diagonally positioned switches have similar fault characteristics. By analyzing the output characteristics under fault, it is found that the position and the cause of similar faults are related to the permutation of carrier in pulse width modulation. A four-state active fault diagnosis method is proposed in this article. First, the permutation of carriers is reconstructed to activate the fault behavior after the fault is detected. Under the specific four modulation states, the fault characteristics can be fully behaved. Next, based on the machine learning algorithm, the effective fault features are extracted from the voltage signals in four modulation states and the fault location is achieved. The experiments show that the proposed method could effectively distinguish similar faults, improve diagnostic accuracy and have better robustness.