Open Switch Fault Diagnosis Method for PWM Voltage Source Rectifier Based on Deep Learning Approach

Abstract

With the development of machine learning technology, numerous studies have been proposed to diagnose the open circuit (OC) faults in the pulse width modulation (PWM) voltage source rectifier (VSR) systems. However, most methods require system signals of more than one current period, which show poor real-time performance. Aiming at this problem, this paper presents an improved diagnosis system based on deep belief networks (DBN) and least square support vector machine (LSSVM). First, the double chain quantum genetic algorithm (DCQGA) is employed to obtain the proper length of measured signals and DBN structure parameters. Then, the fault features are extracted from the signals through DBN. Finally, these features are used to train the LSSVM fault classifier to construct the diagnosis model. The experimental results show that the proposed method can achieve the fault diagnosis including six kinds of single switch faults and 15 kinds of different double switches faults correctly. Besides, the proposed method also shows the superior anti-interference performance and high robustness on abrupt load transient conditions, unbalanced, and/or distorted grid voltage conditions, as well as, different power factor conditions. Furthermore, the average diagnostic time of this method is only 2.57 ms.