Research On Fault Diagnosis Of Three-level Inverter Based On Adaptive Network-based Fuzzy Inference System

In recent years, there exists a contradiction between the fact that voltage level of power electronics device does not make a major breakthrough and the fact that there is a growing demand of higher voltage and higher power application. Therefore, the focus of scholars is on ameliorating inverter topology and control strategy to improve output voltage and power levels under the voltage level of power electronic device. Thus multilevel inverter technology emerges.Because multilevel inverter contains a lot of power devices and the power devices work at the environment of high voltage and high frequency for a long time, inverter power device failure has become a problem that cannot be ignored. For multilevel inverter structure is complex, it is difficult to quickly determine power device failure position. So, fault diagnosis of multilevel inverter has been a popular research interest.The most widely used diode clamping three level inverter is adopted as the object of study. Fault types and causes of inverter are analyzed in this thesis. The open circuit faults of inverter are separated and numbered. By Matlab simulation, fault feature space is generated, and four conclusions are made. Secondly, the fault diagnosis method based on Subtractive Cluster-Adaptive Network-based Fuzzy Inference System(SC-ANFIS) is studied. The three-phase average output current is used as the input of fault diagnosis. By Clarke transform, the dimensionality of fault information is reduced. Considering the cluster characteristic of the fault information, the author adopts the method of fuzzy clustering to obtain fault center and membership functions for each fault type, and uses Adaptive Network-based Fuzzy Inference System as recognition tool. By simulation, the effectiveness of the method is verified. In addition, to solve the problem that different loads have different fault characteristic value, this thesis introduces two methods to transform the characteristic value of different loads into saved characteristic space, which avoid training the fuzzy neural network in different loads and optimize the efficiency of the method. Lastly, the author develops inverter fault diagnosis system by using C#, Oracle database and Matlab. The program has the function of fault record information saving and query, fuzzy neural network training and testing, and fault diagnosis of inverter online or offline. Through testing, it is concluded that based on SC-ANFIS, the method fulfills the task of multilevel inverter fault diagnosis and records the time of fault occurrence, which provide the basis for repairing the fault and improving the circuit. The way of hybrid programming method provides a certain reference value for multilevel inverter fault diagnosis.