Research On Inverter Fault Classification Based On Evidence Fusion

Inverter is a common power conversion device,It's widely used in locomotives,ele-vators,robots,electric vehicles and other fields,which plays a vital role in our life and indus-trial production.Power converter in the variable-frequency speed control system,especially of IGBT in the inverter,becomes a weak part in the whole system and more prone to fail-ure,because of high electric and thermal stress.Once the inverter IGBT gets faulty,it will cause the abnormal output current of the inverter,and even cause the whole system fault and paralysis,so it is very important to study the fault diagnosis and health management of inverter IGBT.Inverter fault diagnosis is divided into two steps: fault monitoring and pattern classifi-cation.This thesis mainly studies the method of pattern classification and the way to improve accuracy of classification.The knowledge-based fault diagnosis is used to classification of inverter IGBT fault type.Firstly,wavelet transform is used to analyze the three phase current,and to extract useful fault feature information and structure fault feature vector.Then,the artificial neural network,support vector machine and decision tree are used to find the mapping relationship between the fault feature vector and the fault type.The performance of those three machine learning methods is compared and analyzed.In the end,the theory of evidence is applied to the classification of inverter failure mode.the accuracy of based on evidence fusion is higher than the artificial neural network,support vector machine and decision tree,respectively..The thesis analyzes the accuracy of several different pattern classification methods in inverter IGBT fault classification of inverter.It has some guiding significance for the re-search of evidence fusion method in the field of fault diagnosis.