| Asynchronous motor is a very widely used industrial equipment,is a necessary part of modern production,once there is a fault,not only will cause damage to the motor itself,but also will bring a very significant negative impact on the whole production system,so the study of asynchronous motor fault diagnosis method has important significance.Motor in the actual operation is often in a complex,variable load or speed control state,for this complex and variable operating state,the conventional diagnosis method based on signal processing is often aimed at a single fault form,cannot solve the problem of multiple faults.To solve this problem,an intelligent diagnosis method is proposed in this thesis.The simulation model is built by combining T-distributed random neighborhood embedding(t-SNE)and support vector machine(SVM)for rotor broken bar fault and bearing fault of induction motor,to study the intelligent diagnosis method of multiple faults of induction motor in early stage.The main work and research are as follows:(1)Research on feature vector extraction and preprocessing methods.The experimental results show that the current signal of the motor during operation contains a variety of characteristic parameters.Since the accuracy of motor fault diagnosis is related to the selection of characteristic parameters,to ensure the accuracy of fault diagnosis,this paper selects the method of time and frequency domain analysis to extract fault characteristics.To solve the problems such as dimensionality disaster caused by excessive information and precision reduction caused by redundant feature information,it is necessary to preprocess the extracted fault features.This paper selects four feature processing methods for dimensionality reduction of feature vectors,namely independent component analysis(ICA),principal component analysis(PCA),local linear embedding(LLE)and T-distributed random neighborhood embedding(t-SNE).The experimental results show that t-SNE can solve the problem of dimension disaster and feature information redundancy more effectively.(2)Select an intelligent classifier.Support vector machine(SVM)is based on the structural risk minimization criterion to obtain the minimum risk reality,and its topology is determined by the support vector,which can better solve the practical problems such as small samples,high dimension,and local minimum point.Therefore,the diagnosis model of support vector machine(SVM)is proposed and designed,and the feature vector processed by t-SNE is input into the SVM model for fault diagnosis.The classification accuracy of the fault motor reaches 94.67%.At the same time,it is compared with the extreme learning machine(ELM),probabilistic neural network(PNN)and Gaussian mixture classifier(GMM),which are three popular classifiers at present.The results show that the overall performance of SVM is obviously better than the other three intelligent classifiers.(3)Research on the application of intelligent diagnostic methods.To solve the problem that the classification performance of SVM is affected by parameter selection,whale optimization algorithm(WOA)is used in this thesis to optimize and select the parameters of SVM again,and WOA-SVM diagnosis model is obtained.The optimized classifier was used for fault diagnosis of rotor broken bar fault and bearing fault under different loads.Five load states were selected for detection,and the diagnosis results reached high accuracy.Especially under the condition of low load and no load,it can also diagnose the fault,which solves the problem that the traditional motor current signal characteristic analysis method cannot diagnose the fault under the condition of low load and no load. |
