Research On Fault Method Of Three-phase Asynchronous Motor Based On SSA And Multi-working Conditions

Three-phase asynchronous motor is the main power source of industrial production,providing a constant source of power for mechanical equipment,promoting the vigorous development of the economy and society,and driving the development process of the new era.However,three-phase asynchronous motors are often under high load,long working hours and other harsh working conditions,which can easily cause motor failure and lead to production accidents,resulting in economic losses.The fault diagnosis of three-phase asynchronous motor can detect early faults,eliminate them in time and reduce the accidents caused by motor faults.In this paper,the three-phase asynchronous motor model YE2-100L2-4 is used as the research object.By building an experimental platform for the three-phase asynchronous motor,designing the motor state and collecting vibration data according to the actual fault type of the motor,and using research methods such as variational modal decomposition(VMD),convolutional neural network(CNN),sparrow search algorithm(SSA),support vector machine(SVM),and feature fusion,we achieve fault diagnosis of three-phase asynchronous motor based on SSA and multiple operating conditions.The main contents of this paper are as follows.Firstly,based on the working principle of three-phase asynchronous motor and common fault types and their fault mechanisms,this paper designs three kinds of faults: bearing fault,broken rotor strip,and air gap eccentricity,builds an experimental platform for three-phase asynchronous motor,and completes the data acquisition of corresponding working conditions.A motor fault diagnosis method based on the fusion of sparrow search algorithm(SSA),variational modal decomposition(VMD)and support vector machine(SVM)is adopted to address the instability of three-phase asynchronous motor signals and the difficulty of fault feature extraction.The method optimizes the hyperparameters Alpha and K of VMD by SSA,and then uses the optimized VMD to decompose the signal,and then finds the energy of each IMFs after decomposition as features for SVM identification.After experimental analysis,the motor vibration signal verifies that the used model has good recognition capability.Secondly,for the motor signal features are not obvious and easily disturbed by background noise,there are limitations in analyzing faults by empirical extraction of features using the traditional method,which relies on the experience of experts or engineers,so the F-CNN-SVM based three-phase asynchronous motor fault diagnosis method is used.The FFT transform is performed on the data first,and then the CNN can be used to extract features adaptively through the internal convolution layer,and the dimensionality of the features can be reduced through the pooling layer,while the generalization ability of the model is improved.The method of automatic extraction of fault features using CNN circumvents the subjectivity and limitations of manual feature selection,while combining the advantages of SVM’s stronger multi-classification performance than softmax.And the feasibility of the method is verified by the vibration data and current data collected from the motor test bench,respectively.Finally,before using CNN for feature extraction,the hyperparameters of its model need to be set,however,the performance of models with different hyperparameter configurations often varies greatly,while there are problems such as the complexity of multiple sources of feature signals.Based on the above problems,an SSA-CNN-SVM based three-phase asynchronous motor fault diagnosis method is used.The method uses SSA for CNN hyperparameter search,then extracts the vibration feature vectors and current feature vectors of multiple source heterogeneity by CNN,and fuses them using the weighted parallel feature fusion method,and finally uses SVM for fault diagnosis.It is verified using the vibration and current datasets collected from the motor test bench,and the recognition rate of the method is100%,which proves the superiority of the method.