Demagnetization Fault Diagnosis Of Permanent Magnet Synchronous Motor Based On Sparse Self Coding

Permanent Magnet Synchronous Motors have the advantages of small footprint,high operating efficiency,strong power and large starting torque,and are widely used in high-tech industries such as aerospace,navigation and automobiles.However,due to the permanent magnet materials used in the permanent magnets,demagnetization failures are prone to occur in the case of high temperature and short-circuit between turns,which greatly affects the stable operation of the Permanent Magnet Synchronous Motor,and even leads to life safety in severe cases.This thesis studied the fault diagnosis of Permanent Magnet Synchronous Motor demagnetization has important practical significance for ensuring the long-term effective operation of Permanent Magnet Synchronous Motor and improving production safety and stability.This thesis aimed at the problem of demagnetization recognition caused by Permanent Magnet Synchronous Motor sample data,low availability,weak feature and complex structure,this thesis proposed a demagnetization fault diagnosis method that combines sparse self-encoding and least squares generative countermeasure network.On the one hand,in view of the long and difficult data collection process of demagnetization failure,a generation network is proposed.The network learns from real motor data samples and generates pseudo sample data that conforms to the distribution of real motor samples;On the other hand,in view of the unstable training and low accuracy of the traditional fault diagnosis network,sparseness is introduced into the self-encoding network as the permanent magnet synchronous motor demagnetization fault diagnosis model.The experimental results prove that the permanent magnet synchronous motor demagnetization fault is finally realized diagnosis.This article explains the fault diagnosis method of permanent magnet synchronous motor from the following three parts:(1)Characteristic analysis of Permanent Magnet Synchronous Motor.First,this thesis discussed the three mathematical models of permanent magnet synchronous motors,which analyze the structure and characteristics of permanent magnet synchronous motors,Then,this thesis studied the characteristic parameters of the permanent magnet synchronous motor,and the magnetomotive force,electromagnetic torque and other parameters are collected to form a limited sample set combination.This thesis analyzes and discuss the relationship between each feature and the demagnetization of the permanent magnet synchronous motor using mathematical methods to prove the rationality of the selected feature parameters.(2)Research on generative countermeasures of network data expansion.This thesis selected a generative countermeasure network to generate generated data with the same distribution as the real motor sample data,and clarify the advantages and disadvantages of the generated network.This thesis in view of the shortcomings,two improved generative confrontation networks are proposed,the least squares generative confrontation network and the conditional generative confrontation network.This thesis using the experimental results,the least squares generative countermeasure network is compared with the original generative countermeasure network,input real motor sample data,optimize the network structure,fine-tune parameters,and perform iterative training to obtain the best network model.Since the least squares generative confrontation network has a faster convergence rate,the generated data is of high quality,and the training is more stable,this thesis selected the least squares generative confrontation network is finally as the expansion network.(3)Research on sparse self-encoding network fault diagnosis method.The introduction of sparsity in the traditional self-encoding network is more helpful to express the structural characteristics of the input data.The weights are continuously modified by the gradient descent algorithm,iterative training is carried out,the weights are optimized,and the objective function is minimized.And this thesis finally get the network parameters in different states.This thesis by comparing the test results,the use of sparse self-encoding network as a fault diagnosis network is more accurate,more feasible,and more maneuverable than traditional networks.