Research On Feature Optimization Method And Its Application In Fault Diagnosis Of Motor System

Nowadays,the motor is an important motive force and driving device in production activities and daily life.The failure of motor or the stop of operations will not only cause damage to the motor itself,but also affect the whole system,even cause social problems such as casualties.Motor fault diagnosis technology can find the problem at the initial stage of the fualt.Hence,targeted maintenance can be performed timely,which can save a lot of time and money for the repair of fault,avoid the stop of production and improve economic benefits at the same time.The feature optimization is a useful pre-processing step for fault diagnosis,where the irrelevant and redundant information in the data can be reduced.In this case,not only the computation complexity can be reduced,but also the better classification performance can be obtained.Nowadays,there are many dimension reduction methods,such as locally linear embedding,principal component analysis,linear discriminant analysis and deep learning.This paper focuses on the principal component analysis algorithm,linear discriminant analysis algorithm,deep learning algorithm and their application in fault diagnosis of motor system.The main research contents are as follows:1.This paper focuses on the common faults of frequency conversion motor,the mechanism analysis of five common faults and the fault characteristics are summarized;2.The principle of principal component analysis algorithm and linear discriminant analysis algorithm is described and summarized.Linear discriminant analysis algorithm is improved based on the predecessors,where the singular value decomposition is introduced to solve the problem of singularity,and the between class divergence matrix is redefining to solve the problem that the similar classes are not easy to separate.Finally,the improved linear discriminant analysis algorithm and principal component analysis algorithm are combined to obtain a new hybrid dimension reduction algorithm to optimize the features;3.The principles and modeling methods of four types of deep learning models(deep belief networks,self-encoding networks,convolutional neural networks,and recurrent neural networks)that are currently common and widely used are detailed described.The advantages and disadvantages of these four types of deep learning model are compared.The convolutional neural network algorithm and the Long-short-time memory algorithm of deep learning are selected for motor fault diagnosis.4.The motor fault diagnosis based on feature optimization is simulated by MATLAB/simulink.An experimental platform is designed and built to further verify the effectiveness of the proposed feature optimization algorithm.