Research On Multi State Fault Identification Of Gear Box Based On State On-line Monitoring

With the continuous progress of science and technology,the mechanical system becomes more and more complex and automated.As the indispensable key element in the mechanical equipment,the gear has multiple states during its operation.How to identify faults quickly and accurately has become an important issue at present.Because of the complex structure and bad working environment,the gear transmission system has high failure probability during its operation.Many of the equipment failures are caused by the failure of gears,which cause great losses.Therefore,this paper takes the gear box as the research object,and extracts the characteristic parameters that can represent the running state from the vibration signal.Then it is used as the input information of the established deep convolution neural network(DCNN)model to identify the gearbox fault.It is proved by experiments that the deep convolution neural network is an effective fault identification method.The main work of this paper is as follows:(1)The gear box vibration test was completed on the UT6818 machine fault simulation test bed and the test data were analyzed preliminarily.Through analyzing the failure mode and failure state of gear box,the tooth crack is chosen as the failure mode of experiment,and according to its crack length,it is divided into five states.The radial and axial vibration signals of the gear of different states are collected with different loads.The collected signals are analyzed from two aspects of the time domain and the frequency domain.(2)The feature vectors that can characterize the state of the gear are extracted by a variety of methods.Firstly,twenty kinds of feature parameters are extracted from the vibration signal by time and frequency domain analysis method.The principal component analysis method is used to optimize its dimension,and four principal elements are selected with the cumulative contribution rate of 90.33%.Secondly,the wavelet packet decomposition method is used to decompose it in five layers,and 32 frequency bands energy are extracted as the characteristic parameter.Thirdly,the linear spectrum analysis method is used to extract rotational frequency,two times rotation frequency,three times rotation frequency and meshing frequency as characteristic parameters.Finally,the feature parameters extracted from the three analytical methods are combined into one feature vector to characterize the gear state.(3)A deep convolution neural network(DCNN)model is established to identify the gear state.The established DCNN model is trained and tested by the extracted feature vectors.The support vector machine(SVM)theory is compared with DCNN model to verify the effectiveness of the established DCNN model.It is concluded that the correlation between radial vibration signal and gear tooth crack fault is higher by comparing the recognition accuracy of radial and axial vibration signal.It is known that the appropriate load can improve the accuracy of state recognition by comparing the signals collected under different loads.