Rotating Machinery Fault Diagnosis Research Based On Wavelet And Artificial Neural Network

Rotating machinery is widely applied in machinery, metallurgy, electric power, chemicaland other industries. If the rotating machinery failure occurred without timely control andelimination, it could led to equipment damage, not only causing enormous economic losses,and even endangering the personal safety, thus the consequences is very serious. Therefore itis of great significance to research diagnosis method of rotating machinery fault. Because therotor axis orbit can reflect the type of rotor fault, this paper research on the automaticdiagnosis method of rotor fault starting from the axis track. The main contents andconclusions are summarized as follows:(1) Axis orbit purification: Using the wavelet method to purify the shaft orbit. Firstly thepurification effects are compared when choosing different wavelet functions and decomposedto different layers, and the best wavelet basis and decomposition level is selected out. Then,according to the characteristics of the rotor vibration signal, the influence of samplingfrequency and rotation speed for purification effect is analyzed, and signal resampling methodis introduced to eliminate the affect, ensuring the purification level;(2) Axis orbit feature extraction: On the base of resampling during the step of axis orbitpurification, the signal is down-sampled further. And the polar radius of axis orbit (thedistance between points on axis orbit and basis points O) is calculated for each rotation cycle,then it is normalized about pan, stretching and rotation. The normalized polar radius sequenceis used as the characteristics of axis orbit. The result of emulation experiment shows that thenormalized polar radius sequence have different variation for different axis orbits, and thesequence of the polar radius which has been down-sampled remain the same samples lengthin each rotation cycle, which facilitate to identify of the next step;(3) Identification of axis orbit: With polar radius axis orbit sequence used as samples, the establishment process of the corresponding BP neural network is discussed. Then the networkis applied to train and recognize the samples, and the results show that the network coulddistinguish between different types of axis orbits effectively;(4) Test validation: Through the rotor-bearing test rig, several axis orbits of actualfailure are measured, then, purification, feature extraction and recognition are carried out forthem. The results prove that the methods in this paper are effective.