Research And Application Of Fault Diagnosis Method For Rotating Machinery Based On Time-frequency Analysis

Rotating machinery is very important in production field. Because of the various excitation sources and the complexity of the rotating machinery, the vibration signals are usually nonstationary and nonlinear multi-component signals, of which the different nonstationary characteristics correspond to different mechanical failure. In order to achieve better fault diagnosis of rotating machinery, it is necessary to do research on the method of time-frequency analysis. The research of this paper is carried out according to this need. Based on time-frequency analysis theories like time-frequency distribution, wavelet scales spectrum and Hilbert time-frequency spectrum, combined with blind source identification, synchronous average and multi-scale entropy methods, research and application of fault diagnosis method for rotating machinery is carried out.In order to extract the independent sources of rotating machinery, a new blind source identification method based on time-frequency analysis is proposed to perform more accurate equipment fault diagnosis. This method is firstly deducted in theory. The effect of blind source identifications based on different time-frequency distributions is analyzed using simulation signals, with which the results of independent component analysis are compared. At last, three different failures of unbalance, misalignment and stents loose are simulated on the rotor workbench. This experiment showed that this method worked very well for the diagnosis of these three different failures.For the actual rotating machinery vibration signals are often mixed with noise and the characteristics of cyclostationary, a novel method, average across multi-scales based on wavelet reassigned scalogram, which combines the time synchronous average’s merit of reducing the interference of noise with wavelet transform’s merit of performing multi-scale analysis. The signal is firstly processed by wavelet transform and rearranged. Then the signals on different scales are processed by synchronous average in order to get the averaged wavelet reassigned scalogram. The validity of this method is verified through simulation analysis and the diagnostic experiment of rolling bearing.Hilbert time-frequency spectrum of rotating machinery’s vibration signal consists a lot of characteristic information, which is related to equipment’s working condition and also usually very hard to recognize. Since multi-scale entropy can effectively be utilized to describe the complexity, a method for working classification is proposed, based on Hilbert time-frequency and multi-scale entropy. The signal is processed by Hilbert-Huang Transform in order to get Hilbert time-frequency spectrum, which is divided and reduced to1-D sequence to get the multi-scale entropy. Through analyzing of the multi-scale entropy curves of the Hilbert time-frequency of different working conditions, The scales, where the sample entropies of different working condition can be separated effectively, are selected to construct the feature vector to recognize working condition. This method is used to extract the feature vectors of different bearing failures working conditions, and bearing’s working conditions are effectively classified.Rotating machinery vibration measurement and analysis system is developed based on virtual instrument. This system can realize eight channels’ vibration signal acquisition, and the data can be transmitted to computer by wireless data transmission module, then displayed, analyzed and stored real time. It has time domain and frequency domain analysis functions, also has time-frequency analysis functions like Hilbert time-frequency spectrum. This system is verified practicality and effectiveness through the practical application.