Application And Research Of Mutual Information Learning To Fault Diagnosis Of Rotating Machinery

With the progress of scientific and technological and the development of socialproduction, modern industrial equipment put forward higher requirements of theproduction efficiency. Hydraulic pump is the "heart" of the hydraulic system, and rollingbearings is widely used in machinery and equipment, their working status plays a veryimportant and even decisive role on the working status of the entire apparatus. Therefore,itis very necessary to monitor and diagnose the hydraulic pum and rolling bearings.In axialpiston pumb and rolling bearings,the fault signal is usually performed as nonlinear andnon-stationary,showing complex motion characteristics,a method based on the combiningof the wavelet packet decomposition and mutual information is used to process thevibration signal and to achieve recognition and diagnosis of the faults in the paper.Wavelet packet analysis is the extension of wavelet analysis, it can achieve moreelaborate signal decomposed and reconstruction, the low-frequency portion of the timedecompose is re-decomposition by wavelet analysis, but the high-frequency part of thedecomposition will not be decomposed, resulting in the poor resolution ratio of the highfrequency part,however, the wavelet packet analysis can achieve the re-decompose both ofthe low-frequency portion and the high-frequency part, so the resolution ratio ofhigh-frequency part is improved,and the information of the signal can be more clearly.A faults diagnosis method based on the combining of the wavelet packetdecomposition and mutual information is presented in this paper, through the signalwavelet packet decomposition,combined with kurtosis criteria to select the sub-frequencyband which includes more information about the faults to reconstruct, the scale featureparameters and the sub-band energy of the reconstructed signal are extracted as the featurevectors.The feature selection algorithm based on mutual information is used to achieve thereduction of the dimensionality of the characteristic vector,and the purposes of the theremoval of redundant and irrelevant characteristic vectors.Finally, the fuzzy C-means clustering method is used on the pattern recognition of avariety of fault conditions of the hydraulic pump and rolling bearings, comparing the results with the conventional signal processing and Principal Component Analysis showedthat the method based on the combining of the the wavelet packet decomposition andmutual information is efficient and accurate in the fault identification of hydraulic pumpand rolling bearings.