Research On Short-time Singular Value Decomposition Algorithm For Rolling Bearing Fault Diagnosis

Rotating machinery are extensively used in the manufacturing of industrial products.The rolling bearing as the supporting part of the rotating machinery plays an important role in the safe and efficient operation of the equipment.However,rolling bearing failure accounts for a high percentage of the mechanical equipment breakdowns,and some cases will result in fatal accidents.Thus,the rolling bearing fault diagnosis,especially the initial faults,is important to ensure the safe operation of the rotating machinery and avoid catastrophic accidents.To diagnose the initial fault of rolling bearings,many feasible feature extraction techniques have been tried.Those noise reduction methods based on singular value decomposition(SVD)are recognised to be promising for its effectiveness and utility.Therefore,a local feature analysis method,which is named as short-time singular value decomposition(ST-SVD),is proposed based on the in-depth understanding of failure mechanism of rolling bearings,the characteristics of fault vibration signals.The core thought of this method is to intercept the integral signal using the sliding windows and analyse each obtained signal segments with SVD method.The main contents of this paper are as follows:1)The failure mechanism of rolling bearings,the characteristics of fault vibration signals and the corresponding spectral distributions are discussed.The SNR differences between the local fault signals are introduced in detail.On these basis,a short-time singular value decomposition(ST-SVD)algorithm is proposed to research the features of local signals.This SVD based algorithm can effectively solve the problems of signal global analysis,such as trajectory matrix construction and singular value order selection.The joint distribution information of time-domain and eigenvalue-domain of the fault signal can be revealed by constructing the short-time eigenvalue spectrum(STES),which are suitable for the visualization of the fault impulse characteristics.2)In order to excavate the implicit information of STES,the main singular value feature series(MSVFs)are extracted from the STES by time-dimension projection of impulse characteristics.In order to extract the fault information embedded in the MSVFs,the short time sequence which can represent the fault impulse component is reconstructed.Then,the natural frequency of this bearing system can be extracted by spectrum analysis.Based on the principle of FIR filter,an optimal filter design scheme with natural frequency as the center frequency is proposed for rolling bearing fault diagnosis.The selection of filter bandwidth ?6,sliding window length L?and other parameters are all analyzed in detail.In order to overcome the shortage of bandpass filter,adaptive filter design method based on noise power estimation and Least Mean Square(LMS)algorithm is also proposed in this article.3)According to the fault information extraction from the MSVFs,a new signal sparse representation method based on adaptive dictionary construction are proposed.The core of this method is to construct the impulse atom dictionary by extracting the reconstructed short time sequence which can express fault impulse component.Then,K-SVD algorithm is used to realize the training of learning dictionary.After that,the adaptive dictionary can be obtained by dimension expansion and time delay operation of atoms of learning dictionary.Additionally,the reconstructed short-time sequence genearted by the minimum main singular value can extract noise information in the original fault signal.On this basis,the iterative termination condition of Matching Pursuit algorithm based on noise power estimation can be established.The simulation results show that the proposed signal sparse representation is effective for rolling bearing fault feature extraction.In summary,the proposed short-time singular value decomposition based rolling bearing fault feature extraction and fault diagnosis methods have a good effect on the weak fault feature extraction,accurate fault diagnosis and intelligent assessment.It also provides a reliable thinking to the problem of slip fluctuation and nonstationarity in the rolling element bearings.