Research On Fault Feature Extraction And Diagnosis Method Of Rolling Element Bearing

With the rapid development of science and the increasing popularity of modernization production, rotating machinery is constantly developing in the direction of large-scale, complex, high speed and automatic. Hence higher requirement to the operational safety of the equipments is put forward. Rolling element bearing, as one of the parts widely used in rotating machinery, directly decides the running state of the whole mechanical system. Consequently, research on techniques of fault diagnosis and condition monitoring for bearing can effectively avoid the occurrence of major accident. Therefore it has important academic significance and application value.This thesis discusses the research status of fault signal de-noising, feature extraction, compound and intelligent fault diagnosis for rolling element bearing, and takes the vibration signal analysis as the main research means. This article puts forward the solution methods for fault feature extraction, weak fault feature enhanced detection, compound fault feature separation, fault pattern recognition and running state detection aiming at several key technologies which need to be solved in the process of fault feature extraction and diagnosis of bearing. The innovation points and main work are as follows:(1)The frequency components of bearing faulty signal collected by sensor are very complicated. Fault character frequency components are difficult to be extracted accurately because of the interference of the unrelated frequencies and noise. Whereas a method based on cepstrum pre-whitening technology and singular value decomposition theory is presented to extract the fault character frequencies of bearing. The method can remove the discrete frequencies and harmonic components by means of cepstrum editing procedure, and then eliminate noise in signal through singular value decomposition and reconstruction signal based on the maximum difference spectrum. At last the test confirms the fault characteristic frequencies of can be acquired clearly.(2)Fault vibration signal of bearing usually contains strong background noise under the poor working environment. Weak fault character components of bearing, easily submerged by noise, are difficult to be detected accurately. Hence a method named adaptive multi-scale self-complementary Top-Hat transformation is presented to diagnosis weak fault of bearing. Morphological self-complementary Top-Hat transformation can not only enhance the impulsiveness in the vibration signal of faulty bearing, but also depress strong background noise. In addition, multi-scale morphological filter is constructed in order to depress noise and retain signal details at the same time. The proposed method adaptively determines the optimal scale of structure element by comparing to fault feature amplitude energy radio.(3)It is different to separate compound fault features of rolling element bearing in single channel signal, because they usually mix each other when compound fault of bearing happening. A compound fault feature separation method based on improved harmonic wavelet packet decomposition is presented in this thesis. The improved harmonic wavelet packet decomposition can arbitrarily divide frequency band of vibration signal for bearing, and overcome the disadvantage that the sub band number and bandwidth range acquired by using conventional harmonic wavelet packet decomposition are subject to the limitation of the binary decomposition. We calculate weight factor for every single point fault in each sub band signal, and make the single channel signal of compound fault reconstruct into different channels. The proposed method can effectively separate the compound fault feature of bearing.(4)Fault mode recognition and running state detection are main contents of the intelligent diagnosis of rolling element bearing. This thesis proposes a novel character parameter named time-wavelet energy spectrum sample entropy through combining Hermitian continuous wavelet transform and sample entropy theory. The parameter adopted as fault character vectors of different fault mode is put into support vector machine (SVM). The output of SVM can achieve an ideal result of fault mode identification for bearing. Next the time-wavelet energy spectrum sample entropy as fault character parameter is used to detect the running state of rolling element bearing. We calculate time-wavelet energy spectrum sample entropy of data from the whole life cycle test rig for bearing. The trend of running state for bearing is acquired by arranging time-wavelet energy entropy chronologically. The proposed method can detect the early damage occurring in bearing effectively by judging the running state trend of bearing.