Research On Fast Fault Feature Extraction Method Of Rolling Bearing And Its Application

Rolling bearing is one of the most essential components in the mechanical systems.It is widely used and usually accompanied by a complex working environment.Simultaneously,it is also a component with high frequency of failure.Once the fault occurs,it will develop rapidly and worsen easily.If we cannot detect the fault quickly and accurately,and take the right remedy in time,then it might affect the running of other components,and eventually paralyze the entire system.Therefore,real-time monitoring and online diagnosis of the bearings during operation have always been the hot spots issue in the field.When a rolling bearing fails,it will produce periodic shock vibrations.The proper signal processing technology will help us to extract the bearing fault characteristics from vibration signals quickly.This is also the key to real-time monitoring and online diagnosis of rolling bearings.However,in practice,due to the complex working environment,the signals collected by the sensors are often mixed with a large amount of background noise and interference.So that the impacts signal that can characterize the bearing failure is submerged and cannot be identified.This largely affects the effectiveness of real-time monitoring and online diagnosis of bearings.In response to the above problems,this thesis systematically studied the method that can extract rapidly the early fault features of rolling bearings under a complex environment consisting of strong noise and large interference,and proposes three solutions from the aspects of rapid signal noise reduction,energy operator demodulation,and feature frequency extraction.(1)For the case with severe background noise during rolling bearing operation,which make it difficult to demodulate fault signals and the its feature,a solution that combining improved fast non-local means(IFNLM)filtering and symmetric high-order difference analytic energy operator(SHO-AEO)is proposed in this thesis.First of all,by optimizing the similarity measurement standard and the kernel function,we proposed an IFNLM filtering algorithm,in which the weighted average calculation of the distance between any two similar blocks only needs to be carried out once.Preprocessing the original signal with this algorithm not only improves the accuracy of noise reduction but also reduces the computational complexity.Secondly,based on the analytic energy operator(AEO),and drawing on the ideas of high-order difference and symmetric difference,we propose the SHO-AEO,which can be used to demodulate the denoised signal and identify the characteristic frequency of bearing fault from its energy spectrum.Moreover,the effectiveness and advantage of the IFNLM-SHOAEO method are verified by simulations and test bench data in the thesis.(2)For the difficulty in extracting fault features of rolling bearings in the complex background with both working environment noise and vibration interference,a novel fast mode decomposition method named hard thresholding fast iterative filtering(HTFIF)is first introduced to rapidly decompose the original composite signal into a set of intrinsic mode functions(IMF).Then a L-KCA indictor considering the characteristics of signal itself and statistical characteristics is developed for sensitive IMF selection.At last by using three sampling points that spaced k apart in the signal to improve the symmetrical difference sequence,an improved k-value symmetrical difference analytic energy operator(k-SDAEO)is presented to demodulate the sensitive IMF and the fault characteristic frequency can be identified in the k-SDAEO spectrum.Also,the effectiveness and advantage of the HTFIF-kSDAEO method are verified by simulations and test bench data.(3)For the problem of early weak fault feature extraction of the bearings,a method based on SOSO enhanced filtering was proposed.The bearing vibration signal is decomposed into a series of IMFs by the HTFIF algorithm we introduced before,and weighted reconstruction is carried out according to the L-KCA value of each IMF to filter out some interference in advance.Then a SOSO?IFNLM enhancement filter structure is constructed to remove a large amount of noise while maintaining the smoothness of the original vibration signal,simultaneously enhancing the fault impact characteristics.Incorporating the symmetric high-order idea into the frequency weighted energy operator technology,a symmetric high-order frequency weighted energy operator(SHFWEO)is proposed to demodulate the noise-reduced signal while improving the signal-to-interference ratio(SIR)of the signal.The effectiveness and advantage of the HTFIF-SOSO?IFNLM-SHFWEO method for extracting bearing early weak faults under complex operating conditions are verified by simulations and accelerated life cycle experiments.Finally in this thesis,taking the SDM00 vibrating screen as an example,the application of the three proposed fault feature extraction methods in the fault diagnosis of vibrating machinery was carried out.Based on the vibrating screen,the above three methods and the three proposed energy operators are compared and evaluated.It is shown that in practical applications,the method of "purification" and demodulation steps can optimize the matching according to the specific situation and provide the best solution for the fault diagnosis of rolling bearings.