Deconvolution-based Rolling Bearing Fauit Feature Extraction And Intelligent Diagnosis Algorithm Research

Rolling bearings are widely used in large industrial equipment by virtue of their high performance and compact design.But rolling bearings are very susceptible to damage,failure to detect them in time can lead to poor working conditions or even the cessation of work,seriously affecting production life.It is therefore of great practical importance to be able to achieve an accurate diagnosis of the type and extent of rolling bearing faults.This paper takes the deconvolution method as the starting point for the extraction of fault characteristics,and carries out a systematic study of the fault diagnosis method of rolling bearings through theoretical and experimental analysis for different problems.In order to solve the problem that the correlation kurtosis index in the(MCKD)algorithm is easily influenced by noise and interference components,resulting in the inaccurate selection of the key parameters fault period T,filter length L and shift number M in the algorithm,a new optimization index is proposed--A new optimization index,the signal cycle kurtosis to noise ratio(SCKNR),is proposed as the optimization objective function of the MCKD algorithm,and the key modal parameters of the MCKD algorithm are automatically found in conjunction with the cuckoo(CSA)search algorithm to save operational costs.Simulated signals and actual engineering signals validate the effectiveness and superiority of the proposed method.To address the problems of bearing fault vibration signals being disturbed by transmission paths,background noise and occasional shocks,a composite model for bearing fault diagnosis based on multi-point optimal minimum entropy deconvolution(MOMEDA)and spectral kurtosis is proposed.The former suppresses the noise components by filtering,while enhancing the weak fault characteristics;the latter can find the optimal filter according to the signal characteristics,and carries out a secondary filtering process on the signal after noise reduction,while obtaining a more intuitive fault characteristic band,finally achieving the extraction of the fault characteristics of the bearing fault vibration signal.Public data and actual engineering data prove the superiority of the proposed method.For the classification of rolling bearing faults in strong noise environments,the traditional signal processing method MOMEDA is combined with Deep Residual Systolic Network(DRSN)for the classification of bearing faults.By inputting the original signal into MOMEDA for reconstruction as the input signal,the generalization performance of the input signal is enhanced;then the gradient disappearance and explosion problem is solved by adding the crosslayer constant connection in the residual structure on top of the multilayer convolutional neural network.Finally,the time-frequency map of the fault samples is fed into the DRSN as a diagnostic model to complete the fault classification.Using engineering data for validation and comparison with a variety of common network models,it is demonstrated that the proposed fault diagnosis method is easier to identify fault features and achieves a classification accuracy of 99.82%.