Study On Vibration Signal Filtering And Its Application For Bearing Fault Diagnosis

Fault detection and diagnosis of the rotating machinery is a vital problem in the industry,such as,aerospace,transportation and wind turbine and so on.Due to the long term operation,high speed and heavy load working condition,the rotating machinery is prone to suffer from initial defects.Among these rotating machines,the bearing is one of the most vulnerable components.Early detection and diagnosis of bearing faults while the machinery is still operating in a controllable region can avoid abnormal event progression and reduce economic loss.Thus,it is urgent to conduct condition monitoring and fault diagnosis on the bearings.In fact,the bearings generally operate under severe conditions,and normally join forces with other rotating parts,such as,gears,shafts,etc.in the real situation.These rotating components bring strong interference to the diagnosis of bearings,and the vibration signatures of bearing can be easily masked by other vibrations generated from gears and shafts.Therefore,signal filtering and information extraction are the prerequisite to improve the accuracy of fault diagnosis,which are also the research aim of this thesis.In this thesis,some researches have been done on the bearing initial fault diagnosis.Based on filter structure and filter parameter optimization,this thesis proposes three different methods.And the main work and contribution of this thesis are as follows,(1)As a new adaptive filter,local mean decomposition(LMD)has several problems,whose decomposition accuracy is affected by the default parameters of the sifting stopping criterion.Then,this thesis proposes an adaptive sifting stopping criterion.And three crucial parameters,boundary condition,envelope estimation and sifting stopping criterion,are all taken into consideration to comprehensively improve LMD,thereby increasing the accuracy of the decomposition of LMD.(2)Spectral kurtosis is an effective indicator to guide the frequency band selection at present.As the time-frequency representation of vibration signal is one of the key factors to affect the fault frequency band selection.And the time-frequency division is very important for the calculation of spectral kurtosis.Therefore,this thesis utilized the improved local mean decomposition as a way to obtain the better time-frequency distribution,then applying it to calculate the spectral kurtosis.Thereby the optimal filter parameters are obtained.(3)In the filter parameter optimization,the health reference is important for the informative frequency band selection.This thesis proposes a new indicator named accuracy rate,which can fuse multi-dimensional information(impact and cyclic impact)and the health reference.With the improvement,the frequency band that contains the fault transient could be located.Finally,this thesis validated the three different methods on two kinds of datasets,i.e.benchmark dataset and experimental dataset.And the advantages and disadvantages of the three methods are analyzed.At the same time,some suggestions are put forward for the further improvement and research direction.