Sparse Representation-oriented Research On Diagnosis Of Rolling Bearing-shaft Compound Faults

Extensively used in modern rotating machineries,the rolling bearings have evolved into a crucial element for promising system's safe and steady operation,but the ones that break down most easily and commonly,accounting for the occurrence of major mechanical faults.Thus,taking targeted condition monitoring and fault diagnosis for rolling bearing operation performance will not only effectively prevent serious accidents,but also benefit the decision-making of reasonable maintenance strategies,and decrease of insufficient or excessive maintenance.This thesis treats the rolling bearing-shaft system with compound faults as the research object,according to the different diagnosis needs under different occasions,proposes the sparse representation-oriented fault diagnosis approaches,and realizes the qualitative and quantitative diagnosis for rolling bearing-shaft systems so that a new technology is provided for condition monitoring and fault diagnosis.Based on the typical bearing outer race single fault,the ball's motion trial when travelling across the fault region,and the features of excited signals corresponding to entry and exit events,are both analyzed.Through identifying the excited ‘step-impact' signal features,a theoretical formula is deduced for the size estimation of bearing outer race fault.Taking the compound bearing-shaft faults into consideration,a representative model for compound fault excited signals is then formed.Further according to the specific qualitative and quantitative diagnosis needs,the signal model is refined in the aspects of signal component detailed features.Tackled with the sparse representation for compound fault excited ‘impact' features considered in the qualitative diagnosis approach,through constructing three redundant dictionaries with different resonance properties and developing a set of fast iterative algorithm assisted by linear time-invariant(LTI)filter and scaled augmented Lagrangian algorithm,this thesis realize the perfect separation for rolling bearing fault-related resonance components and shaft fault-related rotation component.In this manner,the qualitative diagnosis result is provided,including the evaluation for the bearing fault number,severity and the shaft state.The numerical simulation experiment validates its effectiveness.Also,the correlation coefficient is taken to quantitatively evaluate the main parameters' influence laws on the final representation results.To sparsely representing the signals with ‘step-impact' characteristics considered in the quantitative approach,the pre-processing method based on the LTI filter and AR model is firstly put forward to enhance the signal-to-noise ratio and eliminate the inherent rotation component that may result in the ‘step peak dimming' phenomenon.Then,the ADOMP-cored sparse representation and signal separation approach is proposed.Combined with the shift factors of obtained Asymmetric Gaussian Chirplet Model(AGCM)atoms that serve to locate the entry and exit event,bearing compound fault sizes can be accurately and automatically estimated,also known as the quantitative diagnosis.Experimental and comparative analysis results validate the effectiveness and superiority of proposed bearing qualitative and quantitative diagnosis approaches.Meanwhile,the experimental data collected under different rotation speeds and compound fault size combinations are further analyzed to verify the universality of quantitative diagnosis approach.Facing the deviations,reasonable explanations are given with respect to the bearing fault vibration signal excitation mechanism,which provides the theoretical reference for revising the fault size estimation formula.