Research On Fault Diagnosis Method For Reciprocating Compressor Bearings Based On Improved LMD

Reciprocating compressor is the core equipment in petroleum and chemical industry.If it fails,it will not only cause serious economic loss to the enterprise,but also bring serious harm to employees' personal safety.Therefore,fault diagnosis of reciprocating compressor has become the focus of people's research.However,the vibration signal of reciprocating compressor has nonlinear,non-stationary and multi-component coupling characteristics,and the traditional time-frequency analysis method represented by Fourier transform has some limitations on its analysis.Local mean decomposition(LMD)is a signal adaptive analysis method,which has unique advantages in processing non-stationary signals.Therefore,after improving the envelope construction method for LMD,this paper proposes a new fault feature extraction method based on the improved LMD and wavelet packet fuzzy entropy,then realizes the accurate diagnosis of the oversized bearing clearance fault for reciprocating compressor.Firstly,this paper reviewed the development of fault diagnosis technology,and summarized the common techniques,methods,means and main problems in fault diagnosis field for reciprocating compressor.It also expounded common non-stationary signal adaptive decomposition method,then analyzed the achievements and problems for LMD method in end point effect,mode aliasing,envelope structure,and its application in fault diagnosis field.Secondly,after a deeply research on the LMD method,an improved LMD method based on tangent points and monotonically cubic Hermite interpolation(MPCHI)was proposed.One key point of this method is to calculate the offset between the tangent point and the local extreme point by Taylor series,and use the tangent point to replace the extremum point as the envelope interpolation point.This can solve the intersect problem between the constructed envelope and the original signal;Another key point is to construct the envelope by the monotone cubic Hermite interpolation instead of cubic spline interpolation(CSI),and it can avoid the "over-envelope" and "under envelop" problem of cubic spline interpolation,therefore improve the accuracy of envelope.The results of simulation and experimental results show that the method can effectively improve the signal decomposition accuracy.Moreover,according to nonlinear characteristic in the reciprocating compressor vibration signal,wavelet packet fuzzy entropy was employed as quantitative indicator.Then a feature extraction method based on improved LMD and wavelet packet fuzzy entropy was proposed.In this method,the signals were decomposed into a series PF components by improved LMD,and the PF components were filtered through the correlation coefficient,then the selected PF components were quantified by the wavelet packet fuzzy entropy to form the feature matrix,finally the matrix was optimized by singular value decomposition to form the feature vectors with good separability.Finally,the bearing of the transmission mechanism in 2D12 reciprocating compressor is used as the object,and for t oversized bearing clearance faults in different positions were tested,and the fault vibration signal is collected.The fault feature vectors were extracted by the proposed method,and then the faults were recognized accurately by the support vector machine(SVM).The superiority of this method were verified by the comparison with the other feature extraction methods.