Research On Reciprocating Compressor Fault Diagnosis Approach Based On The Fusion Of VMD With Adaptive Parameter Adjustment And RCMWMFE

Reciprocating compressor is the core of the petroleum,chemical,and other fields of key equipment,bearing the heavy production tasks,its working medium usually has high temperature and high pressure,toxic and harmful,flammable and explosive,easy to corrode,and other characteristics.Once a failure occurs,it will lead to production downtime and economic losses or cause major accidents and casualties,therefore,there is an urgent need to carry out fault diagnosis research on reciprocating compressors.Due to the complex structure of the reciprocating compressor,the operation form is diverse,and its vibration signal shows strong multi-source impact,nonlinear,nonstationary complex characteristics,the traditional signal processing methods are difficult to carry out effective fault feature extraction.To address the above problems,this paper proposes an approach based on the fusion of variational mode decomposition(VMD)with adaptive parameter adjustment and refined composite multiscale weighted multidimensional feature entropy(RCMWMFE)for reciprocating compressor fault diagnosis.This method can extract the fault feature information more accurately and has higher fault diagnosis accuracy compared with other methods.(1)For the problem of strong nonstationary reciprocating compressor vibration signal with multiple sources,this paper proposes an adaptive parameter adjustment of the variational mode decomposition algorithm,which uses the empirical mode decomposition-mean permutation entropy(EMD-APE)algorithm for the optimization of the parameter K.After determining K,the mean permutation entropy(APE)algorithm is used for the optimization of α,and the Hilbert transform is combined with the signal.The time-frequency analysis is performed.The simulation signal and the analysis of the rolling bearing fault signal of Case Western Reserve University,USA show that the proposed method can extract the fault characteristics more accurately.(2)For the quantitative description of reciprocating compressor vibration signals with nonlinear complexity,an RCMMWMFE algorithm is proposed in this paper,which incorporates three entropy information such as refined composite multiscale scattering entropy(RCMDE),refined composite multiscale fuzzy entropy(RCMFE),and refined composite multiscale permutation entropy(RCMPE)to obtain a more comprehensive vibration signal characterization.The comparative analysis of the simulation signals shows that the proposed method has a better feature differentiation compared with that of using only a single entropy,which lays a good foundation for the nonlinear metric description of vibration signals.(3)In order to deal with both nonstationary and nonlinear feature extraction of reciprocating compressor vibration signal,this paper organically fuses the above two methods,firstly,the VMD with adaptive parameter adjustment is performed on the original vibration signal,and each IMF component is determined by the number of least common multiplication layers,and the main IMF components are filtered by the correlation coefficient method;then the quantitative calculation of each IMF component’s RCMWMFE to construct the feature vector;finally,a convolutional neural network(CNN)is used for classification and identification.The experimental results of comparing 200,500,and 2000 mixed sample sets of different fault signals of reciprocating compressors show that VMD-RCMWMFE has the highest accuracy of 96.5%,100%,and 100% on the three data sets,respectively,compared with other methods,which verifies the effectiveness of the proposed method and provides a new way for reciprocating compressor fault diagnosis.