Research On Fault Feature Extraction Of Reciprocating Compressor Based On Time-varying Singular Spectrum

In the field of petrochemical,reciprocating compressors are mainly responsible for refining oil and gas,transporting flammable and explosive gases like natural gas and ethylene,and they always work in poor conditions.In addition,because the structure of reciprocating compressor is complex and there are vulnerable parts,the fault occurrence frequency is high,and fault types are complex and diverse.In order to avoid accidents,ensure the safety of staff and reduce the economic loss of enterprises,the fault detection and diagnosis technology of reciprocating compressor has become the focus of people’s research.In this paper,aim at the non-stationary and non-linear characteristic of reciprocating compressor vibration signal,this paper introduces time information into the multifractal theory,and puts forward a fault feature extraction method of reciprocating compressor based on time-varying singular spectrum to describe the whole and detailed information of the vibration signal.Finally,to identify the extracted feature vectors using SVM.The results show that this method can express fault information more accurately and accurately,and is helpful to improve the accuracy of fault diagnosis.Firstly,this paper summarizes,studies and compares the fault diagnosis technologies of reciprocating compressor by collecting and reading the relevant data,and puts forward the research conception of fault feature extraction method.And it also summarizes,studies and compares present situation of intelligent pattern recognition methods,which provides the basis for verifying the effect of feature extraction method.Then,It analyses the main structure,working principle and working process of the reciprocating compressor,studies the fault form and failure mechanism deeply,and establishes the dynamic model of the key components.The phase space reconstruction theory is used to calculate the correlation dimension of the reciprocating compressor which can describe the chaos of its signal.And then,It introduces the theories and algorithms of fractal and multifractal.It also explains some important spectral parameters’ meanings in the fault signal analysis of reciprocating compressor.In addition,the time information is introduced into the multifractal theory to establish the time-varying singular spectrum theory model,and it puts forward the calculation method of the time-varying singular spectrum according to the working process of the reciprocating compressor.According to the fault characteristics of the reciprocating compressor,the SVM model is established by using the one-to-many classification method after studied support vector machine classification theory,and it chooses the best parameters for the SVM.Finally,the fault feature extraction method based on time-varying singular spectrum is applied to the fault diagnosis of 2D12 reciprocating compressor.It sets the diagnostic process of reciprocating compressor,including vibration acceleration signal acquisition,denoising signal using wavelet decomposition combined with LMD decomposition method,extracting fault feature vector by time-varying singular spectrum and classifying fault types by SVM classifier.The classification results show that the valve fault’s classification accuracy is 100% and the bearing fault’s classification accuracy is 93%.It verifies the validity of the fault feature based on the time-varying singular spectrum method,and this method can be used to make accurate fault diagnosis on reciprocating compressors.