Typical Faults Prognosis For Reciprocating Compressors Based On Variational Mode Decomposition And Singular Spectrum Analysis

As the key equipment in the petroleum,petrochemical and some other industries,reciprocating compressors lose efficacy can bring significant economic losses and catastrophes in human casualties in the event of an accident.Therefore,it is quite essential to conduct fault prognosis for reciprocating compressor.Herein,we takes the sliding bearing abrasion and typical fault of the gas valves for the reciprocating compressors as the researching object.We systematically analyzed the several key issues from the perspective of adaptive vibration signal decomposition,such as,the response relationship between the failure developing mechanism and vibration signals,the scale characteristics of signal adaptive decomposition,multifractal spectrum characteristics analysis,the selection of performance degradation evaluation assessment indicators for the key component and the applicability of chaotic dynamics prediction.Combined with variational mode decomposition(VMD)and multifractal spectrum analysis,we conducted feature extraction and pattern recognition with the vibration signal based on the nonlinear signal refinement analysis.We evaluated and predicted the operating state of the sliding bearing for the reciprocating compressor by establishing the parameter index of singular spectrum.Aiming at the typical mechanical failure and operation cycle of a certain 2D12 reciprocating compressors,we proposed an advanced method for assessment and prediction of failure state.The detailed works are as follows:Owing to the concealment and complexity of vibration transmission path of the sliding bearing failure for the reciprocating compressor,it is intractable to predict the failure efficiently.To improve diagnosis accuracy of the bearing clearance fault between the indiscernible cross head shoe and connecting rod head,we analyzed the response relationship of vibration and fault by considering the mechanical failure separation capability in the course of the band-pass filtering based on the VMD algorithm,and then explored the sensitive point to enhance the signal identification and sampling consistency.Combining with transient frequencies and cross-correlation information,the VMD decomposition number was optimized by introducing multifractal generalized spectrum theory from the viewpoint of interstate characteristics severability.The generalized spectral eigenvectors of modal components for each state were extracted with variable order integer optimization.Considering the characteristics difference of different decomposition levels,the supporting vector machine method was introduced and the stratification based incremental learning L nearest neighbor model method(IKNNModel)was established during the failure model recognition.By analyzing the fault simulation and tested data,it proved that the optimizedunsupervised classification IKNNModel algorithm has a good adaptability.The extracted characteristics vector show good separability by combing the VMD and multifractal generalized spectrum analysis,which can efficiently identify the bearing fault features at sensitive measuring points.Gas valve failure is a typical multiple functional failures for reciprocating compressors.Owing to their causality and the tiny differences of the multiple failures,the fault types are difficult to identify.Considering the high nonlinearity and fluctuating behavior of vibration response,the identification method for valve failure was proposed based on VMD and multifractal detrending fluctuation analysis(MFDFA)from the perspective of the common failure mechanism of the valve plate and the response relationship of the vibration signal fluctuation characteristics.VMD_MFDFA algorithm utilized the maximum related minimum redundancy(mRMR)method to unify the VMD decomposition modes of different failures.Combined with the singular spectrum analysis,the 6-dimensional feature vector was established.The singular spectral eigenvalues of the main modes for each state was extracted based on fractal analysis.The spectral vector differences between modes was improved with principal component analysis,and the inter-class separability and robustness of fault features was enhanced with decreasing dimensionality.As the binary tree support vector machine and deep learning based convolution neural network algorithm were simultaneously introduced in pattern recognition,it proved to be the adaptability of convolution neural network in spectrum vector identification and the good identification accuracy of VMD_MFDFA for different gas valve failure with experiments.The high-risk and severe hazard of sliding bearing failure for the compressor make it essential to identify the performance degradation and evaluation prediction.In addition,the fault diagnosis refers to the whole lifecycle of equipment due to the essential characteristic of the coexistence of state and process manifested during the equipment failure.Combining singular value decomposition(SVD)and kernel fuzzy C-means clustering(KFCM)with the VMD and multifractal analysis methods,the evaluation index and state classification algorithm model were built by introducing fractal singular spectrum parameter evaluation.The main modality was maintained and the continuously truncated reconstruction matrix was contructed with VMD.The singular spectral parameter indicators stability between failures was improved based on SVD signal-noise separation and the central difference quotient method.Finally,the KFCM algorithm is exploited to train and establish clustering centers for each state's spectral parameters.After compressor bearing failure simulation,the spectral parameters were optimized and the fuzzy binary tree SVM algorithm was utilized to realize the wear classification and performance degradation evaluation for sliding bearing.Life prediction is an extension of fault assessment,which enriches the connotation of fault prediction.The equipment lifecycle analysis for reciprocating compressors consists of fault diagnosis,prediction and evaluation process.As for the prediction model adaptability and initial sensitivity of nonlinear system,an improved k-neighbor dynamic prediction model was presented based on maximum prediction by taking multifractal singularity spectrum asprediction parameter.The reliability of the prediction results was improved by introducing information entropy saturation into the maximum predictive credible scale.The establishment of phase space reconstruction dynamic modeling domain based on the spectral parameters of different modal components endowed real-time characteristics for complex dynamic evolution system,highlighting the independent influence of each modal component on the prediction.The validity of the prediction model was verified with fitting regression and error analysis.