Research On Compound Fault Diagnosis Method Of Crank Linkage Mechanism Of Reciprocating Compressor

Reciprocating compressor is the key equipment in petroleum refining,oil and gas storage and transportation,metallurgy,refrigeration engineering and other process industries.It has the outstanding advantages of free discharge pressure and high fluid compression efficiency,and has been widely used.However,the reciprocating compressor has a complex structure,numerous incentive sources,and is more likely to occur a variety of failures at the same time.Once an accident occurs,it will lead to shutdown,which may cause huge economic losses,environmental damage and social problems.Therefore,it is necessary to study the compound fault diagnosis method of reciprocating compressor in order to provide an effective method for its comprehensive diagnosis.In this paper,data acquisition,underdetermined blind source separation of compound faults,feature extraction and fault recognition are carried out for the compound faults of the crankshaft linkage mechanism of the reciprocating compressor.In addition,an effective diagnosis method for the compound faults of the crankshaft linkage mechanism of the reciprocating compressor is proposed according to the characteristics of the vibration signals of compound faults.The main work carried out is as follows:Due to the complex working conditions of reciprocating compressors,it is difficult to carry out multi-type field fault test and obtain the compound fault test data of the crankshaft connecting rod mechanism of reciprocating compressors.Therefore,based on the kinematics and dynamics analysis of the crankshaft connecting rod mechanism of reciprocating compressor,a rigid-flexible coupling multi-body dynamics model of the crankshaft connecting rod mechanism complex failure of reciprocating compressor,including the crack of connecting rod and the gap of multiple motion pairs,was established by using software and technology such as Solidworks modeling,Ansys finite element analysis and Adams mechanical system simulation.The dynamic simulation of the compound fault of the crankshaft connecting rod mechanism was realized,and the vibration response signals of the corresponding sensitive measuring points were obtained.The compound fault state database of the crankshaft connecting rod mechanism of the reciprocating compressor was constructed.In the process of on-line monitoring of reciprocating compressor,the number and position of sensor measuring points are usually fixed.When the number of fault vibration sources is larger than the number of sensor measuring points,it is difficult to determine the fault type and position of reciprocating compressor.In this paper,the sparse representation of time-frequency energy is studied,and a clustering algorithm based on K adjacent relative density optimization peak density clustering and fuzzy C-means clustering is proposed to estimate the underdetermined blind source separation mixed matrix of crankshaft linkage compound fault excitation signals.Finally,the compressed sensing method is used to recover the source signals,and the comparative analysis shows that the method can be effectively applied to the underdetermined blind source separation of the combined fault excitation signals of the crankshaft linkage mechanism of the reciprocating compressor.Aiming at the nonlinear characteristics of vibration signals induced by compound faults of the crankshaft linkage mechanism of reciprocating compressor,and the problem that the spread entropy of a single scale can not identify the compound fault characteristic mode well,a quantitative characterization method of compound fault characteristics of the crankshaft linkage mechanism of reciprocating compressor was proposed based on the refined time-shift multiscale fluctuation-based reverse dispersion entropy.The values of embedded dimension m,number of categories c,time delay d,scale factor and sequence length N,which need to be set artificially are discussed.The refined time-shift multiscale fluctuation-based reverse dispersion entropy is applied to the feature extraction of the compound fault of the crankshaft linkage mechanism of reciprocating compressor,and the quantitative characterization of the fault features of different vibration sources is realized.Finally,compared with other entropy methods,the results show that the proposed method can accurately show the uncertainty and complexity of fault signals.Aiming at the problem that fault feature vectors of different fault types have great correlation and are not easy to identify,a pattern recognition method based on PSO and Radam optimization of bidirectional long short-term memory neural network is proposed.Firstly,Radam is selected as the optimizer of the long short time memory network to update the weight of the network.Secondly,particle swarm optimization algorithm is used to optimize the main parameters of the bidirectional long-short time memory network(learning rate and number of hidden layer neural units),and it is applied to the pattern recognition of the compound fault feature vector of the crankshaft linkage mechanism of reciprocating compressor.The accurate diagnosis of compound fault of crankshaft linkage mechanism mechanism of reciprocating compressor is realized.Finally,the influence of feature vectors of different fault excitation signals obtained by different feature extraction methods on the recognition effect of OBi LSTM model is analyzed.The identification accuracy of the compound fault feature vector of the crankshaft linkage mechanism of reciprocating compressor obtained by RTSMFRDE feature extraction method is analyzed,and the superiority of the new method is verified.