Research On Fault Diagnosis For Sliding Bearing Clearance Of Reciprocating Compressor With Compound Faults

Reciprocating compressor is a key equipment used to compress and transport gas in the fields of petroleum,chemicals,etc.Once an accident occurs,it will cause huge economic losses and casualties.Therefore,it is of great significance to study the fault diagnosis technology of reciprocating compressor to ensure its normal operation.From the perspective of the data acquisition,feature extraction and assessment of state recognition during the process of fault diagnosis,this article takes the sliding bearing clearance of reciprocating compressor transmission mechanism as the research object,lucubrating the multi-body dynamics simulation,signal processing method,blind source signal processing technology and intelligent pattern recognition.Combining the vibration signal characteristic of reciprocating compressor sliding bearing clearance composite fault,this article presents a set of effective diagnosis methods of the reciprocating compressor sliding bearing clearance composite faults.The main content is as follows.The sliding bearing clearance fault of reciprocating compressor is difficult to carry out multi-type faults compound experiment due to its concealment,long forming period and high test cost.As a result,fault data acquisition is insufficient.To solve this problem,the research of multi-component dynamic simulation method is carried out on bearing composite fault state of reciprocating compressor with clearance kinematic pair.Selecting the clearance fault of big end bearing shell and small end bearing shell of the connecting rod,analyze the factors that affect the clearance fault of sliding bearing firstly.Establish the kinematic analysis model of bearing clearance fault,using antigravity connection.Then apply nonlinear contact collision model theory to establish the dynamic model of bearing clearance fault.Eventually establish the dynamic simulation model of the bearing clearance composite fault using multi-component dynamic analysis software.The data acquisition of bearing clearance composite fault of reciprocating compressor is realized.Vibration signals of the composite fault of reciprocating compressor extracted by sensor are mutually coupling formed by vibration source signals of several faults and noise.With the influence of actual working environment and transmission path of vibration,the number of vibration sources of reciprocating compressor bearing clearance composite faults is unpredictable,which has a strong impact on the validity of separation result of bearingclearance composite fault signals.To solve this problem,estimation method of fault vibration source amount based on time-varying zero-phase filtration and singular value decomposition is put forward.Empirical Mode Decomposition(EMD)is suitable to analyze non-stationary multivariate coupled signals,but end effect,data envelopment fitting,modal aliasing still exist,for which the method that designing time-varying filter using the time scale of high frequency signals and low frequency signals to replace time-varying zero-phase filtrating pattern decomposition is proposed.The validity of this method is verified by simulation signals and measured signals.At the meantime,considering the number of fault vibration sources is effected by the number of sensors,make use of time-varying zero-phase filtrating pattern decomposition to extend single channel signal to multichannel signal,forming intrinsic mode function recombination signal,for which singular value decomposition is performed.Ascertain the number of fault sources according to the eigenvalue distribution.Through comparative analysis of simulation signal and measured signal,this method can estimate accurately the amount of fault vibration sources of reciprocating compressor bearing clearance composite fault signal.Since the vibration signal of reciprocating compressor bearing clearance composite fault is mostly coupling formed by several fault sources,and the difference in the morphology of bearing clearance fault trait is small.The aliasing of different fault characteristic frequencies may occur and the weaker fault characteristics may be concealed by the stronger fault characteristics as conventional signal decomposition method can not effectively separate the bearing clearance composite fault characteristics with similar morphology.To solve the above problem,an extraction method of reciprocating compressor bearing clearance composite fault characteristics based on deep dictionary learning morphological component analysis(MCA)is proposed.Through the research on theoretical knowledge of MCA method firstly,aiming at the limitation that dictionary can not best match the structural characteristics of complex signals,MCA method of deep dictionary learning is proposed,whose validity is verified by simulation signals.The method is applied to the feature extraction of reciprocating compressor bearing clearance composite faults.Through learning the deep dictionary of different kinds of bearing clearance fault vibration signals,acquire the dictionary best matching fault signal characteristics.Comparing with EMD,verify the accuracy of reciprocating compressor bearing composite fault vibration signal feature extraction with MCA method based on deep dictionary learning.The effective separation and extraction of reciprocating compressor bearing composite fault characteristic components are realized.Master the fault category at the same time is needed in order to grasp health status of reciprocating compressor sliding bearing accurately.According to the position of fault,judge the key parts of maintenance.For the feature that the non-linear relationship between reciprocating compressor bearing clearance degree and vibration response is exist,a quantitative expression method of reciprocating compressor bearing clearance fault feature based on refined composite multiscale dispersed entropy(RCMDE)is proposed,realizing the representation of different degrees of fault feature.Because of correlation between fault feature vectors of different fault category and fault feature vectors of different fault degree is so strong,they are hard to identify.Introducing a deep learning method with feature extraction ability,a improved method of deep belief network(DBN)based on maximum information coefficient(MIC)and genetic algorithm(GA)is proposed.Through verifying the validity and superiority of this method by UCI standard data set,and applying it to reciprocating compressor bearing composite fault diagnosis,correct identification and diagnosis of bearing clearance composite fault category and bearing attrition rate of reciprocating compressor is realized.