| Reciprocating compressor as a modern industrial transport,compression of gas core equipment,in the petroleum and petrochemical as well as manufacturing and other national economic fields are widely used.The key components of the reciprocating compressor rod bearing,long-term high pressure,high speed,heavy load in the harsh operating conditions,due to assembly errors and operating collision friction caused by excessive bearing clearance,resulting in abnormal compressor vibration,threatening equipment safety.Due to the complex structure of compressor and many excitation sources,the vibration signal of bearing gap fault presents strong time-varying and non-smooth characteristics,and the traditional single feature extraction method cannot extract the fault features in a comprehensive and detailed way to achieve accurate fault diagnosis.Therefore,this paper proposes a variable-length feature selection method and subset evaluation criterion by carrying out the research of feature selection method,fusing the multi-domain feature extraction method,constructing a compressor high Vetter feature set,realizing comprehensive and detailed feature extraction,combining the variable-length feature selection method and subset evaluation criterion to select a high-quality feature subset,and carrying out pattern recognition to realize the accurate diagnosis of reciprocating compressor bearing clearance fault.The multi-domain feature analysis method is studied to construct the equipment state high Vetter set and perform the fault feature separability analysis.Based on the extraction of fault features in the time domain,frequency domain,time-frequency domain and entropy value features,a reciprocating compressor state Govett set is constructed,and the separability of features is analyzed based on the dispersion of feature sample distribution.The analysis results show that some features in the time domain,frequency domain features and entropy features have good separability,which can provide guidance basis for the subsequent feature selection.To address the irrelevant and redundant information in the high-dimensional feature set,a feature optimization strategy and subset evaluation criterion based on bounded variable-length genetic algorithm are proposed.On the one hand,to overcome the problem that the genetic algorithm is easy to fall into local optimal solutions,the strategy of searching the feature space with variable length and quadratic optimization within the optimal feature interval is adopted,and the repair operator is introduced to control the size of the feature space;on the other hand,to solve the shortage of the distance measure criterion,the subset evaluation criterion is constructed by using the maximum information coefficient and feature ranking fusion method.On the other hand,in order to solve the shortage of the distance metric criterion,the maximum information coefficient and the feature ranking fusion method are used to construct the subset evaluation criterion.The application of fault diagnosis based on variable-length feature selection is studied for reciprocating compressor connecting rod bearing clearance faults.A bounded variable-length genetic algorithm is used to find the optimal feature number interval of the reciprocating compressor feature set,and a high-quality low-dimensional fault feature subset is selected within the optimal feature number interval and input to a support vector machine classifier.The experimental application shows that the selected subset of fault features has a smaller number of features and a higher fault recognition accuracy than other selection algorithms. |
