Research And Application Of Feature Migration In Screw Compressor Fault Diagnosis

The establishment of fault diagnosis model for mechanical equipment needs a lot of condition monitoring data.Fault diagnosis of equipment lacking historical monitoring data usually relies on existing fault diagnosis models,while the distribution of monitoring sample data of different equipment is often different.At the same time,irrelevant and redundant fault features also bring difficulties to the establishment of diagnosis models.Aiming at the above problems,this paper establishes a migration diagnosis model through feature selection and migration learning method to improve the accuracy of screw compressor fault diagnosis which lacks historical monitoring data.The main research contents are as follows:(1)Selection method of Maximal Information Coefficient(MIC)of screw compressor fault characteristics.By analyzing the fault features,the optimal features are found according to the difference between the intra-class distance and the inter-class distance.The maximum information coefficient is used to measure the correlation between features,and the feature subset is screened.The analysis results show that the aggregation discrimination effect is better after feature selection.(2)Unsupervised K-means clustering and semi-supervised self-training sample selection methods.The unsupervised K-means clustering method is applied to the target domain samples.At the same time,the existing fault monitoring model is used to train the clustering results.The candidate samples are obtained,the sample confidence is evaluated,and the high confidence samples are obtained after screening.Case analysis shows that this method can obtain high confidence samples and solve the problem of using unlabeled samples in target domain.(3)Research on fault feature migration method of screw compressor.Transfer Component Analysis(TCA)is used to reduce the difference of data distribution among monitoring samples of different devices,and centralize the sample set in the new feature space to realize the migration of fault features.The results show that the method reduces the data distribution,reduces the distance of sample clusters of the same kind,and improves the diagnostic accuracy.(4)Fault diagnosis and verification of screw compressor.Aiming at the screw compressor in Tarim Oilfield,the feature migration method is used to diagnose the screw rotor profile roughening fault,which provides a basis for predictive maintenance of field equipment,and shows the effectiveness and adaptability of this method.