Intermittent Process Fault Diagnosis Based On K-means Optimal Clustering And Random Projection

As the complexity of modern industry continues to increase,the links in production are becoming more and more close.Tight production linkages can significantly reduce construction time,reduce labor costs,and modernize production,but also mean high standards of operational detail.If a small part of the system fails,the small will affect the accuracy of the product,and the larger will cause the whole system to collapse.Both of these aspects will have a huge impact on industrial production.In order to ensure the smooth progress of the industrial production process,research on fault diagnosis methods has become inevitable.The sophisticated information extraction and storage technology in modern industry provides a guarantee for collecting high-dimensional,multi-class massive data.Therefore,the focus of fault diagnosis in modern intermittent industrial production processes has shifted from the diagnosis of two-dimensional data to the diagnosis of multidimensional data-driven.The intermittent fault diagnosis method based on multi-dimensional data can diagnose the fault location,eliminate the interference,and ensure the normal operation of the industrial production process by online analysis and working condition judgment of the data.There are many methods for intermittent fault diagnosis.This paper mainly studies data-driven multivariate statistical analysis and machine learning fault diagnosis methods,and moderately integrates the two methods into the intermittent industrial production process of machine tools.On the one hand,based on the traditional principal component analysis PCA(Principal Component Analysis),the multivariate Principal Component Analysis(MPCA)based on intermittent data and the segmentation softening partitioning method are studied.On the other hand,based on machine learning,the K-proximity algorithm is combined with the random projection JL(Johnson-Lindenstrauss)conversion to complete the fault diagnosis of the transition period in the intermittent process.The main contents of the thesis are as follows:First,Aiming at the problem of time division of intermittent industrial process,this paper adds the optimal clustering algorithm of K-means based on the soft segmentation of traditional time.The method increases the scientificity of initializing cluster center point selection,improves the reliability of sub-period partitioning,and makes the transition period division more accurate.Secondly,based on the problem of intermittent industrial process fault diagnosis,based on the screening of k-means optimal cluster transition interval,this paper proposes the fault diagnosis of 3D data in batch production process using MPCA.The method overcomes the inaccurate time division caused by randomly selecting the initial cluster center point,reduces the false positive rate and false negative rate of the fault,and improves the accuracy of fault diagnosis.third,For the problem of intermittent industrial process transition data processing,this paper cites the kNN transition period fault detection based on random projection.This method transforms the transition modal data into a de-randomized JL(Johnson-Lindenstrauss)to quickly generate a projection matrix.The transition modal data is projected from the high-dimensional space to the low-dimensional space,and the distance information between any two points is saved,and then the kNN method is used for fault diagnosis.fourth,Summary and prospects,summarize the fault diagnosis problems of intermittent industrial processes,and make the next development plan according to the development trend and research status.