Research On Industrial Process Quality Monitoring Based On Correlation Statistical Analysis

With the complexity of industrial equipment,fault diagnosis technology has become an indispensable part of the current industrial process.In this context,the growing technology of sensors and digital equipment has made it possible for large amounts of data generated by industrial processes to be captured and utilized by factories.Data-driven industrial process fault diagnosis has become one of the main research directions.Correlation statistical analysis is a kind of method to analyze variables by using the correlation between variables.Canonical correlation analysis is an analytical method to find the correlation of canonical variables,and it is an indispensable multivariate statistical analysis method based on data-driven fault diagnosis.Canonical correlation analysis makes full use of the characteristics of correlation between two sets of variables,so that the extracted main element components are more thorough.This paper is based on the correlation statistical analysis algorithm,in the KPI-related fault detection and initial small fault detection two aspects,and get some results.In the fault detection of KPI-relation,this paper uses the canonical correlation analysis to extract the maximum correlation between variables,decomposes the extracted correlation coefficient,and obtains the corresponding projection matrix.Then KPI-related and KPI-unrelated fault detection are carried out.In this paper,two decomposition methods for correlation coefficients are proposed,one is using singular value decomposition,the other is generalized singular value decomposition.By using these two decomposition methods,different projection spaces are obtained and fault detection is carried out.Finally,the effectiveness of the two methods is proved by numerical simulation and TE process simulation.These two different methods are named for KPI-CCA1 and KPI-CCA2.At the same time,the two methods are compared and analyzed from the point of view of computational complexity and fault detectability,and the similarities and differences between the two methods are summarized through simulation experiments.The sliding window can accumulate the fault amplitude in the form of a small fault.Therefore,in the initial micro fault detection analysis,this paper proposes a method combining statistical analysis and sliding window for fault detection.Because the fault detection based on sliding window needs to remove the old data constantly and add new data to form sliding window,it undoubtedly adds a lot of calculation to the calculation,so a recursive method is proposed in this paper.It is used for fault detection of sliding window,and the advantages and disadvantages of recursive and non-recursive in different cases are obtained by analyzing the computational complexity.Finally,the validity of this method for micro fault detection is proved by simulation.