Research On Nonlinear Process State Monitoring Based On Correlation Analysis

With the development of the era and the advancement of technology,people’s demand for process safety and product quality has gradually increased.Data-based process state monitoring has attracted more and more attention and research by scholars.Among them,multivariate statistical analysis algorithms with many advantages are prepared favored.This paper mainly makes some improvements based on canonical correlation analysis algorithm belong to statistical methods.The optimization goal of canonical correlation analysis is to find a pair of transformation vectors to maximize the correlation between the linear combination of process variables and the linear combination of key performance indicators.However,the canonical correlation analysis only considers the linear relationship between the variables and the projection vector,and when applied to process state monitoring,it cannot judge whether the found abnormality will affect key performance indicators such as product quality.Taking the industrial process as the application object,the main work of this paper is summarized as follows:(1)This paper firstly summarizes the previous research work on state monitoring,and introduces the algorithm principle,solution idea and how it is applied to condition monitoring in detail.(2)Secondly,in order to deal with the nonlinear characteristics common in industrial processes,this paper proposes a novel nonlinear process state monitoring method that combines neural network and canonical correlation analysis in principle,and then also gives the proposed network parameter update method,including strategy and detailed derivation expressions.Finally,we demonstrate the feasibility of the proposed method through a constructed numerical example,a classic industrial simulation process,and a fault case of a real industrial process.(3)Finally,considering the final product quality requirements in the production process,it is necessary to classify the abnormalities detected in the process,that is,whether it will affect the key performance indicators.This paper proposes a process variables space division method based on kernel canonical correlation analysis for state monitoring of key performance indicators.First,the issue of the described kernel canonical correlation analysis is improved,and then the generalized singular value decomposition method is applied to the obtained correlation matrix to achieve the purpose of dividing the process variables space,and the effectiveness of the method is verified by simulation test.