PLS Algorithm Based On Mutual Information And Its Application In Quality-related Process Monitoring

Quality-related fault detection plays a vital role in the industrial production process.Its purpose is to detect the faults that affect the normal production in the industrial production process,and to improve the qualityrelated fault detection rate as high as possible to ensure that the staff can deal with these kinds of faults in time,so as to ensure that the production can be carried out safely and effectively and improve the quality of products.Thanks to the development of computer technology and intelligent instrument technology,multivariate statistical analysis method,as a data-driven monitoring method,has been widely used in industrial production process monitoring.However,traditional multivariate statistical process monitoring methods use all process variables for modeling,some of which may have no correlation with quality variables.Moreover,the premise of these methods is to assume that the original data meet the linear relationship and Gaussian distributionr,while the data generated in the actual production process often has certain nonlinear and non-Gaussian characteristics.Therefore,it is very significance to consider the nonlinear and non-Gaussian structure characteristics of modeling data for quality-related fault detection in actual production process.In response to the above problems,we have made the following improvements based on the partial least squares(PLS)method:1.To solve the problem that the traditional PLS method use all the process variables in the process of extracting principal components for modeling,we use the method of mutual information combined with PLS to eliminate the process variables of irrelevant quality,and construct a sparse PLS method based on mutual information.By using the data generated by the Tennessee Eastman(TE)process to conduct modeling and simulation experiments,the prediction analysis and fault monitoring of the output of the simulation process are carried out,and the effectiveness of the proposed method is verified by comparing with some existing improved PLS methods.2.Considering that traditional monitoring methods does not consider the problem of non-Gaussian structure information in data preprocessing and feature extraction,combined with mutual information can capture the non-Gaussian statistical correlation between variables,this paper proposes a double-mutual information PLS method based on Bayesian fusion.By using the data generated by the TE process to conduct modeling and simulation experiments,the prediction analysis and fault monitoring of the output of the simulation process are carried out,and the effectiveness of the proposed method is verified by comparing with some existing improved PLS methods.