Non-gaussian Industrial Process Monitoring Based On Data-driven

With the continuous improvement of science and technology,industrial production technology has also developed rapidly,and the industrial production process has become more and more complicated.In addition,with huge investment in industry,industrial production safety is becoming more and more important,then process monitoring is becoming an effective way to ensure industrial production safety.Among the various methods of process monitoring,data-driven process monitoring is a more popular method,and characterized by no need for models and experience,which makes it have good applicability and generalization.Statistical process monitoring is a data-driven monitoring method that can be divided into Univariate Process Control(USPC)and Multiple Statistical Process Control(MSPC).Independent Component Analysis(ICA)is one of the important methods of MSPC,which can not only process non-Gaussian distribution process data,but also use higher order statistical information of process data.However,the application of ICA to process monitoring has two problems: One is how to select independent component(IC),the other one is the calculation efficiency is lower and the number of modeling samples is large when use kernel density estimation(KDE).In view of this problem,two improved methods are proposed in this paper,one is the Adaptive Independent Analysis(AICA),and the other is the Adaptation Independent Analysis-Support Vector Data Description(AICA-SVDD).For how to choose IC when ICA method is applied to process monitoring based on AICA process monitoring,a separation matrix is used to build an association matrix in offline modeling,which can represent the similarity of IC;when online monitoring,the components of the online samples are obtained,and the IC with the smallest probability density is selected as the Particular Component(PIC)by kernel density estimation;then some common independent components(CICs)similar to PIC are selected by the correlation matrix;finally,PIC and CICs are used to construct statistics at the same time,and control limits are determined by KDE.By applying AICA to numerical simulation and Tennessee Eastman(TE)simulation process,test results have shown that the proposed method is effective.For the control limit needs to be calculated in real time when using AICA,and the calculation efficiency is lower and the number of modeling samples is large when using KDE,a process monitoring method based on AICA-SVDD is proposed to solve these.Support Vector Data Description(SVDD)is used to replace the KDE to calculate the control limit;after the introduction of the SVDD,the control limit calculation efficiency is improved and the number of modeling samples is reduced;after numerical simulation of AICA-SVDD and its application to TE simulation process,test results have shown that the proposed method is effective for process monitoring.Finally,based on the summary of the whole work,the future research in the field of process monitoring is prospected.