A Novel Independent Component Analysis based Process Monitoring Scheme

Abstract

Traditional Independent component analysis (ICA) based process monitoring method used an elliptical measure as monitoring statistic. However, the extracted ICA components usually exhibit skewed distribution and it will decrease the fault detection rate if the elliptical measure is applied. Thus, this study aims to develop a rectangular measure for ICA based process monitoring. The basic idea of proposed monitoring scheme is first to screen out outliers in order to describe well the data majority. Second, the ICA algorithm is used to extract the features of variables and perform dimension reduction. Finally, the extracted ICA components are combined into the rectangular measure as the monitoring statistic. The efficiency of proposed monitoring scheme will be implemented via a five variables simulation example and a case study of Tennessee Eastman process. Results indicate the proposed method outperform several traditional monitoring methods in terms of fault detection rate.