Multivariate statistical process monitoring of propylene polymerization with principal component analysis and support vector data description

This paper addresses fault diagnosis and identification of propylene polymerization process for which the recorded variables follow non-Gaussian distributions. Recent work has demonstrated the effectiveness of principal component analysis (PCA) in dimension reduction and support vector data description (SVDD) in non-Gaussian monitoring statistics. This article extends this work by combining principal component analysis with support vector data description and introducing a fault identification technique to diagnose abnormal process cause. The research results confirm the utility of the proposed method.