Improvement on multivariate statistical process monitoring using multi-scale ICA

Abstract

A multi-scale independent component analysis (ICA) approach is investigated for industrial process monitoring. By integrating the ability of wavelet on multi-scale analysis and that of ICA on extracting independent components for non-Gaussian process variables, the multivariate statistical monitoring techniques can obtain improved performance. Contrastive tests have been carried out on the famous benchmark chemical plant among ICA-like and PCA-like methods, which reveals that multi-scale ICA approach has lower missed detection rate of faults. © Springer-Verlag Berlin Heidelberg 2006.