A hybrid approach for statistical process control using independent component analysis, support vector machine and local outlier factor

Abstract

We propose a new approach to Statistical Process Control (SPC) by integrating Independent Component Analysis (ICA), Local Outlier Factor (LOF) and Support Vector Machine (SVM). Since LOF value has been calculated based on density, it has high discriminative power regardless of the statistical distribution of data. Therefore, the performance of existing ICA and SVM based SPC approaches can be improved by adopting LOF as the monitoring statistic of SPC. The comparison experiments with several existing SPC approaches were conducted on widely used benchmark dataset, Tennessee Eastman process, in which the proposed approach showed the best fault detection rate. By the result, it was concluded that integrating ICA, LOF and SVM enhanced the process monitoring performance of SPC. © 2011 ISSN 1881-803X.