An integrated process monitoring approach combining dynamic independent component analysis and local outlier factor

In this paper a novel process monitoring scheme for reducing the type І and type ІІ error rates in the monitoring phase is proposed. First، the proposed approach uses an augmented data matrix to implement the process dynamic. Then، we apply independent component analysis (ICA) transformation to the augmented data matrix، and eliminate the outliers using the local outlier factor (LOF) algorithm. Finally، the control limit based on the LOF value of the cleaned data are obtained. In the monitoring phase، if the LOF value of each sample exceeds the control limit، fault has occurred; otherwise، data is normal. The proposed method is applied to fault detection in both a simple multivariate dynamic process and the Tennessee Eastman process. In both processes، type І and type ІІ error rates are witnessed to reduce by considering the process dynamic and performing the LOF algorithm. Results clearly indicate better performance of the proposed scheme compared to the alternative methods.