Skip to Main Content
A new on-line predictive monitoring methodology is introduced, based on a combined adaptive principal component analysis (PCA) and an adaptive predictor filter in an autoregressive (AR) formulation. An adaptive PCA-based monitoring scheme is developed utilizing the process historical data to recursively track the dynamic behavior of the industrial process plant based on statistical Hotelling and squared prediction error measures. An adaptive AR filter is proposed to predict the transformed samples in PCA subspaces determined by adaptive PCA algorithm. The difference between the actual transformed process measurements and the predicted measurements in the PCA subspaces is then utilized to discriminate among possible fault occurrences. Performances of the proposed algorithm have been evaluated on Tennessee Eastman (TE) benchmark process plant. The resulting outcomes demonstrate the promising capabilities of the proposed methodology.