Sparse dynamic inner principal component analysis for process monitoring

Abstract

In this paper, a novel sparse dynamic inner principal component analysis (SDiPCA) method is proposed for process monitoring. First, a simple regression-type approach of dynamic inner principal component analysis (DiPCA) is discussed. To derive sparse principal components, an elastic net regularization is imposed on this regression-type problem. Then a new optimization criterion is established and solved through an alternating algorithm. On the basis of the SDiPCA model, four monitoring statistics are constructed to reflect the process status. Also, the reconstruction-based contribution (RBC) method is employed to isolate faulty variables. Finally, a case study on the Tennessee Eastman process is conducted to illustrate the superior performance of the proposed SDiPCA method compared with DiPCA method.