An extension to RPCA parameter selection and process monitoring
A. Rehmer, and A. Kroll. Preprints of the 20th IFAC World Congress, page 15329-15334. IFAC, (2017)
Multivariate Statistical Process Control (MSPC) techniques such as Principal Component Analysis (PCA) and Partial Least Squares (PLS) have found wide application especially in the statistical modeling and monitoring of chemical processes. However, real industrial processes often violate the assumptions underlying MSPC since they exhibit timevarying and non-stationary behavior. Adaptive PCA-based monitoring procedures such as Moving Window PCA (MWPCA) and Recursive PCA (RPCA) have been proposed to tackle this issue. Although the parameter selection for those procedures is critical to their proper implementation, this topic is rarely covered in the literature. This paper examines two methods for MWPCA and RPCA parameter selection using the Tennessee Eastman process as an example. Based on the findings a novel procedure for RPCA parameter selection as well as a extension to RPCA will be proposed and demonstrated.