An integrated multiresolution framework for quality prediction and process monitoring in batch processes. (October 2020)

• Record Type: Journal Article
• Authors: Rato, Tiago J.; Reis, Marco S.
• Is Part Of: Journal of manufacturing systems. Volume 57(2020)
• Page Start: 198
• Page End: 216
• Publication Date: 2020-10
• Subjects: Multiresolution soft sensor -- Multiway partial least squares -- Variable selection -- Process monitoring -- Industry 4.0 -- Industrial data science; Manufacturing processes -- Periodicals; Production engineering -- Data processing -- Periodicals; Robots, Industrial -- Periodicals; Production, Technique de la -- Informatique -- Périodiques; Robots industriels -- Périodiques; Electronic journals; 670.42
• DOI: 10.1016/j.jmsy.2020.09.007
• Languages: English
• ISSNs: 0278-6125

Abstract

Highlights: End-batch quality is predicted by a Multiresolution Partial Least Squares approach: MR-PLS. MR-PLS selects the variables and time intervals with critical impact on end-batch quality. Process monitoring is done by a translation-invariant wavelet decomposition approach: TIME-PCA. TIME-PCA can adapt to the characteristics of each fault and thus have higher sensitivity. The reliability of MR-PLS predictions is assessed by combining MR-PLS and TIME-PCA. Abstract: This article presents a new integrated multiresolution framework for quality prediction and process monitoring in complex batch operations: MR-QP&PM. For quality prediction, the proposed framework uses a multiresolution soft sensor that identifies the Critical to Quality (CTQ) variables and optimizes their resolution. This later tuning dimension has been overlooked in previous works and has a significant impact on the predictive accuracy of soft sensors, especially for batch processes. To complement the soft sensor and effectively assess the validity of its predictions and whether the process is operating under normal operating conditions (NOC), a monitoring scheme with multiresolution adaptability is applied. In case a fault is detected, MR-QP&PM is capable to discern whether the fault is CTQ and possibly estimate its impact. MR-QP&PM was tested and compared with current benchmark methods using several batch testing systems, including the well-known PENSIM simulator, where its superior predictive performance and detection ability was demonstrated. A critical analysis of when the quality predictions of MR-QP&PM can and cannot be trusted is also made, in order to set the boundaries where the proposed methodology can be safely applied.