Batch process control and monitoring: a Dual STATIS and Parallel Coordinates (DS-PC) approach

Abstract

Multivariate data collected from batches is usually monitored via control charts (CCs) based on MPCA and MPLS for batch to batch comparison. In addition, distribution free approaches include other dimensionality reduction methods for batch and time-wise analysis. However, techniques for multivariate data focused on variable-wise analysis haven’t been widely developed. Here, we propose a nonparametric quality control strategy for off-line monitoring of batches and variables, besides visual clustering of observations within batches. In our approach, CCs based on Dual STATIS are created using robust bagplots to enhance signal detection in batch and variable-wise analysis, while parallel coordinate plots are used in identification of unusual observations’ behavior per variable, regardless distributional assumptions. This proposed strategy poses the main advantage of detecting different type of changes through meaningful visualization tools, allowing easier interpretation of results in industrial settings.