Analysis of Multivariate Industrial Process Data for Quality Pattern Detection in Masterbatch Coloring

L. Hisgen (THM Univ. of Applied Sciences and G.E. Habich Farben GmbH, Germany), T. Kubik (THM Univ. of Applied Sciences, Germany), S. Garbe (G.E. Habich Farben GmbH, Germany), J. Fischer (superus Datenmanagement GmbH, Germany), A. Kloes (THM Univ. of Applied Sciences, Germany), B. Iniguez (Univ. Rovira i Virgili, Spain), M. Schwarz (THM Univ. of Applied Sciences, Germany)

This paper analyzes clustering techniques for identifying quality-related structures in multivariate industrial process data, demonstrated through a case study on masterbatch data from the coloring industry. The dataset consists of batches labeled as successful or unsuccessful based on quality inspection results. The proposed approach combines dimensionality reduction, clustering, and statistical analysis to examine relationships between ingredient composition and quality outcomes. Multiple clustering methods are compared to identify characteristic differences in ingredient usage. The results demonstrate that clustering methods reveal distinct patterns in ingredient composition associated with quality outcomes.