Process monitoring in polymer extrusion by multivariate data analysis

In the context of Industry 4.0, this thesis, developed in collaboration with Sirmax S.p.A., proposes a multivariate statistical approach to optimize process control and product quality in polymer compounding. The research focuses on the manufacturing of customized polypropylene (PP) compounds, where a major industrial challenge is managing the high variability of incoming raw materials, which inherently exhibit fluctuations in their physical and rheological properties. The core and most time-consuming phase of this study involved constructing a comprehensive data engineering pipeline to bridge the gap between Information Technology (IT) and Operational Technology (OT). This was achieved by systematically extracting and aligning data from two primary corporate sources: high-frequency time-series sensor data directly monitored by the machine PLCs, and structured administrative and logistical records managed within the SAP ERP system. The consolidated dataset was utilized to address two distinct analytical goals. First, Principal Component Analysis (PCA) was implemented as an unsupervised method for real-time process monitoring, evaluating machine parameters to identify anomalies and deviations from standard operations. Second, Partial Least Squares (PLS) regression was developed as a supervised framework to predict final product quality metrics directly from online process variables. Ultimately, this data-driven integration provides the company with robust analytical instruments to master process variability and ensure strict product standardization.