Time Series Classification in High-Pressure Die Casting Manufacturing using Dynamic Time Warping

This paper presents a comprehensive approach to time series classification in high-pressure die casting (HPDC) manufacturing using Dynamic Time Warping (DTW) algorithms for automated quality prediction and process optimisation. As part of the metaFacturing European project, we developed machine learning (ML) models that analyse velocity and pressure profiles from HPDC machines to predict casting quality and identify critical process parameters influencing defect formation. We evaluated three classification models—Support Vector Machine (SVM), Linear SVM, and Gradient Boosting Classifier (GBC)—with GBC demonstrating superior performance (weighted F1 score reaching up to 92%). Though, optimizing the weighted F1 score resulted in lower recall (28%) and F1 score (42.42%) for the minority ‘Bad’ class. To ensure practical applicability in industrial settings, the methodology incorporates expert feedback and explainable AI techniques, providing process engineers with transparent, actionable insights for informed decision-making. Although the analysis across different products revealed performance variations, indicating the product-specific nature of velocity and pressure signature patterns, the results still demonstrated the applicability of DTW-based feature extraction combined with classifier models. The approach enables real-time data-driven quality assessment and yields deeper insights supporting process parameter optimisation in industrial manufacturing environments.