• Citer

On the integration of domain adaptation and causal discovery in digital twins: A case study for plastic injection molding

par Paldino, Gian Marco ;Caelen, Olivier ;Oueslati, Marouene;Ansay, Marc;Johanesa, Tojo Valisoa T.V.A.;Bontempi, Gianluca
Référence Future generations computer systems, 182, 108427
Publication Publié, 2026-01-01

Article révisé par les pairs

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

Digital Twin (DT) technology is foundational to Industry 4.0, yet its transition from single-asset replicas to interconnected Digital Twin Ecosystems presents significant engineering challenges. In manufacturing, these ecosystems face constant domain shifts caused by variations in raw materials, environmental conditions, and machine configurations, which degrade the performance and reliability of predictive models. Furthermore, the black-box nature of many machine learning models hinders trust and adoption on the factory floor. This paper proposes a novel engineering framework for developing scalable and interpretable Digital Twin Ecosystems that are robust to domain shifts. We address these challenges by integrating two key technologies at the core of the DT design: (1) Domain Adaptation, to ensure models remain accurate when faced with new, unseen production conditions with minimal labeled data, and (2) Causal Discovery, to learn interpretable models of the underlying process dynamics, enhancing model robustness and operator trust. We validate our framework using real-world data from the plastic injection molding industry. Our experimental results demonstrate that the synergy between domain adaptation and causal feature selection significantly improves predictive performance, paving the way for more resilient, flexible, and trustworthy DT ecosystems in complex manufacturing environments.