Adam Obrusnik, Kristina Tomankova, Krystof Mrozek, Martin Kubecka, Petr Bilek, Petr Zikan, PlasmaSolve s.r.o., Brno, Czech Republic
Digital twin models are revolutionizing both thin film R&D and production environments in the coatings industry. While the definition of a “digital twin” in our industry is still being defined, most authors agree that it is a conceptual agglomerate of simulation, process mining, and machine learning that can predict and optimizing key performance properties of thin film coatings. This work presents an implementation of a Digital Twin framework by PlasmaSolve accompanied by practical use cases where digital twin models have significantly improved production outcomes.
First, we explore the development of automated recipe design for Ti-Al-Si-N coatings, demonstrating how combining physics simulations with machine learning enables accurate prediction of composition, hardness, and stress. These models reduce reliance on extensive experimental datasets while maintaining robust predictions, even beyond the training dataset.
Second, we address troubleshooting production challenges, highlighting how digital twins and process mining uncover root causes of unexpected property deviations. A real-world example showcases how a variation of a cooling protocol has been linked to quasi-random variations in coating performance. By integrating diverse data sources and leveraging advanced analytics, the digital twin approach delivers actionable insights and ensures consistent product quality.
Key lessons include the necessity for transparent, well-structured data pipelines, cross-disciplinary expertise in data science and physics, and proactive collaboration with stakeholders. The presentation concludes with actionable recommendations for OEMs, operators, and digital services providers to maximize the potential of digital twins in production environments.