13–15 Oct 2026
Hotel Caesius Terme & SPA Resort - Bardolino - Verona
Europe/Rome timezone

Physics Based Modelling as an Enabler for Machine Learning Defect Prevention in Casting

13 Oct 2026, 16:30
20m
Gardenia room (Hotel Caesius Thermae & Spa Resort)

Gardenia room

Hotel Caesius Thermae & Spa Resort

Via Peschiera, 3, 37011 Bardolino VR

Speaker

Orlando Di Pietro (RINA Consulting - Centro Sviluppo Materiali SpA)

Description

The growing use of machine learning (ML) in continuous casting offers new opportunities for process monitoring and quality optimisation. However, ML approaches that do not explicitly incorporate physical knowledge often show limited robustness, interpretability and extrapolation capability when dealing with complex defect phenomena such as crack formation.
This contribution reports original work carried out within the RFCS SUNSHINE project, Here, physics‑based numerical modelling is used as a backbone to support data‑driven approaches for surface quality improvement and crack prevention. The work focuses on integrating models describing key physical phenomena during continuous casting — including fluid flow, heat transfer, solidification and shell growth — with ML workflows for process and quality optimisation, but it can properly addresses to ingot casting too. Thermo‑fluid‑dynamic models capture melt flow behaviour, turbulence, meniscus conditions and local heat flux distributions, while thermo‑mechanical and solidification models provide insight into shell growth, thermal gradients and stress–strain development relevant to crack susceptibility.
Attention is devoted to techniques used to merge fluid-dynamics with thermodynamics and solidification, for a physically coherent description of flow‑driven heat transfer and phase‑change. Rather than acting as standalone predictor, coupled modelling generates physically meaningful indicators — such as critical thermal and mechanical conditions, and heat extraction patterns —exploited as high‑quality features for ML algorithms developed within SUNSHINE.
The resulting hybrid modelling–ML framework improves prediction robustness and generalization across steel grades, caster configurations and operating conditions. The contribution highlights how knowledge consolidation activities, accelerating the deployment of explainable and industrially applicable digital tools for quality, productivity and sustainability improvement in casting and solidification frame, either in a continuous or ingot production frame.

Speaker Company/University RINA Consulting - Centro Sviluppo Materiali SpA

Author

Michele De Santis (Rina Consulting - Centro Sviluppo Materiali SpA)

Co-authors

Franco Macci (RINA Consulting - Centro Sviluppo Materiali SpA) Julius Norrena (University of Oulu) Massimo Milone (RINA Consulting - Centro Sviluppo Materiali SpA) Orlando Di Pietro (RINA Consulting - Centro Sviluppo Materiali SpA) Tuomas Alatarvas (University of Oulu)

Presentation materials

There are no materials yet.