Trusted AI-Generated Digital Twins for Research InfraStructurEs

Objective

TwinRISE will deliver a European reference framework for Trusted AI-Generated Digital Twins across research, healthcare, and energy infrastructures. It responds to the Horizon Europe INFRA-TECH-04 call by developing a distributed, AI-native digital twin engine that is modular, interoperable, and trustworthy by design. European research infrastructures face shared challenges: high operational costs, safety-critical constraints, and massive heterogeneous data flows. TwinRISE addresses these by integrating physics-based models with AI surrogates, multimodal foundation models, and federated learning pipelines, hosted on EuroHPC AI Factories. Privacy-preserving methods and enhanced auditability ensure GDPR compliance, data sovereignty, and transparent collaboration.

The project will validate its framework in seven high-impact use cases spanning three domains: (i) healthcare applications such as proton therapy planning and MRI-CT translation (CFB, UCAEN, ULIV, UOX), (ii) accelerator operations and radiation safety (GANIL, KIT, GSI, IFMIF-DONES), and (iii) energy-aware twins for fusion and accelerator-driven systems. Each case will demonstrate predictive maintenance, treatment optimisation, or real-time diagnostic improvements, advancing TRLs from 4-5 to 6-7.

TwinRISE advances the state of the art by combining: generative AI pipelines for automated twin assembly; federated infrastructures enabling collaborative training without raw data exchange; explainability and reliability frameworks aligned with the EU AI Act; and agentic, human-in-the-loop interfaces for transparent decision support. Expected impacts include: up to 30% downtime and commissioning reduction in accelerator and medical systems; 15–20% energy savings via optimised HPC training and control; enhanced patient outcomes through safer and faster planning; and broad uptake of FAIR-compliant models and workflows.