Aims to deliver a secure, trustworthy, and reusable platform for AI-generated digital twins (DTs) across European Research Infrastructures.
TwinRISE (Trusted AI-Generated Digital Twins for Research Infrastructures & Scientific Excellence) integrates generative AI, federated learning, explainability, and uncertainty quantification into a distributed, AI-native Digital Twin Engine, supporting human-in-the-loop operations while ensuring compliance with the EU AI Act and GDPR.
Objectives
- Develop a modular, interoperable Digital Twin Engine spanning accelerators, healthcare, and radiation safety.
- Advance seven use cases across three domains, with three selected as integrated pilots.
- Ensure trust-by-design, embedding safety, explainability, and regulatory compliance at every stage.
- Leverage EuroHPC AI Factories for energy-efficient large-scale model training.
- Release open and FAIR-compliant datasets, models, and workflows via EOSC and AI-on-Demand.
Innovation Beyond State of the Art
TwinRISE moves beyond domain-specific digital twins by creating a cross-domain, interoperable framework. It extends the interTwin Digital Twin Engine with:
- Federated learning and privacy-preserving training across clinical and physics infrastructures.
- Explainable, agent-based control pipelines for safety-critical operations.
- Energy-aware AI, targeting ≥15–20% reduction in HPC training and inference costs.
- A unified trust and audit toolkit for AI compliance and user transparency.
Expected Impacts
- Scientific: Faster discovery through predictive surrogates, reproducibility via FAIR repositories, and ≥60% faster simulations with ≥95% accuracy.
- Technological & Economic: Industry-ready modules for predictive maintenance, medical imaging, and HPC; contribution to European standards for interoperability.
- Societal: Safer and faster proton therapy, reduced downtime in accelerators (≥30%), radiation protection with auditable monitoring, and improved sustainability.
- Policy: Direct contribution to EU AI Act evidence base, FAIR adoption (>80% datasets/models), and integration with GenAI4EU, EuroHPC, and EOSC.
Supporting Institutions
This initiative is backed by strong institutional support: