ANR Project: GUNSROSES (2027-2031)

Guarantees and UNcertainties on expensive Simulations Reach Operational SafEty Studies


Project summary

The objective of the GUNSROSES project is to solve several methodological challenges which remain in uncertainty quantification and metamodeling methodologies for costly computer codes simulating physical phenomena, in particular in the context of operational safety studies. For example, in the field of nuclear reactor operation or design, thermal-hydraulic simulations are often used to support safety studies, but the use of CFD tools is not yet possible due to insufficiently effective methods.

A first research axis focuses on efficient input sampling for uncertainty propagation, exploring deterministic and Bayesian quadrature methods, handling hidden simulation constraints, parallel computing, and robustness to uncertainty misspecification. A second research axis addresses metamodeling approaches (especially Gaussian Process models) to reduce computational cost while maintaining confidence in predictions. This includes defining trust criteria and developing physics-informed metamodels that combine the strengths of deep learning and probabilistic modeling. A third research axis aims to extend these techniques to high-dimensional and functional outputs, developing dimension-reduction strategies and adaptive design methods suitable for realistic simulations. In addition, two real-scale industrial applications will be considered. The algorithms that will have proven the most efficient on these applications will be integrated in the industrial software of the different partners.

People

Internal meetings

xxx

Publications

Communications

External workshops and Symposium

Software

Position offers

Master internship offers

PhD Thesis position

Postdoc position