Proceedings of the
35th European Safety and Reliability Conference (ESREL2025) and
the 33rd Society for Risk Analysis Europe Conference (SRA-E 2025)
15 – 19 June 2025, Stavanger, Norway
Adaptive Gaussian Process-Based strategies for Solving the NASA-DNV UQ Challenge 2025
1ONERA/DTIS, Université de Toulouse, Toulouse, F-31055, France.
2Fédération ENAC ISAE-SUPAERO ONERA, Université de Toulouse, 31000, Toulouse, France
3ONERA/DTIS, Université Paris Saclay, Palaiseau, F-91123, France.
ABSTRACT
This paper describes a dedicated approach to solve the 2025 NASA-DNV UQ challenge problem using adaptive Gaussian process strategies. The uncertainty model is determined through a calibration problem using an optimization approach to identify the aleatory variable joint distribution and the epistemic variable uncertainties. The estimation of the prediction interval for the model output components consists of a quantile estimation problem based on an adaptive Gaussian process strategy. Eventually, the design optimization problems are solved using Bayesian optimization controlling the noise level involved in the estimation of the objective and constraint functions.
Keywords: Uncertainty quantification, Optimization, Gaussian process, Active learning.
