Sensitivity analysis for multi-physics systems

Explainability of numerical models has become a major challenge for engineers. In this context, sensitivity analysis is a key tool: it makes it possible to identify and rank the most influential parameters on a model’s or system’s performance, thereby strengthening confidence in the results and supporting decision-making. By retaining only the most determinant variables, it also contributes to model simplification and to a significant reduction of computational costs.
This issue is particularly critical for the design of multidisciplinary systems such as electrical machines, floating wind turbines or aerospace vehicles. These systems rely on the integration and coupling of numerical simulators stemming from several physics (electromagnetics, thermal, structural…), which makes uncertainty propagation both complex and expensive, as it combines the management of interdisciplinary couplings with the variability of the inputs.
Conventional sensitivity-analysis methods are generally developed for a single numerical simulator. Their transposition to strongly coupled multidisciplinary systems raises several challenges: a variable that is highly influential in one isolated discipline may have only a minor impact on the overall performance (and vice versa); the evaluation of sensitivity indices (Sobol’, Shapley, HSIC, etc.) involves costly multidimensional integrals compounded by repeated execution of the coupled system; and finally, the transfer of information between disciplines further increases computational complexity.
This PhD project aims to design and adapt sensitivity-analysis methods specifically dedicated to coupled multidisciplinary systems, while limiting the number of expensive calls to multidisciplinary solvers.

Mots clefs: Sensitivity Analysis, Machine Learning, Data Science, Multidisciplinary Design Optimization

• Academic supervisor    Enseignant-chercheur (HDR) Sebastien DA VEIGA, ENSAI-CREST, ORCID: 0009-0004-1637-7942
• Doctoral School    ED MATISSE, Université de Rennes
• IFPEN supervisor    Dr Mohamed Reda EL AMRI, mohamed-reda.el-amri@ifpen.fr, ORCID: 0000-0002-7641-9807 & Dr Adrien SPAGNOL, adrien.spagnol@ifpen.fr
• PhD location    IFPEN, Lyon, France  & ONERA, Palaiseau, France
• Duration and start date    3 years, starting in the fourth quarter 2026 (Novembre 2)
• Employer    IFPEN
• Academic requirements    Master’s degree in engineering, applied mathematics or a related discipline
• Language requirements    French, English level B2 (CEFR)    
• Other requirements    Machine Learning, Uncertainty Quantification, Optimization or related fields

To apply, please send your cover letter and CV to the IFPEN supervisor indicated here below.