Séminaire en format hybride au GERAD local 4488 ou Zoom.

This research introduces advanced methodologies for optimizing computationally intensive and complex systems, with a particular focus on aeronautical engineering applications. The core approach utilizes surrogate-based optimization, specifically Bayesian Optimization (BO), which employs adaptive sampling to effectively manage the trade-off between exploration and exploitation. The developed in-house BO method, SEGOMOE, is designed to handle a wide array of design variables, including continuous, discrete, categorical, and hierarchical types.

SEGOMOE relies on a mixture of experts, comprising local surrogate models, to address nonlinearities for both the objective and constraints functions, these surrogates are constructed using the open-source Surrogate Modeling Toolbox (SMT), which incorporates adapted kernels for mixed and hierarchical variables. Additionally, SMT supports multi-fidelity extensions, enabling the integration of diverse information sources. Recent results suggest that high-dimensional problems may be efficiently handled using Cooperative Components Kriging and the associated cooperative approach to efficient global optimization. SEGOMOE is designed to solve both single- and multi-objective optimization problems, it also includes mechanisms to handle efficiently hidden constraints. SEGOMOE has been rigorously validated through a benchmark of analytical problems and practical aeronautical applications, showcasing their robustness and efficiency. This work contributes to the field by providing a flexible and powerful BO tool adaptable to a wide range of complex engineering challenges.