Multi-fidelity bayesian optimization using model-order reduction for viscoplastic structures

Stéphane Nachar (1) , Pierre-Alain Boucard (1) , David Néron (1) , Christian Rey (2)

1 LMT - Laboratoire de Mécanique et Technologie
2 Safran Tech

The numerical optimization of a mechanical part requires a balance between computation time and model accuracy. The work presented herein aims at accelerate global optimization problem by using the framework of Bayesian optimization on a quantity of interest with multiple levels of fidelity. These multi-fidelity data are generated from a quality-driven model-order reduction framework: the LATIN Proper Generalized Decomposition. Within this framework, a reduced-order basis is generated on-the-fly and re-exploited to reduce the computational cost of observations. This strategy is tested on two elasto-viscoplastic test cases: a rocket damper and an aircraft blade and gives significant speedups.

Mots clés: Bayesian optimization; Multi-fidelity kriging; Reduced-order models

Domaines: Mécanique des structures [physics.class-ph]

Stéphane Nachar, Pierre-Alain Boucard, David Néron, Christian Rey. Multi-fidelity bayesian optimization using model-order reduction for viscoplastic structures. 2019. ⟨hal-02396283⟩