Manifold Learning for model order reduction in mechanics of materials

Abstract: 

The talk involves a survey of key methods of model order reduction applied to model-based engineering and digital twining, by learning linear or nonlinear latent spaces. In this survey, projection-based reduced order models are the projection of mechanical equations on a latent space that have been learnt from both synthetic data and experimental data. Various descriptions and representations of structured data for model reduction are presented . Image-based digital twins are developed in a reduced setting. Reduced order models of as-manufactured components predict the mechanical effects of shape variations. A similar workflow is extended to multiphysics or coupled problems, with high dimensional input fields. Practical techniques are proposed for data augmentation and also for hyper-reduction, which is a key point to speed up projection-based model order reduction of finite element models.

    
Manifold Learning, Model Reduction in Engineering
David Ryckelynck , Fabien Casenave , Nissrine Akkari
Springer Nature Switzerland, 2024, SpringerBriefs in Computer Science, ⟨10.1007/978-3-031-52764-7⟩

Le mardi 2 décembre 2025 à 11h00 / Amphithéâtre François Canac, LMA

Voir en ligne : plus d’informations concernant l’orateur