A hybrid machine learning and evolutionary approach to material selection and design optimization for eco-friendly structures - Archive ouverte HAL Accéder directement au contenu
Article Dans Une Revue Structural and Multidisciplinary Optimization Année : 2024

A hybrid machine learning and evolutionary approach to material selection and design optimization for eco-friendly structures

Résumé

In an increasingly competitive and digital industrial environment, the optimization of structures is a key point not only to reduce costs but also to reduce the consumption of natural resources. To this end, different approaches have emerged throughout history based on the tools available at the time. With the current rise of artificial intelligence and the concept of machine learning, revolutionary ideas are emerging that allow an optimal dimensioning of structures in record time. This work presents the use of variational autoencoders and mixed-variable solvers as a proposal for structural optimization and material selection. It has expanded upon previous research by advancing in three directions: (1) incorporating more material attributes, particularly relevant for environmental considerations; (2) analyzing in more detail aspects of VAEs such as the dimensionality of the latent space; and (3) a two-step hybrid approach to select the optimal candidate: preliminary filtering with VAE and final design via mixed-variable model. Various examples demonstrate the applicability of the proposed method.


Mots clés

Fichier principal
Vignette du fichier
16208-Llorente_15961.pdf (4.48 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04599849 , version 1 (04-06-2024)

Identifiants

Citer

Luis Yepes Llorente, Joseph Morlier, Saketh Sridhara, Krishnan Suresh. A hybrid machine learning and evolutionary approach to material selection and design optimization for eco-friendly structures. Structural and Multidisciplinary Optimization, 2024, 67 (5), pp.69. ⟨10.1007/s00158-024-03777-z⟩. ⟨hal-04599849⟩
15 Consultations
10 Téléchargements

Altmetric

Partager

Gmail Mastodon Facebook X LinkedIn More