Skip to Main content Skip to Navigation

Réduction de modèle dans une intégration locale-globale, application aux points de soudure pour la simulation des crash-tests.

Abstract : Performing a crash test simulation quickly and accurately remains a challenge for car manufacturers despite the generalization of high-performance computing and the progress in numerical methods. Indeed, engineers introduce mesoscale models of small components in the car model to obtain excellent prediction accuracy, but the integration of these local models increases computational time. Model order reduction methods that exploit the generally low dimensionality of the loading space of a physical phenomenon and the correlations in the loading are interesting to improve the accuracy of a crash test model at a low computational cost. In this thesis, we address the reduction of mesoscale spot weld models in ESI Group's high-fidelity crash test model to decrease the computational time of a simulation. Professor Chinesta and Dr. Daim supervised this work conducted inside the consortium dedicated to model order reduction for industrial problems, comprising Gestamp, Renault, ESI Group, and Arts et Métiers.We propose (i) a new data-driven reduction technique called incremental Dynamic Mode Decomposition (iDMD) and (ii) an integration scheme, the local-global approach, to introduce this reduced model into a finite element model. We applied these techniques to reduce a mesoscale model of a spot weld. The use case represents an assembly of two metal sheets subject to tensile and shear stresses until failure. In less than 10 minutes, we performed a simulation comparable to the reference simulation performed in 1h30, with the same computing power. The accuracy of the reduced model is excellent as long as the damage is not advanced. During the rupture, the prediction error remains low even with a reduced training compared to the problem's dimensionality. Additionally, we integrated the spot weld reduced model into a global model representing two sheets using the local-global approach. We obtained results comparable to a high-fidelity simulation integrating a spot weld mesoscale model. Thus, the iDMD and the local-global approach associated with a crash-test model would integrate the spot welds' local behaviors without explicitly describing them. Following these encouraging results, we started a more extensive study with Renault on national supercomputers (Jean Zay). These supercomputers allow us to explore the parametric loading space encountered during a crash test to integrate our approach in the dedicated ESI Group software.This thesis is divided into four chapters. In the first chapter, we introduce the crash test simulation and present our state of the art on model order reduction. In the second chapter, we present the iDMD and the local-global approach. In the third chapter, we present the use case on spot welds. Finally, the last chapter is the conclusion of our approach.
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Friday, June 3, 2022 - 3:42:11 PM
Last modification on : Wednesday, September 28, 2022 - 5:57:49 AM
Long-term archiving on: : Sunday, September 4, 2022 - 7:27:47 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03687825, version 1


Agathe Reille. Réduction de modèle dans une intégration locale-globale, application aux points de soudure pour la simulation des crash-tests.. Génie mécanique [physics.class-ph]. HESAM Université, 2021. Français. ⟨NNT : 2021HESAE074⟩. ⟨tel-03687825⟩



Record views


Files downloads