Akce ČVUT

Matti Schneider (KIT) Component-scale upscaling in solid mechanics with Deep Material Networks

06.09.2022   14:00-15:00

Computational micromechanics permits to obtain the mechanical response of materials with complex microstructure provided the geometrical representation of the microstructure is known, together with suitable material models for the phases constituting the heterogeneous material. Although accurate computational methods are available to resolve the corresponding equations, using such technology in a concurrent two-scale setting, i.e., for component-scale simulations, is non-trivial due to the inherent computational challenges involved.

This talk covers a neural-network inspired microstructure surrogate to be used in concurrent multiscale simulations. Based on a hierarchy of laminates, these so-called deep material networks (DMNs) are trained on linear elastic data and applied to inelastic constitutive laws with astonishing accuracy. We report on improvements of the original proposition by Zeliang Liu and coworkers, in particular discussing modifications to handle large-scale problems.

Místo konání

Room B-365, Thákurova 7, 166 29 Prague 6

Kontaktní osoba

Jan Zeman, jan.zeman@cvut.cz

Podrobnější informace

https://mat.fsv.cvut.cz/seminar/