Energy-conserving equivariant GNN predictions of stiffness for lattice materials

Add to your list(s) Download to your calendar using vCal

• Ivan Grega, CUED
• Friday 26 January 2024, 16:00-17:00
• JDB Seminar Room, CUED.

If you have a question about this talk, please contact div-c.

Lattices emerged in recent decades as a promising class of architected materials with a vast design space. Many machine learning models have been proposed as surrogate to numerical modelling in predicting their mechanical properties for rapid design applications. However, they are often not scalable, lack the appropriate physical constraints and hence are limited to a small fragment of the vast design space. Here we develop a graph based neural network to predict the fourth-order stiffness tensor of any arbitrary periodic lattice. We build upon the equivariant MACE model (Batatia, Kovács, Csányi et al.) and introduce positive semi-definite constraints that ensure energy conservation. We trained the model on a generalised dataset of unit cells and demonstrate an example application of the model in structural optimization.

This talk is part of the Engineering - Dynamics and Vibration Tea Time Talks series.

This talk is included in these lists:

• All Talks (aka the CURE list)
• Cambridge University Engineering Department Talks
• Cambridge talks
• Centre for Smart Infrastructure & Construction
• Civil Engineering Talks
• Computational Continuum Mechanics Group Seminars
• Engineering - Dynamics and Vibration Tea Time Talks
• Engineering - Mechanics, Materials and Design (Div C) - talks and events
• Featured lists
• Interested Talks
• JDB Seminar Room, CUED
• School of Technology
• Trust & Technology Initiative - interesting events
• bld31

Note that ex-directory lists are not shown.