University of Cambridge > > Isaac Newton Institute Seminar Series > Generative hyperplasticity with physics-informed probabilistic diffusion fields

Generative hyperplasticity with physics-informed probabilistic diffusion fields

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

If you have a question about this talk, please contact nobody.

USMW02 - Mathematical mechanical biology: old school and new school, methods and applications

Complex materials such as soft tissues exhibit nonlinear anisotropic response and hetergeneous mechanical properties. Data-driven methods have been recently developed to capture the rich mechanical behavior of these materials under extreme deformations. In particular, we have contributed to the field by leveraging neural ordinary differential equations (NODEs) as the building blocks of strain energy density functions that automatically satisfy polyconvexity, objectivity, material symmetry and positive energy dissipation requirements for realistic and physically plausible material models. However, these data-driven models have lacked consideration of uncertainty. This is particularly problematic for soft tissues which exhibit a large variation in mechanical properties from one individual to another. Here we establish a generative modeling framework based on stable diffusion to model distributions of materials while satisfying physics constraints. We use NOD Es to describe the material response. Because the NODE framework automatically satisfies the desired physics, any samples of parameters of the NODE produces feasible  materials. For a given material of interest e.g. skin, we assume that stress-strain curves from the population are available. Fitting a subset of the NODE parameters to the stress-strain data yields samples over the parameter space of the NOD Es. Diffusion probabilistic models are then employed to learn that distribution over these NODE parameters and, inplicitly, over the constitutive models. We showcase the ability of the framework to learn the distribution of material behavior for both syntethic examples and murine skin data, outperforming standard density estimation techniques. We anticipate that this work will further establish the use of data-driven methods for materials that exhibit large variation across a population for which uncertainty quantification is essential.      Co-authors: Vahidullah Tac, Manuel Rausch, Ilias Bilionis, Francisco Sahli Costabal

This talk is part of the Isaac Newton Institute Seminar Series series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2024, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity