BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Laura De Lorenzis (ETH Zurich) DTSTART:20210608T070000Z DTEND:20210608T080000Z DTSTAMP:20260423T052759Z UID:CMWebinar/6 DESCRIPTION:Title: Unsupervised discovery of constitutive laws\nby Laura De Lorenzis (E TH Zurich) as part of Australian Seminar on Computational Mathematics\n\n\ nAbstract\nThe speaker and her group recently proposed a new approach for data-driven automated discovery of constitutive laws. The approach is unsu pervised\, i.e.\, it requires no stress data but only displacement and glo bal force data\, which are realistically available through mechanical test ing and digital image correlation techniques\; it delivers interpretable m odels\, i.e.\, models that are embodied by parsimonious mathematical expre ssions discovered through sparse regression of a large catalogue of candid ate functions\; it is one-shot\, i.e.\, discovery only needs one experimen t — but can use more if available. The problem of unsupervised discovery is solved by enforcing equilibrium constraints in the bulk and at the loa ded boundary of the domain. Sparsity of the solution is achieved by Lp reg ularization combined with thresholding\, which calls for a non-linear opti mization scheme. The ensuing fully automated algorithm leverages physics-b ased constraints for the automatic determination of the penalty parameter in the regularization term. We focus on isotropic hyperelasticity and\, us ing numerically generated data including artificial noise\, we demonstrate the ability of the approach to accurately discover five hyperelastic mode ls of different complexity. We also show that\, if a “true” feature is missing in the function library\, the proposed approach is able to surrog ate it in such a way that the actual response is still accurately predicte d. We finally outline the first steps in the direction of extending the ap proach to more complex types of constitutive laws.\n LOCATION:https://researchseminars.org/talk/CMWebinar/6/ END:VEVENT END:VCALENDAR