This is a past event.

MAE Graduate Seminar Speaker Series

proudly presents

Ryan Sills, PhD

Associate Professor
Rutger University-New Brunswick

Abstract

Neural networks (NNs) provide exciting tools for surrogate modeling applications given their extreme flexibility to learn and their extreme computational efficiency utilizing GPUs. However, in order to enable wide-spread use of NNs for surrogate modeling, two major challenges must be overcome: (1) NN architectures must be developed which can flexibly and efficiently learn the desired physics and (2) training schemes must be developed that do not require overly burdensome amounts of data. The finite-element-based physics-informed neural network (FE-PINN) methodology developed by our group overcomes both challenges. The method is based on a new class of neural network operators called stenciled operators which utilize the inverse isoparametric map in conjunction with the finite element approximation, providing a flexible class of operators with length-scale-invariant computational cost. In comparison, graph-based operators are intrinsically scale dependent in their cost. Furthermore, FE-PINNs do not require labeled data for training, eliminating the data burden. After introducing the method and its advantages, we demonstrate its performance on surrogate modeling tasks in both small-deformation linear elasticity and large-deformation hyper elasticity.

Bio

Ryan Sills received his B.S from Cornell University, his M.S. from the University of Michigan, and his Ph.D. from Stanford University, all in Mechanical Engineering. Prior to joining the faculty in Materials Science and Engineering at Rutgers University in September 2019, he was a Staff Member at Sandia National Laboratories in Livermore, CA for 9 years. At Sandia, he served as a design engineer, researcher, and Principal Investigator for a research program on the mechanics and aging of hydrogen and tritium storage materials. As an Associate Professor at
Rutgers he presently leads the microMechanics of Deformation (μMOD) Research Group, which uses multiscale materials modeling to understand and predict how micromechanics influences the mechanical behaviors of materials. Dr. Sills received his Young Investigator Program award from the Army Research Office in 2021 and his NSF CAREER award in 2023. He became Editor-in-Chief of the Journal of Materials Science: Materials Theory, a Springer Nature journal, in 2024.

Invited by: Trisha Sain