Mechanical and Civil Engineering Seminar

Mechanical and Civil Engineering Seminar Series

Title: Beyond Black-Box Machine Learning for Mechanics

Abstract: Machine learning (ML) is a growing research area of algorithm development and its application to a multitude of science and engineering disciplines. Poorly applied, black-box applications of ML with model overfitting and nonphysical behavior may give mechanics researchers pause to attempt ML. This hesitancy can lead us instead to seek and develop appropriate and perhaps more nuanced uses of ML to mechanics where we incorporate first-principles physics and often difficult-to-use microstructural or full-field experimental data into our modeling. This presentation includes three examples of such research at Sandia alongside university collaborators. First, our physics-informed neural networks (PINNs) approach to constitutive model calibration embeds the principle of stationary potential energy into the loss (i.e. error) function and is trained using full-field experimental surface displacement data of heterogeneous deformation. Second, we developed a physics-informed Reinforcement-Learning Design-of-Experiment (piRL-DOE) algorithm for optimal model calibration of time-dependent plasticity models, using the uncertainty of the model parameters to drive the selection of optimal experimental steps to reduce that uncertainty. Third, we proposed a neural network framework to model general inelastic materials that has recognizable stress and flow rules, is extensible to modeling complex behavior and microstructure, and prevents violations of the second law of thermodynamics by construction. The presentation ends with an outlook of the ML work for mechanics at Sandia. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-NA0003525.

Bio: Dr. Sharlotte Kramer is a Distinguished Member of the Technical Staff at Sandia National Laboratories. She leads a research program in the Engineering Sciences Center that advances predictive solid mechanics for rapid transformation of qualification and sustainment of national security systems. She is the principal investigator of multi-disciplinary projects, transitioning fundamental research to mission applications at the intersection of solid mechanics, materials science, computer science, and data science, with internal Sandia collaborators and external collaborators around the globe. Sharlotte received her bachelor's degree in Aerospace Engineering from the University of Virginia with Ioannis Chasiotis in 2004 and her master's and Ph.D. degrees in Aeronautics from the California Institute of Technology with G. Ravichandran and Kaushik Bhattacharya in 2005 and 2009, respectively. She was a postdoctoral researcher in Materials Science and Engineering at the University of Illinois at Urbana-Champaign with Nancy Sottos before joining Sandia National Laboratories in 2011. She has been a member of the Society of Experimental Mechanics since 2007 and serves as a Member-at-Large on their Executive Board. She is on the Strain Journal Editorial Board and has been a guest editor for the Strain Journal and for the International Journal of Fracture.

NOTE: At this time, in-person Mechanical and Civil Engineering Lectures are open to all Caltech students/staff/faculty/visitors.