MSE Colloquium: Nicholas Zabaras, An information theoretic approach to predictive materials modeling

Abstract

We will briefly discuss a number of fundamental problems in predictive materials modeling to account for uncertainties in models, microstructure data, and coarse graining. As an example, we will introduce predictive multiscale models for deformation processes of polycrystalline materials. We will address methods for quantifying uncertainty in polycrystal microstructures and computing the probability distribution of the observed macroscale properties. A surrogate reduced order stochastic model will be introduced to address location-dependence ofmicrostructures. A multiscale forging problem will be discussed to study the effects of uncertain initial grain size distribution and texture on the macroscopic properties. To address issues of complexity, we will finally pose such multiscale stochastic problems as inference problems in graphs.

Bio

Nicholas Zabaras' research group performs applied and basic research in the interface of computational mathematics and materials. His main research focus is on multiscale modeling of materials (from ab initio to continuum) and multiscale design of materials processes for control of microstructure-sensitive properties. Emphasis of recent work is in quantifying and controlling uncertainty propagation in materials across scales. For achieving these objectives, his laboratory is developing a variety of mathematical and statistical computational tools that include multiscale and multiphysics materials simulators, homogenization techniques, machine learning techniques for exploring large multiscale material databases, level set and phase field methods, techniques for modeling stochastic PDEs in high-dimensions, Bayesian inference approaches and non-linear model reduction algorithms.

Professor N. Zabaras is the director of the Materials Process Design and Control Laboratory and detailed information on current research interests and list of publications and other activities can be found at the MPDCL web site.