Using pattern recognition to solve mechanics problems
Sanjay R. Arwade
Assistant Professor
Dept. of Civil & Environmental Engineering
University of Massachusetts, Amherst
Abstract
Computational mechanicians have been developing increasingly sophisticated tools for the solution of difficult mechanics problems such as those involving heterogeneous materials. At the same time, computer scientists have developed very sophisticated pattern recognition tools that are capable of identifying individual faces in large crowds. In this presentation I will show how the tools of pattern recognition can be adapted to give approximate solutions to some problems in the solid mechanics of heterogeneous materials. The basic approach is to train a recognition algorithm, a classifier, to associate certain patters with certain kinds of response, and use the trained algorithm to predict the response of material samples.
My discussion will center around the prediction of stress and strain concentration in randomly heterogeneous two dimensional particle reinforced composites and polycrystals. My students and I use support vector machine and decision tree classifiers as pattern recognition algorithms. In closing I will discuss how these tools can be used in the analysis of civil structures as well as materials, and give indications of other directions the research may take.
++++
Sanjay R. Arwade received his Ph.D. and M.S. from Cornell University and B.S.E. from Princeton University, all in Civil Engineering. He taught as an assistant professor at Johns Hopkins University for four years before moving to New England and the University of Massachusetts, Amherst, in 2006. His work in stochastic material mechanics and the analysis and evaluation of historic structures is and has been supported by the National Science Foundation. He roots for the Yankees.
Back to Seminars
