On the Design of Polycrystalline Materials with an Integration of Multiscale Modeling and Statistical Learning

Cornell University, Ithaca NY

Investigators

Abstract

The research project provides a new outlook to materials design problems by integrating microstructural models with virtual databases and statistical learning tools. In particular, emphasis is given to (1) Development of innovative computational techniques for microstructure interrogation and multi-scaling of micro-scale deformation and failure, (2) Development of hierarchical microstructure libraries of microstructural signatures (crystal orientations and grain size distributions) with associations to properties (such as strength, toughness and formability) using statistical learning algorithms, (3) Design techniques for identification of the best set of initial features in the microstructure in particular applications, (4) Statistical tools for real-time selection of thermo-mechanical processing sequences to customize microstructures for achieving desired property distributions, and (5) Performing multi-scale finite element analysis for testing and experimentally verifying microstructure design solutions in complex engineering applications. The proposed multi-scale design framework can lead to significant reduction of computational overhead in materials design, thus allowing accelerated insertion of materials and materials processes. While these techniques and libraries will be implemented for polycrystalline FCC aluminum alloys, they will also impact microstructure-sensitive design of many other polycrystalline systems. Improved methods for materials and process design would produce far-reaching benefits to the materials industry and economy. The proposed developments are not specific to materials design; if successful, they could be applied to other complicated design problems in a variety of scientific fields. Dissemination of the algorithms and software tools will enable broad research communities to harness databases of individual length scales and to infer how structures and properties are linked. The problems addressed provide a unique and valuable opportunity for undergraduate and graduate students to work in a multidisciplinary environment that emphasizes the significant and growing roles of multiscale modeling and statistical learning in materials development and design.