Wednesday  |  Monongahela  |  09:20 AM–09:40 AM

#13816, Machine Learning Tools for the Synthesis of Targeted Microstructure Volumes

Across experimental mechanics and materials characterization, many situations of interest arise from the interactions between local material structure and loading conditions. Understanding the details of microstructure/deformation interactions is challenging, as material structures are often complex, and the full 3D volumetric characterization of materials during deformation is rarely straightforward. For these reasons, it is common to rely on experimental repetition to reinforce results, but this can be prohibitively costly in terms of time and/or capital. Here, we present the use of machine learning and network-based approaches to address this challenge for crystalline materials, where we use deep learning networks to synthesize 3D microstructural volumes based on example microstructures gathered from experiments using EBSD. These examples are used as a guide to dictate the general structural appearance seen in the synthetic microstructures, such that newly generated microstructures have similar local features to experiment without being directly identical. These synthetic microstructures can be used for high throughput simulations, enabling a significant leap forward in our ability to map the relationships between material structure and their resultant properties.

Neal Brodnik   University of California Santa Barbara
Devendra Jangid   University of California Santa Barbara
McLean Echlin   University of California Santa Barbara
Bangalore Manjunath   University of California Santa Barbara
Tresa Pollock   University of California Santa Barbara
Samantha Daly   University of California Santa Barbara