|About this Abstract
||2018 TMS Annual Meeting & Exhibition
||Computational Materials Discovery and Optimization
||Predicting Ferroelectric Properties from Microstructures with Deep Learning
||Isaac Curtis, Vishnu Boddeti, Samrat Choudhury
|On-Site Speaker (Planned)
Microstructure is known to ultimately determine the physical properties of materials. The typical Edisonian trial-and-error approach to determine the optimal microstructure for targeted properties can be a non-trivial and costly process. Using a combination of phase-field simulation, modern computer vision and machine learning tools, we present an alternative pathway to determine the processing-microstructure-property relationship. We will focus on the processing parameters of bulk ferroelectrics to obtain the desired domain structures for a targeted dielectric and/or ferroelectric property. We will show that, despite the small quantity of experimental microstructures available, hundreds of thousands of phase-field generated microstructures can be an effective alternate approach to successfully train a deep neural network, a critical step in determining the linkage between processing parameters, domain structure and electrical properties. Finally, we will demonstrate the universality of our approach to design microstructures beyond ferroelectric materials.
||Planned: Supplemental Proceedings volume