Machine Learning for Predicting Properties of Silicon Carbide Grain Boundaries
Technical Report,01 May 2019,30 Sep 2019
US Army Combat Capabilities and Development Command, Army Research Laboratory Aberdeen Proving Ground United States
Pagination or Media Count:
Statistical techniques are utilized to determine the efficacy of physics-based descriptors to predict the energetic properties of silicon carbide grain boundaries. These descriptors are utilized in kernel ridge regression models with a radial basis function kernel for the prediction of grain boundary energetics. Models derived from this approach have been implemented as a replacement to the insertion, removal, and replacement probability functions in a Monte Carlo based selection scheme for sampling the microscopic degrees of freedom in silicon carbide grain boundaries. Preliminary results show these models increase the overall computational efficiency of finding low-energy minimized states compared to current techniques.
- Inorganic Chemistry
- Properties of Metals and Alloys
- Ceramics, Refractories and Glass