About this Abstract |
Meeting |
MS&T24: Materials Science & Technology
|
Symposium
|
Machine Learning and Simulations
|
Presentation Title |
Understanding Grain-Boundary Structure Using Strain Functional Descriptors and Unsupervised Machine Learning |
Author(s) |
Nithin Mathew |
On-Site Speaker (Planned) |
Nithin Mathew |
Abstract Scope |
Properties of grain boundaries (GBs) are controlled by the types and arrangement of local atomic environments at the boundary. Many methods have been proposed for characterization of GB structure using structural/polyhedral units, which have been applied to both minimum energy and metastable structures. We will present a method to characterize GB structure using a complete and symmetry-adapted set of atomic environment descriptors, namely the Strain functional descriptors (SFD), in conjunction with unsupervised machine learning methods such as Gaussian Mixture Models (GMM). Structural units are identified for GBs, reducing thousands (~5000) of metastable states to six different classes. Identified classes and the similarity metric are shown to be good descriptors for developing supervised machine learning models of GB properties and analyzing GB-dislocation interactions. |