About this Abstract |
Meeting |
TMS Specialty Congress 2024
|
Symposium
|
2nd World Congress on Artificial Intelligence in Materials & Manufacturing (AIM 2024)
|
Presentation Title |
An Advanced Machine Learning Approach for Identification of Grain Boundaries in Atomistic Simulation Data |
Author(s) |
Saksham Singh, Akash Gupta, Sumit Kumar Maurya, Surya Ardham |
On-Site Speaker (Planned) |
Surya Ardham |
Abstract Scope |
Atomistic simulations are considered a useful tool to study the defects in material. However, the analysis of results is always a tedious and time-consuming process. We present a deep learning-based approach for automated analysis of atomistic simulation data. The identification of grain boundaries (GBs) in molecular dynamics (MD) simulations of aluminum is chosen as test case. The atomic configurations representing GB are translated into a 3-dimensional array. The length and size of each element in the array are carefully optimized to create a dense representation of each atom's neighborhood while minimizing the memory and computational demands. Two variants of 3-D Convolutional Neural Networks (CNN) are trained on this dataset. The final model shows a test accuracy of more than 95% by using only the atomic coordinates as input feature. The model is further validated on an MD trajectory featuring two GBs in an aluminum supercell, showcasing its automated analysis capability. |
Proceedings Inclusion? |
Definite: Other |