Data Pipeline Development for Grain Boundary Structures Classification

Abstract

Grain Boundaries govern many properties of polycrystalline materials, including the vast majority of engineering materials. Evolutionary algorithm can be applied to predict the grain boundary structures in different systems. However, the recognition and classification of thousands of predicted structures is a very challenging work for eye detection in terms of efficiency and accuracy. A data pipeline is developed to accelerate the classification and recognition of grain boundary structures predicted by Evolutionary Algorithm. The data pipeline has three main components including feature engineering of grain boundary structures, density-based clustering analysis and parallel K-Means clustering analysis. With this data pipeline, we could automate the structure analysis and develop better structural and physical understanding of grain boundaries.