Microstructure representations

applied computer vision methods for microstructure characterization

Recent advances in computing power and automated microstructural image acquisition have opened the doors to data-driven quantitative microstructure analysis. Extraction of salient microstructure features is a crucial enabling component in this rapidly developing field of research; in the past decade the computer vision community has made enormous progress in this area, much of which has gone relatively unexplored by the quantitative microstructure analysis community. My research explores applications of image texture recognition algorithms to engineer efficiently computable generic microstructure descriptors, enabling quantitative microstructure comparisons between and across a wide variety of materials systems. Novel materials science applications include characterization and qualification of powder materials, exploratory analysis of large microstructure datasets, and extraction of quantitative relationships between materials processing and properties metadata and microstructural image features. The fusion of microstructure image analysis and contemporary machine vision techniques will facilitate development of robust autonomous microscopy systems, and may support quantitative engineering standards for complex hierarchical microstructure systems.
A computer vision approach for automated analysis and classification of microstructural image data
In Computational Materials Science (2015) Bag of words representation for microstructure

Straightforward application of the bag of visual words to microstructure images.

Characterizing powder materials using keypoint-based computer vision methods
In Computational Materials Science (2016) Synthetic SEM powder micrographs with Blender.

Exploring a synthetic powder materials dataset, created with Blender. Check out the Blender scripts and the dataset (writeup).

Computer vision and machine learning for autonomous characterization of AM powder feedstocks
In JOM (2017) Schematic of VLAD encoding for powder micrographs.

Exploring real powder micrographs.

Exploring the microstructure manifold: image texture representations applied to ultrahigh carbon steel microstructures
Currently under review t-SNE embedding of CNN codes for ultrahigh carbon steel micrographs.

Exploring ultrahigh carbon steel microstructures with SIFT, convolutional neural networks, and t-SNE.

Monte Carlo simulations of abnormal grain growth

Phenomenology of abnormal grain growth in systems with non-uniform grain boundary mobility
In Metallurgical and Materials Transactions A (2015) initiation of abnormal grain growth

The goal of this project is to estimate the rate of occurrence of abnormal grain growth in microstructures with various crystallographic textures, and to identify local microstructural features and conditions associated with the development of abnormal grains. Deeper understanding of the factors responsible for abnormal grain growth could suggest strategies for controlling abnormal grain growth in nanocrystalline metals, widening their applicability in extremely tough, wear-resistant coatings.

My main research tool is the Monte Carlo Potts model for grain growth as implemented in the SPPARKS software package from Sandia.