An image-driven machine learning approach to kinetic modeling of a discontinuous precipitation reaction

Abstract

Microstructure quantification is an essential component of materials science studies, yet, there are no widely applicable, standard methodologies, for image data representation in complex microstructures. Recently, machine learning methods have demonstrated success in image recognition tasks across disciplines, including materials science. In this work, we develop an approach for microstructure quantification for the purpose of kinetic modeling of a discontinuous precipitation reaction. We develop our approach in a case study on a U-Mo alloy which experiences this phase transformation during sub-eutectoid annealing. Prediction of material processing history based on image data (classification), calculation of area fraction of phases present in the micrographs (segmentation), and kinetic modeling from segmentation results were performed as part of this study. Results indicate that features extracted using a convolutional neural network (CNN) represent microstructure image data well, and segmentation via k-means clustering agree well with manually annotated images. Classification accuracy of original and segmented images is both 94% for a 5-class classification problem. Kinetic modeling results are consistent with previously reported data that employed manual thresholding. The image quantification and kinetic modeling approach developed and presented here aims to reduce researcher bias introduced into the characterization process, and allows for efficiently leveraging information in limited, unbalanced image data sets.