OPMD microstructure image data version 1.0

This data is used in the article "Towards Inverse Microstructure-Centered Materials Design using Generative Phase-field Modeling
and Deep Variational AutoEncoders" for building microstructure-sensitive materials design frameworks.

Microstructure data provided here is generated using high-throughput phase-field modeling at Texas A&M University's Terra supercomputing facility. The dataset is called Open Phase-field Microstructure Dataset (OPMD) version 1.0 and it contains 20,000×40 or 80,000 computationally generated microstructure images that are obtained from 20,000 distinct phase-field simulations where 40 time sequences are saved from each simulation.

This dataset is split into two equally-sized sets called proposal samples and target samples for the UP study that are generated via sampling from different probability density functions.

Images are 1167x875 pixels with a Red-Green-Blue (RGB) color scheme, according to composition order. The extraction of QoI is performed using the original raw data. Images were further resized to 64×64 before feeding them into the encoder model. The microstructure dataset presents a diverse set of chemical spinodal morphologies.

The images uploaded here are a small subset of the original data that is used to build the uncertainty propagation method. There are 6040 microstructures images in the attached folder. This data is extracted from the final time step of the phase-field solver. We carried out 10,000 simulations and omitted microstructures that did not show a dual-phase structure. As a result, 6,040 microstructures have been preserved.

All of the raw data on which the results of the manuscript are based are publicly available via the OPMD website accessible via the https://microstructures.net link.