|About this Abstract
||MS&T22: Materials Science & Technology
||Additive Manufacturing Modeling, Simulation, and Machine Learning: Microstructure, Mechanics, and Process
||A-9: Reducing the Order of a Kinetic Monte Carlo Potts Solidification Model with Machine Learning
||Gregory D. Wong, Anthony D Rollett, Gregory S Rohrer
|On-Site Speaker (Planned)
||Gregory D. Wong
Current microstructure and grain orientation models are expensive to run, and this work seeks to reduce their order using machine learning. Categorical generative adversarial networks (cGAN) and variational autoencoders (VAE) are examined as methods to reduce the order of a Monte Carlo Potts Solidification model. This work is a proof of concept for use with larger models for metal additive manufacturing. A cGAN is a network of convolutional neural networks (CNNs) that can generate an image that the computer has classified as being from a specific labeled group. A VAE, another network of CNNs, correlates image statistics with labels to produce images with specific labels. Both models have been trained to produce images with varied nuclei count. These methods will allow for printing parameters to be fed into the model to create a specific microstructure in future work. Training and output images along with model structure will be presented.