![Electronic Theses and Dissertations](../../assets/md5images/8bf39a2f649bfcc97f3e39e2323994d5.png)
Date of Award
1-1-2023
Document Type
Thesis
Degree Name
M.S. in Engineering Science
First Advisor
Samrat Choudhury
Second Advisor
Dave Harrison
Third Advisor
Yiwei Han
Relational Format
dissertation/thesis
Abstract
Modeling the microstructure evolution and linking them to materials properties is critical in establishing processing-microstructure-property linkage in materials. In this project, the evolution of microstructure of Fe-C alloys is modeled using phase field simulations to understand the behavior of these alloys under various cooling temperatures and initial carbon concentrations. However, these simulations require significant computational resources. Therefore, in this thesis, a convolutional neural network (CNN) and gated recurrent unit (GRU) are coupled with phase field simulations to predict the solidification of Fe-C alloys. In addition to the data-driven approach, the solid-liquid interfacial energy is incorporated during the training process, giving the model additional information regarding its performance, forcing it to learn more accurately using a smaller dataset. Additionally, this thesis provides insight regarding how to improve the performance of a neural network for a microstructure evolution dataset by adjusting the dataset size, model complexity, and loss function. Overall, this thesis provides insight for machine learning tailored for applications in materials science, including a physics-informed approach that can improve model performance when a dataset is sparse.
Recommended Citation
Rhoads, Benjamin Mullins, "Predicting Microstructure Evolution in Alloys Using Deep Learning" (2023). Electronic Theses and Dissertations. 2768.
https://egrove.olemiss.edu/etd/2768