Materials design studio is a collaborative class, taught by Dr. Raymundo Arroyave, in which lectures are given on the cutting edge of materials design, optimization, and uncertainty in experimentation. The class lecture mostly centers around Gaussian processes and other methods of modeling uncertainty. However, on occasion, we have fun lectures in which we discuss a topic such as grant writing, or networking.
The class puts you into a group and allows you the freedom to discuss and select a group project for yourself. Once you're in a group you can select a research topic. For the most part, people come in with some level of knowledge about what they want to research. However, some come in and are on the fence and fall into another group, which typically results in someone expanding their area of expertise. We had this occur with my group, where two of us were reasonable strong in machine learning and CNNs, and one was not. Over the course of the semester he learned quite a bit and now has some expertise he didn't have coming into the semester.
Our project has been centered around a few different pieces, one of which is featurizing images using CNNs to develop effective regressors. The other is developing GAN-based methodologies to generate high-quality microstructure images. I have been working the latter piece of the project, and so far the results have been promising. We have been able to generate a few promising images, but really there is a lot of work to be done.
Was able to get a progressively trained GAN working to generate SEM quality microstructures. I was able to do this using the 8-GPU DGX-style server at Army Research Lab. The results can be seen in the activities tab, under the Summer 2019 Internship.
One of the group members, Richard Couperthwaite, will be presenting the CNN featurization work at TMS in San Diego.
MSEN 619 is a course centered around phase field generation and FEM-based simulations. It uses a software suite called MOOSE (Multiphysics Object Oriented Simulation Environment) which contains modules capable of solving many facets of physical world modeling. Specifically, we have been working in phase field modeling of materials, although the course takes a look at numerical methods and other applications of physics based simulation.
Dr. Karim Ahmed has only offered this course a few times, and he is constantly growing and learning with us. We have enjoyed his expertise and willingness to help us learn. He is very available as a professor and is far more concerned about learning than exams and grades.
Over the course of the semester we have been able to work with researchers from Idaho National Lab. They have a substantial amount of expertise is modeling of fracture, especially of uranium dioxide (UO2) pellets - typically found in nuclear reactors. We have been collaborating weekly and been moving along nicely - although the model is proving quite complex.
We are nearing the completion of the semester now, and preparing our final report and presentation. We intend on continuing this work into the summer. Although my interests and background lie more in machine learning than computational materials, I have rather enjoyed this experience and I think it will set me up well for some future research in the physics based machine learning space.
Admittedly, I've found myself researching thermodynamics, kinetics, and classical numerical methods more and more. I've grown quite fond of the material, and hopefully the interdisciplinary lens will bring a new light to some algorithmic methodologies in my research in the next few years.