My senior capstone project is on GPU-based modeling of material degradation inside of a nuclear reactor. My team coined the product name GPU-Parallelized Irradiation Environment Simulation (G-PIES). G-PIES has a working prototype interleaved with the following resources:
Kohnert A. et al. Modeling Microstructural Evolution in Irradiated Materials with Cluster Dynamics Methods: A Review
Sakaguchi N. et al. HETEROGENEOUS DISLOCATION FORMATION AND SOLUTE REDISTRIBUTION NEAR GRAIN BOUNDARIES IN AUSTENITIC STAINLESS STEEL UNDER ELECTRON IRRADIATION
Pokor C. et al. Irradiation damage in 304 and 316 stainless steels:experimental investigation and modeling.Part I: Evolution of the microstructure
Was G. Fundamentals of Radiation Materials Science
From Kohnert’s review, we determined that Cluster Dynamics and OKMC are the most feasible, untouched territories of materials science that have yet to be exploited with GPU parallelism; acknowledging the renowned molecular dynamics GPU/CPU-parallelized simulator LAMMPS.
Our Cluster Dynamics simulation currently models the clustering of interstitial and vacancy defects, in with 1 spatial dimension. We achieve this with two 2D arrays, representing the concentration of interstitials and vacancies in a given cluster per cubic centimeter at a given space slice. This arbitrary spatial division is to merely model the relationship defects can have on their close neighbors, which is fairly intuitive; if a crack appears in a material, it will likely grow larger overtime.
Our end goal is to provide a GPU-parallelized cluster dynamics simulation that is flexible and powerful for the user. With the understanding that our users will likely be researchers extending from the agency of our mentor, Dr. Tianyi Chen at Oregon State University. By Summer 2024, we will have a product offered as a Command Line Interface, Graphical User Interface, and a C++ API (possibly python too), all which make use of the GPU through CUDA (Nvidia) and Metal (MacOS) kernels.
G-PIES will empower researchers in materials science to closely examine defect generation under significantly larger time scales with optimizations that range up to 10,000x in performance (CUDA/Nvidia). Researchers will be able to fully parameterize the simulation and even fine-tune the inner-calculations of the cluster dynamics modeling in the engine (C++ API & Command Line Interface). I personally hope that G-PIES exemplifies the potential of GPU computation and how this optimization can help us achieve more accurate estimates in our modeling of the physical world. Go G-PIES!!!
