Title:

Nuclear Reactor Modeling & Simulation: Harnessing AI and Computer Vision for Enhanced Efficiency and Future Applications

Speaker:

Dr. Forrest Shriver

Affiliation:

Sentinel Devices LLC

When:

Thursday, February 8, 2024 at 11:00:00 AM   Add to Calendar

Where:

Boggs Building, Room 3-47

Host:

Fan Zhang
fan.zhang@me.gatech.edu

Abstract

Modeling & simulation of nuclear reactors is at the core of nuclear power today, both in academia and industry. In both applications, millions of intermediate calculations are typically needed, such as iterative neutronics solves or non-viable core loading calculations. AI & ML techniques have the potential for outsized impact in reducing the time taken for these simulations, as they can be used as a “stand in” for these intermediate calculations, providing the same quality of answer for a fraction of the compute cost and time. Computer vision-based neural networks, in particular, show incredible emergent capabilities in both predicting reactor parameters at a high level of detail, and in being resilient against out-of-distribution data and novel inputs that have never been seen before. In this seminar, Dr. Shriver will be discussing his foundational work in developing these computer vision models, identified and resolved issues in scaling this work up to handle full-core reactor problems, and his findings around model robustness and model quality from an engineering perspective. He will also discuss potential future applications of these technologies and needed future work in bringing these technologies to the nuclear industry.

Biography

Dr. Forrest Shriver is the CEO of Sentinel Devices LLC and a DOE Innovation Crossroads Fellow. He has over six years of experience in applying AI & machine learning to nuclear power. He has worked with large-scale supercomputers for HPC projects, including the Titan and Summit supercomputers at ORNL, and developing cutting-edge technologies supporting the next generations of supercomputers. He has worked on developing and scaling up neural network-based reactor core simulators, replicating the results from large-scale supercomputing codes for less than 1% of the compute cost. He has also applied this technology to nuclear nonproliferation objectives, using AI to achieve results that would have been impossible to achieve using classical simulation approaches. Most recently, he is working with one of the largest suppliers in the nuclear industry on the application of AI to both traditional LWR and advanced reactor supply chain problems, identifying latent information from text and numerical records that is otherwise impossible to retrieve or analyze. He received the Defense Innovation Award in 2022 and Best Poster in Reactor Physics division in American Nuclear Society in 2020. He obtained his PhD in Nuclear Engineering from University of Florida in 2021.

Notes

Meet the speaker