Text-generating AI, which you might usually think of for tasks like answering questions, is making its way into nuclear science. Zavier Ndum, a graduate student in nuclear engineering, is on a mission to see how large-language models (LLMs) can change the game in nuclear research. These LLMs, like ChatGPT, are known for generating text from vast data sources. While they’re often used in software development, their role in nuclear research is still emerging.
One big challenge is the sensitive nature of the data in nuclear science. “There’s a lot of sensitive data in nuclear science, and it’s crucial to ensure security when handling it,” Ndum explained. “Handing over this knowledge to general AI models like ChatGPT isn’t feasible. But using AI to automate workflows within organizations can boost productivity and efficiency.” In his paper, Ndum introduces AutoFLUKA, an AI framework that automates tasks like running computer simulations with FLUKA software. This tool processes input files, executes simulations, and analyzes results, providing graphs for visualization. Researchers can securely input their data into AutoFLUKA, ensuring it remains confined to their systems.
Ndum faced hurdles, like not having access to the regulated Monte Carlo N-Particle (MCNP) simulation code. But with FLUKA’s similarities to MCNP, he adapted his model for other programs. Transitioning to AI research from health physics was another challenge for Ndum, who previously focused on radiation dose studies. At Texas A&M, he worked with Dr. John Ford and Dr. Yang Liu to explore AI applications in nuclear science. “Venturing into new areas is tough, but perseverance leads to valuable outcomes,” Ndum remarked.
Ndum, president of the State of Texas Chapter of the Health Physics Society, extends this AI approach to health physics. At a recent conference, he showed how LLMs could act as virtual assistants, significantly reducing the time needed to retrieve information from regulatory documents, which is a big help for radiation safety officers (RSOs). “Sorting through RSO guides can be tedious, but AI can simplify the process,” Ndum noted.
As part of Texas A&M’s AI initiative, Ndum shared his findings and plans to discuss these developments at upcoming conferences. His ongoing work involves creating an advanced LLM application for answering complex questions in nuclear science. This tool can process various file formats, offering comprehensive support for research tasks. “Exploring AI’s potential in nuclear science is essential,” Ndum said. “I am committed to discovering what we can achieve.” Dr. Liu highlights the significance of Ndum’s work, stating, “Zavier’s integration of AI in nuclear research exemplifies the forward-thinking needed in this field. Secure, domain-specific automation is transformative, and his contributions are setting the stage for more efficient innovations in reactor modeling and nuclear safety.”
