In the southeastern region of France, a team of 2,000 physicists, scientists, and workers are busy building the world’s largest nuclear fusion plant, the International Thermonuclear Experimental Reactor or ITER.
The aim is to bring the facility online by 2033, and a host of artificial intelligence (AI) tools, in particular Microsoft, are being deployed to achieve the high precision work being done at an unprecedented scale.
Fusion energy aims to replicate the conditions of the Sun to fuse hydrogen isotopes, deuterium, and tritium and form helium while releasing large amounts of energy. Much of the progress in this direction has been made in a reactor vessel called the tokamak. In this donut-shaped vessel, fusion fuel is heated to a hot plasma state where temperatures cross ten times that on the core of the Sun and make nuclear fusion possible.
At ITER, researchers are working to build the world’s largest such reactor vessel to study the nuclear fusion process in greater detail. However, achieving this is a mammoth task considering the scale of the build and the large number of components that are going into making this facility. This is where AI can help.
High precision build
The tokamak at ITER will be assembled using nine sectors. Five sectors are being built in Europe, four in South Korea, while Russia and India will supply some components.
Made from a special kind of stainless steel, these components must be machined and welded together. A series of sophisticated ultrasound scans ensure the welds are without defects.
This generates large amounts of data that need to be analyzed. So, researchers have built an AI model in Microsoft Visual Studio Code, where multiple coding languages can be used to probe the data and save hours of work needed to check the quality and precision of the welds.
The team has also been able to use the model to determine the materials that will be used to line the reactor vessel from the inside.
The collaborative effort of ITER spans more than 30 countries. It demands a standardized document management system that has been in place for more than 20 years. More than 1.5 million documents have been created during this time, and searching for information is like finding ’a needle in a haystack’.
In Q1 2024, the IT Project Tools section at ITER built a Proof of Concept (PoC) AI chatbot to do this job and iteratively improved it as the year progressed. Earlier this year, the research team moved from standalone architecture to a multi-agent approach.
In this approach, the AI tool summarizes the document and indexes the information in a large vector database. The advantage of doing this is that users can send queries to the database and retrieve meaningful responses instead of getting responses matched by keywords.
Powered by tools from OpenAI, the chatbot supports questions in Mandarin, Korean, Japanese, Russian, and Hindi and is accessible to 120 partners of the ITER project. Since ITER documentation also contains many acronyms, the tools team built another chatbot that answers questions specifically about this.
The team is also working on making the document system interfaceable with other large language models that teams from other regions in the world might be using, enabling easier flow of information and reducing friction in achieving the ultimate goal of successfully achieving nuclear fusion inside ITER.
