© iStock/BlackJack3D
Prozesstechnik-Konzept für künstliche neuronale Netze

AI technologies for more efficient dismantling of nuclear power plants

With the shutdown of the last German nuclear power plants, the decommissioning of these installations is now at the centre of attention. However, the impending shortage of skilled labour poses major challenges for the industry. Experts from GRS are therefore working with their partners in a research project that has just been launched to train specialists more efficiently in the future and to promote young talent in a more targeted manner: A hybrid learning platform with VR and AR elements and an innovative AI application are intended to optimise dismantling processes and prepare the industry for the future. In addition, a digital platform is being developed to make licensing processes more transparent and efficient.

The last three nuclear power plants (NPPs) in Germany were shut down on 15 April 2023. Like the German power, prototype and research reactors that were previously shut down, they will now be decommissioned. According to the Federal Office for the Safety of Nuclear Waste Management (Bundesamt für die Sicherheit der nuklearen Entsorgung), 36 plants are currently in various stages of decommissioning; although three NPPs and three research reactors have been shut down, a decommissioning licence has not yet been issued or used; three further NPPs and 31 research reactors have already been completely dismantled and released from the scope of the Atomic Energy Act - so there is more than enough experience with the decommissioning of nuclear installations in Germany.

Dismantling will take several decades yet

As the dismantling of all power, prototype and research reactors is set to take several decades, it is important to identify potential for optimisation and make profitable use of new technologies that can accelerate or simplify processes. This is the starting point for a research project funded by the Federal Ministry of Education and Research (BMBF) in which colleagues from various GRS departments have been contributing their expertise since the beginning of May together with their partners actimondo eG (coordinator), Dornier Nuclear Services, Advanced Nuclear Fuels, Forschungszentrum Jülich, FIR e. V. at RWTH Aachen and the Chair of Educational Technology at TU Dresden.

Over the next 36 months, they want to improve knowledge and project management in the area of dismantling and licensing in three work packages and adapt process management to current conditions with practical solutions. The background to these considerations is primarily the impending shortage of young professionals and skilled labour, which is already becoming increasingly apparent. Recognised as a problematic factor in other sectors of the economy, it is particularly affecting the nuclear industry in Germany, not least because of the phase-out of this technology. This is despite the fact that, as mentioned above, decommissioning projects will continue for decades to come. Accordingly, it makes sense to make the wealth of knowledge that exists today available and easily accessible to the next generation of decommissioning specialists.

Hybrid learning platform with VR and AR elements

The first work package is primarily concerned with providing better and more efficient training for lateral entrants and newcomers - across the entire industry, i.e. for companies, authorities and other organisations. To this end, a hybrid learning platform is being developed that combines face-to-face events with modern learning formats such as e-learning videos, podcasts, and animated films. Training concepts with virtual and augmented reality (VR and AR) will also be utilised. The centralised learning platform is intended to bundle the various educational options that exist in the field of decommissioning nuclear installations and make them more readily available.

AI application

In a further work package, the experts are developing an AI application that can be used to search the wealth of knowledge, i.e. the national and in some cases international specialist literature on the subject of dismantling, in a targeted manner and make it more easily accessible. This AI can be imagined as a Chat GPT specialising in the dismantling of nuclear installations. The basis of this AI will be a so-called Large Language Model (LLM), as is also incorporated in the major well-known commercial chatbots – not only in Chat GPT, but most notably also in Microsoft's Co-Pilot or Google's Gemini. Put simply, LLMs are computational linguistic probability models that are trained using a huge pool of text documents (LLMs can now sometimes also process images, graphics and the like) in order to be able to provide answers to a question on a statistical basis that match the documents with which they were trained as closely as possible.

In contrast to the aforementioned, the AI that is to be programmed will not use the entire internet as a knowledge base, but rather the specialist literature mentioned above; on the other hand, this knowledge base will not be used to train an LLM, as is the case with the chatbots mentioned; instead, the specialist literature will be prepared by GRS so that it can be processed by the LLM on which the AI is based. To do this, the researchers use various techniques, such as Retrieval Augmented Generation. This involves providing an existing LLM, which has been trained on the basis of any large data set, with new external data to take into account - in our case, the specialist literature on dismantling.

This enables the LLM to extract information on any queries from this external data. Another challenge is to update the external data in such a way that the LLM can access the latest data without having to rebuild the entire system - after all, the answers should correspond to the current state of research. In order to process the data accordingly, the researchers want to set up so-called data pipelines. Such data pipelines function like a kind of pipeline system for raw data that flows from various data sources into a common data pool, where it is analysed and curated. This approach is intended to ensure that the results are up-to-date, correct, clear and comprehensible.

The aim of the work package is to develop a corresponding AI application at GRS in order to be able to reliably access non-public data, i.e. the specialist literature mentioned above. Such an AI system would make work considerably easier for experts from industry, authorities and research. The technology could then also be adapted to other specialist areas, such as reactor safety or radioactive waste disposal.

Digital platform for licensing processes

In a third work package, a new digital platform for licensing management is being developed. The aim is to make licensing processes for decommissioning activities more transparent and efficient in the future. With the increasing digitalisation of individual processes, ensuring data security is also not insignificant, for example to guarantee the revision-proof transmission of procedural documents.

The technical basis for the new platform will be blockchain technology, which has become familiar to a wider audience in recent years, particularly through its use in various cryptocurrencies. With the help of the new system, industry experts, for example, will be able to track the exact processing status of their respective dismantling applications. Ideally, this licensing platform should also be applied to other areas at a later date.

The research project will run until June 2027. The project receives funding from the „Forka – Forschung für den Rückbau kerntechnischer Anlagen“ (Research for the Dismantling of Nuclear Installations) funding concept.