A digital twin for final disposal TU Braunschweig research project on the deep disposal of highly radioactive waste

The last three German nuclear power plants have been shut down. What remains is around 27,000 cubic metres of highly radioactive waste. Where this will be safely stored in the future is still open. Originally, a suitable site was to be found in Germany by 2031. In the meantime, a decision is being sought in the 2040s. Researchers at Technische Universität Braunschweig are now investigating in their SEMOTI project how artificial intelligence methods can support the evaluation of possible deep geological disposal sites. The research project is funded by the Federal Ministry for the Environment, Nature Conservation, Nuclear Safety and Consumer Protection with around one million euros.

The basis for the search for a final disposal is the Standortauswahlgesetz (Site Selection Act). It stipulates that a site suitable for licensing must be found with the best possible safety. In concrete terms, this means that it must be possible to safely store radioactive waste for a period of one million years. Only deep underground disposal, so-called deep geological disposal in a host rock such as rock salt, clay or crystalline rock, can be considered.

Since 2017, the Federal Disposal Society has been examining various regions throughout Germany for their suitability for final disposal. In the process, “final disposal with reversibility” is favoured: The entire process – from site selection to planning, construction and operation of the final disposal site to closure of the mine workings – is to be designed in such a way that decisions that have already been made can be reversed.

The aim is therefore to learn continuously within the process, and to do so in a self-learning or self-correcting manner as far as possible. This is where TU Braunschweig’s research project comes in. With the project “Entwicklung einer selbstlernenden Modellierungsmethodik zu geomechanischen und geotechnischen Prozessen am Beispiel der Planungs- und Auffahrungsphase einer Einlagerungsstrecke eines Tiefenlagers“ (Development of a self-learning modelling methodology for geomechanical and geotechnical processes using the example of the planning and excavation phase of a deep geological disposal) (SEMOTI), the researchers want to contribute to the improvement of planning and evaluation.

Machine Learning in Rock Mechanics

“Using the example of an emplacement section, we want to determine whether we can also apply Machine Learning to processes in rock mechanics,” says Professor Joachim Stahlmann from the Institute of Geomechanics and Geotechnics at TU Braunschweig. “Our ambition is to achieve better results with artificial intelligence methods in the field of geotechnics.”

Through monitoring during the individual project phases, the knowledge and the data basis about the deep geological disposal system are constantly increasing. In the self-learning process, undesirable developments can be recognised more quickly and, if necessary, consequences can be derived from them. For example, whether the deep geological disposal can be converted into a final disposal, whether additional measures are required or even whether the stored waste has to be retrieved.

“This can create a tool that can be used for further decision-making,” says Professor Henning Wessels from the Institute for Computational Modeling in Civil Engineering. “AI can also go some way to reducing the subjectivity in the verification process by engineers or scientists“.

The digital twin of a mine

Since there is still no final disposal for highly radioactive waste in Germany, the scientists are developing a fictitious final disposal model, the virtual demonstrator, for their research work. They are looking at the digital twin of an emplacement section of a deep geological disposal in rock salt.

“With the current state of the art, however, we cannot expect the learning processes to be fully automated,” says Professor Stahlmann. “The expertise of engineers and scientists will continue to be used in the development of a deep geological disposal.” But AI methods could help improve both the planning and the assessment of reactions to changes in the state of the final disposal system. In addition, machine learning could produce unexpected solutions that are directly ruled out in manual planning.

Project data

The research project “Entwicklung einer selbstlernenden Modellierungsmethodik zu geomechanischen und geotechnischen Prozessen am Beispiel der Planungs- und Auffahrungsphase einer Einlagerungsstrecke eines Tiefenlagers“ (Development of a self-learning modelling methodology for geomechanical and geotechnical processes using the example of the planning and excavation phase of a deep geological disposal) (SEMOTI) is funded by the Federal Ministry for the Environment, Nature Conservation, Nuclear Safety and Consumer Protection (BMUV) with 997,000 euros over a period of three years, starting in May 2023. The Institute of Geomechanics and Geotechnics (Professor Joachim Stahlmann), the working group “Data-driven Modeling and Simulation of Mechanical Systems“ (Professor Henning Wessels) and the Institute of Dynamics and Vibrations (Professor Ulrich Römer) are involved in the project. The researchers are looking forward to further reinforcements in the team.