Joint project SEMOTI

Motivation and goals

By monitoring the deep geological repository system during the individual project phases, the level of knowledge and data bases are constantly increasing, so that the measurement data obtained as part of a self-learning process offer the possibility of recognizing undesirable developments and, if necessary, deriving consequences from them. However, the potential of AI methods to improve the prediction of scientific and technological aspects in site selection as well as in the construction and (post-)operational phase of a deep geological repository has not been systematically investigated yet (Krafczyk et al. 2021).

The SEMOTI research project aims to create the foundations for the essential geomechanical and geotechnical processes of a machine-learning based modelling methodology enabling an optimization process during the planning phase and a calibration process during the excavation phase for an emplacement drift in rock salt, which is one of the potential host rocks in Germany considered for secure long-term nuclear waste storage due to their distinctive mechanical and hydraulic properties (StandAG 2017). A virtual demonstrator of an emplacement drift is used for this purpose. This work introduces a surrogate modeling approach based on Gaussian Processes (GPs), which are machine learning tools for the regression of unstructured data that provide a quantification of uncertainty (Williams & Rasmussen 2006).