Dual AI Models Predict Nuclear Waste Repository Safety with New Accuracy
Researchers combined physics-based and machine learning models to forecast temperature changes in deep underground nuclear waste storage, achieving reliable predictions critical for repository design and long-term safety monitoring. The hybrid approach could accelerate licensing approvals and reduce operational risks for waste management operators worldwide.
Originaltitel: Physics-based and data-driven digital twins for temperature evolution in Full-scale Emplacement experiment
This study details the creation and analysis of physics-based and data-driven digital twins (DT) for heat transport in the Full-scale Emplacement (FE) experiment at Mont Terri Underground Research Laboratory, aimed at using continuously incoming temperature and humidity sensor data from a long-term heater experiment for a physics-based modelling digital twin (PBM-DT) and data-driven modelling digital twin (DDM-DT) to predict further temperature evolution in the near-field and highlighting their respective strengths. The physics-based modelling (PBM) allows mechanistic insights into thermal processes whereas the data-driven modelling (DDM) shows easy adaptability and high efficiency in temperature prediction. The predicted temperature evolution shows that both models can be used for repository monitoring and safety analysis in the early repository phase, underlining the potential of PBM-DT and DDM-DT approaches to support optimized repository design and safety. This research argues for the combined application of these methods to refine DT technologies in nuclear waste management.