Digital Twin and Surrogate Model for Long-Term Geochemical Processes in Nuclear Waste Management

Abstract This study presents an advance in predicting long-term stability in waste containment systems. Employing a surrogate model based on the 25 GEMS Python scripts, we simulate geochemical interactions within waste degradation processes over 100 years. The model simplifies complex full-scale geochemical models using a mixing tank approach, primarily examining the evolution of material properties influenced by uncertain surface characteristics and reaction kinetics. 1 million cases are generated using the neural network-based surrogate model, drastically reducing computational time (∼1.9 seconds) compared to traditional methods (∼78.4 days). The model evaluates the deterioration mechanisms of various materials like iron, aluminum, zinc, and brass, in cementitious waste packages, crucial for assessing their impact on the integrity of waste containment over extended periods. Our findings with the neural network-based surrogate model, including ion concentrations and mass changes in materials like iron, brass, aluminum, and copper, offer detailed insights into chemical changes in the system. Incorporating a sensitivity analysis with 1 million cases generated by the surrogate model, the study underscores the interplay between chemical reactions and material properties, establishing a digital twin that links reaction rates to the stability of nuclear waste repositories. This study presents key indicators of potential integrity threats due to material expansion, contraction, or gas-induced pressure variations.