Speaker
Description
In support of the LIBRTI programme, we have developed a multi-physics simulation tool for generic breeder systems, initially focused on liquid lithium technologies. A key objective of the work is the ability to accelerate the optimisation of a breeder system design with many free parameters (related to geometry, chemical compositions or operational conditions), and in the presence of various competing requirements. These requirements include tritium generation, heat extraction targets and corrosion resistance. Ultimately, this tool is planned to undergo further developments to serve as a digital twin to support operation, maintenance and decommissioning of the breeder system, so has been built to be adaptable. It can integrate with various software packages and security considered design has been utilised to support the ultimate goal of supporting an operational plant. The implemented simulations include neutronics calculations based on Monte Carlo methods coupled with nuclear activation calculations as well as analytical models. Machine Learning based surrogate models have been adopted for the neutronics/activation calculations. Trained with simulated data, these Machine Learning models are integrated into the simulation of the entire liquid lithium system and enable the multi-targeted optimisation of the system. In addition, we are exploring Machine Learning models in support of the simulation of corrosion effects. The Machine Learning models predict, along with the response values, the uncertainties on these values. This enables the targeted provision of additional training data to reduce uncertainties in regions of the parameter space where it is of most interest.
| Speaker affiliation | Amentum |
|---|
