Physics-Informed Neural Networks for Neutronic Heating

Developing components for fusion reactors poses a complex challenge due to the need to consider a large design space and tight interdependence between different systems. Fast prediction of specific physics fields is crucial for thoroughly exploring this space, ensuring candidate designs align with the specified requirements. Traditional numerical methods fall short of meeting the demand for rapid predictions and probably offer an unnecessary level of accuracy for exploratory work. Hence, there is justification for adopting a surrogate modelling strategy to fulfill this purpose.



In this presentation, we will describe a case study that explores the use of Physics Informed Neural Networks (PINNs) to estimate the temperature field in a spherical tokamak caused by neutronics heating. The work uses OpenMC to compute the neutronic heating, one of the inputs into the PINN. The standard governing equations for heat transfer and the associated thermal boundary conditions are used for training the PINN. Validation using standard finite element simulation demonstrates alignment between the results of finite element simulations and PINNs, highlighting the predictive capability of PINNs in neutronics heating thermal analysis. The pre-trained surrogate model shows the capability of making predictions much more quickly than the finite element method.



The results demonstrate the substantial potential of PINNs to speed up simulations for engineers involved in simulation work and to accelerate the design process of components subjected to neutron interactions.