Designing optimised experiments for the international fusion materials irradiation facility

The development of fusion power requires a facility for assessing the behaviour of materials subjected to damage from 14 MeV neutrons with displacement damage levels of up to about 150 displacements per atom (dpa). The proposed International Fusion Materials Irradiation Facility (IFMIF) will enable experiments to be conducted that closely fit these requirements. When designing experiments it is at first sight natural to suggest a uniformly sampled test matrix over the domain of interest. However, as the irradiation volume in IFMIF will be limited, it is appropriate to consider how the set of experiments may be optimised. In the present work we suggest a set of experiments designed using predictive Bayesian neural network models created using published data for irradiated reduced-activation martensitic steels. It has been possible to identify gaps in knowledge on the basis of model-perceived uncertainties, and to assess trends in order to determine more optimal experimental sampling.