Students

Aidan Furlong

Summary

This research work focuses on using neural networks to predict Pressurized Water Reactor (PWR) neutronics features generated by the OpenMC modeling package, an open-source Monte Carlo code. The network is trained using various thermal hydraulic and neutronics parameters such as fuel, cladding, and moderator temperatures to predict pin powers in high fidelity for a reflective 17×17 fuel assembly. This research builds on previous work, which used the export controlled MPACT package to accomplish similar training goals. The networks trained with open-source datasets performed very similarly to those trained with MPACT, exhibiting many of the same features albeit with modestly worse losses. This paper successfully replicates the previous findings and indicates the potential use of neural networks as surrogate models aimed at reducing the large computational costs needed for conventional simulations