The National Spherical Torus Experiment Upgrade (NSTX-U) at the U.S. Department of Energy’s Princeton Plasma Physics Laboratory (PPPL) is investigating the spherical tokamak concept as an attractive path towards developing fusion energy.  Recently, new Plasma Facing Component (PFC) tile designs were completed in order to accommodate higher heat fluxes, longer pulse lengths, and increased halo current forces compared to the previous generation machine.  These PFCs feature a castellated design with embedded thermocouples which allow shot-integrated energy to be measured. 

As a means of reconstructing the heat flux incident upon the tile surface from subsurface thermocouple data, an artificial neural network learned the relationship by studying simulated examples.  More specifically, a convolutional neural network (CNN) was developed that reconstructs the heat flux profile incident upon the divertor tiles from time varying subsurface thermocouple data and a few 0-D machine parameters.  The proof of concept CNN was trained on thermocouple data generated by approximated NSTX-U heat loads applied to real PFC designs in ANSYS.  Once trained, the CNN is capable of high precision reconstruction for heat flux profiles expected in NSTX-U.  Additionally, to test the system’s ability to cope with noise and systematic error, pseudo-noise was injected into the simulated data. The CNN can predict the incident heat flux despite this noise and error.