Adding image diagnostics to the prediction of future ICF experiments at NIF

Recently, a new machine learning workflow has been developed to predict the outcome of the future indirect drive ICF experiments at the National Ignition Facility. Given the design parameters of an upcoming experiment, a 2-D integrated hohlraum-capsule simulation is first run to make an initial prediction of multiple diagnostic measurements. Then, a machine learning model improves this prediction using a database of the past experiments. In this presentation, we extend this workflow to match not only scalar but also image data. We will discuss technical challenges triggered by including image data and compare the new, image-informed predictions with the scalar-only predictions and with the real experimental data. We will investigate how matching more types of diagnostics helps break degeneracies in the model and reduce error bars. Ultimately, an improved model will be used in the exploration of new ICF designs.

LLNL-ABS-824173. Prepared by LLNL under Contract DE- AC52-07NA27344.