Using Deep Learning to Investigate Performance and Variability in Inertial Confinement Fusion Experiments

On August 8^th, 2021 an inertial confinement fusion (ICF) experiment (N210808) at the National Ignition Facility (NIF) achieved a record-breaking fusion yield of 1.35 Megajoules. Follow up shots, investigating the sensitivity of N210808 implosion performance to engineering and design variability, demonstrated the importance of designing robust experiments.

Here we present our findings using physically constrained deep learning models with experimental data, recorded from ICF experiments at the NIF, to predict multiple ICF performance metrics and investigate variability in shot performance. While recent variability studies concentrate on the hybrid-E platform using features that can be difficult to control, we investigate variability using multiple campaigns with well controlled design parameters as features. We highlight features that drive ICF performance and/or variability and try to create a “one shot measurement”, where we map outputs to variability.

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. LLNL-ABS-836664