Predictive Machine Learning Model of Stimulated Brillouin Backscatter at the National Ignition Facility*

Due to the increasing laser drive energy and stringent control of hohlraum x-ray drive symmetry needed for high-yield (igniting) inertial confinement fusion (ICF) implosions, predictive models of laser-plasma interactions (LPI) become increasingly important. Stimulated Brillouin backscatter (SBS) is a particularly problematic LPI phenomenon as it can redirect 3ω light back along the beam path, potentially resulting in optics damage, and can impact drive symmetry by reducing the drive on the capsule in a space- and time-varying fashion. A predictive model of SBS would allow us to reduce “walk-up shots” (reduced energy and/or power shots used to test whether optics may be damaged by a similar shot taken at full energy and power), allowing more laser time devoted to key physics experiments, and would improve our pre-shot modeling efforts, as we could better account for drive symmetry implications under design changes and quantify the expected variability in implosion performance due to shot-to-shot variations in SBS.

In this talk we detail our efforts in building a machine learning (ML) model to predict SBS. The model is trained on over 800 previous NIF ICF shots, where SBS was recorded by drive diagnostics (DrDs¹) and the full aperture backscatter station (FABS²). The model learns the impact on the SBS signals from a large number of laser, capsule, and hohlraum design parameters, which allows it to make predictions on future shots. We discuss the performance of the model, the uncertainty in its predictions, and how we can leverage it to enhance our predictive capabilities.

¹B. J. MacGowan Anomalous Absorption Conference June 9-14, 2019, Telluride, CO, United States.

²J. D. Moody, et al. Review of scientific instruments 81.10 (2010)