Using Deep Learning to Analyze Thomson Scattering Diagnostic Data in Laboratory Astrophysics Experiments

AI and machine learning are becoming excellent tools to complement more common analysis methods used in supporting laboratory experiments. One machine learning application that has experienced several successes in the high energy density science field is the use of neural network (NN) surrogate models to approximate potentially costly and complex mathematical models.

Here we present our work using a NN surrogate model to analyze ion acoustic wave (IAW) features from Thomson scattering diagnostic data, collected from a laboratory astrophysics campaign at the OMEGA laser facility. The NN was trained on a large dataset of experimentally relevant Thomson scattered light spectra, generated from the Thomson model in the open-source code PlasmaPy. We include both self-validation of the NN by using train and test metrics, and external validation of the NN by extracting plasma parameters from a 1D kinetic Particle-In-Cell (PIC) Chicago simulation that is used to forward model the associated Thomson spectra. We discuss both model effectiveness and model limitations. We compare the NN predictions with a Markov-Chain Monte Carlo (MCMC) analysis of this simulated data. Finally, we compare NN predictions and MCMC analysis on one of the IAW images collected during the experimental campaign.