Development of Neural Networks for Rapid Analysis of a High Repetition Rate X-Ray Diagnostic

Many high repetition rate capable (0.1 – 10 Hz) PW-class laser facilities now operate across the world. While these facilities can produce data at high rates, conventional diagnostics and their accompanying on-shot data analysis techniques are not able to keep up. For this reason, it would be beneficial to automate the data analysis with algorithms that can keep up with the high repetition rate of these facilities while maintaining accuracy on par with traditional analysis. Neural networks (NN) have been used in a variety of fields to analyze large data sets and have been useful for problems of image classification, object recognition, natural language processing, and more recently data analysis from scientific experiments. Here we present results on training a NN to analyze data from UCXS (Ultra-Compact X-ray Spectrometer). UCXS is a high-repetition-rate diagnostic that uses a combination of step-wedge and ross pair filtration that operates in the soft X-ray regime (1 keV - 25 keV) to determine parameters such as plasma temperature. In this presentation, we will discuss preparation of synthetic datasets for initial NN training, the development of deep and convolutional NN's, and the performance on experimental data.*