Reconstruction of Three-Dimensional Core Structures in Inertial Confinement Fusion Implosion Experiments Using a Convolutional Neural Network Model

The performance of inertial confinement fusion (ICF) implosions is highly dependent on the properties of the core, the high-temperature central region. The capability of reconstructing 3-D core structures is crucial for understanding 3-D hot-spot formation and providing 3-D metrics to quantify ICF implosion performance. A deep-learning convolutional neural network (CNN) model was developed to reconstruct 3-D hot-spot and shell structures in ICF experiments. The training data were provided by DEC3D deceleration-phase simulations, which model perturbed hot spots in OMEGA cryogenic implosions. The CNN model is trained with synthetic x-ray images at multiple lines of sight to create a nonlinear mapping of 3-D hydrodynamic profiles. The model was validated using a 3-D hot-spot reconstruction method.[1] Good agreement was obtained in reconstructed 3-D hot-spot plasma emissivity profiles by mapping 2-D measured x-ray images from different lines of sight. The physics-informed CNN model is applied to reconstruct 3-D mass-density profiles based on x-ray image measurements, providing a new pathway to study 3-D areal-density and hot-spot shape asymmetries.