Magnetized Liner Inertial Fusion: Developing a data-driven understanding of magnetic flux compression from computation through experiment

The Magnetized Liner Inertial Fusion (MagLIF) concept studied at Sandia National Laboratories Z facility achieves fusion relevant conditions by delivering current to magnetically compress a cylindrical beryllium liner containing a preheated and pre-magnetized fusion fuel. Improvements to magnetic field coils, laser preheat, and current coupling have led to increased performance. However, experimental measurements are non-linearly dependent on several physical parameters and are highly integrated in the spatial and/or temporal domains. Furthermore, regular application of high-fidelity simulations that capture such unresolved effects in analyzing this data is hampered by high computational costs. The application of modern data-driven techniques is critical to addressing these challenges and advancing our understanding. We demonstrate several data science tools that enable the extraction and interpretation of flux compression at stagnation and inform our understanding of current open questions. We present a Bayesian inference of magnetization of the stagnated fuel plasma for an ensemble of MagLIF experiments, achieved by the application of a deep-learned surrogate of a costly physics simulation. For a subset of the experiments, flux is consistent with scaling of the Nernst effect with preheat-specific-energy (PSE) from 2D simulations. We also observe significantly increased fuel magnetization at the same PSE in experiments using dielectric-coated liners with a higher liner aspect ratio. This may indicate the importance of 3D effects and/or effective compression of the fuel. We discuss a recently developed deep-learning-based image preprocessing tool, an experimental-data-driven image metric design, and surrogate modeling efforts relevant for understanding open questions around impact of 3D effects on flux compression from both simulation and experiment.