Applying Machine Learning Techniques on ICF Performance using Experimental Data

At the National Ignition Facility (NIF), the largest and most energetic laser in the world, cutting edge research is undertaken towards the goal of sustained nuclear fusion reactions.  This research could lead to a remarkable energy source with an energy density far greater than nuclear fission.  However, achieving a sustained burning plasma is difficult due to several possible implosion degradation mechanisms and limited laser energy. Recent breakthrough experiments are a testament that we have not fully optimized the experimental design in ICF experiments.  Simulations, the traditional way of designing experiments, is limited in capturing all the physics involved in actual experiments, making optimization a challenging task. A machine learning tool can provide new insight into poorly understood physics and lead targeted exploration of the parameter space and guide future experimental design and simulation.

Here we present our findings using multiple machine learning algorithms to analyze experimental data from NIF ICF experiments, focusing on the recent high yield experiments that approached the burning plasma regime.  Two physical models are presented, where the model incorporates features related to either pre-shot or mostly post shot information, to predict important metrics related to ICF experiments.  Finally, we use these models to investigate effects of variations in design parameters with the goal of improving performance in ICF experiments.

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.  LLNL-ABS-824477