Deep learning model successfully predicted ignition in inertial confinement fusion experiment at the U.S. NIF

Good news! The fusion of ML & AI with nuclear fusion is only the beginning!

"... Predicting the results of this process without AI are prohibitively intensive, taking hours for each simulation and requiring many simulations for a strong result. So researchers created a fusion model based on the outcomes of NIF [U.S. National Ignition Facility] experiments from 2021-2022, then combined it with generative machine learning to predict the possible outcomes of successive experiments based on the model’s previous results. The machine learning model can run suites of simulations of complicated physics processes like hydrodynamics, thermodynamics, and nuclear activity, as well as the setup and results of the earlier fusion experiments, within a few days total.

All told, the AI model estimated that the NIF’s next big fusion test would have a 74% chance of success—and it turned out to be right. “This outcome demonstrates a promising approach to predictive modeling of [fusion] experiments and provides a framework for developing data-driven models for other complex systems,” ..."

From the editor's summary and abstract:

"Editor’s summary

The inertial confinement fusion experiment uses powerful lasers to induce nuclear reactions and produce fusion energy. The design of the experiment is complex, and computer simulations are routinely used to optimize it. Typically, these simulations require manual tuning, which limits their predictive power. Spears et al. built a generative machine learning, physics-informed model that successfully predicted the initial ignition shot and the statistics of subsequent experiments with the same design. ...

Abstract

An inertial confinement fusion experiment, carried out at the National Ignition Facility, has achieved ignition by generating fusion energy exceeding the laser energy that drove the experiment. Prior to the experiment, a generative machine learning model that combines radiation hydrodynamics simulations, deep learning, experimental data, and Bayesian statistics was used to predict, with a probability greater than 70%, that ignition was the most likely outcome for this shot."

ScienceAdviser

Deep learning model successfully predicted ignition in inertial confinement fusion experiment

Predicting fusion ignition at the National Ignition Facility with physics-informed deep learning (no public access)