Source shape estimation for neutron imaging systems using convolutional neural networks.

Area of Science:

• Nuclear Fusion Physics
• Diagnostic Imaging
• Computational Science

Background:

• Neutron imaging is crucial for diagnosing inertial confinement fusion (ICF) at the National Ignition Facility (NIF).
• Current methods reconstruct fusion source geometry using computationally intensive maximum likelihood estimation.
• Faster methods are needed for simplified representations of fusion source geometry.

Purpose of the Study:

• To develop and evaluate convolutional neural networks (CNNs) for reconstructing fusion source geometry.
• To compare CNN performance with traditional methods for neutron imaging data.
• To demonstrate the application of CNNs on both penumbral and pinhole imaging data.

Main Methods:

• Development of convolutional neural networks (CNNs) for image reconstruction.
• Utilizing neutron flux data from aperture arrays and scintillator detectors.
• Testing CNN performance on simulated and experimental neutron imaging data, including noisy conditions.

Main Results:

• CNNs successfully reconstruct outer contours of simple fusion source geometries.
• The developed CNNs provide a computationally efficient alternative to traditional methods.
• Demonstrated effectiveness of CNNs on both penumbral and pinhole imaging configurations, even with noise.

Conclusions:

• CNNs provide a rapid and effective tool for analyzing neutron imaging data in ICF research.
• This method enables quicker characterization of fusion source geometry, aiding experimental analysis.
• The CNN approach shows promise for real-time or near-real-time diagnostics in fusion experiments.