Source localization for neutron imaging systems using convolutional neural networks.

Area of Science:

• Nuclear Fusion Science
• Machine Learning Applications
• Diagnostic Imaging

Background:

• Nuclear imaging is vital for diagnosing inertial confinement fusion (ICF) implosions at the National Ignition Facility (NIF).
• Accurate source localization within neutron aperture images is critical for reconstructing ICF implosion geometry.
• Current source localization relies on iterative optimization algorithms, which can be computationally intensive.

Purpose of the Study:

• To introduce and evaluate a machine learning-based approach for fusion source localization in NIF nuclear imaging.
• To compare the performance of the machine learning method against traditional optimization-based techniques.
• To assess the impact of the new method on computation time and accuracy for ICF diagnostics.

Main Methods:

• Training a convolutional neural network (CNN) to predict fusion source locations from neutron aperture images.
• Utilizing synthetic data and actual NIF deuterium-tritium (DT) shot data for model training and validation.
• Comparing the computational efficiency and localization accuracy of the CNN approach with existing iterative optimization algorithms.

Main Results:

• The developed machine learning approach significantly decreases computation time by several orders of magnitude.
• The CNN-based source localization achieves accuracy comparable to the current optimization-based methods.
• The method demonstrates effectiveness on both synthetic datasets and real-world NIF DT experimental data.

Conclusions:

• Machine learning, specifically CNNs, offers a highly efficient and accurate alternative for fusion source localization in NIF nuclear imaging.
• This advancement can accelerate the analysis of ICF implosion geometry, improving diagnostic capabilities.
• The findings pave the way for faster, more streamlined fusion research diagnostics.