Abstract
Tomography serves as an advanced diagnostic tool for analyzing plasma fluctuations and turbulence. However, it requires time-consuming calculations, which impede rapid analysis. Integrating a neural network into tomography offers a potential solution. In this work, we present a trial conducted on a tomography system installed on the Plasma Assembly for Nonlinear Turbulence Analysis, a cylindrical plasma device designed for plasma turbulence research. This article reports on the optimization process of a neural network algorithm and on its excellent properties for tomographic reconstruction, including the extraction of plasma fluctuation properties. The neural network-aided tomography is 25 times faster than and provides comparable accuracy to, the standard algorithm, Maximum Likelihood Expectation Maximization.
Funding Information
  • Japan Science and Technology Agency (JPMJSP2136)
  • Japan Society for the Promotion of Science (JP25H00619)
  • Research Institute for Applied Mechanics, Kyushu University (2025EC-NU-2)