Development of tomographic reconstruction method for ITER VUV spectroscopic diagnostic system with a limited field of view

Abstract

The impurities in fusion plasmas are an indispensable part. Since tokamak is confining high-temperature plasma, high heat flux is necessarily applied to the plasma-facing components. In the future fusion devices such as ITER, high-Z plasma-facing components like tungsten divertor are designed to be installed because of its low erosion rate, etching rate, and high melting point. However, high-Z atoms, especially tungsten, emits high Bremsstrahlung radiation and line radiation in the high-temperature plasma due to their high atomic number. So, if high-Z atoms are injected into the Tokamak plasma as an impurity, it can cause high radiation cooling, leading to degradation of confinement performance and even plasma disruption. On the other hand, impurity-injecting scenarios are being studied to cause detachment between the divertor and plasma to reduce the heat flux applied to the divertor, so impurities are expected to be an essential part of the future fusion plasma. For stable plasma operation, it is necessary to diagnose the change in the density distribution of these impurities to control the impurities in plasma. The spectroscopic system is widely used to diagnose the plasma impurities by each nuclide and charge state. In the ITER device, a vacuum ultraviolet (VUV) spectroscopic system is planning to be installed in which field of view (FoV) covers the divertor area, plasma edge area, and plasma core area. ITER divertor VUV spectrometer covers 14.6 nm to 32 nm of wavelength to investigate impurities, such as Be, O, Kr, and W. To investigate impurities over time and space, this spectrometer has 21$^o$ spread angle for spatial resolution and 10 ms of time resolution. For better time resolution, the number of detector channels is divided by 10 at 21o spread angle. The prototype of this spectroscopic system is installed in KSTAR, which successfully get spatio-temporal spectra from several KSTAR experiments. However, the spectroscopic systems used in KSTAR and will be used in ITER have a relatively narrow FoV, which chord-integrated spectrum provides an inaccurate reconstructed distribution of local plasma impurities with conventional tomography method. Therefore, additional information on plasma impurity density distribution is needed to identify the local density distribution of impurities. This thesis uses a magnetic equilibrium assumption and expected poloidal asymmetry distribution from other diagnostics such as Thomson scattering system and charge exchange spectroscopy system for additional information. Because of high toroidal rotational speed of plasma, the impurity distribution exhibits poloidal asymmetry due to centrifugal force when the Mach number is the order of one. In JET and ASDEX-U with a tungsten divertor, this phenomenon has been confidently investigated with the tomographic reconstruction of soft X-ray and bolometer measurements. In ITER tokamak, the effect of centrifugal force is also essential because of its big radius of about 6m and toroidal rotation speed of about 6kHz. With the momentum balance equation of plasma impurities, the poloidal asymmetry distribution of each species can be expected from basic diagnostics in the tokamak. Thus, the new tomography method will exploit the chord-integrated spectrums and the predicted poloidal asymmetry distribution to identify the local two-dimensional (2D) poloidal impurity distribution. This tomographic methodology was first developed in the spectroscopic system of KSTAR and validated using the phantom impurity distribution predicted in KSTAR. Also, the real distribution of impurities was obtained using the measured spectrum in the KSTAR experiment, and the developed tomography method proved to be useful in the real experiment. In the ITER VUV spectroscopic system, the improved tomography method was developed and applied. Since the tomography method developed in the KSTAR spectroscopic systems multiplies the expected distribution of impurities in a one-dimensional (1D) reconstructed profile according to a magnetic field to obtain the 2D impurity distribution, the proportion of the expected impurity distribution is high. So, it results in a large error in the reconstructed profile even if only a small error occurs in the expected distribution. Thus, the improved tomography method developed a methodology to control the weight of the expected impurity distribution. The improved tomography method was validated using phantom ITER impurity distributions, and it was found that the improved tomography method could be applied to the ITER spectroscopic system. The improved tomography method can control the weight of the expected impurity distribution compared to the tomography method developed for the KSTAR spectrometer. It can be confirmed that the distribution of impurities is reconstructed more accurately than the conventional method by lowering the weight when high noise occurs in spectrometer spectra. In ITER, the high neutron flux is predicted, which can cause high noise on the spectroscopic system. Therefore, the improved tomography method is more suitable for use in ITER because it can control the weight of the expected impurity distribution and is predicted to be useful in ITER VUV spectroscopic systems. However, the developed tomography methods are based on magnetic equilibrium, so magnetic field information varies by time. It causes a large calculation time, which is unsuitable for monitoring impurities. To obtain a real-time distribution of impurities, algorithms are needed to reduce calculation time significantly. Neural networks are currently in the spotlight in real-time tomography and have the advantage of drastically reducing calculation time by using learned data. In this thesis, research to imitate the developed tomography method using deep neural network technology has been studied. This neural network model learned magnetic field information obtained from the KSTAR experiment and FoV of the spectroscopic system, significantly reducing the calculation time of obtaining detector geometric matrix w that occurs whenever magnetic field information changes. The verification of this neural network model using the phantom KSTAR impurity distribution confirmed a fast calculation speed and accurate accuracy. Therefore, this neural network technology is essential for real-time impurity distribution monitoring using spectrometer systems. Using the spectroscopic systems, this thesis developed tomographic methods to obtain impurity distributions. First, the tomography method, which used the expected impurity distribution by magnetic equilibrium and centrifugal force as additional information, was developed and utilized in KSTAR spectroscopic system, and the improved tomography method, which can respond more flexibly to the detector noise, was developed and verified in ITER spectroscopic system. Finally, neural network technology, which imitates the developed tomography methods and allows real-time monitoring by reducing calculation time, was developed and verified in the KSTAR spectrometer system. It is expected that the developed tomography methods will be useful for reconstructing the local distribution of impurities using the spectroscopic system with limited FoV not only in KSTAR and ITER but also in next-generation fusion devices.