Real-time kinetic profile reconstruction and Adaptive ELM Control on the DIII-D and KSTAR Tokamaks

Abstract

This PhD thesis focuses on the development and improvement of various control algorithms and modeling techniques for real-time plasma control and analysis in fusion experiments. Throughout this thesis, we develop the real-time kinetic equilibrium reconstruction tools that exceed the performance of any existing alternatives. Similarly, we develop an ELM controller that achieved the highest confinement with stable ELM suppression to date. The first part of the thesis introduces a profile fitting algorithm that accurately models electron density, temperature, ion temperature, and rotation profiles. The algorithm incorporates different fitting methods and constraints to enhance the reliability and robustness of the fits. The resulting profile fits are used to compute pressure constraints and are employed in equilibrium reconstruction. However, the trade-off between robustness and accuracy presents challenges, and the availability of reliable data at predefined intervals remains a concern. To address these issues, the second part of the thesis discusses the development of real-time plasma equilibrium reconstructions using the newly developed RTCAKENN algorithm. This machine learning based algorithm demonstrates its capability to generate high-quality real-time reconstructions of various plasma parameters, surpassing other real-time alternatives. It operates efficiently and exhibits robustness even in the absence of certain diagnostic data. However, further potential improvements have been identified, such as minimizing data pre-processing and exploring direct feeding of raw data to the algorithm. The third part presents the Feedback Adaptive RMP ELM Controller, which achieves and sustains Edge Localized Mode (ELM) suppression in experimental discharges. The controller minimizes the applied RMP current during ELM suppression, resulting in longer discharges and improved plasma performance. In the fourth part, the thesis highlights the development of an ELM-suppression-loss precursor detector, the cross-device implementation of the Feedback Adaptive RMP ELM Controller on KSTAR and DIII-D, as well as its multi-n capabilities and amplitude and phasing feedback. These results enhance confidence in the controller's general applicability. The final part of the thesis introduces the ML-IPEC surrogate model, which utilizes real-time information to adjust amplitude ratios and phasings, resulting in improved confinement and ELM suppression. These initial results demonstrate the potential of incorporating more model knowledge through machine learning-based surrogate models. Overall, this PhD thesis contributes to the advancement of real-time plasma control and analysis in fusion experiments through the development of profile fitting algorithms, ML-based RT equilibrium reconstruction algorithms, ELM suppression controllers, and surrogate models. The research presents promising results, identifies challenges, and suggests future directions for further improvement and applicability in ITER and other scenarios.