Tokamak Control Research Articles

Classification of the L-, H-mode, and plasma-free state: Convolutional neural networks and variational autoencoders on the edge reflectometer for KSTAR.

Classifying and monitoring the L-, H-mode, and plasma-free state are essential for the stable operational control of tokamaks. Edge reflectometry measures plasma density profiles, but the large volume of data and complexity in reconstruction pose significant challenges. There is a need for efficient methods to analyze complex reflectometer data in real-time, which can be addressed using advanced computational techniques. Here, we show that machine learning (ML) techniques can classify discharge states using raw signal data from an edge reflectometer installed on the Korea Superconducting Tokamak Advanced Research. The deep convolutional neural network models achieved classification accuracy of up to 99% when using 2D spectrogram inputs, demonstrating a significant improvement over 1D raw signal inputs. Additionally, the variational autoencoder model effectively clustered the discharge states in the latent space without any label information, further validating the model's capability to classify discharge states. These results suggest that the ML model can effectively handle the complexity of reflectometer data and accurately classify plasma discharge states. This approach not only facilitates real-time diagnosis but also reduces the need for manual data processing.

Low latency optical-based mode tracking with machine learning deployed on FPGAs on a tokamak.

Active feedback control in magnetic confinement fusion devices is desirable to mitigate plasma instabilities and enable robust operation. Optical high-speed cameras provide a powerful, non-invasive diagnostic and can be suitable for these applications. In this study, we process high-speed camera data, at rates exceeding 100 kfps, on in situ field-programmable gate array (FPGA) hardware to track magnetohydrodynamic (MHD) mode evolution and generate control signals in real time. Our system utilizes a convolutional neural network (CNN) model, which predicts the n = 1 MHD mode amplitude and phase using camera images with better accuracy than other tested non-deep-learning-based methods. By implementing this model directly within the standard FPGA readout hardware of the high-speed camera diagnostic, our mode tracking system achieves a total trigger-to-output latency of 17.6 μs and a throughput of up to 120 kfps. This study at the High Beta Tokamak-Extended Pulse (HBT-EP) experiment demonstrates an FPGA-based high-speed camera data acquisition and processing system, enabling application in real-time machine-learning-based tokamak diagnostic and control as well as potential applications in other scientific domains.

Correction: Optimizing 3D magnetic perturbations for edge instability control in the KSTAR tokamak

Correction: Optimizing 3D magnetic perturbations for edge instability control in the KSTAR tokamak

Optimizing 3D magnetic perturbations for edge instability control in the KSTAR tokamak

Optimizing 3D magnetic perturbations for edge instability control in the KSTAR tokamak

Machine learning-based real-time kinetic profile reconstruction in DIII-D

Kinetic equilibrium reconstruction plays a vital role in the physical analysis of plasma stability and control in fusion tokamaks. However, the traditional approach is subjective and prone to human biases. To address this, the consistent automatic kinetic equilibrium reconstruction (CAKE) method was introduced, providing objective results. Nonetheless, its offline nature limits its application in real-time plasma control systems (PCSs). To address this limitation, we present RTCAKENN, a machine learning model that approximates 7 CAKE-level output profiles, namely pressure, inverse q, toroidal current density, electron temperature and density, carbon ion impurity temperature and rotation profiles, using real-time available inputs. The deep neural network consists of an encoder layer, where the scalars and interdependent inputs such as plasma boundary coordinates and motional Stark effect data are encoded using multi-layer perceptrons (MLPs), while profile inputs are encoded by 1D convolutional layers. The encoded data is passed through a MLP for latent feature extraction, before being decoded in the decoding layers, which consist of upsampling and convolutional layers. RTCAKENN has been implemented in the DIII-D PCS and our model achieves accuracy comparable to CAKE and surpasses existing real-time alternatives. Through clever dropout training, RTCAKENN exhibits robustness and can operate even in the absence of Thomson scattering data or charge exchange recombination data. It executes in under 8 ms in the real-time environment, enabling future application in real-time control and analysis.

MEPhIST-0 integrated control and data acquisition system

The growing fusion research industry faces an extensive need for specialists with practical skills. However, the high complexity of the relevant experiments makes it difficult to maintain laboratories in which students can get their first hands-on experience of interacting with relevant devices. One of the options is using small, simplified machines that are designed for low-qualified personnel. This paper presents the architecture and implementation of a control and data acquisition system for an educational tokamak MEPhIST-0. It integrates different plasma diagnostic systems as well as tokamak's control nodes using the EPICS framework. Different approaches to data archiving and access are discussed, and the overall design is presented from the perspective of a primarily educational installation, as well as the preliminary results.

Retraction Note to: Edge-Localized Modes (ELM) Modification with Transport Barrier Control in Tokamak by Analytical Strategies

Retraction Note to: Edge-Localized Modes (ELM) Modification with Transport Barrier Control in Tokamak by Analytical Strategies

Model predictive current profile control in tokamaks by exploiting spatially moving electron cyclotron current drives

The plasma control systems in next-generation tokamaks like ITER will balance competing control objectives to achieve the desired level of performance in advanced scenarios while preventing magnetohydrodynamic instabilities and disruptions. During normal tokamak operation, the points of incidence of the electromagnetic waves generated by the electron cyclotron heating and current drives (EC H&CDs) are usually fixed in space. However, the points of incidence can be modified in real-time by changing the angles of the mirrors that reflect the EC H&CD waves. Altering the points of incidence, in turn, varies the ability of the plasma control system to regulate a plasma property. For instance, changing the EC H&CD wave incidence location may place the power demands necessary to achieve a particular plasma target within saturation limits. Therefore, using the EC H&CD deposition location, which is related to the EC H&CD mirror angle, as a supplementary controllable variable may facilitate access to a given target scenario. However, active scenario-control algorithms have not been designed so far to fully exploit this capability in real time. In this work, a model predictive controller that can handle actuation locations as control inputs is developed. In particular, the controller is designed to regulate both the auxiliary powers and the EC H&CD deposition locations in a pre-defined optimal sense to achieve the control objective of attaining and sustaining a target current profile. The proposed controller is tested for a DIII-D tokamak scenario in nonlinear simulations using the Control Oriented Transport SIMulator (COTSIM).

O-mode reflectometry on Aditya-U

Density profile measurements using a FMCW reflectometer have been done for the first time on Aditya-U tokamak at IPR. Fast swept frequencies in the K-Band (18–28 GHz) were transmitted in 10 μs and the reflected waves were collected via a high gain (20 dBi) pyramidal horn placed outside the vacuum vessel in O-Mode of operation. System was synchronised with the tokamak control using an custom trigger engine with each trigger generating a non-linear ramp of 0–20 V in 10μs and output I and Q data recorded at high speeds of 200 MSps with a memory depth of 1 GS of 16 bits each. Reflections from the viewport which masked the plasma signal were mitigated by tilting the viewport to 25 degrees. Beat frequencies were estimated using the spectogram method and inversion was done by Bottolier-Curtet algorithm. First results from the profile measurements on Aditya-U are presented.

Mixed-Integer MPC Strategies for Fueling and Density Control in Fusion Tokamaks

Model predictive control (MPC) is promising for fueling and core density feedback control in nuclear fusion tokamaks, where the primary actuators, frozen hydrogen fuel pellets fired into the plasma, are discrete. Previous density feedback control approaches have only approximated pellet injection as a continuous input due to the complexity that it introduces. In this letter, we model plasma density and pellet injection as a hybrid system and propose two MPC strategies for density control: mixed-integer (MI) MPC using a conventional mixed-integer programming (MIP) solver and MPC utilizing our novel modification of the penalty term homotopy (PTH) algorithm. By relaxing the integer requirements, the PTH algorithm transforms the MIP problem into a series of continuous optimization problems, reducing computational complexity. Our novel modification to the PTH algorithm ensures that it can handle path constraints, making it viable for constrained hybrid MPC in general. Both strategies perform well with regards to reference tracking without violating path constraints and satisfy the computation time limit for real-time control of the pellet injection system. However, the computation time of the PTH-based MPC strategy consistently outpaces the conventional MI-MPC strategy.

Impact of Neutral Beam and Plasma Shaping on the Current Drive Efficiency and Fast Ion Losses in a Compact Tokamak FNS-ST

Fusion neutron source (FNS)-spherical tokamak (ST) will be a FNS based on ST ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$R =0.5$ </tex-math></inline-formula> m, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$a =0.3$ </tex-math></inline-formula> m, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$k =2.75$ </tex-math></inline-formula> , and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$B =1.5$ </tex-math></inline-formula> T) with a steady-state neutron yield ~1018 n/s. The noninductive current drive and neutron generation is to be maintained by neutral beam injection (NBI). Beam in toroids (BTOR) code (Python) is used to reproduce NBI and plasma shaping with high degree of detail and to perform parametric studies of neutral beam (NB) driven effects. The beam model light neutral beam (LNB) is inherited from the original beam transmission (BTR) code, used since 2005 for NBI design and optimizations. The BTOR methodology combines NB statistical description with high-performance analytical models of particle tracing, which allows one to get results several orders faster, if compared to the conventional (Monte–Carlo) approach, and still with a good agreement. The results obtained for the FNS-ST device prove that the entire beam efficiency highly depends on the beam–plasma mutual size and shaping, including the beam cross section and aiming, plasma aspect ratio, elongation, and triangularity. The results are especially important for a steady-state current drive and fusion control in low aspect ratio tokamaks; the BTOR method is also applicable to a conventional tokamak design.

Real-time density feedback control in ADITYA-U Tokamak

A real-time density feedback control system with a 100 GHz heterodyne interferometer has been designed, developed, and commissioned in ADITYA-U tokamak. It consists of three subsystems i.e. density measurement, feedback control, and gas fuelling. A proportional feedback controller is configured in voltage amplitude mode to operate the piezo valve for gas injection. Experiments are performed during plasma discharges to achieve a predefined density evolution as well as constant density. Stickiness of the piezo valve restricts the temporal response of the valve. It is required to minimize during the plasma discharge for a fast density response which is achieved by a short preceding voltage pulse known as a stick pulse. This paper describes the implementation of a real-time density feedback control system which is successfully validated using another existing 140 GHz interferometer system in ADITYA-U. Moreover, the developed system consumes low power, is cost-effective and re-programmable, can be easily upgraded, and provides interlock with plasma parameters.

The modeling of a tokamak plasma discharge, from first principles to a flight simulator

A newly developed tool to simulate a tokamak full discharge is presented. The tokamak ‘flight simulator’ Fenix couples the tokamak control system with a fast and reduced plasma model, which is realistic enough to take into account several of the plasma non-linearities. A distinguishing feature of this modeling tool is that it only requires the pulse schedule (PS) as input to the simulator. The output is a virtual realization of the full discharge, whose time traces can then be used to judge if the PS satisfies control/physics goals or needs to be revised. This tool is envisioned for routine use in the control room before each pulse is performed, but can also be used off-line to correct PS in advance, or to develop and validate reduced models, control schemes for future machines like a commercial reactor, simulating realistic actuators and sensors behavior.

Advances in prediction of tokamak experiments with theory-based models

The successful validation of theory-based models of transport, magnetohydrodynamic stability, heating and current drive, with tokamak measurements over the last 20 years, has laid the foundation for a new era where these models can be routinely used in a ‘predict first’ approach to design and predict the outcomes of experiments on tokamaks today. The capability to predict the plasma confinement and core profiles with a quantified uncertainty, based on a multi-machine, international, database of experience, will provide confidence that a proposed discharge will remain within the operational limits of the tokamak. Developing this predictive capability for the first generation of burning plasma devices, beginning with ITER, and progressing to tokamak demonstration reactors, is a critical mission of fusion energy research. Major advances have been made implementing this predict first methodology on today’s tokamaks. An overview of several of these recent advances will be presented, providing the integrated modeling foundations of the experimental successes. The first steps to include boundary plasmas, and tokamak control systems, have been made. A commitment to predicting experiments as part of the planning process is needed in order to collect predictive accuracy data and evolve the models and software into a robust whole discharge pulse design simulator.

New horizontal and vertical field coils with optimised location for robust decentralized plasma position control in the IGNITOR tokamak

The paper deals with the improvement of the poloidal system of the IGNITOR tokamak, and the design and simulation of the robust decentralised control systems of the plasma’s vertical and horizontal position. The results were obtained on a linear model that is constructed from the circuit equations for the coils, plasma, and vacuum vessel (VV) currents, and from the force balance equations for the plasma. Two new coils that are located close to the VV have been introduced in the IGNITOR poloidal system; specifically, a Horizontal Field Coil (HFC) that is designated for the plasma vertical position control and a Vertical Field Coil (VFC) that is used for the horizontal position control. For the first time in the design of D-shaped tokamaks poloidal systems, the location of the HFC was optimised to achieve the maximum size of the plasma controllability region in the vertical direction and the location of the VFC was determined from the optimisation of the plasma position response to the voltage step function. Although the ideal position of the VFC with fast response time is on the equatorial plane, its location usually is occupied by the other devices such as ports, diagnostic system, and heating systems, so the separated two-section VFC is used. Careful calculation shows the performance is degraded by only 20 % in this layout. For the improved IGNITOR poloidal system, the new robust decentralised plasma position control systems were designed by means of two robust approaches: H∞-optimisation and Linear Matrix Inequalities (LMI). The results of the simulation of the control systems with two types of actuator models as the multiphase thyristor rectifier and the voltage inverter in Pulse- Width Modulation (PWM) mode for the new IGNITOR poloidal system are presented.

Feedforward beta control in the KSTAR tokamak by deep reinforcement learning

In this work, we address a new feedforward control scheme for the normalized beta (β N) in tokamak plasmas, using the deep reinforcement learning (RL) technique. The deep RL algorithm optimizes an artificial decision-making agent that adjusts the discharge scenario to obtain a given target β N from the state–action–reward sets explored by its own trial and error in a virtual tokamak environment. The virtual environment for the RL training is constructed using a long short-term memory (LSTM) network that imitates the plasma responses to external actuator controls, which is trained using five years’ worth of KSTAR experimental data. The RL agent then experiences numerous discharges with different actuator controls in the LSTM simulator, and its internal parameters are optimized in the direction of maximizing the reward. We analyze a series of KSTAR experiments conducted with the RL-determined scenarios to validate the feasibility of the beta control scheme in a real device. We discuss the successes and limitations of feedforward beta control by RL, and suggest a future research path for this area of study.

Plasma initiation and preliminary magnetic control in the HL-2M tokamak

The first plasmas have been achieved in the HL-2M tokamak by fulfilling the usual criteria for plasma breakdown, i.e. optimal working gas pressure, field null configuration of poloidal magnetic field, reasonable toroidal electrical field and clean wall conditions. The characterization of plasma initiation, such as identification of initiation time and position, is performed by diagnostic data analysis and numerical simulation. A model-based method, which enables dedicated pick-up sensors to estimate plasma radial position in real time, is applied for the limiter plasma shape in the first plasma commissioning phase. For a circular cross-section plasma, feedback control on plasma current and radial position allowed us to extend discharge durations to ∼250 ms associated with a 120 kA plasma current in flat-top. Preliminary plasma equilibrium reconstruction by the NICE code from pick-up sensors and flux loops, as well as interferometry measurements, gives plasma boundaries, plasma profiles of toroidal current density and safety factor, as well as electron density, in an offline way.

Nonlinear feedback algorithm for tearing-mode magnetic control in tokamak

Results of calculations and analysis of the tearing-mode evolution under the effect of resonant magnetic perturbation produced by a nonlinear feedback system in rotating tokamak plasma are presented. The TEAR-code used for calculations is based on the visco-resistive magnetohydrodynamic approximation that gives coupled diffusion-type equations for the magnetic flux perturbation and for the plasma rotation velocities in toroidal and poloidal directions. The code is supplemented by a computational unit simulating the nonlinear algorithm of the feedback system. The feedback gain and phase shift are automatically regulated in real time. The gain depends on the tearing-mode amplitude. At high feedback input signals, the gain does not exceed some finite value to reduce technical requirements for the feedback actuator and to mitigate the phase instability. In the process of the tearing-mode suppression, the gain gradually increases to provide the mode control under reduction of feedback input signal and rise of the intrinsic tearing-mode stability index. Besides that, the real-time control of the feedback phase shift is provided to additionally mitigate the phase instability. According to simulation, the used nonlinear characteristics of the feedback algorithm improve the efficiency of the tearing-mode suppression.

Real-time feedback control system for ADITYA-U horizontal plasma position stabilisation

The ADITYA-U tokamak (R0 = 0.75 m, a = 0.25 m) is designed to shape plasma column in both single and double null diverter configurations. It is quite well known that sustaining a shaped plasma in tokamak requires very good plasma column position control, both horizontal and vertical. An FPGA-based proportional–integral–derivative (PID) control system has been designed and operated to achieve horizontal plasma position control in ADITYA-U tokamak. The complete system has been rigorously tested with sample signals before implementing to the ADITYA-U plasma discharges. The control system is integrated and time-synchronized with the plasma discharge operation of ADITYA –U. Furthermore, the system has been trained to take appropriate actions during the disruption or plasma failure in the tokamak operation. Detailed experimental results have been obtained by the operation of the digital PID controller. The complete design, installation, operation, tuning of the system along with all the relevant testing and operating experience of the digital PID controller for real-time horizontal plasma position control is presented in the paper.

FULLY CONVOLUTIONAL SPATIO-TEMPORAL MODELS FOR REPRESENTATION LEARNING IN PLASMA SCIENCE

We have trained a fully convolutional spatio-temporal model for fast and accurate representation learning in the challenging exemplar application area of fusion energy plasma science. The onset of major disruptions is a critically important fusion energy science issue that must be resolved for advanced tokamak plasmas such as the $25B burning plasma international thermonuclear experimental reactor (ITER) experiment. While a variety of statistical methods have been used to address the problem of tokamak disruption prediction and control, recent approaches based on deep learning have proven particularly compelling. In the present paper, we introduce further improvements to the fusion recurrent neural network (FRNN) software suite, which delivered cross-machine disruption predictions with unprecedented accuracy using a large database of experimental signals from two major tokamaks. Up to now, FRNN was based on the long short-term memory (LSTM) variant of recurrent neural networks to leverage the temporal information in the data. Here, we implement and apply the "temporal convolutional neural network (TCN)" architecture to the time-dependent input signals. This allows highly optimized convolution operations to carry the majority of the computational load of training, thus enabling a reduction in training time, and the effective use of high-performance computing resources for hyperparameter tuning. At the same time, the TCN-based architecture achieves better predictive performance when compared with the LSTM architecture for various tasks for a representative fusion database.