Next-generation fusion facilities like ITER face a "data deluge," generating petabytes of multi-diagnostic signals daily that challenge manual analysis. We present a "signals-first" self-supervised framework for the automated extraction of coherent and transient modes from high-noise time-frequency data across a variety of sensors. We also develop a general-purpose method and tool for extracting coherent, quasi-coherent, and transient modes for fluctuation measurements in tokamaks by employing non-linear optimal techniques in multichannel signal processing with a fast neural network surrogate on fast magnetics, electron cyclotron emission, CO2 interferometers, and beam emission spectroscopy measurements from DIII-D. Results are tested on data from DIII-D, TJ-II, and non-fusion spectrograms. With an inference latency of 0.5 seconds, this framework enables real-time mode identification and large-scale automated database generation for advanced plasma control. Repository is in https://github.com/PlasmaControl/TokEye.

Plasma-facing components (PFCs) that incorporate flowing liquid lithium surfaces are a promising technology because they can provide a renewable interface, modify recycling, support power exhaust, and potentially connect PFCs with fuel recovery. The deployment of liquid lithium PFC technology, however, is constrained by lithium chemical reactivity, fire and aerosol hazards, vacuum and inert gas operation, maintainability, and the need for rapid experimental iteration. Part 1 of this work develops a hazard-complexity metric for selecting secondary-containment architectures for flowing liquid-lithium systems. The framework is applied to six representative containment scenarios and to the Lithium Experimental Application Platform (LEAP), currently under construction at Princeton Plasma Physics Laboratory. LEAP uses a modular, room-sized argon glovebox as an inert secondary containment boundary for a staged flowing lithium program incorporating heating, diagnostics, magnetic field exposure, and future device interface capability. The analysis indicates that an inert, airtight secondary enclosure without scrubbers around an enclosed lithium loop provides a practical balance between hazard reduction and facility complexity for the present LEAP design requirements. This architecture offers a deployable pathway for lithium PFC development and a transferable design logic for other reactive or conductive liquid metal systems.

Recent advances in high-temperature-superconductor technology have made substantially higher toroidal magnetic fields technologically accessible, reopening the design space for compact, high-field tokamak reactors. Because reactor performance projections remain anchored to empirical confinement scalings, the recent update to the ITPA global H-mode confinement database raises an important question: what does the present experimental record and its uncertainty imply for the path to reactor-grade fusion performance? In this work, we revisit confinement extrapolation from an explicitly extrapolation-oriented perspective and, to complement its implications in terms of a direct reactor performance measure, present a cross-machine empirical scaling for fusion power. We systematically search for a minimally complex confinement scaling that optimizes the tradeoff between variance capture and extrapolative robustness. We find that low-order models centered near N=3 to N=4 optimize this tradeoff, with plasma current, machine size, heating power, and elongation emerging as the dominant engineering levers, together with an empirically inferred confinement penalty associated with metallic walls. Recast in reactor-performance terms, the results indicate that both the fusion triple product and fusion power are governed primarily by plasma current: the triple product scales approximately as I_p^2, and the empirical fusion power scaling exhibits a similarly near-quadratic dependence over a survey of the highest performing discharges across several machines. Projecting to reactors, these results suggest that high-field devices with metal walls may require higher plasma current than standard IPB98(y,2)-based expectations imply, and that gigawatt-class tokamak performance likely demands operation at I_p ≳ 20MA.

As contemporary experimental tokamaks are pushed toward reactor-relevant operation, they provide essential testbeds for demonstrating and exploring edge-localized mode (ELM) control strategies for deployment in future fusion pilot plants. Accurate predictions of plasma responses are essential to guide and optimize such demonstrations. This paper introduces and validates an integrated modeling framework over a diverse range of DIII-D operational space. The integrated modeling framework uses only scalar plasma parameters and optional reference boundary to self-consistently and flexibly scan through tokamak operational space and estimate the ELM-suppression relevant plasma response over 3D coil phase space. The framework generates tightly converged equilibria that satisfy a target set of plasma parameters ( $I_\mathrm p$ , $\beta_\mathrm N$ , $I_\mathrm i$ ), with kinetic profiles constrained by an EPED(NN)-computed pedestal and empirical core model. The ideal plasma response of these modeled equilibria is calculated with generalized perturbed equilibrium code (GPEC). Validated against 55 distinct DIII-D equilibria, the framework consistently reproduces pressure, $q$ , and other representative profiles from kinetically constrained equilibrium reconstructions using the time-varying information from only evolving scalar plasma parameters; plasma response validation is performed for the $n = 3$ perturbation. In particular, the plasma response of a DIII-D discharge scanning $q_{95}$ in search of ELM suppression windows is quantitatively reproduced. By accurately reproducing reconstructed equilibria and their corresponding calculated plasma responses across wide parameter variation, the framework supports potential for synthetic parameter scans in key operational, stability, and plasma response dimensions to investigate resonant magnetic perturbation (RMP) ELM-suppression experiments and inform predictive RMP scenario optimization.

Stellarator optimization is a multi-objective, non-convex problem characterized by a complex objective landscape containing many local minima. The solution resulting from a single optimization is highly sensitive to factors such as the initial guess, objective weights, and the optimization method employed. However, merely varying these factors does not guarantee that a physically distinct minimum will be found; optimizations often fail to converge to good minima or simply return to the same or very similar local minima despite large-scale parameter scans. This paper presents a novel application of deflation methods to effectively explore this landscape. By modifying the objective function to penalize and "deflate" away already-found solutions, this technique encourages the optimizer towards attractive, distinct new minima while using a single initial guess and optimization setup. We provide a primer on deflation for nonlinear systems and non-convex optimization before applying it to non-axisymmetric equilibrium and stellarator optimization problems. Key results include the discovery of families of global equilibria with similar core characteristics and the convergence to helical core equilibria without prescient initial guesses. Furthermore, we demonstrate that augmenting stage-one stellarator and stage-two coil optimization with deflation constraints readily produces multiple high-quality, distinct solutions, establishing the method's efficacy and ease of use.

While artificial intelligence (AI) has been promising for fusion control, its inherent black-box nature will make compliant implementation in regulatory environments a challenge. This study implements and validates a real-time AI-enabled linear and interpretable control system for successful divertor detachment control with the DIII-D lower divertor camera. Using D _2 gas, we demonstrate successful feedback divertor detachment control with a mean absolute difference of 2% from the target for both detachment and reattachment. This automatic training and linear processing framework can be extended to any image-based diagnostic for future fusion reactors.

Tokamaks remain leading candidates for achieving practical fusion energy, yet many important control problems inside these devices are still difficult or unsolved. One such challenge is controlling the plasma rotation profile, which strongly influences stability, confinement, and transport. While the average rotation can be controlled, controlling the full profile is challenging due to high dimensionality, response to multiple actuators and dependence on plasma condition. Learning-based control methods, such as reinforcement learning (RL), provide a potential solution to this challenging problem with ability to model complex interactions leading to effective multi-input multi-output control. However, learning such policies is challenging due to the lack of accurate simulators that can model the rotation profile dynamics. In this work, we investigate the use of offline RL and offline model-based RL algorithms for rotation profile control, training them solely on historical data from the DIII-D tokamak. Our final method uses probabilistic models of plasma dynamics to generate rollouts for RL training. We deploy this policy on the DIII-D Tokamak and observe promising real-world results. We conclude by highlighting key challenges and insights from training and deploying an RL policy on a complex physical device while using only limited past data.

Training effective foundation models requires massive and organized datasets, yet scientific domains such as nuclear fusion present unique challenges due to largely heterogeneous and sparse data. Here we characterize the data used in developing such a model: with over 20 sensor types spanning 5 orders of magnitude in sampling rate, mixed tensor structures (point measurements, spectrograms, images), and nonstationary physics. We analyze our input complexity and discuss trade-offs between temporal context and frequency resolution. Our analysis provides a template for representing multi-modal fluctuation data at scale, with implications for both multi-modal control systems and nuclear fusion.

The edge-localized mode (ELM) frequency (f_ELM) was successfully controlled in real time on KSTAR using a proportional-integral (PI) feedback controller, employing a D_2 divertor gas puff as the actuator under tungsten lower-divertor conditions. The controller accurately tracked a two-step target—a 30 Hz increase in f_ELM for 4 s, followed by a 30 Hz decrease for 3 s—yielding mean and median absolute percentage errors of approximately 13

In this work, we demonstrate the deployment of a hardware-accelerated machine learning (ML) inference system integrated into a real-time processing at the DIII-D tokamak fusion reactor. The team has successfully deployed an AMD/Xilinx KCU1500 field-programmable gate array (FPGA) into the realtime Plasma Control System (PCS) nodes that receives the live Beam Emission Spectroscopy (BES) signal used for Edge Localized Mode (ELM) forecasting. The FPGA hosts a dense neural network using the SLAC Neural Network Library (SNL) that has been trained to infer the likelihood of disruptive ELM conditions. This likelihood then feeds a separate plasma controller that uses Resonant Magnetic Perturbation coils to suppress the predicted disruptive condition. The SNL allows for on-the-fly updates of the neural network weights and biases without requiring full hardware resynthesis for the FPGA. Judicious design of the neural-network architecture can further allow for the hot-swapping of multiple classification tasks to be executed on the single FPGA, significantly enhancing the real-time adaptability of the system for context-aware control strategies that respond in real-time to evolving reactor conditions. These adaptive weights naturally support continuous model refinement and seamless task switching during live experimental operation. This use case is chosen as a high rate signal processing example that can serve as a template for general ML-based reactor diagnostic processing for active reactor control systems. We see this as an essential development for achieving reactor relevant operation in future continuous operation fusion devices.

Data-driven acceleration of scientific computing workflows has been a high-profile aim of machine learning (ML) for science, with numerical simulation of transient partial differential equations (PDEs) being one of the main applications. The focus thus far has been on methods that require classical simulations to train, which when combined with the data-hungriness and optimization challenges of neural networks has caused difficulties in demonstrating a convincing advantage against strong classical baselines. We consider an alternative paradigm in which the learner uses a classical solver's own data to accelerate it, enabling a one-shot speedup of the simulation. Concretely, since transient PDEs often require solving a sequence of related linear systems, the feedback from repeated calls to a linear solver such as preconditioned conjugate gradient (PCG) can be used by a bandit algorithm to online-learn an adaptive sequence of solver configurations (e.g. preconditioners). The method we develop, PCGBandit, is implemented directly on top of the popular open source software OpenFOAM, which we use to show its effectiveness on a set of fluid and magnetohydrodynamics (MHD) problems.

We report the experimental results of controlling the pedestal-top electron density by applying resonant magnetic perturbation (RMP) with in-vessel control coils and the main gas puff in the 2024-2025 KSTAR experimental campaign. The density is reconstructed using a parameterized ψ N grid and five channels of line-averaged density measured by the two-colored interferometer (TCI). The reconstruction procedure is accelerated by deploying a multi-layer perceptron to run in approximately 120 µ s and is sufficiently fast for real-time control. A proportional-integral controller was adopted, with the controller gains estimated from the system identification procedure. The experimental results demonstrate that the developed controller can follow a dynamic target while exclusively using both actuators. The absolute percentage errors between the electron density at ψ N = 0.89 and the target were approximately 1.5% median and a 2.5% average, respectively. The developed controller can even lower the density by using the pump-out mechanism under RMP, and it can follow a more dynamic range of density targets than a single actuator controller. The developed controller will enable experimental scenario exploration within a shot by dynamically setting the density target or maintaining a constant electron density within a discharge.

Edge localized mode (ELM) pacing via vertical plasma oscillations or jogging has been successfully demonstrated on DIII-D. Rapid vertical movement of the plasma toward the X-point has been shown to effectively trigger ELMs. By vertically oscillating the plasma at a rate of 20 Hz, the ELM frequency increased from ∼5 Hz, the natural ELM frequency in similar DIII-D discharges, to 20 Hz. Downward jogs have been observed to trigger multiple ELMs in one cycle. ELMs triggered at higher than natural frequencies lead to smaller decreases in stored energy, from 10% to as little as below 1%. As a consequence, the peak heat flux to the divertor has been observed to be reduced by a factor of ∼2. In addition, a reduction in the carbon impurity concentration has been observed. During downward jogs in the lower single null (LSN) configuration, the X-point movement is slower and smaller than the top of the plasma. As a result, a reduction in the plasma cross section and hence volume has been observed. To understand the mechanism of ELM triggering by jogging, a toy model of the edge toroidal current has been built and tested with DIII-D experiment data. The experimental data and model suggest that when the plasma moves down towards the X-point, a net positive toroidal current is locally induced in the edge region. ELITE stability analysis suggests that this current pushes the plasma state across the peeling side of the peeling-ballooning stability boundary into the unstable region triggering ELMs.

Real-time and adaptive plasma control is crucial for robust tokamak operation, requiring sensitivity and tolerance measurements of the plasma state. This paper presents the implementation of an integrated realtime plasma monitoring framework on the DIII-D tokamak to support advanced control approaches, including machine-learning (ML) methods. The system is built on the SHIELD framework, a high-performance modular architecture that provides a unified pipeline for integrating diverse diagnostics. The framework leverages high-bandwidth digitizers, fast numerical processing, and deterministic, low-latency interconnects to stream high-fidelity data from diagnostics such as electron cyclotron emission (ECE), beam emission spectroscopy (BES), CO2 interferometers, and a visible tangential divertor camera (TangTV). The system's validity is demonstrated through direct comparisons of real-time and offline data. Furthermore, we present two key applications of the developed plasma monitoring system with ML-based plasma control strategies, including real-time divertor detachment and active Alfv & eacute;n Eigenmode control. This work presents a robust and scalable approach for integrating high-frequency, multidimensional diagnostics into advanced control algorithms for future fusion devices.

Understanding complex physical systems often requires integrating data from multiple diagnostics, each with limited resolution or coverage. We present a machine learning framework that reconstructs synthetic high-temporal-resolution data for a target diagnostic using information from other diagnostics, without direct target measurements during the inference. This multimodal super-resolution technique improves diagnostic robustness and enables monitoring even in case of measurement failures or degradation. Applied to fusion plasmas, our method targets edge-localized modes (ELMs), which can damage plasma-facing materials. By reconstructing super-resolution Thomson Scattering data from complementary diagnostics, we uncover fine-scale plasma dynamics and validate the role of resonant magnetic perturbations (RMPs) in ELM suppression through magnetic island formation. The approach provides new observation supporting the plasma profile flattening due to these islands. Our results demonstrate the framework's ability to generate high-fidelity synthetic diagnostics, offering a powerful tool for ELM control development in future reactors like ITER. The approach is broadly transferable to other domains facing sparse, incomplete, or degraded diagnostic data, opening new avenues for discovery.

Artificial intelligence models have demonstrated strong predictive capabilities for various instabilities in fusion devices such as Tokamaks, including tearing modes (TM), edge localized modes, and disruptive events, but their opaque nature raises concerns about safety and trustworthiness when applied to fusion power plants. Here, we present a physics-based interpretation framework using a TM prediction model as a demonstration that is validated through a dedicated DIII-D TM avoidance experiment. By applying Shapley analysis, we identify how profiles such as rotation, temperature, and density contribute to the model's prediction of TM stability. Our analysis shows that in our experimental scenario, core electron temperature and rotation peaking play the primary role in TM stability, while density changes have smaller effects on stability. We show that off-axis ion temperature stabilizes TMs, suggesting that off-axis neutral beam heating can further stabilize this scenario. This work presents a generalizable ML-based event prediction methodology, from training to physics-driven interpretation, bridging the gap between physics understanding and opaque ML models.

The near-axis description of optimised stellarator fields has proven to be a powerful tool both for the design and understanding of this magnetic confinement concept. The description consists of an asymptotic model of the equilibrium in the distance from its centremost axis, and is thus only approximate. Any practical application therefore requires the eventual construction of a global equilibrium. This paper presents a novel way of constructing global equilibria using the DESC code that guarantees the correct asymptotic behaviour imposed by a given near-axis construction. The theoretical underpinnings of this construction are carefully presented, and benchmarking examples provided. This opens the door to an efficient coupling of the near-axis framework and that of global equilibria for future optimisation efforts.

How close are neural networks to the best they could possibly do? Standard benchmarks cannot answer this because they lack access to the true posterior p(y|x). We use class-conditional normalizing flows as oracles that make exact posteriors tractable on realistic images (AFHQ, ImageNet). This enables five lines of investigation. Scaling laws: Prediction error decomposes into irreducible aleatoric uncertainty and reducible epistemic error; the epistemic component follows a power law in dataset size, continuing to shrink even when total loss plateaus. Limits of learning: The aleatoric floor is exactly measurable, and architectures differ markedly in how they approach it: ResNets exhibit clean power-law scaling while Vision Transformers stall in low-data regimes. Soft labels: Oracle posteriors contain learnable structure beyond class labels: training with exact posteriors outperforms hard labels and yields near-perfect calibration. Distribution shift: The oracle computes exact KL divergence of controlled perturbations, revealing that shift type matters more than shift magnitude: class imbalance barely affects accuracy at divergence values where input noise causes catastrophic degradation. Active learning: Exact epistemic uncertainty distinguishes genuinely informative samples from inherently ambiguous ones, improving sample efficiency. Our framework reveals that standard metrics hide ongoing learning, mask architectural differences, and cannot diagnose the nature of distribution shift.

We present the design and application of a general algorithm for Prediction And Control using MAchiNe learning (PACMAN) in DIII-D. Machine learning (ML)-based predictors and controllers have shown great promise in achieving regimes in which traditional controllers fail, such as tearing mode (TM) free scenarios, ELM-free scenarios and stable advanced tokamak conditions. The architecture presented here was deployed on DIII-D to facilitate the end-to-end implementation of advanced control experiments, from diagnostic processing to final actuation commands. This paper describes the detailed design of the algorithm and explains the motivation behind each design point. We also describe several successful ML control experiments in DIII-D using this algorithm, including a reinforcement learning controller targeting advanced non-inductive plasmas, a wide-pedestal quiescent H-mode ELM predictor, an Alfv & eacute;n Eigenmode controller, a Model Predictive Control plasma profile controller and a state-machine TM predictor-controller. There is also discussion on guiding principles for real-time ML controller design and implementation.