arXiv subjects

S. Pamela

Publications and source records attributed to S. Pamela.

9 recordsLinked to original sources

arXiv2025

Neural operator surrogate models of plasma edge simulations: feasibility and data efficiency

The inclusion of high-fidelity simulations of SOL turbulence and transient MHD events such as ELMs in highly iterative applications remains computationally prohibitive, limiting their use in design and control workflows. Understanding these phenomena is vital, as they govern heat flux on plasma-facing components, influencing reactor performance and material lifetime. This study explored FNOs as surrogate models to accelerate plasma simulations from the JOREK MHD and STORM turbulence codes. FNOs were trained on single-step rollouts and evaluated in terms of long-term predictive accuracy in an auto-regressive manner. To mitigate the computational burden of dataset generation, a transfer learning strategy was explored, leveraging low-fidelity simulations to improve performance on high-fidelity datasets. These results showed that FNOs effectively captured initial plasma evolution, including blob movement and density source localization. However, long rollouts accumulated errors and exhibited sensitivity to certain physical phenomena, leading to non-monotonic error spikes. Transfer learning significantly reduced errors for small dataset sizes and short rollouts, achieving an order-of-magnitude reduction when transfering from low- to high-fidelity datasets. However, its effectiveness diminished with longer rollouts and larger dataset sizes, especially when applied to datasets with significantly different dynamics. Attempts to transfer models to previously unseen variables in simulations were unsuccessful, underscoring the limitations in this context. These findings demonstrate the promise of NOs for accelerating fusion-relevant PDE simulations. However, they also highlight key challenges: improving long-term accuracy to mitigate error accumulation, capturing critical physical behaviors, and developing robust surrogates that effectively leverage multi-fidelity, multi-physics datasets.

physics.plasm-ph↗

arXiv2024

Emulation Techniques for Scenario and Classical Control Design of Tokamak Plasmas

The optimisation of scenarios and design of real-time-control in tokamaks, especially for machines still in design phase, requires a comprehensive exploration of solutions to the Grad-Shafranov (GS) equation over a high-dimensional space of plasma and coil parameters. Emulators can bypass the numerical issues in the GS equation, if a large enough library of equilibria is available. We train an ensemble of neural networks to emulate the typical shape-control targets (separatrix at midplane, X-points, divertor strike point, flux expansion, poloidal beta) as a function of plasma parameters and active coil currents for the range of plasma configurations relevant to spherical tokamaks with a super-X divertor, with percent-level accuracy. This allows a quick calculation of the classical-control shape matrices, potentially allowing real-time calculation at any point in a shot with sub-ms latency. We devise a hyperparameter sampler to select the optimal network architectures and quantify uncertainties on the model predictions. To generate the relevant training set, we devise a Markov-Chain Monte Carlo algorithm to produce large libraries of forward Grad-Shafranov solutions without the need for user intervention. The algorithm promotes equilibria with desirable properties, while avoiding parameter combinations resulting in problematic profiles or numerical issues in the integration of the GS equation.

physics.plasm-ph↗

arXiv2024

Data efficiency and long term prediction capabilities for neural operator surrogate models of core and edge plasma codes

Simulation-based plasma scenario development, optimization and control are crucial elements towards the successful deployment of next-generation experimental tokamaks and Fusion power plants. Current simulation codes require extremely intensive use of HPC resources that make them unsuitable for iterative or real time applications. Neural network based surrogate models of expensive simulators have been proposed to speed up such costly workflows. Current efforts in this direction in the Fusion community are mostly limited to point estimates of quantities of interest or simple 1D PDE models, with a few notable exceptions. While the AI literature on methods for neural PDE surrogate models is rich, performance benchmarks for Fusion-relevant 2D fields has so far remained flimited. In this work neural PDE surrogates are trained for the JOREK MHD code and the STORM scrape-off layer code using the PDEArena library (https://github.com/microsoft/pdearena). The performance of these surrogate models is investigated as a function of training set size as well as for long-term predictions. The performance of surrogate models that are trained on either one variable or multiple variables at once is also considered. It is found that surrogates that are trained on more data perform best for both long- and short-term predictions. Additionally, surrogate models trained on multiple variables achieve higher accuracy and more stable performance. Downsampling the training set in time may provide stability in the long term at the expense of the short term predictive capability, but visual inspection of the resulting fields suggests that multiple metrics should be used to evaluate performance.

physics.plasm-ph↗

arXiv2023

Efficient training sets for surrogate models of tokamak turbulence with Active Deep Ensembles

Model-based plasma scenario development lies at the heart of the design and operation of future fusion powerplants. Including turbulent transport in integrated models is essential for delivering a successful roadmap towards operation of ITER and the design of DEMO-class devices. Given the highly iterative nature of integrated models, fast machine-learning-based surrogates of turbulent transport are fundamental to fulfil the pressing need for faster simulations opening up pulse design, optimization, and flight simulator applications. A significant bottleneck is the generation of suitably large training datasets covering a large volume in parameter space, which can be prohibitively expensive to obtain for higher fidelity codes. In this work, we propose ADEPT (Active Deep Ensembles for Plasma Turbulence), a physics-informed, two-stage Active Learning strategy to ease this challenge. Active Learning queries a given model by means of an acquisition function that identifies regions where additional data would improve the surrogate model. We provide a benchmark study using available data from the literature for the QuaLiKiz quasilinear transport model. We demonstrate quantitatively that the physics-informed nature of the proposed workflow reduces the need to perform simulations in stable regions of the parameter space, resulting in significantly improved data efficiency. We show an up to a factor of 20 reduction in training dataset size needed to achieve the same performance as random sampling. We then validate the surrogates on multichannel integrated modelling of ITG-dominated JET scenarios and demonstrate that they recover the performance of QuaLiKiz to better than 10\%. This matches the performance obtained in previous work, but with two orders of magnitude fewer training data points.

physics.plasm-ph↗

arXiv2019

Non-linear modeling of the threshold between ELM mitigation and ELM suppression by Resonant Magnetic Perturbations in ASDEX Upgrade

The interaction between Edge Localized Modes (ELMs) and Resonant Magnetic Perturbations (RMPs) is modeled with the magnetohydrodynamic code JOREK using experimental parameters from ASDEX Upgrade discharges. The ELM mitigation or suppression is optimal when the amplification of both tearing and peeling-kink responses result in a better RMP penetration. The ELM mitigation or suppression is not only due to the reduction of the pressure gradient, but predominantly arises from the toroidal coupling between the ELMs and the RMP-induced mode at the plasma edge, forcing the edge modes to saturate at a low level. The bifurcation from ELM mitigation to ELM suppression is observed when the RMP amplitude is increased. ELM mitigation is characterized by rotating modes at the edge, while the mode locking to RMPs is induced by the resonant braking of the electron perpendicular flow in the ELM suppression regime.

physics.plasm-ph↗

arXiv2016

Non-linear modeling of the plasma response to RMPs in ASDEX Upgrade

The plasma response to Resonant Magnetic Perturbations (RMPs) in ASDEX Upgrade is modeled with the non-linear resistive MHD code JOREK, using input profiles that match those of the experiments as closely as possible. The RMP configuration for which Edge Localized Modes are best mitigated in experiments is related to the largest edge kink response observed near the X-point in modeling. On the edge resonant surfaces $q = m/n$, the coupling between the m + 2 kink component and the m resonant component is found to induce the amplification of the resonant magnetic perturbation. The ergodicity and the 3D-displacement near the X-point induced by the resonant amplification can only partly explain the density pumpout observed in experiments.

physics.plasm-ph↗

arXiv2014

Effect of resonant magnetic perturbations on low collisionality discharges in MAST and a comparison with ASDEX Upgrade

Sustained ELM mitigation has been achieved on MAST and AUG using RMPs with a range of toroidal mode numbers over a wide region of low to medium collisionality discharges. The ELM energy loss and peak heat loads at the divertor targets have been reduced. The ELM mitigation phase is typically associated with a drop in plasma density and overall stored energy. In one particular scenario on MAST, by carefully adjusting the fuelling it has been possible to counteract the drop in density and to produce plasmas with mitigated ELMs, reduced peak divertor heat flux and with minimal degradation in pedestal height and confined energy. While the applied resonant magnetic perturbation field can be a good indicator for the onset of ELM mitigation on MAST and AUG there are some cases where this is not the case and which clearly emphasise the need to take into account the plasma response to the applied perturbations. The plasma response calculations show that the increase in ELM frequency is correlated with the size of the edge peeling-tearing like response of the plasma and the distortions of the plasma boundary in the X-point region.

physics.plasm-ph↗

arXiv2013

Current understanding of the processes underlying the triggering of and energy loss associated with type I ELMs

The type I ELMy H-mode is the baseline operating scenario for ITER. While it is known that the type I ELM ultimately results from the peeling-ballooning instability, there is growing experimental evidence that a mode grows up before the ELM crash that may modify the edge plasma, which then leads to the ELM event due to the peeling-ballooning mode. The triggered mode results in the release of a large number of particles and energy from the core plasma but the precise mechanism by which these losses occur is still not fully understood and hence makes predictions for future devices uncertain. Our current understanding of the processes that trigger type I ELMs and the size of the resultant energy loss are reviewed and compared to experimental data and ideas for further development are discussed.

physics.plasm-ph↗

arXiv2013

Towards understanding edge localised mode mitigation by resonant magnetic perturbations in MAST

Type-I Edge Localised Modes (ELMs) have been mitigated in MAST through the application of n = 3, 4 and 6 resonant magnetic perturbations (RMPs). For each toroidal mode number of the non-axisymmetric applied fields, the frequency of the ELMs has been increased significantly, and the peak heat flux on the divertor plates reduced commensurately. This increase in ELM frequency occurs despite a significant drop in the edge pressure gradient, which would be expected to stabilise the peeling-ballooning modes thought to be responsible for type-I ELMs. Various mechanisms which could cause a destabilisation of the peeling-ballooning modes are presented, including pedestal widening, plasma rotation braking, three dimensional corrugation of the plasma boundary and the existence of radially extended lobe structures near to the X-point. This leads to a model aimed at resolving the apparent dichotomy of ELM control, that is to say ELM suppression occurring due to the pedestal pressure reduction below the peeling-ballooning stability boundary, whilst the reduction in pressure can also lead to ELM mitigation, which is ostensibly a destabilisation of peeling-ballooning modes. In the case of ELM mitigation, the pedestal broadening, 3d corrugation or lobes near the X-point degrade ballooning stability so much that the pedestal recovers rapidly to cross the new stability boundary at lower pressure more frequently, whilst in the case of suppression, the plasma parameters are such that the particle transport reduces the edge pressure below the stability boundary which is only mildly affected by negligible rotation braking, small edge corrugation or short, broad lobe structures.

physics.plasm-ph↗