Enhanced RMP hysteresis via ω_E-based adaptive ELM control via ML-accelerated real-time turbulence analysis

Utilizing a state-of-the-art feedback controller, one can adaptively modify the RMP currents depending on the plasma response to recover performance while avoiding ELMs. This controller, however, still suffers from suppression loss when the RMP current is lowered past some critical threshold­, whose value is difficult to predict beforehand due to the highly nonlinear physics governing the plasma edge. RMP ELM suppression hinges on the alignment of rational surfaces at the pedestal top, where lower E Χ B rotation frequency ω_E allows the applied perturbation to penetrate rather than being screened out. Moreover, one can exploit the phenomena of RMP hysteresis which allows for access and sustainment of suppression at lower RMP currents, to bolster and recover plasma performance [1]. Therefore, we monitor the poloidal turbulence group velocity v_θ as measured by the 2D Beam Emission Spectroscopy (BES) system [2] as a fast indicator of ω_E in order to exploit this RMP hysteresis effect. We trained a machine learning (ML) model using ground truth labels generated from the cross-correlation time delay technique to track the ion-scale turbulent eddies in RMP suppressed DIII-D plasmas to infer v_θ from raw BES data. With a real-time indication of v_θ near the pedestal top, one can preemptively inform the plasma control system to increase RMP currents to avoid suppression loss, improving and facilitating robust adaptive ELM control, crucial for reactor relevant devices.

References

[1] S. Kim et al., (2024). Nature communications, 15(1), 3990.

[2] G. McKee et al., (1999). Review of Scientific Instruments 70, 913.