Excellent reproducibility of KSTAR resonant magnetic perturbation (RMP)-driven, edge-localized mode (ELM)-crash suppression enables us to construct a database reliably for the study of ELM-crash suppression conditions. To establish a high-fidelity database, we have selected one of the frequently used RMP configurations in KSTAR, n = 1, +90° phasing RMP-coil configuration. A series of fitting curves based on edge profile diagnostics data is used for normalized electron collisionality (νe*) and plasma toroidal rotation (Vtor) at pedestal top. Since ITER is expected to employ slowly rotating, low-collisionality, high-density plasmas whose conditions are not readily accessible in the existing devices, the exploration and understanding of these two parameters would be important not only for RMP-driven, ELM-crash-suppression physics but also for the success of ITER. The data points for the ELM-crash suppression phase are in the range of 0.2 < νe,ped*< 1.1 with Zeff = 2 assumption and Vtor,ped> 40 km/s. Suppression thresholds or boundaries in νe,ped* or Vtor,ped are not confirmed in the investigated parameter ranges so far. The KSTAR database still needs additional experimental datasets in ITER-relevant conditions (νe,ped* ∼ 0.1 and low-torque low-rotation) to confirm the boundary of the ELM-crash suppression window in KSTAR and reduce the uncertainties of the RMP ELM-crash control technique in ITER. In both νe,ped* and Vtor,ped parameters, the phase-space distribution of the ELM-crash suppression has no clear distinction from that of the ELM-crash mitigation. Linear discriminant analysis provides a linear combination of parameters relevant to ELM-crash suppression, best separating two data categories. Recursive feature elimination indicates that ne,ped and IRMP, as well as νe,ped* and Vtor,ped, are critical variables in the separation of the data groups.
Read full abstract