April 4-6

Abstract Details

files Add files


Author: Cihan Akcay
Requested Type: Poster
Submitted: 2022-02-28 18:58:59

Co-authors: E. Olofsson, J. M. Finn, D. P. Brennan

Contact Info:
General Atomics
3550 General Atomics Court
San Diego, California   92122
United States

Abstract Text:
We have extended the mode-locking model due to error fields, which was previously based on a 3rd order ODE[1]. The extension includes the effects of a resistive wall in addition to an error field. This modification leads to a 5th order ODE system, in which the resistive plasma - resistive wall stability index t(rw) replaces the resistive plasma - ideal wall index t in the third order system. The new model also includes a term that represents the quasilinear saturation of the tearing perturbation by finite island width. We choose the error field and the externally applied torque as the control parameters to be varied. The order parameters are the time-asymptotic values of the dependent variables of the 5th order ODE system, which reduce to two order parameters, independent of the control parameters, after a certain normalization. As with the third order system of Ref. [1], first, clustering algorithms are used to classify the solutions into locked and unlocked states, facilitated by the aforementioned parameter normalization. This classification splits the 2D control parameter space into three regions featuring: (I) only locked states, (II) only unlocked states, and (III) both locked and unlocked states co-existing in a region of hysteresis. The second step in our machine learning approach is to use the locked/unlocked classifications to determine the probability of locking in the hysteresis region. Locking probabilities from various stages of disrupting DIII-D discharges are also presented.

* Work supported by US DOE under DE-FG02-95ER54309, DE-FC02-04ER54698, and DE-SC0022031.

[1] C. Akcay, J. M. Finn, D. P. Brennan, T. Burr, and D. M. Kurkcuoglu, "Machine learning methods for probabilistic locked-mode predictors in tokamak plasmas", Phys. Plasmas 28, 082106 (2021).