DECAF multi-device database analysis of an MHD mode locking forecaster with real-time application

Tokamak reactors require low disruptivity to support commercial viability. An important precursor to disruptions is the locking dynamic of rotating MHD events that are often neoclassical tearing modes (NTM). The drag of electromagnetic and fluid viscosity torques can cause the slowing down of NTM's and lock them to the device conducting structure. A balance of the driving torque from the NBI, and drag from perpendicular viscous diffusion drag, and electromagnetic forces on the mode, as well as its inertia, are used to model the mode rotation dynamics. Prediction, observation, and avoidance of these modes are, therefore, essential in the operation of tokamaks. A mode locking forecaster based on the torque balance model has been developed across devices of ranging aspect ratio and error fields. A real-time application of the model in the KSTAR tokamak was able to achieve 100% accuracy in forecasting for a set of 50 shots of varying scenarios. Analysis of the KSTAR, NSTX, MAST-U, and DIII-D databases is conducted, and differences in performance of the forecaster are compared. Explanations are given for variation in performance and possible solutions to achieve higher accuracy given