Development and preliminary calibration of an off-normal warning system for SPARC

This work explores the development and preliminary calibration of an off-normal warning system for SPARC, the aim of which is to minimize disruption loads and maximize operation time via the detection, interpretation, and pacification (i.e. avoidance and mitigation) of anomalous events. The detection and interpretation of this system will be facilitated via both physics-based warning thresholds as well as machine learning-based Proximity-to-Instability Algorithms, while the pacification will be handled by equilibrium steering, soft-landings, and the disruption mitigation system. The implementation of the system will initially focus on developing physics-based warnings, which are expected to be more reliable than ML-based alternatives early in operation and can provide more interpretable results to use in pulse-planning. The preliminary calibration of these warnings will be performed using a novel technique that trains individual warning modules targeted at specific off-normal events (e.g. impurity accumulation, vertical displacement events, locked modes, etc.) on both simulated examples of these events in a SPARC-like environment as well as events from the Alcator C-Mod database. The validation of warning modules for several events using this technique will be presented here.