Computing Library › AI Plasma Control
AI Plasma Control

Disruption Avoidance

Acting on early warning to steer the plasma away from a disruption while keeping the discharge alive.

Avoid before you mitigate

Mitigation limits the damage of a disruption; avoidance prevents it. Avoidance is always preferable because it keeps the plasma useful. When the predictor and physics indicators show the plasma approaching a limit, the avoidance logic applies controlled changes that increase margin without terminating the discharge.

Levers of avoidance

Kronos motion — early warning

Graded response

Avoidance is graduated. Early, gentle warning triggers small corrections that a human would barely notice. As proximity grows, the response escalates: larger corrections, then a controlled ramp-down, and only if avoidance fails does the stack hand off to mitigation. Each escalation step is pre-validated so it does not itself create a new problem.

The trade against false alarms

Aggressive avoidance that fires on weak signals wastes discharges and erodes trust; timid avoidance that waits for certainty acts too late. Tuning this trade - the operating point on the alarm-versus-miss curve - is a core design decision, made explicit rather than buried in a threshold.

Where avoidance meets scenario control

Avoidance actions are implemented through the same actuator-management and scenario layers as normal control, using pre-validated alternate branches. This reuse means an avoidance maneuver is a known, tested trajectory, not an improvised command issued in a crisis.