Disruption Avoidance and Mitigation
Turning a disruption prediction into an action: steering the plasma away from the boundary or safely shutting it down.
Prediction is not the goal
A disruptivity alarm is only useful if it triggers an effective response. Avoidance means adjusting actuators to move the plasma back into a safe operating region; mitigation means, when avoidance fails, dissipating stored energy in a controlled way to protect the machine.
Avoidance strategies
- Reduce plasma current or density before a limit is reached
- Adjust heating and current drive to suppress tearing modes
- Modify shape or position to restore vertical stability
- Inject controlled amounts of gas to manage impurity content
Mitigation strategies
When a disruption is unavoidable, massive gas injection or shattered-pellet injection radiates thermal energy over the whole wall rather than a hot spot, and raises density to suppress runaway electrons. The trigger timing is a control decision informed by the predictor.
The ML role
ML contributes the state estimate and the disruptivity forecast, and can learn a policy mapping states to actuator commands. Because the consequences of a bad action are severe, avoidance policies are usually validated in simulation and against historical data before any closed-loop deployment, and constrained to physically safe command ranges.
Design context
Avoidance and mitigation are studied for any tokamak concept, including spherical tokamaks. For the Kronos breeder design (Hyperion), disruption handling is part of the modeling and control-study scope; the device is not built, so all such work is simulation-based. Mirror configurations such as the Kronos burner have a different stability problem set and do not disrupt in the tokamak sense.
Good avoidance shrinks the number of times mitigation is needed, and good mitigation bounds the damage when avoidance fails.