Quench & Disruption Precursor Features

Sub-threshold signatures engineered from magnetics, strain, and profile telemetry give the anomaly ensembles a head start on quench and disruption.

Seeing trouble early

A quench in a REBCO magnet or a disruption in the breeder plasma is preceded by faint, sub-threshold structure in the telemetry: a growing Mirnov mode, a locking magnetic island, a strain delta drifting, a pressure profile peaking toward a stability limit. The fabric engineers these into explicit precursor features so the L3 anomaly ensembles operate on physics-shaped inputs rather than raw traces.

Breeder precursors

• Mode amplitude and rotation frequency from the Mirnov array, tracking MHD growth and locking.
• Pressure-gradient proximity to stability limits from the core pressure map.
• Edge behavior specific to the negative-triangularity ELM-free regime.
• n=1 locked-mode indicators that presage disruption at 9.66 MA.

Burner precursors

• End-plug density falling toward the confinement threshold.
• Ambipolar potential dropping, weakening the mirror's electrostatic plugging.
• Strain deltas at the 26.49 T plug coils approaching envelope.

Why engineer, not just learn

The anomaly ensembles could in principle learn from raw data, but physics-grounded precursor features make them faster, more sample-efficient, and interpretable. When an ensemble fires, an operator can see which precursor drove it. The features also give the twin's predictive shadow (running 50-100 ms ahead) something concrete to forecast.

These are the features that make the sub-10 microsecond control boundary meaningful: the fabric surfaces the precursor early enough that L1 actuation, or the autonomous failsafe, can respond. Everything here is validated in simulation for machines whose FOAK operation is expected near 2030.