Chapter 6
Standardized operational protocols for plasma startup, steady-state operation, shutdown, and emergency procedures ensure safe and reproducible fusion operations.
A fusion plasma shot involves carefully orchestrated sequences of events, from initial vacuum preparation through plasma termination. WIA-FUSION defines standard protocols for each phase.
The startup protocol brings the reactor from cold standby to burning plasma conditions:
Verify vacuum (<10⁻⁶ Pa), activate wall conditioning, check diagnostics
Toroidal field: 0 → 5.3 T at 1 T/s, establish poloidal shaping
Puff 10²⁰ deuterium atoms, achieve 0.1-1 Pa pressure
ECRH pre-ionization (1-2 MW), loop voltage 10-20 V, achieve Ip > 100 kA
Ramp to target current (15 MA at 1 MA/s), monitor flux consumption
Activate NBI, tune ICRH/ECRH, reach steady-state conditions
During steady-state, the AI control system maintains plasma parameters within target ranges:
STEADY-STATE CONTROL TARGETS:
├── Core Temperature: 10-15 keV (±10%)
│ └── Feedback: ECRH power modulation
├── Density: 1.0×10²⁰/m³ (±15%)
│ └── Feedback: Pellet injection rate
├── Plasma Current: 15 MA (±5%)
│ └── Feedback: Inductive + bootstrap
└── Beta: 2.5% (limit < 3.5%)
└── Feedback: Heating power reduction
AI Control Loop:
├── Cycle time: 1 ms
├── Inputs: 50+ diagnostic signals
├── Optimization: Maximize Q, minimize disruption risk
└── Actions: Heating, shape, fueling adjustments
Controlled shutdown ("soft landing") safely terminates the plasma:
Reduce all heating at 50%/minute while maintaining plasma control
Reduce plasma current at 0.5 MA/s, maintain vertical stability
Stop gas puffing and pellet injection
Ramp down toroidal and poloidal fields
Archive data, inspect components, log performance
System receives emergency trigger
Massive gas injection activated
NBI, ECRH, ICRH immediately terminated
Plasma extinguished via radiative collapse
Magnetic field dumped, facility in safe state
The AI control protocol defines how machine learning systems interact with the plasma:
AI DISRUPTION PREDICTION (AI-DISRUPT-001):
├── Latency: < 10 ms (data to action)
├── Inputs: Ip, li, β, locked modes, radiation, q95, etc.
├── Model: LSTM + Attention (50,000+ training shots)
├── Accuracy: > 95% detection, < 5% false positive
└── Thresholds:
├── GREEN (< 30%): Monitor at 100 Hz
├── YELLOW (30-70%): Alert, prepare mitigation
└── RED (> 70%): Execute mitigation immediately