Use Cases¶
Fusion Reactor Control¶
Tokamak Control Algorithm Prototyping¶
Status: Research / Alpha. Not a production PCS. Real hardware integration requires significant additional work (EPICS/CODAC interface, deterministic OS, safety certification).
scpn-control provides a prototyping platform for control algorithm development. The 11.9 µs kernel step (Criterion-verified) enables:
- Shape control algorithm R&D — develop and test at high loop rates
- Disruption prediction prototyping — ML inference pipeline (synthetic training data)
- SPI mitigation modeling — halo current and runaway electron physics
- Profile control R&D — neuro-cybernetic dual R+Z SNN feedback
Target users: Control algorithm researchers, fusion startups in early design, graduate students.
ITER / DEMO (Future, Speculative)¶
The contract-based pre/post-condition checking layer provides runtime assertion checking — not formal theorem-proved verification. The SNN controller includes a Lyapunov stability guard, but this has not been validated against real disruption scenarios or certified for nuclear safety.
Digital Twin & Commissioning¶
Offline Plant Commissioning¶
Before first plasma, use the digital twin to:
- Commission control algorithms against synthetic plasma scenarios
- Train operators on disruption response procedures
- Validate sensor/actuator configurations
- Stress-test control logic with adversarial perturbations
The digital twin runs in real-time on a single CPU core, enabling rapid iteration without dedicated HPC resources.
Hardware-in-the-Loop Testing¶
The HIL harness (scpn-control hil-test) replays reference shot data
through the full controller pipeline, verifying end-to-end latency
and control fidelity against experimental baselines.
Research & Education¶
Neuro-Symbolic Control Research¶
scpn-control is the only open-source implementation of:
- Stochastic Petri Net to SNN compilation for fusion control
- Kuramoto-Sakaguchi phase dynamics with Lyapunov stability monitoring
- Phase-amplitude coupling (PAC) between oscillator populations
These are active research frontiers. The implementation is tested but not yet validated in peer-reviewed fusion publications.
Graduate-Level Course Material¶
The codebase includes:
- 5 tutorial notebooks (Jupyter) with step-by-step walkthroughs
- A live Streamlit dashboard for interactive exploration
- 2,417 tests (99.99% coverage) demonstrating expected behavior and edge cases
- Competitive analysis against state-of-the-art codes
Edge & Embedded Deployment¶
No GPU Required¶
The neural equilibrium kernel achieves 0.39 ms on CPU only (not head-to-head validated against P-EFIT on identical equilibria). This potentially enables deployment on:
- ARM-based edge controllers (not tested on ARM)
- Embedded systems (not tested in radiation-hard environments)
- Air-gapped control networks (no cloud dependency)
Rust Native Backend¶
The 5-crate Rust workspace provides:
- Zero-copy PyO3 bindings for Python interop
- No runtime garbage collection (deterministic latency)
- Rayon parallelism for multi-core scaling
- Criterion benchmarks for regression testing
Comparison Matrix¶
| Use Case | scpn-control | TORAX | FUSE | FreeGS |
|---|---|---|---|---|
| Real-time kernel | 11.9 µs (bare step) | No | No | No |
| Disruption prediction | Experimental (synthetic) | No | No | No |
| SPI mitigation | Experimental | No | No | No |
| Digital twin | Experimental | No | No | No |
| Neural equilibrium | 0.39 ms (CPU, not cross-validated) | No | No | No |
| SNN controller | Yes (pure LIF+NEF, mocked CI) | No | No | No |
| Contract checking | Yes (runtime assertions) | No | No | No |
| QLKNN-10D transport | Yes (trained MLP) | Yes (QLKNN10D) | No | No |
| PPO RL agent | Yes (beats MPC + PID) | No | No | No |
| Edge deployment | Possible (Rust, untested on ARM) | No (JAX) | No (Julia) | Partial |
| Autodifferentiation | Yes (JAX) | Yes (JAX) | Yes (Julia) | No |
| GPU transport | Yes (JAX) | Yes | Yes | No |
| Real tokamak data | No | Yes | Yes | Yes |
| Peer-reviewed papers | No | Yes | Yes | Yes |
Get Started¶
For commercial licensing: protoscience@anulum.li