Quick update on the SCPN Fusion Core:
I just shipped the canonical neuro-symbolic control demo.
https://lnkd.in/eb-vSv4r
One notebook. Zero setup. Runs in <2 min.
What it shows:
- Stochastic Petri net → verified SNN compiler (LIF + bitstream path)
- Formal topology/liveness/boundedness proofs (with SHA256 proof bundle)
- Closed-loop on *real DIII-D shot 166000* (beta-limit disruption precursor)
- Full FusionKernel digital twin evolution under SNN control
- Side-by-side: SNN vs PID vs MPC — RMSE, disruption flags, actuator commands, latency (sub-ms p95)
- Deterministic artifact export → replay with identical states/actions/proofs
Notebook (executed outputs baked in, Colab button too):
https://lnkd.in/eb-vSv4r
Full repo + RESULTS.md (honest metrics, DIII-D/SPARC validation, limitations section):
https://lnkd.in/eTJMfWC8
I’m pruning the kitchen-sink modules this week (legacy/ folder incoming) so it becomes a clean control-only package.
Would love brutal feedback:
- Does the formal verification approach look credible for real-time safety?
- Anyone at DIII-D National Fusion Facility-D / Princeton Plasma Physics Laboratory (PPPL) / ITER Organization willing to throw more shots at it?
- What’s the biggest red flag you see for actual hardware-in-loop?
AGPL, fully reproducible, happy to hop on a call or add features.
Cheers
hashtag#FusionEnergy
hashtag#NuclearFusion
hashtag#AI
hashtag#MachineLearning
hashtag#Tokamak
hashtag#PlasmaPhysics
hashtag#ControlSystems
hashtag#Neuromorphic
hashtag#SNN
hashtag#PetriNets
hashtag#OpenSource
hashtag#EnergyTech
