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