Skip to content
Technology How it works Breeder — Hyperion Burner — Aegis Burner — MetroVolt AI-Native Architecture Magnets Fuel cycle Safety Roadmap
Solutions AI & Data Centers Defense & Government Grid & Baseload Neutron Detection Quantum
Learn Technical Library
Proof Publications Whitepapers Technical Library Open Science & Reproducibility The Honest Gates
Company About / Mission Leadership Environment Health & Safety Investors Careers Press Contact
3D Model
AI Architecture › L3 · Twin Modeling & AI
L3 · Twin Modeling & AI

L3 Twin Modeling and AI: Architecture Overview

Layer 3 is where Kronos turns validated telemetry into predictive physics: GNNs, PINNs, anomaly ensembles, MPC agents, and the KRONOS-CTRL digital twin.

THE STACK · click to jumpL7Ecosystem & StrategyL6Experience & VisualizationL5Applications & CopilotsL4OrchestrationL3Twin Modeling & AIL2Data FabricL1Control PlaneL0Foundation▲tlmctl▼L3 · TWIN MODELING & AIThe KRONOS-CTRL digital twin and its predictive shadow.1KRONOS-CTRL Twinlive plant state2GNNscoupled subsystems3PINNsphysics-constrained4Anomaly Ensemblesdrift & fault detection5MPCreceding-horizon control6Predictive Shadowruns seconds aheadMACHINE TIEState estimate descends to L1 control; alerts rise to L4 / L5.KRONOS FUSION ENERGYAI-NATIVE S.M.A.R.T. GENERATORTWIN MODELING & AISHEET 05REV. 2026-08L3 · AI-NATIVE STACK
L3 · Twin Modeling & AI — its place in the stack (left, click any layer) and its internal components (right). Telemetry rises; control descends.

Where L3 sits in the stack

Layer 3 (L3) consumes the clean, spatially normalized feature streams published by the L2 Data Fabric and returns physics-grade state estimates, forecasts, and actuation intents to the L1 Control Plane. It never touches the microsecond hard-real-time path directly: the L1 edge FPGAs and the autonomous hardware failsafe remain authoritative. L3 operates one tier up, in the 1-100 ms regime, producing the model of the plant that L1 executes against.

Five capabilities compose L3. Graph neural networks represent the diagnostic constellation as a dynamic graph and impute dropped or degraded signals. Physics-informed neural networks solve the Grad-Shafranov equilibrium and MHD stability natively, without a mesh solver in the loop. Anomaly-detection ensembles surface sub-threshold quench and disruption precursors. Model-predictive-control agents plan actuation inside a certified safe operating envelope. The KRONOS-CTRL digital twin binds these into a coupled Power / Neutronics / Thermomechanics / MHD state that runs a 50-100 ms predictive shadow ahead of the real plant.

Both machines, one architecture

L3 is instantiated twice from the same framework. For the breeder (Hyperion), a D-T spherical tokamak with Q_sci 3.076, 85.0 MW fusion power, 9.66 MA plasma current, 16.84 T peak field and negative triangularity delta -0.30, the hard problems are equilibrium reconstruction, the ELM-free negative-triangularity shape, and disruption avoidance. For the burner (Aegis / MetroVolt), a D-3He tandem-mirror generator with a 26.49 T plug and 17 T throat, the hard problems are end-plug density, the ambipolar potential, and direct energy conversion of the 5.44%-neutron-fraction output.

These machines are design and simulation studies. Breeder construction begins Q2 2027 with first-of-a-kind first tritium targeted ~2030; no hardware net-gain is claimed before FOAK. L3 is therefore validated today against multi-physics simulation and legacy device data, with a defined path to plant synchronization once hardware exists.

Content reviewed August 2026 · design-and-simulation stage