Spatial Normalization

Raw channels live at physical sensor positions; normalization maps them into the flux and field coordinates the twin reasons in, common across both machines.

From sensor space to physics space

A Mirnov coil reports at its poloidal angle; an interferometry chord reports a line integral along its path; a strain gauge reports at a magnet location. The twin does not reason in sensor indices — it reasons in flux coordinates for the breeder and in axial-and-radial mirror coordinates for the burner. Spatial normalization is the transform that carries each raw channel into that shared frame.

Breeder: flux coordinates

For the breeder (Hyperion), a spherical tokamak with R0 1.2 m, aspect ratio 2.5, and negative triangularity delta -0.30, normalization maps magnetics and profiles onto normalized flux (psi_N). This makes a core pressure sample comparable across shots regardless of small equilibrium shifts, and it is the natural frame for the PINN that solves Grad-Shafranov (see equilibrium telemetry).

Burner: mirror coordinates

For the burner, the tandem mirror, normalization uses axial position along the field line and radius from the axis, so end-plug density and the ambipolar potential are expressed where the confinement physics happens — at the 26.49 T plugs and the 17 T throats. This lets the same feature definition describe the plug on either end symmetrically.

Line integrals and inversions

• Chord-integrated diagnostics (interferometry) are inverted to local quantities using the reconstructed geometry.
• Positions are corrected for thermal and magnetic motion of the structure, tracked by structural-health monitoring.
• Normalized coordinates are versioned so an archived feature can always be placed back in physical space.

A single normalization layer serving both machines is what lets the same feature-engineering and twin code run on either device with only a coordinate module swapped.