# Source reconstruction and modeling decisions

## Recovered original operating point

The **current page** of `deliverables/Simulation Results_240524.one` gives:

| Item | Heat (W) | Autodesk FR-4 (°C) | Autodesk metal core (°C) |
|---|---:|---:|---:|
| D1 | 1.00 | 120.8 | 97.1 |
| Q1 | 0.05 | 107.6 | 98.0 |
| R1 | 0.45 | 109.2 | 95.3 |
| U1 | 0.10 | 94.6 | 94.5 |
| Distributed PCB | 0.40 | — | — |
| Total | **2.00** | | |

Ambient: **30°C**, sea level. The temperature table is explicitly for horizontal orientation. The four JPGs have no readable numerical scale and are not numerical validation data. “Horizontal” does not explicitly establish which board face points upward; the generated workflow supports both directions.

The distributed-power row has `0.40` in the Power column and `2.0` in Power Total. The latter is interpreted as the grand total, since 1 + .05 + .45 + .10 + .40 = 2.00. It is not an additional 2 W board heat source.

The OneNote containers include old, deleted material from an unrelated optical network terminal. Raw string searches alone mix those projects. `sources/troy-results.xml` and `sources/troy-setup.xml` contain the current pages, decoded with [onenote2xml](https://github.com/alegrigoriev/onenote2xml). The temporary parser required two macOS portability substitutions: default language ID 1033 and `cp1252` decoding instead of Windows `mbcs`. `sources/troy-results-readable.txt` preserves the table structure.

## Geometry correction

Both STEP files have identical SHA-256:

`66171eb96984af01dab4481e11873e1583c14c9162eb1391c329a033666f3c92`

The supplied CAD board thickness is **0.4103624 mm**. Both fabrication-order PDFs specify **1.6 mm**, and Troy's FR-4 material screenshot also specifies 1.6 mm. The new board solid uses the documented 1.6 mm thickness. The archived material screenshot specifies that thickness, but the actual Autodesk solved geometry is not available to confirm whether Troy corrected the thin STEP. This is a potentially material difference from the historical result, separate from package modeling. Its rounded outline and through-hole coordinates come from the supplied Gerber and Excellon files, sampled onto the mesh. Component footprints come from the supplied CAD envelopes. Small passive details and connector interiors are simplified. The connector housings are approximate backside blockers; pins, solder joints, and attached wire heat sinking are omitted.

Coordinates: X/Y are the native board plane, Z is its normal, component side +Z. The component mounting surface is Z=0; the back of the board is Z=-1.6 mm. Horizontal-up gravity is -Z, horizontal-down gravity is +Z, and vertical gravity is -Y. Changing gravity does not rotate the conductivity tensor away from the board.

## Material evidence and gaps

The embedded screenshot `sources/Simulation Setup_240524/image_10.png` gives Troy's FR-4 effective properties:

| Property | Value |
|---|---:|
| In-plane conductivity | 16.9951 W/(m K) |
| Through-plane conductivity | 0.283244 W/(m K) |
| Density | 2131.48 kg/m³ |
| Heat capacity | 559.764 J/(kg K) |
| Copper layer thickness | 0.035 mm each, two layers |
| Copper coverage assumed by Troy | 75% each |

The normal conductivity above is **visually transcribed**, because OneNote's OCR gets one digit wrong. The reconstruction baseline retains this homogenized anisotropic conductivity. It does not claim to resolve Gerber traces.

The supplied `LED board image 4.png` describes Cu / 150 µm FR-4 / Cu / 100–150 µm bond insulation / aluminum. This is not a confirmed manufacturing stackup. The metal-core model uses an explicit, editable nominal assumption in `config/baseline.json`, reduced to series through-plane and parallel in-plane conductivities. This preserves the dielectric bottleneck instead of treating the board as solid aluminum. Dielectric grade, actual thermal conductivity, layer placement, and aluminum alloy remain unconfirmed.

The original setup mentions two-resistor compact thermal models. The Q1 screenshot shows 74.95 K/W junction-to-lead and 45 K/W junction-to-case, consistent with the supplied datasheet. It does not supply a complete verified compact model for all packages. The current CFD baseline uses **effective solid package envelopes with assumed conductivity**. Reported package extrema/means are **not junction temperatures** and are not directly interchangeable with Troy's CTM temperatures. In particular, an LED junction estimate needs the correct solder-point temperature and the manufacturer's definition of thermal resistance, including whether it uses electrical or dissipated thermal power.

Radiation was mentioned in the old setup instructions, but its enablement and emissivities were not recorded. The new model explicitly includes diffuse-gray radiation through nonparticipating air; emissivity is an assumption to sweep. The remote open boundary has a fixed 30°C radiation temperature and permits inflow/outflow of air.

## Verification required before design decisions

1. All region meshes pass `checkMesh`; preserve maximum skewness and nonorthogonality in the run record.
2. Temperatures and velocities are finite; no below-ambient solid minimum beyond numerical tolerance for this heated, open-air case.
3. Total convective plus radiative heat leaving all solid/air interfaces agrees with 2 W to within 1%, after accounting for numerical convergence.
4. Package and board temperatures change by less than 0.1 K over the final 200 steady iterations; peak speed changes by less than 1% over that interval and open-boundary mass imbalance is below 0.1%. Also inspect momentum, pressure, and energy residuals. Steady iteration count is not elapsed physical time.
5. Run at least three systematically refined meshes; report changes in package maximum/mean temperatures and plume speed. Do not claim mesh independence from a single mesh or residual convergence alone.
6. Repeat with a larger ambient domain and finer radiation angular resolution; establish that open boundaries and radiation discretization do not control the answer.
7. Sweep uncertain MC dielectric conductivity, package resistance, emissivity, and orientation. Numerical precision cannot remove these input uncertainties.
8. Validate against measured board/solder-point temperatures at the same 2 W heat load, ambient, mounting, wiring, and orientation. The Autodesk results are a historical comparison, not ground truth.

The next physical-fidelity step is explicit copper topology and calibrated package/lead/solder thermal paths. The Gerbers and plated drill report are present, but the present baseline deliberately does not invent solder geometry, plating thickness, or an MC dielectric product.

## Solver provenance

- OpenCFD OpenFOAM **v2412**, official ARM64 container pinned by digest in `scripts/foam.sh`.
- `chtMultiRegionSimpleFoam`: steady, compressible, laminar buoyant conjugate heat transfer.
- Ideal-gas air, Sutherland viscosity; solid/air temperature and flux coupling. Absolute pressure is initialized hydrostatically in both cells and boundary faces. The open reduced-pressure boundary uses a hydrostatic atmospheric reference and incoming dynamic-pressure correction.
- Full anisotropic board conduction, with the directional boundary diffusivity `Anialpha`; segregated thermal coupling. The attempted implicit assembly failed heat-conservation checks and is excluded.
- `fvDOM` radiation, configurable angular resolution; radiative energy coupled at the fluid/solid interface.
- CAD envelope extraction: [Gmsh/OpenCASCADE](https://gmsh.info/doc/texinfo/). Production meshing: native graded orthogonal hexahedra written by scripts/cartesian_mesh.py, then split into eight regions with OpenFOAM. Curved edges and sub-cell holes are approximated.
- Official [CHT solver documentation](https://doc.openfoam.com/2212/tools/processing/solvers/rtm/heat-transfer/chtMultiRegionFoam/) describes the transient member of the same solver family; executable behavior and dictionary syntax here were checked against the installed v2412 source and tutorials.

## Startup equilibrium check

With all heat sources disabled and radiation disabled, the 40-iteration hydrostatic check remained at 303.15 K to printed precision, with maximum air speed 0.0001001 m/s and net interface heat transfer below 1e-11 W. This is a pressure/startup check, not a validation of the powered radiation model. Earlier uniform absolute-pressure startup runs produced artificial flow and were rejected.