AERO / HEAT-EXCHANGER BENCHMARK PROGRAM
EVIDENCE RELEASE COMPLETE / VALIDATION INCOMPLETETwo evidence slices. The gates stay honest.
This report distinguishes actual runs from candidates that stopped before execution. HX-S01 is a source-informed synthetic PCHE study with numerical gates that failed. HX-B01 adds a separately disclosed analytical comparison that also failed its revealed correlation screen. Neither is a claim of validated heat-exchanger performance.
HX-S01: 3-grid CHT numerical FAIL · HX-B01: −30.86% correlation screen
Next maturity pass: failure → truth → transfer
The program now has a clear next sequence. HX-S01 remains the retained numerical failure. HX-02 is an input-only, publication-grounded two-stream plate-and-shell benchmark selected for thermal-hydraulic truth; its result reveal is not authorized yet. HX-03 is a separate, employer-independent high-temperature transfer case whose first-order screen precedes any justified CFD or FEA.
HX-B01 scoped analytical comparison
After two preserved Luna feasibility STOPs, Aero ran a narrower, source-grounded Seo et al. straight-microchannel screen. The visible packet contained geometry, water properties, and Re = 300; the fitted source correlation and outcome material were revealed only after the prediction freeze. The implementing agent had seen the public article, so this is not claimed as a fully human-custodian blind experiment.
| Quantity | Frozen analytical value | Revealed comparison |
|---|---|---|
| Hydraulic diameter reconciliation | 0.1963% difference | PASS vs ≤1% screen |
| Fanning friction factor | 0.0533333 | 0.0771347 from source correlation |
| Channel pressure drop | 2,810.20 Pa | 4,064.32 Pa, same channel-only definition |
| Relative error | — | FAIL −30.86% underprediction |
The source correlation is reported over Re = 100–850 with an approximately ±8% experiment/correlation envelope, so this screen is outside that envelope. The paper does not provide a verified machine-readable point for a pointwise test here, and its apparatus pressure drop includes port effects not represented by this channel-only comparison. Pointwise experiment validation remains NOT ESTABLISHED; CFD and FEA were not run by scope.
Synthetic companions for the remaining paths
The publication candidates B02–B05 remain source/independent-truth gated. Rather than invent publication inputs, Aero runs a separate synthetic suite that exercises the same capability intents with every assumption disclosed.
| Companion | Capability exercised | Status |
|---|---|---|
| HX-S02 | Coupled transfer and structural-contract handling | COMPLETE · PCHE-specific CFD/FEA not established |
| HX-S03 | 1-D → 2-D → 3-D fidelity selection | COMPLETE · synthetic surrogate |
| HX-S04 | 81-case geometry/thermal sensitivity sweep | COMPLETE · synthetic equations |
| HX-S05 | Architecture/material trade with retained alternatives | COMPLETE · preliminary only |
| HX-D01 | New design-transfer requirement with controlled escalation | COMPLETE · no solver claim |
These companions use public papers only for methodology context. They do not use publication results or author rankings and do not establish validation, qualification, or design readiness.
Numerical failure is not the same as a failed run
All three OpenFOAM Foundation 10 CHT grids reached the original 250-iteration horizon. Mass and energy balances met their stated bounds, but every grid missed the predeclared residual limit and the medium-to-fine pressure-drop changes exceeded 5%. A solver process ending normally does not establish convergence.
Not established: experimental agreement, PCHE-specific structural performance, design readiness, or autonomous local-model piloting.
HX-S01 frozen 250-iteration baseline
| Grid | Cells | Max residual | Mass imbalance | Energy imbalance | Hot / cold Delta-p | Disposition |
|---|---|---|---|---|---|---|
| Coarse | 494,670 | 2.1375e-4 | 0.0669% | 3.671% | 1,319.4 / 1,340.9 Pa | FAIL residual |
| Medium | 1,003,677 | 5.7688e-6 | 0.0896% | 2.492% | 1,393.7 / 1,407.9 Pa | FAIL residual |
| Fine | 2,770,939 | 6.6989e-6 | 0.0624% | 1.496% | 1,800.9 / 1,808.9 Pa | FAIL residual |
Medium-to-fine pressure-drop changes were 22.61% hot-side and 22.17% cold-side. Same-settings 1,000-iteration continuations also miss residual convergence; fine-to-medium pressure-drop differences remain 28.7% and 28.1%. Those continuations are diagnostic and do not replace the frozen baseline.
Post-baseline diagnostic: stable solve still not established
After the frozen 250-iteration baseline failed, Aero ran separately labeled, non-blind coarse-grid diagnostics. LTOL-003 tightened the copied linear solves and retained every checkpoint from iterations 1,100 through 1,600. It ended normally and passed residual and conservation checks, but the engineering monitors still moved too much to declare a stationary solution.
| LTOL-003 gate | Observed | Disposition |
|---|---|---|
| Solver end | NORMAL_END | PASS |
| Final equation residual | 5.8496e-9 | PASS vs 1e-6 |
| Mass / energy balance | 0.067% / 2.181% | PASS |
| Pressure-drop / duty stationarity | Cold Δp 1.81–4.42%; duties 0.79–0.89% per 25 steps | FAIL vs 0.5% |
LTOL-002 also reached 1,600 normally, but output purging removed two required checkpoints, so its full stability gate is reported as NOT EVALUABLE. The medium/fine diagnostic follow-up was not run. These runs are synthetic diagnostics—not experimental validation, structural proof, or design readiness.
Study dispositions
| Study | Status | What was established |
|---|---|---|
| HX-B01 | SCOPED ANALYTICAL COMPARISON COMPLETE POINTWISE EXPERIMENT NOT ESTABLISHED | Two frozen input-only Luna STOPs are preserved; the scoped Seo analytical comparison failed its revealed correlation screen, with pointwise experiment, CFD, and FEA unestablished. |
| HX-B02 | SOURCE AND STRUCTURAL TRUTH GATE OPEN | Lawful full-text inputs and independent structural truth not established; no run. |
| HX-B03 | ACCESS AND VERSION GATE OPEN | Version-matched full methods/input source and independent structural truth not established; no run. |
| HX-B04 | FULL TEXT AND REPRODUCIBILITY GATE OPEN | Complete lawful inputs and independent structural truth not established; no run. |
| HX-B05 | STOP BEFORE PREREGISTRATION | Luna feasibility plan stopped before preregistration; no ranking or solver run. |
| HX-D01 | NOT STARTED ORDER GATE OPEN | Not started; predecessor disposition gate remains open. |
Why HX-S01 is synthetic
Samarmad and Jaffal (2023) grounds the PCHE configuration context; Seo et al. (2015) informs straight-microchannel methodology; NIST water correlations ground the nominal fluid-property range. The reconstructed solver condition had no verified, matched pointwise experimental target. Conflicting cold-inlet values and incomplete structural data are documented rather than silently filled with purported measurements.
The retained first-principles screen gives hydraulic diameter 1.52754 mm, Reynolds number 519.12, straight-duct pressure drop 576.20 Pa per channel, and a limited 27.46-30.90 W idealized heat-duty band per channel pair. These are analytical screening outputs, not experimental targets or uncertainty bounds.
The full report lists geometry, property and model assumptions; frozen gates; baseline and diagnostic results; custody limitations; and references. No source article PDF, figure, or table is redistributed here.
What remains before program closure
- Resolve or formally replace HX-B01 with matched geometry, measurements, and uncertainty; keep both previous stop plans intact.
- Close the source-access/reproducibility and independent structural-truth gates for HX-B02 through HX-B04.
- Supply HX-B05's missing requirements and verified material/property/geometry inputs before any new ranking attempt.
- Only then open HX-D01, with a broad-requirements stage and controlled fidelity escalation.
No result here establishes employer-hardware suitability, qualification, or design release. Hashes identify artifacts; they do not establish scientific validity or trusted time.