diff --git a/docs/hydrogen-tank.html b/docs/hydrogen-tank.html new file mode 100644 index 0000000..e4a4ad9 --- /dev/null +++ b/docs/hydrogen-tank.html @@ -0,0 +1,306 @@ + + + + + +Hydrogen Tank SHM — Design Note + + + + + +
+ ← Roadmap +
+
GNN-SHM · Design Note
+

Hydrogen Tank SHM (LH2)

+

Extending the established H3 fairing GNN-SHM stack to the cryogenic hydrogen tank. The damage physics differ from the fairing, but the pipeline transfers with a small, targeted set of additions.

+
Status: Design phase — not yet implemented. Numbers below are design assumptions (require FEM / literature verification).
+
+ +
+
01 — Geometry
+

Geometry

+
+ + + + + + + + + +
Table 1 — Modeling assumptions
ItemDesign assumptionNote
TargetH3 upper-stage LH2 tank — cylindrical shell + domesExact dimensions non-public → representative values
Diameterφ ≈ 5.2 mRepresentative upper-structure diameter
ModelSymmetric sector (1/6–1/12) + circumferential symmetry BCSame approach as the fairing generator
RegionsBarrel + dome + weld landWeld lines modeled explicitly (defect-prone)
+
+

Reuses the sector generation, symmetry BCs, and mesh control logic from generate_fairing_dataset.py / generate_realistic_fairing.py.

+
+ +
+
02 — Materials
+

Cryogenic material shift

+

Primary target: Al-Li alloy (H-IIA/H3 tank heritage, e.g. 2219 / 2195). The essence is the property shift from room temperature (RT) to liquid-hydrogen temperature (LH2, −253 °C / 20 K).

+
+ + + + + + + + + +
Table 2 — RT → cryogenic property shift (design assumption)
PropertyRT (design)Cryo (LH2, design)Trend
Young's modulus E~70–78 GPa+5–15 %Increase
Yield strength σy~380–450 MPa+15–30 %Increase
Elongation~10–12 %DecreaseEmbrittlement
CTE α~23×10⁻⁶/℃Integrated CTE drops (use contraction)Decrease
+
+
⚠ Design assumption. Ranges are from literature and must be fixed against material data (JAXA / references) before implementation. Future line — CFRP composite cryotank — swaps the material model (microcracking + hydrogen permeability).
+
+ +
+
03 — Loading
+

Loading

+

Unlike the fairing (thermal CTE + static), the tank superimposes:

+ +

Static analysis (defect stress concentration) and GW dynamic analysis (sensor time-histories) run in parallel — the same two-branch structure as the fairing.

+
+ +
+
04 — Defect models
+

Defect models

+
+ + + + + + + + + +
Table 3 — Defect physics & modeling
DefectPhysicsModelingFairing analog
Weld flawPorosity / lack-of-fusion / cracks on weld lineStiffness/continuity loss on weld line (element weakening or cohesive)— (new)
Thermal-cycle microcrackMicro-cracking from cryogenic cyclingLocal stiffness loss + partial contact discontinuitySimilar to delamination
Insulation debondFoam / MLI insulation separationInterface cohesive degradationSame type as skin-core debond
H-embrittlementToughness loss from hydrogen embrittlementMaterial toughness reduction (sensitivity study)Handled on the UQ side
+
+

The cohesive/CZM implementations in generate_cohesive_fairing.py / generate_czm_sector12.py apply directly to insulation debond and weld-flaw interfaces.

+
+ +
+
05 — Graph & features
+

Graph & features — deltas only

+

The existing static graph uses 34-dim node features (build_graph.py). For the tank, keep the base schema and add only a few dimensions:

+ +
+ Design principle. Do not create a new schema — add minimal deltas to the existing 34 dims, so train.py / models.py / DA / conformal connect with no modification. Fiber-orientation dims are zero-filled for isotropic Al-Li (revived for the CFRP cryotank). +
+
+ +
+
06 — Reuse map
+

Reuse map

+
+ + + + + + + + + + + + + +
Table 4 — What connects, and how much changes
Existing assetRole for the tankChange
build_graph.py / build_gw_graph.pyGraph construction+few feature dims only
train.py / train_gw.pyTrainingNo change (swap --data_dir)
models.py (GAT/GCN/GIN/SAGE …)ModelsNo change
domain_adapt.py / payload_da_gw.pyAmbient-test → cryo-operation sim2realNo change (pass X matrix)
OGW conformalLeak-risk decision with FPR guaranteeRe-set threshold / cost
fairing_stage2.pyStage-2 characterization (crack size / leak rate)Tank-specific labels
Temperature-robustness frameworkCryo = extreme operating pointExtend operating point to 20 K
pce_driver.py / reliabilityUQ for embrittlement / toughnessTank-specific uncertainty variables
+
+

The only substantially new code is the FEM generator (see below).

+
+ +
+
07 — One-sample plan
+

How the tank line starts

+

Per project practice — validate one sample first, then batch.

+
+
+

FEM generation

+

New src/generate_cryotank_dataset.py (based on the fairing generator): Al-Li cylinder with internal pressure + cryogenic thermal stress, weld-line and thermal-cycle crack defects. Healthy, 1 sample.

+
+

ODB extraction

+

Reuse extract_odb_results.py as-is.

+
+

Graph & visual check

+

Reuse build_graph.py; eyeball one graph (feature distributions, defect labels) before scaling.

+
+

Batch → same flow

+

If OK, expand the DOE → run_batch.py, then the identical fairing flow (train → DA → conformal → Stage-2).

+
+ +

Milestones

+
+
M1

Healthy 1-sample generation → graph visual OK (this note's validation).

+
M2

One sample per defect type → verify physical plausibility of stress concentration / waveform change.

+
M3

DOE batch (N ≈ 100) → detection baseline with train.py.

+
M4

Ambient ↔ cryo domain adaptation + conformal → leak-risk FPR guarantee.

+
M5

Stage-2 characterization (crack size / leak rate) → go/no-go prognosis.

+
+
+ +
+
08 — Open items
+

Risks & open items

+ +
+ + +
+ + + diff --git a/docs/index.html b/docs/index.html index 54a27c1..8c4c3ba 100644 --- a/docs/index.html +++ b/docs/index.html @@ -311,6 +311,7 @@

Training

sim2real + guarantee

Bridge ambient test data → cryogenic operation with the existing DA toolkit; wrap leak-risk decisions in split-conformal FPR guarantees.

+

Read the full design note →