feat: Capacity Manager: reconstruct DNV-RP-C201 stiffened-panel + girder capacity models from a Sesam SIN#233
Open
oleandor wants to merge 57 commits into
Open
feat: Capacity Manager: reconstruct DNV-RP-C201 stiffened-panel + girder capacity models from a Sesam SIN#233oleandor wants to merge 57 commits into
oleandor wants to merge 57 commits into
Conversation
…m a SIN New ada.fem.capacity subpackage: reads a Sesam .SIN, assembles panel groups (grouping seeded from a Genie model.json source for now), extracts plate/ stiffener geometry, and resolves FE membrane stresses into DNV-RP-C201 design variables. Validated against the Mini-topside reference: plate geometry 354/354 vs model.json; transverse-membrane + shear stresses exact (median rel 1e-6 over 540 cases); axial reproduces Genie's AxialLoads decomposition (~8% residual). Emits a neutral CapacityModel and a Genie-compatible model.json mirror. Abstract CapacityModelBuilder leaves a girder seam. 5 tests (skip-if-no-reference). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
A Sesam SIN set's GSETMEMB block can carry an uninitialised cap/dim slot (~1e8 on the Ruben 399 MB topside) that tripped the _MAX_DIM_VALUE guard; the PTAB decode loop then swallowed the ValueError and dropped the whole block, so mesh.sets came back without those sets and named-group scoping failed. The pointer table is authoritative — _read_pointer_table already clamps the read to the file size and truncates at the first invalid pointer — so cap only the requested slot count instead of rejecting the block. All 104 Ruben sets now resolve (allPlates=58809 shells, Pl_T100=2372 plates, etc.). Also give a clear error when a SIN genuinely has no set membership. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Capacity reconstruction scanned every element block (np.where per call, O(n_elements)) for each node/coords/tributary lookup, making panel assembly O(n^2) — ~23 h on a 128k-element topside. Build a node->element / element->node index once (extract._ensure_index, cached on the mesh): the same lookups are now O(1) and tributary is O(node degree). Tributary for all 69k beams drops from a ~23 h extrapolation to ~0.4 s; the Mini suite runs ~2x faster with identical results. Also: - scope a plate-only set (e.g. Pl_T100) to the stiffeners bordering its plates so it still yields stiffened panels (_beams_bordering_plates); - when concept names carry no _sbm role suffix, classify the secondary stiffener as the dominant plate-bordering profile rather than the shallowest — real models have tiny stub beams below the real stiffeners. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Scope-aware + cached so a scoped code-check on a 128k-element topside is seconds, not infeasible (Pl_T100: ~87 s -> ~13 s): - restrict the plate side to the scoped region (its plate members + the stiffeners' tributary plates) so plate-field bounding and the merge do not scan every shell in the model; - cache each plate's (frame-invariant) area on the node index and sum the cache in the rectangularity test instead of re-shoelacing every plate on every merge attempt; - prune merge candidate pairs by bounding-box adjacency before the expensive geometry test; - use elementwise dot, not @, in the rectangularity bbox (the local BLAS faults on batched matmuls of that shape). Classification is now purely geometric and naming-independent: the secondary stiffener is the dominant plate-bordering profile, not the _sbm-tagged concept. (Stiffener/cell *labels* still use concept names where present, but no grouping decision depends on the naming.) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
_concept_role / _CONCEPT_ROLE_RE were only used by the secondary-stiffener classification, which is now geometric (dominant plate-bordering profile). Remove the dead _sbm/_gbm role convention; the remaining _BEAM_SUFFIX_RE is used only to derive readable labels from concept names, never for grouping. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace the concept-name-based atomic grouping (_concept_groups grouped beams by concept base / _stiffener_specs_for_beams grouped by concept name) with a geometric one: each secondary beam is a stiffener run, its bounded rectangular plate strip is an atomic panel, and _merge_panels fuses adjacent strips into Genie's panels. Concept names are now used only to LABEL panels/stiffeners — verified the grouping (plate + stiffener element sets) is identical with names suppressed. Mini stays exact: Mini_grid_x100 reproduces Genie's 10 panels with the same names, stiffener tuples and plate tuples; codecheck2 still matches 49 panels; stress-resolution tests unchanged. Full-Mini panel count rises (per-beam merges slightly less than the old per-cell pre-clustering in unvalidated regions) but every panel stays unique and rectangular. Multi-element stiffener chaining is intentionally not done here: a physical stiffener is split by Genie at every support (girders AND perpendicular- plate floors the FE mesh doesn't expose as beams), so a purely geometric chain over-merges across floor lines; per-beam keeps the panels correct. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
SinSource.group now accepts several set names (their union via _scoped_set_members), so a code-check can be scoped to a few selected sets at once; a single name still works unchanged. get_field_shell_data dropped the whole result read with a KeyError when an RVSTRESS shell record referenced an element with no RDPOINTS entry (seen on the 399 MB Ruben topside). Skip those rows — without a result-point definition there is no integration-point field to build — so the viewer bake of real models does not fail. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The solver pulled the MKL 2026 preview (mkl_avx10.3) as numpy's BLAS/LAPACK backend, which SEH-crashes (0xc06d007f) on np.linalg.svd / gemm on Windows here — taking down the FEA-artefact bake (test_fea_artefacts) and any svd path. Force the OpenBLAS-backed libblas in the prod solve-group. svd/gemm now work; test_fea_artefacts goes from a hard crash to 86/87 (the remaining failure is an unrelated freetype poster-rendering dep). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
capacity_models() and resolve_cases() (and stress_resolve.resolve_cases) now accept a keyword-only on_progress(completed, total) callback, ticked once per built model / per resolved result case. This lets a caller drive a progress bar without adapy taking any UI dependency. Backward compatible: the parameter defaults to None. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
New toggle next to "Show definitions"/"Results" that hides every FEA draw range not belonging to a capacity model, leaving only the capacity panels (their faces plus the boundary/colour overlays when enabled). The pre-isolation hidden-range set is snapshotted per mesh so toggling it off restores exactly what was visible before. Driven by a new isolateDefinitions store flag and applyCapacityIsolation/ clearCapacityIsolation handlers built on CustomBatchedMesh's per-range hide. Frontend rebuilt (embedded index.zip). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The capacity result colour overlay was a transparent (opacity 0.96) layer coplanar with the lit base shell mesh and only a -2 polygon offset. On large models the base z-fights through and the <1 opacity lets the lit base bleed, so the UF colour read differently on each face (or appeared on one side only). Make the overlay fully opaque to the UF colour and give it a robust polygon offset (factor -4, units -16) so it reliably wins against the coplanar base on both faces. depthWrite stays off so the capacity-boundary lines still layer on top. Frontend rebuilt. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
resolve_cases now raises a clear ValueError listing the available result cases when a requested case is not present, instead of silently resolving to all-zero (the Ruben example asked for cases 14/15 on a SIN with only 1-13). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
When a SIN has no element stresses (SESTRA ISEL4=-1), resolve_cases now reconstructs shell membrane stresses from nodal displacements instead of resolving to all-zero. New stress_recovery module computes per-corner plane-stress membrane tensors (sigma = D*B*u) for TRI3/QUAD4 in the element local frame, emitted as a synthetic STRESS block (corners as bottom/top pairs, with result-point coords) so the existing resolver pipeline -- including the Section-5 start/mid/end station logic -- works unchanged. Verified against the Mini model (which has real RVSTRESS): per-element membrane tensors match exactly (median rel err 0.000); governing UFs match to median abs 0.0009 / p95 0.025, residual only in the start/end peaks (slightly conservative). Closed-form unit tests added. The stress-bearing path is untouched (recovery only triggers when STRESS is absent). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
get_field_line_data bailed when the RDPOINTS map was empty, so SINs that ship beam forces but no RDPOINTS (e.g. SESTRA smart-load-combination runs) lost all element forces and resolved stiffener axial/bending to zero. RDPOINTS is only needed for the shell-stress path; for line elements the element type comes from the converted mesh and the result-point count from the record length (component count from RDFORCES). The shell path and the RDPOINTS-present (e.g. Mini) path are unchanged. Verified on Ruben: FORCES now materialises (69564 beams x 3 stations), stiffener Nsd is non-zero, governing UF ~0.21 (was 0). Sesam reader regression tests and the Mini capacity integration test still pass. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
resolve_cases re-opened and re-decoded the SIN and rebuilt the full (step-invariant) mesh for every result case, which dominates on multi-hundred-MB models. Route it through a single SinStreamReader via a new iter_sin_step_results generator: the file is opened and decoded once, the static blocks and mesh are built once, and only each step's RV* tables are re-read. Results are unchanged (Mini capacity integration still passes); the win scales with the number of cases. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The resolve loop looked up element stresses/forces via ElementFieldData.get_by_element_id, which scans the whole block; calling it once per element across hundreds of capacity models made resolution quadratic and dominated runtime on large SINs. Index each field block's rows by element id once (cached on the block, which is rebuilt per step) and look up in O(1). Results unchanged (Mini capacity integration still passes); the Mini integration test dropped from ~23s to ~9s, and the win grows with model count. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The membrane recovery called np.linalg.inv/det per shell element (thousands of LAPACK calls), which crashes with a native SEH exception (0xC06D007F) on some Sesam-env BLAS/LAPACK backends -- the same failure class as the earlier svd crash. The operands are tiny (2x2 Jacobian, 3-vectors), so replace inv/det and all @/np.dot/np.linalg.norm with closed-form scalar + elementwise arithmetic. Crash-safe regardless of BLAS backend, and faster. Results unchanged (recovery unit tests + Mini capacity integration still pass). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The capacity side panel showed usage factors and the governing checks for the selected stiffener; add a per-row "Input" toggle that reveals the structured numeric input the DNV-RP-C201 stiffened-plate check actually used, grouped: Geometry, Stiffener section, Material, Design loads, Resolved design (Section 6), Options. Values are read from the existing sidecar (model geometry/section/material + per-row loads/resolved variables) and shown in engineering units (mm, MPa, kN, kN·m, kPa). No figures, numbers only; default view is unchanged until the button is clicked. Frontend rebuilt. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The inline Input details overflowed the (cramped, unscrollable) capacity side panel. Render it instead as a floating, scrollable panel pinned to the top-right of the 3D view via a portal (max-h 80vh, overflow-y-auto), with a header (stiffener name + close button). The "Input" button still toggles it. Frontend rebuilt. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add stiffener_stations() (start/mid/end coords along the span) and render them in the viewer: a "Points" toggle in the floating Input panel marks positions 1/2/3 with numbered, colour-coded sprites on the selected capacity model in the 3D view. The per-position load rows (sigma_y1/2/3, tau, N, M) carry a matching colour dot so it's clear which point each component is taken from. Markers are screen-space sprites (always visible), children of the FEA mesh (share its transform). Frontend rebuilt. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Stale v1 sidecars (no station coords) were being reused, so the Points markers had nothing to place. Bump the sidecar schema to v2 (adds the Section-5 stations) so the reuse matcher refreshes old sidecars automatically; sync the frontend validator + test. Also enlarge the station marker sprites (0.03 -> 0.05) for visibility. Frontend rebuilt. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Input panel now shows axial as membrane stress σ_x1/2/3 (from AverageLongitudinalMembraneStresses) instead of force, aligning with the FE stiffened-panel (Section 5) inputs in aibel_dnv_rp_c201_ui. - Tag resolved/extracted values with DNV-RP-C201 refs: σ_ySd (6.17), τ_Sd (6.16), N_Sd (6.2); per-position [5.3.x] clauses. - [6.4.3] mesh-discretization check per stiffener (extract.stiffener_ discretization): a warning banner on the main panel counts under-meshed models (e.g. Mini's single first-order element stiffeners; need >=4), plus a "Modelling [6.4.3]" group in the input panel. - "Individual UF" toggle on the main panel colours the selected panel by the selected stiffener's UF instead of the panel-governing maximum. - Sidecar schema -> v3; test tracks the constant. Frontend rebuilt. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The previous "Individual UF" coloured the whole selected panel by one (governing) stiffener's UF. Rework it to a per-stiffener view: each stiffener's own line + tributary plate strip is coloured by that stiffener's UF (max where strips overlap), revealing the within-panel UF variation. Uses the new per-stiffener tributary_plate_ids from the sidecar; the toggle is global (all panels). Sidecar -> v4. Frontend rebuilt. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
tributary_plate_ids only returned plates sharing a stiffener's full edge, so triangles and mid-bay elements were left uncoloured in the individual UF view. Add partition_plates_by_nearest_stiffener: assign every panel plate element (quad or triangle) to its nearest stiffener segment, giving gap-free per-stiffener coverage. Sidecar -> v5. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
SESTRA "smart load combinations" (e.g. Ruben case 14) are defined in the
SIN's RDRESCMB records but not stored as basic RV* result cases, so the
capacity check could not request them.
Read the combination definitions (combination IRES -> {basic IRES: factor})
and resolve a combination by superposing its basic cases' STRESS/FORCES/
RVNODDIS field blocks before the Section-6 resolve. Resolution is linear in
the FE field up to the per-station min/max, so superposing fields (and
running stress recovery on the combined displacement field) is exact —
unlike linearly combining the resolved scalars.
resolve_cases now splits requests into directly-stored basic cases (streamed
and resolved one at a time, memory-bounded as before) and combinations (their
basic blocks accumulated, then resolved once via the extracted _resolve_step).
SinMetadata gains combinations/result_names plus combination_ids and
selectable_cases. Verified on Mini: superposing lcc1/lcc2 reproduces the
fields SESTRA also stored for them to the float32 noise floor (< 1e-4 rel)
across every resolved design variable.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…etail) Matches the dnv-rp-c201 v6 sidecar. The frontend no longer expects every per-element UF value and case row inline: - applyCapacityVisualField is now async; for storage === "afel" it Range-fetches one (field, case) step from the capacity AFEL blob (reusing fetchElemFieldStep) and maps it onto the run's shared element_axis, instead of reading inline field.cases values. - loadCapacityCaseDetail lazy-loads capacity.detail/<case>.json into a new caseDetail map in the store when a case becomes active; CapacityControls sources its rows (list, selection, individual-UF, metric scores) from it. - active.capacityFetch retains the manifest-relative fetchers so the overlay can fetch blobs/detail after the initial load. Legacy (<=v5) inline sidecars still load via the json fallback path. CAPACITY_RESULTS_VERSION -> 6 in lockstep. index.zip rebuilt from source. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…cases Three capacity-panel UX changes (sidecar v7): "Only definitions" now shows *only* the capacity models: - hides the whole-model element-edge wireframe (was drawn over the isolated scene as "rest as wireframe"), and - collapses every non-capacity triangle in the colour overlay to zero area so the rest no longer shows as grey shells when results are on. - New "Show rest as wireframe" toggle brings the wireframe back for context. Worst over selected cases: - "Worst (over selected cases)" is the first Case option; per-case checkboxes (All/None, default all) choose the subset. - Colours by the max UF per element over the subset for the active check (Range-fetching each case's AFEL step and reducing element-wise). - The list shows worst UF per (model, stiffener) and the case it came from (new Case column), backed by the compact capacity.summary.json. - Clicking a worst row drills into that governing case. Also: rename the "Metric" dropdown to "Check"; lazy per-case detail and AFEL fetch retained. CAPACITY_RESULTS_VERSION -> 7 in lockstep. index.zip rebuilt. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Equal-height button row (items-stretch); "Only definitions" and "Show rest as wireframe" now default on. - "Individual UF" colouring now also works in the worst-over-cases view. - Input/Results sidecars get an "Input"/"Results" heading with the Case and capacity-model name beneath. - Input sidecar "Export" button downloads the inputs as an aibel_dnv_rp_c201_ui case file (fe_stiffened) for import there. - Failed capacity checks (runs[].errors) are flagged in red. - Design-loads inputs regrouped under Axial/Transverse/Shear/Moment sub-headings so the long names fit the sidecar width. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Paint the definition edges of capacity models whose check raised (for the active case, or any selected case in the worst view) red in the 3D scene, via per-vertex colours on the boundary overlay — matching the red error banner in the sidecar. - Move the under-meshed warning up next to the error banner at the top of the Capacity panel and present both as the same full-width collapsible notices (error red, warning amber), instead of the warning sitting below the run/case selectors. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The input export wrote full float precision (e.g. "hw": 180.00000715255737) from the metre->mm unit conversions. Round each numeric field to the same precision the Input sidebar shows (1 dp >=100, 2 dp >=1, else 3 sig figs) so the exported case reads like the panel (180.0, 5.94, ...). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Read the new per-case check_inputs (the exact geometry/material the engine consumed) and prefer it over the display capacity_model dict when showing the Input sidebar and building the fe_stiffened export. This makes the exported case reproduce the viewer's UF in aibel_dnv_rp_c201_ui — the stiffener span feeding sigma_y,R (eq. 4.6) is now the value the check used (e.g. 2.6 m, not the dict's 3.9 m), so C_ys [6.21] no longer flips negative on import. Pin CAPACITY_RESULTS_VERSION to 8 in lockstep; falls back to the display dict for pre-v8 bundles. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Input panel "Resolved design (Section 6)" now reads the engine's resolved_design (sigma_y,Sd eq 6.17, tau_Sd eq 6.16, N_Sd eq 6.2) instead of the Genie reference SigmaYSd, so the figures are DNV-RP-C201-correct (e.g. sigma_y,Sd 63.6 MPa, not the mid-point 27.0). Falls back to resolved_variables for pre-v8 bundles. - Carry optimize_z_star from check_inputs into the Options line and the fe_stiffened export, so an exported case imports with the same z* optimisation the viewer used. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Export loads.N_1..3 (kN) in the fe_stiffened case so the UI uses the FE axial-force resultants directly instead of re-deriving N from the membrane sigma_x — the two differ several-fold for stiffeners carrying global bending, which made the imported beam-column UF far too low. Pin CAPACITY_RESULTS_VERSION to 9 in lockstep (negative interaction UFs are now floored at 0 engine-side). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- UF colour coding is now the Genie "UfTot" discrete 6-band scheme (no fading): <0.2 light-blue, >=0.2 cyan, >=0.4 green, >=0.6 yellow, >=0.8 orange, >=1 red. Applied to the 3D overlay (capacityUfColor), the legend gradient (hard colour stops), and the result-table text colours. Sampled from .local/reference/genie_uf_color_scheme/genie_uf_color_scheme.png. - Drop loads.N_1..3 from the fe_stiffened export — that check is stress-only now (Stipla Section-5); the membrane sigma_x is exported as before. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…ve-stresses perf Two faults made the stiffener axial force (and beam-column UF) several-fold too high on some panels: - The panel merge bucketed only by (beam section, plane, offset), so one "panel" could span different plate thicknesses (e.g. 8/20/40 mm). A plate-thickness change is a buckling-panel boundary, so the merge now keys on plate thickness too (_merge_panels takes aux for the GELTH lookup) and never fuses across it. - _build_stiff_geom picked an arbitrary bordering plate via next() for the axial tributary (t, s_spacing) — it could grab a multi-spacing / wrong-thickness plate, so AxialLoadsPlate diverged from sigma_x*t*s by that factor. It now uses the panel's own plate (model.plates[0]) — the same one the capacity check consumes — so the integral matches the check exactly. Verified on DBSW OPE / Pl_L500: the bogus governing UF 1.385 becomes 0.226/0.450, and the invariant AxialLoadsPlate == sigma_x*t*s holds for all 1923 resolved stiffener-cases (worst ratio 1.0000). Adds test_plate_axial_force_matches_membrane_stress_times_tributary. This commit also lands the pre-existing in-tree "resolve design stresses" perf refactor (per-element rotation cache, geom_cache step-invariant geometry) that the tributary fix is built on, and updates its _resolve_step test for the geom_cache argument. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Bump CAPACITY_RESULTS_VERSION to 10 in lockstep with viewer_export.py so cached v9 bundles regenerate with the corrected stiffener axial forces (topology fix). Rebuilt the embedded frontend bundle (index.zip). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
iter_sin_step_results / SinStreamReader / _records_for take an optional `forces_elements` set that narrows the per-step RVFORCES decode to the given IELNO (stiffener beam elements). The capacity resolve only needs line forces for those elements, so this skips decoding the whole model's forces every step. Required by the already-committed stress_resolve.py, which calls iter_sin_step_results(..., forces_elements=...) — committing it completes the in-tree resolve-stresses perf refactor (the rest landed with the topology fix). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…presentation - ada.fem.capacity.girder: girder bays from primary plate-bordering beam runs (coplanar, perpendicular-stiffener flanking with line-contact coverage); rectangular per-side tributary strips partitioned nearest-girder-wins (parallel girders never overlap; perpendicular crossings share corners) - CapacityManager.girder_models()/resolve_girder_cases() - frontend: capacity sidecar v11 (second run, scope 'girder', g-prefixed case ids), girder line drawn amber / UF-coloured, tributary plates shown for the selected girder only, panel stiffeners as dashed lines, Individual UF always on (button removed), per-run store reset Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
# Conflicts: # src/frontend/package-lock.json
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ride)
The ':'-joined /opt/adacpp-overlay prefix from the Linux deploy makes the
whole PYTHONPATH one bogus entry on win-64 (';' separator), so
'python -m ada.comms.rest' failed with ModuleNotFoundError: ada. The overlay
path is a Linux-image concern only; the win-64 override drops it.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ection - GirderCapacityModel.stiffener_stations: supported-stiffener 2-point lines clipped to the tributary strip (drawn dashed in the viewer, like panels) - CapacityManager.panel_groups(): grouping exposed as its own step so callers can report the slow SIN grouping as a separate progress phase - capacity colour overlay is RGBA: elements without a result are transparent (shaded base mesh shows through) instead of flat grey - capacityResultsStore: per-run UI memory - switching Run (panel <-> girder) stashes and restores case/metric/selection/worst subset - click inside an already-selected panel selects the individual stiffener strip under the cursor (repeat clicks re-trigger via clickCoordinate); amber outline highlights the selected strip; Points/Input/Results follow - Export from a girder row emits an aibel fe_girder case (sigma_y,i = N_Gi/(A_G+l*t) inversion of eq. 5.2); girder-specific Input panel groups - rebuild embedded frontend bundle Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…le for Stipla - dominant_flange(): Genie models T-girders as unsymmetrical I-sections with a dummy flange (width == tw, token thickness); the GIORH parser and _section_from_mesh took the top slot literally, checking e.g. TG850x300x16x20 with a 16 x 0.1 mm flange (UF 8.13 vs 1.52 corrected). Both paths now read both flange slots and pick the governing real one. - Girder case export (buildUiGirderValues) emits the stiffener profile (stiffener_type + st_hw/st_tw/st_bf/st_tf, from sidecar v11+ check_inputs) so the dnv-rp-c201 UI can export the case to Stipla 3 DNV-G without manual profile entry; viewer Input panel lists the profile dims under Supported stiffeners. Frontend dist rebuilt. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
oleandor
changed the title
|
👋 Hi there! I have checked your PR and found the following: PR Review:I found no pr-related issues.
Python Review:I found some python-related issues: Python Linting results:
Python Packaging results:
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds a Capacity Manager (
ada.fem.capacity) that reconstructs Genie-style buckling capacity models straight from a Sesam.SINresult file and resolves their DNV-RP-C201 design loads, plus the viewer integration to inspect the results:The branch already contains upstream
main(0.27.0) as its merge base and is 0 behind — it merges clean.Testing
tests/core/fem/capacity: 36 tests (grouping/geometry vs a reference Geniemodel.json, SIN-native model reconstruction, stress-resolve values vs Genie criteria results, section parsing, flange-pick rules).tsc --noEmitclean,npm run buildregenerates the embedded bundle.🤖 Generated with Claude Code