Repository navigation
fix(outputs): node channels report the deck node on refined lines - #23
Conversation
The node output channels Ten<L>N<J>, Curv<L>N<J>, BendMom<L>N<J>,
L<L>N<J>p/v/a{x,y,z}, L<L>N<J>Dec/Azi, Torq<L>N<J> and Twist<L>N<J> name a
deck node (1 + the total NumSegs of the line's sections) and are checked
against that count at parse time, but when the static solve refined a
finite-EI line's mesh (adaptive, stability or curvature refinement) they
silently read node J of the refined mesh. On the lazy-wave cable of
tests/data/lazy_wave_touchdown_planar_coarse.dat with the top section cut
from 20 to 4 elements (refined from 71 to 1136 elements), Ten4N34, the deck
node at the 220 m touchdown, reported 5488 N, the tension at 41 m, against
1544 N, and BendMom4N54, on the seabed at 420 m, 1214 N.m against 0.
The cable builder now maps each deck node onto the final mesh by its
unstretched arc length from End A (an exact match within 1e-9 of the line
length, computed from the arc positions, so a non-uniform local refinement
maps too) and stores the map on the cable; a deck node with no mesh node
there fails closed when its channel is evaluated. The finite-EI channel
evaluator reads deck nodes through the map, so every route that reaches it
reports the deck node: the standalone static and dynamic cubic-Hermite
routes, the mixed EI = 0 + finite-EI route, the multibody march, the
OpenFAST aggregate (CD_AGG_EvalChannel) and the C API CableDyn_EvalChannel.
The static profile writers and CableDyn_GetLineValues keep listing every
node of the refined mesh and evaluate by mesh node; the static profile then
carries a DeckNode column naming the deck node of each row (0 for a node the
refinement added). Range graphs, .elements.out and the per-line files are
unchanged. EI = 0 lines and the two-moving-end compatibility route are never
refined and are unchanged. The map is rebuilt with the cable on every
initialisation, so a checkpoint restart, ptfm_init and an InitDeckAt
re-solve carry it.
The Python StaticProfile reads and validates the DeckNode column and gains
deck_node_arc_lengths; node_range_graph, profile_at/profiles_at and
line_range_graph place main-output node channels at their deck node, from a
static profile or from a per-line position file beside a refined profile.
seabed_friction_aniso read the sway of refined-mesh nodes 38 and 42 (37 and
41 m along the line) through L1N38py/L1N42py while meaning deck nodes; its
gate now reads deck nodes 20 and 21 (38 and 40 m), the grounded run it was
written for.
New ctest node_channels_refined_line runs the reproducer on the static,
dynamic and mixed routes and through the C API. Every output file of the
examples is byte-identical (no example refines). doc/outputs.rst documents
the deck-node numbering on a refined line and how to find the deck nodes in
the full-mesh files; doc/capi.rst, the options and post-processing pages and
the CHANGELOG follow.
validation/PAPER_REPRODUCTION.md adds element OutputRecovery to the paper
mode settings: the article reported the element-mean axial force and the
geometric Hermite curvature, which that option restores.
Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The article's table was re-run in both modes on the development source at 71158d5. The record now gives those values: - Tension-only lines no longer use the same scheme in both modes: since the force-blended EI = 0 step, alpha_force_blend selects it too, and False gives the article's scheme. In paper mode every EI = 0 and finite-EI result equals v0.1.0 in every printed digit except the two-point axial rule row. - Two-point axial rule: -922.7 / -810.3 / -264.4 kN (80 / 200 / 800 m) after the seabed anchor-end weight-moment fix, from -937 / -820 / -280.5 kN; the spurious compression under under-integration stands (F). - Slack-to-taut rows, defaults: changes 0.0010-1.5410 %, peaks 10.2246 ... 10.3821 MN, no apparent order; a note states that both blends converge to the same 0.25 ms peak and that the article's order 2.69 refers to the configuration blend (D). - One-hour VolturnUS-S, defaults: end-element SD / min / max 0.27 / 1.46 / 1.05 %, end force 0.27 / 0.26 / 0.50 / 0.43 % and its 12 s Fourier amplitude 1.14 % (D); the mean and the end-element Fourier values are unchanged. - Mooring motion envelope: with the defaults a different set of motions is subdivided at 0.1 s and the severe-motion peaks changed. Known difference gives the peaks against converged 0.5 / 0.25 ms references, and says that the force blend removes the spurious semitaut spikes but underestimates the all-chain heave peaks at that step, so peak tensions need a smaller dtM. - How to reproduce: paper mode for the EI = 0 test programs (GenAlphaConfig%force_blend = .FALSE.), and a toolchain note: gfortran gives the v0.1.0 paper-mode values, with 0 rather than 1 of 1200 800 m intervals subdivided. Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The deck-node map of a refined finite-EI line was available only through the DeckNode column of .static.out, which the multibody march does not write, and not at all to in-process users. Python then fell back to mesh numbering when it placed main-output node channels with a per-line position file, LinePositions read from one, or position arrays. - .Line<L>.p.out of a refined line carries a second comment line, "# DeckNode" with the deck node of each mesh node (0 for an added node), on every route that writes it, the multibody march included. Unrefined lines write the file unchanged. - The C API GetLineValues takes a new quantity CD_C_LINE_DECK_NODE (10): the deck node of each node of the final mesh, 1 .. n_nodes on a line meshed as the deck. No function is added, so the ABI minor level is unchanged; an earlier library rejects the code with CD_C_BAD_INPUT. The header comments now say that the line arrays cover every node of the final mesh while node channels name deck nodes. - Python: LineNodeHistory.deck_nodes reads and validates the comment line, and Line.deck_nodes queries the C API. profile_at, profiles_at and line_range_graph place main-output node channels with a per-line file or LinePositions read from one through the file's own map, then through the static profile beside it. They fail on a LinePositions taken from a multi-line refined static profile instead of guessing. The docs state that plain position arrays are numbered as given. - PAPER_REPRODUCTION.md names its value columns after revision 71158d5. - test_node_channels_refined.py fails when --library is given but is not a file. It checks the .p.out map, placement from a lone per-line file and from LinePositions, and Line.deck_nodes. test_c_api_shared checks the new quantity on an EI = 0 line. - orcaflex_discrete_modules_reference.py reads .static.out columns by header name, since a refined line has 13 columns. The other .static.out readers in validation/ and tests/ already read by name or by leading index. test_anchor_uplift skips every comment line of a .p.out file. Examples remain byte-identical (no example refines). Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
|
@codex review |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 33cc02d7b3
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
line_positions(profile, line_id=L) on a multi-line static profile with a DeckNode column kept only the shared source path, so placing main-output node channels with the result looked up the profile without a line and raised for every multi-line profile. LinePositions now carries the selected line_id (also through window), and the deck-node lookup uses that line's map. A LinePositions built without a line still refuses a multi-line profile, as it cannot name one map. Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
|
@codex review |
|
Codex Review: Didn't find any major issues. Another round soon, please! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
Summary
On a finite-EI line whose mesh was refined automatically, the node-indexed OUTPUT channels (
Ten<L>N<J>,Curv<L>N<J>,BendMom<L>N<J>,L<L>N<J>p/v/a,Dec,Azi,Torq,Twist) read node J of the refined mesh instead of deck node J, silently. Names are validated against the deck node count, so after refinement a channel pointed somewhere else on the line:Ten4N5(sag bend)Ten4N34(touchdown)BendMom4N54(on the seabed)Fix
PtfmInitandInitDeckAtcarry it.CD_AGG_EvalChannel) and the C APIEvalChannel..static.outgains aDeckNodecolumn, appended last and only when a line was refined..Line<L>.p.outfiles gain a# DeckNodeline, also only when refined.GetLineValuesquantityCD_C_LINE_DECK_NODE. An older library rejects it withCD_C_BAD_INPUT; no function is added.Line.deck_nodesandStaticProfile.deck_node_arc_lengths.profile_at,profiles_at,node_range_graphandline_range_graphplace main-output channels at their deck node, and raise rather than guess when no map is available.validation/PAPER_REPRODUCTION.mdis brought up to date with a re-run of the article's table on dev 71158d5:False alpha_force_blend,sequenced cable_staticsandelement OutputRecovery. It reproduces v0.1.0 in every printed digit except the two-point axial-rule row, which moved with the seabed anchor-end weight moment fix.Tests
node_channels_refined_line: the reproducer at deck nodes 1, 2, 5, 18, 34, 54 and 72:.elements.out;L4N<J>pagainst.p.outat the mapped node, on every row;EvalChannelagainstGetLineValues,CD_C_LINE_DECK_NODE, and rejection of node 73;LinePositions.It fails on dev.
test_results.py,test_profiles.py, andtest_c_api_sharedfor the new quantity.seabed_friction_aniso: its gate read refined nodes it took for deck nodes; it now reads deck nodes 20/21 (38/40 m).pre-commit run --all-filesand the Sphinx build with-Wpass.Checklist
pre-commit run --all-filespassesctest --test-dir build --output-on-failurepassespython -m pytest tests python/testspasses (if Python or validation records changed)validation/PAPER_REPRODUCTION.md[Unreleased](heading added if missing) for user-visible changesdoc/updated for changed inputs, options, or outputsfeat:,fix:,docs:, ...)git commit -s) under theDeveloper Certificate of Origin