Repository navigation
feat: VIV FRAME for the synchronisation model's amplitude estimate and damping split - #166
Conversation
…d damping split The synchronisation model defined its ambient-frame quantities across the ambient flow (current and held field) only. A cable moved through still water, by a tow or by its floater, therefore kept y0 = 0, Cv(0) and CdAmp(0), and its VIV DAMPING split blended to Morison's drag, while the same problem in the cable's frame (the water moving past the held cable) took the amplitude-dependent values. A VIV FRAME row (TypeName Tf) now also acts on a type with a VIV AMPLITUDE table. With Tf > 0 the ambient-frame quantities take the ambient flow relative to the node velocity low-pass filtered over Tf, the filter of #165: vbar_{n+1} = vbar_n + (1 - exp(-dt/Tf)) (v_{n+1} - vbar_n) - Amplitude estimate: y' = (v - vbar) . t x (u - vbar)_n/|(u - vbar)_n|, and with it the table's Cv, the drag factor CdAmp and C_dy. - VIV DAMPING split: the axes e_x, e_y and the blend beta take (u - vbar)_n, an element the mean vbar of its end nodes. vbar is the step-start value within a step, so the Jacobians are unchanged. - The phase rate keeps the published form (U the relative normal speed), and the lift and drag the relative velocity. For a frame at constant velocity (a steady tow) the quantities equal those of the held cable exactly; the estimate loses the part of the VIV velocity that passes the filter, a factor wTf/sqrt(1 + (wTf)^2). - Finite-EI cables: Set_VIV takes tf(ne) for the synchronisation model (refused with the wake oscillator, which keeps wake(4, :), and where an element with tf > 0 has no amplitude table). viv_ms grows from 6 to 9 rows (CD_VIV_NMS_FRAME), rows 7-9 the filtered velocity, so the recovery, snapshot, mirror and checkpoint copies carry it. The drag kernel takes the frame velocity as an optional frame_v. - EI = 0 lines: CD_Set_Model_VIV takes tf(n_elem); rows 7-9 of the VIV motion statistics hold the filtered smoothed velocity, so the state get/set, rollback and checkpoint carry it (CD_Model_VIV_NMS 9). - Deck reader: a VIV FRAME row needs a WAKE OSCILLATOR row or a VIV AMPLITUDE table; the Python reader and builder follow. - Tf 0 or no row leaves runs unchanged to the bit. Gates: - test_hermite_viv_frame: a 20 m pipe of the displaced mass (so the in-line inertia of the two frames is the same; another mass differs by the Froude-Krylov term) with the amplitude table and Thorsen's damping, moved at 0.6 m/s amplitude and 40 s period through still water, against the same pipe held in the water moving the other way (held field with its acceleration). With Tf 0.6 s: peak A/D 0.838 against 0.825 (0.834 Debug, 0.843 on the Prescott kernel; gate 5 %), RMS 0.415 against 0.445 (gate 12 %; the remainder is the phase input, see below). Without the frame: 0.711 against 0.850 peak, 0.349 against 0.450 RMS. Refusals, an all-zero tf bit for bit, and a snapshot replay bit for bit. - test_viv_frame_deck: a Tf 0 row is the deck without it, bit for bit, on both cores; a finite-EI cable towed at 0.15 m/s with Tf 10 s gives A/D 0.780 against 0.792 held in the current (0.665 without the frame); held in the current Tf changes A/D by 0.16 % (finite-EI) and less than 0.01 % (EI = 0); the refusal of a type without a table. - test_viv_coupled: with a VIV FRAME on amplitude-table types (and VIV DAMPING on the cable) both checkpoint blocks grow by 3 per node, the restarted run continues bit for bit, a rebuild without the VIV state does not, and a rewound step repeats. - Python reader, builder and native-parity cases. The phase input y' = r'.e, e across the relative flow, equals u.e and so is zero in still water: the moving pipe's phase follows y'' alone. It is left in the published form here; taking y' relative to the frame closes the RMS gap of the pipe check to 0.1 %. Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
… filtered motion The synchronisation phase read the cross-flow velocity r'.e across the relative flow, which is u.e and so zero in still water: a cable moved through still water locked its phase on y'' alone, while the held cable in the moving water did not. The in-line phase input x' = r'.v_n/U carried the imposed motion along the flow. With a VIV FRAME row and Tf > 0 the cross-flow and in-line phase inputs now take the node's velocity relative to its filtered velocity vbar (viv_node_motion on finite-EI cables; the smoothed velocity in the phase step on EI = 0 lines). The relative flow, and so the axes and the shedding speed U of the phase rate, are unchanged. The acceleration stays the node's: vbar changes slowly, and its rate (v - vbar)/Tf would carry the VIV velocity into y''. The frame then acts on every synchronisation type, so a VIV FRAME row now needs LINE TYPES Cl > 0 or a WAKE OSCILLATOR row, not a VIV AMPLITUDE table; the setters take tf > 0 where cl > 0. Without a table the 9 motion statistics keep rows 3-6 at 0 and vbar in rows 7-9. The Python reader and builder follow. Without a row, or with Tf 0, runs are unchanged to the bit. Gates: - test_hermite_viv_frame, the 20 m pipe of the displaced mass moved at 0.6 m/s amplitude and 40 s period in still water against the same pipe held in the water moving the other way, Tf 0.6 s: | pipe | frame | peak A/D moving / held | RMS A/D moving / held | |-------------------|-------|------------------------|-----------------------| | table and damping | Tf | 0.8191 / 0.8189 | 0.4464 / 0.4470 | | table and damping | none | 0.7108 / 0.8502 | 0.3486 / 0.4500 | | Cl alone | Tf | 0.7127 / 0.7136 | 0.3722 / 0.3710 | | Cl alone | none | 0.7108 / 0.7364 | 0.3486 / 0.3749 | Gate 1 % on peak and RMS with the frame (at most 0.7 % over the Release and Debug builds and both OpenBLAS kernels; the rest is the filter lag, omega Tf = 0.094); without it more than 10 % / 15 % with the table, more than 4 % RMS with Cl alone. - test_viv_frame_deck: towed at 0.15 m/s with Tf 10 s, A/D 0.7905 against 0.7906 held in the current with the same row (0.7924 without it) with the table and damping, 0.6764 against 0.6764 (0.6775) with Cl alone; gates 0.2 % and 1 %. Held, Tf changes A/D by at most 0.23 % on both cores (gate 0.5 %). A Tf 0 row is bit-identical on both cores with and without the table. A row on a type without VIV is refused by name. - test_viv_coupled: the frame on the lift alone, restart and rewind bit for bit with the 9-row VIV block on both lines. - test_viv_wake_deck W7 and the Python cases follow the new rule. Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
|
@codex review |
Codex Review Summary: ✅ CompletedThis 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. |
|
Codex Review: Didn't find any major issues. Chef's kiss. 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". |
On Linux with the Prescott OpenBLAS kernel the moving pipe's peak A/D was 0.8106 against 0.8189 held (1.01 %), above the 1 % gate, while the RMS agreed within 0.2 %. The peak is a single extreme of the response and ranges 0.811-0.819 over builds, kernels and platforms; the RMS is the robust measure. The theory section states the spread. Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
|
@codex review |
|
Codex Review: Didn't find any major issues. Hooray! 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
#165 made the wake-oscillator frequency use the flow relative to the node velocity low-pass filtered over
Tf(VIV FRAME). The synchronisation model still defined its ambient-frame features across the ambient flow only:VIV AMPLITUDE(Cv(A/D)and the drag factorCdAmp);VIV DAMPINGsplit.A cylinder moved through still water therefore got
y0 = 0,Cv(0),CdAmp(0)and a pure Morison split, while the same problem in the cylinder's frame got the amplitude-dependent features. For a floater-driven cable (the hang-off moves at wave frequency) the result depended on the frame.With a$\bar v$ :
VIV FRAMErow andTf > 0, these features now use the flow relative to the filtered node velocityon finite-EI Hermite cables and
EI = 0lines. Without the row (or withTf = 0) every result is bit-identical todev.Implementation
viv_msgrows from 6 to 9 rows (rows 7-9 holdframe_vof the drag kernel).CD_HermiteCable_Dyn_Set_VIV(tf=)andCD_Set_Model_VIV(tf=)takeTfper element.viv_frame_tf) follow the same rules.theory-viv-frame, cross-references from the amplitude and damping sections, capabilities rows,driver_format.md, CHANGELOG.Verification
Frame check: a 20 m finite-EI pipe with the displaced-water mass, moved at 0.6 m/s amplitude and 40 s period through still water, against the same pipe held in water moving the other way (Froude-Krylov included),
Tf0.6 s:VIV DAMPINGTf0.6 sVIV DAMPINGClaloneTf0.6 sClaloneTf10 s: towed 0.7905 vs held 0.7906 (table + damping), 0.6764 vs 0.6764 (Clalone).Tf10 s); documented in the theory section.VIV FRAMErow needsCl > 0or aWAKE OSCILLATORrow; a type with neither is refused by name.ctest: 426/426 on the default and the Prescott OpenBLAS kernels; Debug (-fcheck=all) VIV tests pass.dev7d3998f: 10 decks, 26/26 output files byte-identical.check_example_fallbacks.py: 66 decks, no fallback notes;pre-commitclean; Sphinx-W -ngives only the 3 warningsdevalso gives.