Repository navigation
feat(openfast): linearise the line dynamics, as MoorDyn-F - #66
Conversation
With Linearize = True the module now registers continuous states: the interior-node positions of every EI = 0 line (MoorDyn's names and order) and the free DOFs of every finite-EI cable (node positions and tangents), then their velocities. lin_eval loads a state and an input into the committed mirror, solves the accelerations without a time step and restores the committed state; CD_CalcContStateDeriv, CD_JacobianPContState and the input Jacobian central-difference it for A, B, C and D. This replaces the quasi-static linearisation without states. Positions, tangents and coupled-point displacements are stepped by 1 um: a millimetre step lifted a light cable's grounded nodes out of their 0.3 mm seabed penetration on one side of the difference, and the former 5 cm input step slackened a chain's last element, each halving a stiffness. Velocities take MoorDyn-F's 0.1 m/s, which gives the quadratic drag its equivalent linear damping. On the UMaine twin decks the reduction D - C A^-1 B matches MoorDyn-F's to 0.1 % per fairlead and on the platform 6x6, and the lowest nine line-mode frequencies to 0.3 %; both state matrices are stable. On the mixed deck with a lazy-wave power cable every coupled point matches central differences of the deck's own static solutions to 3.3e-4, also with a hang-off spring. openfast_lin_ab.py gates the frequencies, the stability and (with --cable-deck and --driver) the mixed deck, and solves an ill-conditioned A plainly instead of truncating it. A numeric-stiffness cable end connection now linearises. Free/Connect points and a Rigid cable end connection fail closed with named messages, as bodies, rods, Point3 buoys, torsion and CONTROL already did. Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
Every Linearize call now takes the line-state branch of the input Jacobian, dXdu included only when the glue asks for it (a deck without line states has none), so quasistatic_dydu and its scratch-aggregate re-solves are unreachable and removed. Outside a linearisation run the input Jacobian is the runtime fixed-state derivative for any state count, and dXdu or a state Jacobian fails with a named message. The module comments no longer describe the zero-state reduction. 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: 3a7eb208d2
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".
The viscoelastic slow-branch stretch dl_1 and the Syrope slow strain are dynamic states with their own rate laws, but the linearisation held them, so D - C A^-1 B gave these ropes' instantaneous stiffness instead of their relaxed static one. They are now first-order linearisation states (VF_DerivOrder1, "Line l segment e dl1, m" / "slow strain, -"): lin_eval loads them from x and writes their rates, which the new CD_Get_Model_Internal_Rates evaluates with the instantaneous element laws (the taut series-Kelvin ODE or the slack zero-load relaxation; the Syrope kernel at dt = 0). A viscoelastic variant of the UMaine chains (150 stretches) now matches its static solutions to 7e-5 (held: 0.6 % too stiff); a taut three-leg Syrope spread with a SYROPE IC history (60 slow strains) to 6e-6 (held: 2.9 times too stiff). The Syrope running maximum and the seabed-friction stick anchors stay held: load histories, whose operating-point values are the small-motion tangent. A line or cable with VIV fails closed: it oscillates about a lock-in cycle, not an equilibrium. The A/B script's static reference adds TMax = 0 to a deck that leaves TMax to the host. Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
|
@codex review |
|
Codex Review: Didn't find any major issues. Nice work! 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". |
OpenFAST linearisation with the line dynamics, as MoorDyn-F
With
Linearize = Truethe module now registers continuous states and linearises the line dynamics, replacing the quasi-static linearisation without states.States
EI = 0line's interior-node positions, with MoorDyn's names and order (Line 1 node 1 Px, m...Vz, m/s);m = dr/ds(Cable 1 node 2 Px, m,Tx, -,dTx, 1/s, node 1 at End A);Matrices.
lin_evalloads a state and an input into the committed mirror and solves the accelerations without a time step. It then restores the committed state.CD_CalcContStateDeriv,CD_JacobianPContStateand the input Jacobian central-difference it forA,B,CandD.Steps.
Validation (
validation/scripts/openfast_lin_ab.py)D − C A⁻¹ Bvs MoorDyn-F (882 states each)--cable-deck, 2976 cable states) vs central differences of the deck's own static solvesThe damping ratios are printed, not gated, and come out at half MoorDyn-F's. Both codes difference the drag over ±0.1 m/s. CableDyn integrates the drag over the element, and for a one-node perturbation that gives half the slope. The script now solves an ill-conditioned
Aplainly (cond ~1e11 from a frictionless laid cable is physical). It switches to the pseudo-inverse only above 1e14.Scope
Rigidcable end connection fail closed with named messages, as bodies, rods, Point3 buoys, torsion,CONTROLand FAST.Farm already did. These carry states that are not linearisation states.quasistatic_dydu) is removed. Outside a linearisation run the input Jacobian is the runtime fixed-state derivative as before.Checks
deck_driverin the deep scratch build tree, whose Syrope long-path case exceeds Windows' path limit there and passes in the repository build.doc/openfast.rst(new Linearisation section),doc/troubleshooting.rst,VALIDATION.md,CHANGELOG.md.