Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 10 additions & 1 deletion ARCHITECTURE.md
Original file line number Diff line number Diff line change
Expand Up @@ -193,7 +193,16 @@ The effective tangent is `(1 − α_m)/(β Δt²) M + (1 − α_f) K_q +
1e-14, 30 iterations, 12 backtracks. A non-converged step, or on the Hermite path
an under-resolved one, is repeated by internal subdivision (4, 16, … up to
`recovery_max_substeps`) along a C² quintic boundary trajectory that ends on the
host time.
host time. On `EI = 0` lines `CD_Step_Model` also re-solves a converged step in
`max(4, ceil(turn/ei0_substep_angle))` sub-steps on the same trajectory when a
taut segment turned by more than `ei0_substep_angle` or a slack segment snapped
taut while the step does not resolve its axial wave (`ei0_substeps auto`, the
default; at most `ei0_substep_max`). It restores the committed kinematics and
constitutive and friction states first (`commit_state_copy`, the one list of
that state in `CableDyn_Model`) and keeps the plain step when a sub-step fails.
The sub-steps of a recovered step, the halves of a line-system step halved by
`CD_Step_System`, and the lines attached to the objects of the monolithic
body/point step are taken as given.

### Bodies, rods, and points

Expand Down
124 changes: 114 additions & 10 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -100,10 +100,13 @@ All notable changes to CableDyn are recorded in this file. The format follows
step before and after, ten times the largest tension anywhere on the line one step before and
after, and the largest tension of the initial state. The first spike of a line is reported as
a warning, once per line, and the run summary counts them. The run ends with a note when a
taut segment turned by more than 1° in one step (the mean and standard deviation of the
tension err by about 1 % and more above it; 1.3 % at 2.9° on a heaving chain), and when
segments snapped taut while the axial wave crosses a segment more than once per step, so the
snap peaks are not resolved in time. The standalone driver also prints a note for a dynamic
taut segment turned by more than 1° in one solver step (with `ei0_substeps auto`, see Changed,
1.1 times the larger of 1° and `ei0_substep_angle`; the sub-steps of a sub-stepped or
recovered step count as steps; the mean and standard deviation of the tension err by about
1 % and more above it, 1.3 % at 2.9° on a heaving chain, and near 6° per step snap-load peaks
come out 10 to 20 % low with either generalised-α blend), and when segments snapped taut
while the axial wave crosses a segment more than once per solver step, so the snap peaks are
not resolved in time. The standalone driver also prints a note for a dynamic
deck with `rhoInf` above 0.95, on the missing numerical dissipation. See the manual's
troubleshooting page.
- Anchor uplift channels for the drag-anchor check of a catenary mooring (no vertical load at
Expand Down Expand Up @@ -168,6 +171,92 @@ All notable changes to CableDyn are recorded in this file. The format follows

### Changed

- Tension-only (`EI = 0`) lines resolve fast segment rotation and slack-to-taut snaps by
sub-stepping, on by default (new OPTIONS `ei0_substeps auto|off`, `ei0_substep_angle`
(default 1°) and `ei0_substep_max` (default 16), checked when the line is initialised; the
Python deck reader applies the same rules). A converged step in which a segment that ends it
taut (above 0.1 % of the largest step-start tension) turned by more than
`ei0_substep_angle`, or a segment slack at its start ends it above a quarter of that tension
while the step does not resolve the axial wave over it (the wave crosses the segment more
than once per step), is solved again from the same committed state in
`max(4, ceil(turn/angle))` equal sub-steps, at most `ei0_substep_max`, with the end motion on
the quintic of the recovering step and the last sub-step on the given end state; output and
coupling times are unchanged. The tensions are those of the run audit (the elastic tension,
or that of the viscoelastic or Syrope law), and the step-start tension is taken as at least
0.1 % of the line's initial tension, so a line slack all over does not count barely taut
segments. At `dtM` 0.1 s a taut segment of the Lozon 80 m
all-chain mooring turns by up to 6° per step in 5 m / 7 s heave, and the snap-load peaks of
both generalised-α blends came out 10 to 20 % low. The release cap was not the cause: it acts
in 0.3 % of the element steps, never within 0.5 s of a peak, and the cycle maxima with and
without it agree to within 1 %; the peaks followed the algorithmic damping (28 and 31 % low at
`rhoInf` 0, 10 and 40 % high at 0.6). Over seven runs whose amplitude differs by up to 3e-4,
against 0.5 and 0.25 ms runs, the fairlead and anchor maxima move from 15 and 17 % low to 4 and
8 % low (the configuration blend of 0.1.1 gave 10 and 18 % low), and the anchor maxima of
the 200 m semitaut mooring in 10 m / 14 s and 5 m / 7 s surge from 16 and 26 % high to 4 %
high and 2 % low (over thirteen runs 5 % high and 0.3 % low, single runs from 4 % low to
21 % high). A single run at that step varies by 15 to 20 points, so a single comparison
can make either blend look accurate. The decision depends only on the committed state, the
step inputs and the line's initial tension, so restarts repeat it; when a sub-step does not
converge the plain step's own result is kept, and a step recovered after a stall or a half
of a line-system step halved after a failure takes its sub-steps as given (when the halves
fail too, the full step is re-run as the nominal step, sub-steps included). It applies on
every route that steps `EI = 0` lines (standalone lines and line systems, OpenFAST,
FAST.Farm and the C API, whose raw lines always take it) except the lines attached to the
bodies, rods and points of the monolithic body and point coupling, which needs the end
reaction and tangent of the nominal step. `monolithic bodyScheme` is the default of every
standalone deck with `Rigid6` bodies, `Point3` buoys, rods or `Free`/`Connect` points on
`EI = 0` lines, so those lines are not sub-stepped there: 7 of the 21 shipped `EI = 0`
dynamic example decks run this way, and none of them notes an unresolved segment turn. A
line of such a deck whose ends are all fixed or anchored is sub-stepped, `staggered
bodyScheme` sub-steps them all, and the standalone driver notes the exclusion once per run.
`False alpha_force_blend` (paper mode) turns it off unless the deck sets it, so the paper
reproduction is unchanged. A step that triggers costs its plain step and usually four
sub-steps, at most the plain step and three passes of `ei0_substep_max` sub-steps (49 line
solves at the default), and that once for every sub-step where bodies or rods sub-cycle the
lines (`staggered bodyScheme`, coupled runs); the OpenMP loop over the lines of a system
hands them out dynamically, which leaves every line's result unchanged. The standalone
driver reports the sub-stepped steps per line ("solved N of the M line step(s) it
examined"), OpenFAST and FAST.Farm print the counts (steps sub-stepped, largest count, steps
that needed more than `ei0_substep_max`, steps that kept the plain step) at the end of the
run, and `CD_AGG_Get_Substep_Counts` returns them to an aggregate host as 64-bit integers.
Only committed steps count: a step rolled back and taken again (a line-system halving, a
stall recovery, a corrector or a junction iteration) counts once. Over the shipped examples
19 of the 21 `EI = 0` dynamic decks are unchanged bit for bit; in the two `als_volturnus`
line-break decks the pre-failure anchor snap at 250 s, a single-step peak at `dtM` 0.05 s,
moves from 2.145 to 1.918 MN (1.963 MN converged) and from 2.262 to 2.158 MN (2.334 MN
converged), with the other tension statistics within 0.02 %. The 20-run Lozon mooring
envelope at `dtM` 0.1 s takes 1.67 times as long (1.6 to 2.5 times on the 5 m / 7 s and
10 m / 14 s motions; the steps that do not trigger cost only the measurement), and the
examples whose steps never trigger run as before. Of the reproducers of the force-blend
entry below, the semi-taut chain–polyester leg in 15 m surge now peaks at 4.65 MN at the
anchor at `dtM` 0.05 s (5.74 MN before, 4.6 MN at 0.0025 s), the Syrope line at 1.81 MN
(1.64 MN before, 1.78 MN at 1 and 2.5 ms), the lightly damped viscoelastic light cable at
0.062 MN at the anchor (0.085 MN before, 0.082 to 0.095 MN at 0.5 to 2.5 ms; its snap is not
resolved at 0.1 s, and over seven runs whose amplitude differs by up to 3e-4 it peaks at
0.056 to 0.079 MN with the sub-steps and 0.056 to 0.085 MN without) and the shipped
viscoelastic polyester at 1.80 MN as before (in its steps recovered after a stall, taken as
given, a segment now turns by up to 2.0° per solver step, which the run notes); the light
cable stays at 0.092 / 0.106 MN at `dtM` 0.1 s and its largest segment tension at `dtM`
0.01 s moves from 0.107 to 0.095 MN. Its fairlead and anchor peaks at `dtM` 2.5 ms move from
0.099 / 0.091 to 0.102 / 0.119 MN, and on the stiff seabed at `dtM` 0.05 s from 0.52 / 0.67
to 0.62 / 0.86 MN. None of these is a converged peak: the undamped `EI = 0` cable has no
converged snap peak (see the force-blend entry below; converged finite-EI runs give 0.093 /
0.057 MN). At a step that does not resolve the axial wave the sub-steps resolve more of the
segment rattle of this undamped cable and can raise isolated spikes, as its anchor peak at
2.5 ms shows. At `dtM` 0.5 ms, where the step resolves the wave, the cable takes no sub-step,
with or without physical axial damping (`BA` 5.06e5 N·s), and its results equal those taken
as given bit for bit; without the axial-wave condition on the snap trigger `auto` sub-stepped
8081 of the 60000 steps of that run and raised its anchor peak from 0.116 to 0.52 MN. `ei0_substep_max` and `ei0_substep_angle` do not change these runs. Validation:
the Bergdahl matrix at its 0.3125 ms step resolves the axial wave over a segment and takes no
sub-step, so its maxima are unchanged bit for bit (RMSE 1.58 N, MAPE 3.13 %, Pearson 0.993,
Lin concordance 0.989); of its refinement runs the 0.1 m mesh moves the T = 1.25 s mean
cycle maximum from 68.33 to 68.35 N (largest mesh-halving change 0.52 % to 0.56 %) and the
0.625 ms step from 72.2 to 69.5 N. The WD0050/A2 cell of the L2-2 parity moves within its
recorded range (largest OrcaFlex summary difference 0.091 to 0.093, MoorDyn-C metric 0.078
to 0.076); the other L2-2 cells, the one-hour VolturnUS-S record, the L3-2 regular-wave
comparison and the snap-load verification are unchanged bit for bit (the latter two step
the generalised-α kernel directly). `off ei0_substeps` gives the previous results bit for
bit.
- Finite-EI node tension, curvature and bend moment are recovered from nodal equilibrium. The
`Ten<L>N<J>`, `Curv<L>N<J>` and `BendMom<L>N<J>` channels, the `Tension`, `Curvature` and
`BendMoment` columns of the static profile and the range graph, and the `.t.out` segment
Expand Down Expand Up @@ -368,8 +457,9 @@ All notable changes to CableDyn are recorded in this file. The format follows
work to the motion than its strain energy. Both happened silently at the default
`rhoInf = 0.4`, at every `dtM` from 0.002 to 0.2 s, on realistic input: a light, axially stiff
cable (80 kg/m, EA 8e8 N) under 3 m heave reported 4.2 MN at the fairlead and 3.4 MN at the
anchor (`dtM` 0.0025 s) and 20 MN at the anchor (`dtM` 0.1 s), against 0.1 to 0.18 MN at
`rhoInf = 0`, at `dtM` down to 0.25 ms and with a finite-EI model; the same cable in an `Hs`
anchor (`dtM` 0.0025 s) and 20 MN at the anchor (`dtM` 0.1 s), where converged finite-EI
runs of the same cable give 0.093 MN at the fairlead and 0.057 MN at the anchor; the same
cable in an `Hs`
8 m JONSWAP sea 0.41 MN at the anchor against 0.043 MN; the shipped viscoelastic polyester in
15 m surge 2.69 MN against 1.80 MN; a semi-taut chain–polyester leg 9.5 MN at the anchor
against 4.6 MN; and a line on a stiff seabed failed the plausibility guard. The step now blends
Expand All @@ -393,10 +483,21 @@ All notable changes to CableDyn are recorded in this file. The format follows
the dashpot's end share, so that the whole blend of a segment ending slack returns at most its
strain energy unless that end share alone returns more (in the test chain it returned up to
2750 times that with only the elastic share capped); a damped segment that ends taut beyond
that band keeps the plain force blend. The seabed contact of
that band keeps the plain force blend. A damped segment slack at the step start keeps its
step-start dashpot force under a cap too, so that this kept share returns at most the
dashpot's own one-step dissipation `T_d,n |l̇_n| Δt` plus the end share's credit (up to 46
times that in the test chain before). The seabed contact of
a node that lifts off within a step is not capped (see the manual's theory page). At the
default `rhoInf` the reproducers now give 0.10 / 0.09 MN, 0.092 / 0.106 MN, 0.075 / 0.043 MN
and 1.80 MN, as their references, and the stiff-seabed line runs to completion; the
default `rhoInf` the light cable now peaks at 0.10 / 0.09 MN at `dtM` 0.0025 s and at 0.092 /
0.106 MN at 0.1 s, the same cable in the JONSWAP sea at 0.075 / 0.043 MN and the
viscoelastic polyester at 1.80 MN, as at 0.0025 s; the stiff-seabed line runs to completion.
The light-cable figures are not converged peaks. The undamped `EI = 0` deck has no converged
snap peak: the peak rises as `dtM` falls (0.32 to 0.46 MN at microsecond steps, also with an
explicit lumped-mass solver), as single segments rattle, and falls with mesh refinement at a
fixed `dtM`; its 20 ms averages converge to 0.083 / 0.047 MN, the slack-to-taut event, and
converged finite-EI runs give 0.093 MN at the fairlead and 0.057 MN at the anchor. On the
stiff seabed the `EI = 0` peak does not converge either (0.74 to 0.95 MN at coarse steps,
1.2 to 1.4 MN at 0.025 to 0.05 ms). The
chain–polyester anchor snap peaks at 5.7 MN at `dtM` 0.05 s, a single-step peak (4.6 MN at
0.0025 s) that the new snap note reports. On a linear element the scheme is unchanged, so taut
lines keep their accuracy: every MoorDyn-C, OrcaFlex, analytic and experimental comparison that
Expand All @@ -410,7 +511,10 @@ All notable changes to CableDyn are recorded in this file. The format follows
`EI = 0` dynamic decks (one more force evaluation per step and about 4 % more Newton
iterations). `False alpha_force_blend`, which until now applied to finite-EI lines only,
selects the former scheme for `EI = 0` lines as well. Lines that the C API builds without a
deck (`CableDyn_InitLine`, `CableDyn_InitLines`) always take the force blend.
deck (`CableDyn_InitLine`, `CableDyn_InitLines`) always take the force blend. The figures
of this entry take every step as given, as `off ei0_substeps` now does; the rotation- and
snap-resolved sub-steps of the default (under Changed) move the snap peaks among them, for
example the chain–polyester anchor to 4.65 MN and the viscoelastic light cable to 0.062 MN.
- `modified_newton` on `EI = 0` lines (with or without the force blend, so also under
`False alpha_force_blend`) rebuilds the tangent when an iterate reduces the residual by less
than a factor four, as the manual states; it had kept the first tangent for the whole step,
Expand Down
Loading
Loading