Skip to content

feat: scipy 5-DOF optimizer, depowering, and input validation - #58

Merged
marinlauber merged 24 commits into
masterfrom
pr/core-physics
Oct 17, 2026
Merged

feat: scipy 5-DOF optimizer, depowering, and input validation#58
marinlauber merged 24 commits into
masterfrom
pr/core-physics

Conversation

@TAJD

@TAJD TAJD commented Jun 17, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • Replace NLopt with scipy optimizer — resolves Python 3.13 incompatibility
  • Iterative depowering with configurable heel limit (phi_max); fixes solver progression past first solution
  • Leeway fix — remove clamp, clarify radians conversion, drop dead _cl
  • Input validation and error handling across the full VPP stack
  • Comprehensive new test suite: hydro, aero, resistance, API, utils

Part of

Second of four PRs from #55. Base is pr/uv-ci (stacked); merge after #57.

Test plan

  • uv run pytest tests/test_vpp.py tests/test_hydro.py tests/test_aero.py tests/test_resistance.py tests/test_api.py tests/test_utils.py -v
  • Polars output visually unchanged vs main

@TAJD @claude
feat: scipy 5-DOF optimizer, depowering, and input validation
59aeeae 
- Replace NLopt with scipy optimizer for Python 3.13 compatibility
- Add iterative depowering with heel limit enforcement (phi_max)
- Fix leeway clamp and radians conversion in force balance
- Fix depowering solver progression past first phi_max solution
- Add input validation and error handling across VPP stack
- Add comprehensive test suite (hydro, aero, resistance, API, utils)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This was referenced Jun 17, 2026
Merged
Closed
TAJD and others added 3 commits June 17, 2026 20:42
@TAJD
fix: revert api.py to master — ShortKeel belongs in sail-daring PR
1c3ab04
@TAJD @claude
feat: ORC sail variants, ShortKeel appendage, and Daring yacht
b2ac11c 
- Add ORC spinnaker variants and low-performance sail coefficient data
- Add ShortKeel appendage class for hull-integrated keels (Daring class)
- Add Daring 5.5m yacht definition with estimated parameters
- Add Daring baseline performance results and righting moment data
- Add yacht and Daring-specific test suites

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@TAJD
feat: ShortKeel API detection and extended API tests
aab8d20
Base automatically changed from pr/uv-ci to master October 13, 2026 11:26
marinlauber and others added 20 commits October 13, 2026 13:33
@marinlauber @claude
Merge master into pr/core-physics
c34c774 
Resolve conflicts with #51: keep this branch's scipy interpolation and
solver rewrite, restore master's VPPResults loader for the dict format
written by VPP.write.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@marinlauber @claude
docs: update install and run instructions for uv
b428e9d 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@marinlauber @claude
Merge pr/core-physics into pr/sail-daring
96dad5b 
Brings in master (uv/CI, #51) via #58. Conflicts in SailMod, YachtMod
and api.py resolved in favour of this branch, which supersedes master's
Akima interpolation and whitespace changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@marinlauber @claude
Remove Daring-specific example, tests and polar data
bb7c70a 
ShortKeel stays as a generic low-aspect-ratio appendage; its unit tests
move to test_yacht.py.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@marinlauber @claude
Remove unused polars_YD41.json results file
c42a327 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@marinlauber
Merge pull request #59 from marinlauber/pr/sail-daring
5da143e 
feat: ORC sail variants, ShortKeel appendage, and Daring yacht
@marinlauber @claude
fix: default hull roughness to smooth, as in the ORC VPP
148bce2 
The ITTC 1978 roughness allowance is calibrated on ships; at 150 um it
added ~80% to the friction of a 12 m yacht and slowed light-air speeds
by 1-1.5 kn versus master and WinVPP. The ORC VPP has no roughness term,
so default to 0 (Yacht and API) and keep it as an opt-in.

Also use the passed length in the allowance (appendages used the hull
length) and label the friction line as Hughes, which is what it is.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@marinlauber @claude
fix: don't pad polar curves over unsolved sail points
65ecd4a 
_get_cross padded each sail's fastest-TWA range by two indices, which
could reach TWAs where the sail is skipped (e.g. kites at TWA <= lim_up)
and still stores zeros, so the curve dropped to the origin (MN1 + A5 at
18 kn in runVPP). Only pad over points with a non-zero speed, and make
the upper padding match the lower one.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@marinlauber @claude
fix: realistic A5 dimensions in the YD-41 example
fd460a3 
The A5 had vce=2.75 m (the A2 is 9.55 m; ORC gives ~0.565*ISP), so it
produced under a third of the heeling moment and won down to 64 deg TWA
in a breeze. Use area=110 m^2, vce=9.0 m, which restricts it to the
~100-125 deg heavy-air reaching band. Regenerate the example outputs
and the README figure.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@marinlauber @claude
fix: keep rudder side-force factor at 2 above 30 deg heel
3a56892 
Rudder._Ksff ramped from 1 to 2 up to 30 deg heel and then dropped back
to 1, halving the rudder lift slope past 30 deg. The jump made the
equilibrium discontinuous and trapped the optimiser at exactly 30 deg.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@marinlauber @claude
feat: maximise boat speed over depowering with SLSQP
b765693 
Replace the flatten-then-reef heuristic with a single point solver that
maximises boat speed over [vb, phi, leeway, flat, red] subject to scaled,
unsquared force/moment equilibrium and phi <= phi_max. The old solver
took the first setting under the heel limit rather than the fastest one,
so boat speed could drop as TWS rose (up to 0.6 kn at 14->16 kn), and it
stored non-equilibrium points.

- warm start along TWA, keep the fastest converged guess
- check convergence; sail sets that cannot be carried are left at 0
- sequential and parallel share the same row solver (identical results)
- 'iterative' and '5dof' remain as aliases of 'sequential'
- drop tqdm, no longer used
- tests: speed monotonic in TWS, stored points in equilibrium,
  infeasible kite left unsolved

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@marinlauber @claude
fix: plot sail chart TWS in knots, not m/s
44ec568 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JN4y2NWa8kFMBZ5FV92jkj
@marinlauber @claude
fix: draw TWS legend in black, colours mean sail sets
8c8bf3b 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JN4y2NWa8kFMBZ5FV92jkj
@marinlauber @claude
feat: align aero and stability models with the ORC VPP 2026
f50a15a 
Bugs fixed against the documentation:
- heeling moment uses the full heeling force (no extra cos(heel)) and the
  centre of effort height above the waterline, HBI + ZCE (5.56); the main
  and kite CE were measured from the deck, missing the freeboard
- effective span coefficient uses 0.075 * overlap (5.42), not 0.0075
- hull windage blends head-on to side area with the AWA in the right
  units, and is no longer scaled by cos(AWA) twice
- dynamic righting moment follows eq. 4.32 (incl. the -2.1 term and SLR
  in knots/sqrt(ft)); it was ~0 N.m
- fractionality uses the reefed P and is capped at 1

ORC 2024-2026 changes:
- minimum flat 0.53
- twist function (5.40) with the overlap term; the 1/3 factor follows
  Figure 5.14, the printed equation gives no lowering at 150% overlap
- jib twist delta CEH removed, as in the 2026 document
- kites depower by reef: area * reef^2, CE * reef, reef >= Kite.min_reef
  (default 0.85)
- reduced jib keeps its full hoist down to LPG = J (luff no longer goes
  to zero with ftj)
- sail coefficient tables updated to ORC 2026 (kites, low main/jib sets);
  the default kite is the symmetric spinnaker

Also adds a low-speed fallback start for the first point of a row.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JN4y2NWa8kFMBZ5FV92jkj
@marinlauber @claude
feat: save figures as PDF for sharper curves
1100109 
Figure.png stays as the README image, since GitHub cannot inline a PDF.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JN4y2NWa8kFMBZ5FV92jkj
@marinlauber @claude
feat: cap heel under spinnaker at the ORC 21.5 deg
cbe0ef4 
Kites take a max_heel (default 21.5) which tightens the heel bound of
the equilibrium solve; ORC models it as a soft resistance boundary.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JN4y2NWa8kFMBZ5FV92jkj
@marinlauber @claude
feat: heeled wind triangle with the ORC log wind gradient
e2237ec 
TWS is given at 10 m and scaled with a log profile (z0 = 0.005 m) to
the sail plan's centre of effort, and to half the freeboard for hull
windage. The cross-track true wind is projected on the heeled mast
plane with cos(heel) (ORC VPP 2026, 7.1).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JN4y2NWa8kFMBZ5FV92jkj
@marinlauber @claude
feat: ORC appendage drag and induced drag corrections
490b996 
- Rudder lift area ramps from 0 upright to twice its area at 30 deg.
- Induced drag heel multiplier (ORC Fig. 6.10), faded out with the
  drive to side force ratio from 0.2 to 0.4.
- Keel root free surface and funsteady factors on the appendage
  effective span (ORC 6.5.1); HydroMod.update takes an optional tws.
- Appendage friction from five strips with the ORC friction line, the
  t/c form factor and lift-dependent drag (ORC 6.1.2).
- Bulbs no longer carry side force or double their wetted area.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JN4y2NWa8kFMBZ5FV92jkj
@marinlauber @claude
fix: appendage aspect ratio span^2/area and lifting-line lift slope
249c684 
Ar was span/area (dimensional). It is now span^2/area, doubled for the
hull reflection, and the lift slope uses 2*pi*Ae/(Ae+2), which grows with
aspect ratio. The short keel Jones slope uses the same Ae.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JN4y2NWa8kFMBZ5FV92jkj
@marinlauber @claude
chore: release 1.0.0, drop module headers and unused verbose arg
44a96d5 
- bump version to 1.0.0 and update the author email
- remove the per-module __author__/__version__ headers
- drop the unused verbose argument from VPP.run and its callers
- README: document the PDF outputs, benchmark figures and appendage lift

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JN4y2NWa8kFMBZ5FV92jkj
@marinlauber
marinlauber merged commit bd97605 into master Oct 17, 2026
3 checks passed
@marinlauber
marinlauber deleted the pr/core-physics branch October 17, 2026 08:49
@marinlauber

marinlauber commented Oct 17, 2026

Copy link
Copy Markdown
Owner

This fixes most parts of #55, I'll close the old one.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@marinlauber marinlauber marinlauber approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@TAJD @marinlauber