The Fujin experiment

Updated March 29, 2021

------------------ WORK IN PROGRESS ------------------


Back to main table of contents

The Fujin experiment

The purpose of this page is to present a comparison of flow5 predictions to the experimental data published in

  • [1] Tahara, Y. Masuyama, T. Fukasawa and M. Katori (2012).
    Sail Performance Analysis of Sailing Yachts by Numerical Calculations and Experiments, Fluid Dynamics, Computational Modeling and Applications, Dr. L. Hector Juarez (Ed.), ISBN: 978-953-51-0052-2, InTech
  • [2] Yusuke Tahara and Toichi Fukasawa.
    Database of sail shapes versus sail performance and validation of numerical calculations for the upwind condition, Journal of Marine Science and Technology, June 2009.

The authors have carried a comprehensive campaign of performance measurements abord their sailboat "Fujin" in real sailing conditions for different configurations of main sail and jib. The full sets of geometric data and experimental results are summarized in their publication.

The test cases that have been modelled and simulated in flow5 are

  • Case 9807172B
  • Case 98110105
  • Case 96092335
  • Case 96080248
This selection has been made to cover a large range of sail and wind configurations.

The Fujin - partial extract from Fig. 1 in ref. [1]

flow5 calculations

The calculations were performed with flow5 v7.08.
The corresponding project file can be downloaded here: Fujin.fl5.


Both sail have been modelled using the spline type sail, with each horizontal section defined as a cubic spline interpolating the geometric points measured during the experiment.

Although shown in the images, the hull was excluded from the analysis since the intersection of hull and water is not managed in flow5 v7.07.


The mesh of both sails is of the free type, preferred for the better quality of the triangular panels it provides at the gaff.

Sensivity analyses were performed with decreasing triangle sizes and increasing panel count. Also a test with a ruled mesh has been carried out in one case for comparison. All results are presented below.


Panel method

The analysis is of the triangular uniform panel type. This method is the recommended type in flow5 for its robustness, speed and versatility.
The reasons for this preference are that the linear method is not significantly more accurate, and that the errors due to the approximative wake model are significantly higher than the precision of the panel method.
A comparative analysis of the two panel orders has nonetheless been carried out and the results are presented below.

Wake model

The wake model is the Vortex Particle Wake (VPW) which is the recommended type in flow5 whenever accuracy is preferred over analysis times.

The settings of the VPW have been adjusted to minimize the vorton core size while maintaining a stable wake shape, in accordance with these recommendations.

Wind gradient

The wind gradient parabola has been adjusted to represent approximately the gradient presented in Figure 7 of ref. [2].

Extra drag

Since the sail calculations in flow5 are of the inviscid type, the additional component given in equation 2 of ref. [2] has been added to the results.



The results presented below are those obtained with flow5 7.08 and using the meshes and the analysis setttings defined in the project file. The intent is to build experience with the sail module and try to improve these predictions.


The moments acting on the boat are not given explicitely in the publications, but are instead used to calculate the position of the Center of Effort (CE) using the following formulas.

Test case 96092335

AWA (°) Twist (°) Draft (%) AWS (m/s) Heel (°) VB (kt)
30.7 15.5 8.6 6.9 15.1 5.0

This is a standard case with average AWA. No simple explanation could be found for the difference between the measured and predicted lift coefficients.

This test case was used to evaluate the influence of the analysis settings. It shows that the results are not too sensitive to

  • the number of panels and the type of mesh: free or ruled
  • the type of surface densities, i.e. uniform or linear
  • the wake model
It is difficult to say which settings give the best results and to give recommendations given the uncertainties on both the measurement and the calculations.

Measured 1.44 0.28 0.50 1.39 0.41 4.17
free mesh
1359 uniform triangles
1.15 0.14 -0.54 1.02 0.89 4.63
free mesh
2078 uniform triangles
1.18 0.14 -0.55 1.05 0.80 4.36
free mesh
2078 linear triangles
1.20 0.15 -0.57 1.07 0.71 4.29
free mesh
2078 linear triangles
Flat wake
1.18 0.17 -0.53 1.05 0.38 4.04
ruled mesh
1188 uniform triangles
1.16 0.14 -0.54 1.02 0.90 4.72

Test case 9807172B

AWA (°) Twist (°) Draft (%) AWS (m/s) Heel (°) VB (kt)
29.8 10.9 9.3 7.2 8.3 4.2

This configuration was with the main sail only and no jib. The authors caution in [2] that the measurement of the AWA was made uncertain due to difficulties to steer the boat, and that flow separation on the sail was greater than in the case with the jib. The results should therefore be considered with caution.

Measured 1.25 0.45 0.19 1.31 1.68 5.86
free mesh
1485 uniform triangles
1.17 0.14 0.57 1.04 1.90 5.97

Test case 98110105

AWA (°) Twist (°) Draft (%) AWS (m/s) Heel (°) VB (kt)
20.5 14.5 7.9 8.6 11.6 4.8

This is the measurement performed with the lowest AWA, i.e. in theory the condition closest to those of the panel method's ideal potential flow.

Measured 1.15 0.20 0.22 1.15 0.65 4.73
free mesh
1376 uniform triangles
1.19 0.12 -0.36 1.14 0.86 5.59

Test case 96080248

AWA (°) Twist (°) Draft (%) AWS (m/s) Heel (°) VB (kt)
37.9 14.5 7.2 7.5 19.6 6.0

This test case is for a large angle of attack, for which the authors state that both the main sail and the jib were not eased sufficiently. This may have caused flow conditions far off from the ideal laminar flow and hence the results are not significant.

Measured 1.58 0.45 0.62 1.52 0.34 4.17
free mesh
2088 uniform triangles
1.51 0.19 -0.86 1.24 1.02 4.61
Back to top

Analysis of the results

The first and main observation is that the predictions are not as close to the measurements as they are in the case of the Jibe 2 experiment. There are several potential explanations for this difference.

On the measurement side:

The details of how these measurements were performed and of the difficulties involved are given in the two publications in reference.

More importantly, the cause of the discrepancies is likely to be found on the analysis side:

As mentioned above the test case 98110105, is the one most likely to give results closest to the predictions, with indeed the measured and predicted lift coefficients close within a few percent. This however may not be significant since flow separation is likely to be present at this AWA, which is confirmed by the under-prediction of the drag coefficient.

Conclusion and recommendations

This study tends to show that flow separation plays a significant role in the lift and drag of sails, something which panel methods such as those implemented in flow5 do not account for. This causes the predictions to be relatively imprecise.

flow5 should only be used to analyze cases where flow separation is limited, which corresponds to upwind conditions with low AWA.

Note: the author will be very interested in, and thankful for, other numerical or experimental test cases which could be used to validate or improve the predictions. For users with access to such test cases and willing to share the experience, a license can eventually be provided free of charge.

Back to top