Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: Solutions obtained from a numerical method based on Prandtl’s lifting-line theory, valid for multiple lift-
Received 26 July 2012 ing surfaces with arbitrary sweep, are obtained for a number of rigid wing and sail geometries. The results
Received in revised form 26 March 2013 are compared against solutions obtained using established vortex-lattice methods, and computational
Accepted 17 April 2013
fluid dynamics solutions to the Euler equations. For the case of an untwisted, rectangular wing, numerical
Available online 3 May 2013
lifting-line, vortex-lattice, and Euler solutions were all in reasonable agreement. However, the numerical
lifting-line method was the only method to predict the constant ratio of induced-drag coefficient to lift
Keywords:
coefficient squared, which has been predicted from the analytic solution and confirmed by well estab-
Potential flow
Aerodynamics
lished experimental data. Results are also presented for a forward-swept, tapered wing. Additional results
Lifting line theory are presented in terms of lift and induced-drag coefficients for an isolated mainsail, and mainsail/jib com-
binations with sails representative of both a standard and tall rig Catalina 27. The influence of the non-
linear terms in the lifting-line solution appears minimal, with the exception of mainsail results when
considering jib/mainsail combinations.
Ó 2013 Elsevier Ltd. All rights reserved.
1. Introduction equations. These include lifting-line methods (c.f., Refs. [1–4]), pa-
nel methods (c.f., Ref. [5]), and vortex-lattice methods (c.f., Ref.
A wide range of computational tools are available for the analysis [6]). In the present work, potential flow solutions are obtained
of fluid flow about three-dimensional sails, including solutions to using numerical lifting-line method (NLLM) [4] and a vortex-lattice
the Reynolds-averaged Navier–Stokes (RANS) equations. However, method (VLM) [6]. Potential flow solutions also find utility when
for wide-ranging parameter studies and preliminary design, pre- coupled to boundary-layer equations, where overall solution times
mium is placed on ease of solution, accuracy, and minimizing the are significantly below those required to solve the full Reynolds-
required computational costs. Three-dimensional RANS solutions averaged Navier–Stokes equations.
are computationally expensive, and hence may not represent the Lifting-line theory was first developed by Prandtl [1,2] in 1918
ideal tool for preliminary design studies, particularly for sails in up- for a single lifting surface with no sweep or dihedral and has been
wind configurations for which the flow remains primarily attached. used for the analysis of isolated sails by a number of authors [7–9].
Solutions to the Euler equations, which govern the velocity and The theory was extended to a pair of parallel lifting lines based on
pressure fields for a fluid in which viscous forces are negligible, Munk’s equivalence theorem [3] and solutions were later pre-
represent a viable alternative. There are several approaches to sented in Glauert [10]. Subsequently, Milgram [11] applied the the-
obtaining solutions to the Euler equations. One involves a direct ory to the analysis of multiple sails. However, the series solution
computational fluid dynamics (CFD) solution to the equations method presented by Milgram [11] requires that the lifting lines
which is still a relatively expensive endeavor. However, if an invis- are parallel, that the sails are without sweep, and have an elliptic
cid, constant density fluid flow about a body originates from a irro- lift distribution. More recently Berbente and Maraloi [12] applied
tational flow, such as a uniform flow, then Kelvin’s theorem shows Prandtl’s lifting-line theory to multiple sails with fewer restrictions
the fluid will remain irrotational everywhere. This permits the on the configuration. However, their approach still utilizes an infi-
introduction of a velocity potential with the resulting potential nite series solution, which is strictly valid only for a single, un-
flow equation which may be solved in lieu of the Euler equations. swept lifting surface. As such, multiple sails were modeled by
There are a number of approaches to solving the potential flow Berbente and Maraloi [12] as a single hybrid lifting surface, which
does not account for three-dimensional mainsail/jib interactions.
This is particularly problematic for fractional rigs, which were
⇑ Corresponding author. Tel.: +1 435 797 2878.
not considered by Berbente and Maraloi [12]. Consequently, the
E-mail address: spall@engineering.usu.edu (R.E. Spall).
0045-7930/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compfluid.2013.04.020
30 R.E. Spall et al. / Computers & Fluids 82 (2013) 29–37
Nomenclature
above lifting-line approaches are not ideal for the general analysis strengths Gi ¼ Ci =ðci V 1 Þ for each horseshoe vortex and the proper-
of multiple sails. ties of the sail given as:
Phillips and Snyder [4] developed a numerical method for !
XN
obtaining solutions to Prandtl’s lifting-line equation for systems e Li ¼ 0
2 u1 þ
v ji Gj fi Gi C
ð1Þ
of lifting surfaces of arbitrary location and orientation. However, j¼1
their application was for wings of significant thickness and rela-
tively small camber, not the geometries that characterize sails: where
very thin, swept, and with large camber. The application to sail " P #
ðu1 þ Nj¼1 v ji Gj Þ uni
analysis is significant as this extension permits the modeling of ai ¼ tan1 P ð2Þ
multiple sails using Prandtl’s lifting-line theory without restric- ðu1 þ Nj¼1 v ji Gj Þ uai
tions regarding sweep. In addition, the three-dimensional interac- e Li is a function of the local angle of
and the section lift coefficient C
tions between the upwind and downwind sails are properly
attack. Eq. (1) is applied at N control points located on the quarter
accounted for. The purpose of this work is to evaluate the viability
chord resulting in a nonlinear system of N equations and N un-
of the lifting-line method of Phillips and Snyder for preliminary sail
knowns which may be rapidly solved using Newton’s method.
analysis. Results are compared with solutions resulting from vor-
If a small-angle-of-attack approximation for both the geometric
tex-lattice methods and Euler CFD solutions.
and induced angles of attack is imposed, and all 2nd-order terms
Commonly, vortex-lattice methods with only a single lattice
are neglected, Eq. (1) reduces to:
element in the chordwise direction are termed lifting-line meth-
ods. For instance, Milgram [11], in addition to series solutions of X
N
Prandtl’s lifting-line theory, also utilized this approach for the anal-
e Li;a
2ju1 fi jGi C v ji uni Gj ¼ Ce Li;a ðu1 uni aL0i Þ ð3Þ
ysis of jib/mainsail interactions, and termed the procedure a lift- j¼1
ing-line method. More recently Bertin and Cummings [13] stated The above equation requires the lift-slope coefficient and the
that the lifting-line method of Phillips and Snyder [4] is equivalent zero-lift angle of attack at each section, and is linear in the vortex
to vortex-lattice methods that use only a single lattice element in strength Gi. It is expected to provide good results for slightly swept,
the chordwise direction. However, calling an under-resolved vor- high aspect ratio sails at small angles of attack. The total nondi-
tex-lattice method a lifting-line method seems questionable at mensional aerodynamic force vector is computed from the
best, as the vortex-lattice method is not based on Prandtl’s lift- relation:
ing-line theory. As described in the following section, the method !
of Phillips and Snyder [4], and a single chordwise element vor- Fa XN X
N
dSi
¼2 G i u1 þ Gi Gj v ji fi ð4Þ
tex-lattice method, are fundamentally quite distinct. 1
qV 21 Sr Sr
2 i¼1 j¼1
are available in the AVL code to impose an angle of attack. The user where
may align the mean chord line with the x-axis and adjust the angle
of the free stream, or align the free stream with the x-axis and ad-
A
B¼ ð6Þ
just the angle of the chord line. In either case, both the bound and ðAV þ 2ÞðAV þ 1Þ
trailing vorticity are always oriented parallel to the x-axis. Both ap-
and
proaches give similar results at low angles of attack.
We again note that the vortex-lattice method, using a single AR
vortex-lattice element in the chordwise direction, is not equivalent C¼ B ð7Þ
6
to the lifting-line method of Phillips and Snyder [4]. In particular,
In addition, AV and AR are related to the entrance angle at the
with the VLM, using a single element in the chordwise direction,
leading and trailing edges, respectively. The constant K is deter-
the bound vortex and control points are (usually) placed at one-
mined to provide a pre-determined maximum depth y ^max at the
and three-quarter chord locations, respectively. A zero normal
location where dy ^=d^
x ¼ 0. This involves finding the root of the
velocity (Neumann) condition is specified at the control point
equation dy ^=d^
x ¼ 0, and substituting the resulting value for ^ x,
which provides the system of N equations and N unknowns. Conse-
along with the desired y ^max , into Eq. (5) to solve for K. The constant
quently, with a single vortex-lattice element in the chordwise
A alters the distribution of sail fullness fore and aft. Laine and Laine
direction the solution is dependent only on the position and slope
recommend A = 1 for sails used in light air conditions, and
of the sail at the control point. Sail curvature upstream or down-
A = 1 + AV/4, which provides for more fullness forward and a flatter
stream of this control point does not influence the solution. With
leech, for sails used in heavier air (and represents the definition
the lifting-line method of Phillips and Snyder [4], the chordwise
used in this work).
variation in the shape of the sail is taken into account through sec-
The three dimensional sail shape is developed by specifying sec-
tion lift coefficients as indicated in Eq. (1), or section lift coefficient
tion shapes at the foot, approximate vertical mid-point, and head
slopes as indicated in Eq. (3). These values may be available
of the sail. Quadratic interpolation is used to determine necessary
through experimental data, or, in the present work, computed
values of AV, AR, and the depth y ^max to define intermediate section
using either thin airfoil theory or a two-dimensional vortex panel
profiles.
method. The means by which these are acquired influences the
execution speed of the code.
Solutions to the Euler equations using the general purpose, 4. Results
unstructured mesh CFD solver STAR-CCM+ [14] are also presented.
However, we note that solutions to the Euler equations will always Results are first presented for two different thin wings, distin-
be subject to the influences of numerical diffusion, and hence will guished by their taper ratio, in which the section shapes were de-
not represent true potential flow solutions. fined by a NACA 9400 airfoil. Subsequently, we present results for
In terms of computational efficiency, Phillips and Snyder [4] an isolated mainsail, and finally two mainsail/jib combinations,
performed a careful analysis of computation time for the numerical with section shapes defined by the procedure of Laine and Laine
lifting-line method, a panel method, and a structured CFD Euler [15]. The mainsail and jib dimensions are representative of those
solver. Their conclusions were that the nonlinear lifting-line solu- for both a standard and tall rig Catalina 27. Results for all cases
tions were four and six orders of magnitude faster than the panel are presented for lifting-line, vortex-lattice, and Euler solutions.
method and CFD Euler solvers, respectively. However, we caution The ranges of angles of attack presented include levels above and
that computation time for codes is very subjective, varying be- below what would be feasible for a flexible sail. This is done simply
tween methods with respect to the size of the system of unknowns to extend the range of comparison between the different solution
(e.g., grid resolution), levels of convergence specified, compilers methodologies.
and optimization levels, and programming skills. In terms of the We first consider an isolated, thin, rectangular wing of aspect
lifting-line method as coded for the current work results for the ratio 3 with section shapes defined by a NACA 9400 airfoil. For
solutions to the linearized equations are achieved in less than the Euler solutions, the sail was located within a rectangular box
0.1 s. The nonlinear solver employing a series solution for section defining the outer boundaries of the computational domain with
lift coefficients converges in approximately 1 s. If a two-dimen- the boat centerline aligned along the x-axis. The dimensions are gi-
sional panel method is utilized to obtain the section lift coeffi- ven as 20 m 6 x 6 21 m, 20 m 6 y 6 20 m, 20 m 6 z 6 23 m,
cients, the compute time increases to several minutes, depending where the leading edge (luff) of the sail was located at x = 0 m,
on the streamwise resolution of the panel method solver. Although the lateral location of the luff at y = 0 m, and the foot of the sail
not utilized in the present work, should experimental values of the at z = 0 m. Consequently, the influence of the water surface is not
section lift coefficients be available, compute time would be on the accounted for. Symmetry boundary conditions were imposed
order of 1 s. around the perimeter, while a fixed (uniform) velocity condition
u = 1 m/s was imposed at the inlet boundary (x = 20 m). Zero-
3. Sail shapes derivative boundary conditions were imposed at the outlet bound-
ary (x = 21 m). Angles of attack were changed by rotating the wing
Two approaches to define section camber shapes were followed. about its leading edge.
Initial calculations for isolated rectangular and tapered wings are The steady Euler equations were solved using the unstructured,
presented for camber lines derived from NACA 9400 airfoil sec- general purpose CFD code STAR-CCM+ [14]. Second-order upwind-
tions. This provides a maximum camber of 9%, occurring at 40% ing interpolation was applied to the convective terms and the SIM-
chord, representing a reasonable approximation of a section profile PLE procedure was employed for pressure–velocity coupling. A
for a generic mainsail. For the Catalina 27 sails, section shapes fol- segregated multi-grid solver was employed and solutions were
lowing the work of Laine and Laine [15] were used. In particular, considered converged when all residuals were decreased to
the equation giving the sail depth ðy^Þ at any point ^
x along a section approximately 106, and lift and induced-drag coefficients no long-
was given by: er changed with further iterations. Solutions were obtained on
!
Að1 ^xÞAV þ2 AR 3 three different mesh sizes for each angle of attack studied. In gen-
^ð^xÞ ¼ K
y ^x þ C ^x þ B ð5Þ eral, mesh sizes ranged from approximately 700,000 to 5.5 million
ðAV þ 2ÞðAV þ 1Þ 6
cells for the coarsest to finest meshes, respectively. Results pre-
32 R.E. Spall et al. / Computers & Fluids 82 (2013) 29–37
Lift coefficients for the jib are shown in Fig. 14. These results
show smaller differences between the different methods relative
to the mainsail results, likely due to the fact that the jib is not in
the wake of the mainsail. The linear lifting-line method predicts
larger lift values than does the nonlinear method which is opposite
the trend for the mainsail. Higher lift values on the jib result in
greater downwash on the main, which results in lower lift values,
and vice versa. The results for low wind angles relative to the boat
centerline that are shown in Fig. 14 are not realistic for flexible
sails, because the jib would luff under the influence of the negative
lift. The Euler solutions predict lower values for the lift coefficient
than the other methods, and reveal a large drop-off at the higher
angles of attack. This is due to separation at the leading edge, a re-
sult of numerical viscosity.
The induced-drag polars for the mainsail are shown in Fig. 15.
The figure reveals that for a given lift coefficient, the predicted in-
duced-drag coefficients are highest for the Euler and vortex-lattice
solutions, and lowest for the nonlinear lifting-line method. The fig-
ure also reveals a negative induced-drag coefficient for the nonlin-
ear lifting-line method at small angles of attack. This may at first
seem untenable, however, the overall induced-drag coefficient
(mainsail and jib combined) is positive. The negative mainsail in-
duced-drag shown in Fig. 15 is a result of upwash induced on the
Fig. 12. Mainsail/jib combination with coarse Euler solution surface mesh. mainsail by negative lift on the jib.
36 R.E. Spall et al. / Computers & Fluids 82 (2013) 29–37
Fig. 15. Drag polars for mainsail. Fig. 17. Comparison of lift coefficients between tall and standard rig sails computed
using nonlinear lifting-line method.
Shown in Fig. 16 are the induced-drag polars for the jib. The re- at a given lift coefficient than does the tall rig. The opposite trend
sults for the linear and nonlinear lifting-line methods are in good is observed for the jib. Again, these results are consistent with the
agreement. The vortex-lattice method predicts a smaller induced- differing aspect ratios of the sails. Results for the combined sails
drag coefficient at a given lift coefficient, than either of the lift- are also shown. Here it is clear that overall, the tall rig has a lower
ing-line methods. The induced-drag coefficients at a given lift coef- induced-drag coefficient for a given lift coefficient, and hence one
ficient are much higher for the Euler results. This is again the would expect improved sailing performance.
influence of numerical viscosity errors on the solution, which can-
not be completely eliminated regardless of the mesh density. 5. Conclusions
The final set of results considers only the nonlinear lifting-line
method in a comparison between standard and tall rig Catalina A numerical method based on Prandtl’s lifting-line theory has
27 configurations. The aspect ratios for the tall rig mainsail and been applied to isolated and multiple thin-sail lifting surfaces
jib are 5.92 and 5.9, respectively. Consequently, the aspect ratio and the results compared with solutions obtained using a vortex-
of the mainsail is increased compared to the standard rig, but the lattice method and CFD solutions to the Euler equations. Where
aspect ratio of the jib is decreased. Results in terms of lift coeffi- solution behavior was known, such as for an untwisted, rectangu-
cients and induced-drag polars are shown in Figs. 17 and 18, lar wing, the lifting-line method performed quite well. In fact, the
respectively. In terms of lift coefficients for the mainsail, the tall numerical lifting-line method was the only method to predict a
rig values are slightly higher than the standard rig. For the jib, constant ratio of induced-drag coefficient to lift coefficient
the lift coefficients for the standard rig are slightly higher than squared. The Euler solutions suffered from the effects of numerical
the tall rig. These results are consistent with the differing aspect diffusion, resulting in artificial flow separations at angles of attack
ratios for the tall and standard rig sails. In terms of the induced- that require rapid acceleration around the sail leading edge. The
drag polars, the standard rig mainsail has a higher induced-drag vortex-lattice solutions obtained using the AVL code were limited
R.E. Spall et al. / Computers & Fluids 82 (2013) 29–37 37
to small angles of attack due to the alignment of the bound and References
trailing vorticity with the x-axis. This is not a limitation of the
vortex-lattice method in general, but is a result of a linearization [1] Prandtl L. Tragflügel Theorie. Nachrichten von der Gesellschaft der
Wisseschaften zu Göttingen, Geschäeftliche Mitteilungen, Klasse; 1918.
procedure used in the AVL code. The linear and nonlinear lifting- [2] Prandtl L. Theory of lifting surfaces. NACA TN 9; July 1920.
line results are quite comparable for isolated sails or for the lead [3] Prandtl L. Induced drag of multiplanes. NACA TN 182; July 1920.
sail when both the jib and mainsail are considered. However, for [4] Phillips WF, Snyder DO. Modern adaptation of Prandtl’s classic lifting-line
theory. J Aircraft 2000;37(4):662–70.
the jib/main combination, the results for linear and nonlinear [5] Phillips WF. Mechanics of flight. 2nd ed.. John Wiley and Sons Inc.; 2010.
lifting-line method differ significantly for the mainsail. Nonlinear [6] Drela M. <http://web.mit.edu/drela/Public/web/avl>.
lifting-line method was also applied to compare the difference in [7] Wood CJ, Tan SH. Towards an optimal yacht sail. J Fluid Mech 1978;85:459–77.
[8] Sugimoto T. A first course in optimum design of yacht sails. Fluid Dyn Res
lift and induced-drag coefficients between standard and tall rig
1993;11:153–70.
Catalina 27 sails. The results indicated an increase in the lift [9] Jackson PS. Modeling the aerodynamics of upwind sails. J Wind Eng Ind
coefficient at a given angle of attack for the mainsail, but a de- Aerodyn 1996;63:17–34.
creased value for the jib. This is consistent with the larger mainsail [10] Glauert H. The elements of airfoil and airscrew theory. 2nd ed. Cambridge
Science Classics; 1947.
aspect ratio for the tall rig relative to the standard rig, but a smaller [11] Milgram JH. The analytical design of yacht sails. Trans – Soc Naval Architects
aspect ratio for the jib. The tall rig displays a lower induced-drag Mar Eng 1968;76:118–60.
coefficient for a given lift coefficient, which should improve sailing [12] Berbente C, Maraloi C. Theoretical and experimental research regarding the
aerodynamics of a ship sail system. UPB Sci Bull Ser D 2006;68:17–32.
performance. Overall conclusions are that the accuracy realized [13] Bertin J, Cummings R. Aerodynamics for engineers. 5th ed. Prentice-Hall; 2009.
from the numerical lifting-line method is as good as or better than [14] CD-Adapco. STAR-CCM+ users manual; 2011.
that obtained from the vortex-lattice method or inviscid CFD solu- [15] Laine R, Laine J. The sailcut CAD handbook. <http://sailcut.sourceforge.net/
docs/en>; 2009.
tions, but particularly for the linearized equations, at a small frac- [16] Gentry A. The aerodynamics of sail interaction. In: Proceedings of the third
tion of the computational cost. AIAA symposium on aero/hydronautics of sailing, Redondo Beach, CA;
November 1971.