Ocean Engineering 33 (2006) 234246 www.elsevier.

com/locate/oceaneng

Parametric design of sailing hull shapes

Abstract The author presents an original algorithm aimed at automatically generating the hull shape of a sailing yacht starting from an initial set of parameters. The procedure consists of two steps. First one keel line and a Designed Water Line (DWL in the following) are faired according to a set of parameters, say length of water line, canoe body draft, stem angle and some adimensional coefcients. This information is then used to fair the hull surface, which must in turn full more prerequisites (parameters like displaty cement, oatation area and related coefcients). The hull is dened by means of a B-spline surface, the fairing of which is ensured by allowing for all the imposed objectives and constraints. An optimisation technique based upon the gradient method ensures that a reliable solution is obtained in a very short time. q 2005 Elsevier Ltd. All rights reserved.
Keywords: CAD; Yacht design; Surface fairing; Optimisation

1. Introduction Hull fairing is a difcult task, a problem, though, which can be solved elegantly and efciently, thanks to the evolution of new technologies. Naval designers of the past tried to set up the so called Hull Equation, an expression that should be able to give a mathematical description of the hull. To solve such a problem in a closed form, however, is complicated, rst of all because describing all possible hulls by a single equation is
* Tel: C39 091 6657145; fax: C39 091 484344. E-mail address: [email protected].

0029-8018/$ - see front matter q 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.oceaneng.2005.03.007

A. Mancuso / Ocean Engineering 33 (2006) 234246

235

Nomenclature lWL Bmax AS AWL Tc Pc length Water Line maximum Beam maximum transverse Area oatation Area canoe body draft canoe volume displacement

a real challenge. In fact it is well known that the hull shape depends on several parameters, many of which interrelated. Traditionally, the design of a hull (Larsson and Eliasson, 1996) can be obtained by means of two set of mutually perpendicular curves (Fig. 1): the so called waterlines and sections. Waterlines lye on horizontal planes, whereas sections are drawn vertically through the hull, perpendicular to the bow-stern axis. Of course the nal design of the hull can be obtained by a trial and error technique. Such a procedure it is time consuming; it requires a great skill and the whole design can be affected by even a simple modication of a data point on a curve (surface). According to the specic design requirements, the designer tries to endow all the curves with the desired level of continuity and curvature. However, even if all the curves were well faired, this would not yet be enough to state that the entire surface is faired as well. This is why the designer generally works with one more set of curves, the so called buttocks, which are sections through the hull drawn vertically but parallel to the longitudinal plane of symmetry (Fig. 1). These curves help to check the simultaneous fairing of waterlines and sections. Nowadays designers can rely on new means provided by informatics, as CAD technologies, and there are good reasons for that. First of all, a virtual model allows a large number of preliminary calculations (e.g. volumes, weights, inertia moments, etc.) to be made in a short time. This information can also be useful to rene the hull design according to the specications. Solid modellers of the new

Buttock Section

Water line

Fig. 1. Main curves used for the hull denition.

236

A. Mancuso / Ocean Engineering 33 (2006) 234246

generation afford an immediate control of geometric characteristics, such as tangency and curvature, which play an important role in the behaviour and performance of a yacht. By what precedes, the use of parametric curves (Piegl and Tiller, 1997) and surfaces ` zier, B-spline, NURBS) looks like an efcient approach (Mancuso, 2001). A (e.g. Be researcher aims at setting up a means of obtaining a reliable CAD model of the hull surface starting from a set of data. By relaying on a CAD model allows the solution of several problems related to the design and simulation on a virtual model without yet realising a physical prototype. It also allows checking numerically the behaviour of a yacht from the uid dynamic standpoint (Percival et al., 2001; Mancuso and Mancuso, 2004; Dejhalla et al., 2001; Valorani et al., 2003), from the structural one (Cappello and Mancuso, 2001; Augusto and Kawano, 1998) and so on, postponing the more expensive experimental tests which are anyway essential to validate a design. In the literature several applications of these concepts are reported concerning the automatic generation of the hull form. Islam et al. (2001), for instance, apply neural networks to a preliminary evaluation of the shape, and then genetic algorithms to optimise the hull according to a number of parameters. Though this procedure does not require an initial sketch, it needs, however, the use of a surface smoothing technique to ensure a well faired hull. Calkins et al. (2001) develop a tool for chined hull generation in concept design; this may help the designer during the early stage of the project. At any rate it requires the denition of several parameters (ca. 40) and the three-dimensional hull fairing is not ensured. Yilmatz and Kukner (1999), succeed in obtaining cross sectional curves of a desired hull-which requires only a few parameters (ca. 5) to be assigned-by applying a regression technique to a large database of hulls. Harries et al. (2001) have integrated various commercial software to optimise the design of a sailing yacht from both the geometrical and hydrodynamic point of view. In the present paper an optimisation technique is described aiming at automatically designing the immersed part of a sailing yacht hull, on the basis of a few main parameters like length of water line, canoe body draft, stem angle and some adimensional coefcients. The algorithm tries to obtain a good compromise among these parameters by minimising an error function which depends on target values. Initially some of these data are assigned both to keel and DWL. An initial B-spline curve is drawn out and automatically optimised in order to match the assigned data for both the curves. With this information a B-spline surface is set up to be optimised by a second step in the optimisation process. It is no use to perform surface fairing, because this is ensured by boundary conditions and constraints on the design variables. The application will show how the proposed method affords a reliable solution in a very short time.

2. Procedure set up As has been said before, the procedure for the hull fairing goes through two steps. Fig. 2 (top) shows the algorithm ow-chart. During the rst step, assuming a set of data, the boundary curves, DWL and keel, are optimised by means of B-spline curves. These curves are also used to build a B-spline surface of the hull which will be itself optimised by allowing for the remaining data. In the two steps the fundamental structure of

A. Mancuso / Ocean Engineering 33 (2006) 234246

237

Main parameters definition I DWL and Keel optimisation II Hull surface optimisation

Control points

Optimising function
IGES file of the hull

optimiser

new set of control points Optimisation Loop

no
convergence

yes exit

Fig. 2. The two phases of the optimisation process (top); the structure of the common optimisation loop (down).

the optimisation loop (although slightly different) is quite similar to the one depicted in Fig. 2 (down). The design variables of the problem are assumed to be, in both steps, the Cartesian coordinate of the control points. The objective is to minimise a function (optimising function, in the following) depending on target values assigned to the main parameters. The optimiser, according to objective and constraint, attempts to modify the position of the control points. In this way a new set of control points is generated, and a new value of the optimising function can be established. The procedure comes to a halt once convergence is achieved (for a given tolerance) or some maximum number of iterations has been executed. At each main iteration, the algorithm writes the IGES le of the hull (in terms of NURBS curves and surfaces); it is then possible to check the surface evolution during the optimisation process by means of any CAD software. 2.1. Design variables The design variables are different, depending on the kind of optimisation to be performed (i.e. DWL, keel, Hull). For the DWL, if a number of control points, N, is assumed, the total number of variables will be 2!(NK2). In fact, being the lWL a design data, both the rst and last points of the curve are xed; moreover, each control point may move in the x and y directions, z being 0. As for the keel, there are (NK2) variables. In this case the interior control points may only move along z, because the x location has been

238

A. Mancuso / Ocean Engineering 33 (2006) 234246

already established from the DWL. Finally, as regards the hull, the total number of design variables, DV (Fig. 5) is: DV Z 2 ! N K 2 ! M K 1 (1)

Specically, each control point, excluding the boundaries, may move along both the y and z directions, the x direction being xed. The control points thus belong to planes parallel to the yz plane. The boundaries control points may only change along one direction; the DWL control points move along y, while the keel control points move along z. This approach has two main advantages; rstly, the whole optimisation has been separated into three separate processes, thus reducing the number of variables with reference to the hull surface optimisation. Secondly, the x coordinate being xed, each row of control points denes a section, and so the automatic generation follows the technique usually adopted by the designer. Taking into account also boundaries and constraints (described in Section 2.3), the procedure leads to a faired surface hull. 2.2. Objective functions and optimising function The problem we are dealing with is a typical multi-objective optimisation, which can be solved in several different ways. In this work the weighted sum approach has been used (Vanderplast, 1984). In the general case of multi-objective problems the optimising function can be written like: fxi Z
m X jZ1

wj gj xi

(2)

where: xi are the design variables; gj(xi) are the objective functions; wj are the designer-selected weights; m is the number of objectives.

2.3. Boundaries and constraints In order to better direct the solution of the optimisation, boundaries and constraints on the design variable can be applied. In what follows, boundaries and constraints referring to the hull optimisation are detailed, those on DWL and Keel being quite similar. Particularly the hull boundaries have been dened by offsetting each control point lying on the DWL and the keel line. This offset has been set equal to G5% on both y and z directions. Thus little adjustments of DWL and keel, during the hull optimisation, are possible. In Fig. 3 lower and upper bounds are shown. The solid line represents the initial estimate of the design variables (as dened in Section 4.3). Lower bounds on y and upper bounds on z have been set equal to zero. Geometrical constraints are useful to avoid interweaving of some segments of the control surface patch; they may also favour curve and surface fairing. Mathematically,

A. Mancuso / Ocean Engineering 33 (2006) 234246

239

100

80

60 y values

design variable lower bound upper bound

40

20

0 z values

-20

number of variable -40 0 10 20 30 40 50 60

Fig. 3. Boundaries of the variables adopted for the hull optimisation.

these constraints can be supplied in a matrix form like: A !X % B (3)

A is a sparse matrix containing 1 and K1, X the column vector containing the design variables and B a column vector (made of zeros, in this application). Fig. 4 shows an example in which some constraints are violated (a and b) or not (c). Eq. (3) has been specied for each of the three optimisations.

constraint violation

constraint violation

z y

(a)

(b)

(c)

Fig. 4. Some example of control polygons showing: (a) constraint violation on y; (b) constraint violation on z; (c) any constraint violation.

240

A. Mancuso / Ocean Engineering 33 (2006) 234246

2.4. Algorithm (SQP) The problem (which, as already said, is a multi-objective optimisation problem) has been solved with the aid of the optimisation toolbox of MatLab. In particular, efcient algorithms are made available to the optimiser by the BFGS quasi-Newton method, with a mixed quadratic and cubic line search procedure. This quasi-Newton method uses the BFGS formula to update the approximation of the Hessian matrix, since no information about the gradient of the optimising function can be supplied (a detailed description can be found in Matlab (2001)). The problem to be solved may have many local optima. That is why the initial set of design variables to be subjected to the optimisation process may inuence the nal result. It is good practice to start the optimisations from different sets of design variables in order to better explore the domain space.

3. Denition of the main parameters Typically, a sailing hull can be dened by a set of parameters whose values mainly depend on the type of boat (e.g. cruising/regatta) and the lWL. In the following the parameters used for the purpose of this work will be described and discussed. The numerical values assigned to some of these parameters are reported for a modern light/medium displacement yacht. Many other parameters can be dened, but they are related to the emerging part of the hull (Larsson and Eliasson, 1996; Mancuso, 2001), so they have not been used here. Cp Z Vc lWL ! As (4)

this ratio is the prismatic coefcient and is a measure of the fullness of the yacht. The fuller the ends, the larger will be the Cp. Its optimum value is difcult to be predicted, as it depends on the speed of the yacht, on sailing conditions and on wind force. As a general rule, yachts are designed to sail close-hauled in a breeze; in this case the optimum Cp is close to 0.56. lWL C0 Z p 3 Vc (5)

this ratio inuences both habitability and resistance of the yacht. It is a good practice to assign values in the range of 56. In this way the Froude number Fn overcomes its critical value which is Fny0.45. In some extreme cases like, for instance, the IACC (International Americas Cup Class) yacht, the above ratio may attain higher values (77.5) (IACC rules, 2003). C1 Z AWL lWL Bmax (6)

this is the ratio between the area inside the DWL (the water plane area AWL) and the box around it. Its value is in the range of about 0.60.8. Like the Cp value, the fuller the ends of the yacht the larger the C1. The centre of gravity of this area is called Centre of Floatation

A. Mancuso / Ocean Engineering 33 (2006) 234246

241

and is located on the transverse axis around which the hull is trimmed when a weight is moved longitudinally. The longitudinal location of this centre (LCF in the following) can be expressed as a percentage of the lWL. C2 Z As Bmax Tc (7)

this is the ratio between the maximum transverse area, AS, and the box around it. Its value is in the range of about 0.60.9. A typical value for a modern yacht is w0.75.

4. Denition of curves and surface In the following a description of the technique applied to model boundary curves and surface is detailed. Such an approach is, of course, adapted to the problem we are dealing with but, it does not change the generality of the procedure which can be suited to different cases. 4.1. DWL boundary curve The strategy applied to fair the DWL is focused on the determination of a curve shape that simultaneously reects the target values for both C1 and LCF. Fig. 5 shows a typical shape of the DWL of a modern sailing yacht. The design data are the length of the water line and the maximum beam. Moreover, where the beam reaches its maximum value, the tangent to the curve must vanish (i.e. be horizontal); similarly, when xZlWL it must generally be vertical. The problem is now well dened, and it is possible to nd a suitable curve shape obeying the imposed geometrical constraints (and for which LCF and C1 approach the correspondent target values LCF* and C1*). Consequently the optimising function of Eq. (2) will be written like:
f Z w1 LCF K LCF2 C w2 C1 K C1 2

(8)

As already said, such result can be obtained by changing the position (x, y) of the B-spline control points.
100 50 Bmax 0 LCF 0 100 200 300 lWL 400 500 600 700 control polygon DWL

Fig. 5. DWL shape described by a B-spline curve.

242

A. Mancuso / Ocean Engineering 33 (2006) 234246

4.2. Keel boundary curve The keel line has been treated differently from the DWL. Typical values for the coefcients are not available, like there were for the DWL. For a sailing yacht (Fig. 6) it is better to assign geometrical constraints like the Tc, the stem angle q, as in Calkins et al. (2001) and, in this case too, when the draft attains its minimum value, the tangent to the curve must be horizontal. On the basis of these data a cubic interpolating parametric curve can be drawn out. For convenience, by the procedure described in Cappello and Mancuso (2004), the curve is converted into a B-spline curve showing (i) the same number of control points as dened for the DWL, (ii) the longitudinal position of these points xed and equal to that of the corresponding ones on the DWL. The goal is to nd a B-spline curve that is as close as possible to the interpolating curve previously found. Specically, the interpolating curve is rstly evaluated over R points (e.g. RZ200 to 300). During the optimisation the B-spline control points are moved until the sum of the squared distance between each data point and the curve is lowest; hence the objective function, f, will be: fZ
R X rZ0

D Sr

(9)

where it is assumed that: q DSr Z x K xr 2 C z K zr 2 Here xr, zr, the coordinates of the rth data point, show the lowest distance from the P(x*,z*) curve point. 4.3. Hull surface denition Once the boundary curves are found, an initial surface can be constructed. The number of control points in the bow-stern direction, N, is equal to that used for the boundaries; whereas in the direction keel-DWL, M, it is free and can be assumed arbitrarily. Fig. 7 shows the initial estimate with a net of N!M control points. The intermediate position (between the boundaries) of each control point has been obtained by assigning it the same x coordinate of the corresponding boundary control point, and linearly spacing the y and z values between these boundaries; this position makes the initial surface always ruled. This set of control points will be subjected to the optimisation process with the aim of nding
0 -50 0 100 200

keel line control polygon 300 400 500 600

Tc 700

lWL
Fig. 6. Keel line described by a B-spline curve.

A. Mancuso / Ocean Engineering 33 (2006) 234246

243

control point surface patch

Fig. 7. Control patch describing the hull shape.

a well faired hull. In this case the goal is simultaneous compliance with the target values for all the parameters dened by Eqs. (4)(7) and the optimising function becomes:
K Cp 2 C w4 C0 K C0 2 C w5 C1 K C1 2 C w6 C2 K C2 2 f Z w3 Cp

(10)

5. Case study The optimisation procedure has been applied to a hull having lWLZ6.65 m, designed with the traditional method (Larsson and Eliasson, 1996). The main characteristics of the yacht are listed in the third column of Table 1 (more details can be found in Mancuso, 2001). These parameters (from row 1 to row 7) refer to a sailing yacht designed to sail close-hauled in a breeze condition. From Eq. (5) one can nd the volume displacement P c
Table 1 Yacht parameters assigned and found after optimisations Parameter (unit) 1 2 3 4 5 6 7 8 9 Cp C0 C1 C2 Tc (m) LCF (m) q (deg) Pc (m3) Bmax (m) Assigned value 0.56 5.70 0.70 0.74 0.29 3.79 K12 1.59 1.98 Value after curve optim. 0.6965 0.29 3.79 K11.6 1.99 Value after hull optim. 0.560 5.70 0.701 0.740 0.285 3.798 K12.5 1.588 2.02

244

A. Mancuso / Ocean Engineering 33 (2006) 234246

(listed in row 8); while Eqs. (4) and (7) yield the maximum beam (row 9). Once all the parameters have been chosen and/or calculated the optimisation procedure can begin. DWL and keel have been modelled by cubic B-spline curves with 10 control points and uniform knot vector. The hull has been modelled by means of a 10!5 control points bicubic B-spline surface with uniform knot vector in both the parametric directions. The total number of variables is 16 for the DWL, 8 for the keel and, according to Eq. (1), 64 (2!8!4) for the hull optimisation. Several tests have been carried out to calibrate the weights of the optimising functions. No remarkable effects have been found in the nal solutions, hence all the weights have been set equal to 1. Values listed in column 4 of Table 1 have been found after optimising the curves, while those in column 5 followed the surface optimisation. Notice how stem angle, maximum beam and draft are slightly different, following surface optimisation, form those found previously. As a result a little adjustment of these values was enough to match all the assigned objectives, as shown in the last column of Table 1. In the following some results concerning the hull optimisation are shown. Fig. 8 shows the trend during the optimisation process of Cp, C0, C1 and C2. The prismatic coefcient at the beginning of the optimisation had a value very close to the target one. This is simply due to the initial surface having low values of both P c and AS, whereas the values of C0 and C2 are quite different from the target ones. After a few iterations C0 and C2 approach the target values, while Cp is kept under control by the optimiser. The C1 component of the optimising function of Eq. (10) does not show
0.568 0.566 0.564 0.562 0.56 0.558 0.556 0.703 0.702 0.701 0.7 0.6 0.699 0.698 0.697 0.696 0 5 10 15 20 25 30 0.55 0.5 0.45 0 5 10 15 20 25 30 C1 0.7 0.65 0 5 10 15 20 25 30 Cp 6.6 6.5 6.4 6.3 6.2 6.1 6 5.9 5.8 5.7 5.6 0 5 10 15 20 25 30 C0

0.75 C2

Fig. 8. Components of the optimising function of Eq. (10) vs. iteration number.

A. Mancuso / Ocean Engineering 33 (2006) 234246

245

Fig. 9. Perspective view of the optimised hull with the main curve section (top); enlarged view of the sections (down).

appreciable variations, because it moves around the value previously found while optimising the curves. However, if this last objective is neglected (i.e. it is assumed that w5Z0), undesired at regions can be detected close to the bow and centre bottom of the yacht, which conrms the need to add this objective in the optimising function. The algorithm converges toward a minimum value after 30 iterations. As a matter of fact, after the rst iteration the value of Eq. (10) is f1Z0.81, while at the end of the optimisation is f30Z2!10K6. Fig. 9 (top) shows the rendered image of the optimised hull with the main curve sections superimposed (water lines, sections, buttocks and diagonal); while Fig. 9 (down) shows an enlarged view of the sections. All analyses have been performed on a PC operating with CPU of 2.4 GHz. The total required time is a few minutes (less than 15).

6. Conclusion This paper describes an automatic procedure for generating a hull shape, based on an optimisation technique. The procedure is focused on the determination of reliable CAD models of sailing yachts. Starting from an assigned set of parameters and certain geometrical conditions, it is possible rst to optimise the boundary curves (designed water line and keel) and then to optimise the whole hull surface. The method is fully automatic and leads to a reasonable solution in a very short time (a few minutes). The hull surface, which is stored in IGES format, can be imported into any CAD software, so it may be

246

A. Mancuso / Ocean Engineering 33 (2006) 234246

checked and/or parameters may be possibly adjusted; besides it can be used for further investigations like, for instance, uid dynamics or structural analyses.

References
Americas Cup Class Rules version 5.0, 15 December 2003. Augusto, O.B., Kawano, A., 1998. A mixed continuous and discrete nonlinear constrained algorithm for optimizing ship hull structural design. Ocean Engineering 25 (9), 793811. Calkins, D.E., Schachter, R.D., Olivera, L.T., 2001. An automated computational method for planning hull form denition in concept design. Ocean Engineering 28, 297327. Cappello, F., Mancuso, A., 2001. Lay-up optimisation for the hull of a racing sailing yacht. ADES, International Journal Advances in Engineering Software 32 (1), 133139. Cappello, F., Mancuso, A., 2004. Curve and surface tting via optimisation technique. CAD04 International CAD Conference and Exhibition May 2428, 2004 Pattaya Beach, Thailand. Dejhalla, R., Mrsa, Z., Vulkovic, S., 2001. Application of genetic algorithm for ship hull form optimisation II. Shipbuild Program 48, 117133. Harries, S., Abt, C., Hochkirch, K., 2001. Hydrodynamic Modeling of Sailing Yachts. The 15th Chesapeake Sailing Yacht Symposium; Annapolis. Islam, M.M., Khondoker, M.R.H., Rahman, C.M., 2001. Application of articial intelligence techniques in automatic hull form generation. Ocean Engineering 28, 15311544. Larsson, L., Eliasson, R.E., 1996. Principles of Yacht Design. Adlard Coles Nautical, London. Mancuso, A., 2001. Modellazione 3D di Carene di Barche a Vela Il. Progettista Industriale, 106113. Mancuso, A., Mancuso, U., 2004. Ottimizzazione di Forma di Carene di Barche a Vela Mediante Analisi CFD Il. Progettista Industriale 2004, 7077. MatLab, 2001. R12 Optimization ToolBox User Guide. Piegl, L., Tiller, W., 1997. The NURBS Book. Springer, Berlin. Percival, S., Hendrix, D., Noblesse, D., 2001. Hydrodynamic optimization of ship hull forms. Applied Ocean Research 23, 337355. Valorani, M., Peri, D., Campana, E.F., 2003. Sensitivity analysis methods to design optimal ship hulls. Optimization and Engineering 4, 337364. Vanderplast, G.N., 1984. Numerical Optimization Techniques for Engineering Design, Monterey, CA 1984. Yilmatz, H., Kukner, A., 1999. Evaluation of cross curves of shing vessels at the preliminary design stage. Ocean Engineering 26, 979990.