You are on page 1of 11

SEAREV : A FULLY INTEGRATED WAVE ENERGY CONVERTER Aurlien Babarit(1), Alain Clment(1), Jacques Ruer(2), Catherine Tartivel (2)

(1) Laboratoire de Mcanique des Fluides, CNRS UMR6598 ,Ecole Centrale de Nantes , 1, Rue de la Noe, 44300 Nantes France : Alain.Clement@ec-nantes.fr (2) SAIPEM SA, 1 Av. San Fernando, 78884 St Quentin en Yvelinnes Cedex France : Jacques.Ruer@saipem-sa.com

ABSTRACT
A new wave energy converter (WEC), called SEAREV1, is presented. The mathematical models and numerical simulators that have been developped, both in frequency and time domains, to assess the electricity production on a given test site are described. Genetic algorithm using these numerical models have then been implemented in order to optimize the shape and the mechanical parameters of the device. A discrete real-time control of bang-bang type (latching control) permits to enhance the response of the system to the waves, and therefore the annual production of energy.

INTRODUCTION
The SEAREV wave energy converter is a floating device enclosing a heavy horizontal axis wheel serving as an internal gravity reference (see Fig.2). The centre of gravity of the wheel being off-centered, this component behaves mechanically like a pendulum. The rotational motion of this pendular wheel relative to the hull activates an hydraulic Power Take Off (PTO) which, in turn, set an electric generator into motion. Two major advantages of this arrangement are that, first: all the moving parts (mechanic, hydraulic, electric, components) are sheltered from the action of the sea inside a closed, waterproof shell; and secundly that the choice of a wheel working as a pendulum involve neither endstop nor any security system limiting the stroke.

Figure 1 The SEAREV device: external view

This freely floating device is kept on site by a slack single line mooring which enable self alignment of the device in the dominant wave direction due to the general shape of the hull. This allows the device complying easily with tide The system being based on the strong coupling of two nonlinear mechanical oscillators, an analytical study of its behavior in regular and irregular waves has been performed. A complete mechanical model of the system has been written, including hydrodynamics of wave device interaction and nonlinear mechanics of the internal PTO. Only the linear version of the equations will be presented here. We will show how the shape and all the mechanical parameters of the system have been optimized using a general purpose multi-parameter multi-criterion software, based on genetic algorithms. A specific real time control has been developed in order to enhance the production in low energetic sea states. Numerical time-domain simulations are presented to exemplify the benefit that latching control can bring to the system.

MATHEMATICAL MODELLING
The floating body is assumed to have two vertical planes of symmetry. We assume that the rotation axis of the internal cylinder is perpendicular to the main symmetry plane of the floating body. We suppose also the direction of wave propagation being parallel to this plane (self alignement). Under these assumptions, in a 2D incident wave train, the floating body will move only in surge, heave and pitch mode in the (xOz) plane (Fig.2).
1

Systme lectrique Autonome de Rcupration dnergie des Vagues

Let

xG the surge motion, z G the heave motion of the centre of gravity G of the floating body, the pitch motion and

the relative motion between the floating body and the inner pendulum. The Power Take Off system (PTO) then works this relative motion . Let mb the floating body mass, I b its inertia , m p the mass of the inner cylindrical pendulum,

Iy

its moment of inertia. Let l = AP the distance between the center of rotation A of the inner pendulum and its own center of gravity P, d = GA the distance between the center of gravity of the floating body and the center of rotation of the pendulum. Wave amplitude and resulting body motion are assumed to be sufficiently small to justify working in the classical linear potential flow theory.

Let us define the four components position vector of the Figure 2 notations system

Z = ( xG , z G , , ) , and the matrices M , B and


t

K as:

m (d l ) p 0 mb + m p 0 0 mb + m p M= I + I + m (d l )2 m p (d l ) 0 b y p 2 m ld 0 mpl I +m l y p p k x 0 K= 0 0 0 0 0 0 0 0 0 m p g (d l ) m p gl 0 m p gl m p gl 0 0 0 0 0 0 0 0 0 0 0 BPTO

m l p 0 I + m l 2 m ld y p p 2 I +m l y p

(1)

(2)

0 0 B= 0 0

(3)

where M is the proper mass and inertia matrix of the system, K the restoring matrix, and B the external damping matrix where the only non zero term is a linear term representing the moment exerted by the PTO system proportional to the & rotational velocity, namely: B PTO (t ) . Frequency domain model Let us first consider the case of a regular wave train with circular frequency exciting the system. Adding the coupling terms between the hull and the cylinder in the classical floating body motion equations, one can derive easily the equations of the system under the matrix form:

[ (M + C
where:

) 2 + i (B + C A ) + (K + K H ) Z = Fex ( )

(4)

Fex is the hydrodynamical excitation vector

C M ( ) and C A ( ) the added mass and wave radiation damping matrices, both function of the hull geometry and wave frequency K H the hydrostatic restoring matrix

For a typical SEAREV device as shown on Fig.1 (25m length, 15m width, 14m depth, 1000T displacement, 400T pendulum mass), the Response Amplitude Operators (RAO) are shown in Fig.3;

Figure 3: RAOs of the SEAREV device left: pitch and wheel motion , right: surge Computations of the hydrodynamic coeficients C M ( ) and computer code. (AQUADYN. Delhommeau et al. 1992).

C A ( ) have been performed using a standard seakeeping

The double-peaked curve of pitch RAO on Fig.3 illustrates the advantage of such a multiple resonance system. We are able to tune the different natural frequencies, (mainly pitch of the floater and pendular wheel rotation) to broaden the bandwidth of the composite system with regard to the wave spectrum. In this frequency domain approach, the mean power absorbed by the system is simply given by:

p ( ) =
where

1 2 BPTO ( ) 2

(5)

( )

is derived from the solution of algebraic system (4).


2

In figure (4), the function p ( ) / A is plotted for the same design as for the RAO above. For the sake of comparison, we have also plotted on this figure the theoretical maximum power absorbed by the body. Since the SEAREV device works in pitch and surge modes, it can be shown that the maximum absorbed energy at frequency is given by (e.g Falnes 2002a): p max = 2

g 2 4 2

which is twice the maximum energy absorbed by a heaving point aborber. Figure 4 absorbed power in regular waves for a typical SEAREV device

Time-domain linear model

In the time domain waves are no longer assumed to be regular, and non linear components of forces (like hydraulic PTO for instance) can be introduced. Hydrodynamics will still be considered as linear here, implying moderate motions of the hull. The time domain counterpart of eq.(4) is :

(6)

& & & (M + C M ())Z& = Fex (t ) BZ(t ) H(t )Z( )d (K + K


t 0

)Z(t )

(7)

where the wave radiation forces are accounted for through a convolution product (Cummins 1962). The computation of the impulse response functions H (t ) has been performed by using a dedicated code ACHIL3D (Clment 1999), but could have been obtained through a Fourier transform of their frequency domain counterpart The mean absorbed power during a time interval T will now be given by

C M ( ) and C A ( ) (Falnes 2002a).

p=
Given an excitation vector

1 &2 BPTO (t )dt T 0

(8)

Fex (t ) and initial conditions, the integro-differential equation (7) can be integrated to simulate

the behaviour of the system in irregular waves. But the presence of the convolution integral in this equation is not well suited to such numerical integration.

Time domain state equation model We have therefore developped an alternative formulation by approximating the convolution products in (8). This identification, in the sense of system theory, is based on the approximation of the elements of matrix H (t ) by series of exponentials, following a method proposed by R de Prony (1795) and implemented by Duclos et al. 2001. Let

S (t ) denotes a scalar convolution integral

S (t ) = g (t ) f ( )d
0

(9)

and suppose that we are able by using Pronys method to find an approximation of

S (t ) under the form:


(10)

g (t ) i e i ( t )
i =1

then

S (t ) can be written:

S (t ) = i e
i =1 0 t

i ( t )

f ( )d = S i (t )
i =1

(11)

with:

S i (t ) = i e i (t ) f ( )d
0

(12)

which satifies : Given the initial conditions

& S i (t ) = i S i (t ) + i f (t )

(13)

S i (0) = 0 , S (t ) can now be computed by integrating the set of ordinary differential equations
H (t ) in order to replace the integro-

(11) (13) rather than computing the integral (9). This identification technique has been applied to all the coefficients of the matrix differential eq.(7) by an equivalent state equation. Defining the vector

S(t ) as:
(14)

& S(t ) = H (t )Z( )d


0

and assuming that all the components have been approximated by series of exponentials we have the system:

3 N1 j S1 jr (t ) j =1 r =1 N2 j 3 S (t ) S(t ) = 3j =1 rN=1 2 jr j =1 r =31j S 3 jr (t ) 0 S = S + Z (t ) & & ijr ijr ijr j ijr


Introducing a vector of radiation states

(15)

I (t ) such that

I (t ) = (S rij ) 1i 3;1 j 3;1 r N ij


t

we obtain finally a state equation for the global system

& X = AX + C
with
Id A= 0 0 0 (M + C M ) 0
1

(16)

0 0 Id 0 0 (K + K H ) B I d 0

Id C= 0 0

0 (M + C M ) 0
1

0 0 0 F t ex I d 0

where the new extended state vector

& X is defined as X = Z t Z t I t

The time-domain numerical simulations of the system as those reported here can now be performed very efficiently by integrating this system of ODE using a 4th order Runge-Kutta algorithm (see e.g Press et al. 1992). For the simulations in irregular waves, the incident wavetrain was obtained by discretizing the ITTC spectrum (Rawson and Tupper 1983):

5 1 5 H1 3 A 4 and B = S ( f ) = 5 e f with A = , where T1 is the peak period of the spectrum (sometimes noted 4 16 T1 4 T1 4 f T p ).
The classical method of random phases is then used to synthetized the irregular wavetrain as a superposition of waves with amplitudes

N regular

a( f ) = 2S ( f )f

DEVICE OPTIMIZATION
method Once validated, the above mathematical model was available to be used for the optimization of the device. Not only the shape of the hull, but also the mechanical parameters such as: mass and inertia of the pendulum cylinder, position of its axis of rotation, position of the centre of gravity of the floater and of the cylinder,must be determined in order to optimize the annual production of energy. For a given site with a probability of occurrence of sea state ( H 1 3 ,T1 ) given by a function

C ( H 1 3 , T1 ) , in the above fully linear approach, the


N 2

annual production to be optimized reads:

j E = C ( H1 3 , T1 ) p ( j ) a( , H1 3 , T1 2 H 1 3 T1 j =1

(18) Figure 5 Scatter diagram at Yeu Island site

In the present study we have used a reference site located near Yeu Island, a French Atlantic coast island, where the mean annual incident power is about 23 kW/m. The scatter diagram Fig.5 gives a repartition of the sea states statistics. So in this optimization, we have to maximize E while, indeed, minimizing a variable representing the cost of the device. At this early stage of the study a precise cost function was not yet established, so we have used the total displacement of the device as a cost function. The optimization problem is thus multi objective, and the number of parameters to be varied is large: say between 10 and 20. Furthermore, there are several constraints about the design that must be respected during the optimization. For example, the static stability of the device must be positive in pitch and roll. The inner pendulum must have a realistic density (we have chosen here the density of concrete). The draught of the floating body must be kept reasonable (15 meters max, say), etc... For such a global optimization problem, one can imagine using genetic algorithms which are known to be very efficient in these circumstances (R. Duvigneau 2002). Genetic algorithms are optimization algorithm which are not properly speaking deterministic. They attempt to reproduce the Darwinian process of species evolution that can be found in the living world. For our problem, a set of parameters is used to define a particular shape of the floating body and the mechanical characteristics of the inner pendulum at PTO; this set constitutes the genome of each individual. From a set of initial individuals, which forms a population, genetic algorithms first make a call to numerical codes which compute the transfer functions, solving eq.4, and the annual energy produced at the reference site (eq.18) for each individual. Then they select the best ones and cross their genome to create a new generation, which is potentially better than the first one (Darwin's hypothesis). Finally, iterating the process leads to the optimal designs (Fig.6). Moreover, to closer mimic living species evolution, genetic algorithms also introduce mutations from time to time in the genome of the population, in order to enhance the potential and to enlarge the field of possibilities. For shape optimization problems, genetic algorithms constitutes the very best choice, but they are known to converge rather slowly. Figure 6: genetic algorithm for SEAREV shape optimization In our optimization, each evaluation (computation of the annual energy produced at the reference site) of an individual is quite CPU time consuming (about 5 mn/individual on a pentium III 1 GHz PC). The computation of the hydrodynamical coefficients was rapidly identified as the most heavy part of these computations. Once it has been performed, the calculation of the absorbed energy afterwards is practically instantaneous. So the optimization process has been split in two sub-process, in order to accelerate the global computation.

The idea here was to compute first the hydrodynamical coefficients for a given hull shape. The shape being given, the displacement is also known. To maximize the absorbed energy, one have to find the optimal internal pendulum for this given hull. To do this sub-optimization, the best way is to use a gradient method. Basically, the method consists in computing the direction , in the parameter space, in which the energy growth rate is maximum, and to move by a step in this direction. Iterating the process leads to the local maximum closest to the initial point. In practice, the algorithm used to do this is the Fletcher-Reeves algorithm (Press et al. 1992). Globally, we have divided by 4 the number of optimization parameters by using this method. So, the inner optimization loop computes the optimal pendulum for a given shape of the floating body. Hull shape optimization, in the outer loop, is performed by genetic algorithms, using a high performance commercial code ModeFrontier (www.esteco.it). results In figure (7), we have plotted an example of results of an optimization run with a population of 40 individuals on 100 generations, which represents about 15 days of computation on a pentium III 1GHz PC. The shape of the floating body was defined by 15 real parameters. For each optimization run, various constraints were put on shape and mechanical parameters to account for technological or practical criterion. The horizontal axis of the figure corresponds to the total displacement in tons, the vertical axis to the mean absorbed power, in kW, at the Yeu island site for this particular run.

Here, the best individuals are those which deliver the most energy with the least displacement (under the prescribed constraints), so the corresponding dots should be located in the left upper part of the plots in figure 7. Naturally, this corner of the plot is empty, but one can clearly identify a frontier on which optimal individuals tends to accumulate, and above which there will be no solution. This is the so-called Pareto front, which is therefore the locus of the optimal individuals. The Pareto front that we obtained may be roughly split in two parts. In the first one, for displacements between, say, 0 and 500 tons, the absorbed power growing rate is about 0.1 kW/t whereas in the second, for displacement larger than 500 t, it is about 0.05 kW/t. In the upper plot in figure (7), the colour field represent the ratio between the mass of the pendulum over the total mass of the device. One can notice that on the Pareto front, the ratio of the pendulum mass on the floating body mass is about 0.5. In the middle plot of Fig.7, the color indicates the draught T of the device. It appears clearly that the best designs are those having the largest allowable draught, which was set to 15 meters in this run. This constraint on the draught, and all the dimensional constraints on the other parameters was automatically accounted for by the optimization software. In the last plot, we show the ratio of the depth of cylinder centre d over the device draught T. Globally it appears that the optimizer always choose the deeper possible location for the pendulum cylinder. Several tens of thousand of individuals has been generated and evaluated (i.e full year production at the reference site) in this study. The best design was shown to have a mean annual capture width of 4 meters when installed on this 23 kW/m site, resulting in a capture ratio of 35% when we reduce the capture width not to the breadth of the device, but to the cubic root of its displacement.

LATCHING CONTROL
These results are already promising for a site featuring such a moderate resource level. At this stage, without control, the device traps energy only from the wave components whose frequency lays in the band of the natural frequencies of the system, (see Fig.4). Figure 7 results of an optimization run Nevertheless, this system is potentially able to catch energy in the wave spectrum outside this band provided it can be dynamically controlled to adapt its bandwidth in real time. We therefore equipped it with a latching control system. Let us first recall here the principle of this control, introduced by Budal and Falnes in 1980 and 1982 for the control of point absorbers. Condidering a single degree of freedom (DoF) wave energy converter, we introduce a technological system which is able to lock the motion of this productive DoF when its velocity vanishes. While the motion is locked, the wave exciting force continue to evolve. The motion is then released after a while in such a way that during its next oscillation, the velocity and the excitation force are in phase. (this control is sometimes named: phase control). This kind of control enable to create parametric resonance in the dynamics of the device, and thus can improve its efficiency by a constant adaptation of its dynamics to the incoming wave. We have shown in previous publication (A. Babarit et al. 2003b) how it can be applied to a single DoF heaving point absorber regular waves The problem to be solved to implement latching control in a wave energy device is the in-line determination of the optimal duration of the latching phase. In other word, the system itself says when it should be locked (velocity vanishes), but you have to decide when to release it to obtain the best performance from its motion afterwards. In regular waves, and provided the system equation have been expressed in a state equation form as eq.(16), a semi-analytical method has been proposed and validated in A. Babarit et al. 2003b. The method has been applied here to the four degrees of freedom SEAREV device The potential enhancement brought by latching control to our device is clearly illustrated by Fig.8 where we have plotted the result of a time-domain simulation of the response of the SEAREV device to a regular incident wavetrain. This was obtained by integrating eq.16 by a 4th order Runge-Kutta algorithm from the instant t = 0 when the body is released in a regular established wave field. From 0 to 300 sec, we can see (Fig.8) a period of transients after which the regular oscillating regime is established. At t=360 s, the latching control is switched on; the pitching motion of the device is practically not affected whereas the cylinder motion (t ) is dramatically magnified, after a long transient period. The slope of the energy curve

show the gain in term of extracted power. Here the latching delay was determined by the above mentionned semi-analytical method. The same sequence is reported in Fig.9, but here the optimal response was found to have a period triple of the incident wave period. This remarkable non-linear feature was already pointed out for a single DoF system in Babarit et al. 2003a. The magnification of the productive motion is again very impressive here, and we can now observe the effect of the triple period motion of the cylinder on the pitch motion of the hull which exhibits now higher order components. This is due to the coupling terms between surge pitch and cylinder rotation in mechanical equations (1), (2).

Figure 8 Switching latching control on at t=360s in regular waves irregular waves

Figure 9 Switching latching control on at t=360s in s; period tripling of the pendulum response

The problem of determining in line the optimal latching delay when waves are irregular, like in the real world, is far more difficult to solve. The semi-analytical method used in regular waves no longer works here, and alternative approaches should be considered. Using non-causal approaches, we were able to assess different latching strategies for a single DoF system (heaving point absorber) in Babarit et al. 2003a, 2003b. In order to extend this to the SEAREV device, we have implemented a method of optimal command (Hoskin and Nichols 1986) based on Pontryagin maximum principle. Let u be the control variable. We will consider then that the latching of the system is performed by adding an very large extra damping term of the form

& F = Gu (t )
in the RHS of system (16), now written in the general form

(19)

& X = f (t , X, u )

(20)

Ideally, G should be infinite to obtain a perfect and instantaneous latching. Practically, for numerical stability arguments, it is given a finite value, very large compared to the PTO damping. The optimal command problem can then be formulated as follows. Given a long sequence wavetrain, find the sequence

u (t ) over the interval [0, T ] which maximizes the integral of the absorbed power:

[0, T ] of an incident irregular

& max E (u ) = BPTO 2 dt


u
0

(21)

A Hamiltonian H is then defined: equation

& = H ( X, u, ) X
*

with the end condition

(T ) = 0 . If u * (t ) is the optimal sequence, the maximum principle states that is should maximize

the Hamiltonian at each time step. This problem (of finding u (t ) given the irregular wave field over 0, T ) is solved iteratively, integrating alternatively (20) and (22) in, respectively, direct and reverse time direction, and maximizing H by a new sequence of u(t) during each iteration. This process converges in 50 to 60 iterations, the convergence depending on the value of G.

Figure 10: SEAREV_01 Production matrix; without (left) and with (right) control Applying this method to our device, with irregular wave trains synthetized from the ITTC spectrum mentionned above, we were able to construct two production matrices for the device: one when no control is applied, the other one with optimal latching control computed by this method. These matrices showing the extracted power normalized by the rated device power are shown on Figure 10. Applying both these matrices to the scatter diagram of our test site (Yeu Island) we observed an overall magnification by a factor around 2 for the energy producted yearly. From these figures therefore, it is clear that real time control is not simply an improvement for this kind of resonant wave energy converter, but a real breakthrough.

CONCLUSION
A new patented wave energy device has been developed. From a robust principle, optimizations studies has been performed in order to maximize the annual energy production at a reference site (Yeu Island) were the wave energy resource is moderate. Results are promising especially when latching control of the internal pendulum is active. This system has some advantages that we can sum up here: It is completely closed; all the mechanical parts, hydraulic and electric components, etc.. are located inside, then sheltered from sea action. This will certainly increase their lifetime and reduce maintenance costs. The choice of a wheel as a pendulum makes the internal system very robust due to the absence of any endstop (the technical design was done such that a complete turn of the wheel is allowed and will not damage the system) Use of low technology components available on the shelf, in a floating body (rather than submerged high technology) result in global robustness and cost reduction Being self-referenced, the system reacts against gravity, instead of moorings. A slack single line mooring is then sufficient to keep it on site, and being freely floating, it can accommodate any tidal range It is easy to take away and tow to shore for off-site maintenance which will then be easier (then cheaper) than for fixed bottom seated systems. A real time (wave to wave) sophisticated control can bring a considerable increase in energy production (this system having the ability to lock the pendulum will be used also for security procedures).

& H = B PTO 2 +

f , with

(t ) an adjoint state vector, solution of the differential


(22)

As other recent wave energy systems, it is modular and will be deployed offshore in wave farms of several tenth of individuals, (representing typically 20 to 30 MW/km2 installed power density) in water depth of (typically) 50m, inducing therefore practically no visual impact from the shore.

Tests with a model at 1/12 scale are being performed in the multidirectional 50m by 30m wave tank of Ecole Centrale de Nantes, and a full scale prototype (1000T, 500kW) to be tested at sea in 2009 is currently at the pre-design stage.

ACKNOWLEDGMENT
This study was financially supported by the CNRS, the Regional Council of Pays de la Loire and the French Energy Agency ADEME.

REFERENCES
A. Babarit, G. Duclos, and A.H. Clement. Benefit of latching control for heaving wave energy device in random sea. Proc. 13th Int Offshore and Polar Engineering Conf.ISOPE2003, 1:341348, 2003. ISBN 1-880653-60-5. A. Babarit, G. Duclos and A.H. Clment: Comparison of Latching Control Strategies for a heaving wave energy Device in random sea. Proceedings of the fifth European Wave Energy Conference, Cork, 2003, pp.245-257. A. Babarit, A.H. Clment: Optimal Latching Control of a Wave Energy Converter. Proceedings of the sixth European Wave and Tidal Energy Conference, Glasgow, 2005. Figure 11: the 1/12 scale PTO model being tested on moving table

A. Babarit, (2005), Optimisation hydrodynamique et contrle optimal dun rcuprateur d nergie des vagues. PhD thesis report, Ecole Centrale de Nantes. K. Budal and J. Falnes. Interacting point absorbers with controlled motion, in Power from Sea Waves. BM Count, Academic Press, 1980. K. Budal, J. Falnes, L.C. Iversen, P.M. Lillebeken, G. Oltedal, T. Hals, T. Onshus, and A.S. Hoy. The Norwegian wave power buoy project. The Second International Symposium on Wave Energy Utilization, pages 323344, June 1982. A.H. Clment. Using differential properties of the green function in seakeeping computational codes. Proc. 7th Intern. Conf. Numer. Ship Hydrod., 6(5):115, 1999. W.E. Cummins. The impulse response function and ship motions. Schiffstechnik, p. 491502, 1962. Riche de Prony. Essai exprimental et analytique sur les lois de la dilatabilit des fluides lastiques et sur celles de la force expansive de la vapeur de leau et de la vapeur de lalcool diffrentes tempratures. Ecole polytechnique 1, cahier 2, pages 2476, 1795. G. Delhommeau, P. Ferrant, and M. Guilbaud. Calculation and measurement of forces on a high speed vehicle in forced pitch and heave. Applied Ocean Research, 14(2):119126, 1992. G. Duclos, A.H. Clment, and G. Chatry. Absorption of outgoing waves in a numerical wave tank using a self-adaptive boundary condition. Int. Journal of Offshore and Polar Engineering, 11(3):168175, 2001. R. Duvigneau. Contribution loptimisation de formes pour des coulements forts nombres de Reynolds autour de gomtries complexes. PhD thesis, Ecole Centrale de Nantes et Universit de Nantes, 2002. J. Falnes. Ocean Waves and Oscillating Systems, Linear interactions including wave-energy extraction. Cambridge University Press, 1 edition, 2002. J. Falnes. Optimum control of oscillation of wave-energy converters. International Journal of Offshore and Polar Engineering, 12(2):147155, 2002.

R.E. Hoskin and N.K. Nichols. Optimal strategies for phase control of wave energy devices. In Michael E. McCormick and Young C. Kim, editors, Utilization of Ocean Waves :Wave to Energy Conversion, pages 184199, 345 East 47th Street New York,New york 10017-2398, 1986. American Society of Civil Engineers. W.H. Press, S.A. Teukolsky ,W.T. Vetterling, and B.P. Flannery. Numerical Recipes in Fortran 77. Cambridge University Press, 1992. K.J. Rawson and F.C. Tupper. Basic ship theory, volume 1. Longman, 3 edition, 1983.

You might also like