You are on page 1of 11

Marine Structures 12 (1999) 199}209

Wind loads on marine structures


M.R. Haddara!, C. Guedes Soares",*
!Memorial University of Newfoundland, Faculty of Engineering and Applied Science, St. John's, Nyd,
Canada A1B 3X5
"Unit of Marine Technology and Engineering, Technical University of Lisbon, Instituto Superior Te& cnico,
Av. Rovisco Pais, 1049-001 Lisboa, Codex, Portugal
Received 29 January 1999; received in revised form 15 March 1999; accepted 9 April 1999

Abstract

A review of the available methods for the calculation of wind loads on ships and o!shore
structures is presented. Based on the review, four methods were selected and implemented
in order to carry out a comparative study of the wind loads on ships. Namely, the ahead force,
the side force and the yawing moment were calculated and compared. A large tanker both
in the loaded and ballast conditions was analysed using the four methods. In the second
part, an expression for the estimation of wind loads on ships is proposed and compared with
the experimental data. A neural network technique was used to obtain this expression.
( 1999 Elsevier Science Ltd. All rights reserved.

Keywords: Wind loads; Loads on ships

1. General

The wind loads do not usually play a major role in the structural design of ships and
o!shore structures. The magnitude of the mean static forces and the moments caused
by the wind amount to only a fraction of the total loading on the structure. However,
there are instances when the e!ects of these forces and moments become critical.
Cracks occurring in #are booms due to dynamic wind loading have been reported in
the literature [1]. Wind loads play important roles in the e$cient operation of a ship's
propulsion plant or on her manoeuvrability. Also in certain particular situations such
as the towing of ships, moored ships and dynamic positioning, wind loads play an
important role.

* Corresponding author. Tel.: 00351-1-841-7607; fax: 00351-1-847-4015.


E-mail address: guedess@alfa.ist.utl.pt (C. Guedes Soares)

0951-8339/99/$ - see front matter ( 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 9 5 1 - 8 3 3 9 ( 9 9 ) 0 0 0 2 3 - 4
200 M.R. Haddara, C. Guedes Soares / Marine Structures 12 (1999) 199}209

Wind tunnel tests are the most accurate procedure to estimate the wind forces on
ships and o!shore structures. However, these tests are time consuming and very
expensive, thus the prediction methods are generally used complementary or alterna-
tively to wind tunnel tests. Several numerical methods have been developed for the
prediction of wind forces on #oating structures without recurring to direct model
testing.
Isherwood [2] proposed expressions for the forces and moment coe$cients, derived
from multiple regression analysis of previously published experimental results. The
wind coe$cients are given by equations, depending on the basic characteristics of the
water above the hull. This method can be applied to a wide range of ship types and
con"gurations and for all angles of the wind relative to the bow.
Gould [3] presented a numerical procedure for the estimation of the wind
forces and moment on the superstructures of ships. The e!ect of wind speed over
the sea on the e!ective relative wind speed acting on the ship is discussed. In fact, the
wind pro"le over the sea depends on the sea surface, which moves itself and presents
increasing roughness with increasing wind speed. A logarithmic pro"le for wind
speed is proposed by the author. The e!ective wind speed and the lateral centre of
pressure for a gradient wind, is then estimated by subdividing the frontal and lateral
projections into `universal elementsa. The wind tunnel data for ships covering a wide
range of designs are given with advice on their application to other ships. This data
can be used to estimate the forces and moment coe$cients for a wide range of ship
types.
Blendermann [4] has presented a systematic collection of wind load data. Continu-
ing this work, Blendermann [5,6] derived expressions for the coe$cients of the wind
forces acting in the longitudinal and transverse directions and the yaw and rolling
moments. These expressions are functions of a drag coe$cient, the angle of attack, the
frontal and transverse areas subject to wind, the coordinates of the centre of area and
a parameter called the cross-force parameter. This parameter is obtained statistically
from the experimental data.
Blendermann [7] also proposed a method for the prediction of wind loads on ships
in a non-uniform air #ow using experimental data. The non-uniformity of the air #ow
is accounted for an e!ective dynamic pressure.
Brix [8,9] has presented a selection of the wind load data from Blendermann [4,5],
together with methods for their use in manoeuvring problems.
Lee and Low [10] obtained results of wind tunnel tests on models of rigid o!shore
platforms. The pressure transducers located at 141 locations on the 1 : 268 scale model
were used to measure wind loads. The wind velocity was measured using hot wire
anemometer. The wind tunnel results show an overriding in#uence of the legs. They
accounted for about 70% of all drag and about 80% of all over turning moments in
the #oating mode.
OCIMF [11] presented coe$cients and procedures for computing wind loads on
very large crude carriers (VLCCs), i.e. tankers in the 150,000}500,000 DWT range.
However, the coe$cients and procedures can be applied to smaller tankers with
similar geometry. The wind coe$cients are based upon data obtained from wind
tunnel tests conducted at the University of Michigan in the 1960s.
M.R. Haddara, C. Guedes Soares / Marine Structures 12 (1999) 199}209 201

A very long mat-like platform was used to study the wind forces on a #oating
airport by Ohmatsu et al. [12]. Experiments were conducted in a closed-return wind
tunnel. A distorted model was used where only the fore and aft parts were accurately
scaled. Wind pressure distribution, wind forces and wind velocity distribution around
the model were measured. It has been found that the total drag force coe$cient
increases with the increase in the model length and decreases with the increase in draft.
Chen et al. [13] carried out preliminary experiments to study the wind environment
over the helideck of an o!shore platform in the Beihai sea, China. Tests were carried
out in an atmospheric boundary layer wind tunnel using a 1 : 100 scale model. A hot-
wire anemometer was used to measure the velocity over the helideck. The velocity was
measured at equally spaced stations, ten in the horizontal direction and eight in the
vertical direction. The results for "ve di!erent angles of attack were obtained. These
experiments indicate that the distributions of the mean as well as the #uctuating
velocities are quite sensitive to the direction of the wind. They also found that the
separation occurs generally at the front edge of the helideck. In some cases, the mean
velocity distribution at the centre of the deck has a turning point, which may suggest
an instability.
The vibration characteristics of several Norwegian o!shore #are booms have been
discussed by Oppen and Kvitrud [13]. A survey of the cracks produced by this
resonant vibration was also given. The cross wind vortex-induced vibration and wake
interaction with the chord members were the main causes for cracks. The methods for
mitigating these e!ects were described and a recommended design practice was
suggested.

2. Comparative study on wind loads

Three methods available in the literature were used to estimate the wind loads on
a 351.0 meter tanker, in the loaded and ballast conditions. Table 1 shows the principal
dimensions of the tanker. The methods applied are those described in [2,3,11]. The
results of the calculations are shown in Figs. 2}4 for the loaded tanker and in Figs. 5}7
for the tanker in ballast. The experimental data obtained by Blendermann [4] is
plotted together for comparative purposes.

Table 1
Main particulars of the tanker

Loaded In ballast

Overall length (m) 351.40


Beam (m) 55.40
Draft (m) 23.50 10.625
Lateral projected area (m2) 3401.47 7839.63
Height of centre of lateral area above 6.83 12.34
Distance of centre of lateral from midship 24.45 (forward) 8.32 (forward)
Transverse area (m2) 1131.79 1803.93
202 M.R. Haddara, C. Guedes Soares / Marine Structures 12 (1999) 199}209

Fig. 1. Convention for positive quantities.

Fig. 2. Longitudinal force coe$cient, loaded tanker.

The convention for positive forces and moment are presented in Fig. 1, where F ,
x
F and N represent, respectively, the longitudinal force, side force and yaw moment
y
and b is the angle between the wind speed and the ship bow.
Fig. 2 represents the longitudinal force coe$cient for the loaded tanker. Although
one can observe some dispersion on the numerical results, in general the predictions
compare well with the experimental results. For wind coming from the stern, all
methods tend to overpredict the experimental results. Among the three methods, the
OCIMF is the one which compares better with experiments.
Fig. 3, shows the results of the side force coe$cients for the loaded ship. One can
observe that both Isherwood's and the OCIMF methods agree with the experimental
values obtained for the side force by Blendermann over the whole range of the wind
angle. Values for the lateral force coe$cient obtained by Gould's method are highly
overestimated.
In Fig. 4 the yaw moment coe$cients results are presented. One can "nd qualitative
di!erences between Blendermann's experimental results and OCIMF values for the
yaw moment coe$cient on one side and estimates obtained by Gould's and Isher-
wood's methods on the other side. The yaw moment coe$cients obtained using the
"rst two methods have negative values over the whole range of the wind angle. The
M.R. Haddara, C. Guedes Soares / Marine Structures 12 (1999) 199}209 203

Fig. 3. Side force coe$cient, loaded tanker.

Fig. 4. Yaw moment coe$cient, loaded tanker.

values obtained using Gould's and Isherwood's methods display a change in sign of
the coe$cient around a wind angle of 703. Large quantitative discrepancies exist
between the results obtained using the three methods and Blendermann's experi-
mental results.
Figs. 5}7 represent comparisons between the numerical predictions and experi-
mental results, respectively of the longitudinal force, side force and yaw moment
coe$cients for the tanker in the ballast condition. The conclusions for the longitudi-
nal force coe$cients are similar to those obtained for the loaded condition. Again the
OCIMF method tends to agree better with the experiments than the other methods.
In the case of the side force coe$cients all methods overestimate the experimental
data, the Isherwood results being closer to the experiments than the Gould and
OCIMF results. The numerical results for the yaw moment coe$cients agree qualitat-
ively with those obtained from Blendermann's experimental values. For wind angles
below 803, OCIMF is the closest to Blendermann's experimental values. For angles
above 803, Gould's estimates are the closest.
204 M.R. Haddara, C. Guedes Soares / Marine Structures 12 (1999) 199}209

Fig. 5. Longitudinal force coe$cient, tanker in ballast.

Fig. 6. Side force coe$cient, tanker in ballast.

Fig. 7. Yaw moment coe$cient, tanker in ballast.


M.R. Haddara, C. Guedes Soares / Marine Structures 12 (1999) 199}209 205

In conclusion, it seems that this comparative study indicates that there is no general
agreement between the four methods available for the estimation of wind forces on
ships. The discrepancy between the estimates obtained using the other three methods
and Blendermann's experimental results may be caused by the limited range of
experiments carried out in these studies and the statistical analysis carried on the data
to obtain formulas for the estimation of the forces.

3. An universal model for the estimation of wind loads on ships

The trend of using neural network techniques in the parametric identi"cation of


mathematical models using measured input and output of the system is growing. One
of the main advantages of this method is that no functional relationship between the
dependent and the independent variables has to be assumed a priori. This technique
was used to obtain a universal expression for the wind loads. Blendermann's [4]
experimental data were used for training a feed-forward neural network. Optimisation
of the network was obtained using the back-propagation method. The data for
a tanker, which were not used in the training of the network, was used to test the
ability of the network to predict wind loads for a new ship. This approach yields
a mathematical expression which can be used to estimate loads on any ship. This
means that this is a universal mathematical expression which does not depend on the
type of ship.
The following expressions are used to calculate the coe$cients of the wind forces:
m
C " + c H , k"1, 2, 3, (1)
k ki ki
i/1
where k"1, 2, 3 refer to the longitudinal force, transverse force and yaw moment,
respectively, and
[1!e~Gki]
H " ,
ki [1#e~Gki]

5
G " + w x , k"1, 2, 3 (2)
ki kij j
j/l
and
A A ¸
x " L, x " r , x " ,
- ¸2 2 B2 3 B

S
x " , x "e, x "1, (3)
4 ¸ 5 6

where A is the lateral projected area, A is the transverse projected area, s is the
L T
distance between the centre of the lateral projected area and the midship section of
the ship, e is the angle between the centre line of the ship and the wind velocity, ¸ is
206 M.R. Haddara, C. Guedes Soares / Marine Structures 12 (1999) 199}209

the ship's length and B is the ship's beam. The values for the weights, c and w , are
ki kij
calculated by the neural network.
A neural network having one hidden layer was used. The input layer has six inputs
as shown in Eq. (3). The hidden layer has 11 nodes. The experimental data for 19 ships
were used to train the network. Table 2 shows a list of the ships used in the training of
the network. The weights, which were calculated by the neural network, are used to
predict the wind forces acting on the same tanker that has been used in the compara-
tive study in the previous section. This ship was not included in the data used to train
the network. It should be noted that the training data did not include any tankers.
Furthermore, all the ships used in the training of the network were smaller than the
tanker used to test the generalisation ability of the network.
Figs. 8}13 show a comparison between the experimental and the predicted wind
forces coe$cients for the tanker. The predicted data agreed qualitatively with the
experimental data in all cases. The errors in estimating the maximum forces on the
tanker using the neural network are much smaller than those produced by the other
methods. These results show that this method produces much better predictions than
that given by the methods currently available in the literature.

4. Conclusions

A review of the available methods for the calculation of wind loads on ships and
o!shore structures is presented. A comparative study was carried out, where the

Table 2
Ships used in the training of the neural network

Ship Load (m) B (m) D (m)

Container ship (full) 210.75 30.50 11.6


Container ship (empty) 210.75 30.50 9.6
Container ship (loaded) 210.75 30.50 9.6
Container ship (empty) 216.40 23.77 6.94
Drill ship 150.1 21.35 7.00
Cruise ship 143.90 17.35 5.90
Cruise ship 161.00 29.00 6.05
Cutter 25.05 5.80 2.50
Cargo ship (loaded) 141.1 18.50 7.32
Cargo ship (empty) 141.1 18.50 4.43
Cargo ship (loaded) 155.45 23.10 8.69
Cargo ship (container on deck) 155.45 23.10 8.69
Research vessel: Wind from Port 55.00 12.50 3.95
Research vessel: Wind from Starboard 55.00 12.50 3.95
Speed boat 53.60 9.20 2.50
O!shore supply vessel 61.95 13.00 4.85
O!shore supply vessel 61.00 13.00 4.85
Gas tanker (loaded) 274.00 47.20 10.95
Gas tanker (ballast) 274.00 47.20 8.04
M.R. Haddara, C. Guedes Soares / Marine Structures 12 (1999) 199}209 207

Fig. 8. Predicted and measured longitudinal force coe$cient (loaded tanker).

Fig. 9. Predicted and measured side force coe$cient (loaded tanker).

Fig. 10. Predicted and measured yaw moment coe$cient (loaded tanker).
208 M.R. Haddara, C. Guedes Soares / Marine Structures 12 (1999) 199}209

Fig. 11. Predicted and measured longitudinal force coe$cient (tanker in ballast).

Fig. 12. Predicted and measured side force coe$cient (tanker in ballast).

Fig. 13. Predicted and measured yaw moment coe$cient (tanker in ballast).
M.R. Haddara, C. Guedes Soares / Marine Structures 12 (1999) 199}209 209

ahead force, side force and yawing moment were calculated for a tanker in loaded and
ballast conditions using four methods. The quality of the predictions is assessed by
comparing the numerical results with experimentally obtained results.
Generally speaking the comparative study indicates that there is no general agree-
ment between the methods used for the estimation of wind forces on ships. The
experimental results obtained by Blendermann are the most comprehensive and
reliable among the four methods. The discrepancy between the estimates obtained
using the other three methods and Blendermann's experimental results may be caused
by the limited range of experiments carried out in these studies and the statistical
analysis carried on the data to obtain formulas for the estimation of the forces.
In the second part of the paper, a neural network technique was used to obtain
a universal expression for the estimation of wind loads on ships. The tanker, which
was used for the comparative study, was chosen to test the ability of the network to
predict wind loads on ships. It should be noted that this tanker was not used in the
training of the network. It was found that the numerical predictions agree well with
the experimental results, suggesting that the neural network technique can be used to
generate expressions, which produce better results than those given by the methods
currently available in the literature.

References

[1] Oppen, AN, Kvitrud A. Wind induced resonant cross #ow vibrations on Norwegian o!shore #are
booms. Proceedings of the 14th International Conference on O!shore Mechanics and Arctic Engin-
eering (OMAE), v1-A, New York: ASME, 1995, p. 341}54.
[2] Isherwood RM. Wind resistance of merchant ships. Trans. Roy. Inst. Naval Architects 1972;
114:327}38.
[3] Gould RWF. The estimation of wind loads on ship superstructures. The Royal Institution of Naval
Architects, monograph, No. 8, 1982. p. 34.
[4] Blendermann W. Schi!sform und Windlast-Korrelations-und Regressionanalyse von Windkanalmes-
sungen am Modell. Report No. 533. Institut fur Schi!bau der Universitat Hamburg, 1993a. 99 pages
plus Appendix.
[5] Blendermann W. Wind loads on moored and manoeuvring vessels. Proceedings 12th International
Conference on O!shore Mechanics and Arctic Engineering (OMAE), New York: ASME, 1993, v1, p. 183.
[6] Blendermann W. Parameter identi"cation of wind loads on ships. J. Wind Engng. Ind. Aerodyn.
1994;51:339}51.
[7] Blendermann W. Estimation of wind loads on ships in wind with a strong gradient. Proceedings of the
14th International Conference on O!shore Mechanics and Arctic Engineering (OMAE), New York:
ASME, 1995, v1-A, p. 271}7.
[8] Brix J. editor. Manoeuvring Technical Manual. Appendix I: Wind forces and moments in dimension-
less form, Part 1, Schi! & Hafen 42, 1990. No 2, p. 41, No 3, p. 55, No 4, p. 47.
[9] Brix J. editor. Manoeuvring technical manual. Hamburg: Seehafen, 1993. p. 147 and 236.
[10] Lee TS, Low HT. Wind e!ects on o!shore platforms: a wind tunnel model study. Proceedings of the
Third International O!shore and Polar Engineering Conference, Singapore, 1993. p. 466}470.
[11] OCIMF. Prediction of wind loads and current loads on VLCCs. 1994.
[12] Ohmatsu S, Takai R, Sato H. On the wind and current forces acting on a ultra large #oating platform.
Proceedings of the 14th International Conference on O!shore Mechanics and Arctic Engineering
(OMAE), New York: ASME, 1995, v1-A, p. 475}81.
[13] Chen Q, Gu Z, Sun T, Song S. Wind environment over the Helideck of an o!shore platform. J Wind
Engng Ind Aerodyn. 1995;54/55:621}31.

You might also like