You are on page 1of 18

Applied Ocean Research 21 (1999) 311–328

www.elsevier.com/locate/apor

On the estimation of Morison force coefficients and their predictive


accuracy for very rough circular cylinders
J. Wolfram 1,*, M. Naghipour 2
Department of Civil and Offshore Engineering, Heriot-Watt University, Edinburgh EH14 4AS, UK
Received 28 August 1998; received in revised form 7 July 1999

Abstract
This paper makes an assessment of the various method that may be used to analyse experiment data on the force experienced by a circular
cylinder in waves and combined wave and current flows to estimate drag and inertia coefficients for use in Morison’s equation. Most of the
widely used techniques are considered together with a weighted least squares approach for time domain analysis. A set of data obtained from
experiments on heavily roughened circular cylinders of diameters 0.513 and 0.216 m in the Delta wave flume at De Voorst in Holland in
waves and simulated current has been analysed in turn by all these techniques. The experiment data was split into two halves. The first was
used for the analyses and the second was used to assess the predictive accuracy of Morison’s equation. Using the force coefficients obtained
from the different analysis techniques corresponding predicted force time series were constructed using the particle kinematics measured in
the second parts of the data sets. These predicted time series were then compared with the corresponding measured force time histories. The
root mean square error and the bias in the estimation of maximum force in each wave cycle are used as measures of predictive accuracy and
as a basis for comparing the efficiency of the different analysis techniques. It was found that the weighted least square method generally gave
the best predictive accuracy, but only by a small margin. q 1999 Elsevier Science Ltd. All rights reserved.
Keywords: Morison’s equation; Surface roughness; Keulegan–Carpenter numbers; Hydrodynamic force coefficients

Nomenclature KDp linearised drag term


KM inertia term
CD drag coefficient N number of observations
Cf frms =…0:5rDu2rms †; total force coefficient Nw number of waves of above average height
CM inertia coefficient R Dean’s reliability ratio
D cylinder diameter Re denotes real part of function
ef error term Sx …v† one-sided energy spectra
E mean square error Sxy …v† cross-spectral density of x and y
E‰xŠ expectation of x t time
f Morison force T time period of wave cycle
fe estimated or predicted Morison force u horizontal wave particle velocity
fm measured Morison force um maximum horizontal water particle velocity in
frms root mean square force wave cycle
g acceleration
p due to gravity urms root mean square horizontal wave particle velocity
i 21 u_ horizontal wave particle acceleration
Im denotes imaginary part of function uv amplitude of horizontal wave particle velocity at
k index of weighting term and roughness height frequency v
KC …um T=D†, Keulegan–Carpenter number U uniform steady current velocity
KD drag term m mean value
r density of water
s standard deviation
* Corresponding author.
v wave frequency
1
Total Oil Marine Professor of Offshore Research and Development. kxl time average of x
2
Formerly PhD student at Heriot-Watt University. uxu modulus of x
0141-1187/99/$ - see front matter q 1999 Elsevier Science Ltd. All rights reserved.
PII: S0141-118 7(99)00018-8

Downloaded from http://www.elearnica.ir


312 J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328

1. Introduction of diameters 0.513 and 0.216 m, that have been tested in


long-crested random waves in the Delta wave flume at De
The domain of Morison’s equation [28] covers cylinders Voorst in Holland. The experiments were at a large scale
with diameters up to around 20% of wavelength. Thus the with Reynolds numbers up to 5 × 105 and both cylinders had
hydrodynamic loading on jacket structures and many other a surface roughness coefficient …k=D† of 0.038 that is typical
cylindrical components such as pipelines, risers and umbi- of the hard fouling conditions found in the North Sea. These
licals fall within the Morison regime. When assessing the experiments are described in more detail in Section 2.
maximum hydrodynamic loading on a jacket structure the The data from each random wave experiment has been
designer will usually have at his disposal the estimated divided into two parts. The first has been used for the deter-
maximum wave height and corresponding period for the mination of force coefficients in the context of Morison’s
structures proposed location and estimated values of CD equation by a variety of different methods that are listed in
and CM to use in Morison’s equation. He may well wish Section 3. The different method produced somewhat differ-
to have a measure of the error or uncertainty associated ent force coefficients from the same set of data. In each case
with this prediction particularly if doing a structural relia- these force coefficients have then been used in Morison’s
bility assessment or a risk analysis. The sources of uncer- equation to predict the force time history for the second half
tainty, or error, in the predicted maximum force can be split of the set of data based on the particle kinematics measured
into three parts. The first is associated with the prediction of in this part of the data. These force predictions are then
the extreme wave and its characteristics, including height, compared with the measured force time histories and the
period and surface profile. The second is associated with the discrepancies used as measures of predictive accuracy.
prediction of the particle kinematics beneath the wave The two halves of each data set had the same nominal
surface in the field of the structure; and the third is asso- JONSWAP spectrum (Joint North Sea Wave Project) (as
ciated with the predictive accuracy of Morison’s equation described in the Section 2) but in each case the actual
given the force coefficients chosen. sample wave spectra and the individual time series for the
This paper is primarily concerned with the last of these two halves were quite different i.e. there is no repetition in
sources of error. It is not practically possible to dissociate the time series. Thus the tests of predictive accuracy were
the modelling error intrinsic to Morison’s equation from that independent of the analysis process to a much greater degree
associated from the force coefficients used within it; and than when the analysed time series is refitted with a recon-
here they are considered together. There has been a consid- structed time series using Morison’s equation. Ideally the
erable volume of experimental research undertaken to esti- data used for testing predictive accuracy should be obtained
mate the force coefficients in Morison’s equation. The early from a number of quite different wave spectra and the fit to a
results obtained from small scale experiments and in steady variety of different force time series examined. Unfortu-
flows have now been largely discarded in the light of results nately such experimental data were not available.
from larger scale experiments, at more appropriate In order to estimate the predictive accuracy a measure is
Reynolds numbers, in oscillatory and wave flow regimes needed of how well the predicted force maps onto the
(e.g. Refs. [1,33,15]). Cylinder surface conditions more recorded force. One measure would be the root mean square
representative of the marine growth coverage found error normalised by some function of the magnitude of the
offshore have also been used in some experiments (e.g. recorded signal. This would measure the quality of the
Ref. [37]). Notable experiments undertaken offshore to mapping at all points of the time series. However it is
measure forces in the Morison regime include those at the the maximum magnitude of the force involving a single
Christchurch Bay Tower off the south coast of England [5] extreme wave that is of interest in the ultimate limit-state
and at the Ocean Test Structure (OTS) in the Gulf of Mexico design assessment. In the fatigue limit-state it is the range of
[22]. Whilst these are, in principle, the most realistic experi- the force produced by each wave that is of concern. Thus the
ments, the resulting force coefficients show considerable measure of predictive accuracy used here is based on the
scatter associated with the difficulties of accurate simulta- difference between the measured maximum fm and
neous measurements of force and particle kinematics at the the predicted maximum force fe in each wave cycle.
same location in uncontrollable conditions. As a result of all In a statistical sense a good estimator should be unbiased
these activities, there are available in the literature, values of and of minimum variance. Thus two parameters may be
CD and CM for a variety of surface roughness conditions and used to assess how well a predicted force time series
Keulegan–Carpenter numbers. However there are few compares with the corresponding measured force time
measures of their predictive accuracy when used in Mori- series. The chosen non-dimensional parameters are the
son’s equation. normalised mean error (MNE) or bias and root mean square
This paper addresses the question of predictive accuracy error (RMSE):
of Morison’s equation and its relationship with the analysis
method used to obtain force coefficients from experimental
data. This has been done in the context of a particular set of 100 X
Nw
…fm †i 2 …fe †i
MNE ˆ …1†
data obtained with two heavily roughened circular cylinders Nw iˆ1 …fm †i
J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328 313

wave maker carriage fixed pile

beach

Fig. 1. Schematic diagram of fixed and mobile cylinders in the large wave flume.

and techniques. Section 6 presents a discussion of the results


v from the analysis of the experimental data using the various
u Nw  
u 1 X …fm †i 2 …fe †i 2 approaches and their associated limitations. Finally some
RMSE ˆ 100t ; …2†
Nw iˆ1 …fm †i conclusions are drawn.

where Nw is the number of waves of above average height.


As these parameters can be unduly influenced when fm is 2. Description of the experiments
small and the absolute error is large it was thought desirable
not to consider small waves and their corresponding forces During September and October 1993 a series of experi-
when predicting the measured time series. For jacket type ments were undertaken to examine the wave loading on two
offshore structures the ultimate loading involves very large large scale circular cylinders in the Delft Hydraulic Labor-
waves and most of the fatigue damage also occurs in larger atory’s Delta wave flume in the Netherlands (DHL). This
waves; and so the ability to predict the forces due to small flume is 230 m long, 5 m wide, 7 m deep and during the
waves is of little practical interest. Therefore it was decided tests, was filled with water to a depth of about 5 m. The
to see how well the measured force due to waves of above waves were generated by a programmable, hydraulically
average height could be predicted. driven, piston type wave-maker and their energy was dissi-
The Keulegan–Carpenter number (KC—defined in the pated at the other end of the flume through the use of a 1:6
nomenclature) for the tubular members of an offshore struc- sloping concrete beach. This facility is capable of generat-
ture varies not only with the height and period of each wave ing regular and random waves with a range of periods of
but also with the diameter of the various members and their about 3–10 s and wave heights up to about 2 m over most of
location relative to the free surface. In the design of offshore the range of periods.
structures it is usual to ignore the variation of hydrodynamic For the random wave experiments the JONSWAP [12,21]
coefficients with KC and simply to use the same single spectrum was used and the results presented in this paper are
values for the drag and inertia coefficients for the whole for experiments in long crested random waves with a signif-
structure. That approach is reflected in the analysis under- icant wave height of 1.5 m and a peak period of 5.9 s.
taken here where the emphasis has been on determining a During the random wave experiments occasionally indivi-
single mean value for each coefficient from each random dual wave crests over-topped the flume implying a wave
wave experiment. It is these single values that are then used crest elevation of more than 2 m. The larger waves were
in the prediction of the force time histories. Furthermore, visibly non-linear with more sharply peaked, and sometimes
variations of force coefficients with KC are not readily breaking, crests.
discernible from frequency domain analysis of random For simulating the effects of current and combined wave/
wave data. As one of the objectives of the study was an current flows the flume is equipped with an 8 m by 6 m
assessment of the relative merits of the various time domain towing carriage that can attain a steady velocity of 1 m/s
and frequency domain analysis techniques a common basis and runs on a set of rails on the top of the flume walls. The
for comparison was required and hence the mean values of maximum towing distance is dependent upon the test set up.
CD and CM have been used. The other objective was to give For these experiments the carriage speeds were ^ 1 m/s
some measure of the uncertainty and bias involved in using (for the larger cylinder) and ^ 0.5 m/s (for the smaller
Morison’s equation for the prediction of in-line forces that cylinder) and the towing distance was approximately
would be helpful in structural reliability calculations and 110 m. Fig. 1 shows a schematic longitudinal section of
structural assessments and therefore following typical the flume with a cylinder mounted on the moving carriage,
design practice was desirable. a fixed cylinder, the beach and the wave-maker.
Section 2 of the paper describes the experiments that were The two vertical cylinders used for the experiments had
undertaken in the Delft Hydraulics Laboratories (DHL) long base diameters of 0.21 m (small) and 0.5 m (large) and were
wave flume at De Voorst in Holland as part of an EC/ mounted in turn on the towing carriage and at fixed loca-
EPSRC funded project. Section 3 overviews the various tions in the flume. Both cylinders were manufactured from
methods for the prediction of force coefficients from experi- stainless steel and were covered with the roughness pattern
mental data. Section 4 looks in detail at the time domain that was originally developed by Wolfram [38] which has
analysis approaches and Section 5 at the frequency domain been shown to simulate well the effect of hard marine
314 J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328

small pile large pile


5
MWL MWL

4 1.5m
2 2.5m
3

2 5m 7m 1

Fig. 2. Schematic diagram of cylinders in the flume showing the force sleeve positions.

growth. This roughness consists of a pseudo-random rigidly attached at the bottom. Braces were fitted between
arrangement of three different sizes of right square pyramids the cylinder and the bottom support so that the estimated
which in each case have heights of the same dimensions as natural frequency of the system as a whole was about 4 Hz.
the base. The roughness elements were cast in fibreglass on This frequency is some 16 times higher than the highest
semi-circular cylindrical shells which were strapped around wave frequency tested but sometimes some vibration energy
the cylinder giving an effective roughness ratio …k=D† of was input to the cylinder by the motion of the carriage and
0.038 and corresponding effective diameters for the large was evident in the force signal traces [26].
and small rough cylinders of 0.513 and 0.216 m, respec- The ambient flow velocity was always measured at the
tively [26]. The schematic general elevation of the cylinders longitudinal location of the cylinder and at the elevation of
and location of the force sleeves in the cross sections of force sleeves using either electromagnetic flow meters
flume is shown in Fig. 2. (EMF) or perforated ball velocity meters. (PVM). The
Measurements of the wave force time series for both EMF meters are very sensitive to magnetic fields and the
cylinders were obtained using strain-gauged force measur- output signals had to be filtered to reduce noise. The EMF
ing sections, each half a diameter in length and capable of meters were generally used except for the large mobile
measuring both the in-line and transverse forces. The small cylinder and for the upper force sleeves on the small fixed
cylinder had five forces sleeves; one centred 0.5 m above cylinder where PVMs were used. The original form of the
the mean water level (MWL) and four with centres at PVM was first used by Bishop [7] in wave force investiga-
distances of 0.5, 1.5, 2.5 and 3.5 m below the MWL as tions at the second Christchurch Bay Tower. These instru-
shown in Fig. 2. The large cylinder had two forces measur- ments have also been used to measure particle velocities in
ing sleeves centred at distances of 1.5 and 2.5 m below the three-dimensions in waves at laboratory scale by Chaplin
MWL. The measurements at the force sleeves at position 2 and Subbiah [16]. The signals from the PVMs represent the
(see Fig. 2) on both the small and large cylinders were used Morison-type forces on the ball and these have been cali-
for the analyses described in this paper. These sleeves were brated in oscillatory flow. The wave surface elevation was
sufficiently close to the water surface to give reasonably also measured continuously at the same sections of the wave
large measured forces but were always submerged during flume as the axes of the cylinders using surface-following
the experiments ensuring continuity of the force measure- capacitance-type wave probes. The analogue voltages from
ments. the flow meters, wave probes and load cells were digitised
When mounted on the carriage the cylinder was rigidly using an A/D converter at a sampling frequency of 40 Hz.
fixed to the carriage at the top and had a horizontal plate The errors associated with the measurements of force,

Table 1
Details of all experiments analysed (R is Dean’s reliability ratio and KC is Keulegan–Carpenter number)

Run no. Current (m/s) Pile diameter (mm) No. of waves R KC

Min Mean Max Min Mean Max

1 0 513 291 0.03 0.74 1.46 0.2 5.5 17.5


2 1 513 130 0.72 3.79 15.5 1.92 13.5 25.3
3 21 513 102 3.22 5.66 24.8 1.43 17.4 32.3
4 0 216 286 0.08 1.72 3.21 0.4 12.8 37.9
5 0.5 216 152 0.99 3.63 6.0 0.75 21.8 45.9
6 2 0.5 216 141 0.63 5.18 10.5 1.92 22.9 57
J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328 315

16 S (ω) (second part of data) Run1

14
S (ω) (first part of data)

S (ω) (m 2 sec)
12

10

0
0 0.1 0.2 0.3 0.4 0.5 0.6
Frequency (Hz)

Fig. 3. Comparison between surface elevation spectra in the two halves of the random wave experiment for the fixed large pile.

water particle velocity and wave surface elevation have not The authors discuss below all the techniques that have been
been quantified explicitly in a formal manner. But the proce- used, at least to their knowledge. However they have not
dures used are well established and the techniques have considered approaches which seek to modify the form of
been employed for over ten years satisfactorily. It is consid- Morison’s equation, using systems identification and other
ered that the measurement errors are small compared to the techniques, by the addition of further terms (see for example
predictive errors that are the subject of this paper. Refs. [34,39]). Such modified forms of Morison’s equation
The details of the six experiment runs considered here are are still the subject of research and are not currently used in
given in Table 1 from which it can be seen that there were practice.
experiments with a current in the wave direction, with a The various techniques considered here can be cate-
current opposing the wave direction and with no current gorised according to type and those that have been used in
for both the small and large cylinders. The currents were the present study are identified as Method 1, Method 2, etc.
achieved by translating the cylinder on the moving carriage in the following list.
either away from the wave-maker (i.e. current in the wave
1. Time domain techniques:
direction) or towards the wave-maker (i.e. current opposing
the wave direction). Several such translations needed to be Wave by wave analyses
patched together to produce a complete run of 10 min
Fitting at maximum velocity and acceleration (Evans)
duration for each experiment condition. The experiments
Fourier averaging
with the stationary cylinders were nominally of 30 min
duration. Bearman, Chaplin et al. [2] (Method 1)
While the two cylinders and their roughness patterns were Klopman and Kostense (Method 2) [24]
geometrically similar, the current speeds were not in exactly
Ordinary least squares (Method 3)
the same scale relationship and, as the same spectrum was
Weighted least squares
used throughout, all the experiments are distinctly different
rather than being scaled up (or down) versions of one Whole record analyses
another. The small cylinder was originally introduced into
Mean square method (Bishop and Shipway)
the programme to extend the range of KC investigated. The
Ordinary least squares (Method 4)
time series corresponding to the wave spectrum was simu-
Weighted least squares (Method 5)
lated by a sum of 200 sinusoidal components so the first and
Method of moments (Pierson and Holmes) (Method 6)
second halves of each experiment run were independent and
had significantly different sample spectra; as can be seen 2. Frequency domain techniques
from Fig. 3 which is a typical example.
Linear models (i.e. with a linearised drag term)
Cross spectral density (force and particle velocity)
3. Overview of methods for estimating force coefficients (Method 7)
Cross spectral density (force and surface elevation)
A wide variety of approaches have been used to analyse (Method 8)
experiment data to determine drag and inertia coefficients. Least squares fitting to force spectrum (Method 9)
316 J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328

Non-linear models CD can be on the order of 85 and 50%, respectively. Dean


suggested that data will be well-conditioned for evaluating
Cubic drag model of Bendat and Piersol [3] (Method
both CM and CD together when 0:25 , R , 4 and for CM
10)
only when 0 , R , 0:25 and for CD only when R . 4. The
In wave by wave analysis CD and CM are assumed constant mean and range of the reliability ratio for all the experi-
over each wave cycle and the output of the analysis can be ments analysed here are presented in Table 1 and as will
presented as a plot against KC. The alternative is to assume be seen later some of the data are poorly conditioned for the
that CD and CM are constant over the whole record and to estimation of CD and some for CM.
analysis this as a single entity. In this latter case no explicit
measure of the variance CD and CM is obtained. In random
waves the resultant average values of CD and CM obtained 4. Time domain analysis techniques
from these two approaches will be different even if essen-
The general form of Morison’s equation for a vertical
tially the same analysis technique is used. This is because in
cylinder adapted to allow for the presence of a current U
the former case each wave has an equal weighting when
is given below together with its shorthand form.
finding the mean whereas in the latter case the influence
of each wave on the outcome will be proportional to its f ˆ 0:5CD rD…u 1 U†u…u 1 U†u 1 0:25CM rpD2 u_
wave period (or more specifically the number of data points
within the wave cycle). _
ˆ KD …u 1 U†u…u 1 U†u 1 KM u: …4†
The frequency domain analysis methods all use the entire
Here f is the Morison force, r is water density, CM and CD
record and use spectra and cross spectra in various ways to
are the drag and inertia coefficients, respectively, and u and
estimate the force coefficients. In some cases the force coeffi-
u_ are the horizontal components of water particle velocity
cients are obtained as functions of frequency and in others as
and acceleration, respectively. In the absence of current the
single values. Most of the analysis techniques examined here
term U is removed.
involve linearisation of the drag term. However one non-linear
approach involving a cubic expansion is examined.
4.1. Wave by wave analyses
Most of the methods listed above use the velocity time
series measured beside the force sleeve on the test cylinder The simplest approach is to find the points of zero accel-
to estimate particle kinematics. Occasionally, both offshore eration where the corresponding force is purely drag and
and in the laboratory this is not available and the particle calculate CD directly. Similarly at the points where the velo-
kinematics must be estimated directly from the surface city is zero the force will be purely inertial and CM can be
elevation by a wave theory. Method 8 in the above list calculated (e.g. Ref. [20]). However this approach uses very
incorporates this directly in the analysis but this additional little of the data and may be subject to significant errors if
aspect could be included in any of the methods using there is noise on the velocity and force time series. It is not
random linear wave theory as discussed later. Most of the widely used and is not considered here.
methods have been extended to include the effects of a Keulegan and Carpenter [23] introduced a method to find
steady current of uniform depth-wise profile and the corre- the inertia and drag force coefficients in each cycle using
sponding expressions for the coefficients are also presented Fourier or time averaging [13]. This basic approach is
in the following sections. widely adopted and methods based on it have been devel-
Theerrorassociatedwithanyestimateofthe forcecoefficient oped and used by Bearman et al. [1,2], Bishop [6], Klopman
will depend in part on the assumptions implicit in the analysis and Kostense [24] and Davies [18].
technique used. It will also depend upon the data and whether In the method used by Bearman et al. [2] for the case
these are well or poorly conditioned for resolving CM and CD. without current, Morison’s equation is multiplied by u and
Dean [19], in the context of least squares time domain analysis, the time average taken over a wave cycle. Assuming that the
suggested that much of the scatter in the reported coefficients time-averaged cross product kuul _ is zero over a wave cycle
canbedue topoorlyconditioneddata.Hepresenteda criteria for this yields the following expression:
evaluating the suitability of data for determining CM and CD in kful
either time or frequency domain analysis. This involves CD ˆ 3
: …5†
0:5rDkuuu l
calculating the following ‘reliability ratio’
Similarly multiplying Morison’s equation by u; _ taking the
s
time average over each wave cycle and assuming kuuuuul _ is
2 ku4 l
Rˆ ; …3† zero yields:
pD ku_ 2 l
_
kf ul
CM ˆ : …6†
where kl indicates time averaging of the enclosed quantity 0:25rpD2 ku_2 l
over the wave cycle. He showed with a particular example The approach implicitly assumes that the velocity and accel-
that if the mean square error is of the order of 10% of the eration time series are orthogonal. On essentially the same
measured force, then the error in estimated values of CM and basis Klopman and Kostense [24] modified this approach
J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328 317

and multiplied Morison’s equation first by uuuu and then by u_ yields better results. So although these terms are, generally,
before time averaging to give the following pair of comparatively small they do have some effect and as they are
equations. readily calculated there is a little point in not including them.
The least-squares approach can be applied in a wave by
kfuuuul
CD ˆ …7† wave analysis or to a whole record or, indeed, in the
0:5rDku4 l frequency domain. In the time domain CD and CM are chosen
and to minimize the sum of the squares of the difference
between the measured and the estimated force at each
_
kf ul measurement point. This yields two simultaneous equations
CM ˆ …8†
0:25rpD2 ku_ 2 l that can be solved to give the following expressions:
_
Here term just the term kuuuuul is assumed to be zero. P P P P
fuuuu u_ 2 2 f u_ uuuuu_
Both of these Fourier averaging approaches can easily be CD ˆ P P P ; …13†
0:5rD{ u4 u_ 2 2 … uuuuu† _ 2
}
extended to include the effect of a uniform steady current U.
However the time averages considered above to be zero will P P P P
no longer be so and the pair of equations obtained after time _
f u_ u4 2 fuuuu uuuuu
CM ˆ P P P …14†
averaging must be solved simultaneously. The equivalent 0:25rpD { u_
2 2 4 _
u 2 … uuuuu† 2}

expressions for Eqs. (5) and (6) are then found to be:
The corresponding equations for the case when there is a
kf …u 1 U†lku_2 l 2 kf ulk…u
_ _
1 U†ul steady uniform current are obtained simply by substituting
CD ˆ …9†
0:5rD…ku…u 1 U†u3 lku_ 2 l 1 k…u 1 U†u…u 1 U†uul†
_ U 1 u for u on each occurrence of u in the Eqs. (13) and
(14). The least-squares method can be applied to any length
CM ˆ of time series. There need not be a whole number of waves
or even a whole wave cycle.
_
kf ulku…u 1 U†u3 l 2 kf …u 1 U†lku…u
_ 1 U†u…u 1 U†ul If a single wave cycle P is considered then for, say, the
particle velocity …1=N† u ˆ kul; and similarly for all
0:25rpD2 …ku_2 lku…u 1 U†u3 l 2 ku…u
_ 1 U†lku…u
_ 1 U†u…u 1 U†ul†
other time averages over a wave cycle. Thus Eqs. (13) and
…10† (14) are seen to be identical to Eqs. (11) and (12) when U ˆ
For Eqs. (7) and (8) they become: 0; for wave by wave analysis. Furthermore, if the second
terms in both the numerators and denominators of Eqs. (13)
kf …u 1 U†uu 1 Uulku_ 2 l 2 kf ulk…u
_ _
1 U†u…u 1 U†uul and (14) are considered negligible then these equations
CD ˆ
0:5rD……k…u 1 U† lku l 2 …kuu 1 Uu…u 1 U†ul†
4 _2 _ †
2
become the same as those used by Klopman and Kostense;
…11† Eqs. (7) and (8) above.
In the least-squares approach all data points in the
_
kf ulk…u _ 1 U†uu 1 Uul
1 U†4 l 2 kf …u 1 U†uu 1 Uulku…u measured force time series have an equal influence on the
CM ˆ determination of CD and CM. This may be thought a limita-
0:25rpD ……ku_ lk…u 1 U† l 2 …kuu 1 Uu…u 1 U†ul†
2 2 4 _ 2†
tion when there is little interest in predicting forces close to
…12†
zero and much more interest in predicting maximum forces.
In the derivation of Eqs. (9)–(12) no assumptions or restric- To ensure more emphasis is given to finding force coeffi-
tions have been placed on the velocity or acceleration time cients which will predict maximum forces accurately a
series or their time averages. Hence the shape of the wave is weighted least-squares approach where the weights are
unrestricted and it will be noticed that when U ˆ 0 (i.e. there is related to the magnitude of the force can be adopted. Thus
no current) these equations are different from the correspond- at each point in the time series the difference between the
ing Eqs. (5)–(8) in that there is a second term in both the measured and predicted force is multiplied by f k , where k is
numerator and the denominator of Eqs. (9)–(12). These a positive index. This means that those parts of the time
terms involve the time averages which Bearman et al. [2] series where the measured force is small have very little
and Klopman and Kostense [24] assumed were zero in their influence on the estimated values of CD and CM but when
approaches. More interestingly it is found that Eqs. (11) and the magnitude of the force is large the influence is consider-
(12) are identical to those obtained using a least-squares able. The expressions for the force coefficients when there is
approach and as will be seen later the least-squares approach a steady current U then become:
P P P P
f 2k f …u 1 U†uu 1 Uu f 2k u_2 2 f 2k f u_ f 2k …u 1 U†uu 1 Uuu_
CD ˆ P 2k P P ; …15†
0:5rD{ f …u 1 U†4 f 2k u_ 2 2 … f 2k u…u _ 1 U†uu 1 Uu†2 }
P P P P
f 2k f u_ f 2k …u 1 U†4 2 f 2k f …u 1 U†uu 1 Uu f 2k u…u
_ 1 U†uu 1 Uu
CM ˆ P 2k 2 P 2k P 2k …16†
0:25rpD { f u_
2 f …u 1 U† 2 … f u…u
4 _ 1 U†uu 1 Uu†2 }
318 J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328

Table 2
Values of CD and CM from the analysis methods in the random waves for a fixed large pile—Run 1

Method of analysis CD (mean) s cd CM (mean) s cm MNE RMSE

Time domain: wave by wave


1. Bearman et al. [2] 1.55 1.01 2.04 0.17 5.28 11.57
2. Klopman and Kostense [24] 1.38 1.11 2.04 0.17 6.78 12.31
3. Ordinary least-squares 1.88 1.54 2.08 0.15 1.32 11.41
Time domain: whole record
4. Ordinary least-squares …k ˆ 0† 1.57 – 2.04 – 5.08 11.51
5. Weighted least-squares k ˆ 1 1.54 – 2.14 – 1.23 10.82
Kˆ2 1.52 – 2.22 – 2 1.64 11.28
Kˆ3 1.50 – 2.28 – 2 4.18 12.24
6. Method of moments 1.73 – 2.06 – 2.78 11.08
Frequency domain
7. Cross spectra—force and particle vel. 1.41 0.40 1.93 0.44 11.03 14.70
8. Cross spectra—force and surface elevations 1.57 0.69 1.95 1.08 8.70 13.24
9. Least-squares fit to spectra 2.26 – 1.87 – 1.94 14.54
10. Cubic model of Bendat and Piersol 1.42 0.58 1.93 0.44 10.93 14.63

For a wave by wave analysis, in the limit when k is very Bishop and Shipway [7] to analyse the data collected in
large, CD and CM will be found such that the reconstructed the offshore experiments at the Christchurch Bay Tower.
time series exactly matches the original but only at the This approach involves estimating moments from the data
points of maximum and minimum (maximum negative) and then using a regression analysis. It is useful where cycle
force. However when predicting the maxima and minima by cycle analysis is not possible due to an unknown phase
for a different time series large values of k are found to lag between the force and velocity measurements; as
produce poor results. When k ˆ 0 then this method clearly occurred at the Christchurch Bay Tower experiments.
becomes the same as the simple least squares method. However with scatter in the data it can lead to some negative
As the wave by wave analysis usually involves simple estimates of CD, as discussed by Davies [18], and it is not
averaging of the force coefficients found in each wave cycle used here. The techniques used in this study for whole
some of the weighting effects will be reduced and very large record time domain analysis are the weighted least-squares
waves will have no more influence on the final outcome than approach (outlined above) and the method of moments.
very small ones. So in the study here the authors have In the weighted least-squares approach the authors have
applied this approach predominantly to whole experiment used various values of k in Eqs. (15) and (16) and have
records. found that this can be optimised in an iterative manner to
give a minimum predictive error in the peak force regions.
4.2. Whole records time series analysis Results for various values of k from 0 (corresponding to
ordinary least squares) to three are presented in Tables 2–
Some techniques require the variability of irregular 7 and for these sets of data a value of two gives, on average,
waves in a whole time series to work at all. One such is the best results.
the mean square method developed by Bishop [6] and The method of moments was introduced by Pierson and

Table 3
Values of CD and CM from the analysis methods in the random waves for a mobile large pile—Run 2

Method of analysis CD (mean) s cd CM (mean) s cm MNE RMSE

Time domain: wave by wave


1. Bearman et al. [2] 1.32 0.37 1.49 0.56 15.18 16.69
2. Klopman and Kostense [24] 1.32 0.36 1.50 0.57 15.54 17.01
3. Ordinary least-squares 1.51 0.37 1.72 0.72 2.82 8.44
Time domain: whole record
4. Ordinary least-squares …k ˆ 0† 1.42 – 1.59 – 8.74 11.50
5. Weighted least-squares k ˆ 1 1.44 – 1.82 – 7.29 10.49
kˆ2 1.44 – 1.98 – 6.26 9.85
kˆ3 1.45 – 2.13 – 5.36 9.35
6. Method of moments 1.46 – 0.53 – 7.80 11.01
Frequency domain
8. Cross spectra—force and surface elevations 1.10 0.49 2.34 1.31 24.81 25.55
9. Least-squares fit to spectra 1.14 – 2.32 – 22.6 23.4
J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328 319

Table 4
Values of CD and CM from the analysis methods in the random waves for a mobile large pile—Run 3

Method of analysis CD (mean) s cd CM (mean) s cm MNE RMSE

Time domain: wave by wave


1. Bearman et al. [2] 1.45 0.33 1.36 0.62 1.28 7.50
2. Klopman and Kostense [24] 1.44 0.32 1.37 0.60 2.06 7.60
3. Ordinary least squares 1.54 0.30 1.69 1.81 2 4.84 9.14
Time domain: whole record
4. Ordinary least squares …k ˆ 0† 1.48 – 1.55 – 2 0.89 7.55
5. Weighted least squares k ˆ 1 1.47 – 1.62 – 2 0.14 7.40
kˆ2 1.46 – 1.63 – 0.34 7.36
kˆ3 1.45 – 1.64 – 0.69 7.36
6. Method of moments 1.55 – 1.68 – 2 5.54 9.58
Frequency domain
8. Cross spectra—force and surface elevations 1.43 0.05 1.43 0.27 2.40 7.68
9. Least squares fit to spectra 1.55 – 1.68 – 2 5.54 9.58

Holmes [32] and used by Muga and Wilson [29]. It starts by Tung and Huang [35] found:
assuming that the velocity and acceleration, u and u; _ are m1
independent normal random variables with zero mean. KD ˆ …19†
2s2u ‰…1 1 g2 †Z…g† 1 gz…g†Š
Then expressions are obtained for the expected values,
E‰f 2 Š and E‰f 4 Š from Morison’s equation in terms of KD and
and KM and the moments of u and u: _ These expected values s
m22 2 KD2 s4u …3 1 6g 1 g4 †
are then estimated with the means of the square and the KM ˆ ; …20†
fourth power of the measured force, denoted here by m 2 s2u_
and m 4, and the resultant pair of simultaneous equations is in which m 1 is the mean value
R of the measured force, g ˆ
solved to yield the following expressions: m u = su and Z…g† ˆ g0 z…x† dx and z…g† ˆ
s …2p†20:5 exp…20:5g2 †:
m2 2 3KM su
2 4
As well as the assumption of a normal distribution for
KD ˆ …17†
s2u_ wave particle velocity and acceleration, the method of
moments also requires long records and significant irregu-
and larity to allow the accurate estimation of the higher
moments. However the method is robust in that any
v
u unknown phase lags between force and velocity measure-
u …m 2 3m2 †0:5
KM ˆ t 4 p 42 …18† ments will not influence the results provided the time series
78su_ is long enough.
A few points can be made to summarise the time domain
For the corresponding case of a co-existing steady current analysis techniques. The least squares method requires no

Table 5
Values of CD and CM from the analysis methods in the random waves for a fixed small pile—Run 4

Method of analysis CD (mean) s cd CM (mean) s cm MNE RMSE

Time domain: wave by wave


1. Bearman et al. [2] 2.05 1.44 1.91 0.40 2 18.3 27.76
2. Klopman and Kostense [24] 1.95 1.34 1.91 0.40 2 13.5 23.75
3. Ordinary least squares 1.95 1.34 1.91 0.40 2 13.5 23.75
Time domain: whole record
4. Ordinary least squares …k ˆ 0† 1.61 – 1.93 – 2.30 15.41
5. Weighted least squares k ˆ 1 1.55 – 2.22 – 0.83 14.48
kˆ2 1.51 – 2.35 – 2 1.56 14.79
kˆ3 1.49 – 2.57 – 2 4.43 16.14
6. Method of moments 1.46 – 2.45 – 0.92 14.47
Frequency domain
7. Cross spectra—force and particle vel 1.74 0.15 1.70 0.52 5.67 16.95
8. Cross spectra—force and surface elevations 1.66 0.51 2.31 2.4 2 5.14 16.25
9. Least squares fit to spectra 1.85 – 1.84 – 2 7.81 20.03
10. Cubic model of Bendat and Piersol 1.28 0.66 1.85 0.51 21.93 24.73
320 J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328

Table 6
Values of CD and CM from the analysis methods in the random waves for a mobile large pile—Run 5

Method of analysis CD (mean) s cd CM (mean) s cm MNE RMSE

Time domain: wave by wave


1. Bearman et al. [2] 1.53 0.35 1.56 0.42 2 6.11 13.75
2. Klopman and Kostense [24] 1.47 0.34 1.56 0.44 2 2.48 12.14
3. Ordinary least squares 1.60 0.28 1.61 0.48 2 10.9 16.85
Time domain: whole record
4. Ordinary least squares …k ˆ 0† 1.48 – 1.74 – 2 3.88 12.62
5. Weighted least squares k ˆ 1 1.44 – 2.07 – 2 2.34 11.98
kˆ2 1.42 – 2.22 – 2 1.83 11.80
kˆ3 1.41 – 2.31 – 2 1.83 11.79
6. Method of moments 1.71 – 2.56 – 2 22.3 26.4
Frequency domain
8. Cross spectra—force and surface elevations 1.23 0.67 2.19 3.03 10.53 14.67
9. Least squares fit to spectra 1.47 – 1.59 – 2 2.26 12.06

specific assumptions concerning the data and can be applied categories; those that involve the linearisation of Morison’s
to any waves regular or irregular, linear or non-linear, and to equation, and those that are non-linear in some way. Three
a record of any (sensible) length. The Fourier averaging of the former type and one of the latter are considered here.
methods require whole wave cycles and assume that velo- Not considered here are models that depart from the original
city and acceleration time series are orthogonal. The method form of Morison’s equation by adding additional terms.
of moments further requires that the velocity and accelera- There have been many such attempts to improve Morison’s
tion time series are zero mean random processes with Gaus- equation most recently using systems identification
sian distributions. Finally the weighted least squares techniques (see for example Refs. [34,39]).
approach allows emphasis to be placed on force maxima All the frequency domain techniques assume that the
at the expense of smaller values of force. water particle velocity is a random Gaussian process
The data analysed here were obtained in waves that were which, in the absence of current, is also a zero mean process,
significantly non-linear and in some cases sufficiently steep and further, that uuuu in the drag term can be approximated
to have breaking crests (like real storm waves!). The results by a polynomial yielding:
of the analysis for the six experiments are shown in Tables r
2–7 and, not surprisingly, show that the analysis methods 8
uuuu ˆ usu …21†
making fewer assumptions give better predictive accuracy. p
As accuracy is measured in term of ability to predict as the linear approximation, and
maxima it is not surprising that the weighted least squares (r"   #)
gives on average (see Table 8) the best results. 2 u 1 u 3
uuuu ˆ usu2
1 …22†
p su 3 su

5. Frequency domain methods as the cubic approximation. See Refs. [3,9,10,11] for more
details. In the presence of a uniform current a similar poly-
Frequency domain methods may be divided into two nomial fitting procedure can be used and then the mean of

Table 7
Values of CD and CM from the analysis methods in the random waves for a mobile large pile—Run 6

Method of analysis CD (mean) s cd CM (mean) s cm MNE RMSE

Time domain: wave by wave


1. Bearman et al. [2] 1.04 0.34 1.56 0.46 14.99 17.31
2. Klopman and Kostense [24] 1.05 0.32 1.56 0.46 14.17 16.64
3. Ordinary least squares 1.16 0.25 1.76 0.81 4.38 10.71
Time domain: Whole record
4. Ordinary least squares …k ˆ 0† 1.21 – 1.74 – 1.06 10.06
5. Weighted least squares k ˆ 1 1.21 – 1.86 – 0.35 10.25
kˆ2 1.21 – 1.94 – 0.14 10.34
kˆ3 1.21 – 1.94 – 0.06 10.34
6. Method of moments 1.15 – 2.99 – 2 0.10 12.58
Frequency domain
8. Cross spectra—force and surface elevations 1.80 0.81 1.63 0.50 2 44.0 46.2
9. Least-squares fit to spectra 1.21 – 1.74 – 1.06 10.06
J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328 321

Table 8
Values of CD and CM from the analysis methods in the random waves (averaged in the six runs)

Method of analysis %MNE %RMSE CD (mean) CM (mean)

Time domain: wave by wave


1. Bearman et al. [2] 10.19 15.76 1.49 1.65
2. Klopman and Kostense [24] 9.09 14.91 1.44 1.66
3. Ordinary least squares 5.56 12.83 1.42 1.80
Time domain: whole record
4. Ordinary least-squares …k ˆ 0† 3.66 11.44 1.46 1.77
5. Weighted least-squares k ˆ 2 2.03 10.90 1.51 2.04
6. Method of moments 6.57 14.19 1.51 2.05
Frequency domain
8. Cross spectra—force and surface elevations 15.93 20.6 1.47 1.98
9. Least-squares fit to spectra 24.14 30.80 1.82 1.79
Averages 10.14 16.67 1.54 1.86

the water particle velocity is no longer zero but equal to the where Su …v† is the one-sided spectral density function of the
current speed U. This yields [10,14] input i.e. measured particle velocity and Suf …v† is the cross-
spectral density of the input and output i.e. measured parti-
uuuu ˆ s2u ‰…1 2 g2 †…2Z…g† 2 1† 1 2gz…g†Š
cle velocity and measured force.
1 2su ‰g…2Z…g† 2 1† 1 2z…g†Šu …23† Borgman [11] proposed another type of linear cross-spec-
tral density model for use when particle kinematics have to
for the linear approximation where the various terms have be inferred from the wave surface elevation. Linear random
been defined above after
pEq. (20). If the current is zero then wave theory and the linear dispersion relationship are used
g ˆ 0; and z…g† ˆ 1= 2pand the expression above reduces to establish time series for the wave particle kinematics
to the linear approximation in Eq. (21). from the wave surface elevation time series. The cross spec-
tra in the expression below (obtained by Borgman) are then
5.1. Linear models computed.
r
There are a variety of related analysis methods based on 8
Shf …v† ˆ K s S …v† 1 KM Shu_ …v†: …28†
the use of spectra and cross spectra that have been devel- p D u hu
oped by Borgman [9,11] and Bendat and Piersol [3] among
Each of the cross spectra in this equation has a real and
others. The first considered here involves the cross-spectra
imaginary part yielding two equations that can be solved
of measured force and wave particle velocity and is the
simultaneously to provide the following expressions for
linear counterpart of the non-linear model considered in
the force coefficients as functions of frequency:
the Section 5.2. In the linear model Morison’s equation
becomes; r !
p 2 Chu_ Qhf 2 Qhu_ Chf
r CD …v† ˆ ; …29†
8 8 rDsu Chu_ Qhu 2 Qhu_ Chu
_ 1 KD
F…t† ˆ KM u…t† _ 1 KDp u…t†: …24†
s u…t† ˆ KM u…t†
p u !
By assuming random linear wave theory an expression is 4 Chf Qhu 2 Chu Qhf
CM …v† ˆ : …30†
obtained linking Morison force and wave particle velocity. rpD2 Chu_ Qhu 2 Qhu_ Chu
Taking Fourier transforms the corresponding frequency
domain expression is found with force and velocity linked Here C and Q indicate the real and imaginary parts of cross-
by the transfer function Huf …v† from which the force coeffi- spectra in Eq. (28).
cients can be estimated as follows: In the case where there is current as well as waves then
r Eq. (28) may be written as:
2
KDp …v† ˆ rDCD …v† s ˆ Re‰Huf …v†Š …25† Shf …v† ˆ 4…z…g† 1 uguZ…g††KD su Shu …v† 1 KM Shu_ …v†; …31†
p u
and the drag coefficient is then given by:
pD2 Im‰Huf …v†Š !
KM …v† ˆ CM …v† ˆ : …26† Chu_ Qhf 2 Qhu_ Chf
4 v CD …v† ˆ
1
:
The transfer function Huf …v†can be estimated from (see for 2rDsu …z…g† 1 uguZ…g†† Chu_ Qhu 2 Qhu_ Chu
example Refs. [4,31]): …32†
Suf …v† The presence of the current does effect the equation for the
Huf …v† ˆ …27†
Su …v† inertia coefficient.
322 J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328

Although they may not appear so at first sight Eqs. (29) kinematic time series. In this respect the method is similar
and (30) are very similar to Eqs. (25) and (26) and this to those of Bishop [6] and Pierson and Holmes [32] and
method (denoted as Method 8) is quite similar to the one hence may be useful when there is an unknown phase lag
above (Method 7). In both methods the drag and inertia between force and velocity measurements. However the
coefficients are obtained as functions of frequency, rather overall the predictive accuracy of this method (Method 9)
than Keulegan–Carpenter. Here the results have been aver- is seen to be poorest of all in Table 8.
aged over the frequency range from 0.1 to 0.4 Hz (where
from Fig. 3 most of the energy is seen to lie) to provide
single values of CD and CM to be used in the prediction 5.2. Non-linear models
study. The predictive accuracy of these two methods may Non-linear frequency domain models have been consid-
be judged from the results presented in Tables 2–7 and ered theoretically by Bendat and Piersol [3]. These were
summarised in Table 8. Overall both are seen to be signifi- applied to small scale experiments by Vugts and Bouquet
cantly poorer than the time domain approaches. [36], and to large scale experiment data, obtained in the De
Borgman [11] also proposed a least-squares frequency Voorst wave flume, by Bliek and Klopman [8]. More
domain method for determining force coefficients using a recently non-linear extensions to Morison’s equation have
linear model based on the following equation for power been considered by Worden et al. [39], among others, using
spectral densities. systems identification techniques applied to a variety of
8KD2 s2u experiment data.
Sf …v† ˆ Su …v† 1 KM
2
Su_ …v†: …33† The model considered here is shown in Fig. 4 and was
p
originally developed by Bendat and Piersol. It uses a cubic
The spectra for the particle kinematics can be obtained approximation in which the terms Tl …v† and Tn …v† represent
from the surface elevation or, alternatively, they may be the transfer function of the linear and non-linear parts of
computed more directly, and usually more accurately, Morison’s equation. To restrict this model to the usual Mori-
from the velocity time series. However for the analysis son form Tl …v† is considered to be purely imaginary (i.e.
here they have been computed from the wave surface eleva- with 908 phase), corresponding to the inertia term and Tn …v†
tion time series. purely real corresponding to the drag term. The case for
The coefficients KM and KD are estimated such that the doing this is discussed by Vugts and Bouquet. The corre-
predicted force spectrum Sf …v† from Eq. (23) is the best sponding form of Morison’s equation is then:
least squares fit to the measured force spectrum. This yields
the following equations: _ 1 KD1 u…t† 1 KD2 u…t†3 ;
f …t† ˆ KM u…t† …37†

X
N X
N X
N X
N
Su …vi †Su_ …vi † Su …vi †Sf …vi † 2 S2u …vi † Su_ …vi †Sf …vi †
iˆ1 iˆ1 iˆ1 iˆ1
2
KM ˆ !2 ; …34†
X
N X
N X
N
Su …vi †Su_ …vi † 2 S2u …vi † S2u_ …vi †
iˆ1 iˆ1 iˆ1

X
N X
N X
N X
N
Su …vi †Su_ …vi † Su_ …vi †Sf …vi † 2 S2u_ …vi † Su …vi †Sf …vi †
KD2 ˆ …p=8† iˆ1 iˆ1
!2
iˆ1 iˆ1
; …35†
X
N X
N X
N
Su …vi †Su_ …vi † 2 S2u …vi † S2u_ …vi †
iˆ1 iˆ1 iˆ1

where N is the number of discrete frequencies at which the where


spectrum is computed from the time series. Note that in this
single value of each of the force coefficients is obtained. CM rpD2 3as2u CD rD aC D r D
KM ˆ ; KD1 ˆ ; KD2 ˆ
When there is a current in addition to waves then Eq. (33) 4 2 2
may be shown [25] to take the form:
and
S…v† ˆ 16…z…g† 1 uguZ…g†† 2
KD2 s2u Su …v† 1 2
KM Su_ …v† …36† r
1 2
aˆ :
3 su p
where the terms have the same meaning as in Eqs. (19) and
(20). The corresponding pair of equations for the force coef- The corresponding frequency domain expression is found to
ficients, similar in form to Eqs. (34) and (35), is readily be
obtained. In this method it is worth noting that no account
is taken of the phase between the force and particle F…v† ˆ ivKM u…v† 1 KD a…3s2u u…v† 1 up3 …v†† …38†
J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328 323

Tl n(t)
f(t)

u(t) Σ
ασ 23

Σ Tn

u3 α

Fig. 4. Non-linear model for Morison’s equation after Bendat and Piersol.

where up3 …v† is a triple convolution of u…v† with itself (see different ranges of KC at each frequency. However as can
Refs. [3,39] for full and interesting discussions of this and be seen from Tables 2 and 5 there is no improvement in the
related models). CM can be estimated directly from the predictive accuracy. As a result the authors did not consider
imaginary part of the transfer function in exactly the same it worth applying this non-linear model to cases where
way as described for Method 7 using Eqs. (25) and (26) current was present. The extension to the case with current
above. is explained in Bendat and Pierson [3].
To estimate CD the non-linear term must first be estimated The frequency range over which the averages for CD …v†
and to do this accurately requires a long time series; as noted and CM …v† are computed will affect the predictive accuracy,
by Vugts and Bouquet and Bliek and Klopman. There are as is obvious from Figs. 5 and 6. It is interesting to note that
various possible procedures for obtaining the estimate (see for a given cylinder diameter KC is directly proportional to
Ref. [8]) and the one used here is that described in Bendat wave amplitude and thus the square root of the wave energy
and Pierson [3]. Briefly the velocity cubed time series is spectral density; if the random wave is considered as the
Fourier transformed and the corresponding spectral density sum of sinusoidal waves with one wave for each elemental
Su3 …v† found. Then the following expression is used to esti- frequency range. So the variation of KC with frequency is
mate Tn …v† : directly related to the spectral shape and the same KC values
will occur twice, once at the low frequency end and once at
Sf …v† ˆ a2 uTn …v†u2 {27s4u Su …v† 1 Su3 …v†}; …39† the high frequency end. In principle some weighted average
could be used to reflect this fact and also the distribution of
whence energy in the wave spectrum. However the authors have not
KD …v† ˆ Re‰Tn …v†Š: …40† done this as they consider that generally frequency domain
methods are not really worth pursuing in the context of
The variation of these force coefficients with frequency is estimating the Morison force coefficients. To summarise,
seen to be significant in Figs. 5 and 6 for the large and small the frequency domain methods make more assumptions
cylinder, respectively, in the case where there is no current. than the time domain methods. Most notable is the linear-
It is interesting to note the variation of CD with frequency is isation (or cubic approximation) of the drag term. In addi-
quite different for the two cases reflecting, perhaps, the tion all frequency domain methods assume random linear

5.5
CM
5

4.5
4

3.5
CD and CM

3 CD
2.5

2
1.5
1
0.5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Frequency (Hz)

Fig. 5. Variation in CD and CM with frequency using the non-linear model (Method 10) for Run 1.
324 J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328

3.5
CM
3

2.5
CD and CM
2

1.5

CD
1

0.5

0
0.1 0.15 0.2 0.25 0.3 0.35 0.4
Frequency (Hz)

Fig. 6. Variation in CD and CM with frequency using the non-linear model (Method 10) for Run 4.

wave theory, and it is seen in the time domain methods without current a large number of correspondingly small
when this assumption is relaxed predictive accuracy waves are needed to estimate a mean value of Cf with statis-
improves. Finally the variation of CD and CM with frequency tical accuracy. Importantly, given the scatter in Cf at low KC
is of no apparent practical value; indeed it is a nuisance. It is then there will be at least similar scatter in CD and CM
therefore not surprising that time domain methods appear to irrespective of how the total force is split into the corre-
have greater predictive accuracy. sponding component parts. Interestingly the effect of intro-
ducing a positive current is to reduce not only the scatter in
Cf at low KC values but also its average magnitude as can be
6. Discussion of the results seen from Fig. 8. A negative current reduces the average
magnitude even further and across a wider range of KC as
The principal objective of this paper is to examine the can be seen in Fig. 9. Similar effects are observed for the
efficiency of the various method of analysis rather than to large cylinder.
present an extensive set of experiment data and so only a Dean’s reliability ratio R is shown for the large and small
subset of a larger programme of experiments are considered. cylinders, respectively, in Figs. 10 and 12. The correspond-
The results of the other experiments in the same programme, ing plots of CD and CM (estimated using least squares wave
on smooth cylinders and those with slight roughness are by wave analysis) against KC are shown in Figs. 11 and 13.
presented in Ref. [27]. These clearly show the effect of the reliability ratio upon the
Before examining the efficiency of the various methods scatter in the estimated force coefficients. For example the
for estimating CD and CM from random wave data it is generally higher R values for the smaller cylinder are
interesting to look at some plots of the total force coefficient reflected in greater scatter in the corresponding CM values
Cf (defined in the nomenclature) against KC. The results when compared to those for the larger cylinder. So not only
obtained with no current are shown in Fig. 7 for the small is there scatter from wave to wave in the total non-
cylinder. The inference that can be drawn is that at low KC dimensional force, but the data are often not well conditioned

45
Run4
40
35
30
25
Cf

20
15
10
5
0 5 10 15 20 25 30 35
KC

Fig. 7. Variation of Cf with KC in the random waves for the fixed small pile.
J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328 325

8
Run5
7
6
5

Cf
4
3
2
1
0 10 20 30 40 50
KC

Fig. 8. Variation of Cf with KC in random waves for the mobile small pile, U ˆ 10:5 m=s.

to allocate this total force into component parts. Thus irre- magnitude of the drag term relative to the inertia term
spective of the analysis method employed to predict CD and when a current is present and thus the proportional reduc-
CM the scatter endemic in the data ensures that there will tion in any relative error when estimating it from the total
inevitably be uncertainty in the estimates found. force. Turning to predictive accuracy few clear trends
Tables 2–7 show the mean values of CD and CM obtained occur. The RMSE tends to reduce somewhat in most
using the various analysis methods for the six experiment cases when current is added but the bias (MNE) shows
runs. The standard deviations quoted for the wave by wave no particular trend.
analyses (Methods 1–3) express the wave to wave variabil- 2. In the time domain, the wave by wave analyses give
ity; whereas those for the frequency domain analyses (Meth- similar results as would be expected in light of the simi-
ods 7, 8 and 10) express the variation with frequency across larity among these methods highlighted in Section 4. The
the range from 0.1 to 0.4 Hz. Also shown in these Tables are least squares method (Method 3) comes out slightly
the corresponding mean bias and standard error when these ahead overall (but not consistently so) in terms of predic-
coefficients are used for predicting the second, unanalysed, tive accuracy as might be expected as it makes fewer
part of the measured force time series. There are several assumptions. Given the improvement seen in the whole
points worth noting about these results. record analysis by weighting the least squares it was
decided to do a wave by wave analysis using this
1. Irrespective of the method of analysis both CD and CM are approach, with k ˆ 2; and the results are summarised
reduced by the presence of a current. Interestingly the in Table 8; from which it is seen to be generally superior
larger reduction occurs for the larger pile when the current to all the other methods. The time domain, whole record,
is in the same direction as the waves but the reverse is least squares analysis yields similar results to the least
true for the small pile. It is not immediately clear why squares wave by wave analysis. However the method of
this should be the case but it is worth remembering that moments (Method 6) is seen to be significantly poorer
the range of KC is quite different for the two piles and and for the small fixed pile gave a bias of 2 22% and a
hence so will be the effects of wake re-encounter when standard error of 26.4%; a plausible explanation for
the wave flow reverses in each wave cycle. For the wave which is beyond the authors.
by wave analysis methods it is noticeable that the stan- 3. The frequency domain methods generally give poorer
dard deviation of the CD reduces significantly in all cases predictive accuracy than the time domain methods.
with the addition of current but for the CM values the This can be explained by the additional assumptions impli-
reverse is true. This may be explained by the increased cit in approximating the drag term and the necessarily

8
7 Run6
6
5
Cf

4
3
2
1

0 10 20 30 40 50
KC

Fig. 9. Variation of Cf with KC in random waves for the mobile small pile, U ˆ 20:5 m=s.
326 J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328

1.5
R = 0.74 (average for all data) Run1

R 0.5

0 2 4 6 8 10 12 14 16
KC

Fig. 10. Variation of Dean’s reliability ratio R with KC for the fixed large pile.

arbitrary choice of spectral width over which to average be due to the relative short length of the records of
the frequency dependent force coefficients. It is surpris- around 15 min or 140 waves. However the results
ing that the cross spectral method involving the surface obtained by Bliek and Klopman and Vugts and Bouquet
elevation (Method 8) is marginally better than the one with addition of non-linear terms were not much more
involving the water particle velocity (Method 7) for both encouraging than those obtained here; despite in the
the fixed cylinders (see Tables 2 and 5), as particle kine- former case of having a record length of 189 min.
matics predictions from surface elevation using linear
theory are often poor. When a current is present the latter The results presented here are for very rough cylinders.
approach is seen to give very variable results (see parti- The question arises would similar results for predictive
cularly Table 7) implying a lack of numerical robustness. accuracy have been obtained with smooth cylinders. In the
The frequency domain least-squares approach suffers in a authors’ view the ranking of the analysis methods would not
similar, but more extreme manner (see Table 7 again). In have been altered because this is largely dictated by the
this case the high frequency cut-off was found to have a assumptions and approximations implicit in each of the
significant effect upon the denominators in Eqs. (34) and methods as discussed above. The predictive accuracy
(35). would probably be similar, but not the same. Even at the
4. The introduction of a non-linear term into the frequency largest scale so far employed for model experiments on
domain analysis makes little difference for the large pile smooth circular cylinders in random waves there is doubt
case as can be seen in Table 2; whereas for the small pile concerning the flow regime. Particularly in the smaller
the result is a very distinct decrease in predictive accu- waves there is the possibility that the flow is not fully turbu-
racy both in terms of RMSE and bias. This may, in part, lent throughout the wave cycle. This is likely to lead to

6
1< KC< 18, mean(KC)= 5.5, mean(CD)= 1.88, Run1
4
CD

0 2 4 6 8 10 12 14 16
KC

3 1< KC< 18, mean(KC)= 5.5, mean(CM)= 2.08, Run1

2
CM

0
0 2 4 6 8 10 12 14 16
KC

Fig. 11. Variation of CD and CM obtained using wave by wave least square method (Method 3), with KC in random waves for the fixed large pile.
J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328 327

3.5
R= 1.72 (average for all data) Run4
3

2.5

2
R
1.5

0.5

0
0 5 10 15 20 25 30 35
KC

Fig. 12. Variation of Dean’s reliability ratio R with KC for fixed small pile.

greater uncertainty in the prediction of peak forces, particu- 7. Conclusions


larly when all the data are used in the analysis.
One of the problem facing someone undertaking a It is clear that the method used to analyse experimental
risk assessment or a structural reliability calculation data in terms of Morison’s equation has a significant effect
for an offshore jacket structure is to estimate the on both the force coefficients obtained and their predictive
uncertainty in the wave/current loading predicted by accuracy. It is found that no single method is consistently
Morison’s equation. The values of MNE and RMSE better under all circumstances but on average the wave by
quoted in Table 8 probably represent the lower wave weighted least squares method gives both the lowest
bounds of the uncertainty that it is possible to achieve bias (2%) and root mean square error (11%) as can be seen
currently. One problem with laboratory experiments is in Table 8. The other time domain analysis methods whilst a
that in order to achieve acceptably high Reynolds little poorer worked reasonably satisfactorily. The
numbers the range of KC is lower than that which frequency domain techniques were found to be less satisfac-
occurs in practice so there is inevitably some extrapolation. tory and, on occasions, were not robust and so are not
However this error is thought to be small (particular for the recommended unless there are particular reasons for their
small cylinder data considered here) as at large KC drag adoptions.
coefficients generally approach their steady flow values The force coefficients obtained by the various methods
asymptotically. varied significantly but there was a clear trend which
Space does not permit presentation and discussion of all showed that the addition of current significantly decreased
the results obtained from the analyses undertaken and the the drag coefficient and to a lesser extent the inertia coeffi-
reader is referred to Ref. [30] for further details. cient. For KC values above around 10 use of mean values
6
1< KC< 38, mean(CD)= 1.85, mean(KC)= 12.8 Run4
4
CD

0
5 10 15 20 25 30 35
KC

3 1< KC< 38, mean(CM)= 1.92, mean(KC)= 12.8 Run4

2
CM

0
5 10 15 20 25 30 35
KC

Fig. 13. Variation of CD and CM, obtained using wave by wave least square method (Method 3), with KC in the random waves for a fixed small pile.
328 J. Wolfram, M. Naghipour / Applied Ocean Research 21 (1999) 311–328

for drag and inertia coefficients (about 1.7 and 2, respec- of the First International Conference on Behaviour of Offshore Struc-
tively) for heavily marine roughened cylinders in waves tures, Trondheim, Norway, 1976:40–64.
[20] Evans DJ. Analysis of wave force data, Offshore Technology Confer-
without current seems reasonably satisfactory. When ence, paper number OTC 1005, 1969.
current is present both coefficients should be significantly [21] Hasselmann K, et al. Measurements of wind-wave growth and swell
less. decay during the Joint North Sea Wave Project (JONSWAP).
Deutschen Hydrographischen Zeitschrift, A 1973;27–95.
[22] Heideman JC, Olsen OA, Johansson PI. Local wave force coefficients.
Proceedings of the Speciality Conference, Civil Engineering in the
References Oceans IV, 1979;2:684–99.
[23] Keulegan GH, Carpenter LH. Forces on cylinders and plates in an
[1] Bearman PW. Wave loading experiments on circular cylinders at oscillating fluid. Journal of Research of the National Bureau of Stan-
large scale. Proceedings of the Fifth International Conference on dards 1958;60(5):423–40.
Behaviour of Offshore structures, Trondhein, Norway, Tapir, 1988: [24] Klopman G, Kostense JK. The loading on a vertical cylinder in
471–87. random waves at high Reynolds numbers, Water wave kinematics,
[2] Bearman PW, Chaplin JR, Graham JMR, Kostense JK, Hall PF, Klop- Proceedings of NATO Advanced Workshop on Water Wave Kine-
man G. The loading on a cylinder in post-critical flow beneath peri- matics, Molde, Norway, (NATO ASI Series E: Applied Sciences),
odic and random waves. Behaviour of Offshore Structures 1985:213– 178. The Netherlands: Kluwer Academic, 1989. p. 679–700.
25. [25] Li YC, Kang HG. Wave current forces on slender cylinders. Proceed-
[3] Bendat JS, Piersol AG. Decomposition of wave forces into linear and ings of the 11th International Conference of Offshore Mechanics and
non-linear components. Journal of Sound and Vibration 1986; Arctic Engineering (OMAE) 1992;1:117–25.
106(3):391–408. [26] Mackwood PR. Wave and current flows around circular cylinders at
[4] Bendat JS, Piersol AG. Random data, analysis and measurement large scale, LIP Project 10D, 1993:27.
procedures, 2. New York: Wiley, 1986. [27] Mackwood PR, Bearman PW, Graham JMR. Wave loading data from
[5] Bishop JR. Wave force investigation at the second Christchurch bay fixed vertical cylinders with simulated hard marine fouling, HSE
tower, Summary report, NMI R177, OT-O-82100, 1984. Report OTH94 438, ISBN 0-7176-0999-5, 1995.
[6] Bishop JR. The mean square value of wave force based on the Mori- [28] Morison JR, O’Brien MP, Johnson JW, Schaaf S. The force exerted
son’s equation, NMI R 40, also OT-R-7811, 1978. by surface wave on piles. Transactions of the American Institute of
[7] Bishop JR, Shipway JC. Wave force coefficients from the second Mining and Metallurgical Engineers 1950;189:147–54.
Christchurch Bay Tower, NMI R 178, also OT-O-82101, Part 1, 1984. [29] Muga BJ, Wilson JF. Dynamic Analysis of Ocean Structures
[8] Bliek A, Klopman G. Non-linear frequency modelling of wave forces 1970:134–50.
on large vertical and horizontal cylinders in random waves, BOSS’88, [30] Naghipour M. The accuracy of hydrodynamic force prediction for
1988: 821–40. offshore structures and Morison’s equation, PhD thesis, Heriot-Watt
[9] Borgman LE. Spectral analysis of ocean wave forces on piling. Jour- University, Edinburgh, 1996.
nal of the Waterways and Harbours Division, Proceedings of the [31] Newland DE. An introduction to random vibrations and spectral
ASCE 1967;93(WW2):129–56. analysis, 3. New York: Longman, 1993.
[10] Borgman LE. Ocean wave simulation for engineering design. Journal [32] Pierson WJ, Holmes P. Irregular wave forces on a pile. Journal of the
of the Waterways and Harbours Division, Proceedings of the ASCE Waterways and Harbours Division, Proceedings of the American
1969;95(WW4):557–83. Society of Civil Engineers (ASCE) 1965;91(WW4):1–10.
[11] Borgman LE. Statistical models for ocean waves and wave forces. [33] Rodenbusch G, Gutierrez CA. Forces on cylinders in two-dimensional
Advances in Hydroscience 1972;8:139–81. flows, Technical progress report, BRC 13-83, Shell Development Co,
[12] Chakrabarti SK. Technical note: on the formulation of Jonswap spec- 1983.
trum. Applied Ocean Research 1984;6(3):175–6. [34] Standsby PK, Worden K, Tomlinson GR, Billings SA. Improved
[13] Chakrabarti SK. Hydrodynamics of offshore structures, New York: wave force classification using system identification. Applied Ocean
Computational Mechanics Publications, 1987. Research 1992;14:107–18.
[14] Chakrabarti SK, Johnson JG. Random wave–current interaction— [35] Tung CC, Huang NE. Interactions between waves and currents and
theory and experiment. Proceedings of the 14th International Confer- their influence on fluid forces, BOSS’76, 1976:129–43.
ence on Offshore Mechanics and Arctic Engineering (OMAE), [36] Vugts JH, Bouquet AG. A non-linear, frequency domain description
vol. I/A, Offshore technology, 1995:25–36. of wave forces on an element of a vertical pile in random seas, Beha-
[15] Chaplin JR. Non-linear forces on horizontal cylinders in the inertia viour of offshore structures. Amsterdam: Elsevier Science, 1985.
regime in waves at high Reynolds numbers. Proceedings of the Fifth p. 239–53.
International Conference on Behaviour of Offshore structures, 1988: [37] Wolfram J. A novel underwater hydrodynamic experiment facility
505–18. and first results with marine growth covered cylinders, Part B, Trans-
[16] Chaplin JR, Subbiah K. Velocity measurements in multi-directional actions of Royal Institution of Naval Architects, London, 1992.
waves using a perforated-ball velocity meter. Applied Ocean [38] Wolfram J. Marine fouling and fluid loading, Final Report on Project
Research 1994;16:223–34. FLE6, MTD Programme Fluid Loading in Offshore Engineering,
[18] Davies MJS. Wave loading data from fixed vertical cylinders, OTI 92 1987.
558, 1993:89. [39] Worden K, Stansby PK, Tomlinson GR, Billings SA. Identification of
[19] Dean RG. Methodology for evaluating suitability of wave and wave non-linear wave forces. Journal of Fluids and Structures 1994;8:19–
force data for determining drag and inertia coefficients, Proceedings 71.

You might also like