On the time domain simulation of second order

wave forces and induced responses

R. S. L A N G L E Y

Offshore Strttetttres Group, College o f Aeronautics, Cranfield Institute o f Technology, Cranfield,

Bedford MK43 OAL, UK

This paper is concerned with the time domain simulation of the second order motions of a moored
vessel when the random seastate is represented as a sum of harmonic components. It is known that in
these circumstances successive runs of a simulation program produce different results for the statis-
tical moments of the response. Here, the variation of the first four statistical moments of the response
over an ensemble of program runs is investigated, leading to an assessment of the likely accuracy of
these quantities as predicted by a limited number of program runs. Also, it is shown that an approxi-
mate simulation method which uses deterministic wave amplitudes and random phase angles does not
correctly predict the fourth moment of the response.

INTRODUCTION minimum duration of 40 times the surge natural period

One of the major considerations in the design of an off. of the vessel to obtain r.m.s, responses which are indepen-
shore mooring system is the large amplitude slow drift dent of the phasing of the waves. This corresponds to a
motions which may be exhibited by the moored vessel. duration of around 4 h full scale or 30 min model scale.
These motions are a resonant response to low frequency The aim of the present paper is to examine the variability
second order wave forces, which have been the subject of in computed results for the slow drift response of a moored
much research effort in recent years, a The problem facing vessel when the random seastate is modelled as a super-
the designer is to estimate the largest slow drift response position of regular wave components. A technique is devel-
which is likely to be exhibited by the vessel, in order to oped whereby the first four statistical moments of the slow
ensure that the mooring system will not be overstressed. drift response of a linear system, as given by time domain
This can be done either by performing model tests, which simulation, can be efficiently calculated. Using this tech.
can be cumbersome if a parametric study is required at the nique the variation of the statistical moments over an
initial design stage, or by employing a theoretical analysis ensemble of program runs is examined. Theoretical expres-
with the aid of a digital computer. The theoretical analysis sions for the likely accuracy of the statistical moments as
can be either a frequency domain analysis which yields predicted by a limited number of program runs are also
spectra and r.m.s, values of the response, or a time domain developed and compared with simulation results.
analysis which yields a time history of the response. In the
study of slow drift responses the more costly time domain REPRESENTATION OF THE RANDOM SEASTATE
analysis is usually preferred, as this can more readily deal
with the nonlinearities which are inherent in the problem, The random surface elevation of the sea, 7/(t), is often
two examples being the nonlinear restoring force properties represented as a sum of regular wave components in the
of the moorings and the non-Gaussian nature of the excita- form
tion force. Such a time domain computer program must N
firstly generate a time history of the random second order ~/(t)--- • c n cos(cont+en) (1)
wave forces and then compute the motions of the vessel n=l
using a time step integration procedure. Generally, the where Cn, Wn and en are the amplitude, frequency and phase
second order forces are sinmlated by considering the of the nth component. Strictly, equation (1) is based on a
random seastate to "be a superposition of regular wave Fourier series expansion of r/(t) over a limited time period T:
components, each having a random phase angle. 2-s One
problem with this approach is that the slow drift forces, N
being a second order effect, are very sensitive to wave n(t) = ~, [an cos COnt+ bn sin COnt] (2)
grouping 6'7 and subsequent runs of the computer program n=l
can produce different r.m.s, values of the slow drift force where a n and bn are Fourier coefficients and con = (21rn/T).
and response. This effect has also been noted in model Such an expansion was used by Rice t~ to represent a
tests 22 where the 'shift register technique '8 is generally general Gaussian process, and as pointed out by Middleton H
used to generate the random seastate. Standing et al. 22 the expansion becomes orthogonal as T ~ co. Under these
have estimated that a wave tank test needs to have a circumstances a n and b n are independent Gaussian random
variables in the ensemble sense having zero mean and
E[a2n] = E[b2n] = Gnn(con) dco. Here Gnn(co) is the one-sided
Accepted October 1985. Discussion closes September 1986. wave spectrum, dco = (2rr]T) and E [ ] represents the en-

0141-I 187/86/030134-10 $2.00

9134 Applied Ocean Research, 1986, Vol. 8, No. 3 9 1986 Computational Mechanics Publications
 On the time domahl simulation of second order wave forces and induced responses: R. S. Langley

semble average. Thus to simulate a time history of r~(t) of The double summation appearing in equation (7) can be
duration T, the coefficients an and bn Ol = 1,2 . . . . . N ) are divided into three regions, namely n = m , n > m and
chosen at random from a Gaussian distribution having zero n < m . Letting k = n --m, this gives:
mean and variance Grm(Oan)doa, and the time history is N N--I N--k
calculated from equation (2). By equating equations (1) F = ~ lanl2Hnn + ~ ~ Hm+k.mam+ka* e t~
and (2) it is found that c n cos en = an and cn sin en = --bn. n=l k=l m=l
By transforming from the Gaussian variables a n and b n to
N--I N--k
the variables Cn and en it can be shown that c n has a Ray-
+ Z • Hm,m+k~ma*m+k e -iwkt (9)
leigh distribution with E [C2n]= 2Gnn(Oan)doaand en has a k=l m=l
uniform distribution over a range of 2/r. ~2 Often equation
(1) is used to simulate ~(.t) with the approximation that where, as in all the following equations, real parts are
c n is given the deterministic value Cn= [2Gnn(Oan)dw] in assumed. Noting from equations (6) and (8) that Hn*m=
rather than being chosen at random from a Rayleigh distri- Hmn , taking the complex conjugate of the last term in
bution. The Gaussian nature of r/(t) then relies upon the equation (9) gives:
central limit theorem rather than the statistical properties N N--1 N--k
of c n and en. The effects of this approximation have been F = Z lanl'Ynn + 2 Z Z Hm+x, ma*arn+X eiwxt
discussed by Tucker et al., ~2 the most interesting being that n=l k=l m=X (10)
the approximate simulation always has the spectrum
Gnn(oa), whereas a particular time history generated using The first term in the above equation represents a mean
the more exact method has a spectrum with a random drift force whereas the second term is slowly varying.
variation about Gnn(oa). The effect of this approximation Writing these terms as F s and Fa gives:
on the predicted slow drift response of a moored vessel is
N
discussed in following sections of the present paper.
It is convenient for the analysis of the following sections Fs = Z lanl2Hnn (lla)
n=l
to rewrite equation (2) in the form:
N--1
N Fa = ~. X k e it~ (llb):
r/(t) = Re ~ an eiWnt; an = an-- ibn (3) k=l
n=l
where Xk is given by:
From the properties ofa n and bn it follows that:
N--k
E[a.am] = 0; E[a.a*m] = o m ~ n; Xtc = 2 Z "*
Hm+k, mamam+ k (12)
(4)
E [la,,121 = 2Gnn(~n) d ~ = A n m=l

where a * denotes the complex conjugate of a m, and the Now, from equation ( l l b ) the spectrum of the slowly
abbreviation A n = 2Gnn(COn)d~ has been introduced. Note varying drift force is:
that equation (4) is valid whether the more exact simula- IXkl ~ 2 N--/r N--k
tion technique or the approximation involving a deter- Y
ministic value of Cn is used. The approximation does effect GFF(O)k)= 2dw =1 n=l

the higher statistical moments of an, h o w e v e r - f o r example $ AS A A AS

XHm+k, mHn+k, namam+kanan+ k (13)
the exact result f o r E[lan[ 4] is 2A2n whereas the approxi-
mate technique would give A2n. Since equations (1 la) and (13) involve the terms an which
are random in the ensemble sense it follows that in general
the mean drift force and the drift force spectrum will also
REPRESENTATION OF THE SECOND ORDER FORCE be random in the ensemble sense. When the approximate
In the literature (for example, refs. i, 4, 13) the low fre- method of simulating the surface elevation r/(t) is used,
quency second order force acting on a moored vessel in a which involves using deterministic values of Cn, it can be
random sea is usually written as: shown that although the drift force spectrum is random
the mean drift force takes a deterministic value due to the
N N
fact that lanl2=cn. Using equation (4), the ensemble
F= Z ~ enem{T~m c~ +e,,--e,,d averages of the mean force and the drift force spectrum in
n=l m=l
the general case are given by:
+ rf~n sin [(Oan-- ~m) t + en-- em] } (5)
N
where the random surface elevation is represented by equa- E[Fs] = 2 Z H . . a n n ( r 1 7 6 dco (14)
tion (1) and Tern and TSnm are second order transfer rl=l
functions which satisfy: N--k
TCnm= TCmn; TSm =--TSmn (6) E[GFF(WD]=8 ~,, IHm+k,,,,12G,m(~Om)
m=l
Using the notation of equation (3), equation (5) can be
X Gnn(Wm+k) dw (15)
rewritten as:
Equations (14) and (15) are in fact discrete forms of the
N N
analytical results which can be derived for the mean drift
F = Re Z Z a.a*H,,,, exp [i(w.--a~m)tI (7)
force and the drift force spectrum: 14
n=l m=l
where the complex second order transfer function Hnm is
given by: Fs = 2 f H(oa, oa) Gnn(oa) dw (16)
Hnm = renm -- iTem (8) 0

Applied Ocean Research, 1986, Vol. 8, No. 3 135

 On the thne domah) simulation of second order wave forces and induced responses: R. S. Langley

oQ oo

GFF(~g) = 8.1 In(w + ~k, ~)l 2 G~n(6~ Grp~(r + cox) dr zs = K f H(r ~o) Gnn(r ) dr
(24)
o
o 07)
Gzz(~Ok) = IT(~lc)I2GFIz(~k)
wher e H(~ol, 6o2)is the continuous form of the second order
transfer function. The accuracy of equations (14) and (15) where GFy(COk)is given by equation (17).
will depend upon the number of terms considered, N, and To summarise the results of both this section and the
the frequency interval dco, which is in turn related to the previous one, the second order force and response produced
simulation time T via dr (2rr]T). In general, the larger by the random seastate represented by equation (3) are
the simulation time the more accurate the equations given by equations (11) and (19) respectively, where the
will be. terms Xk and T k are shown in equations (12) and (20).
The spectra of the force and response for any particular
realisation are given by equations (13) and (21) and the
REPRESENTATION OF THE SLOW DRIFT RESPONSE ensemble averages of these spectra by equations (15) and
This section considers the response of a linear single degree (23). To produce a time history of either the force or the
of freedom system to the forces discussed in the previous response, the summations involved in equations (11) and
section. This may represent the slow drift surge motion of (19) can be performed very efficiently using a Fast Fourier
a Single Anchor Leg Storage (SALS) system is or a Tension Transform algorithm 17 once the terms Xtc and Tk have
Leg Platform (TLP), ~6 where the moorings consist of taut been evaluated. This offers an alternative efficient sinmla-
tethers and are predominantly linear. The equation of tion scheme to the one suggested by Marthinsen a which
motion is written as: is based on a heuristic approximation to the slow drift
force. In simulating the force, only those frequency com-
M'5 + B~ + Kz = F (18) ponents which are near to the natural frequency of the
where M, B and K are the mass, damping and stiffness of vessel need to be included, thus giving an upper bound on
the system and z is the surge displacement. The steady state the number of terms used in equation (1 lb).
response to the drift force F = F ~ + F a can be written as
a mean offset z s plus a dynamic response z a. Using equa- MOMENTS OF THE SLOWLY VARYING FORCE AND
tions (11) and (18) gives: RESPONSE
1 N The slowly varying second order force and response can
z s =-~ ~ lanl2Hnn (19a) both be represented in the following way:

N--I Rd = Re {Y} (25a)

Zcl = ~ Z k S k e itOkt (19b)
/V--1
k=l
Y = ~ Yk eiWkt (25b)
where Tk is tlae frequency response function associated k=l
with equation (18):
Here Ra stands for either the force or the response and
1 2 2 from equations (11) and (19) YIc = X e when the force is
:irk = ~ [(r --cog) + 2iOWzWt:]-' (20) considered and Yk = TkXk when the response is con-
sidered. The present section considers the second, third
Here r z is the natural frequency of the system, r 2 =K/M, and fourth statistical moments of the quantity R d. From
and/3 is the damping ratio, [3= B](2Mwzl. The spectrum of the theory of complex numbers:
the slowly varying drift response can be found from equa-
n~r = 89 {Y2 + YY*) (26)
tion (19b) to be:
R~t = ~Re ( y a + 3 Y ' Y * } (27)
[TkXkl 2 2 N--k N--k
azz(Ogk) = - -2=dw
- ~ W m ~ = l = n=l
~. ITkl2Hm+k,m R~t = IRe {y4 + 4 y a y , + 3y2 r , 2 } (28)
Using equation (25b), the temporal averages of R~, R,]
"* .am. +kanan+k
. . * 4
*
• Hn+k,n am (21) and Rd can now be written as:

As was the case for the forces, the mean offset and the 1 N--I
response spectrum will be random in the ensemble sense. (R2a)= 2 e~=1 IYal2 (29 /
Using equation (4) the ensemble averages of these quan-
tities are: 3 N--I N--I
(R~0=-Re E E YkYtY~+t (30)
N 1 4 k=l i=l
E [zs] = 2 ~ :HrmGnn(Wn) dw (22)
n=l K 1 N--1 N--1 N~I

2V--k
(R~r) = - R e Z Z ~ Y~YtYmY~+t+rn
2 k=l /=1 m=l
E[Czz(,~k)] = 8 y . IZkl' IHm+k,ml2
In =1 3 N--I N--1 N--I
+-ReE E E YkYtYZ,* YL~-.,
* (31)
x Gnn(Wm) G)~(O)m+kl dt.o (23) 8 k=l I=1 m=l

Again, these are discrete forms of analytical results which The procedure for calculating the temporal averages is then
can be derived for the mean offset and response spectrum: as follows. Firstly, the terms a n appearing in equation (3)

136 Applied Ocean Research, 1986, Vol. 8, No. 3

 On the time don~ain simulation of second order wave forces and induced responses: R. S. Langley

are chosen in the random manner which has been discussed 100 f o \
previously and inserted into equation (12) to give Xk.
Secondly, given Xk (and Tk) the required moments can be
calculated from equations (29) to (31). The moments Equation (37) can be used to estimate the likely accuracy
calculated in this way are equivalent to those which would of a moment of the slow drift force or response which has
be obtained from a single 'run' of a time domain program. been calculated from M program runs. As would be ex-
Repeating the procedure with a different choice of d n will pected, the error tends to zero as the number of program
in general lead to different moments, and by following the runs becomes large. In other cases, a knowledge of a and bt
procedure a large number of times the way in which the is required if the accuracy of S is to be assessed, for which
moments vary over an ensemble of program runs can be it is necessary to return to equations (29)-(31). When the
investigated. Bowers 7 has shown using the law of large second moment of the force or response is considered,
numbers that the result obtained from equation (29) for equation (29) implies that:
the mean squared response tends to the exact analytical
I N-~ i N-I
result as the number of termsNconsidered tends to infinity.
P=2 E E[IYk121=2 E fk (38)
The accuracy of equations (29)-(31) for a finite value of N k=l k=l
is discussed in the following sections. Note that the com-
putation time involved in evaluating equations (30) and
o2+p 2= E E e[IYkl2lYtl2l= E rkt
(31) can be reduced by making use of the symmetries of k=l l=l = l=l
the various terms, as discussed in Appendix A.
(39)
THE ACCURACY OF A LIMITED NUMBER OF where expressions for fk and rkt are derived in Appendix
PROGRAM RUNS B by using equations (12) and (4). In theory similar ex-
pressions can be derived when the higher moments are
Consider the case where the moments given by equations considered, although the algebra involved is extremely
(29)-(31) are calculated for M independent program runs. lengthy. For this reason, estimates of (o/p) for the higher
Let the values of one of these moments as given by the M moments are obtained here from numerical simulations.
runs be xt, i = 1, 2 . . . . . M. The final estimate for this
moment, S say, will be:
NUMERICAL RESULTS AND DISCUSSION
i M
S = - - Z xi (32) This section presents numerical results for the slow drift
M i=1 forces and motions of a system governed by equation (18).
The equation is cast into the form:
I f p is the exact ensemble average of this moment, such that
p =E [xt], then the error in the estimate S will be: + 2[3Wz~ + co2zz = G(t) (40)
e = S -- p (33) where COz is the natural frequency,/3 is the damping ratio
and G = F/M. It is assumed that the slowly varying part of
Consider now the statistics of the error e. This error will in
the force G(t) can be written in the form of equation
general be different if a different group of M program runs
(1 lb), meaning that the response is given by equation (19b)
is used and thus e is a random variable when viewed over a
with M = 1.0. A value of coz = 0.02 is used here, which is
large number of such groups. The average value of e is
representative of the surge natural frequency of a moored
given by:
offshore structure. For simplicity a constant value of the
1 M function H m +k, m appearing in equation (12) is used, which
e [e] = ~ ~,, E [xi] -- U = 0 (34) means that the slow drift force is directly proportional to
jv~' /=1 the square of the wave envelopeJ s A value of 10 -4 is
and the mean squared value of e is: chosen, as this gives slow drift responses which are repre-
sentative of those exhibited by moored vessels. The incident
1 seastate is taken to have a JONSWAP 19 wave spectrum, with
e[d] Z ZE[x xjl E[xd+ a significant wave height of 15 m, which is often considered
i=x j=x M i---'1
to be the design spectrum for North Sea conditions:
M 2 M M Ca) - 4
1 Z E[x~] +--~,2 ~, Z E[xixl]
M2i=1 J~l i=l / = i + l
x 3' exp [--(~--O)p)Z/202092p] (41)
2p M "
Z E [xil + p2 (35) where a = 0.0081,13 = 1.25, wp = 0.3656 and o = 0.07 for
Ml=l w~<wp, a = 0.09 for r wp. For most of the results
Noting that the terms x t are statistically independent, and presented here the frequency interval used is dco = 0.0004,
defining o such that E[x]] = o 2 +p2, equation (35) leads to which gives a simulation time of 4.36h, corresponding to
the final result: 50 cycles of free vibration of the system. This gives 1750
frequency components in equation (3), taken between 0.2
1 2 2 1 2 2 /u2 = _o_2 and 0.9 rad/s. In order to simulate the second order forces
E[e2I=M(O +p ) + M ( M - - 1 ) p - - 2 # + M and motions it is necessary to generate the random variables
an appearing in equation (3), which is done using the NAG
(36) scientific subroutines 2~ GO5CAF and GO5DDF. The
The r.m.s, percentage error of the estimate S, ~ say, can former is used to generate the phase angles en from a
now be written as: uniform distribution when the approximate simulation

Applied Ocean Research, 1986, Vol. 8, No. 3 137

 On the time don~aflz s#nulation of second order wave forces and induced responses: R. S. Langley

61 frequency components o f the force which are likely to

induce a resonant response were included, meaning that
100 terms were used giving an upper frequency of 0.04
rad/s = 2Wz. Using a Fast Fourier transform algorithm 17
#C to perform the summation appearing in equation (1 lb) at
t'o
o
a number of discrete times, it was found that 4.36 h o f
the slow drift force could be simulated using 80 s CPU
,a-
v time on a VAX 750 digital computer. The resonant nature
LIJ
o f the slow drift response is demonstrated by these two
ur 20 figures, as a scaled force of around 6 x 10 -a N/kg produces
tD
U-- a response in the order of 30 m. For a moored VLCC with
o M = 2.5 • 106 kg this force is in the region o f 150 tonnes,
u.J
_..I which is slight when compared to the mass o f the vessel.
~ 00 Figure 3 shows the spectrum of the slow drift force. The
t.o
irregular curve in this figure is the spectrum as given by
equation (13) for one program run while the smooth curve
is the ensemble average, given by equation (15). It can be
seen that there is a significant variation about the ensemble
-2-0 average result. Figures 4 and 5 show the corresponding
response spectra for/3 = 0.1 and 0.05 respectively, where
again it can be seen that there is a great deal of scatter
about the ensemble average curves. Bowers 7 has shown
-t~.O-- using the law of large numbers that the area of the two
0.0 250 50"0 spectra should tend to the same value as the number of
TIME (MINS.) frequency components considered tends to infinity. From
Figs. 3-5 it can be seen that for the finite number of
Figure 1. Sample time history o f the slow drift force
frequency components considered here the areas differ
considerably, a fact which is further investigated by the
numerical simulations discussed below.
T 1 1
Using equations (29)-(31) it is possible to calculate the
30 moments of the slow drift force or response as given by a
particular program run. In the present investigation
program runs have been performed in groups of 50. For
each program run the moments of the slowly varying force
are calculated, together with the moments of the response
for /3 = 0.01 and 13 = 0.05. The mean values and the
variances o f the moments over the 50 runs are then calcu-

LU
I T T
Z
o
(21-
0
tlJ
r

-10
• 30

-20
9 (o)

2OJ-
t .J

;11
O0 2 5-0 500
TIME (HINS.)
Figure 2. Sample time history of the slow drift response
for13= 0.05

method is used, whereas the latter is used to generate an

and bn from a Gaussian distribution when the more exact
method is used (see section on Representation o f the
Random Seastate). These subroutines are themselves widely
used and are thought to be representative of other routines
"~t
00
0-0 002
FREQUENCY (RAD/S)
0-04
which are available.
Figures 1 and 2 show typical time histories o f the slow Figure 3. Spectra of the slow drift force. (a) Spectrum
drift force and response when 13= 0.05, as calculated from front one program nm, area = 215. 7; (b ) ensemble average,
equations ( l l b ) and (19b). In equation ( l i b ) only those area = 19Z3

138 Applied Ocean Research, 1986, Vol. 8, No. 3

 On the time domahz simulation o f second order wave forces and htduced responses: R. S. Langley

30 obtained from equation (39) are in reasonable agreement

with those obtained from the simulations. Considering the
case /3 = 0.1 when the more exact simulation method is
o used, inserting the theoretical results for the mean value
x
A and the variance of the second moment into equation (37)
yields the result that in 50 runs the r.m.s, error in the esti-
Z
mated second moment is 3.36%. The actual percentage
. z0 (o) errors in the two groups o f 50 simulations which were
I'-- performed were 1.08% and 2.15% which are o f a similar
LJ
LU
Q- b) magnitude. A similar level o f agreement is found for the
t.O other cases presented in Table 1, suggesting that equation
LLI
(37) gives a useful method of estimating the number o f
C) program runs needed to provide a sufficiently accurate
O...
(/1 result. Figures 6-8 show the results of 50 program runs
LLI
for the moments o f the response when/3 = 0.05, together
10 with the cumulative average o f the moments. The scatter
represented by the variances of the moments given in
Table 1 is highly noticeable in these figures, where it can

1"00

0.0 002 O-Or,

(RAD/S)
FREQUENEY %
Figzlre 4. Spectra o f the slow drift response f o r [J = O.1. 0"75
(a) Spectntm f r o m one program ntn, area = 71.49; (b) r

ensemble average, area = 59.32

~
lated. In all, four groups o f 50 program runs have been L)
UJ
(a)
performed, two using the approximate simulation tech-
(3.
U9
0"50
nique involving a deterministic wave amplitude and two IdJ (b)
using the more exact simulation technique in which the
n
terms an and bn appearing in equation (3) are chosen from 12%

a Gaussian distribution. The results o f these simulations are r

shown in Table 1. Also shown in this table are theoretical

values for the ensemble average of the second moment,
given by equation (38), and the variance of the second
o-2s 1
moment, given by equatiori (39). A general trend in this
table is that the variances of the moments of the response
are greater than those o f the force, and further the vari- 000 " - ..A ^..-~.,=,-u v ,

ances for /~= 0.05 are greater than those for /3= 0.i. 00 002 OOl,
Referring to Figs. 3-5 this might be expected, as the FREQUENCY ( R A D I S )
variances are likely to increase as the number of frequency Figure 5. Spectra o f tile slow drift response f o r [3= O.05.
components having a significant contribution to the (a) Spectrum f r o m one program ntn, area = 160.8; (b) en-
moments decrease. The variances of the second moments semble average, area = 120.6

Table 1. Simulation results/or the slowly varybzg force and response, m n is the nth statistical moment, (a/12)n is the ratio appearing in
equation (37) for the nth moment. $1 represents the exact simulation technique, $2 the approximate technique
Simulation results - two sets of 50 runs Theory

m2 (a/l~)2 m, Ca/#), "14 X 1 0 -4 (o.Lu)4 m, (a/tJ)~

S1 : Force 196.6 0.159 21.74 0.431 17.58 0.406 200.4 0.182
205.4 0.174 27.90 0.548 22.96 0.656
SI: Response/~ = 0.1 58.84 0.239 29.70 1.307 1.352 0.696 59.48 0.238
60.76 0.216 30.25 1.329 1.447 0.566
SI: Response # =0.05 127.1 0.319 --9.224 -5.950 5.548 0.969 120.8 0.296
123.6 0.252 --21.88 --3.240 5.104 0.559
$2: Force 197.3 0.139 23.51 0.503 18.88 0.565 200.4 0.137
195.6 0.118 22.44 0.420 17.63 0.466
$2: Response # = 0.i 59.32 0.217 28.61 1.600 1.321 0.545 59.48 0.204
59.12 0.177 33.57 --1.167 1.249 0.409
$2: Response # = 0.05 120.6 0.274 --8.752 --7.740 4.918 0.618 120.8 0.269
122.2 0.221 --7.832 - 4.808 0.491

Applied Ocean Research, 1986, Vol. 8, No. 3 139

On the time domahz shnulation o f second order wave forces and htduced responses: R. S. Langley

//
process this ratio is 3.0. For a constant value of the second
order transfer function Hm+x,m, as has been considered
here, the slow drift force is proportional to the square of
the wave envelope and as such has an exponential proba-
bility density function. In this case the ratio m4[m~ should
be 9.0. The reason why this value has not been obtained
here is that only those frequency components of the slow
drift force up to 0.04 rad/s have been considered, meaning
that the resulting force is not exactly the square of the
wave envelope. Turning to the response, the ratio m4/m~is
t--, nearer to 3.0 than that of the force, meaning that the
812c response is closer to having a Gaussian distribution. This
agrees with the fact that the response of a linear system
becomes Gaussian as the damping tends to zero, regardless
of the statistics of the excitation force, which can be
proved qualitatively by applying the central limit theorem
to the Duhammel integral representation of the response.
gC This explains why the ratio m4/m~ is nearer to 3.0 for the
lower value of damping, 18= 0.05, than for 18= 0.1. Table 2
shows that the approximate simulation method consistently
predicts a lower value of the ratio m4/m~ than the more
exact method. This is because the approximate simulation
60,1 I I I I
method does not give the correct value of terms of the form
1 25 50 E [ l a m l 4] which appear in the fourth moment, and thus
NUMBER OF RUNS does not correctly predict m4. This suggests that the
Figure 6. Second moment o f the slow drift response for
18= 0.05. (a) Results front indivMual ntns; (b) cumulative
average 15 I I I

I I

Ln
'o

Z
w

0
0
i'~ I / (a)

r~ -I00 os
t~
"I-

O0 I ! I
-200
25 50
NUMBER OF RUNS
I - I I Figure 8. Fourth moment of the slow drift response for
25 50 18= 0.05. (a) Results from individual runs; (b) cumulative
NUMBER OF RUNS average
Figure 7. Third moment o f the slow drift response for
18= 0.05. (a) Results from flldividual ntns; (b} cumulative Table 2. The ratio m4[m] as given by the exact and approximate
average simulation methods
Exact Approximate
be seen that at least 25 runs are required before the cumu- simulation simulation
lative averages start to level off.
Force 5.00 4.73
Table 2 shows the ratio m4/m~ for the slow drift force Response # = 0.1 3.89 3.66
and response, where m n is the nth moment given in Table 1 Response ~ = 0.05 3.50 3.29
(averaged o v e r the two sets of 50 runs). For a Gaussian

140 Applied Ocean Research, 1986, Vol. 8, No. 3

 On the time domahz simulation o f second order wave forces and indttced responses: R. S. Langley

! ! !
technique which uses deterministic wave amplitudes and
0"7- random phase angles does not accurately predict the fourth
moments o f the force and response for the simulation times
considered here, which is in agreement with comments
~ I(1)
made by Tucker et al. t2 concerning the wave group statistics
N (b) produced by this technique.
Possible ways of overcoming the problem o f the varia-
~ (el bility o f the moments of the slow drift force and response
are as follows:
1. The simulation time could be considerably increased
until a point is reached where the moments are accur-
ately predicted by a single simulation. However, the
computing costs required to do this may be prohibi-
F.-
tively large, and a number o f simulations would be
"~ 03 required to check that the moments were correct.
2. A limited number o f simulations of moderate length
could be performed and a factor of safety based on
I I I equation (37) could be applied to the predicted r.m.s.
10 20 30 /,,-0 50 response. The problem here is that the expressions
NUHBER OF CYCLES SIMULATED given by equations (38) and (39) for the terms g and
o which appear in equation (37) are valid only for a
Figure 9. The ratio (o/la)2 for the drift response. [a)
linear system, which may not always be the case,
fl = 0.05 exact method; (b) fl = 0.05 approximate method;
particularly if catenary moorings are used.
(c} fl= 0.01 exact method; {d} fl = 0.1 approximate method
3. An alternative simulation scheme could be used
whereby a time history o f the surface elevation o f
the sea is generated using an ARMA algorithm, aa
approximate method produces a force and response which This time history could be converted into a time
are statistically incorrect, which may have significant conse- history o f the slow drift force using the second order
quences if extreme values are estimated from the resulting impulse response function, which can be found
time histories. numerically as the Fourier transform o f the quadratic
Finally, Fig. 9 shows the effect o f the frequency interval frequency response function. 9 This is equivalent to
dw on the variance o f the mean squared value. The vari- representing the slow drift force as the second term
ances are plotted against the approximate number of slow in a Volterra series, at The advantage o f this approach
drift response cycles which are simulated, No, which can be is that the simulation can be continued until the
found from the frequency interval as N e = r As response moments reach a converged value which,
would be expected the variances decrease as the simulation due to the nature of the ARMA algorithm, should
time increases, meaning that as N c increases fewer runs will be independent of wave grouping.
be requked to give a good estimate of the mean squared
value. However, longer simulation times require a smaller On a practical note, environmental conditions in the
frequency interval dw and thus a greater number o f terms North Sea are stationary for a maximum period of around
N to cover the required frequency range, leading to an three hours or so meaning that the simulations performed
increase in computer costs for a single run. It is therefore here are not necessarily realistic. Rather than concentrate
necessary to strike a balance between the length o f the runs on refining the results of a stationary analysis to a fine
and the number o f runs which are considered. degree, it may be more fruitful in the long term to consider
the non-stationary analysis o f the slow drift responses.
CONCLUSIONS
This paper has been concerned with the time domain simu- REFERENCES
lation o f second order forces and motions when the random
seastate is represented as a sum of sine terms, a technique 1 Pinkster, J. A. Low frequency second order wave exciting
forces on floating structures, Publication No. 650 of the
which is often used in practice as evidenced by refs. 2-5. Netherlands Ship Model Basin, 1980
It has been shown that different results for the spectra and 2 SCrheim, H. Analysis of motion in single point mooring sys-
statistical moments of the force are obtained for each simu- tems, Norwegian Maritime Research, 1981, 2
lation which is performed in this way, meaning that a 3 Marthinsen, T. Calculation of slowly varying drift forces,
number of simulations are required if these quantities are to Applied Ocean Research 1983, 5,141
4 Faltinsen, O. M. and Ll~ken, A. E. Slow drift oscillations of a
be accurately predicted. The variation o f the statistical ship in irregular waves, Applied Ocean Research 1979, 1, 21
moments over an ensemble o f simulations has been exam- 5 Rainey, R. C. T. and Cash, D. G. F. Experience in analysis of
ined for a particular linear system, where it was found that SPM systems, Paper No. OTC 4346, Offshore Technology
at least 25 simulations each o f duration 4 h (50 natural Conference, Houston, Texas, 1982
6 Spangenberg, S. and Kofoed Jacobsen, B. The effect of wave
periods) would be needed to accurately predict the grouping on slow drift oscillations of an offshore structure,
moments. This does not agree with the prediction o f Paper No. 8, International Symposium on Ocean Engineering
Standing et al. 22 that the r.m.s, response will be inde- Ship Handling, Swedish Maritime Research Centre, 1980
pendent of the phasing of the waves for run times greater 7 Bowers, E. C. Long period oscillations of moored ships subject
to short wave seas, Transactions Royal Society of Naval Archi-
than 40 times the natural period, although this statement tects, London, 1976, 118, 181
was directed towards a different simulation technique. 8 Fryer, D. K., Gilbert, G. and Wilkie, M. J. A wave spectrum
It has also been found that an approximate simulation synthesizer, Journal oflt)'draulic Research 1973, 11 (3)

Applied Ocean Research, 1986, Vol. 8, No. 3 141

On the time domahz simulation of second order wave forces and induced responses: R. S. Langley

9 Dalzell, J. F. and Kim, C. H. An analysis of the quadratic

frequency response for added resistance, Journal o f Ship k k !:/=Ir
Research 1979, 23,198
10 Rice, S. O. Mathematical analysis of random noise, in Selected
Papers on Noise and Stochastic Processes,Wax, N. (ed.), Dover + 2 Z Z Z gklrn (A4)
Publications Inc., New York, 1954 k 1 m>l
11 Middleton, D. An hTtroduction to Statistical Communication
Theory, McGraw-Hill Book Company, New York, 1960
The triple sum in the last term of the above equation can
12 Tucker, M. J., ChaUenor, P. G. and Carter, D. J. T. Numerical be divided into five regions, namely k =m, k =l, k < l ,
simulation of a random sea: a common error and its effect 1< k < m and k > m . Writing the contribution from each
upon wave group statistics, Applied Ocean Research 1984, of the regions separately and then making use of the sym-
6, 118 metries ofgktm leads to the final result:
13 Newman, J. N. Second order slowly varying forces on vessels in
irregular waves, Proc. Int. Symp. on Dynamics, Marble Vehicles
and Offshore Structures in leaves, University College London, ~. ~. Z gktm = Z gkkk + 3 E Z gkl,
London, 1974 k 1 m k k 1:/:Ir
14 Neal, E. Second order hydrodynamic forces due to stochastic
excitation, Proc. lOth ONR Symposium, Cambridge, Mass., +61 I Z g,a, (AS)
1974 k l>k m>l
15 Langley, R. S. and Kirk, C. L. Random dynamic analysis of an
offshore single anchor leg storage system, Applied Ocean The second term in equation (31) also has the form of
Research 1982, 4,232 equation (A3), but in this case germ is symmetric in only
16 Godfrey, P. S. Compliant drilling and production platforms, k and l. By dividing the double sum over k and l into the
design and construction of offshore structures, Proe. of the
hmitution of Civil Engineers, London, 1977, 161 regions k =l, k > l and k < l and making use of the sym.
17 Cooley, J. W. and Tukey, J. W. An algorithm for the machine metry, the following result can be derived:
calculation of complex Fourier series, Mathematics for Com-
putation 1965, 19, 297 L E Z gklm = Z E gkkm + 2 ~ ~ ~ gklm (A6)
18 Langley, R. S. The statistics of second order wave forces, k l m m k m k l>k
Applied Ocean Research 1984, 6, 182
19 Spidsoe, N. and Sigbjornsson, R. On the reliability of standaxd Equations (A2), (A5) and (A6) can then be used to effi-
wave spectra in structural response analysis, Engineering ciently evaluate equations (30) and (31).
Structures 1980, 2, 123
20 FORTRAN Library Manual, Numerical Algorithms Group Ltd,
Oxford, UK, 1983, Mark 10, Vol. 5
21 Bedrosian, E. and Rice, S. O. The output properties of Volterra APPENDIX B - EVALUATION OF THE TERMS .fk
systems (nonlinear systems with memory) driven by harmonic
and Gaussian inputs, Proc. ofthelEEE 1971,59, 1688 AND rig
22 Standing, R. G., Dacunha, N. M. C. and Matten, R. B. Slowly- This appendix is concerned with the evaluation of the terms
varying second-order wave forces: theory and experiment,
National Maritime Institute, 1981, Report No. N3H138, fk and rkZ appearing in equations (38) and (39). The analysis
OT-R-8211 below relates specifically to the slow drift force (Yk = Xk),
23 Spanos, P-T. D. ARMA algorithms for ocean wave modelling, but can readily be converted to the slow drift response by
Trang ASME Journal of Energy Resources Technology 1983, introducing the factors Tk, as discussed in the section on
105,300 Moments of the Slowly Varying Force and Response.
Using equations (38) and (12) the term ft~ can be writ-
ten as:
APPENDIX A - EVALUATION OF MULTIPLE SUMS
N--k N--k
This appendix discusses the way in which the computer fk = 4 Z ~. Hb+tc,bH*+ic,cE[fl~b+k~c~*+k] (B1)
time needed to evaluate equations (30) and (31) can be b=l c=l
reduced by the use of symmetry. Firstly, equation (30)
From equation (4) the ensemble average appearing in the
involves a double sum of the form
above equation is only non-zero for b = c, which implies:
Z Z gta (AI) N--k
k I fk = 4 ~ IHb+~,bl2AbAb+k (n2)
where get = gig. This summation can be divided into the b=l
regions k = l, k > l, and k < l, and by making use of the where At, is defined by equation (4). Next, from equations
symmetry ofgkt, equation (A1) can be written as: (39) and (12) the term rkt can be written as

~, Z gta= Z gkg + 2 Z Z get (A2) N--k N--I N--k N--I

k 1 k k I>k rkQ = 16 E E E E Hb+k,bHc+l,cH~+lc,a
b=l c=1 d=l e=l
which involves less operations to evaluate than the original * ~* ^ ~,~ ~ ~* ~ ~*
double sum. Turning next to the first term in equation XHe+t, eE[abab+kacac+taaaa+kaeae+t] (B3)
(31), this has the form: Using equation (4) there are three cases where the ensemble
average appearing in the above equation is non-zero, which
E E Z gklm (A3) are as follows (for k:~l):
k I m
Case1: c = b + k , d = b + l , e = b
where gklm is synunetric in k, 1 and m. The double sum over
l and m can be divided into the regions l = m , l > m and Case2: d = b , e = c
1< m, and making use of the symmetry of getm gives: Case3: c = b - - l , d = b - - l , e = b + k - - I
rkt can be written as the sum of these three contributions in
Zk Z Zmg k l m Zk {~l g k l l + 2 Z, m>lE gklm}: the form:

142 Applied Ocean Research, 1986, Vol. 8, No. 3

 On the time domain simulation of second order wave forces and induced responses: R. S. Langley

3
rgk = 32S~,k + 16pQk (B9)
rkt = 16 ~] S~t k q=l (B4)
i=1 Normally lu is unity but a value of zero is to be used for the
approximate simulation technique. Pkt accounts for the
where SAt is the contribution from case i and is given by:
factor of two which appears in the contribution to S~t
when c = b , c + l = b + k , c = b + k and c = b - - l , and i s
sit = = R e F, Zb+t+k,b+tnb§247
b given by:

* k+~ b+kab+IAb+k+lhb Zb+lc

X Hb+ (B5) = (IHt,+~:,bl IHb+t,bl AbAb+tAb+k
b

§ iHt,+g, bl2 iHt,+g, b+k-tl 2AbAb+lc

2 Ab+tc-t
(B6) + 2 2 2
IHb+~,bl IHb+k+l,b+~:l AbAb§
Care must be taken when k = l as the following further case
+ 2 22
is possible: IHb+k, bl IHb,b-tl AbAb+kAb-t} (BIG)
Case4) d = e, e = b Similarly, Qk accounts for the factor of two which appears
in the contribution to S~k when c = b, c = b + k and
which includes cases 1 and 3 and can be shown to give a
c = b - - k . Thus:
contribution S~k =S~k. Thus when k = l the correct
9expression for rkk is: Ok = 2)". {IHt,+k, bl4AbAb+v.
22
rkk = 16(S~ + S~k) = 32S~k (B7) b

Further complications are caused by the fact tlmt terms + inb+k, bl2 lHb+2k, b+kl 2AbAl~+kAb+2k
2
of the form E [lamP] appear, the value of which depends
upon the statistical properties of d m. Normally the result + IHb+k,bl2 IHb,b-kl 2AbAb+kAb-k}
2
(Bll)
is E[laml 4] = 2A~, but when the approximate simulation A factor of four needs to be introduced for the term involv-
technique is used, which involves using deterministic wave ing AbAb+k,
2 2 and thus it may be thought that a factor of
amplitudes Cn, the result is E [ laml 4] = A2m. six is needed for this term in equation (BI 1). However, the
These effects lead to the final results:
term corresponds to the case c = b, where S~k and S~r are
rta = 16(S~t + S~t +.S~t) + 16pPkt k~ 1 (B8) not distinct cases and only one of them should be included.

Applied Ocean Research, i986, Vol. 8, No. 3 143