Vous êtes sur la page 1sur 11

66

Identification of excitation and reaction forces


spectra for offshore platforms
A.A. Elshafey, M.R. Haddara, and H. Marzouk

Abstract: A simple method is used to predict damping and natural frequencies of a fixed offshore platform subjected to
wind-generated random waves using its stationary response. The method also predicts the reaction forces and bending mo-
ment at the foundation of the platform. The method uses the random decrement signature as a tool to identify the parame-
Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by 184.175.30.163 on 05/26/13

ters in the equation of motion. Excellent agreements were obtained between the estimated and actual values of the
parameters as well as for the reaction and moment at the platform foundation. The method can be applied without any in-
terruption to the operation of the offshore structure. The method is easy to apply and uses inexpensive motion measure-
ment instruments. The estimated forces and bending moments can be used as a tool for an on-line foundation check. A
study on the limitations of applying the random decrement technique is also presented.
Key words: offshore platform, excitation force, random decrement, reaction force.
Résumé : Une méthode simple est utilisée pour prédire l’amortissement et les fréquences naturelles d’une plate-forme de
forage en mer soumise à de la houle aléatoire générée par le vent par l’emploi d’une réponse stationnaire. Cette méthode
prédit également les forces de réaction et le moment de flexion à la base de la plate-forme. La méthode utilise la méthode
du décrément aléatoire comme outil pour déterminer les paramètres dans l’équation du mouvement. D’excellentes corréla-
tions ont été obtenues entre les valeurs estimées et réelles des paramètres et de la réaction et du moment à la base de la
plate-forme. La méthode peut être appliquée sans interrompre les opérations de la structure de forage en mer. La méthode
For personal use only.

est facile à appliquer et utilise des instruments peu dispendieux de mesure du mouvement. Les forces et les moments de
flexion estimés peuvent être utilisés comme outil de vérification de la base en ligne. Une étude des limites d’application
de la technique de décrément aléatoire est également présentée.
Mots-clés : plate-forme de forage en mer, force d’excitation, décrément aléatoire, force de réaction.
[Traduit par la Rédaction]

Introduction vibration-based damage detection techniques have also been


An early detection of fatigue cracks occurring in offshore suggested (Budipriyanto et al. 2007). Other methods that de-
platform members is critical to the safe, efficient, and eco- pend on the measurement of the vibratory response of struc-
nomic operation of the platform. Several approaches have tures only can also be used e.g., operational modal analysis
been suggested to rationalize inspection of offshore plat- technique (Brincker et al. 2001). The random decrement
forms using reliability-based methods (Onoufriou 1999; technique has been applied successfully to multi-degree of
Madhavan Pillai and Prasad 2000) and damage detection freedom systems to predict early damage occurrence (Zu-
techniques (Viero and Roitman 1999). Offshore platforms baydi et al. 2000). Van Overschee and De Moor (1996) pre-
subjected to random waves are modeled usually as multi-de- sented subspace identification algorithm for the
gree of freedom systems. The forces caused by the random identification of multi-input multi-output linear systems
waves can excite certain vibratory modes corresponding to from the measured frequency response data.
frequencies near or equal to wave frequencies. A number of The random decrement (RD) is an averaging technique
that can be used to extract the free decaying response of a
vibrating body from its random excited stationary response.
Received 2 July 2008. Revision accepted 22 July 2009. It was first introduced by Cole (1968) to identify the damp-
Published on the NRC Research Press Web site at cjce.nrc.ca on ing of an aerospace structure using stationary random re-
15 January 2010. sponse. The RD can be obtained without a prior knowledge
A. Elshafey. Department of Civil Engineering, Faculty of of the excitation forces under the assumption that the forces
Engineering, Minufiya University, Shebin Elkom, Egypt. are zero-mean, stationary Gaussian random process. Owing
M. Haddara.1 Department Civil Engineering, Faculty of to its efficiency and simplicity in processing vibration meas-
Engineering, Architecture and Science Ryerson University, 350 urements and the lack of requirements for input excitation
Victoria Street, Toronto, ON M5B 2K3, Canada. measurements, the method is applied extensively to detect
H. Marzouk. Faculty of Engineering and Applied Science, damage in civil and offshore structures (Yang et al. 1980,
Memorial University of Newfoundland, St. John’s, NL, Canada. 1984; Zubaydi et al. 2002; Budipriyanto et al. 2007; Elsha-
Written discussion of this article is welcomed and will be fey et al. 2008). The method can also be used to identify
received by the Editor until 31 May 2010. mode shapes and frequencies of multi-degree of freedom
systems (Ibrahim and Mikulcik 1977). Vandiver et al.
1Corresponding author (e-mail: mhaddara@mun.ca). (1982) showed that the random decrement can be obtained

Can. J. Civ. Eng. 37: 66–76 (2010) doi:10.1139/L09-116 Published by NRC Research Press
Elshafey et al. 67

from the auto-correlation function by multiplying the auto- The symbol P is used in place of P(Y,t|Yo). The solution
correlation function by the threshold or triggering level. of eq. [7] subject to the initial condition lim P(Y,t|Yo) = d(t)
Haddara (1992) extended the random decrement technique as t ? 0 yields an expression for the conditional probability
to nonlinear systems. Zubaydi et al. (2000, 2002) and Budi- density function that governs the process Y(t). Instead of
priyanto et al. (2007) used the random decrement signature solving eq. [7], one can use it to derive expressions that de-
to identify damage in the side shell of a ship. In this work, scribe the propagation of the mean and variance of the proc-
the random decrement was used for its simplicity. It does ess, Y(t), as functions of time.
not need prior knowledge of the excitation force and takes a
short computer running time. Equations of the means and variances
By multiplying both sides of eq. [7] by y1 and y2, respec-
Equation of motion
tively, and integrating the whole equation over the complete
The response, x, of a single degree of freedom linear sys- domain of the variables y1 and y2, it can be shown that
Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by 184.175.30.163 on 05/26/13

tem is governed by the following basic dynamic equation:


_ 1 ¼ m2
m
½1 m€x ðtÞ þ c_xðtÞ þ kxðtÞ ¼ FðtÞ ½8
_ 2 ¼  < 2zuo y2 þ u2o y1 >
m
where m stands for the total virtual mass, c for the damping,
k for the stiffness, t for time, and F(t) is the external force. where m1 and m2 stand for the mean values of the displace-
The total virtual mass is the sum of the physical and the hy- ment and the velocity, respectively.
drodynamic added mass of the system. Multiplying eq. [7] by y21 , y22 , and y1y2, respectively, and
Equation [1] can be normalized with respect to the total integrating the whole equation over the complete domain of
virtual mass m as the variables y1 and y2, we get (Appendix A)
c k FðtÞ n_ 11 ¼ 2n12
½2 x€ ðtÞ þ x_ ðtÞ þ xðtÞ ¼
m m m ½9 n_ 22 ¼ 2 < 2zuo y22 þ u2o y1 y2 > þjo
or n_ 12 ¼ n22  < 2zuo y1 y2 þ u2o y21 >
For personal use only.

½3 x€ ðtÞ þ 2uo z x_ ðtÞ þ uo xðtÞ ¼ f ðtÞ


where n11 is the variance of the displacement, n22 is the var-
where uo is the natural frequency (rad/s), z is the damping iance of the velocity, and n12 is the covariance of the displa-
ratio, f(t) is the force per unit total virtual mass, and x(t) is cement and velocity. Equations [8] and [9] describe the
the response of the system. A dot over the derivative indi- means and the variances of the displacement and velocity
cates differentiation with respect to time. as functions of time. These equations will be used for the
The random excitation f(t) is assumed to satisfy the fol- identification of the parameters in the equation of motion of
lowing conditions: the offshore structure.
Equations [8] can be combined in one equation as
< f ðtÞ > ¼ 0
½4 < f ðtÞf ðt þ tÞ > ¼ jo dðtÞ ½10 _ þ u2o m ¼ 0
€ þ 2zuo m
m

where d is the Kronecker delta and jo is the variance of the where m is the mean value of the displacement. Equation
excitation. [10] shows that the free decay motion can be derived from
The following change of variables is used: the stationary random response. This is the equation of the
random decrement.
½5 y1 ¼ x; y2 ¼ x_

Using the change of variables from eq. [5] in eq. [3], one
Random decrement signature
gets Equation [10] shows that the RD can be used to describe
the free decay response of the system. The advantage of this
y_ 1 ¼ y2 approach is that one can obtain the free response from the
½6
y_ 2 ¼ 2uo z y2  uo y1 þ f ðtÞ stationary random response of the system. To obtain the RD
from the stationary random response, the response is divided
It can be shown (Haddara 2006) that the conditional prob- into a number of segments N, each of length t. All these
ability density function(P(Y,t|Y segments should have the same initial condition, xi(ti) =
) o) governing the vector ran- xs = constant, i = 1,. . ., N. The initial condition is called the
y1
dom process, YðtÞ ¼ , satisfies the Fokker–Planck triggering value. These segments will also have initial slopes
y2 with alternating signs. The ensemble average of the N seg-
equation that is given by ments yields the RD, as shown in Fig. 1.
This approach can be expressed mathematically using the
@P @ @
½7 ¼ ðy2 PÞ þ following equations:
@t @y1 @y2
    j @2P 1X N
 y1 2zuo y2 þ u2o y1 P þ o 2 ½11 xðtÞ ¼ xi ðti þ tÞ
2 @y2 N i¼1

Published by NRC Research Press


68 Can. J. Civ. Eng. Vol. 37, 2010

Fig. 1. Random decrement (RD) approach: (a) Random response, (b) RD after two summations, (c) RD after N summations.
(a) (b) (c)
Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by 184.175.30.163 on 05/26/13

Fig. 2. Pierson–Moskowitz input spectra and the first natural fre- Fig. 3. Random decrement functions corresponding to different
quency. spectra.
For personal use only.

where xi(ti) = xs for i = 1,2,3,. . ., N; x_ i ðti Þ  0 for i = 1,3,5, tions spectra and the magnitude of the first natural fre-
. . ., N – 1; and x_ i ðti Þ  0 for i = 2,4,6, . . ., N. quency of the beam. In Fig. 3, the random decrements
calculated using the responses to the different excitations
Equation [11] gives the random decrement signature. It is
are compared with the free decay response of the beam. It
assumed that using segments with the same initial value and
is clear that the random decrement does not agree with the
alternating slope sign simulates an initial condition where
free decay response if the magnitude of the first natural fre-
the initial displacement is constant and the initial velocity is
quency of the structure is not within the range of the fre-
zero. The triggering level used in this paper is equal to the quencies of the spectrum.
root mean square of the measured parameter. Vandiver et al.
(1982) related the random decrement to the auto-correlation Systems with high damping
function and showed that the random decrement can be ob-
Comparisons between the free decay and the RD for the
tained from the auto-correlation function by multiplying it
beam with two different damping ratios were made. Figure
by the triggering level. The random decrement approach re- 4 shows a comparison between the RD and the free decay
quires no knowledge of the excitation, f(t), as long as it is a response for a beam with a damping ratio of 0.02, while
stationary, zero-mean Gaussian random process. Fig. 5 shows the comparison when the damping ratio is
0.20. It is clear that the RD can only be used for systems
Limitations of the random decrement with low damping ratios, z < 0.02.
technique
Systems with large deformation under Pierson–
Several case studies are considered to illustrate the limita- Moskowitz loading
tions of the RD technique. In this case, we have used a beam subjected to number of
excitations having different magnitude. Figure 6 shows a
First natural frequency is outside the range of comparison of the RD obtained from the different excita-
frequencies of the excitation spectrum tions and the free decay response of the beam. It is clear
In this example, a simple beam is used to discuss the rela- that as the deformation of the beam increases there is a de-
tionship between the natural frequency of the system and the viation between the RD and the free decay.
range of the frequencies of the spectrum of the excitation. A Based on these results, one can conclude that the RD
simple beam was subjected to a series of exciting forces technique can best be applied to lightly damped, linear sys-
having Pierson–Moskowitz spectra. The excitations have tems whose fundamental natural frequency is within the fre-
different range of frequencies. Figure 2 shows the excita- quency range of the excitation spectrum.

Published by NRC Research Press


Elshafey et al. 69

Fig. 4. Comparison between free decay and random decrement Fig. 6. Random decrement (RD) function under magnified loads.
(RD) at z = 0.02.
1.20
Free decay
Normalized displacement

0.80 RD
0.40
0.00
-0.40
-0.80
Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by 184.175.30.163 on 05/26/13

-1.20
0.00 0.10 0.20 0.30 0.40
Time (s)

Fig. 5. Comparison between free decay and random decrement


(RD) at z = 0.20. or
1.20 2p
Free Decay ½13 u¼
Normalized displacement

T
0.80
RD
The damping ratio is calculated using the logarithmic dec-
0.40 rement concept. The logarithmic decrement (d) is given by
 
0.00 1  Ai 
For personal use only.

½14 d ¼ ln 
n Aiþn 
-0.40
where Ai and Ai+n stand for the amplitudes of the ith and
-0.80 (i+n)th cycles, respectively.
The modal damping (z) is related to the logarithmic dec-
-1.20
rement by the equation
0.00 0.10 0.20 0.30 0.40 0.50 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Time (s)
d2
½15 z¼
4p2 þ d2
Analysis of the random decrement signature
The RD signature can be used to determine the natural
frequency and the damping ratio. The RD obtained from the Identification of the excitation and the
stationary response of a multi-degree of freedom system will reaction
contain all the frequencies of that system. In the approach
we are presenting here, we are assuming that we can model In the previous sections, we showed how one can identify
the system as a single degree of freedom system, repre- the natural frequency and the damping ratio using the ran-
sented by its first mode. To isolate the first mode shape, the dom decrement equation. The next step is to identify the ex-
response signal has been passed through a band filter whose citation force f(t).
range encompasses the fundamental frequency of the struc- The variance of the excitation force can be obtained using
ture. Thus, a random decrement is created for the fundamen- eq. [9].
tal mode of the structure. This approach was used In the stationary case, n_ 11 ¼ n_ 12 ¼ n_ 22 ¼ 0, then eq. [9]
successfully by Zubaydi et al. (2000) and Budipriyanto et can be reduced to
al. (2007). The fundamental frequency of existing structures
can be established by building a finite element model for the ½16 jo ¼ 4 zuo n22
as-built structure. A first estimate of the natural frequency and
can be obtained from the random decrement signature. The
filtering range can be based on this estimate. A better esti- ½17 n22 ¼ u2o n11
mate for the natural frequency and the random decrement
can be subsequently obtained from the filtered signal. n11 and n22 can be obtained from the time history of the re-
The natural frequency in Hertz and radians per second can sponse. Equation [16] can then be used to determine the var-
be identified from the RD signature using the measured pe- iance of the exciting force.
riod of one cycle, as The time history of the excitation can be obtained using
eq. [3], rewritten as
1
½12 fr ¼ ½18 f ðtÞ ¼ x€ þ 2zuo x_ þ u2o x
T

Published by NRC Research Press


70 Can. J. Civ. Eng. Vol. 37, 2010

The time history of the reaction force at the foundation, Fig. 7. Random decrement signature of response.
R(t), can be obtained as 1.20
½19 RðtÞ ¼ ð2zuo x_ þ u2o xÞ

Normailzed displacement
0.80

Numerical examples 0.40


Two examples are used to validate the method developed
in this work: a simple single degree of freedom system and 0.00
an offshore platform.
-0.40
Excitation
-0.80
Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by 184.175.30.163 on 05/26/13

The excitation force, f(t), used to generate the response


for the system is represented by the sum of N sinusoidal
terms, as given by -1.20
X
N 0.00 1.00 2.00 3.00 4.00 5.00
½20 f ðtÞ ¼ ai sinðu i tþqi Þ Time (s)
i¼1

The phase angle, qi, is a uniformly distributed random (or numerical integration of the acceleration). In this work,
variable that takes values between 0 and 2p. The amplitudes we assume that we have displacement measurements and
ai are to be determined according to a chosen spectrum. For we use a Lagrange 5-points central differentiation to calcu-
the first example, the excitation force is assumed to have a late the first and second derivatives of the displacement.
wide band spectrum with constant density between the fre- For the first point a forward method is used and for last
quencies, u 1 to u N . This range encompasses the natural fre- point a backward method is used. These quantities are used
quency of the system. In the second example, the excitation in eq. [18] to obtain a predicted record of the excitation
For personal use only.

is assumed to have a Pierson–Moskowitz spectrum. force. To compare the actual and predicted values of the ex-
citation, several measures are used: the variance (s2), and
Example 1: Single degree of freedom system the auto-correlation function, the time history, and spectrum
In this example, we show the results of applying the comparison. The excitation is a zero-mean process. The var-
method to an SDOF system subjected to a broad band exci- iance is calculated for a record having a length of 500 s. The
tation. The force amplitudes per unit mass ai are assumed to variances for the actual and predicted exciting forces are
be 0.0625N/kg. The frequency u is assumed to be 10 rad/s shown in Table 2. The error in the prediction is less than
and the damping ratio z is assumed to be 0.02. The excita- 0.2%. A comparison between the autocorrelation functions
tion is represented by 5000 sinusoidal components over a for the actual and predicted force is given in Fig. 8. It is evi-
frequency range of 5.0 to 15.0 Hz. The response is calcu- dent that the correlation functions for the actual and pre-
lated using a Runge-Kutta method (Shoup 1980; Thomson dicted forces have an excellent agreement. Figure 9 shows
1993). Integration was done using a time increment of the relationship between the actual and estimated forces.
0.02 s. Figure 7 shows a random decrement signature for
this system obtained using a triggering level equal to the Example 2: Offshore structure
standard deviation of the response (s = 0.1898m). Several In this example, we deal with a simple offshore platform
triggering levels were tried; it was found that a triggering shown in Fig. 10. The platform consists of a vertical steel
level of one to one and half the standard deviation gives ex- pipe standing in 26.0 m water depth. The diameter of the
cellent results. The number of segments used to construct pipe is 2.10 m for the part under water and its wall thickness
the random decrement shown in Fig. 7 is 936. Good results is five centimetres. The part outside the water is 1.50 m in
are usually obtained if the number of segments is more than diameter and has a wall thickness of 4 cm. The platform
100. The time series generated is 500 s long that gives 25 supports a mass of 3000.00 kg on deck. It is assumed
000 sampling points. that the platform is subjected to waves having a Pierson–
The period is measured directly from the digitized graph Moskowitz spectrum shown in Fig. 11. The significant wave
using the average of the first five cycles and found to be height (Hs) of the waves is 2.0 m and the average zero-up-
0.6274 s. The natural frequency in Hertz is calculated using crossing period (Tz) is 6.00 s. The response of the platform
the relation: fr = 1/T. The damping is calculated using the and the reaction force and moment were calculated using an
logarithmic decrement, as in eqs. [14] and [15]. Table 1 ANSYS software package. A study was done for choosing a
shows the comparison between the actual values and the suitable number of elements. The wave input was repre-
predicted values. sented by a series of sine waves in the time domain. The am-
plitudes and frequencies were taken from the assumed
Prediction of the excitation spectral density function of the waves and the phase angle
A time history for the exciting force was generated using was assumed to be a random variable having a uniform dis-
eq. [18]. Usually, only the displacement (or acceleration) is tribution between 0 and 2p. The platform is divided into 46
known from the measured response, the velocity can be ob- elements. To allow for water waves, an ANSYS pipe59 ele-
tained using numerical differentiation of the displacement ment was used. The data was recorded at a rate of 20.00 Hz.

Published by NRC Research Press


Elshafey et al. 71

Table 1. Actual and estimated natural frequency and damping. Fig. 10. Offshore platform.
Item Actual (rad/s) Predicted (rad/s) Error (%)
Damped natural 9.998 10.012 0.14
frequency, ud
Damping ratio, z 0.02 0.0203 1.50

Table 2. Variance comparison.

From actual From estimated


Item force force Error (%)
Variance (s2) 9.714 (N2) 9.732 (N2) 0.19
Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by 184.175.30.163 on 05/26/13

Fig. 8. Auto-correlation function of the excitation.


1.20
Auto-correlation function

0.80 Actual force


Estimated force
0.40

0.00

-0.40

-0.80
For personal use only.

-1.20
0.00 1.00 2.00 3.00 4.00
Time lag (s)
Fig. 11. Pierson–Moskowitz spectrum.
Fig. 9. Relationship between actual and estimated forces.
0.50

0.40
Amplitude (m)

0.30

0.20

0.10

0.00
0.00 1.00 2.00 3.00 4.00 5.00 6.00
Frequency (Rad/s)

noise in the data. For the first mode, a filter of band pass of
the range 0.25 to 1.50 Hz is used and for the second mode a
band pass filter of range between 2.5 to 4.5 was used.
The drag and inertia coefficients were assumed to be 1.20 It can be shown from Table 3 that the RD approach can
and 2.00, respectively. The top nodal displacement was re- predict the natural frequencies with sufficient accuracy. The
corded at a frequency of 20.00 Hz. The output data was material damping used in the analysis was 0.02; this means
passed through a band pass filter. The RD signature of the that the water hydrodynamic damping is estimated to be
displacement was extracted for the filtered data containing 0.0576 for the first natural frequency and 0.027 for the sec-
the first natural frequency, as shown in Fig. 12. The RD sig- ond natural frequency. It should be noted that only the
nature for the filtered data containing the second natural fre- modes in the direction of wave propagation are considered.
quency is shown in Fig. 13. The filter is used to separate the The period of one cycle was measured directly from the
response corresponding to the modes and remove unwanted digitized curve and found to be 0.99 s. The total damping

Published by NRC Research Press


72 Can. J. Civ. Eng. Vol. 37, 2010

Table 3. Comparison between estimated and actual frequencies.

Item Actual (Hz) Estimated (Hz) Error (%) Damping ratio


1st natural frequency 1.032 1.01 2.13 0.0776
2nd natural frequency 3.766 3.95 –4.88 0.0470

Table 4. First five frequencies of the platform

Mode No. 1 2 3 4 5
Natural frequency (Hz) 1.0328 3.7661 10.597 19.680 23.413
Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by 184.175.30.163 on 05/26/13

Fig. 12. Random decrement signature for the first mode of plat- Fig. 13. Random decrement signature for the second mode of plat-
form. form.
1.20 1.20

Normailzed displacement
Normalized displacement

0.80 0.80

0.40 0.40

0.00 0.00

-0.40 -0.40

-0.80 -0.80
For personal use only.

-1.20 -1.20
0.00 1.00 2.00 3.00 4.00 5.00 0.00 0.50 1.00 1.50 2.00 2.50 3.00
Time (s) Time (s)

ratio is extracted from the logarithmic decrement and found functions of the estimated and actual reactions are compared
to be 0.0776. Table 4 shows the first five natural frequencies in Fig. 16. It is clear that the method can estimate the reac-
of the platform. tion force per unit mass successfully.
The actual bending moment is plotted against the esti-
Estimation of reaction force and bending mated reaction force per unit virtual mass, the relationship
is found to be linear, as shown in Fig. 17. The slope of this
moment line gives the product of the virtual mass and the moment
For an SDOF system, the reaction force does not depend lever arm. Using the two relations obtained from Figs. 14
on the inertia force; it depends only on the damping and re- and 17, one can obtain an approximate value for the point
storing forces. In this work, we model the platform as a sin- of application of the resultant external force on the platform.
gle degree of freedom system having a natural frequency This point was found to be at a height of 25.66 m above the
equal to the fundamental frequency of the platform. The base, for this example. This is approximately near to the
horizontal reaction per unit mass at the foundation can be height of the mean free water surface. The auto-correlation
estimated from eq. [19]. A comparison between the actual functions of the actual bending moment and the estimated
reaction and estimated reaction force per unit mass is shown normalized bending moment are compared in Fig. 18, show-
in Fig. 14. By plotting the relationship between the actual ing an excellent agreement.
reaction and the predicted reaction per unit virtual mass, This result was further investigated using different water
one is able to determine the value of the total virtual mass depths and significant wave heights. The results of this in-
of the structure. A straight line passing through the origin is vestigation are shown in Figs. 19 and 20. The results shown
fitted to the points taken from time history records of the re- in Fig. 15 indicate that for the cases considered here, the
action forces. The slope of the straight line fit gives an ap- variation in the location of the resultant external force is sig-
proximation of the total virtual mass. nificant. For water depths between 26.00 and 30.00 m, the
The normalized auto-correlation functions for both of the point of application of the resultant external force is located
estimated reaction per unit virtual mass and the total reac- at a point close to the mean free water level. When the depth
tion obtained from the finite element analysis are shown in is decreased to 20.00 m, the point of action of the external
Fig. 15. An excellent agreement between the two auto-corre- resultant force is found to be located at a height of about 0.9
lation functions can be observed. The power spectral density the depth of the platform above the level of the base. When

Published by NRC Research Press


Elshafey et al. 73

Fig. 14. Relationship between total actual reaction and estimated Fig. 17. Relationship between estimated reaction per unit virtual
reaction per unit mass. mass and actual moment.
5
4.00x10 1.20x107
Actual values
Y = 61875.74 X 8.00x106
5 Fitted
Actual reaction (N)

Actual My (N·m)
2.00x10
6
4.00x10

0.00x10
0
0.00x100

-4.00x106
5
-2.00x10 Y = 1587797.9 X
Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by 184.175.30.163 on 05/26/13

Actual points
Fitted -8.00x106
7
-4.00x10
5 -1.20x10
-8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00 6.00 -8.00 -4.00 0.00 4.00 8.00
Estimated reation / unit mass (N/kg) Estimated reaction / unit mass (N/kg)

Fig. 18. Auto-correlation function of moment at base.


Fig. 15. Auto-correlation functions for actual and estimated reac- 1.20
tion forces.
Auto-correlation function Estimated My
1.20 0.80 Actual My
Actual
Auto-correlation function

0.80 Estimated 0.40


For personal use only.

0.40 0.00
0.00 -0.40
-0.40 -0.80

-0.80 -1.20
0.00 1.00 2.00 3.00 4.00 5.00
-1.20
Time-lag (s)
0.00 1.00 2.00 3.00 4.00 5.00
Time-lag (s) Fig. 19. Relationship of water depth and location of force produ-
cing the bending moment.
1.50
Fig. 16. Power spectral density comparison for actual and estimated
Force location / water depth

reactions. Actual
12 Fitted
10 1.25
11
10 Actual
10
10
9 Estimated
1.00
PSD (N2/Hz)

10
8
10
107
6
10 0.75
5
10
4
10
3
10 0.50
2
22.00

24.00

26.00

28.00

30.00
14.00

16.00

18.00

20.00

10
0.00 2.00 4.00 6.00 8.00 10.00
Frequency (Hz)
Water depth (m)

Published by NRC Research Press


74 Can. J. Civ. Eng. Vol. 37, 2010

Fig. 20. Relationship between virtual mass and water depth. tion of output-only systems using frequency domain decomposi-
3 tion. Smart Materials and Structures, 10(3): 441–445. doi:10.
3.60x10
Virtual mass / water depth (kg/m)

1088/0964-1726/10/3/303.
Actual Values Budipriyanto, A., Haddara, M.R., and Swamidas, A.S.J. 2007. Iden-
3 Fitted
3.20x10 tification of damage on ship’s Cross-stiffened plate panels using
vibration response. Ocean Engineering, 34(5-6): 709–716.
3
doi:10.1016/j.oceaneng.2006.05.009.
2.80x10 Cole, H.A. 1968. On the line analysis of random vibration. In Pro-
ceedings of the AIAA/ASME 9th Structures, Structural Dy-
2.40x10
3 namics and Materials Conference, Palm Springs, Calif. 1–3
April 1968. American Institute of Aeronautics and Astronautics,
New York. 8 p.
3
2.00x10 Elshafey, A., Haddara, M.R., and Marzouk, H. 2008. The use of
Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by 184.175.30.163 on 05/26/13

random decrement technique for health monitoring of structures.


3 In Proceedings of the 5th International Conference on Advanced
1.60x10 Composite Materials in Bridges and Structures (ACMBS-V),
18.00

28.00

30.00

32.00
20.00

22.00

24.00

26.00

Winnipeg, Man. 22–24 September 2008.


Haddara, M.R. 1992. On the random decrement for nonlinear roll-
Water depth (m) ing motion. In Proceedings of the 11th International Conference
on Offshore Mechanics and Arctic Engineering, Calgary, Alta.,
7–12 June 1992. American Society of Mechanical Engineers,
New York. Vol. II, pp. 321–324.
the water depth is further decreased to 16.00 m, the point of
Haddara, M.R. 2006. Complete identification of the roll equation
action of the force was lowered to a point 0.85 of the depth
using the stationary random roll response. Journal of Ship Re-
of water above the base. The change of the virtual mass with search, 50(4): 388–397.
depth is shown in Fig. 20. The results show that as the depth Ibrahim, S.R., and Mikulcik, E.C. 1977. A method for the direct
increases the density of the virtual mass decreases. Since the identification of vibration parameters from the free response.
density of the virtual mass is calculated as the total virtual
For personal use only.

Shock and Vibration Bulletin, 47(4): 183–198.


mass divided by the water depth, we expect that as the water Madhavan Pillai, T.M., and Prasad, A.M. 2000. Fatigue reliability
depth increases the density should decrease. It should be analysis in time domain for inspection strategy of fixed offshore
noted that the curves illustrated in this paper represent the structures. Ocean Engineering, 27(2): 167–186. doi:10.1016/
case under study and the effect of some parameters are not S0029-8018(98)00020-1.
included such as the significant wave height. Onoufriou, T. 1999. Reliability based inspection planning of off-
shore structures. Marine Structures, 12(7-8): 521–539. doi:10.
1016/S0951-8339(99)00030-1.
Conclusions Shoup, E.T. 1980. Applied numerical methods for the microcompu-
The random decrement (RD) method is a powerful ter. Prentice-Hall, Englewood Cliffs, NJ.
method that can be used to predict the parameters in the Thomson, W.T. 1993. Theory of vibration with applications. 4th
equation of motion of vibratory systems using their station- ed. Prentice Hall, Englewood Cliffs, NJ.
ary response in random seas with no prior knowledge of the Van Overschee, P., and De Moor, B. 1996. Continuous-time fre-
excitation. This allows the method to be used without inter- quency domain subspace system identification. Signal Proces-
rupting the normal operation of the system. The structure sing, 52(2): 179–194. doi:10.1016/0165-1684(96)00052-7.
was modeled as a single degree of freedom system, and the Vandiver, J.K., Dunwoody, A.B., Campbell, R.B., and Cook, M.F.
1982. A mathematical bases for the random decrement vibration
fundamental mode was used to describe the response of the
signature analysis technique. Journal of Mechanical Design, 104:
structure. The error in the prediction of the fundamental nat-
307–313.
ural frequency is less than 5%. The work reported here de- Viero, P.F., and Roitman, N. 1999. Application of damage identifi-
scribes the validation of the methodology for intact cation methods in offshore platforms. Marine Structures, 12(2):
structures. Several significant results are obtained in this 107–126. doi:10.1016/S0951-8339(99)00007-6.
work. These include a method for the prediction of the Yang, J.C.S., Dagalakis, N.G., and Hirt, M. 1980. Application of
wave exciting force and its location on an offshore struc- the random decrement technique in the detection of induced
ture, a method for the prediction of the reaction force at cracks of an offshore platform model. In Computational Meth-
the foundation of an offshore structure, and a method for ods for Offshore Structures, Special Publication, AMD, Library
the estimation of the total virtual mass of the structure in of congress No. 80–69180. ASME. Vol. 37.
waves. Yang, J.C.S., Chen, J., and Dagalakis, N.G. 1984. Damage detec-
tion in offshore structures by random decrement technique. Jour-
Acknowledgements nal of Energy Resources Technology, 106: 38–42.
Zubaydi, A., Haddara, M.R., and Swamidas, A.S.J. 2000. Damage
We would like to acknowledge the financial support pro-
identification in stiffened plates using the random decrement
vided by the Natural Sciences and Engineering Research
technique. Oceanic Engineering International, 4(1): 22–30.
Council of Cananda (NSERC). Zubaydi, A., Haddara, M.R., and Swamidas, A.S.J. 2002. Damage
identification in a ship’s structure using neural networks. Ocean
References Engineering, 29(10): 1187–1200. doi:10.1016/S0029-8018(01)
Brincker, R., Zhang, L., and Andersen, P. 2001. Modal identifica- 00077-4.

Published by NRC Research Press


Elshafey et al. 75

List of symbols x; x_ ; x€ displacement (m), velocity (m/s), and acceleration


(m/s2), respectively
c damping coefficient (Ns/m) d logarithmic decrement or Kronecker delta
fr natural frequency (Hz) z nondimensional damping coefficient
F(t) excitation force (N) m1,m2 mean values of displacement and velocity, respec-
f(t) reaction force/unit mass tively
k stiffness (N/m) n11,n22 the variance of displacement and velocity, respec-
m mass (Kg) tively
P(Y,t|Yo) conditional probability density s standard deviation
RD random decrement jo variance of excitation
R(t) reaction force (N) uo natural frequencies (rad/s)
Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by 184.175.30.163 on 05/26/13

Appendix A
Details of the derivation of eqs. [8]-[10] are given in this Appendix.
Using the notations N = 2uzy2, F = u2y1, and dP(Y,t|Yo) = P(Y,t+dt|Yo) – P(Y,t|Yo)
Equation [7] can be rewritten as

@ @ j @2P
½A1 PðY; t þ dtjYo Þ  PðY; tjYo Þ ¼  ðy2 PÞ þ fðN þ FÞPg þ o dt
@y1 @y2 2 @y2
Multiplying the two sides of eq. [A-1] by y1, y2, and y1y2, respectively, and integrating the equation w.r.t. y1 and y2 from –?
to ?, we get
Z1Z
For personal use only.

y1 PðY; t þ dtjYo Þdy1 dy2 ¼ m1 ðt þ dtÞ


1
Z1Z
y1 PðY; tjYo Þdy1 dy2 ¼ m1 ðtÞ
1
Z1 Z  Z
Z1

@ y1 ¼1 R y1 ¼1
y1  ðy2 PÞ dy1 dy2 ¼  dy2 y1 y2 P j  y2 P dy2 ¼ m2  dy2 y1 y2 P j ¼ m2
@y1 y1 ¼1 y1 ¼1
1 1
Z1 Z Z1 
@ y2 ¼1
y1 ðN þ FÞPdy1 dy2 ¼ y1 dy1 ðN þ FÞP j ¼0
@y2 y2 ¼1
1 1

Z1 Z  Z1

@ y2 ¼1
y2  ðy2 PÞ dy1 dy2 ¼ y2 dy2 y2 P j ¼0
@y1 y2 ¼1
1 1
2 3
Z1 Z Z1 Z1
@ y2 ¼1
y2 ðN þ FÞPdy1 dy2 ¼ dy1 4y2 ðN þ FÞP j  ðN þ FÞPdy25
@y2 y2 ¼1
1 1 1

R y2 ¼1
¼  < N þ F > þ dy1 y2 ðN þ FÞP j ¼  < 2uzy2 þ u2 y1 >
y2 ¼1

Multiplying the two sides of eq. [A-1] by y21 , y22 , and y1y2, respectively, and integrating the equation w. r. t. y1 and y2 from
–? to ? we get
Z1Z
y21 PðY; t þ dtjYo Þdy1 dy2 ¼ n11 ðt þ dtÞ
1
Z1Z
y21 PðY; tjYo Þdy1 dy2 ¼ n11 ðtÞ
1

Published by NRC Research Press


76 Can. J. Civ. Eng. Vol. 37, 2010

Z1 Z  Z1 Z1 Z
@ y1 ¼1
y21  ðy2 PÞ dy1 dy2 ¼  y21 y2 P j dy2 þ 2 y1 y2 Pdy1 dy2 ¼ 2n12
@y1 y1 ¼1
1 1 1

Z1 Z Z1

j @2P j @P 1
y21 o 2
dy1 dy2 ¼ o y31 j dy1 ¼ 0
2 @y2 2 @y2 1
1 1

Z1Z
y22 PðY; t þ dtjYo Þdy1 dy2 ¼ n22 ðt þ dtÞ
1
Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by 184.175.30.163 on 05/26/13

Z1Z
y22 PðY; tjYo Þdy1 dy2 ¼ n22 ðtÞ
1

Z1 Z Z1 Z1 Z
@   y2 ¼1
y22 ðN þ FÞP dy1 dy2 ¼ ðy22 ðN þ FÞP j dy1  2y2 ðN þ FÞPdy1 dy2
@y2 y2 ¼1
1 1 1
Z1 Z
¼ 2 ð2uzy22 þu2 y1 y2 ÞPdy1 dy2 ¼ 2 < 2uzy22 þ u2 y1 y2 >
1
For personal use only.

Z1 Z Z1 Z1 Z
jo @ 2 P j @P @P
y22 2
dy1 dy2 ¼ o y22 dy1  jo y2 dy1 dy2 ¼ jo
2 @y2 2 @y2 @y2
1 1 1

In deriving eqs. [8] and [9], we assumed the following boundary conditions:

y1 ¼1 y¼1 y1 ¼1 y¼1 y1 ¼1 @P y¼1


y1 y2 P j ¼ ðN þ FÞP j ¼P j ¼P j ¼ yi ðN þ FÞP j ¼ j ¼ 0; i ¼ 1; 2
y1 ¼1 y¼1 y1 ¼1 y¼1 y1 ¼1 @y2 y¼1

Equation [10] can be derived as follows:


€ 1 ¼  < 2zuo y2 þ u2o y1 >
_2 ¼ m
m

_ 1 þ u2o m1 Þ
€ 1 ¼ ð2zuo m
m

_ _ þ u2o m1 ¼ 0
€ 1 þ 2zuo m
m

In eq. [10], the symbol m is used instead of m1, where m is used for the mean value of the displacement.

Published by NRC Research Press

Vous aimerez peut-être aussi