J Apor 2016 10 004

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

Applied Ocean Research 61 (2016) 130–147

Contents lists available at ScienceDirect

Applied Ocean Research


journal homepage: www.elsevier.com/locate/apor

Improved assessments of wave-induced fatigue for free spanning


pipelines
Håvar A. Sollund a,b,∗ , Knut Vedeld b , Olav Fyrileiv b , Jostein Hellesland a
a
Mechanics Division, Department of Mathematics, University of Oslo, Moltke Moes vei 35, Pb. 1053 Blindern, 0316 Oslo, Norway
b
DNV GL A/S, Veritasveien 1, 1364 Høvik, Norway

a r t i c l e i n f o a b s t r a c t

Article history: For subsea pipeline projects, the costs related to seabed correction and free span intervention are often
Received 1 October 2015 considerable. Development of reliable methods for fatigue analyses of pipelines in free spans contributes
Received in revised form 2 September 2016 to minimize costs without compromising pipeline integrity. Assessment of wave-induced fatigue dam-
Accepted 13 October 2016
age on multi-span pipelines is investigated, and improved analysis methods are suggested in this paper.
Available online 16 November 2016
A time-domain (TD) algorithm is developed, which accounts for non-linear hydrodynamic loading and
dynamic interaction between adjacent spans. The proposed TD approach is employed to evaluate lin-
Keywords:
earized frequency-domain (FD) solutions from recognized design standards and to study the dynamic
Free spans
Multi-spans
response of multi-span pipelines to direct wave loading. Differences between multi- and single-span
Structural dynamics analyses are described for the first time, and the common assumption that the main fatigue damage con-
Offshore pipelines tribution comes from the fundamental mode is demonstrated not to hold for multi-spans. An improved
Pipeline design FD solution capable of predicting multi-mode response is derived and demonstrated to give accurate
Wave loading fatigue life estimates for multi-span pipelines.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction mechanisms and response frequencies are different, calculations


of fatigue damage induced by waves and VIV are performed sepa-
Unsupported pipeline spans – conventionally termed free spans rately [1,3]. The widely used recommended practice DNV-RP-F105
– may result from permanent seabed roughness or from seabed [3] includes comprehensive fatigue assessment methodologies for
mobility and scouring processes [1]. Free spans may pose a risk both VIV and direct wave loading.
due to trawl pull-over or hooking [2–4]. Furthermore, vortex shed- For accurate fatigue life predictions, many sea-states must be
ding and direct wave loading may induce a dynamic response in analyzed for each span. Combined with a large number of spans,
the pipeline [5,6], which is commonly assessed by a fatigue limit the amount of analyses required can become massive. Hence, more
state approach [3]. Thus, free spans constitute an integral part of accurate time-domain (TD) simulations using general purpose
design and operations management of submarine pipelines. Expen- finite element analysis (FEA) software are too time-consuming, and
sive span intervention may be required, and pipeline failures have frequency-domain (FD) calculations have remained the industry
been observed when free spans were not appropriately designed standard due to their superior computational efficiency [1,3]. The
for or adequately monitored [7]. FD algorithms require accurate prediction of the pipeline’s modal
Vortex-induced vibrations (VIV) are triggered by resonance response, and the modal analysis should account for dynamic inter-
between pipeline eigenfrequencies and periodic pressure oscilla- action between adjacent spans. Such interacting spans are termed
tions due to vortex shedding. The vibrations may occur both in multi-spans. Both eigenfrequencies and mode shapes are affected
the vertical (cross-flow) and in the horizontal (in-line) direction by span interaction, and it has been found that conventional single-
[3,5,6]. As opposed to VIV, displacements due to direct wave forces span analyses may be inaccurate and even non-conservative [8,9].
are normally considered in the in-line direction only and generally Moreover, separation of eigenfrequencies may be less pronounced
occur at the wave frequency, which typically is much lower than for multi-span pipelines (hereafter referred to simply as multi-
the frequencies associated with in-line VIV [5]. Since the loading spans) as compared to single-span pipelines (hereafter referred to
as single-spans). Several vibration modes may be excited simul-
taneously (i.e., giving a multi-mode response), and the fatigue
damage contributions from higher-order modal response should
∗ Corresponding author at: Mechanics Division, University of Oslo, P. Box 1053-
be accounted for [8].
Blindern, NO-0316 Oslo, Norway.
E-mail address: haavaras@math.uio.no (H.A. Sollund).

http://dx.doi.org/10.1016/j.apor.2016.10.004
0141-1187/© 2016 Elsevier Ltd. All rights reserved.
H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147 131

While there has been extensive focus on improving VIV analy-


ses [1], relatively little attention has been devoted to the study of
wave-induced fatigue. Notably, not a single study has to our knowl-
edge assessed how wave-induced fatigue damage may be affected
by multi-mode response or span interaction. In DNV-RP-F105 [3],
a force model based on the Morison equation [10] is detailed. The
Morison equation contains a non-linear drag term, which makes
the wave-loading and resulting response process non-Gaussian. To
facilitate a stochastic FD solution, the drag term is therefore lin-
earized [3,11]. Another potential source of inaccuracy, particularly
for multi-spans, is the assumption that only the fundamental mode
contributes non-negligibly to fatigue damage. A multi-mode exten-
sion was suggested by Mørk and Fyrileiv [11], but the accuracy of Fig. 1. Definition of multi-span model and coordinate system. The pipeline rests
their algorithm has not been thoroughly evaluated. on shoulders with linear elastic soil (visualized as springs) and is subjected to dis-
An alternative, but quite similar force model methodology was tributed hydrodynamic loading p(x,t).
suggested by Xu et al. [12]. The study outlined both TD and FD
solutions, with the latter based on linearized Morison loading. The The modeled section may, however, comprise an arbitrary number
suggested procedure was based on a modal analysis of an isolated of spans. Span lengths are denoted Li and intermediate span shoul-
span with boundary conditions idealized as single rotational and der lengths are denoted Lint ,i . The longest span in the multi-span
vertical springs at either end of the span. The authors proposed to section is called the main span. The lengths of the outer span shoul-
use the springs to reflect effects of span interaction, but guidance ders are denoted Lshoulder and the pipeline is assumed to be simply
was neither given on how to adjust the spring stiffness to repre- supported at the outermost ends. The boundary conditions are,
sent multi-span conditions nor on how to obtain correct bending however, inconsequential as Lshoulder should be sufficiently long to
stresses. Considering that maximum modal stresses often occur ensure that the pipe dynamic response is not affected by increasing
at the span shoulders [3,8], it is doubtful whether a generalized its value further. The total model length L becomes
spring model is appropriate, particularly for multi-spans where
modal stresses interact on the shoulders. Xu et al. [12] validated 
r

r−1

L = 2Lshoulder + Li + Lint,i . (1)


their model by assessing a single-span using different TD and FD
i=1 i=1
approaches. The fatigue lives varied by up to a factor of 6.5. These
deviations were attributed to the large uncertainty associated with The pipeline is modeled using Euler-Bernoulli beam theory since
fatigue damage assessments in general, without further explana- shear deformations have been shown to have a negligible impact
tions. on the modal response of offshore pipelines [3,13]. Moreover, the
The present study aims to clarify how to estimate wave-induced pipeline is assumed to be initially straight, which is reasonable since
fatigue damage on multi-spans, an area where governing design the direct wave and current loading is restricted to the horizon-
codes until now have been vague. For this purpose, a computation- tal plane. Thus, relevant displacements and bending stresses are
ally efficient TD approach for assessing fatigue damage due to direct decoupled from static curvatures induced by seabed topography
wave loading is derived. The suggested method has three advan- and gravity, which are known to influence the tangent stiffness for
tages compared to current industry practice [3]; (1) linearization cross-flow vibrations [3,8,13].
of the drag force is not required, resulting in more accurate repre- Non-linear geometric effects may influence the dynamic
sentation of the hydrodynamic loading, (2) multi-mode response is pipeline response. These include tension variation due to elonga-
included, and (3) span interaction is accurately accounted for. Com- tion of the pipe, as well as a coupling between the axial and bending
bined, items (2) and (3) enable realistic multi-span analyses. The TD stiffness resulting from large rotations [8,13]. Such effects must
approach employs a semi-analytical modal analysis method previ- be considered for conventional bottom roughness (static) analy-
ously presented by Sollund et al. [8] in conjunction with a mode ses [1,3,14]. However, dynamic displacements due to direct wave
superposition procedure and an explicit time integration scheme. loading are, for most sea-states, likely to be smaller than static ver-
The derived algorithm incorporates the partial safety factor format tical displacements imposed on the pipe by the combined effect
specified by DNV-RP-F105 [3] and is verified against independent, of functional loading, gravity and seabed topography. The dynamic
non-linear TD FEA. Application of the TD algorithm to multi-spans pipeline response is therefore linearized about the (straight) static
is thoroughly investigated, and effects of span interaction on wave- equilibrium position, i.e., the mass and stiffness properties are
induced fatigue described for the first time. Since the quadratic assumed to be time invariant. The linearization will be justified by
drag term in the Morison equation is retained, the TD approach comparison to fully non-linear FE analyses. Non-linear geometric
is also applied to evaluate the linearized FD solution in DNV-RP- effects were also disregarded in the solutions by Mørk and Fyrileiv
F105 [3]. A more accurate FD approach is derived and its suitability [3,11] and Xu et al. [12].
for multi-span analyses is investigated. The span shoulders are modeled using continuously distributed
It should be noted that the hydrodynamic force coefficients for linear elastic soil stiffness, which is visualized using discrete lin-
combined wave and current loading suggested by [3] are adopted ear springs in Fig. 1. Non-linear pipe-soil interaction effects, such
directly, and that the accuracy of the applied Morison force model as sliding on the shoulders due to severe wave loading, are thus
is not a subject of this paper. not reflected in the model. For cases where large lateral displace-
ments can be expected on the span shoulders, the non-linear
force-displacement relationship between pipe and soil may be rep-
2. Problem definition resented using appropriate empirically based models [3,15,16]. In
the present study, linear elastic soil stiffness is chosen to align the
2.1. Geometry and model description proposed algorithm with the FD approach in current industry prac-
tice [3]. Hence, soil stiffness and modal damping coefficients are in
Fig. 1 gives a schematic representation of a multi-span pipeline each case selected based on the guidance provided in DNV-RP-F105
with two spans and associated outer and intermediate shoulders. [3]. Note that axial soil stiffness can be disregarded since non-linear
132 H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147

effects are assumed to be negligible and only lateral displacements The instantaneous water particle velocity U is taken as the sum
are accounted for. of the wave-induced particle velocity Uw and the current velocity
Uc , i.e.,
2.2. Loading conditions
U = Uw + Uc . (8)
Pipelines are subjected to static loads from gravity, internal
U will generally be a function of both location x along the pipe axis
and external pressure, and temperature. In addition, the pipelines
and of time t. Uw and Uc are assumed to be co-linear, and only waves
must resist dynamic hydrodynamic loads from waves and currents.
are assumed to contribute to the hydrodynamic inertia loading, i.e.,
During bending, the internal and external pressures exert a net
the time derivative of U will be taken as equal to the wave-induced
transverse loading directed along the radius of curvature [17,18].
water particle acceleration:
The pressure-induced transverse loading is conveniently accounted
for by letting second-order load effects be governed by an effective U̇ = U̇w . (9)
axial force [2,18] defined as
The hydrodynamic loading is modeled strictly according to the
Seff = N − pi Ai + pe Ae , (2) recommendations in DNV-RP-F105 [3], which reflects free span
where N is the pipe-wall axial force, pi and pe are the internal and design practice. The following assumptions are made to be con-
external pressures, and Ai and Ae are the internal and external cross- sistent with the FD methodology in [3]:
sectional areas.
The effective axial force Seff , span lengths and shoulder lengths • The drag coefficient CD and inertia coefficient CM are taken as
are considered as input parameters in the present study, but functions of the relative pipe-surface roughness ks /D, the gap
should generally be estimated by a preceding bottom roughness ratio eg /D and the sea-state specific flow conditions. The gap eg
(static) analysis, accounting for the functional loading, self-weight, between the pipe and the seabed shall for each span be taken as
seabed topography and non-linear pipe-soil interaction [3,14]. An the average value over the central third of the span.
increased value of Seff (defined as positive in tension) results • The load coefficients are independent of the Reynolds number,
in higher natural frequencies, and also higher associated modal i.e., supercritical flow is assumed. This is an implicit assumption in
stresses. It is often, but not always, conservative to apply a lower- DNV-RP-F105 since pipelines typically have a rough surface and
bound value for the effective axial force. Such a lower-bound value, flow states relevant for fatigue or extreme loading are associated
corresponding to the solution for a pipe that is fully restrained with high flow velocities.
axially, is given by [2,18] • The added mass coefficient Ca is assumed time invariant and
expressed solely as a function of eg /D (by taking the limit of Eq. (5)
Seff = Heff − pi Ai (1 − 2v) − EA˛T (3) for still water conditions). Thus, Ca is taken as unity for a cylinder
where Heff is the effective residual lay tension, pi is the internal in free flow, but gradually increases to 2.29 as the gap is reduced
pressure difference relative to laying, Ai is the internal cross- to zero [3,5,19].
sectional area of the pipe,  is the Poisson’s ratio, E and A are • The water-particle velocity U(t) is independent of x, i.e., the mean
the Young’s modulus and the area of the steel wall cross-section, wave direction is assumed to be perpendicular to the pipe with
respectively, ␣ is the thermal expansion coefficient, and T is the most of the wave energy concentrated around the main wave
change in temperature relative to the temperature during laying. direction.
The distributed hydrodynamic loading p(x,t) is represented by • The independence principle (i.e., applying only the wave com-
the Morison equation [5,10]. By letting U(x,t) denote the instanta- ponent perpendicular to the pipe) may conservatively be used
neous horizontal water particle velocity and v̇0 (x,t) the transverse for other incoming wave directions. Wave directionality and
velocity of a point on the pipe centroidal axis, the Morison load may spreading may be accounted for using a single reduction factor
be expressed as as suggested in [3].
• The load coefficients Ca , CD and CM are assumed to be character-
1   
p(x, t) = DCD (U − v̇0 ) |U − v̇0 | + D2 CM U̇ − (CM − 1) v̈0 istic of the main span and constant along the pipe length.
2 4
(4)
1    The TD algorithm derived in the following is not restricted to
≈ DCD (U|U| − 2|U|v̇0 ) + D2 CM U̇ − (CM − 1) v̈0
2 4 these simplifying assumptions, and more general expressions are
when U » v̇0 , which is a reasonable approximation for free spanning given in the next section. However, comparisons to FD results
pipelines. In Eq. (4),  is the water density, D is the hydrodynamic obtained with the DNV-RP-F105 algorithm become more transpar-
diameter of the pipe (i.e., the outer diameter including all coating ent when these key assumptions are adopted in both approaches.
layers), CD is the drag coefficient, and CM is the inertia coefficient, Observable differences will then be mainly due to simplification in
given by the FD approach regarding multi-mode response, span interaction
and linearization of the drag force.
CM = Ca + 1 , (5) Based on the listed assumptions, the drag and inertia coefficients
may be expressed as
where Ca is the added mass coefficient. The first term of Eq. (4) is
conventionally referred to as the drag force, and the second term as k  e 
s g
the inertia force. The added mass ma , representing the inertia of the CD = CD0 · CD
KC,˛ (KC, ˛) · CD
proxi
,
D D
water surrounding the pipe [5], is proportional to Ca and defined k  e  (10)
by 0 CM s CM g
CM = CM (KC, ˛) · ks
· proxi
.
D D

ma = D2 Ca . (6)
4 The Keulegan-Carpenter number (KC) and the current flow velocity
By applying Eq. (6), Eq. (4) may be rewritten as ratio ␣ for irregular wave sea-states are defined by [3,19]:

1  Us Tu Uc
p(x, t) = DCD (U|U| − 2|U|v̇0 ) + D2 CM U̇ − ma v̈0 (7) KC = and ˛= , (11)
2 4 D Uc + Us
H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147 133

Fig. 2. Schematic illustration of a multi-span section. Different soil regions identified by shoulder coordinates gk and soil stiffness factors ksoil,k .

where Us is the significant wave-induced flow velocity and Tu is the


mean zero up-crossing period of oscillating flow. Explicit expres- 
n     x     nx  
sions for the remaining quantities entering Eq. (10) are given by ix 2x
v0 (x, t) = sin Di (t) = sin sin ··· sin
L L L L
DNV-RP-F105 [3]. Note that CD may be amplified by cross-flow VIV i=1
[3], an effect which has been disregarded in this paper. KC and ␣ are ⎡D ⎤
treated as sea-state specific constants, implying that CD and CM do 1 (t)

not vary with time within a TD simulation, i.e., wake reversal effects ⎢ D2 (t) ⎥
⎢ ⎥
[20], variation in current component and span evolution have not ⎢ . ⎥ = ND, (14)
been considered. This is not a limitation of the model, however, as ⎣ . ⎦
.
the load coefficients could be updated based on the instantaneous Dn (t)
wave and current velocities without additional computational cost.
where a displacement vector D of generalized coordinates Di has
been introduced.
3. Modal analysis and time-domain algorithm The equation of motion for the pipe is derived in Appendix A
based on Eqs. (7) and (12)-(14). The result may be expressed in the
3.1. Equation of motion conventional manner as

The technical theory of beams [21] has been applied for the basic MD̈ + CḊ + KD = R (t) . (15)
displacement field:
The total system mass matrix M in Eq. (15) is given by
∂v ∂v0  L  L
u(x, y, t) = −y (x, y, t) = −y (x, t) ,
∂x ∂x (12) M= (md + mc + ma ) N T Ndx = me N T Ndx
0 0
v (x, y, t) = v0 (x, t) , L
⇒ (M)ii = me , (16)
2
where u and v are the tangential and transverse displacement
components, respectively, and t denotes time. The transverse dis- where me is the effective mass per unit length, including con-
placement of points on the centroidal axis is denoted v0 . The axial tributions from dry structural mass md , content mass mc and
strain due to bending is hence the only non-zero strain component, hydrodynamic added mass ma . The nomenclature “ii” in Eq. (16)
and given by is standard index notation, implying that M is a diagonal matrix.
The total system stiffness matrix K in Eq. (15) is given by
2
∂ v0 K = K struc + Kg + Ksoil
εxx (x, t) = −y (x, t) . (13)
∂x2   
L L L (17)
The transverse displacements are assumed to be so small that = EI NT,xx N,xx dx + Seff NT,x N,x dx + ksoil NT Ndx,
0 0 0
any change in the effective axial force Seff can be neglected. How-
ever, the centroidal axis does expand somewhat due to the lateral where E is the Young’s modulus of the pipe steel and I is the second
deflections, and the associated change in potential energy of ten- moment of area. The pipe cross-section is assumed to be uniform
sion (given as the product of Seff and the increase in length of the along the model length L. Differentiation with respect to the posi-
centroidal axis) is accounted for. tion variable x is denoted with a subscript “,x”. Hence, N,xx is the
The dynamic pipe response will extend some distance into shape function vector N from Eq. (14) differentiated twice with
the shoulder and a certain shoulder length is therefore included regard to x.
in the model, as shown in Fig. 1. Since axial and transverse The effective axial force Seff can also be approximated as uniform
degrees of freedom are decoupled, there is no dependence on the along the length of the local pipe model with small loss of accuracy
degree of axial restraint. Hence, simply supported end conditions for the in-line direction [22]. Consequently, the stiffness matrices
can be assumed at the outermost shoulder ends. The bound- Kstruc and Kg will both be diagonal:
ary conditions require that v0 (0,t) = v0 (L,t) = 0, which is fulfilled by
expressing the transverse displacement as a Fourier half-range sine   (i )
4
(i )
2
(Kstruc )ii + Kg = EI + Seff (18)
series: ii 2L3 2L
134 H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147

The expression for the soil stiffness matrix Ksoil will generally In Eq. (23), 1 comprises structural damping str and soil damping
be more complicated, since the soil stiffness coefficient ksoil will be soil , and the mass matrix M is given by Eq. (16).
equal to the lateral dynamic stiffness coefficient KL on the shoul- The load vector R(t) in Eq. (15) is given by
ders, and drop to zero in the spans. In addition, the shoulder soil  L 
stiffness may not be uniform along a multi-span section. A system- 1 
R (t) = DCD (U|U| − 2|U|v̇0 ) + D2 CM U̇ NT dx . (24)
atic description of the soil variation on the shoulders, similar to 2 4
0
the approach previously described by Sollund et al. [8], is therefore
introduced in order to establish a general expression for the result- Since the load coefficients and the water-particle velocity are
ing stiffness matrix. As illustrated by the schematic multi-span assumed independent of x, the load vector may be rewritten as
section in Fig. 2, coordinate indicators g2k-1 and g2k are associ- 1 

ated with the beginning and end of each shoulder. Whenever the R (t) = DCD U|U| + D2 CM U̇
2 4
soil stiffness changes value on a particular shoulder, the relevant
shoulder will be divided into two shoulders with a non-zero span  L  L 
in-between (reference is made to coordinates g6 and g7 in the fig- NT dx − DCD |U| NT Ndx Ḋ (25)
ure). When all shoulders, separated either by spans or by variation 0 0
1 

DCD
in soil type, have been defined, they are enumerated consecutively
= DCD U|U| + D2 CM U̇ Rc − |U|MḊ.
from left to right with the index k. 2 4 me
After performing the integration in Eq. (17), the soil stiffness
matrix may now be expressed in index notation as The vector Rc can be established prior to commencing the time-
stepping algorithm, which will be presented in Section 3.4. The i-th
i=
/ j: entry is found to be

m  g2k  ix   jx 
L
(Ksoil )ij = ksoil,k sin sin dx (Rc )i = (1 − cos (i )) . (26)
g2k−1
L L i
k=1
⎛  g2k
  g2k−1

Consequently, the load vector may in each time step be estab-
L 
m sin (i − j)  sin (i − j) 
ksoil,k ⎝
L L lished without time-consuming evaluation of integrals.
= −
2 i−j i−j
k=1
 g2k
  g
⎞ 3.2. Mode superposition analysis
sin (i + j)  sin (i + j)  2k−1

L
+
L ⎠,
i+j i+j A number of sea-states must be evaluated when calculating
the accumulated fatigue damage due to wave loading. Considering
i=j: that only a few modes are likely to contribute to fatigue damage

m  g2k  ix  even under the most severe conditions, an increase in computa-
(Ksoil )ii = ksoil,k sin2 dx tional efficiency may be obtained by applying a mode superposition
g2k−1
L
k=1 procedure. This is based on the orthogonality properties of the

m
ksoil,k
 L
  g2k
  g2k−1
 eigenvectors ␸i determined by solving the equation of motion for
= g2k −g2k−1 − sin 2i − sin 2i , undamped free vibrations of the pipe:
2 2i L L
k=1  
(19) K − ωi2 M i = 0. (27)

where m is the total number of shoulders (i.e., m = 5 in Fig. 2).The As a consequence of Betti’s law [23], the eigenvectors obey the
damping matrix C in Eq. (15) is defined by following orthogonality relations:
 L 
T 0 if i =
/ j ,
C= cNT Ndx , (20) ␾i M␾j =
Mi if i = j ,
0  (28)
T 0 if i =
/ j ,
where c is the viscous damping coefficient. For the problem at ␾i K␾j =
ωi2 Mi if i = j .
hand, mass-proportional damping seems reasonable since the
wave period will be increasingly large relative to the periods of The displacement vector D may be expressed as a linear com-
higher-order vibrational modes, thus resulting in quasi-static exci- bination of the eigenvectors. The transverse pipe displacement, Eq.
tation and negligible effect of damping for higher-order modes. (14), then becomes
Throughout the present study, the viscous damping parameter is
therefore assumed to be of the Rayleigh mass-proportional type 
r

and given by v0 (x, t) = N (x) D (t) = N (x) Z (t) = N (x) ␾j Zj (t) , (29)
j=1
c (x) = a0 me = 2 i ωi me , (21)
where Z is the vector of modal coordinates.
where a0 is a constant, i is the modal damping ratio for mode i
The displacement vector D may thus be replaced by the modal
and ωi is the angular frequency for mode i. Typically, an estimated
expansion defined by Eq. (29) in the equation of motion, Eq. (15).
damping ratio is only available for the fundamental mode, as is the
After premultiplication by ␸i T and taking note of Eq. (28), the equa-
case in DNV-RP-F105 [3], in which case
tion of motion becomes uncoupled and may be written on the form
a0 ω1
a0 = 2 1 ω1 ⇒ i = = 1 , (22) T
i R (t)
2ωi ωi Z̈i + 2 + ωi2 Zi = = R̃i (t) .
i ωi Żi (30)
Mi
and the damping matrix becomes
There will be r uncoupled equations, corresponding to the r
C=2 1 ω1 M . (23) contributing eigenvectors (r « n, where n is the number of gener-
H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147 135

alized coordinates in D). The generalized forces R̃i , based on Eq. In Eq. (35), h is the water depth and eg is again the gap between the
(25), become pipe and the seabed. For multi-spans, the characteristic value for
1 
 ␾T R the longest span is chosen. i is a randomly selected phase angle
i c uniformly distributed on the interval [0,2], and G(ωw ) is the fre-
R̃i (t) = DCD U|U| + D2 CM U̇
2 4 Mi quency transfer function from wave elevation at the sea surface to
⎛ ⎞ wave-induced flow velocity at the pipe level. The wave number ki is
DCD 
r
for each wave component determined from the dispersion relation
|U|␾i M ⎝ ␾j Żj (t)⎠ (31)
T

me Mi [3,19].
j=1 The superimposed current velocity Uc and the long-term signif-
1 

DCD icant wave height Hs are both modeled using 3-parameter Weibull
= DCD U|U| + D2 CM U̇ R̃c,i − |U|Żi (t) , distributions, with cumulative distribution function given by
2 4 me
 x − ˇW 
where R̃c,i may be calculated prior to commencing the time- FW (x) = 1 − exp − W
, (36)
stepping procedure. ˛W
Eq. (30) can be rewritten in vector-form as where x is either Uc or Hs , and ˛W is the scale, ˇW is the shape and
Z̈ + C̃Ż + ␻2 Z = R̃ (t) , (32) W is the location parameter.
In the present study, the time between independent current
where events is assumed to be 24 h, while the time between independent
⎡ ⎤
1 ω1 0 ··· 0 sea-states is assumed to be 3 h. For each long-term fatigue assess-
⎢ ···
⎥ ment, the Hs distribution is discretized into 30 subintervals and the
⎢ 0 2 ω2 0 ⎥
C̃ = 2 ⎢ ⎥ , Uc distribution into 10 subintervals, in both cases with the calcu-
⎢ .. .. .. .. ⎥ lated 100-year return-period value as the upper limit. For every
⎣ . . . . ⎦ value of Hs , 3-h simulations are performed with each of the 10
0 0 ··· r ωr different values of Uc as a constant superimposed current. The pro-
⎡ ⎤ ⎡ ⎤ cedure is repeated for 10 different sea-state realizations, giving rise
ω12 0 ··· 0 R̃1 (t) to a total of 100 3-h simulations per Hs , and 3000 simulations for a
⎢ 0 ω22
⎥ ⎢ R̃ (t) ⎥ full fatigue assessment.
⎢ ··· 0 ⎥ ⎢ 2 ⎥
␻2 = ⎢

⎥ , R̃ (t) = ⎢
⎥ ⎢ . ⎥ .
⎥ (33) Throughout the study, the peak period Tp is assumed conditional
⎣ ... ..
.
..
.
..
. ⎦ ⎣ . ⎦
. on Hs according to

0 0 ··· ωr2 R̃r (t) Tp = 7.5 · (Hs )0.3 , (37)

which is consistent with the recommendations in [3].


An important advantage of applying a mode superposition anal-
ysis is that the partial safety factors f for frequency, k for damping
and s for stress range recommended by DNV-RP-F105 [3] may 3.4. Time stepping algorithm
be adopted directly. Thus, the safety factor format in [3] can be
implemented by replacing C̃ and ␻2 with Dynamic analyses of offshore structures subjected to Morison
loading are commonly performed using the implicit and uncondi-
C̃ ␻2 tionally stable Newmark-ˇ scheme 26. A drawback with the use
C̃ = and ␻2 =  2 . (34)
k· f of implicit methods for this purpose is that iteration, due to the
f
presence of the structural velocity in the drag term, is required
The safety factor s is applied to the resulting stress ranges during to accurately predict the force vector. The widely used central
the fatigue analyses as described in Section 3.5. In addition, DNV- difference method [27], which is known to be computationally
RP-F105 recommends an allowable utilization ratio on the total inexpensive for single degree of freedom problems and prob-
fatigue damage. lems with either constant or diagonal mass and damping matrices
[28,29], is therefore preferred here. Although the method is not
3.3. Generation of sea-states and environmental modeling unconditionally stable, the preceding mode superposition analy-
sis filters away the high-frequency modes, thereby ensuring that
Stationary, irregular sea-states are simulated according to the time steps smaller than the critical step size are achievable with-
widely used JONSWAP wave spectrum [3,19]. The wave spectrum out too high computational cost. The suggested central difference
S is characterized by the significant wave height Hs , peak period approach is here adjusted to handle the hydrodynamic damping
Tp and peak-enhancement factor ␥ [3], and it is a function of the term entering R̃(t).
angular wave frequency ωw . For a given wave spectrum, the wave- After introducing finite difference expressions for the time
induced flow velocity and acceleration at the pipe level may be derivatives, Eq. (32) can be rewritten at time step k in the form
simulated according to linear (Airy) wave theory [5,24,25] by

N     
ωw,i cosh (ki · (eg + D)) (Zk+1 − 2Zk + Z k−1 ) (Zk+1 − Zk−1 )
Uw (t) = 2S ωw,i ωw,i sin ωw,i t + i + C̃
sinh (ki · h) 2
(t) 2t
i=1


N
      (Zk+1 − Zk−1 )
= G ωw,i 2S ωw,i ωw,i sin ωw,i t + i , +ω2 Z k = ak R̃c − bk , (38)
2t
i=1
 (35)
 ωw,i
N
2
cosh (ki · (eg + D))     where
U̇w (t) = 2S ωw,i ωw,i cos ωw,i t +
sinh (ki · h)
i
1  DCD,k
i=1 ak = DCD,k Uk |Uk | + D2 CM,k U̇k and bk = |Uk | .
2 4 me
 N
      (39)
= ωw,i G ωw,i 2S ωw,i ωw,i cos ωw,i t + i .
i=1
136 H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147

Solving Eq. (38) for Zk+1 yields ing fatigue life Tlife are calculated according to the Palmgren-Miner
rule [3,33]:
Zk+1 = K−1 R
eff,k eff,k
, (40) n
i
Tlife = where Dyear = , (44)
Dyear Ni
in which i

1 1   and ni is the annual number of cycles with stress range Si . The


K eff,k= 2
I+ C̃ + bk I , allowable utilization ratio is 0.5 for safety class “normal” [3].
(t) 2t
 
2 4. Stochastic frequency-domain analysis
R eff,k = ak R̃c − ␻2 − 2
I Zk (41)
(t)
  Since the wave loading on a structure over its design life is
1 1   specified using probabilistic concepts, a stochastic FD-based fatigue
− 2
I− C̃ + bk I Zk−1 ,
(t) 2t analysis is useful and computationally advantageous. The stochas-
tic analysis method currently implemented in DNV-RP-F105 [3] is
and I is the (r × r) identity matrix. Assuming the initial conditions based on a linearization of the Morison equation and an assumption
at k = 0 are known, it is clear from Eq. (41), that Z-1 is required for of only one contributing vibration mode. While the latter assump-
calculation of Z1 . From the finite difference derivative expressions tion is likely appropriate for conventional single-span analyses, it is
for k = 0, it is trivial to obtain more questionable for multi-spans which may have eigenfrequen-
cies that are not well separated [8]. In this section, the derivation
2
(t) of the response spectrum SSS of wave-induced stresses is outlined,
Z−1 = Z0 − t Ż0 + Z̈0 . (42)
2 and a novel expression is proposed to account for the contribu-
tion of higher-order modes. The new expression for SSS will later
The inversion of Keff ,k in each time step can be avoided by using be evaluated for use in multi-span analyses.
(Zk − Zk-1 )/2 as the time-derivative in the hydrodynamic damping A response process is Gaussian, and thus completely described
term, as suggested by Rio et al. [29]. However, since Keff ,k is diagonal by the response spectrum, if the forcing processes are Gaussian
in the present case, the computational gains would be small. It is [23]. However, even though the wave-induced flow velocity Uw and
therefore preferred to retain second-order accuracy in the time- acceleration U̇w can be considered Gaussian [34], the wave-induced
derivative expression. stress response cannot due to the non-linearity of the drag force
A time-step t between 1/10 and 1/20 of the period T associ- [11,12,24]. The drag term in Morison’s equation must hence be lin-
ated with the highest considered mode was found to give adequate earized to facilitate a stochastic frequency-domain analysis. The
accuracy (the critical step size with regard to numerical stability is linearized equation of motion will therefore be derived, and used
t < T/ [27]). The algorithm was implemented in Matlab [30]. The as a basis for the subsequent derivation of the response spectrum
computational time for a single 3-h sea-state would typically vary SSS .
from ∼1 s to ∼30 s depending on the step size, which is considered The eigenmodes ϕj (x) are defined by
suitable for the present purpose.

r

r
j (x) = N (x) j ⇒ v0 (x, t) = N (x) j Zj (t) = j (x) Zj (t) . (45)
3.5. Fatigue calculations j=1 j=1

Pipelines are typically assembled from 12.2-m (40-ft) long pipe By utilizing the definition of ϕi (x) above, it is readily observed
joints connected by a circumferential (girth) weld at each end. In that Eq. (30) can be expressed as
free span design, it is conservatively assumed that a weld is located 1 
   (x) dx
2 L i
at the position of maximum fatigue damage. Thus, the fatigue dam- Z̈i + 2 i ωi Żi + ωi Zi = DCD U|U| + D2 CM U̇
age resistance is assumed equal to the resistance of the girth weld 2 4 Mi
[3]. DCD
− |U|Żi . (46)
The cyclic axial bending stresses (x,t) due to wave-induced me
transverse displacements occur perpendicularly to the pipe-joint
girth welds and may thus propagate potential cracks or weld It is further observed that the modal mass Mi may be written
defects. The mid-wall bending stresses, as applied in [3], are  L 
obtained from T T
Mi = ␾i M␾i = ␾i (md + mc + ma ) NT Ndx
D − t  0
s s
(x, t) = E
2
N ,xx (x) ˚Z (t)  L
␾i = me i2 (x) dx , (47)
D − t  
r 
r
0
s s
=E N ,xx (x) j Zj (t) = j (x) Zj (t) , (43)
2 where the effective mass me is given by
j=1 j=1
L
0
(md + mc + ma ) i2 (x) dx
where E is the Young’s modulus, Ds is the outer steel diameter, ts is me = L . (48)
the wall thickness and j (x) is the mid-wall modal bending stress. 0
i2 (x) dx
Stress cycles and associated stress ranges Si are extracted from
A mode shape weighting factor is defined by
the bending stress history using a rainflow counting procedure

[31,32]. The stress ranges are subsequently multiplied by the safety
factor s and the number of cycles to failure Ni for each stress range i (x)dx
is estimated using an appropriate SN-curve. In pipeline design, the F i = L , (49)
and D class SN-curves are often used for the weld root and weld cap, i2 (x)dx
respectively [1,33]. Annual fatigue damage Dyear and correspond- L
H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147 137

and for convenience of notation, we introduce the quantities where j (x) is the modal bending stress defined by Eq. (43) and
Hj (ωw ) is the complex frequency response function for mode j given
1  by
gD = DCD ∧ gI = D2 CM . (50)
2 4
1
After linearizing the quadratic drag term and using Eqs. Hj (ωw ) =   where i2 = −1 . (59)
(47)–(50) to simplify the expression, Eq. (46) may be rewritten as Mj ωj2 2
− ωw +2 T,j ωj ωw ·i

  i 2gD |U| The cross-spectral density function SPjPk (ωw ) is defined by [23]
Z̈i + 2 i ωi Żi + ωi2 Zi = bgD U + gI U̇ − Żi . (51)
me me  L L
SPjPk (ωw ) = j (x1 ) k (x2 ) Sp (x1 , x2 , ωw ) dx1 dx2 . (60)
The linearization factor b must be derived based on a suitable 0 0
optimality criterion. Commonly, b is estimated by minimizing the
In Eq. (60), Sp (x1 ,x2 ,ωw ) is the spectral density function for the
mean square of the error in the drag load [12,24,35]. However, Mørk
linearized Morison load. Under the assumption that the load is
and Fyrileiv [11] have presented an alternative expression based on
location-invariant, Sp becomes
minimization of the error in the predicted fatigue damage. Their
 
expression has been incorporated in DNV-RP-F105 [3], and is given Sp (x1 , x2 , ωw ) ≈ Sp (ωw ) = b2 gD2 + ωw
2 2
gI SUU (ωw )
by  
U  = b2 gD2 + ωw gI G2 (ωw ) S
2 2
(ωw ) , (61)
c
b = 2.11 · U · gc , (52) where SUU (ωw ) is the wave-induced flow velocity spectrum at pipe
U
level, G(ωw ) is the transfer function defined by Eq. (35) and S (ωw )
where  U = Us /2 is the standard deviation of the wave-induced flow is the wave spectrum. When Eqs. (60) and (61) are inserted into
velocity. The derivation of b is based on an assumption of a Gaus- Eq. (58) the complete expression for the one-sided stress response
sian, zero-mean flow velocity. The function gc is a correction to spectrum SSS (x,ωw ) is obtained.
account for the effect of a steady current [35]. It can be expressed Assuming that only the fundamental mode contributes to
as fatigue damage, Eq. (58) is reduced to
U   U   U    
c √ c Uc c 1 SSS (x, ωw ) = b2 gD2 + ωw
2 2
gI G2 (ωw ) S (ωw )
gc = 2 ϑ +  − ;
U U U U 2
21 2
(x)
z2  z
(53)  
1
 2  , (62)
1 − m2e ω12 − ωw 2 2 + 2
ϑ (z) = √ e 2 and  (z) = ∗
ϑ (z ) dz . ∗ T,1 ω1 ωw
2 −∞
which is the expression recommended by DNV-RP-F105 [3]. In Eq.
The absolute value of the water particle velocity in Eq. (51) may (62), 1 is the mode shape weighting factor (Eq. (49)) for the fun-
be replaced by the expectation value E[|Uw |] of the wave-induced damental mode.
velocity [24]. After correcting for the effect of a steady current, the An extension of Eq. (62) to account for higher-order modes was
expression is given by proposed by Mørk and Fyrileiv [11] based on the assumption that

U   U 
the cross-terms in Eq. (58) are negligible (which is often justified
c 2 c in the case of well separated eigenfrequencies and light damping
|U| ≈ E [|Uw |] · gc = U · gc . (54)
U  U [23]):
 
SSS (x, ωw ) = b2 gD2 + ωw
2 2
gI G2 (ωw ) S (ωw )
A hydrodynamic damping ratio can then be defined by requiring
that  2j 2
(x)
2gD |U| gD |U|
U   2
j
 . (63)
2 h,i ωi = ⇒ = ≈ √
1 U gD
gc
c
. (55)  2
me h,i
me ωi  2 me fi U j m2e ωj2 − ωw
2 + 2 T,j ωj ωw

By inserting the total damping ratio T ,i = i + h,i into Eq. (51), The response spectrum expression derived by Xu et al. [12] is sim-
the final, linearized form of the equation of motion becomes ilarly based on an assumption of negligible cross-modal terms.
  i In the present study, the influence of the cross-terms
Z̈i + 2 T,i ωi Żi + ωi2 Zi = bgD U + gI U̇ . (56) will be investigated. For increased transparency and ease-of-
me
implementation, Eq. (58) is rewritten as
For the derivation of the response spectrum, it is also useful to  
SSS (x, ωw ) = b2 gD2 + ωw
2 2
gI G2 (ωw ) S (ωw )
introduce the modal load Pi by
 j Tjk − i · Ujk
 j (x) k k (x)
P   L · , (64)
Z̈i + 2 T,i ωi Żi + ωi2 Zi = i ⇒ Pi = bgD U + gI U̇ i (x) dx, m2e 2 + U2
Tjk jk
Mi j k
0
(57) where we have introduced
 
where we have utilized Eq. (47).We will assume that more than Tjk = ωj2 ωk2 − ωw
2 ωj2 + ωk2 4 +4
+ ωw 2
T,j T,k ωj ωk ωw ,
one modal load Pi will contribute to fatigue damage. Thus, the total     (65)
stress response spectrum SSS (x,ωw ) should be obtained from a sum Ujk = 2 T,j ωw ωj ωk2 − ωw
2 −2
T,k ωw ωk ωj2 − ωw
2 .
of the cross-spectral density functions SPjPk (ωw ) for the modal loads
Pj and Pk [23], and can be written as It is noted that
 2  2
 Tjk = Tkj and Tjj = ωj2 − ωw
2 + 2
∗ T,j ωj ωw ,
SSS (x, ωw ) = j (x) k (x) Hj (ωw ) Hk (ωw ) SPjPk (ωw ) , (58) (66)
j k Ujk = −Ukj and Ujj = 0 .
138 H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147

Thus, it is obvious that the imaginary terms in Eq. (64) cancel unknown pipe velocity. The pipe velocity time series is there-
each other out and that Eq. (64) reduces to Eq. (63) if the cross-terms fore extracted from a preceding semi-analytical analysis, and for
are disregarded. each element the magnitude of the load is prescribed based on
When the response spectrum has been established, the response the average velocity at its two nodal positions. Thus, the loading
spectral moments MS ,i can be calculated by prescribed in the TD FEA is identical to the loading applied in the
 ∞ semi-analytical analyses.
MS,i = (ωw )i SSS (x, ωw ) dωw . (67) The TD FEA are performed by means of direct integration, i.e.,
0 no mode superposition analysis is carried out. Abaqus uses implicit
Hilber-Hughes-Taylor time integration (also called the ␣-method),
Specifically, it follows from the Wiener-Khintchine relations
which is an extension of the widely used Newmark ˇ-method that
[23,25] that the standard deviation of the stress response is given
allows algorithmic damping of high-frequency modes while retain-
by
ing second-order accuracy and unconditional stability [26]. The
 default choices of ␣ = −0.05, ˇ = ¼(1 − ␣)2 and = ½(1 − 2␣) are
 ∞
S = MS,0 = SSS (x, ωw ) dωw . (68) used.
0 Crucially, the NLGEOM (non-linear geometry) option is turned
on throughout the FEA, implying that the tangent stiffness matrix
DNV-RP-F105 [3] gives an expression to calculate the fatigue life in each step of the analyses is based on the current deformed con-
as a function of the spectral moments Ms,i for the stress response figuration. Hence, all non-linear geometric effects are accounted
and the relevant SN-curve parameters. The expression, which is for.
omitted here for brevity, is based on the assumption that the
stress response is a narrow-banded Gaussian process. The inter-
ested reader is referred to e.g. Naess and Moan [25] for a derivation 6. Results and discussions
of the expression. DNV-RP-F105 includes an empirical correction
factor, originally proposed by Wirsching and Light [36], to account 6.1. Comparison to finite element analyses
for wide-banded damage (i.e., that the response spectrum covers
a wide range of frequencies). This correction factor has also been TD simulations were carried out for two free span scenarios
included in all FD calculations in the present work. using both the semi-analytical TD algorithm and non-linear FEA.
It can be remarked that when Eq. (63) is used for the response The results are compared in the present section in order to verify
spectrum, the standard deviation of the response, Eq. (68), may be that the semi-analytical TD algorithm has been correctly derived
written as and implemented, as well as to illustrate important aspects of the
 wave-induced dynamic response of free spanning pipelines.
S = 2 ,
S,j (69) A pipeline with a single span is considered first. The selected
j span geometry, later referred to as “Case 1”, is illustrated in Fig. 3a).
The relative span length is L1 /Ds = 75, and the relative shoulder
where  S,j is the standard deviation for mode j. Thus, Eq. (63) lengths are Lshoulder /Ds = 100. The gap is eg = 2 m, and the water
can be recognized as a “square-root of the sum of squares” (SRSS) depth is h = 90 m. The pipe cross-section is a 40-inch gas pipeline
approach, while Eq. (64) may be termed a “complete quadratic com- with a 75-mm concrete coating (Pipe 1 in Table 1) and an effec-
bination” (CQC). Conventionally, the terms SRSS and CQC have been tive axial force Seff = −1 MN (compressive). The chosen soil stiffness
used in the context of earthquake engineering, where the influ- corresponds to stiff clay, with values selected according to DNV-RP-
ence of the cross-modal terms on prediction of maximum responses F105 [3] and listed in Table 2.
has been recognized for some decades [37]. To our knowledge, the Only transverse displacements are considered in the semi-
present study is the first to investigate cross-modal contributions analytical analysis and Seff is assumed to be constant. By contrast,
to wave-induced fatigue of submarine pipelines. the stiffness matrix is updated in each time step of the non-
linear FEA, resulting in the introduction of axial stiffness and
5. Finite element analyses displacement components when the pipe elements rotate. Axial
soil stiffness is therefore included in the FEA model (Table 2). More-
For verification of the presented TD algorithm, TD finite element over, Seff is allowed to change when the pipe elongates, resulting
analyses (FEA) are carried out using the commercially available in additional sensitivity to the degree of axial restraint. In order to
software Abaqus [38]. To distinguish between the FEA and the TD prevent the axially fixed boundary conditions from affecting the
algorithm developed in Section 3, the latter will be termed “semi- non-linear FEA, the shoulders are taken longer than required for
analytical analysis” since it is based on a semi-analytical modal convergence of eigenfrequencies and associated modal stresses.
analysis method. In the FEA, the pipeline is modeled using first- It is natural to first examine the outcome of the modal analysis.
order shear deformable PIPE31 elements. An element length of 0.1 Since the dominant wave frequencies are generally much lower
Ds is chosen. The element length was selected based on conver- than the pipeline eigenfrequencies, it is expected that the lowest
gence studies of modal response [8,13], as well as on convergence eigenmodes contribute most to the pipe response. However, the
of bending stress time histories for a pinned pipeline subjected to importance of including a particular mode in the fatigue analysis is
regular waves. Mass-proportional damping is specified according dictated not only by its eigenvalue (ωi 2 ), but also by its shape (ϕi ).
to Eq. (23). Linear springs are used to model the span shoulders, To illustrate this, we will discuss the mode shapes for the first three
both in the lateral and in the axial direction. When applying the eigenmodes obtained by the FEA and the semi-analytical method,
effective axial force, the springs are removed and the node at the which are displayed in Fig. 3b). The fundamental and third modes
right end of the model is kept temporarily free. When the pipe end are observed to be symmetric relative to the mid-span plane, while
is fixed against translation, the springs are reintroduced strain free. the second mode is anti-symmetric. These symmetry properties are
A modal analysis is also carried out prior to the TD FEA simulation. expected for an initially straight single-span pipe. However, since
In the TD FEA, the hydrodynamic loading is applied as a dis- the hydrodynamic loading is constant along the span, and hence
tributed load with the magnitude prescribed in each time step. The symmetric, the loading will not excite the anti-symmetric sec-
hydrodynamic damping term obviously depends on the a priori ond mode. This can also be deduced from the corresponding mode
H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147 139

Fig. 3. (a) Single-span. (b) Mode shapes from FEA and semi-analytical method. (c) Modal stresses.

Table 1
Pipe cross-sections used in the present work.

Description Pipe 1 Pipe 2 Pipe 3 Unit

D Hydrodynamic diameter 1.1806 0.5338 0.806 m


Ds Outer steel diameter 1.0186 0.5238 0.720 m
ts Steel thickness 0.0261 0.025 0.0231 m
steel Density of steel 7850 7850 7850 kg/m3
tcoat Coating thickness 0.006 0.005 0.003 m
coat Coating density 1300 935.68 1300 kg/m3
tconc Concrete thickness 0.075 0 0.040 m
conc Concrete density 2850 0 2450 kg/m3
cont Content density 180 205 195 kg/m3
me Effective mass 2662.66 581.28 1235.23 kg/m
E Young’s modulus 2.07.1011 2.07.1011 2.07.1011 N/m2

Table 2
Soil types used in the present work.

Description Stiff clay, Very soft clay, Case 2 Medium sand, Case 3 Unit
Case 1

KL Lateral dynamic 7653 716.54 18759 kN/m2


KAX Axial dynamic (only FEA) 7653 716.54 – kN/m2
soil Damping ratio 0.0175 0.0121 0.015 –

shape weighting factors i , which are found to be 1.32, 0 and 0.563, In order to compare the semi-analytical TD approach to TD FEA, a
respectively, when calculated according to Eq. (49). As observed simulation was conducted for a 1-h sea-state of irregular waves cor-
from Eq. (56), there will be no dynamic response if i is zero. The responding to Hs = 8 m. For the verification study, the superimposed
strong dependence on i is also reflected in the expressions for the current velocity was changed every 10th minute (with a 1-min lin-
stress response spectrum, Eqs. (62)–(64). ear interpolation between velocities), and the current flow velocity
Fig. 3c) displays the associated unit diameter amplitude stresses, ratio ␣ updated accordingly. This was done to test the capability of
defined as AIL,i = −D· i (x). It is observed that the symmetric modes the algorithm to include time-varying drag and inertia coefficients.
have modal stress peaks that interact on the shoulders and mid- The velocities were sampled according to the Weibull distribution
span. However, the stress peaks have the same sign on the for Case 1 in Table 3. The semi-analytical analysis was performed
shoulders, and opposite sign mid-span. Consequently, if the third with t = 0.05 s for 1, 2 and 3 modes. Rainflow counting was per-
mode adds non-negligibly to the fatigue damage induced by the formed for 1500 locations along the pipe axis, and the resulting
fundamental mode, the cross-modal terms will increase the dam- annual fatigue damage is plotted in Fig. 4a). The fatigue life at a
age on the shoulders and reduce the damage at mid-span. particular location may be obtained by taking the inverse of the
Importantly, Fig. 3b) and c) also show that the mode shapes plotted value (Eq. (44)).
and stresses obtained by FEA and semi-analytical analysis match Most notably, it is found that by including the third mode, the
perfectly. This is in agreement with previous observations [8,13]. maximum damage on the shoulders (critical location 1) increases
140 H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147

Table 3
Weibull parameters for current velocity and significant wave height at two selected locations.

Shape Scale Location

Case 1 & 2 Current velocity Uc 1.260 0.087 m/s 0 m/s


Significant wave height Hs 1.343 2.057 m 0.939 m
Case 3 Current velocity Uc 1.580 0.120 m/s 0.010 m/s
Significant wave height Hs 1.900 2.810 m 0.500 m

Fig. 4. (a) Annual fatigue damage along Case 1 pipe axis (only semi-an.). (b) Bending stress time series at critical location 1. (c) Bending stress time series at critical location
2.

by 11.9%. Thus, the influence of including higher-order modes, where a large number of spans and sea-states generally must be
although modest compared to the large degree of uncertainty asso- examined.
ciated with fatigue estimations, is not entirely negligible even for Another comparison between TD FEA and semi-analytical anal-
a typical single span. As expected, the second mode does not con- yses was performed for the Case 2 double-span depicted in
tribute to fatigue damage, and so the curves for 1 and 2 modes Fig. 5a). This time, a 20-inch pipe is used (Pipe 2, Table 1), with
in Fig. 4a) cannot be distinguished. By including the third mode, Seff = −200 kn. The two spans, with relative lengths of 145 and 150,
the fatigue damage decreases at mid-span (critical location 2) by are both quite long, and they are separated by a short intermediate
23%, which agrees well with the observation of oppositely directed shoulder with Lint ,1 /Ds = 15. The chosen soil stiffness is representa-
modal stresses in Fig. 3c). However, such a result would not be tive for very soft clay (Table 2).
expected from Eq. (63), which indicates that fatigue damage always Mode shapes and modal stresses for the first two modes are
increases when adding more modes. By contrast, the cross-terms plotted in Fig. 5b) and c). Semi-analytical and FEA results are seen
in Eq. (64) will be negative and thus reduce the predicted fatigue to coincide also in this case. A strong degree of interaction between
damage when the modal stresses have opposite signs. the spans is observed, and it is noteworthy that the fundamen-
An FE simulation with t = 0.2 s was performed in order to tal mode becomes almost anti-symmetric about the intermediate
confirm the semi-analytical results. Bending stress time series are shoulder. By contrast, the second mode has positive peaks in both
compared for critical locations both at the shoulder (critical loca- spans. As a result, a much larger mode shape weighting factor is
tion 1, Fig. 4b) and at mid-span (critical location 2, Fig. 4c). The FEA obtained for mode 2 (2 = 1.48) than for mode 1 (1 = 0.216). Fur-
and semi-analytical time series are seen to be in excellent agree- thermore, the frequencies of the two modes are found to be quite
ment at both locations. The two analysis methods also predict the close (f1 = 0.228 Hz and f2 = 0.285 Hz) since the two spans are of
same maximum bending stress (22.6 MPa, Fig. 4b). Both time series, nearly equal length. Consequently, it is likely that the second mode
and in particular that in Fig. 4c (shown from t = 3000 s), demonstrate will give the dominating fatigue damage contribution.
that the two analyses do not drift apart with time due to for instance A 1-h simulation was again performed using the same Hs as in
period elongation. Case 1 (Fig. 3a), but with the water depth reduced to 50 m. The sim-
On the shoulder, the fatigue lives estimated by the FEA and ulations with the semi-analytical TD method were performed using
the semi-analytical analysis are found to be 21.3 and 21.2 years, 1, 2 and 6 modes, where the sixth mode corresponds to the second
respectively. The corresponding results at mid-span are 45.8 and symmetric mode for the shortest span. The time step ranged from
48.9 years. The predictions by the two methods are close, and must 0.2 s to 0.05 s. Note that the time step is set relative to the period of
be considered to be acceptable. The discrepancy in the latter case the highest considered mode (see discussion in Section 3.4), and a
is likely due to a small influence of the fifth mode, combined with reduction in time-step is therefore necessary when including more
some inaccuracy due to a fairly large time step in the FEA. A high- modes. The annual fatigue damage is plotted as a function of posi-
resolution semi-analytical analysis with 15 modes and t = 0.005 s tion x in Fig. 6a). Results are only shown for 2 and 6 modes, since
gave Tlife = 47.0 years. the fatigue damage induced by the fundamental mode alone would
Note that the choice of time-step in the FEA is limited by a long barely be visible in the figure. The maximum damage for the fun-
computational time. The entire analysis for Case 1 took 14 h. In com- damental mode alone was less than 1% of the damage observed
parison, the semi-analytical analysis with 3 modes took only 2.1 s, when including the second mode. For 6 versus 2 modes, the results
plus about 1.5 s for generation of sea-state and rainflow counting. are reminiscent of those in Case 1, with increased damage on the
Hence, the required computational effort for the proposed TD algo- shoulders and less damage in the middle of the spans when more
rithm is small, making it suitable also in free span design practice, modes are included. The fatigue lives for the three semi-analytical
H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147 141

Fig. 5. (a) Double-span. (b) Mode shapes from FEA and semi-analytical method. (c) Modal stresses.

Fig. 6. (a) Annual fatigue damage along Case 2 pipe axis (only semi-an.). (b) Bending stress time series at critical location.

simulations (1, 2 and 6 modes) are 44.4, 0.377 and 0.324 years, 6.2. Comparison to frequency-domain analyses
respectively.
The bending stresses from the semi-analytical analysis with 6 It is also of interest to compare the semi-analytical TD results for
modes, extracted at the most critical location, are compared to Case 1 and 2 to results of frequency-domain (FD) analyses carried
bending stresses from non-linear FEA in Fig. 6b). As in Case 1, it is out according to DNV-RP-F105 [3]. If only the fundamental mode is
observed that the results are in excellent agreement. The FEA was considered in accordance with Eq. (62), the fatigue lives computed
carried out with t = 0.2 s, and the obtained fatigue life is 0.325 by the FD analyses become 23.1 years for Case 1 and 69.2 years for
years, which is virtually identical to the semi-analytical result. Case 2. These values correspond reasonably well with the single-
Together, the Case 1 and 2 comparisons strongly suggest that mode TD results of 23.7 and 44.4 years. The discrepancy in the latter
the TD algorithm developed in Section 3 has been correctly imple- case is likely predominantly due to a severe sea-state realization
mented. Minimal differences are observed between the results of in the TD analysis. The TD results are sensitive to the randomly
the two approaches, even for scenarios such as Case 2 with harsh selected set of phase angles for the generated waves (Eq. (35)), and
loading conditions and large flexibility in the spans. This also sug- the deviation would probably decrease if the TD fatigue life was
gests that the non-linear geometric effects which were included based on an average of several sea-state realizations.
in the non-linear FEA, but disregarded in the semi-analytical More importantly, the fatigue life for Case 2, as predicted by the
approach, have affected the analyses negligibly. Consequently, the semi-analytical TD analysis, drops to 0.324 years when including 6
simplifications in the semi-analytical TD algorithm appear to give modes. This clearly demonstrates that the fatigue calculation can-
negligible loss of accuracy for most cases, although an impact not be based on the fundamental mode alone if the modal analysis
of non-linear effects in particularly severe sea-states cannot be accounts for multi-span interaction. This observation is important.
excluded. It warrants the conclusion that if a modal analysis is carried out for
a multi-span and wave-induced fatigue assessments are conducted
142 H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147

Fig. 7. (a) Case 3 double-span (b) Mode shapes for mode 1 and 2. Critical location indicated by black line. (c) Unit diameter amplitude stresses. (d) Annual fatigue damage
from semi-an. TD analysis along Case 3 pipe axis.

uncritically according to the method in DNV-RP-F105 [3], there is Case 2 pipeline experienced quite large displacements and was of
a risk of obtaining a fatigue life estimate that is non-conservative interest for comparisons with non-linear TD FEA. However, for the
by two orders of magnitude. The reason for such undesired results remainder of the study it is more appropriate to examine a double-
is the assumption of single-span behavior, which is inherent in the span predominantly subjected to wave-induced fatigue. We retain
design code methodology and clearly not valid for Case 2. However, a pipeline with two spans of similar lengths, and the selected span
there is currently very limited guidance on how to cope with situ- dimensions for this case, labeled Case 3, are shown in Fig. 7a). A
ations where several modes may be responding, particularly when 30-inch, concrete-coated pipe cross-section (Pipe 3, Table 1) with
the fundamental mode is not the most severe. It is a primary objec- Seff = −500 kN is selected. The gap is eg = 1m, the water depth is
tive of the present study to determine how to safely handle such h = 60 m and the soil is taken as medium sand (Table 3). The mode
cases. shapes and associated modal stresses for mode 1 and 2 are dis-
It should be remarked that all safety factors were set to unity played in Fig. 7b) and c) and seen to be quite similar to those in
in the calculations above. The effect of including safety factors is Case 2.
discussed later. The FD calculations were also based on the semi- The annual fatigue damage of the pipeline, when subjected
analytical modal analysis rather than the modal response formulae to a 1-h sea-state with Hs = 7 m, was calculated using the semi-
in [3] in order to avoid that inaccuracies in the modal analyses analytical TD approach. Results along the pipe axis are plotted in
obscured the comparisons. This applies throughout the study. Fig. 7d). For the fundamental mode alone, the fatigue life is found
to be 193 years. A substantially smaller fatigue life of 44.2 years
6.3. Effect of considering the most severe mode only is obtained when including the second mode, and for 10 modes it
converges to 36.5 years. Thus, the second mode again provides the
While direct wave action was the dominant source of fatigue dominating damage contribution.
for Case 1 (Fig. 3a), the pipeline in Case 2 (Fig. 5a) would have A fatigue calculation was performed also for the second mode
experienced onset of cross-flow VIV with associated drag ampli- alone (“Mode 2 only”, Fig. 7d). From the results, a few important
fication. VIV has, however, been disregarded in this study. The observations can be made. Firstly, it is not trivial to infer from the
H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147 143

Fig. 8. Annual fatigue damage as function of Hs . (a) Case 1. (b) Case 3.

damage distributions of the individual modes 1 and 2 how the dam- For the Case 1 single-span, the FD solution agrees reasonably
age distribution becomes for the modes combined. In particular, it well with the single-mode TD solution (Fig. 8a). The TD solution
is seen that the damage from mode 1 and 2 combined (“2 modes”, predicts slightly more damage for Hs < 7, while the TD and FD esti-
Fig. 7d) along the shortest span (60m < x < 120 m) is much smaller mates correspond excellently for Hs > 7. The integrated TD fatigue
than the damage induced by mode 2 alone, which indicates that life is found to be 446 ± 4 years, which is 5.7% smaller than the FD
the signs of the modal stresses (Fig. 7c) influence the analysis. Sec- result of 473 years. The most accurate approach, the 3-mode TD
ondly, the results show that a single-mode assessment based on solution, is observed to generally lie above the FD curve, giving rise
the dominant mode (in this case mode 2) does not accurately rep- to a fatigue life of 399 ± 4 years. Hence, the results are consistent
resent the multi-mode fatigue damage distribution, although the with the previous findings for a single sea-state, indicating that
maximum value for this particular case is reasonably close to the the FD solution agrees well with a single-mode TD solution, but
corresponding value for 10 modes. that higher-order modes may give some additional damage. Calcu-
The significance of the latter observation is that a free spanning lations were also performed with safety factors corresponding to
pipe with multiple responding modes cannot be reliably assessed safety class “normal” and “well defined” span type ( = 0.5, f = 1.1,
by a single-mode calculation – such as Eq. (62) – even if the calcula- s = 1.3, k = 1.15) [3]. The changes in the FD and the 3-mode TD
tion is based on the most severe mode rather than the fundamental fatigue lives were similar, with the former becoming 59.3 years
mode. and the latter 49.2 ± 0.6 years.
For the Case 3 double-span, the fundamental mode has an anti-
symmetric shape. The fatigue calculations based solely on this
6.4. Fatigue assessments for long-term sea-state distributions
mode are highly non-conservative (Fig. 8b), which is in agreement
with the previous observation for a single sea-state (Fig. 7d). It is
The comparisons between semi-analytical TD results and FD
seen, however, that the mode 1 TD and FD calculations give virtu-
results discussed in Section 6.2 were based on analyses of single
ally identical results, with the TD fatigue life being 0.2% higher than
sea-states. However, full fatigue analyses must consider the entire
the FD estimate of 2227 years.
long-term distribution of sea-states representative for a pipeline’s
In practical free-span design, it is common to simply ignore
geographic location. TD and FD results will now be compared for
span interaction and analyze the main span as a single-span.
two full fatigue assessments.
Since the fundamental mode for the single-span is symmetric, this
The analyses were carried out for Case 1 and 3 using Weibull
approach (“FD, single-span”, Fig. 8b) is found to yield a more rea-
probability density functions for waves and current representative
sonable fatigue life estimate of 555 years. In comparison, the full
for two separate North Sea locations. The Weibull parameters are
TD approach accounting for multi-span behavior with 10 modes
specified in Table 3. In each case, the fatigue life is obtained by
gives 422 ± 5 years, corresponding to a 31.5% increase in fatigue
1 damage relative to the single-span FD analysis. If safety factors are
Tlife = ∞ , (70) included, the FD fatigue life drops to 70.8 years, and the 10-mode TD
Dyear (Hs ) · p (Hs ) dHs
0 fatigue life to 53.4 ± 0.4 years. The FD and TD fatigue lives are both
reduced by about 87% by the introduction of safety factors. Thus,
where Dyear (Hs ) is the annual fatigue damage as a function of signif-
the obtained difference between the conventional single-span FD
icant wave height Hs and p(Hs ) is the probability density function.
solution and the full TD solution is much smaller than the safety
The weighted annual fatigue damage distribution (i.e., the inte-
margin. A TD single-span solution, accounting for 5 modes, is also
grand in the denominator of Eq. (70)) is plotted versus Hs for Case 1
included in Fig. 8b). It generally lies above the FD single-span curve
in Fig. 8a) and Case 3 in Fig. 8b). The figures compare conventional
and gives a 12.5% smaller fatigue life, consistent with the results for
fundamental-mode FD solutions, Eq. (62), with single- and multi-
Case 1 (Fig. 8a).
mode TD solutions at the most critical location, established from
It is stressed that the notation Tlife ± st.dev. reflects only the
Fig. 4a) and Fig. 7d). Each data point on the TD curves represents
spread due to random sea-state realizations. It is not intended
an average of ten 3-h sea-states. Standard deviations are shown as
to account for the total model uncertainty, which is much larger
error bars (standard errors were hardly detectable due to the long
and should be evaluated by sensitivity analyses on a case-to-case
simulation time). In addition, the long-term distribution of current
basis.
velocities has been discretized for each sea-state and superimposed
as described in Section 3.3.
144 H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147

Fig. 9. Normalized damage versus KC and ␣. (a) TD solution for 1 mode. (b) TD solution for 3 modes.

6.5. Evaluation of FD solution as function of flow regime fundamental-mode assumption does not in general hold for multi-
span pipelines. It is thus of interest to expand the FD approach
Even when the FD and TD solutions are based on the same modal to account for several modes. In previous studies [11,12] such an
analysis and both account for the fundamental mode only, there expansion has been suggested using the “square root of the sum
are differences between the results. For example, the FD solution of squares” (SRSS) approach represented by Eq. (63). The approach
for Case 1 was observed to be slightly non-conservative as com- disregards the cross-terms in the “complete quadratic combina-
pared to the single-mode TD solution. The differences stem partly tion” (CQC) expression, Eq. (64).
from the linearization of the Morison load, Eq. (56), and partly from In Fig. 10a), the fundamental-mode FD and 3-mode TD results
the semi-empirical correction for wide-banded damage in the FD from Fig. 4a) (Case 1) are replotted along with 3-mode FD solutions
solution [3,11]. Obviously, the inaccuracy due to linearized drag based on both Eqs. (63) and (64). Interestingly, the SRSS solu-
loading will depend on the flow regime, as characterized by the tion cannot be distinguished from the fundamental-mode solution.
␣ and KC parameters (Eq. (11)). Generally, wave-induced dam- Also, the correspondence is excellent between the CQC and TD solu-
age contributes significantly to fatigue accumulation for ␣ in the tions. Hence, it is the cross-terms in Eq. (64) that account for the
range 0.1–0.4 and KC < 20. The peak damage was associated with whole effect of including the third mode.
␣ = 0.24 and KC = 3.0 for Case 1 (Hs ≈ 5.8, Fig. 8a) and with ␣ = 0.21 The CQC solution was also used to repeat the fatigue assess-
and KC = 6.4 for Case 3 (Hs ≈ 5.2, Fig. 8b). ment for the long-term sea-state distribution previously shown in
A set of fatigue analyses was carried out to identify how the lin- Fig. 8a). The results are shown in Fig. 10b). Again the CQC FD algo-
earized FD scheme compares to TD analyses with non-linear drag rithm is observed to agree well with the TD solution. The integrated
loading as a function of ␣ and KC. Case 1 was used as a basis for FD fatigue life is found to be 422 years, which is less than 6% higher
the comparison, with Hs adjusted until the desired KC value was than the TD estimate of 399 ± 4 years.
obtained and a constant Uc superimposed to achieve the relevant Fig. 10c) and d) show the results of a completely similar inves-
current ratio. Fatigue damage results for the single-mode FD solu- tigation for Case 3. In this case, the fundamental-mode FD curve
tion, given relative to a corresponding TD solution, are displayed for fatigue damage along the pipe axis is very different from the
in Fig. 9a). It is seen that the FD and TD results are similar in the SRSS solution for 10 modes. The reason for the strong difference is
most important ␣-range below 0.4. The FD solution is mostly within the dominance of the second mode. By comparing Fig. 10c) with
±10% of the TD results in this range. Interestingly, the FD results are Fig. 7d) it is seen how the resulting damage distribution arises
non-conservative for the lowest and highest KC values examined from the individual contributions of mode 1 and 2. The SRSS curve
(KC = 2.5, 20 and 30), but conservative in-between. For ␣ > 0.3, the is also different from the 10-mode TD curve. Thus, even though
FD solution is seen to become increasingly conservative regardless the SRSS solution does account for the dominant second mode, it
of the KC value. fails to accurately predict modal interaction. The CQC curve, on the
TD simulations were also performed with 3 modes included. other hand, can hardly be distinguished from the TD curve. This
Results are shown in Fig. 9b). Not surprisingly, these results are clearly demonstrates that Eq. (64) captures multi-mode response
qualitatively similar to the results in Fig. 9a). However, in the low- with excellent accuracy.
␣ range the FD solution is now generally non-conservative (by up Fatigue damage for the full Hs distribution for Case 3 is plotted
to 26%). As mentioned previously, the differences between FD and in Fig. 10d). The CQC and TD results are again similar. The fatigue
TD results are small compared to the safety margin and overall life according to the CQC expression is found to be 418 years, which
uncertainty. Nevertheless, a non-negligible impact of higher-order corresponds almost perfectly to the TD solution of 422 ± 5 years.
modes for a typical single-span pipeline has not been described in The SRSS approach is seen to perform poorly both for the single-
previous studies [11,12]. and the double-span, although for different reasons. The single-
span frequencies are widely separated and even the fundamental
frequency is normally much higher than the peak frequency of the
6.6. The influence of cross-terms in FD solution wave spectrum. All the higher-order eigenfrequencies are in the
quasi-static domain. For this situation, the fundamental mode dom-
In DNV-RP-F105 [3], it is assumed that the main damage inates the response and the SRSS solution is inappropriate because
contribution comes from the fundamental mode and an FD algo- the disregarded cross-terms will often be larger than the retained
rithm based on Eq. (62) is outlined. We have observed that the
H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147 145

Fig. 10. (a) and (c) Fatigue damage along pipe axis. (b) and (d) Fatigue damage distribution versus Hs .

higher-order terms. For the double-span, the first and second eigen- The calculations of wave-induced damage are more sensitive to
frequencies are not well separated, and SRSS solutions are known variation of the intermediate shoulder length (Fig. 11b). The single-
to be inaccurate for such cases [23,37]. Hence, it is recommended mode TD curve reflects a change in mode shape. The side-span peak
to base FD evaluations of wave-induced fatigue on the CQC expres- gradually vanishes and then reappears, but now with the same sign
sion, Eq. (64). as the main-span peak (cf. Fig. 7b). Consequently, the weighting
factor 1 steadily increases. The damage predicted by the multi-
mode TD analysis reaches a maximum at Lint ,1 /Ds = 3, where the
6.7. Effect on fatigue damage of inserting a side-span estimated annual damage is 79% higher than the single-span FD
result. In view of this finding, a fatigue analysis for the long-term Hs
The Case 3 double-span pipeline has been investigated in some distribution (Table 3) was performed for Case 3 with Lint ,1 adjusted
detail, and even though the response has been observed to be multi- to 3Ds and safety factors included. The resulting fatigue life was
modal, the predicted fatigue life is only about 25% different from found to be 36.8 ± 0.5 years, which is 48% smaller than the single-
the conventional single-span, single-mode estimate. It is impor- span FD estimate of 70.8 years.
tant to assess whether the deviations can become much larger than The results of the present study indicate that a combination
this. For this purpose, the Case 3 span geometry was systematically of multi-span behavior and multi-mode response may result in
varied. Fig. 11a) shows the fatigue damage results for varying the roughly a doubling of predicted fatigue damage as compared to con-
relative side-span length L1 /Ds and Fig. 11b) shows the results for ventional pipeline design calculations. The results also indicate that
varying the intermediate shoulder length Lint ,1 /Ds . All dimensions large deviations from conventional single-span analyses are rare
except the parameter that was varied were kept constant as given and well within the tolerance of the recommended safety factors.
by Fig. 7a). The TD calculations were performed for the same sea- It is emphasized, however, that quantitatively different estimates
state (Hs = 7 m) chosen previously, taking the average of ten 3-h obviously would result from other combinations of pipe-cross sec-
sea-state realizations. tions, soil stiffness and environmental conditions. Thus, complex
In Fig. 11a), the damage caused by the fundamental mode alone span scenarios should be evaluated for multi-span effects using
declines abruptly as L1 approaches L2 (=70Ds ) and the mode shape for instance the time- or frequency-domain algorithms developed
becomes increasingly anti-symmetric. The fatigue damage deter- herein.
mined by the multi-mode (10 modes) analysis is seen to increase Since this study to our knowledge is the first to investigate
monotonously with L1 , which is not surprising since a longer side- effects of multi-span behavior on wave-induced fatigue, the focus
span is associated with a drop in the eigenfrequencies. However, has been to evaluate the safety of, and suggest improvements to,
the maximum TD damage estimate is only 26% higher than the existing design practice. It should be remarked that the flexibility of
single-span FD solution. the time-domain algorithm may also be used to potentially reduce
146 H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147

Fig. 11. (a) Annual damage as function of side-span length. (b) Annual damage as function of intermediate shoulder length.

the level of conservatism, for instance by implementing hydrody- It is assumed that rotatory inertia can be neglected [21]. For
namic load coefficients that are functions of the instantaneous wave the displacement field in Eq. (12) the kinetic energy is then readily
velocity and by including the variation in loading along the pipe obtained as
axis.  L  L 
1 2 1 T T
T= (md + mc ) v̇0 dx = Ḋ (md + mc ) N Ndx
2 0
2 0
7. Conclusions
1 T
Ḋ = Ḋ Mdc Ḋ, (A.1)
• A complete time-domain algorithm for the analysis of wave- 2
induced fatigue damage of free spanning pipelines has been where md is the dry mass and mc is the content mass per unit
developed, which is capable of handling multi-span behavior and length, as defined previously for Eq. (16), and where the transverse
multi-mode response. The algorithm is computationally efficient, displacement v0 has been expressed using Eq. (14).
it incorporates the safety factor format of recognized design codes The total potential energy ˘ of the system comprises contribu-
[2,3], and it has been verified against detailed finite element anal- tions from the elastic strain energy due to bending, potential energy
yses. of tension and elastic strain energy due to deformation of the soil.
• Disregarding the cross-modal terms in frequency-domain fatigue The potential energy is thus given by
calculations, as suggested by previous studies [11,12], has been   L  2  L
demonstrated to be inaccurate for both single- and multi-span 1 1 ∂v0 1
˘= E(εxx )2 dV + Seff dx + ksoil (v0 )2 dx
pipelines. 2 V
2 0 ∂x 2 0
• Frequency-domain analyses based on the complete response   L   L 
1 T 1 T
spectrum expression, including the cross-modal terms, have been = D EI NT,xx N,xx dx D+ D Seff NT,x N,x dx
shown to be in excellent agreement with time-domain results. 2 0
2 0
• The common assumption that the main damage contribution  L 
1 T
comes from the fundamental mode does not hold for multi-span D+ D ksoil NT Ndx D
2
pipelines. Fatigue life estimates may be highly non-conservative 0
if only the fundamental mode is considered when the modal anal- 1 T  1
= D Kstruc + K g + K soil D = DT KD
ysis accounts for span interaction. 2 2
• A parametric study on an example double-span pipeline indicates (A.2)
that multi-span interaction and multi-mode response can give In Eq. (A.2), E is the Young’s modulus, I is the second moment of area,
rise to a doubling of the estimated fatigue damage as compared and it has been assumed that the pipe cross-section is uniform along
to conventional single-span analyses. The deviations between full the model length L. The structural stiffness matrix for bending of
multi-span/multi-mode analyses and conventional single-span the pipe is denoted Kstruc , the geometric stiffness matrix is denoted
analyses may be larger for other combinations of pipe-cross sec- Kg , and the soil stiffness matrix is denoted Ksoil . Together these
tions, soil stiffness values and environmental conditions. contributions constitute the total system stiffness matrix K.
Prior to applying the extended Hamilton’s principle, the virtual
Acknowledgement work ıWnc due to all forces not accounted for in ˘ must be deter-
mined. In the present context, it is natural to include contributions
The research has been financed by the University of Oslo and from the time-varying hydrodynamic load p(x,t) as well as from a
the DNV GL scholarship fund. viscous damping force proportional to the transverse pipe velocity.
Consequently, the resulting expression becomes

Appendix A.
L
 
ıWnc = p(x, t)ıv0 − c v̇0 ıv0 dx
0
The equation of motion for a pipeline subjected to direct wave 
loading may be derived by applying Hamilton’s principle, which
L
 
= ıDT p (x, t) NT − cNT NḊ dx , (A.3)
requires that the system Lagrangian, dependent on the kinetic and 0
total potential energy, is known.
H.A. Sollund et al. / Applied Ocean Research 61 (2016) 130–147 147

where p(x,t) is given by Eq. (7) and c is the damping coefficient. The [10] J.P. Morison, M.P. O’Brien, J.W. Johnson, S.A. Schaaf, The force exerted by
equations of motion may now be derived from Hamilton’s principle surface waves on piles, Petrol. Trans. AIME 189 (1950) 149–154.
[11] K.J. Mørk, O. Fyrileiv, Fatigue design according to the DNV guideline for free
[21,23] as expressed by spanning pipelines, in: Proc. of 21st Offshore Pipeline Tech. Conf., OPT’98,
 t1  t1 Oslo, Norway, February, 1998, pp. 23–24.
ı (T − ˘) dt + ıWnc dt = 0 , (A.4) [12] T. Xu, B. Lauridsen, Y. Bai, Wave-induced fatigue of multi-span pipelines, Mar.
Struct. 2 (1999) 83–106.
t0 t0
[13] K. Vedeld, H.A. Sollund, J. Hellesland, Free vibrations of free spanning offshore
where ı denotes variation during the time interval from t0 to t1 . pipelines, Eng. Struct. 56 (2013) 68–82.
[14] H.A. Sollund, K. Vedeld, A finite element solver for modal analyses of offshore
After inserting for the kinetic energy T, potential energy ˘ and pipeline multi-spans, in: Research Report in Mechanics, No. 14-1, Mechanics
virtual work ıWnc according to Eqs. (A.1)–(A.3), Eq. (A.4) becomes Division, Dept. of Mathematics, University of Oslo, Norway, 2014.
    
t1
T
t1 L
 
ıḊ Mdc Ḋ − ıDT KD dt + ıDT p (x, t) NT − cNT NḊ dxdt
t0 t0 0
    
d    
t1 L
T
= ıD Mdc Ḋ − ıDT KD − p (x, t) NT − cNT NḊ dx dt
t0
dt 0
      (A.5)
T
!t1 t1
T d
L
T
L
T
= ıD Mdc Ḋ − ıD M Ḋ + cN Ndx Ḋ + KD − p (x, t) N dx dt
t0
t0
dt dc 0 0
   
!t1 t1 L
= ıDT Mdc Ḋ − ıDT Mdc D̈ + CḊ + KD − p (x, t) NT dx dt = 0.
t0
t0 0

In Eq. (A.5), a damping matrix C has been introduced. According [15] R. Verley, T. Sotberg, A soil resistance model for pipelines placed on sandy
soils, J. Offshore Mech. Arct. Eng. 116 (3) (1994) 145–153.
to Hamilton’s principle, the variation ıD vanishes at the integration [16] R. Verley, K.M. Lund, A soil resistance model for pipelines placed on clay soils,
limits t0 and t1 . Thus, in: Proc. of 14th Int. Conf. on Offshore Mechanics and Arctic Engineering,
!t1 OMAE 1995, Copenhagen, Denmark, June, 1995, pp. 18–22.
ıD (t0 ) = ıD (t1 ) = 0 ⇒ ıDT M dc Ḋ = 0. (A.6) [17] O. Fyrileiv, L. Collberg, Influence of pressure in pipeline design-effective axial
t0
force, in: Proc. of 24th Int. Conf. on Offshore Mechanics and Arctic
Since Eq. (A.5) must be satisfied for any admissible variation ıD, Engineering, OMAE 2005, Halkidiki, Greece, June, 2005, pp. 12–17.
[18] K. Vedeld, H.A. Sollund, J. Hellesland, O. Fyrileiv, Effective axial forces in
it follows that the expression inside the parentheses vanishes, offshore lined and clad pipes, Eng. Struct. 66 (2014) 66–80.
thereby resulting in the equation [19] DNV-RP-C205, Environmental Conditions and Environmental Loads, Det
Norske Veritas, Norway, 2010 (October).
 L [20] K.F. Lambrakos, J.C. Chao, H. Beckmann, H.R. Brannon, Wake model of
Mdc D̈ + CḊ + KD − p (x, t) NT dx hydrodynamic forces on pipelines, Ocean Eng. 14 (2) (1987) 117–136.
[21] I.H. Shames, C.L. Dym, Energy and Finite Element Methods in Structural
0
 L 
Mechanics: SI Units Edition, 2nd ed., Taylor & Francis Group, 1991 (ISBN
0891169423).
1 
= Mdc D̈ + CḊ + KD − DCD (U|U| − 2|U|v̇0 ) + D2 CM U̇ − ma v̈0 NT dx [22] H.A. Sollund, K. Vedeld, Effects of seabed topography on modal analyses of
2 4
free spanning pipelines, in: Proc. of 25th Int. Ocean and Polar Engineering
 
0
L Conf., ISOPE 2015, Kona, Hawaii, USA, June, 2015, pp. 21–26.
[23] R.W. Clough, J. Penzien, Dynamics of Structures, 1st ed., McGraw Hill, 1975
= Mdc D̈ + ma NT Ndx D̈ + CḊ + KD
(ISBN 0-07-085098-4).
0 [24] B. Van Dao, J. Penzien, Comparison of treatments of non-linear drag forces
 L  acting on fixed offshore platforms, Appl. Ocean Res. 4 (2) (1982) 66–72.
1  [25] A. Naess, T. Moan, Stochastic Dynamics of Marine Structures, 1st ed.,
− DCD (U|U| − 2|U|v̇0 ) + D2 CM U̇ NT dx
2 4 Cambridge University Press, New York, NY, USA, 2013 (ISBN
0
978-0-521-88155-5).
= MD̈ + CḊ + KD − R (t) = 0. [26] R.D. Cook, D.S. Malkus, M.E. Plesha, R.J. Witt, Concepts and Applications of
(A.7) Finite Element Analysiss, 4th ed., Univ. of Wisconsin, John Wiley & sons,
Madison, 2002 (ISBN 978-0-471-35605-9).
[27] A.K. Chopra, Dynamics of Structures – Theory and Applications to Earthquake
Eq. (A.7) can be recognized as identical to Eq. (15). Engineering, 3rd ed., Pearson Prentice Hall, 2007 (ISBN 0-13-156174-X).
[28] K.J. Bathe, M.M.I. Baig, On a composite implicit time integration procedure for
nonlinear dynamics, Comput. Struct. 83 (2005) 2513–2524.
References [29] G. Rio, A. Soive, V. Grolleau, Comparative study of numerical explicit time
integration algorithms, Adv. Eng. Softw. 36 (2005) 252–265.
[1] M. Drago, M. Mattioli, R. Bruschi, L. Vitali, Insights on the design of [30] The MathWorks Inc, Matlab, v. 7.11.2010b, The MathWorks Inc., Natick, MA,
free-spanning pipelines, Phil. Trans. R. Soc. A 373 (2015) 1–27. USA, 2010.
[2] DNV-OS-F101, Submarine Pipeline Systems, Det Norske Veritas, Norway, [31] ASTM E, Standard practices for cycle counting in fatigue analysis Annual Book
2012. of ASTM Standards, vol. 03.01, ASTM E, Philadelphi, 1999, pp. 710–718,
[3] DNV-RP-F105, Free Spanning Pipelines, Det Norske Veritas, Norway, 2006 1049-85 (Reapproved 1997).
(February). [32] A. Niesłony, Determination of fragments of multiaxial service loading strongly
[4] DNV-RP-F111, Interference Between Trawl Gear and Pipelines, Det Norske influencing the fatigue of machine components, Mech. Syst. Signal. Process.
Veritas, Norway, 2010 (October). 23 (8) (2009) 2712–2721.
[5] B.M. Sumer, J. Fredsøe, Hydrodynamics around cylindrical structures Adv. [33] DNV-RP-C203, Fatigue Design of Offshore Steel Structures, Det Norske
Series on Ocean Eng., vol. 12, 1st ed., World Scientific, 2006. Veritas, Norway, 2012 (October).
[6] R.D. Blevins, Flow-Induced Vibration, 2nd ed., Krieger Publishing, 1990 (ISBN [34] A.K. Malhota, J. Penzien, Stochastic Analysis of Offshore Tower Structures,
978-1575241838). Earthquake Eng. Res. Center, College of Engineering, University of California,
[7] O. Fyrileiv, K.J. Mørk, M. Chezhian, Experiences using DNV-RP-F105 in Berkeley, 1969 (Report No. EERC-69-6, May).
assessment of free spanning pipelines, in: Proc. of 24th Int. Conf. on Offshore [35] L.P. Krolikowski, T.A. Gay, An improved linearization technique for frequency
Mechanics and Arctic Engineering, OMAE 2005, Halkidiki, Greece, June, 2005, domain riser analysis, in: Proc. of 12th Offshore Tech. Conf., OTC 1980,
pp. 12–17. Houston, USA, May, 1980, pp. 5–8.
[8] H.A. Sollund, K. Vedeld, J. Hellesland, O. Fyrileiv, Dynamic response of [36] P.H. Wirsching, M.C. Light, Fatigue under wide band random stresses, J. Struct.
multi-span offshore pipelines, Mar. Struct. 39 (2014) 174–197. Div. ASCE 106 (7) (1980) 1593–1607.
[9] L. Vitali, F. Marchesani, G. Curti, R. Bruschi, Dynamic excitation of offshore [37] E.L. Wilson, A. der Kiureghian, E.P. Bayo, A replacement for the SRSS method
pipelines resting on very uneven seabeds, in: T. Moan, et al. (Eds.), Proc. of in seismic analysis, Earthq. Eng. Struct. D 9 (1981) 187–194.
2nd European Conf. on Struct. Dynamics, EURODYN ’93, Trondheim, Norway, [38] Dassault Systèmes Simulia Corp, Abaqus, v. 6.12, Dassault Systèmes Simulia
June, 1993. Corp., Providence, RI, USA, 2012.

You might also like