You are on page 1of 12

A Semi-Analytical Model for Structural Response Calculations of Subsea Pipelines in Interacting Free

Spans

V International Conference on Computational Methods in Marine Engineering


MARINE 2013
B. Brinkmann and P. Wriggers (Eds)

A SEMI-ANALYTICAL MODEL FOR STRUCTURAL RESPONSE


CALCULATIONS OF SUBSEA PIPELINES IN INTERACTING FREE
SPANS
HÅVAR SOLLUND*† AND KNUT VEDELD†
*†
Mechanics Division,
Department of Mathematics, University of Oslo,
Moltke Moes vei 35, Pb. 1053 Blindern,
0316 Oslo, Norway
E-mail: haavaras@math.uio.no - Web page: http: // www.math.uio.no

Key words: Modal analysis, Rayleigh-Ritz, Pipeline, Free span, Multi-span

1 INTRODUCTION
Pipeline free spans may be caused by uneven seabed, by surrounding subsea infrastructure
such as pipeline crossings or by erosion processes like seabed scouring. On the seabed, the
pipeline is subject to wave and current loading, and in free spans the surrounding flow may
give rise to vortex shedding. The vortex shedding generates oscillations in the drag and lift
forces acting on the pipe [1]. If the frequencies of the force oscillations are close to one of the
eigenfrequencies of the free spanning pipeline, the pipeline may experience large-amplitude
vibrations. Such vibrations are termed vortex-induced vibrations (VIV), and fatigue failure
due to VIV in free spans is a major risk factor for offshore pipelines [2,3]. Moreover, the
natural frequency of a free span decreases quickly with increasing span length, making long
spans more prone to VIV-induced fatigue damage. Since the cost of seabed intervention in
order to reduce span lengths is high, modern design codes, like Det Norske Veritas'
recommended practice provisions, DNV-RP-F105 “Free Spanning Pipelines” [4], allow for
the occurrence of VIV as long as the accumulated fatigue damage is accounted for.
It is essential for reliable free span design to estimate eigenfrequencies and associated
modal stresses with a high degree of accuracy. For multi-span configurations and complicated
single-span configurations, the structural analyses are normally performed by means of
detailed non-linear finite element analysis (FEA). However, such analyses are time-
consuming, and since the number of free spans along a pipeline route may be substantial, fast
and reliable methods for identifying critical span configurations are attractive for the pipeline
engineering community. A simplified method for calculation of eigenfrequencies and
associated stresses was therefore presented by Fyrileiv and Mørk in 2002 [3], and later
included in DNV-RP-F105. The method developed by Fyrileiv and Mørk has a limited range
of application with regard to span length and static deflection into the span [3,5]. Furthermore,
the simplified method can only be applied on single-span scenarios. A semi-analytical method
for static and dynamic free span analyses was recently developed by Sollund and Vedeld [5].
The semi-analytical method has no practical limitations on span length and mid-span
deflections, while still being ~600 times faster than FEA for typical span configurations [5].
The present paper describes an extension of the semi-analytical method, making the

889
Håvar Sollund and Knut Vedeld

method applicable for multi-span configurations. With regard to computational time, the
multi-span model operates equally fast as the single-span model, thereby addressing the need
in the pipeline industry for a screening tool for interacting free spans, and making the method
highly suitable for implementation in engineering software programs.

2 PROBLEM DEFINITION

Figure 1: Multi-span scenario of a pipeline, indicating free span length Ls, adjacent span length La and
intermediate shoulder length Lint.

Figure 1 displays a free span scenario with two adjacent spans. In the figure, the main span
length is Ls, the adjacent span length is La and the shoulder length in-between the span is Lint,
which in the following will be termed the intermediate shoulder length. The combined length
Lsa = La + Lint + Ls will be called the span area length. The regions where the pipe touches
down on the seabed on either side of the span area are labeled span shoulders.
In DNV-RP-F105 [4], some guidance is given on how to determine whether free spans are
influenced by other neighboring free spans or not. The guidance for span classification
consists of a single set of curves (with each curve representing a particular soil type)
separating between interacting and isolated spans, and these curves are replotted in Figure 2.

Figure 2: Span interaction curves, as defined in DNV-RP-F105 [4].

For a given free span length with a specified intermediate shoulder length, the curves in
Figure 2 indicate that there is a limit adjacent span length for which longer adjacent spans
result in interaction between the two spans, and shorter adjacent spans result in no interaction

890
Håvar Sollund and Knut Vedeld

between the two spans. A main objective in the following text is to evaluate the curves and
assess whether the dimensionless parameter representation in Figure 2 is sufficient for a
classification of interacting and isolated spans.
Traditionally, indication of directions in VIV analyses is defined based on the direction of
flow rather than the axis of the excited cylinder. Hence, in the following the horizontal
direction will be termed in-line and the vertical direction will be termed cross-flow.
The geometry and boundary conditions of the model are idealized as shown in Figure 3.
The shoulders are considered to be horizontal and straight, and the pipeline ends are assumed
to be simply supported, allowing no axial or transversal displacements. This assumption is
physically motivated, either by span shoulders that are sufficiently long to provide
approximate axial fixity as a result of axial pipe-soil friction, or that other effects from the
seabed induces axial fixation (such as neighboring spans, rock dumps, pipe crossings, etc.).

Figure 3: Definition of pipeline model and Cartesian coordinate system. (a) Static and dynamic soil springs are
applied axially, laterally and vertically at the span shoulders. (b) Directions of spring forces.

The modeled pipe has a dry mass including mass of fluid content per unit length md and a
submerged weight q. For dynamic response calculations, one also has to include the effect of
the added mass ma due to acceleration of the surrounding water. The effective mass me, given
by the sum of md and ma, is taken as an input parameter to the dynamic response calculations.
The pipe-soil interaction will be modeled according to the recommendations in DNV-RP-
F105 [4], distinguishing between stiffness coefficients for vertical static, vertical dynamic,
lateral dynamic, axial static and axial dynamic displacements. The static and dynamic soil
stiffnesses are termed KVS, KV, KL, KAXS and KAX, respectively. In the numerical computation,
the shoulder lengths are taken as three times the length of the span area Lsa. The axial friction
is ignored when determining the deflected geometry due to the static loading, but dynamic
axial soil stiffness is included in the modal analyses. However, the semi-analytical model
presented below is equally applicable for other choices of span shoulder lengths and axial

891
Håvar Sollund and Knut Vedeld

friction behavior. The effective axial force concept [6] will be applied for calculation of
geometric stiffness effects.
In the semi-analytical method presented in the next section, three solutions to the problem
defined by Figure 3 will be required:
1. A solution for the static equilibrium case, where the pipe is subject to its submerged
weight q and effective axial force Seff.
2. A solution to determine the eigenfrequencies and associated mode shapes for the
linearized harmonic eigenvalue problem in the in-line direction subject to pipe
effective mass me and effective axial force Seff.
3. A solution, dependent on the static vertical configuration, to determine eigen-
frequencies and mode shapes for the linearized harmonic eigenvalue problem in the
cross-flow direction, accounting for the stiffening effect of the vertical static
displacement, effective mass me and effective axial force Seff.

3 THE SEMI-ANALYTICAL METHOD


To include the effects of the static vertical deflections of the model defined in Figure 3 a
curvilinear coordinate system is chosen. The coordinate system is defined in Figure 4, where
en is the direction normal to the pipe axis and es is the direction tangential to the pipe axis.
Unlike the x,y-coordinate system, the es,en-system is not fixed in space, but will rotate along
the pipe length ensuring that the es direction is always tangential to the pipe axis. The
curvilinear coordinate (or arc length coordinate) s denotes the position along the pipe length.
The coordinate directed from the pipe centroidal axis towards the outer circumference is
denoted n. Depending upon the analysis type, this direction will either be in the horizontal
plane (in-line analyses) or the vertical plane (cross-flow analyses). At each position s the
displacement in tangential direction is us while the displacement in normal direction is un. The
radius of curvature R(s) will vary along the pipe length.

Figure 4: Curvilinear coordinate system with unit vectors es in the tangential direction and en in the normal
direction. The static deflection vs and the Cartesian coordinate system oriented along the undeformed pipe axis
are also shown.

Consistent with the Euler-Bernoulli beam theory, the only non-zero component of the
strain tensor is εss. The following expression, as derived by Sollund and Vedeld [5], applies:

892
Håvar Sollund and Knut Vedeld

1  u s  1  u 0   v 
 ss    u n     v0  n  0  u 0  (1)
1  n  s  1  n  s s  s 
The displacements u0 and v0 of a point on the pipe centroidal axis are the unknown
displacements for which we must solve. The boundary conditions require that u0 and v0 must
be zero at positions s = 0 and s = L. Note that for the static analysis and the in-line modal
analysis, this corresponds to the positions x = 0 and x = Li, where Li is the initial pipe length
before deflection into the span. Thus, the following Fourier sine approximation can be applied
for u0 and v0

 is  
u 0 s, t    Du ,i sin  sin t   N u Du sin t 
 L   N u 0  D u 
u    ND (2)
i 1

 is  0 N v   D v 
v0 s, t    Dv ,i sin   sin t   N v D v sin t 
i 1  L  

Based on the strain expression in Eq. (1), a differential operator may be defined as


d  d u0 
d v0 
1 

1   s n  s
 n

s
 s    s   n
2 

s 2 
(3)

The static analysis may thus be performed by solving

KD  R (4)

where the stiffness matrix is obtained from


K  K struc  K g  K soil (5)

and Kstruc, Kg and Ksoil denotes the structural stiffness matrix, the geometric stiffness matrix
and the soil stiffness matrix, respectively.
The structural stiffness matrix is given by
K struc  E  dN  dN dV
T
(6)
V

The geometric stiffness matrix is dependent on the effective axial force Seff and is obtained
in the usual manner [5,7]. In the single-span model previously presented by Sollund and
Vedeld [5], a non-linear static analysis which included the build-up of tension in the pipe due
to pipe lengthening, was incorporated. However, it is often convenient to determine the
distribution of effective axial forces in a global analysis, and taking the relevant value of
effective axial force as an input to the local model of the span area. Therefore, the effective
axial force has been kept constant during the static analysis in the present study.
The soil stiffness matrix can be found by
L
k NT N 0 
K soil    soil, A u u T ds (7)
0 0 k soil N N v 
v

893
Håvar Sollund and Knut Vedeld

where the axial soil stiffness ksoil,A is replaced by KAXS or KAX, and the transverse soil stiffness
ksoil is replaced by KVS, KV, or KL depending on the analysis type (cross-flow/in-line,
static/dynamic). The soil stiffnesses are set to zero in the spans La and Ls.
The load vector R is given by
L
0 (8)
R   N T   ds
0 q1
where 1 is a vector with unit value in all its entries.
The eigenfrequencies and mode shapes are obtained from the static configuration by
solving the equations for free vibration of the pipe

K   MD  0 (9)

2
M D KD  0 
where M is the mass matrix and the eigenvalues ω are the natural circular frequencies of the
pipe. An attractive feature of the Fourier sine expansions applied for expressing the
displacements, Eq. (2), is that the mass matrix M is diagonal.

4 FINITE ELEMENT ANALYSES


The commercially available finite element solution software ABAQUS [8] was used for
finite element (FE) analyses. The FE analyses were performed in order to provide benchmarks
for evaluation of the semi-analytical multi-span model. The beam element PIPE31 was used
for the FE modeling of the pipe. This is a first order shear deformable linear beam element.
Linear springs were used to model the soil stiffness. The model was generated according to
Figure 3, with vertical, lateral and axial springs at each node. Shoulder lengths and span
lengths were set to be equally long in the finite element and corresponding semi-analytical
models. The effective axial force was kept constant during the static analysis, by introducing a
roller support on the right side of the model. The axial displacement due to the introduction of
the roller support was negligible (~0.1 pipe diameter compared to a total model length of
~1000 pipe diameters) and did not affect the outcome of the analysis.
The recommendation for maximum element lengths in DNV-RP-F105 [4] is one outer
pipeline diameter, i.e., the ratio of the element length to the outer pipeline diameter should be
less than 1. Based on convergence studies, an element length equal to 10% of the outer pipe
diameter was chosen throughout in the analyses, ensuring highly accurate estimates of modal
frequencies and associated stresses.

5 RESULTS AND DISCUSSION

5.1 Validation of the semi-analytical method


Detailed FE analyses, as described in the preceding section, were performed in order to
assess the accuracy of the semi-analytical multi-span model. Three different pipe cross-
sections were applied for the comparisons. The three configurations are described by the
parameters listed in Table 1. Pipe 1 is a small-diameter (4-inch internal diameter), thick-
walled oil pipeline with steel diameter-to-thickness (Ds/ts) ratio as small as 7.4, without
external coating. The second pipe cross-section has a 9-inch internal diameter, a Ds/ts ratio of

894
Håvar Sollund and Knut Vedeld

19.8 and a 50-mm thermal insulation coating. The final pipe configuration is a large-diameter
gas pipeline with Ds/ts equal to 35.8 and a 60-mm concrete coating. The effective masses of
the pipes range from 80 kg/m to more than 1300 kg/m. Thus, the three cases span a wide
diameter and mass range, as seen from Table 1.

Table 1: Input parameters for pipe cross-sections.

Input parameter Symbol Unit Pipe 1 Pipe 2 Pipe 3

Outer steel diameter Ds mm 140.7 254.3 720.2


Steel thickness ts mm 19.1 12.85 20.1
Coating thickness tcoat mm 0 50 5.0
Concrete thickness tconc mm 0 0 60
Density of steel ρsteel kg/m3 7850 7850 7850
Density of coating ρcoat kg/m3 - 793 1300
Density of concrete ρconc kg/m3 - - 2250
Density of content ρcont kg/m3 821 821 150
Young’s modulus E GPa 207 207 207
Submerged weight q N/m 471.71 459.75 1652.11
Effective mass me kg/m 80.02 249.37 1334.50

Six FE analyses were performed for each of the three pipe cross-sections. Each set of six
analyses comprised two analyses on very soft clay, two analyses on stiff clay and two
analyses on dense sand. Relevant values of soil stiffnesses were taken from DNV-RPF105
[4]. For each soil type, analyses were performed for two values of relative intermediate
shoulder length Lint/Ls (see Figure 1 for definitions). The first of these intermediate shoulder
lengths would represent a very short shoulder with Lint/Ls of either 0.1 or 0.2, while a longer
intermediate shoulder would be applied for the second analysis, covering Lint/Ls values from
0.3 to 0.8. Based on Figure 2 and on preliminary analyses with the semi-analytical method,
the value of the relative adjacent span length La/Ls was always adjusted in such a manner that
interaction in the first mode would occur, either in the in-line direction or in the cross-flow
direction or both. This resulted in a variation of La/Ls from 0.6 to 0.9. The results of the semi-
analytical multi-span model coincided with the results of the semi-analytical single-span
model when the spans were not interacting. Hence, smaller values of La/Ls were not of interest
for the present study, since the accuracy of the single-span model has been evaluated
previously [5]. The 18 validation cases also covered relative span lengths Ls/Ds of the
dominant span ranging from 60 to 160.
The results of the validation study are displayed in Figure 5. The figure shows the ratios of
the fundamental frequencies predicted by the semi-analytical method to the corresponding
frequencies obtained by FEA, as well as the ratios of predicted modal stresses to modal
stresses from FEA. Results are shown for both the in-line and the cross-flow direction. It is
seen that the accuracies of frequency calculations are within 3-4% for the in-line direction and
within ±1% for the cross-flow direction. The accuracy of the modal stress results is within
±5% as compared to the FE results. The results reveal that there is a small, but consistent stiff
bias in the in-line direction, i.e., the frequencies and stresses are slightly overestimated

895
Håvar Sollund and Knut Vedeld

compared to the results obtained by FEA. The stiff bias appears to be most pronounced for
cases with a short intermediate shoulder length Lint/Ls and a short relative length of the
dominant span Ls/Ds. A similar consistent, but less pronounced, stiff bias in in-line direction
could be observed for short relative span lengths in the assessment of the single-span model
[5]. However, in an engineering context, the semi-analytical method performs excellently and
well within the accuracy that was deemed acceptable for the approximate response calculation
method previously presented by Fyrileiv and Mørk [3], and later incorporated in DNV-RP-
F105 [4]. In fact, also the method developed by Fyrileiv and Mørk exhibited a stiff bias for
short spans, and it is likely that this bias is caused by shear deformations which have been
included in the FE modeling, but ignored in the semi-analytical method as well as in the
method by Fyrileiv and Mørk [3,5].

Figure 5: Ratios of predicted fundamental frequencies and associated stresses to FE results for both the in-line
and the cross-flow direction.

5.2 Assessment of multi-span classification curves


Prior to determining whether two neighboring free spans interact or not, a criterion for
interaction has to be established. A reasonable engineering approach is to classify two spans
as interacting if the presence of an adjacent span influences the frequency or modal stress of
the main span above a certain limit. The limit should be aligned with the accuracy of dynamic
response calculations using FE analysis or approximate calculation methods, like the semi-
analytical method or the single-span method outlined in DNV-RP-F105. Minor inaccuracies
will then be covered by the calibrated safety factors (e.g., 1.1 on natural frequencies and 1.3
on modal stresses for well-defined spans subject to normal safety requirements [4]) in leading
design codes such as DNV-RP-F105 [4]. Based on this reasoning the following definition has
been proposed and applied for the present study:
Two free spans are interacting if the presence of an adjacent span changes the
fundamental frequency, either in-line or cross-flow, of the main span by more than 5% or the

896
Håvar Sollund and Knut Vedeld

associated first-mode dynamic stress of the main span by more than 10%.
While the introduction of an adjacent span may be regarded as a softening of the main-
span shoulder, and consequently, will result in a consistent drop in the natural frequencies, the
effect on the modal stresses is less certain and more dependent on local geometry and possibly
on intermediate shoulder lengths. The 10% limit on modal stresses in the definition above will
account for such variation in stresses.
Based on the proposed span interaction definition, classification curves like illustrated in
Figure 2 were determined with the semi-analytical method for the three pipe configurations
described in Table 1. Four different soil types were applied; very soft clay, stiff clay, loose
sand and dense sand, respectively. For each combination of pipe and soil type, classification
curves were established for ten values of relative length Ls/Ds of the main span, ranging from
30 to 160. Each individual classification curve was constructed from 15 classification points,
i.e., 15 values of required adjacent span length La/Ls, each corresponding to a given
intermediate shoulder length Lint/Ls. The algorithm for identifying the classification points is
illustrated by the flow chart shown in Figure 6. As displayed in Figure 6, the accuracy by
which the classification points are obtained is controlled by the tolerance T. For the purpose
of the present study, T was set to 0.005·Ls. Large numbers of analyses are necessary in order
to establish the classification curves with such high precision. However, due to its
computational efficiency, the semi-analytical method is highly suited for this type of large-
scale parameter studies.

Figure 6: Algorithm for determining the length of the adjacent span La that is required for span interaction.

The results obtained for pipe 1 (see Table 1) on dense sand are displayed in Figure 7.
Results are shown for all the selected values of Ls/Ds. It is observed that the classification
points follow the curve from DNV-RP-F105 for Lint/Ls above 0.4. For intermediate shoulder
lengths above this limit the spans behave as isolated spans independently of the length of the
adjacent span. For values of La/Ls slightly larger than one, the adjacent span will be longer
than the main span, and soon have a fundamental frequency that is 5% smaller than the main

897
Håvar Sollund and Knut Vedeld

span frequency. Hence, the interaction criterion will be fulfilled. The fact that the obtained
classification points coincides with the RP-curve in this region indicates that the definition of
interacting spans applied in the present study is well aligned with the approach previously
applied for establishing the classification curves in [4]. For values of Lint/Ls below 0.4, the
calculated classification points are scattered around the RP-curve, thereby clearly
demonstrating that onset of span interaction will depend on the length of the main span.
Interestingly, the results for the shortest (denoted L1) and the longest (denoted L10) main
spans in the study are both well below the RP-curve, that is, on the “isolated span” side of the
curve. However, while the required adjacent span length for interaction increases smoothly
for the shortest main span length, the required value of La/Ls for the longest main span
stabilizes around 0.8 and then experiences an abrupt jump when Lint/Ls reaches the threshold
value of 0.4. This behavior, with an La/Ls plateau followed by an abrupt change to values of
La/Ls above one for a threshold intermediate shoulder length, was observed many times in this
study and was most pronounced for large values of the main span length. The results for loose
sand (not shown) resembled the results for dense sand, but the threshold value of Lint/Ls
increased to ~0.6 for the longest main span.

Figure 7: Multi-span classification points for pipe 1 on dense sand. Results are shown for 10 values of relative
main span length Ls/Ds ranging from 30 to 160, where L1 corresponds to 30 and L10 to 160.

Figure 8 shows the results obtained for the large-diameter gas pipeline, pipe 3, on very soft
clay. This time the scatter around the RP-curve is significant, and it is clearly demonstrated by
the amount of scatter in Figure 8, that a dimensionless parameter representation based only on
the quantities Lint/Ls and La/Ls will not be able to accurately predict whether two neighboring
free spans interact. It should be noted, however, that Figure 8 exhibits the worst visual fit to
the RP-curve of the cases investigated in this study. For pipe 1 and 2 on very soft clay (results
not shown) only the results for the two longest main spans fell below the RP-curve, and the
results for the shortest main spans followed the slope of the RP-curve quite closely for small
intermediate shoulder lengths. Thus, it appears that the RP-curve may have been established
as a lower bound for La/Ls based on results generated for shorter main span lengths than Ls/Ds
equal to ~130 (corresponding to L8 in Figure 8). In fact, Ls/Ds equal to 130 is the upper limit

10

898
Håvar Sollund and Knut Vedeld

of the application range for the approximate response calculation method in DNV-RP-F105,
and it may have been implicitly assumed that FE analyses are required for all spans exceeding
this limit. However, no validity range is explicitly stated for the multi-span interaction curves.
On the other hand, the RP does emphasize that the curves are provided for indicative purposes
only, and are meant to be used in lieu of detailed data, thereby recognizing the inherent
limitations in the validity of the curves.

Figure 8: Multi-span classification points for pipe 3 on very soft clay. Results are shown for 10 values of
relative main span length Ls/Ds ranging from 30 to 160, where L1 corresponds to 30 and L10 to 160.

Figure 9: Multi-span classification curves for the three selected pipe configurations on stiff clay. Panel A shows
results for Ls/Ds = 30. Panel B shows results for Ls/Ds = 160.

In Figure 9 above, results are compared for the three different pipe cross-sections on stiff
clay. Panel A displays the results for a relative main span length of 30, while panel B displays
the results for a relative main span length of 160. All the curves in Figure 9, with the

11

899
Håvar Sollund and Knut Vedeld

exception of the curve for pipe 3 in panel B, follow the RP-curve quite closely. However, as
seen previously, the curves for the shortest main span length lie above the RP-curve, i.e., on
the “interacting spans” side, while the curves for the longest main span length fall below the
RP-curve. The classification point for pipe 1 at Lint/Ls equal to 0.1 in panel A, may appear as
an outlier, but this behavior was not untypical for the shortest main span length. Indeed a
similar trend is seen for the shortest main span series in Figure 8, but it is obscured by the
large amount of data points in that plot. Based on Figure 9, one may conclude that also the
characteristics of the pipe cross-section, such as diameter and specific gravity, will influence
whether two neighboring free spans interact for a given intermediate shoulder length.

6 CONCLUSIONS
- A semi-analytical method for the dynamics analysis of pipelines with interacting
multi-spans has been developed, and comparisons have been made with FE analyses.
- The semi-analytical method is accurate to within 5% deviation for calculation of
fundamental frequencies and associated modal stresses compared to detailed FE
analyses.
- The curves given in DNV-RP-F105 [4] for classification of free spans into isolated
and interacting spans have been evaluated by performing several thousand analyses
with the semi-analytical method.
- It has been shown that span interaction is a complex issue which depends on more
parameters than the relative length of the adjacent span and the relative length of the
intermediate span shoulder. A strong dependence on the relative length of the main
span has been demonstrated, as well as a dependence on the characteristics of the
pipe cross-section.

REFERENCES
[1] Sumer, B.M. and Fredsøe, J. Hydrodynamics around cylindrical structures. Adv. Series
on Ocean Eng., Vol. 12, World Scientific, (2006)
[2] Fyrileiv, O., Mørk, K. and Chezhian, M. Experiences using DNV-RP-F105 in assessment
of free spanning pipelines. Proc. of 24th Int. Conf. on Offshore Mechanics and Arctic
Engineering, OMAE 2005, Halkidiki, Greece, June 12-17 (2005)
[3] Fyrileiv, O. and Mørk, K. Structural response of pipeline free spans based on beam
theory. Proc. of 21st Int. Conf. on Offshore Mechanics and Arctic Engineering, OMAE
2002. Oslo, Norway, June 23-28 (2002)
[4] DNV-RP-F105. Free Spanning Pipelines. Det Norske Veritas, Norway (2006)
[5] Sollund, H. and Vedeld, K. A semi-analytical model for free vibrations of free spanning
offshore pipelines. Research Report in Mechanics, No. 12-2, Mechanics Division,
Department of Mathematics, University of Oslo, Norway (2012)
[6] Fyrileiv, O. and Collberg, L. Influence of pressure in pipeline design – effective axial
force. Proc. of 24th Int. Conf. on Offshore Mechanics and Arctic Engineering, OMAE
2005, Halkidiki, Greece, June 12-17 (2005)
[7] Cook, R.D., Malkus, D.S, Plesha, M. E. and Witt, R. J. Concepts and applications of finite
element analysis. Wiley, 4th ed., (2002)
[8] ABAQUS, v. 6.12, Dassault Systèmes Simulia Corp., Providence, RI, USA (2012)

12

900

You might also like