You are on page 1of 14

Ocean Engineering 105 (2015) 217–230

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Modal response of short pipeline spans on partial elastic foundations


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

art ic l e i nf o a b s t r a c t

Article history: Very short pipeline free spans in flooded areas, river crossings, straits and inlets have been given a lot of
Received 21 August 2014 attention due to recent failures caused by vortex-induced vibrations (VIV). Performing VIV analyses of
Accepted 15 June 2015 free spanning pipelines requires accurate prediction of modal response quantities. Currently, no simple
Available online 12 July 2015
analytical solutions are available for short free spans supported by soils with a finite stiffness. Based on
Keywords: the Buckingham Pi theorem, the solution space for short free spans is spanned in the present study.
Free spans Closed form analytical solutions are determined based on surface fitting to the developed solution space.
Submarine pipelines The novel analytical solutions are accurate, easily implementable and thoroughly verified, and represent
Analytical solutions a significant improvement to current approaches found in design standards. An accurate expression for
Pipeline engineering
the buckling load of short spans is also derived, and asymptotic properties of the buckling load and
Vortex-induced vibrations
frequency equations as the span length approaches zero are explored.
Modal analysis
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction VIV is an important design issue for pipelines and piping


systems (API RP 1111, 2009; DNV-OS-F101, 2012; DNV-RP-F105,
When external fluid flow crosses a cylinder, periodic formation 2006), both offshore (Fyrileiv et al., 2005; Soni and Larsen, 2005;
and shedding of vortices occurs in the wake (Sumer and Fredsøe, Wang et al., 2009) and onshore (Hart et al., 1992). Predominantly,
2006). The formation and shedding of vortices occur at a fre- VIV of pipes and piping are caused by water flow or wind. The
quency called the Strouhal frequency (White, 1999), which importance of VIV as a design issue is highlighted by numerous
increases proportionally to the flow velocity. Within a vortex, the fatigue failures over the years (Fyrileiv et al., 2005) and significant
local flow velocity is higher than the flow velocity in the spending to mitigate it (Hove et al., 2011). VIV design of offshore
surrounding fluid, causing a local pressure drop. Since the forma- pipelines is generally done according to DNV-RP-F105 (2006), as
tion and shedding of vortices is periodic, such pressure variations recommended by two leading offshore pipeline design codes DNV-
exert harmonic forces in horizontal and vertical directions on the OS-F101 (2012) and API RP 1111 (2009). Onshore pipelines may be
cylinder (Blevins, 1990). With increasing flow velocity, the fre- exposed to wind if they are above ground (Hart et al., 1992), but
quency of the harmonic forces increases along with the Strouhal sometimes floods or rivers may dig up a pipeline, creating a free
frequency. For a threshold flow velocity, called the onset flow span exposed to water flow. Recent failures in Canada and the US
velocity, the frequency of the harmonic forces are sufficiently close have highlighted that such free spans must be designed for and
to the eigenfrequency of the cylinder to cause a dynamic response. intervened on if found unacceptable (Heggen et al., 2014).
When the cylinder starts to respond dynamically, the flow velocity Pipeline free spans which have been caused by floods or
no longer governs the shedding frequency according to the erosion processes in a river are generally very short. Hence, the
Strouhal law. Instead, the response of the cylinder instigates a natural frequencies of such spans are high, requiring significant
fluid structure interaction where the shedding frequency is gov- flow velocities to excite VIV. Similar cases from offshore pipelines
erned by the response frequency of the cylinder (Aronsen, 2007). may be found in inlets and straits where tidal currents and density
This phenomenon is called lock-in (Blevins, 1990). Due to lock-in, a driven currents may be high, imposing strict restrictions on
broad-banded flow velocity spectrum will cause vibrations, called acceptable free span lengths, as illustrated by a series of failures
vortex-induced vibrations (VIV) (Zdravkovich, 2003, 1997). in the Cook inlet in the 1960s (Fyrileiv et al., 2005).
VIV design is generally conducted by extensive parametric
studies. A large number of analyses for various combinations of
n flow velocities, span lengths, soil stiffness properties, etc., must be
Corresponding author at: Mechanics Division, University of Oslo, P Box 1053—
Blindern, NO-0316 Oslo, Norway. Tel.: þ 47 92847245; fax: þ47 22854349. performed to determine which free spans are acceptable and
E-mail address: haavaras@math.uio.no (H.A. Sollund). which are not. The acceptability of a free span is to a large extent

http://dx.doi.org/10.1016/j.oceaneng.2015.06.019
0029-8018/& 2015 Elsevier Ltd. All rights reserved.
218 H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230

Nomenclature me effective pipeline mass, Eq. (2) [kg/m]


ms pipeline structural mass [kg/m]
a,b,c curve-fit constants, Eqs. (52) and (53) [–] M ¼  EI  d2w/dx2, bending moment [N m]
A maximum unit diameter modal amplitude stress, Eq. MAB bending moment at interfaces between span and
(10) [Pa] shoulders [N m]
Ai internal cross-sectional area [m2] Mmax maximum modal bending moment, Eq. (10) [N m]
As steel cross-sectional area [m2] Mms,Msh bending moment in ms and sh coordinate systems
ASA maximum modal stress by semi-analytical [N m]
method [Pa] p polynomial coefficient vector, Eq. (44) [–]
AF105 maximum modal stress according to DNV-RP- P compressive axial load, Eq. (16) [N]
F105 [Pa] Pcr critical buckling load, Eq. (36) [N]
C1–4 undetermined coefficients in expression for wms [m] Pcr,BEF critical buckling load for beam on elastic foundation,
D hydrodynamic diameter [m] Eq. (30) [N]
D diagonal weight coefficient matrix, Eq. (47) [–] PE Euler buckling load, Eq. (31) [N]
Ds outer steel diameter [m] Qw ¼  EI  d3w/dx3 P  dw/dx, vertical shear force [N]
D1–4 undetermined coefficients in expression for wsh [m] Qw,ms,Qw,sh vertical shear force in ms and sh coordinate
E Young’s modulus [Pa] systems [N]
f fundamental frequency [Hz] R concentrated mid-span load [N]
fBEF frequency for beam on an elastic foundation, Eqs. Seff effective axial force, Eq. (1) [N]
(38)-(40) [Hz] ts steel wall thickness [m]
fSA fundamental frequency by semi-analytical u,v,w displacements in x-, y- and z-directions
method [Hz] respectively [m]
fF105 fundamental frequency according to DNV-RP- wms,wsh transverse displacement in ms and sh coordinate
F105 [Hz] systems [m]
Ff non-dimensional frequency response function, Eq. x,y,z Cartesian coordinates [m]
(7) [–] xms x-coordinate with origo at mid-span [m]
Fκ non-dimensional curvature function, Eq. (12) [–] xsh x-coordinate with origo at interface between span and
g expression fitted to Gf/Gf,BEF, Eqs. (49) and (50) [–] shoulder [m]
g vector of discrete evaluations of Gf/Gf,BEF [–] zb buckling parameter, Eq. (24) [–]
gA expression fitted to Gκ/Gκ,0, Eq. (51) [–] zb,BEF ¼ Pcr,BEF/PE, Eq. (31) [–]
Gf fundamental frequency response surface, Eq. (9) [–] zb,PER buckling parameter, asymptotic expansion, Eq. (29) [–]
Gf,BEF response surface for frequency of beam on elastic α temperature expansion coefficient [(1C)  1]
foundation, Eq. (41) [–] β ¼ log10(πk), Eq. (15) [–]
Gκ response surface for maximum modal curvature, Eq. γ displacement function parameter, Eqs. (20) [m  1]
(14) [–] δ displacement function parameter, Eqs. (20) [m  1]
Gκ,0 non-dimensional maximum modal curvature with no Δpi change in internal pressure relative to laying [Pa]
effective axial force, Eq. (42) [–] ΔT change in temperature [1C]
β
h function of β, Eq. (33) [–]. ε ¼ 10 /2, perturbation parameter [–]
Heff residual effective lay tension [N] η displacement function parameter, Eq. (18) [–]
Hf dimensional analysis assumption function for the θ ¼ dw/dx, slope [–]
fundamental frequency, Eq. (3) [–] θms, θsh slope in ms and sh coordinate systems [–]
Hκ dimensional analysis assumption function for the κmax maximum curvature for φ [m  2]
maximum modal curvature, Eq. (11) [–] v Poisson’s ratio [–]
I area moment of inertia [m4] π design vector, Eq. (44) [–]
k dynamic soil stiffness per meter [N/m2] Π design matrix, Eq. (46) [–]
ky dynamic soil stiffness per meter in lateral (y) direction πk ¼ kL4s /EI, dimensionless soil stiffness parameter, Eq.
[N/m2] (8) [–]
kz dynamic soil stiffness per meter in vertical (z) direc- π Seff ¼ SeffL2s /EI, dimensionless effective axial force para-
tion [N/m2] meter, Eq. (8) [–]
Leff effective free span length, Eq. (33) [m] πfIL dimensionless frequency parameter, Eq. (8) [–]
Ls free span length [m] πκ,max dimensionless curvature parameter, Eq. (13) [–]
Lshoulder length of span shoulder [m] φ mode shape for the fundamental mode [–]
ma hydrodynamic added mass [kg/m] ω angular frequency [rad/s]
mc mass of pipeline content [kg/m]

dependent on the modal response parameters for the span. Since analytical methods require time and thorough verification to
the number of analyses is considerable, it is desirable to have a fast implement. As a result, approximate analytical solutions have
and accurate way to predict such modal response parameters. been widely applied in free span analyses for decades (DNV’81,
Generally, the modal response may be determined to any desired 1981; DNV Guideline no. 14, 1998; DNV-RP-F105, 2002, 2006).
accuracy using finite element (FE) software (Sollund and Vedeld, The first attempts at predicting the modal response of free
2014) or appropriate semi-analytical models (Sollund et al., 2014; spans analytically was to interpret the span as a situation some-
Vedeld et al., 2013). On the other hand, FE analyses are cumber- where between a pinned–pinned and a fixed–fixed beam. It was
some and time consuming to perform parametrically and semi- customary to assume that a pinned–fixed boundary condition
H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230 219

could serve as a reasonable approximation (Xiao and Zhao, 2010). soil interaction is generally complex and depends on several
Hobbs (1986) solved for the frequency of a free span resting on soil parameters, e.g., in-situ stress conditions and loading history, the
at each end, where the soil was modeled as a linearly elastic shear strength of the soil, pipe embedment and characteristic soil
foundation. Fyrileiv and Mørk (2002) expanded on the theory of parameters such as submerged unit weight, Poisson’s ratio, void
Hobbs by including an approximation for the effect of axial forces, ratio and plasticity index. Ideally, the specific soil parameters are
based on work by Hetenyi (1946), and also considered the stiffen- determined from representative, undisturbed soil samples, but in-
ing effect of static deflections (Fyrileiv and Mørk, 2002; Vedeld et situ geotechnical testing is costly and guidance on typical para-
al., 2013). Fyrileiv and Mørk (2002) adjusted their approximation meter ranges based on less comprehensive soil classification is
by curve fitting to the results of a large number of FE analyses. given in DNV-RP-F105 (2006).
The current revision of DNV-RP-F105 (2006) includes a Conventionally, the dynamic free span response is assessed in
recommendation for calculating approximate modal response the frequency plane, with eigenfrequencies and associated modal
which is taken directly from the formulation proposed by stresses as essential input parameters. In the present study, the
Fyrileiv and Mørk (2002). Their formulation is highly accurate soil foundation is assumed to behave linearly elastically, which is
for a large number of relevant free span configurations and in line with current industry practice (DNV-RP-F105, 2006), and
pipeline operational conditions. However, the semi-empirical also natural since modal analyses are linearized procedures. The
formulations of Fyrileiv and Mørk (2002) were mainly devel- aim of the present investigation is to provide improved approx-
oped with typical offshore pipeline free spans in mind. As a imate modal response formulae for short spans, with dynamic soil
result, the lower span length limit in their studies was Ls/Ds at stiffness taken as an input parameter. It is assumed that a
30, where Ls is the span length and Ds is the outer steel pipe representative pipe–soil linearization can be achieved for both
diameter. When results are curve fitted, in the case of Fyrileiv span scenarios illustrated in Fig. 1 using e.g., the guidance for
and Mørk’s solution by an inverse second order polynomial, the pipe–soil interaction provided by DNV-RP-F105 (2006). However,
accuracy of the fitted expression generally deteriorates outside it should be recognized that there will generally be significant
the domain of the study. In the case of short spans, the accuracy inherent uncertainty in the estimated dynamic soil stiffness, and
of Fyrieliv and Mørk’s solution becomes poor for spans with sensitivity analyses are therefore appropriate when carrying out
relative lengths Ls/Ds in the region below 30. modal analyses.
As a consequence of recent failures in river crossings, much Pipelines are produced in joints with relatively short lengths.
attention has been given to existing onshore pipelines with For offshore pipelines a joint is typically 12.2 m long (DNV-OS-
river crossings or in areas of high flood risk (Heggen et al., F101, 2012). Each joint is connected by a circumferential girth
2014). There is currently no available design code for VIV design weld. If a pipeline is exposed to external flow causing VIV
of pipelines in river crossings or flooded areas, so the industry excitation, the resulting axial bending stresses occur perpendicu-
has, based on the authors’ experience, generally applied DNV- larly to the girth welds. Hence, if a crack or weld defect is present
RP-F105 (2006). Systematic experience with using DNV-RP-F105 between the girth weld and the pipeline steel material, cyclic
for very short span configurations has highlighted that the bending stresses may propagate such a crack and fatigue failure
solutions of Fyrileiv and Mørk, for the modal response quan- may ensue. Fatigue failure is therefore the most common kind of
tities, are unsuited for such spans. failure related to VIV of pipelines. Local buckling (Kyriakides and
In the present study, novel solutions for the modal response of Corona, 2007) may also occur due to combined static and dynamic
short free spans on partial elastic foundations are developed, loading (DNV-RP-F105, 2006).
responding to the need in the industry for accurate and simple A pipeline must resist a range of different types of loading—
modal response calculation methods. The solutions are developed internal and external pressure forces, thermal loads, hydrody-
by initially spanning the solution space in non-dimensional para- namic loads and static loads from its own weight. Two load
meters determined using the Buckingham Pi theorem (Logan, conditions contribute to second-order load effects for pipelines.
1997; Palmer, 2008). A highly accurate semi-analytical method Compressive or tensile axial forces increase or decrease, respec-
(Vedeld et al., 2013) is applied to create response surfaces in the tively, the bending moments in the pipe (Bazant and Cedolin,
dimensionless solution space, and various surface fitting techni- 1991), and pressure loading causes a net transverse load on the
ques are applied to create simple closed-form analytical expres- tensile side of the pipe during bending (Fyrileiv and Collberg,
sions for these response surfaces. Extensive parametric 2005; Vedeld et al., 2014). The combined second-order load effects
comparisons are made between the novel analytical solutions from axial loading and transverse loads from pressure are gen-
and solutions obtained with the semi-empirical formulae of erally accounted for by use of the effective axial force concept
Fyrileiv and Mørk (2002). (Fyrileiv and Collberg, 2005; Sparks, 1984; Vedeld et al., 2014),
which is treated as an ordinary axial force when geometric
stiffness effects are considered. The effective axial force Seff for
2. Problem definition an axially restrained pipeline is obtained as (DNV-OS-F101, 2012;
DNV-RP-F105, 2006)
2.1. Physical description and loading conditions
Seff ¼ H eff  ð1  2vÞΔpi Ai  EAs αΔT; ð1Þ

If buried pipes are dug up by erosive processes, or an offshore where Heff is the effective residual lay tension (if tension was
pipeline does not touch down on the seabed due to seabed applied during installation), Δpi is the difference in internal
roughness, scouring or an otherwise mobile seabed, the part of pressure relative to laying, Ai is the internal cross-sectional area
the pipe which is left unsupported by the soil foundation is called of the pipe, v is the Poisson’s ratio, E is the Young’s modulus, As is
a free span. Two kinds of free spans will be studied. A physical the steel cross-sectional area of the pipeline, α is the thermal
representation of potential span configurations is given in Fig. 1. expansion coefficient of the steel layer and ΔT is the change in
The areas near the span, where the pipe has contact with the soil, temperature relative to installation.
are called span shoulders, and the distance between the seabed or The mass of the pipe influences the dynamic response through
riverbed to the bottom of the pipe is called the gap. inertia forces, and generally, for two otherwise equivalent systems, a
The dynamic response of free spanning pipelines is strongly heavier system has a lower frequency than a lighter system. It is
dependent on the soil characteristics of the span shoulders. Pipe– therefore important to include all contributions to the system mass
220 H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230

Fig. 1. A pipeline dug up by erosive processes (a) and a short free span on the seabed (b).

steady hydrodynamic loading will cause a static deflection in a free


span. If a pipeline is deformed statically, the curved system
responds differently to dynamic loading than a perfectly straight
configuration. Most notably, the curvature will couple axial and
transverse degrees of freedom, which has a stiffening effect,
increasing both the modal response frequency and the associated
modal stresses (Fyrileiv and Mørk, 2002; Vedeld et al., 2013). In
cases where the deflection is more extreme, the first symmetric
mode shape becomes suppressed, and the first anti-symmetric
mode attains the lowest frequency (Fyrileiv and Mørk, 2002;
Fig. 2. Model of shoulder supports, boundary conditions and free span geometry. Sollund et al., 2014; Vedeld et al., 2013). In the present study,
however, the spans are very short and influence from initial
curvature or static deflections are considered negligible. As a
when predicting the frequency. For free spanning pipelines, the total result, axial deformations and axial soil resistance may be dis-
system effective mass me includes the following components regarded as well. The idealization of the geometry and shoulder
supports given in Fig. 2 will be adopted in this study.
me ¼ ms þ mc þ ma ; ð2Þ
As seen from the figure, shoulders are modeled with a length
where ms is the structural mass, mc is the content mass and ma is the Lshoulder, the horizontal soil stiffness is assumed linear with a stiffness ky
hydrodynamic added mass. Pipelines may have internal corrosion and the vertical soil stiffness is also assumed linear with a stiffness kz.
protection layers (Vedeld et al., 2012a, 2012b), external weight coat- Deformations in the horizontal (y-) direction are denoted v and
ings (Ness and Verley, 1996), thermal insulation coatings (Choqueuse deformations in the vertical (z-) direction are denoted w. In the
et al., 2002; Guan et al., 2005; Vedeld and Sollund, 2014) or idealized model, the shoulders should be long enough for the
combinations thereof. Hence, the total structural mass includes the frequency to be independent of their length (i.e., Lshoulder-1).
weight of the steel wall plus all other layers, and the added mass must Sollund and Vedeld (2014) and Vedeld et al. (2013) demonstrated
be calculated based on the outer hydrodynamic diameter of the pipe. that Euler–Bernoulli beam theory is highly accurate when predicting
Hydrodynamic added mass and hydrodynamic damping are modal response parameters for pipeline free spans. Effects of shear
dependent on the gap between the pipe and the seabed, the flow deformation have therefore been disregarded in the present study. For
velocity and the vibration amplitude (Sumer and Fredsøe, 2006). very short spans, it may be noted that the pipeline frequency will be
Hence, the pipeline’s VIV response frequency will generally differ from expected to approach the frequency of an infinitely long beam on
its still-water eigenfrequency. These effects are implicitly accounted for elastic foundation. Consequently, the pipeline should still be regarded
by the empirically based VIV response models included in DNV-RP- as slender, and shear deformation is not expected to become increas-
F105 (2006). Since the response models take the still-water eigenfre- ingly important with decreasing span length, as would be expected for
quency as input, the present study will focus on modal analyses for a pinned–pinned or fixed–fixed beam.
still-water conditions. Note that experimental studies have also
demonstrated a complex coupled response between the pipe, soil
and surrounding flow, and local scouring beneath the free span has 3. Analysis
been observed to affect the pipeline response frequency and vice versa
(Gao et al., 2006; Yang et al., 2008). Such boundary effects are, 3.1. Dimensional analysis
however, not yet accounted for in conventional pipeline design
practice, and will not be considered herein. The Buckingham Pi theorem (Logan, 1997; Palmer, 2008) is
The present study is concerned with very short span config- applied in order to identify non-dimensional groups of importance
urations under extreme environmental conditions. If carefully for determining frequencies and modal stresses for pipeline free
designed for, VIV response of a long slender structure like for spans. The frequency f ¼ ω/(2π), where ω is the angular frequency,
instance a riser (Mukundan, 2008) may be non-critical, and VIV and the modal bending stress A are observed to depend on the
can in some cases be a daily phenomenon without causing any following input parameters (with units indicated in parentheses):
significant damage to the structure. For a stiff pipeline with a short
span, however, VIV will cause significant bending stresses and in  Effective mass me (kg/m)
some cases the mere onset of VIV may be a critical design  Bending stiffness EI (N m2)
condition. Since the spans considered in the present paper are  Span length Ls (m)
generally very short, only the fundamental frequency and the  Dynamic soil stiffness k (N/m2)
associated modal stress will be considered in this study.  Effective axial force Seff (N)

2.2. Mathematical model For the frequency it is reasonable to assume that there exists a
relation
In general free spans, the combined loading from the effective
axial force, the submerged weight of the pipe and potentially H f ðf ; me ; EI; Ls ; k; Seff Þ ¼ 0 : ð3Þ
H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230 221

According to the Buckingham Pi theorem, an equivalent phy- By assuming a relation on the form
sical law should exist between the non-dimensional groups
H κ ðκ max ; me ; EI; Ls ; k; Seff Þ ¼ 0; ð11Þ
obtained by determining the basis of the null space to the
dimensional matrix (Logan, 1997). For the present problem, the and performing the dimensional analysis in the same way as
dimensional matrix becomes described for frequency, Eqs. (4)–(6), the Buckingham Pi theorem
gives the equivalent relation
me EI Ls KL Seff f IL  
2 3 F κ π k ; π Seff ; π κmax ¼ 0; ð12Þ
kg 1 1 0 1 1 0
6 7 πk and π S are defined by Eq. (8), while
m 4 1 3 1 1 1 0 5; ð4Þ where eff

s 0 2 0 2 2 1 π κ max ¼ κ 2
max Ls : ð13Þ
and the non-dimensional groups are thus obtained by solving the By solving for κmax in Eq. (12) and inserting the result into Eq.
equation (10), the following expression is obtained for the maximum modal
2 3 stress:
1 1 0 1 1 0
6 7 1 DEDs  
4 1 3 1 1 1 0 5x ¼ 0: ð5Þ
A¼ Gκ π k ; π Seff : ð14Þ
0  2 0  2  2 1 2 L2s

The solution to Eq. (5) is Gf and Gκ from Eqs. (9) and (14) are response surfaces for the
2 3 2 3 2 3 frequency and curvature respectively, of a harmonically excited pipe-
0 0 0:5
6 1 7 6 1 7 6  0:5 7 line in a free span. These response surfaces are independent of pipe
6 7 6 7 6 7 geometry and loading conditions and hence, if reasonable approxima-
6 7 6 7 6 7
6 4 7 6 2 7 6 2 7 tions for these response surfaces are developed, the response fre-
6
x ¼ x4 6 7 6
þ x5 6 7 6
þ x6 6 7; ð6Þ
7 7 7
6 1 7 6 0 7 6 0 7 quency and associated modal stress will be determined.
6 7 6 7 6 7
4 0 5 4 1 5 4 0 5 For realistic free-span scenarios, the non-dimensional para-
0 0 1 meter πk will have a range of several orders of magnitude. It is
therefore convenient to introduce a logarithmical scaling of this
and according to the Buckingham Pi theorem, the relation parameter, defined by
 
F f π k ; π Seff ; π f ¼ 0; ð7Þ β ¼ log π k : ð15Þ
is therefore equivalent to Eq. (3), where
sffiffiffiffiffiffiffiffiffiffiffi
3.2. Buckling of short spans
kL4s Seff L2s me L4s
π k ¼ ; π Seff ¼ ; πf ¼ f : ð8Þ
EI EI EI Prior to establishing suitable domains for the response surfaces
By solving for f, we obtain a functional relationship Gf such that Gf and Gκ it is necessary to determine the buckling load Pcr, which
sffiffiffiffiffiffiffiffiffiffiffi will constitute a natural upper limit for compressive values of the
EI   effective axial force Seff. Fig. 3 shows an infinitely long beam on an
f¼ 4 f
G π k ; π Seff : ð9Þ
me L s elastic foundation, with an unsupported span of length Ls and a
concentrated force R acting in the middle of the span. The beam is
The maximum unit diameter amplitude stress A is the max- also subjected to an axial force P, conveniently defined as positive
imum bending stress associated with the particular mode shape in compression in order to obtain a positive value for Pcr (note that
when the maximum displacement is normalized to one outer Seff elsewhere in this study is defined as positive in tension).
diameter. A may also be determined by using the Buckingham Pi The differential equation for transverse displacements w of the
theorem. The maximum stress (at the outer fiber of the pipe steel beam axis is given by (Hetenyi, 1946; Bergan and Syvertsen, 1989)
wall) may then be obtained by
4 2
 2  d w d w
M max Ds 1 ∂ φ 1 EI þP 2 þ kw ¼ 0; ð16Þ
A¼  DEDs ¼ DEDs κ max ; ð10Þ dx4 dx
I 2 2 ∂x2 max 2
where a uniform bending stiffness EI has been assumed, and
where Mmax is the maximum bending moment associated with the where k is the stiffness of the soil. For k ¼0 (i.e., in the span) the
relevant mode, I is the area moment of inertia, Ds is the outer steel solution to Eq. (16) is readily obtained as (Bergan and Syvertsen,
diameter and φ is the mode shape. In DNV-RP-F105 (2006), the 1989)
normalization constant D is the outer hydrodynamic diameter, i.e.,    
xms xms
the outer diameter of the pipe including all coating layers. The VIV wms ðxms Þ ¼ C 1 sin 2η þ C 2 cos 2η þ C 3 xms þ C 4 ; ð17Þ
response amplitudes are calibrated to a normalized displacement Ls Ls
amplitude of a unit hydrodynamic diameter D (DNV-RP-F105,
2006).
We wish to apply the Buckingham Pi theorem in a similar
manner as for frequency, in order to obtain the non-dimensional
groups relevant for obtaining κmax. It should be noted that the
mode shape, prior to normalization, should be considered as a
dimensionless quantity. Therefore, when applying the Bucking-
ham Pi theorem to obtain the relevant non-dimensional groups for
predicting the maximum curvature κmax, the entry corresponding
to κmax in the dimensional matrix will be given the value (  2).
Fig. 3. An infinitely long beam subjected to a concentrated force R and axial force P
Equivalently, one may apply κmax  Ls with dimension (length)  1 resting on partial elastic foundation with a free span of length Ls. The coordinates
and still maintain geometric similarity (since Ls as opposed to D is xms and xsh are directed along the pipe axis, but are zero at mid-span and at the
a length that is characteristic for the particular problem). interface between span and shoulder, respectively.
222 H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230

where we have introduced a longitudinal coordinate xms, defined An approximate solution to Eq. (26) can be obtained by
to be zero in the middle of the span (Fig. 3) and where expanding zb as a power series in ε about the unperturbed root
rffiffiffiffiffi zb ¼0
L P
η¼ s : ð18Þ
zb;PER ðεÞ ¼ εz1 þ ε2 z2 þ ε3 z3 þ… ð27Þ
2 EI
On the elastic shoulders, the solution to Eq. (16) can be It can be shown by numerical solution of Eq. (25) that zb {1 for
expressed as (Hetenyi, 1946) values of β o 0. Thus, in Eq. (26) zb may now be replaced by the
      power series from Eq. (27) and the cosine-term by its Taylor series,
wsh ðxsh Þ ¼ D1 eδxsh þ D2 e  δxsh cos γ xsh þ D3 eδxsh þD4 e  δxsh
  resulting in the transformed equation
sin γ xsh ; ð19Þ  
π 2 εz 1 þ ε2 z 2 þ ε3 z 3 þ …

provided P o2(k  EI)1/2. The coordinate xsh is defined to be zero at 1 π 2   1 π 4
the point B defined in Fig. 3, and the parameters γ and δ are given ¼ ε 1þ 1 U εz1 þ ε2 z2 þ ε3 z3 þ … þ U
2! 2 4! 2
by  2 o
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffiffi ffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffiffi ffi εz1 þ ε z2 þ ε z3 þ … …
2 3

k P k P  4 
γ¼ þ and δ ¼  : ð20Þ π2 π 2 π2
4EI 4EI 4EI 4EI ¼ 2ε  z 1 ε2 þ z 1  z 2 ε3 þ … ð28Þ
8 384 8
From the trivial assumption that the displacement wsh and
By equating powers of ε on both sides of Eq. (28) and
bending moment Msh-0 when xsh-1, it follows that D1 ¼D3 ¼0.
truncating the resulting series after three terms, we obtain the
The remaining six undetermined coefficients in Eqs. (17) and (19)
fourth order approximation
may be obtained from the following boundary conditions and    
continuity requirements: 2 1 1 2 1 1
zb;PER ðεÞ ¼ 2 ε− ε2 þ ε3 ¼ 2 10β=2 − ⋅10β þ ⋅103β=2 : ð29Þ
  π 8 48 π 8 48
ðiÞ θms ð0Þ ¼ 0; ðivÞ θms L2s ¼ θsh ð0Þ;
  It is physically obvious that the buckling load Pcr should
R Ls
ðiiÞ Q w;ms ð0Þ ¼  ; ðvÞ M ms ¼ M sh ð0Þ ¼ M AB ; approach the buckling load for an infinitely long beam on a
2 2
    continuous elastic foundation as Ls-0. Hetenyi (1946) showed
Ls Ls R that the buckling load for such beams is given by
ðiiiÞ wms ¼ wsh ð0Þ; ðviÞ Q w;ms ¼ Q w;sh ð0Þ ¼  :
2 2 2 pffiffiffiffiffiffiffi
ð21Þ P cr;BEF ¼ 2 kEI : ð30Þ

In Eq. (21), θ ¼dw/dx is the slope, M ¼ EI  d2w/dx2 is the From the definition of the parameter zb, Eq. (24), it follows that
pffiffiffiffiffiffiffi sffiffiffiffiffiffiffi
bending moment, and Qw ¼  EI  d3w/dx3  P  dw/dx is the ver-
P cr;BEF 2 kEI 2 kL4s 2
tical shear force. zb;BEF ¼ ¼ ¼ ¼ 10β=2 ð31Þ
The bending moment MAB at the interfaces between the span
PE π 2 EI=L2s π 2 EI π 2
and the shoulders is found to be As expected, it is observed that zb,PER-zb,BEF when ε-0.
 2   DNV-RP-F105 (2006) applies an effective length concept, in
Ls 1  cos η
R 3δ2  γ 2  2η
1
cos η which the effective length Leff is defined as the length of a fixed–
M AB ¼   : ð22Þ
2 2δ
þ 2Lηs tan η fixed beam with the same buckling load as the actual free span.
3δ  γ
2 2
Hence, the exact value of Leff based on the solution to Eq. (25) can
From Eq. (22) it is evident that buckling will occur when the be found from
denominator  2 sffiffiffiffiffi
  P cr 4π 2 EI=L2eff Ls Leff 4
2δ Ls zb ¼ ¼ ¼ 4 ) ¼ : ð32Þ
þ tan η ¼ 0: ð23Þ PE π 2 EI=L2s Leff Ls zb
3δ  γ 2
2 2η
Effective length expressions can similarly be obtained for
Let Pcr denote the value of P fulfilling Eq. (23). After performing the asymptotic solutions by replacing zb in Eq. (32) with z b,PER
the substitution or z b,BEF .
P cr L2s P cr The expression for effective length Leff in DNV-RP-F105 (2006)
zb ¼ ¼ ; ð24Þ may be written as
π 2 EI P E
Leff 4:73
the buckling condition, Eq. (23), may be rewritten as: ¼  ; ð33Þ
sffiffiffiffiffiffiffi Ls h β
π 2 zb kL4s
π pffiffiffiffiffi ¼ ¼ 10β=2 : ð25Þ where
1 þ cos 2 zb EI 8
  <  0:066β þ 1:02β þ 0:63 β Z 2:7;
2
for
The transcendental Eq. (25) has previously been derived (albeit h β ¼ ð34Þ
: 0:036β þ 0:61β þ 1:0
2
for β o 2:7:
differently) by Hetenyi (1946).
The present study deals with short and very short spans, and
In the present study, the exact function h(β) was calculated
consequently it is of interest to investigate the asymptotic proper-
from the condition for buckling by solving Eq. (25) using Newton’s
ties of Eq. (25) as the span length Ls-0, corresponding to β-  1.
method (Adams, 1999). In Fig. 4 the exact function has been
From inspection of Eq. (25) it is clear that zb-0 when β becomes
β plotted along with the corresponding expression from DNV-RP-
increasingly negative. A parameter ε ¼ 10 /2 is introduced, with
F105.
ε{1 (i.e., β o  1). Eq. (25) may then be reformulated as a
It is observed that the recommended practice is highly accurate
perturbation problem:
 π pffiffiffiffiffi in the range 0 o β o8, but deviates from the exact solution outside
π 2 zb ¼ ε 1 þ cos z ð26Þ this interval. Therefore a new expression for h(β) has been derived
2 b for very low β by means of ordinary least squares curve fitting. The
H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230 223

3.3. Domain of the response surfaces

When the Buckingham Pi theorem is applied, the number of


governing parameters for a problem is reduced. In the present
study, the number of independent parameters to determine either
a modal frequency or the associated modal stress was reduced
from five to two. The immediate consequence is that the response
surface is reduced from a five-dimensional hyper-surface (in a six-
dimensional space) to a surface in three-dimensional Euclidian
space which is, contrary to its five-dimensional counterpart, easy
to visualize. The reduction in number of governing parameters
comes at a price, however. The two dimensionless parameters πk
and π Seff are themselves functions of EI, k, me, Seff and Ls, and it is
not trivial to determine reasonable domains for them.
In order to determine a relevant domain, a selection of 48
actual pipelines was chosen, having outer diameters ranging from
Fig. 4. Comparison between h(β) obtained from exact solution and from DNV-RP- 51 mm to 1.2 m. The diameter to wall thickness ratio Ds/ts was
F105. varied between 8 and 111. Details for the pipeline geometries are
given in Appendix A. To account for reasonable effective axial force
loading, the pressures, temperatures and steel wall axial forces
were varied randomly and well outside the ranges of typical
pipeline operational limits. However, all cases where the Von
Mises stress in the pipe wall exceeded 400 MPa were disregarded.
Typical pipelines have yield stresses in the range of 450 MPa, so
combinations of operational parameters which yield higher Von
Mises stresses than 400 MPa are rare since some residual capacity
in the pipe must be assumed, to accommodate safety philosophies
(API RP 1111, 2009; ASME B31.8, 2007; DNV-OS-F101, 2012) and
capacity for bending utilization. An additional limitation was set
on Seff, where the compressive effective axial force was limited to
0.7Pcr, where Pcr was calculated from Eqs. (33)–(36). Span lengths
were varied for Ls/Ds in the range between 1 and 30, where 30 is
the lower limit for the current approximate modal response
quantities developed by Fyrileiv and Mørk (2002). 5 soil types
Fig. 5. Effective length ratio Leff/Ls as a function of β according to the exact solution,
were selected based on guidance given in DNV-RP-F105 (2006),
the expression from DNV-RP-F105, the solution for beam on elastic foundation and randomly varied to account for realistic variation in soil
(BEF) and the new expression (only β o 0). stiffness. Tabulated values for the dynamic soil stiffness for each
soil type are presented in Appendix A.
resulting expression reads The randomized study led to two sets of envelope curves which
  give suitable ranges for β and π Seff . The curves are given by
h β ¼ 0:00861β þ 0:107β þ 0:5145β þ 1:0  3 r β o 0:
3 2
for
ð35Þ 3rβ r2

By using the novel expression given by Eq. (35) one may  0:8 U10β=2 r π Seff r0:8 U 10β=2 ð37Þ
calculate the buckling load for free spans in the low-β range from
the conventional formula for fixed–fixed beams:

4π 2 EI 3.4. Dynamic response of an axially loaded beam on an elastic


P cr ¼ : ð36Þ foundation
L2eff

Fig. 5 compares the effective lengths obtained using the exact The asymptotic behavior for free span frequencies when the
solution to Eq. (25), the expression in DNV-RP-F105, the solution span length Ls approaches zero may, in a similar manner as for the
for an infinitely long beam on a continuous elastic foundation buckling load in Section 3.2, be studied by considering the
(BEF) and the new low-β fit given by Eq. (35). Again it is seen that dynamic response of an infinitely long beam on a continuous
DNV-RP-F105 is inaccurate for β o 0. It is further demonstrated elastic foundation. Obviously, when Ls-0 the physical problem
that the new expression predicts the exact effective length illustrated by Fig. 2 is reduced to vibration of an axially loaded
accurately, and that the exact solution converges quickly toward beam with length 2Lshoulder on an elastic foundation. With a
the BEF solution when β becomes negative. In fact, the relative pinned–pinned configuration, a Rayleigh-Ritz (Shames and Dym,
difference between the BEF and the exact solution becomes less 1991) approach with a Fourier sine series (Dirichlet, 1829) is
than 1% below β E  1.6. The solution corresponding to the appropriate in order to determine a formula for the harmonic
asymptotic expansion zb,PER is omitted from Fig. 5 for clarity. The response frequency and associated mode shape. A pinned–pinned
accuracy of this solution is found to be excellent (within 0.2%) for condition is considered suitable for the present applications since
β o 0, but deteriorates quickly for larger β. An effective length the span shoulders are assumed sufficiently long as for the
based on zb,PER could thus have been applied for all β o0, but the boundary condition to be inconsequential. Timoshenko (1937)
new expression for h(β), Eq. (35), is preferred here due to its has made a similar deduction for the case with zero axial loading
consistency with the current format in DNV-RP-F105. and it is trivial to expand his approach to include an axial load. The
224 H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230

resulting frequency is interpretation of the modal response ceases to make sense (Hobbs,
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1986). Interestingly, if the second mode has reached the frequency
u 4  2
u nπ nπ
1 t 2Lshoulder EI þ 2Lshoulder Seff þ k for an infinitely long beam on an elastic foundation, all frequencies
f BEF ¼ ð38Þ with a higher mode number than two will also have reached the
2π me
same frequency. In a VIV context, when the frequencies of all
where fBEF is the frequency for a beam on an elastic foundation and modes collapse onto one another, in practice any of the modes
n is the number of half wave-lengths in the mode shape. Notably, could be excited. If any of the modes may be excited, the relevant
when Seff Z 0, fBEF converges to modal stress has become undetermined. For an effective axial
sffiffiffiffiffiffi
force equal to zero, Eq. (39) is fulfilled for the second mode when
1 k
f BEF ¼ ð39Þ β E1.2 (Sollund et al., 2015). As a result, it makes little sense to
2 π me
evaluate higher order modal stresses for β o1.2, as the frequency
when the length of the elastic foundation goes to infinity. When of all higher order modes will be the same, and the VIV problem is
Lshoulder-1 and Seff o0 it can be shown (Sollund et al., 2015) that reduced, more or less, to an onset problem. In other words, fatigue
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!ffi vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi calculations should not be performed, but rather a design check of
u u   !
1ut k S2eff 1ut k Seff 2 whether VIV occurs or not will be appropriate, unless more
f BEF ¼ 1 ¼ 1 : ð40Þ
2 π me 4kEI 2 π me P cr;BEF sophisticated dynamic analyses are conducted where damping is
also considered.
Some important observations may be made from Eqs. (39) and
(40):
3.5. Expressions for the response surfaces Gf and Gκ
 The frequency does not depend on the mode number or the
bending stiffness. In tension, the frequency is also independent According to Eqs. (9) and (14), closed analytical expressions for
of the axial load. the modal frequency and associated modal stress may be deter-
 The mode shape does not influence the frequency, implying mined if the surfaces Gf and Gκ can be determined. A transcen-
that all the mode shapes have the same frequency. dental equation for obtaining the exact frequency and mode shape
for an axially loaded free spanning pipeline was recently derived
Of course, in reality, if we include effects of damping and non- by Sollund et al. (2015). However, in the present context a semi-
linear soil behavior the mode number will certainly influence the analytical solution (Vedeld et al., 2013) to the free span problem
frequency. In the idealized condition however, where damping is with the same level of accuracy as finite element (FE) solutions
disregarded and the soil is assumed uniform and linearly elastic, will be preferred for calculation of frequencies and modal stresses.
the frequency is not mode shape dependent. As a result, the This method is significantly faster than FE and hence well suited
associated modal stresses become undetermined when the length for parametric studies to develop accurate representations of Gf
of the beam approaches infinity. Hence, when spans become very and Gκ. It was found that shoulder lengths of 100 m gave
short and the free span modal frequencies converge towards Eq. convergent behavior for the fundamental frequency and its asso-
(39) or Eq. (40), it is important to bear in mind that the associated ciated modal stress. An evenly spaced grid of 1900 points was
stresses are not necessarily relevant, since a number of modes may generated over the surface in the β  π Seff plane defined by Eq. (37),
share the same frequency. and Gf and Gκ were calculated based on the semi-analytical
The implications of the asymptotic behavior of the modal method developed by Vedeld et al. (2013). The resulting surface
frequencies should be emphasized. The fundamental frequency plots of Gf and Gκ are given in Fig. 6.
will, as stated above, converge to the solution for an infinitely long Necessarily, a near perfect relation could be established if for
beam on an elastic foundation when the span length tends to zero. instance all 1900 points were tabulated and one could simply
However, the frequency of an infinitely long beam on an elastic interpolate between them, or a large polynomial surface function
foundation is not mode dependent, implying that the second could be fitted. In the present context the objective is, however, to
mode will have a similar convergent behavior. Since the second create simple analytical formulae, suitable for implementation in
mode has a higher frequency than the fundamental mode, the design standards. Hence, significantly simpler expressions for the
convergence behavior will be more rapid, and the second mode solutions are sought. From Fig. 6 it is observed that both surfaces
will reach the frequency for a beam on an elastic foundation for tend to zero for β in the region below zero. This is not surprising,
spans longer than zero, i.e. the second mode will fulfill Eqs. (39) considering that both Gf and Gκ are multiplied by a factor of L2s , as
and (40). For infinitely long shoulders, the mode shape for the seen from Eqs. (9) and (14), which tends to zero for small β.
second mode will be infinitely long as well, and the physical Surfaces which tend to zero pose a problem for surface fitting

Fig. 6. (a) The non-dimensional frequency function Gf and (b) the non-dimensional curvature function Gκ.
H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230 225

processes since relative deviations to the surfaces will be high in (Tofallis, 2008):
regions where the values are close to zero, even if absolute errors  1
p ¼ Π D2 Π ΠT D2 g;
T
are small. Hence, it is reasonable to normalize the surfaces Gf and ð45Þ
Gκ to make them more constant over their domain.
where the jth row of the design matrix Π and the response vector
For small Ls it is expected that the solution for frequency will
g are given by
converge towards the solution for a beam on a continuous elastic
h i h i
Πj ¼ πTj ¼ 1 βj π Seff ;j βj βj π Seff ;j π 2Seff ;j … ;
foundation. If Eq. (39) is transformed to the same non-dimensional 2

space as Gf and Gκ, a new surface Gf,BEF can be formulated


sffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffi Gf  
sffiffiffiffiffiffiffiffiffiffiffi gj ¼ β ;π for j A f1; …; 1900g; ð46Þ
EI 1 k 1 me L4s k 10β=2 Gf ;BEF j Seff ;j
f BEF ¼ G ;BEF ¼ ) G ;BEF ¼ U ¼
me L4s
f
2π me f
2π EI me 2π
and the matrix D is a diagonal weight coefficient matrix given by
ð41Þ (1
gj ; i ¼ j
No similar solution may be found for the stresses, however, Dji ¼ ð47Þ
0; ia j
since the mode shape for an infinitely long beam on an elastic
surface is undetermined. Hence, a different normalization was Based on the surface fitting process in Eq. (45), and combining
sought for Gκ. It was found that a simple curve fit for Gκ for the line Eqs. (9) and (41), it was ultimately found that the frequency f, for a
π Seff ¼ 0 in the region  3 r β o0 was effective. This curve Gκ,0 is very short span can be calculated as:
given as f ¼ gf BEF ð48Þ
Gκ ;0 ¼ 0:389e1:41β ð42Þ where g, the fitted expression for Gf /Gf,BEF, is given by
If Gf is normalized with respect to Gf,BEF and Gκ is normalized to g ¼ 0:883  0:0361β þ 0:0863π Seff  0:177βπ Seff ; 3rβ o0
Gκ,0, the resulting surfaces are more constant and no longer tend to ð49Þ
zero, as seen from Fig. 7.
and
The surfaces presented in Fig. 7 are considered more suitable
for surface fitting, and a range of different techniques was tried. g ¼ 0:86  0:0865β þ 0:101π Seff  0:0402β  0:0662βπ Seff  0:00907π 2Seff
2

The objective of the surface fitting was to minimize the mean and
þ 0:0135β π Seff þ 0:00398βπ 2Seff ; 0rβr2
2
maximum relative deviations between the approximated surface ð50Þ
and the solutions generated via the semi-analytical method, while
As seen from Fig. 7, there is a sharp sensitivity of the surface Gκ/
keeping the functional relationships as simple as possible. It was,
Gκ,0 to π Seff , particularly in the low β regions. Polynomial surfaces
for those reasons, found reasonable to divide the two surfaces Gf
worked with poor accuracy, so instead a different family of curves
/Gf,BEF and Gκ/Gκ,0 into two parts, and making surface fits for the
was chosen. The approach which gave the best results was to
two regions β o0 and β Z0 separately. For the surface Gf /Gf,BEF, it
curve fit the surface along a set of lines gA where β was kept
was determined that polynomial surfaces struck a good balance
constant. This family of lines gA could all be fitted using second
between accuracy and solution complexity. The surface may be
order polynomials:
expressed as
g A ¼ a þbπ Seff þ cπ 2Seff ð51Þ
Gf  
β; π Seff ¼ πT p ¼ p00 þ p10 β þ p01 π Seff þ…; ð43Þ
Gf ;BEF where the constants a, b and c varied for each constant value of β.
The constants a, b and c were thus curve fitted as functions of β
where a design vector π and a coefficient vector p have been
defined by 3rβ o0 :
a ¼  0:0235β  0:0262β þ0:998
2
h i
πT ¼ 1 β π Seff β βπ Seff π 2Seff ⋯ ;
2
b ¼  0:4167e  1:476β
h i
pT ¼ p00 p10 p01 p20 p11 p02 ⋯ : ð44Þ c ¼ 0:1459e  2:904β ð52Þ
and
The coefficient vector p is obtained by least squares minimiza-
tion of the relative deviation according to the following relation 0rβr2 :

Fig. 7. (a) The normalized non-dimensional frequency function and (b) the normalized non-dimensional curvature function.
226 H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230

a ¼ 0:0678β  0:0926β þ 0:993


2 frequency and modal stress results, given that the total number of
cases is 8000. This notion is supported by the finding that the root
b ¼  0:4216e  1:542β
mean square (RMS) of the relative error is 1.4% and the standard
c ¼ 0:1398e  2:868β ð53Þ deviation (STD) for the relative error is also 1.4% for the predicted
Based on Eqs. (14), (42) and (51), the final expression for the frequencies. Similarly, for the associated modal stresses, the RMS
maximum modal stress A is then relative error is 1.8% and the STD for the relative error is 1.7%.
It was described in Section 3.4 how the fundamental frequency
1 DEDs
A¼ g Gκ ;0 ð54Þ will approach the frequency fBEF of an infinitely long beam on
2 L2s A
elastic foundation, Eqs. (39) and (40), when β becomes very small.
For the same 8000 randomly selected cases considered in Fig. 8,
the frequency was also estimated using the fBEF expressions. The
4. Results and discussions results are shown for β o 0 in Fig. 9. It is observed that the relative
error compared to semi-analytical results as expected decreases
4.1. Accuracy of the frequency and stress functions quickly when β becomes very small. The relative error is generally
less than 2% at β E  3, and Eqs. (39) and (40) may thus be
Novel modal response formulae, given by Eqs. (48) and (54), recommended for β o 3. Modal stresses are not relevant for
were derived in Section 3.5. These formulae are valid on the values of β in that region, as discussed in Section 3.4.
domain established in Section 3.3 and specified by Eq. (37), and
the accuracy of the formulae will now be thoroughly assessed. The
asymptotic behavior for very low values of β will also be
considered. 4.2. Results from the semi-empirical method in DNV-RP-F105 for low
The response surfaces for frequencies and associated modal β
stresses developed in Section 3.5 were modeled based on uni-
formly spaced grids. To achieve a good validation of the proposed In Section 4.1, 8000 cases were randomly selected, and results
formulae for frequency, Eq. (48), and for associated modal stress, were compared between the novel formulae given in Eqs. (48) and
Eq. (54), solutions at random selections of points on the domain (54) and the semi-analytical method of Vedeld et al. (2013). In this
defined by Eq. (37) were applied. To achieve realistic pipeline free section, the semi-empirical method from DNV-RP-F105 (2006) is
span scenarios, the 48 pipelines and 5 soil types given in Appendix compared to the semi-analytical method for the same 8000 cases.
A were randomly combined 8000 times. For each combination of From Figs. 4 and 5, it is evident that the semi-empirical model in
pipe and soil type, a random value for β in the interval (  3, 2) was DNV-RP-F105 ceases to be representative for β o 0. Hence, results
selected and the free span length adjusted such that β achieved have been divided into the two regions  3 r β o0, and 0 r β r2,
the correct value given the selected soil type and pipe cross- to investigate the accuracy in the region where the model is
section. Afterwards, π Seff was randomly chosen from the interval in supposed to be accurate, and in the region where we expect it to
Eq. (37), from which Seff was calculated according to Eq. (8). After deteriorate.
establishing the randomized grid for β and π Seff , along with In the region 0 r β r2, the semi-empirical method in DNV-RP-
randomized realizations of bending stiffness and soil stiffness, a F105 is compared to the semi-analytical method of Vedeld et al.
large-scale check of the proposed formulae in Eqs. (48) and (54) (2013) in Fig. 10. In the figure, fF105 is the fundamental frequency
was conducted by comparing them to the semi-analytical method and AF105 is the unit diameter amplitude stress, both calculated
of Vedeld et al. (2013). according to DNV-RP-F105 (2006).
As functions of β, the error of the approximations in Eqs. (48) From Fig. 10 it is observed that the frequency calculated
and (54) relative to the semi-analytical method are given in Fig. 8 according to DNV-RP-F105 is generally too high compared to the
for the 8000 randomly selected cases. In Fig. 8 fSA and ASA are the more accurate prediction, and furthermore that the relative errors
fundamental frequency and the associated maximum unit dia- are moderately large, ranging from 30% to  37%. The RMS relative
meter modal stress calculated using the semi-analytical method. error is 18% with an STD of 11%. For the stresses, the inaccuracy is
As seen from Fig. 8, the relative error of the frequency significantly larger, but generally the stresses are conservatively
approximation is maximally about 3% and minimally about  8%. overestimated. The RMS relative error is 69% with an STD of 41%.
The relative error of the stress surface has an upper limit of 6.5% Compared to the new formulations given in Eqs. (48) and (54), the
and a lower limit of  10%. The number of cases where the relative existing semi-empirical equations in DNV-RP-F105 are signifi-
error exceeds two percent is, however, quite small both for the cantly less accurate.

Fig. 8. The relative errors of (a) the approximate frequency response surface and (b) the approximate unit diameter amplitude stress surface, for 8000 randomly
selected cases.
H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230 227

For lower values  3 r β o0, the solutions for frequencies and From the figure, it is observed that the inaccuracy of DNV-RP-
stresses are expected to deteriorate, since the effective length F105 for  3 r β o0 is more or less extreme, with up to a factor of
formulation in DNV-RP-F105 was demonstrated in Section 3.2 to 8 on the frequency and a factor of 70 on the modal stress. Hence, it
be highly inaccurate in this region. From Fig. 10 it is also an is not recommended to apply DNV-RP-F105 in this β-region.
apparent trend that the accuracy is decreasing with decreasing β. Comparatively, the inaccuracy of the novel approximate formula-
The results for  3 r β o 0 are presented in Fig. 11. tions given in Eqs. (48) and (54) is limited to only a few percent.

4.3. Simplified boundary condition approaches for short spans

As mentioned in the introduction, simplified boundary condi-


tions such as pinned–pinned or fixed–fixed have been applied
historically to predict response frequencies for free spanning
pipelines. In the present study, the novel analytical solutions are
compared to results for pinned–pinned and fixed–fixed boundary
conditions for the case of zero effective axial force. The results are
given in Fig. 12.
As seen from the figure, using idealized boundary conditions
makes the non-dimensional frequency and modal curvature inde-
pendent of β. As a consequence, frequencies and stresses will be
reasonably approximated only for a limited range of β if an idealized
boundary condition is applied. Interestingly, as β increases, the
solutions for both the frequency and the modal stress converge
towards the fixed–fixed solutions, which is to be expected since
Fig. 9. Relative errors using the frequency fBEF of an infinitely long beam on elastic increasing the span length or the soil stiffness, or both, will cause the
foundation for free spans in the low-β region. elastic foundation to behave more like a fixed condition. More

Fig. 10. Relative differences between the semi-empirical model from DNV-RP-F105 and the semi-analytical model by Vedeld et al. (2013) for (a) fundamental frequencies and
(b) unit diameter modal amplitude stresses, for 0r βr 2.

Fig. 11. Relative differences between the semi-empirical model from DNV-RP-F105 and the semi-analytical model by Vedeld et al. (2013) for a) fundamental frequencies and
b) unit diameter modal amplitude stresses, for  3r βr 0.
228 H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230

Fig. 12. Gf (a) and Gκ (b) are presented as functions of β for free spans resting on a partial elastic foundation (Free span) and for the idealized boundary conditions pinned–
pinned (Pinned) and fixed–fixed (Fixed).

surprisingly, the pinned–pinned condition becomes significantly  The novel analytical formulae are highly accurate (to within a
stiffer than the elastic foundation solution for β lower than about few percent) and constitute a significant improvement to
3. This implies that the elastic foundation response is softer than free current analytical approaches. Hence, the formulae are recom-
rotation at the span ends, and furthermore that any simplified mended for use in future revisions of industrial design
boundary condition will be non-conservative in this β-region since standards.
a higher modal frequency will require higher flow velocities for VIV  Simplified boundary condition approaches, using pinned–
to be excited. Consequently, simplified boundary conditions may in pinned or pinned–fixed boundaries, cannot be applied for
some cases be applied reasonably for β between 3 and 10 but for short spans and influence VIV design in a non-conservative
lower β, simplified boundary conditions will yield non-conservative manner.
estimates for the modal frequency.
On a side note, it may be observed from Fig. 12b) that there
appears to be a discontinuous derivative for β equal to 4.25. This is Acknowledgements
in fact to be expected, since the stress curve is actually composed
of two curves, one for maximum stress at the shoulder and one for
maximum stress mid-span. For β E4.25, these two curves inter- The authors would like to extend their gratitude to Prof.
sect and their difference in slope is visible from the figure. Karsten Trulsen at the University of Oslo for helpful and insightful
comments on the topic of asymptotic analysis. The research has
been financed by the University of Oslo and the DNV Scholarship
Fund.
5. Conclusions

 Easily implementable, closed, analytical formulae for the fun- Appendix A. Pipeline cross-sections and soil stiffness
damental frequency and associated maximum modal stress parameters
have been determined for short pipeline free spans, typically
relevant for free span design in flooded areas, river crossings, The Young’s moduli of all pipelines have been assumed as
inlets or straits. 207 GPa. The pipe cross-sections and effective masses are given in
H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230 229

Table A.1 Aronsen, K.H., 2007. An Experimental Investigation of In-line and Combined In-line
Pipeline cross-sections and effective masses. and Cross-flow Vortex Induced Vibrations. Norwegian University of Science and
Technology (NTNU), Trondheim, Norway, Ph.D. Thesis.
Pipeline number Ds (mm) ts (mm) me (kg/m) Bazant, Z.,P., Cedolin, L., 1991. Stability of Structures: Elastic, Inelastic, Fracture and
Damage Theories. Courier Dover Publications, ISBN: 9-780-486-425689.
1 50 3.91 5.841639 Bergan, P.G., Syvertsen, T.G., 1989. Knekning av søyler og rammer, second ed. Tapir,
2 100 6 19.99121 Trondheim, Norway.
3 150 4.8 32.66997 Blevins, R.D., 1990. Flow-induced Vibration, second ed. Krieger Publishing, ISBN:
978-1575241838.
4 150 10.97 50.49274
Choqueuse, D., Chomard, A., Bucherie, C., 2002. Insulation materials for ultra-deep
5 150 7.9 41.82949
sea flow assurance: Evaluation of the material properties. In: Offshore Tech.
6 215.5 3.58 52.80067
Conf., May 2002, OTC 2002, Houston, TX, USA.
7 215.5 4.8 58.23846 DNV Guideline no. 14, 1998. Free Spanning Pipelines. Det Norske Veritas, Norway.
8 215.5 12.7 91.89991 DNV-OS-F101, 2012. Submarine Pipeline Systems. Det Norske Veritas, Norway.
9 268.75 3.96 79.29171 DNV-RP-F105, 2002. Free Spanning Pipelines. Det Norske Veritas, Norway.
10 268.75 6.85 95.33365 DNV-RP-F105, 2006. Free Spanning Pipelines. Det Norske Veritas, Norway.
11 268.75 12.7 126.7058 DNV’81, 1981. Rules for Submarine Pipeline Systems. Det Norske Veritas Classifica-
12 318.75 4.78 112.0943 tion, Norway.
13 318.75 7.39 129.3139 Dirichlet, P.L., 1829. On the convergence of trigonometric series which serve to
14 318.75 12.7 163.442 represent an arbitrary function between two given limits. J. Reine Angew. Math.
15 350 5.33 135.7453 (Crelle’s j.) 4, 157–169.
16 350 7.14 148.8925 Fyrileiv, O., Collberg, L., 2005. Influence of pressure in pipeline design—effective
17 350 9.53 166.0365 axial force. In: Proc. of 24th Int. Conf. on Offshore Mechanics and Arctic
18 400 8.9 200.57 Engineering, Jun. 12–17, OMAE 2005, Halkidiki, Greece.
19 400 16.67 263.1783 Fyrileiv, O., Mørk, K.J., 2002. Structural response of pipeline free spans based on
beam theory. In: Proc. of 21st Int. Conf. on Offshore Mechanics and Arctic
20 450 5.56 212.2207
Engineering, Jun. 23–28, OMAE 2002, Oslo, Norway.
21 450 7.8 233.2687
Fyrileiv, O., Mørk, K.J., Chezhian, M., 2005. Experiences using DNV-RP-F105 in
22 450 10.31 256.5973
assessment of free spanning pipelines. In: Proc. of 24th Int. Conf. on Offshore
23 500 5.56 255.5096
Mechanics and Arctic Engineering, Jun. 12–17, OMAE 2005, Halkidiki, Greece.
24 500 8.15 282.6138 Gao, F.-P., Yang, B., Ying-Xiang, W., Shu-Ming, Y., 2006. Steady current induced
25 500 15.9 361.9924 seabed scour around a vibrating pipeline. App. Ocean Res. 28, 291–298.
26 550 6.6 314.7628 Guan, S.W., Gritis, N., Jackson, A., Singh, P., 2005. Advanced onshore and offshore
27 550 8.2 333.1907 pipeline coating technologies. In: China Int. Oil and Gas Pipeline Tech.
28 550 9.5 348.0823 (Integrity) Conf., Sept. 2005, Shanghai, China.
29 600 6.35 363.8665 Hart, J.D., Sause, R., Wyche Ford, G., Row, D.G., 1992. Mitigation of wind-induced
30 600 9.9 408.4624 vibration of arctic pipeline systems. In: Proc. of 11th Int. Conf. on Offshore
31 600 17.5 502.1119 Mechanics and Arctic Engineering, Jun. 7–12, OMAE 1992, Calgary, Canada.
32 650 7.92 441.2654 Heggen, H.O., Fletcher, R., Fyrileiv, O., Ferris, G., Ho, M., 2014. Fatigue of pipelines
33 650 12.7 506.0066 subjected to vortex-induced vibrations at river crossings, Proc. of the Rio Oil &
34 750 7.14 555.9285 Gas Expo and Conf. 2014, Sep. 15-18. Rio de Janeiro, Brazil.
35 750 11 616.7218 Hetenyi, M., 1946. Beams on Elastic Foundations: Theory with Applications in the
36 750 16.4 700.6934 Fields of Civil and Mechanical Engineering. University of Michigan press, ISBN-
37 850 8.8 726.7526 10: 0472084453.
Hobbs, R.E., 1986. Influence of structural boundary conditions on pipeline free span
38 850 11.68 778.164
dynamics. In: Proc. of Fifth Int. Conf. on Offshore Mechanics and Arctic
39 850 19 907.228
Engineering, OMAE 1986, Tokyo, Japan.
40 900 8.08 791.2603
Hove, F., Søreide, T.H., Nordsve, A.C., Trier, S.D., 2011. Novel design of small diameter
41 900 12 865.4887 pipelines on uneven seabed. In: Proc. of 30th Int. Conf. on Ocean, Offshore and
42 900 22.2 1055.534 Arctic Engineering, Jun. 19–24, OMAE 2011, Rotterdam, The Netherlands.
43 1050 9 1067.522 Kyriakides, S., Corona, E., 2007. Mechanics of Offshore Pipelines buckling and
44 1050 13.4 1164.822 collapse. vol. 1. Elsevier Science Publications, ISBN: 978-00-8055140-1.
45 1050 26 1438.848 Logan, J.D., 1997. Applied Mathematics. John Wiley & Sons, ISBN: 0-471-16513-1.
46 1200 11.61 1427.888 Mukundan, H., 2008. Vortex Induced Vibrations of Marine Risers: Motion and Force
47 1200 17 1563.76 Reconstruction from Field and Experimental Data. Massachusetts Institute of
48 1200 26.2 1792.787 Technology (MIT), Cambridge, MA, USA, Ph.D. Thesis.
Ness, O.B., Verley, R., 1996. Strain concentrations in pipelines with concrete coating.
J. Offshore Mech. Arct. Eng. 118 (3), 225–231.
Palmer, A.C., 2008. Dimensional Analysis and Intelligent Experimentation. World
Scientific, Singapore ISBN-13 978-981-270-818-2.
Shames, I.H., Dym, C.L., 1991. Energy and Finite Element Methods in Structural
Table A.2 Mechanics: SI Units Edition, second ed. Taylor & Francis Group, ISBN:
Soil types applied in the present work. 0891169423.
Sollund, H.A., Vedeld, K., 2014. A Finite Element Solver for Modal Analyses of Multi-
Very soft clay Soft clay Firm clay Loose sand Dense sand Unit span Offshore Pipelines Research Report in Mechanics, No. 14-1. Mechanics
Division, Department of Mathematics, University of Oslo, Norway.
k 653.0 1567 3396 10,944 21,887 kN/m2 Sollund, H.A., Vedeld, K., Hellesland, J., Fyrileiv, O., 2014. Dynamic response of
multi-span offshore pipelines. Mar. Struct. 39, 174–197.
Sollund, H.A., Vedeld, K., Fyrileiv, O., 2015. Modal response of free spanning
pipelines based on dimensional analysis. App. Ocean Res. 50, 13–29.
Table A.1. Soil types and relevant dynamic stiffness properties are Soni, P.K., Larsen, C.M., 2005. Dynamic interaction between spans in a multi span
given in Table A.2. pipeline subjected to vortex induced vibrations. In: Proc. of 24th Int. Conf. on
Offshore mechanics and Arctic Engineering, Jun. 12–17, OMAE 2005, Halkidiki,
Greece.
Sparks, C.P., 1984. The influence of tension, pressure and weight on pipe and riser
deformations and stresses. ASME trans 106, 46–54.
Sumer, B.M., Fredsøe, J., 2006. Adv. Series on Ocean Eng.. revised ed.Hydrodynamics
References
Around Cylindrical Structures, vol. 26. World Scientific, Singapore, ISBN:
9812700390.
API RP 1111, 2009. Design, Construction, Operation and Maintenance of Offshore Timoshenko, S.P., 1937. Vibration Problems in Engineering, second ed. Nostrand,
Hydrocarbon Pipelines (Limit State Design). American Petroleum Institute, USA. New York, NY.
ASME B31.8, 2007. Gas Transmission and Distribution Piping Systems, ASME Code Tofallis, C., 2008. Least squares percentage regression. J. Mod. Appl. Stat. Methods 7
for Pressure Piping. American Society for Mechanical Engineers, USA, B31. (2), 526–534.
Adams, R.A., 1999. Calculus: A Complete Course, fourth ed. Addison Wesley Vedeld, K., Sollund, H.A., 2014. Stresses in heated pressurized multi-layered
Longman Ltd., ISBN: 0-201-39607-6. cylinders in generalized plane strain conditions. Int. J. Pres. Ves. Pip. 120, 27–35.
230 H.A. Sollund et al. / Ocean Engineering 105 (2015) 217–230

Vedeld, K., Osnes, H., Fyrileiv, O., 2012a. Analytical expressions for stress distribu- White, F.M., 1999. Fluid Mechanics, fourth ed. McGraw Hill, ISBN: 0-07-116848-6.
tions in lined pipes: axial stress and contact pressure interaction. Mar. Struct. Xiao, Z.-G., Zhao, X.-L., 2010. Prediction of natural frequency of free spanning
26 (1), 1–26. subsea pipelines. Int. J. Steel Struct. 10 (1), 81–89.
Vedeld, K., Osnes, H., Fyrileiv, O., 2012b. New interpretations of gripping force tests Yang, B., Gao, F.-P., Dong-Sheng, J., Ying-Xiang, W., 2008. Experimental study of
for lined pipes. Mar. Struct. 29 (1), 152–168. vortex-induced vibrations of a pipeline near an erodible seabed. Ocean Eng. 35,
Vedeld, K., Sollund, H.A., Hellesland, J., 2013. Free vibrations of free spanning 301–309.
offshore pipelines. Eng. Struct. 56, 68–82. Zdravkovich, M.M., 1997. Flow Around Circular Cylinders. vol. I. Fundamentals.
Vedeld, K., Sollund, H., Hellesland, J., Fyrileiv, O., 2014. Effective axial forces in Oxford Science Pub., ISBN: 978-0-19-856396-9.
offshore lined and clad pipelines. Eng. Struct. 66 (1), 66–80. Zdravkovich, M.M., 2003. Flow Around Circular Cylinders, vol. II. Applications.
Wang, J., Wang, S.F., Duan, G., Jukes, P., 2009. VIV analysis of pipelines under Oxford Science Pub, ISBN: 0-19-856561-5.
complex span conditions. In: Deepwater Offshore Specialty Symp., Jan. 16–19,
DOSS-71, Harbin, China.

You might also like