Professional Documents
Culture Documents
tot
(2)
where u
c
is the mean current velocity (normal to the pipe), u
w
the significant wave-induced flow velocity, and
0
an approxi-
mation for the lowest natural frequency given by
0
= 1 +CSF _
EI
m
c
I
c
4
_1 +
F
c
P
c
+ C
3
_
o
o
]
2
_
(3)
4 Copyright 2014 by ASME
Figure 6: Overview of span location, height and length
with CSF the stiffening effect of the concrete coating, I
c
the
effective span length [10], m
c
the effective mass, F
c
the
effective axial force, o the static deflection and C
3
the end
boundary coefficient. The moment of inertia for the hollow
circular pipe is given by
I =
n
64
(
o
2
-
2
) (4)
Figure 7: Long free spanning pipeline vulnerable to VIV
and the critical buckling load can be calculated as
P
c
= (1 + SCF) C
2
_
n
I
c
]
2
EI (5)
where C
2
is an end boundary coefficient as well. In addition to
the reduced velocity (2), the stability parameter
K
s
= 4 n
m
c
1
p
w
o
2
(6)
is calculated for each span, where
1
is the total modal damping
ratio, comprising structural damping, hydrodynamic damping
and soil damping. Based on the values of the reduced velocity
(2) and the corresponding stability parameter (6), the in-line
vibration amplitude can be estimated based on the response
model shown in Figure 8, and presented in the Appendix.
Figure 8: Amplitude Response model for in-line VIV motion
For the (ultra)deep water pipeline, presented in this paper,
the contribution of wave induced velocities is neglected, i.e. we
assume u
w
= u m/s. The current velocity is typically specified
as a Weibull probability density function [6], which can be
estimated from the 1, 10 and 100 y return period. Since no
detailed metocean data was available, and given the magnitude
of the water depth (exceeding 2400 meter), we have used a
uniform current velocity distribution of u
c
= u.1 m/s in this
paper.
5 Copyright 2014 by ASME
Under these hydrodynamic conditions, the reduced
velocity for the long span, shown in Figure 7, exceeds the
threshold for the onset of in-line VIV:
I
R
=
u
c
+u
w
tot
= 1.18 > u.9u9 = I
R,onsct
IL
(7)
SAGE Profile calculates the maximum allowable span
length that satisfies I
R
< I
R,onsct
IL
as I
mux
=65.8 meter. The
in-line vibration amplitude can be determined from the
response model (shown in Figure 8) as the value of that
corresponds with the design value of the reduced velocity
I
Rd
= I
R
y
]
(8)
with y
]
the safety factor for the natural frequency, which
depends on the safety class and whether the span is (very) well
defined or not [5].
The dimensionless vibration amplitude A
tot
can then
be translated into a stress range
S
IL
= 2 A
IL
_
A
tot
]
u
IL
y
s
(9)
with y
s
a safety factor,
u
IL
the reduction factor for the current
flow ratio
o =
u
c
u
c
+u
w
(10)
and A
IL
the unit stress amplitude, i.e. the stress due to unit
diameter in-line mode shape deflection. According to DNV-RP-
F105 [6], the unit stress amplitude may be approximated as
A
IL
= C
4
(1 + CSF)
tot
(
o
- t
s
) E
s
I
c]]
2
(11)
with C
4
the mid-span boundary condition coefficient and E
s
the
stiffness of steel. The number of cycles to failure N
IL
at a stress
range S
IL
is defined by an SN-curve of the form
S = _
o
1
S
-m
1
S > S
sw
o
2
S
-m
2
S S
sw
(12)
where {m
1
, m
2
] are fatigue exponents (i.e. the inverse slope of
the bi-linear SN curve), {o
1
, o
2
] are characteristic fatigue
strength constants, and
S
sw
= 1u
_
Iogu
1
-IogN
sw
m
1
]
(13)
is the stress at the intersection of the two SN-curves, with the
number of cycles for which the change in slope appears.
Typically, log N
sw
is either 6 or 7.
The SN-curves may be determined from dedicated
laboratory test data, accepted fracture mechanics theory, or the
values recommended in [7].
For fatigue calculations in SAGE Profile, the pipeline
design engineer can either define his own SN curve, or select
the SN-curves {F , F
1
, F
2
, F
3
] from DNV-RP-C203. The latter
curves have a different shape for free corrosion (only one slope)
or when cathodic protection is present (two-slope curve). For
instance, the SN curves in seawater when cathodic protection is
present are shown on Figure 14. The change in slope occurs
atN
sw
= 1u
6
. For the fatigue analysis, presented here, we have
used the F
3
SN-curve from [7], assuming cathodic protection is
present.
Figure 9: SN curves for cathodically protected pipelines
The marginal fatigue life capacity against in-line VIV in a
single sea-state is calculated by integrating over the long-term
distribution of the current velocity. As we assume a uniform
current velocity distribution, the fatigue life calculation
simplifies to
I
IL
|y] =
N
IL
0
S6S 24 S6uu
(14)
For the critical free span shown in Figure 12, this leads to a
remaining fatigue life of 116 years, which is well above the
design life of the flowline.
In addition to the in-line VIV assessment, SAGE Profile
also constructs the response model for cross-flow VIV, based
on the current flow ratio (17), and the Keulegan Carpenter
number
KC =
u
w
w
tot
(15)
with u
w
the significant wave-induced velocity, and
w
the
corresponding frequency. Given the significant water depths,
exceeding 2400 meters, fatigue analysis for both cross-flow
VIV and cross-flow induced VIV can be omitted.
6 Copyright 2014 by ASME
SENSITIVITY TO SOIL PROPERTIES
For the deterministic benchmark, presented in the previous
section and elaborated in [4], sensitivity analyses were
performed to investigate the influence of the element length,
the weight of the suspended catenary and the estimation of the
on-bottom residual lay tension on the ability to predict free
spanning pipes. These analyses indicate that
Reducing the element length enhances the accuracy of
the simulated spans. The lower bound for the element
length is governed by the seabed resolution.
The suspended catenary has to be taken into account to
capture the actual pipeline embedment at the touch-
down point.
The applied residual lay tension has a pronounced
influence on the span predictions. The simulated span
lengths increase with increasing lay tension. This is in
line with the observations reported in [8] on influence
of the effective axial force on free spanning pipes.
The predicted span length and height were shown to be
even more sensitive to the constitutive soil model. An elastic
soil model tends to over-estimate the span length. Hence, an
elasto-plastic soil model is recommended. Moreover, the shear
strength proves to have a significant influence on both the
number of predicted spans and the corresponding span lengths.
Indeed, the soil reaction is dictated by a vertical soil spring,
which reflects the bearing capacity
u
and for clays, DNV-RP-
F105 [5] recommends
u
(z
p
) = (S.14 C
u
+y
s
z
p
) B(z
p
) (16)
where z
p
is the pipe penetration, and
B(z
p
) = _
2 _z
p
(
o
- z
p
) u z
p
o
2
o
otheiwise
(17)
the bearing width. The bearing capacity (16) is often calculated
assuming a constant shear strength, which is the average value
of the shear stress profile measured in the top ~400 mm of the
undisturbed seabed.
However, predictions of the initial pipeline embedment
based on a merely static load generally under-estimate the
actual pipe penetration, because they do not account for the
stress concentration at the touchdown point, nor the dynamic
movement induced by the vertical catenary oscillations. Cheuk
and White [9] have suggested to modify the original shear
strength to reflect the level of soil disturbance and remoulding
expected during the lay process. Introducing a soil sensitivity
St, the remoulded shear strength is written as a fraction of the
original undisturbed shear strength:
S
u
cm
=
S
u
St
(18)
The soil sensitivity can vary between St = 1 (insensitive
clays) and St = 16 (quick clays) and hence has a significant
influence on calculated bearing capacity. The influenced of soil
sensitivity on the vertical soil spring (16) is shown in Figure 10.
Figure 10: Influence of soil sensitivity on bearing capacity
On-bottom roughness assessment and free span analysis is
often performed in feasibility studies and during the route
selection process, where detailed geotechnical data and
accurate pipe-soil interaction parameters are not always
available. To study the effect of incomplete data on the soil
properties, we have re-run the pipeline laydown simulations for
different values of the remoulded shear strength.
Table 2: Sensitivity analysis on soil parameters
St
[-]
Sensitivity
S
u
[kPa]
S
u
cm
[kPa]
1 Insensitive 2.2 2.2
2 Medium 2.2 1.1
4 Sensitive 2.2 0.55
The parameters used in the sensitivity analysis are
summarized in Table 2, and the results are shown in Figure 11.
Both the number of identified spans and the total span length
over the proposed pipeline route increase significantly with the
undrained shear strength of the soil. Indeed, considering a
sensitivity of 1 rather than 4 raises the number of spans from 27
to 83, and implies a 20% increase in the accumulated span
length.
From this fairly simple and straightforward sensitivity
analysis, it is obvious that the soil properties are of paramount
importance for the accurate prediction of free spans and the
corresponding fatigue lifetime of the pipeline when subjected to
vortex induced vibrations.
7 Copyright 2014 by ASME
Figure 11: Influence of soil properties on span predictions
Yet, the soil properties are not always known in the early
design stages, or at least subject to uncertainty. In the next
section, we introduce a probabilistic model to estimate in-line
VIV fatigue, where the soil properties are represented by
stochastic variables. This is the perimeter of engineering where
structural reliability analysis provides the tools that allow
calculating the cost of ignorance.
RELIABILITY ANALYSIS APPLIED TO FREE SPANS
Reliability methods deal with the uncertain nature of loads
and resistance, and lead to the assessment of the reliability [10-
11]. Reliability methods are based on analysis models for the
structure in conjunction with available information about loads
and resistances, and their associated uncertainties. The analysis
models are usually imperfect, and the information about loads
and resistances is usually incomplete [12]. The reliability as
assessed by reliability methods is therefore generally not a
purely physical properties of the structure in its environment of
actions, but rather a nominal measure of the safety of the
structure, given a certain analysis model and a certain amount
and quality of information.
According to [12], the structural reliability is defined as the
probability that the system will not attain each specified limit
state. The limit state or performance function g(X) may be
generally defined by the stochastic loads I(X) and resistances
R(X) as the condition where the load equals the systems
resistance:
g(X) = R(X) -I(X) u (19)
The loads and resistance of a system possess inherent
uncertainties in their magnitude for different periods of time in
the design life. For a given time period, the limit state can be
represented graphically as shown in Figure 12-a. The structural
reliability of the system is therefore the area under the
resistance curve that is greater than the load (Figure 12-b).
Mathematically, it is more convenient to calculate the
reliability of a system in terms of its complement: the
probability that failure will occur, like shown on Figure 12-c.
The probability of failure is calculated as
P
]
= P(g u) = _
R
(x)
L
(x) Jx
g<0
(20)
with
R
(x) and
L
(x) the probability density functions for the
resistance and load respectively.
(a) Limit State Definition
(b) Stuctural Reliability
(c) Probability of Failure
Figure 12: Structural reliability and probability of failure
8 Copyright 2014 by ASME
The reliability index is defined as [10-12]
[ = -
-1
(P
]
) (21)
where is the standard normal (cumulative) distribution
function. The probability of failure P
]
and the corresponding
reliability index can be solved by Monte Carlo simulations [13]
or using first or second-order reliability methods [14].
In this paper, application of the First Order Reliability
Method (FORM) is demonstrated to predict the probability of
(fatigue) failure for a free span subjected to in-line VIV, when
the soil properties are treated as stochastic variables.
Stochastic Basic Variables X
|
Standardised Normal Variables U
|
Figure 13: Transformation of the limit state surface
In a FORM analysis, the limit state surface g(X) in the
space spanned by the basic variables X is transformed to a
corresponding limit state function g(u) by normalization of
the random variables X
=
X
-p
X
i
o
X
i
(22)
with p
X
i
and o
X
i
the mean value and standard deviation of the
original stochastic variable X
]ut
(I) =
I
o
1
_ _ s
m
1
s
(s) Js
s
sw
+
o
1
o
2
_ s
m
2
s
(s) Js
s
sw
0
_
where
s
(s) is the stochastic long-term stress range probability density
function, capturing the uncertainty in the amplitude response
model.
The sensitivity analysis on soil properties, summarized in
Figure 11, has shown that the span prediction is to a very large
extent dependent on the accuracy of the geotechnical input
data. To study the importance of uncertainty in soil properties,
we have chosen all variables and parameters in (23) to be
deterministic, except the ones that relate to geotechnical data.
For the damage induced by in-line VIV [6], this reduces to the
coefficient for the lateral soil stiffness C
L
and the uncertainty on
the stability parameter (6), which captures the effect of soil
damping.
For the lateral soil stiffness, we assume that C
L
is normally
distributed with a mean value p
CL
= 1200 kN/m
5/2
and a
coefficient of variability
co:
CL
=
o
CL
p
CL
= u.S
(24)
These values are representative for a soft clay and in line
with the recommendations given in [8] and [16]. As C
L
is
assumed to be normally distributed, it can be converted into a
standard normal variable u
CL
through the transformation (22).
Figure 14 shows the variation of the limit state function (23) as
a function of the expected value for C
L
when all other variables
are fixed. The fatigue damage quickly becomes important for
increasing values of the lateral soil stiffness.
Figure 14: Influence of soil stiffness on limit state function
In order to account for the uncertainty on the model used
for calculating the stability parameter K
s
, an additional
stochastic variable X
KS
is multiplied with (6).
This stochastic variable is assumed to follow a lognormal
distribution with p
Xks
=1 and co:
Xks
=0.12. The lognormal
distribution implies that = ln(X
KS
) follows a normal
distribution:
X
KS
= exp(p
+ o
u) (25)
where u is the standard normal variable and p
and o
are the
mean and the standard deviation of . The mean of a lognormal
variable can be computed as
p
Xks
= exp_p
+
o
2
2
_ (26)
and the variance can be expressed as
o
Xks
= |exp(o
2
) -1] exp(2 p
+o
2
)
(27)
and hence, by very definition of (26),
o
Xks
= |exp(o
2
) -1] p
Xks
2
(28)
Inverting (26) and (28) yields an expression for the variance
o
2
= ln_
o
Xks
2
p
Xks
2
+ 1_ (29)
and the mean value
p
= ln(p
Xks
) -
o
2
2
(28)
of the normally distributed variable . These expressions
finally allow transforming the uncertainty on the stability
parameter X
KS
into the standard normal variable
u
Xks
=
ln(X
KS
) -p
(30)
Figure 15: Influence of stability X
KS
on limit state function
10 Copyright 2014 by ASME
The influence of X
KS
on the performance function (23) is
shown in Figure 15. Since we are only investigating uncertainty
in soil properties, the vector u in the transformed limit state
function 0(u) reduces to u = (u
CL
, u
Xks
). Assuming that
these stochastic variables are not correlated, the joint
probability distribution can be visualized as the product of the
both normal standard probability density functions, like shown
in Figure 16.
Figure 16: Joint probability density function in U-space
The FORM analysis now reduces to the minimization problem
[ = min u
1
u
(31)
provided 0(X) = u. This can be visualized in the space
spanned by the standard normal variables u = (u
CL
, u
Xks
), as
demonstrated in Figure 17. The design point is found as the root
of the minimization problem (31) as
u
= (u
CL
, u
Xks
tot
+ I
R,onsct
IL
(36)
I
R,2
IL
= I
R,cnd
IL
- 2
A
,2
tot
(37)
where A
,1
and A
,2
are the corresponding vibration amplitudes
A
,1
tot
= max _u.18 _1 -
K
sd
1.2
] R
1
I0
;
A
,2
tot
_ (38)
and
A
,2
tot
= u.1S _1 -
K
sd
1.8
] R
2
I0
(39)
These amplitude values depend on the reduction factors
u R
1
I0
(I
c
, 0
cI
) 1 and u R
2
I0
(I
c
, 0
cI
) 1 who account
for the effect of the turbulence intensity I
c
and the angle of
attack (0
cI
, in radiance) for the flow [6].
Also note that DNV-RP-F105 introduces an additional
reduction function to account for reduced in-line VIV in wave
dominated conditions:
u
IL
= _
u.u foi o < u.S
o - u.S
u.S
foi u.S < o < u.8
1.u foi o > u.8
(40)
Thus, if o < u.S, in-line VIV may be ignored.
REFERENCES
[1] Bai Y. and Bai Q., Pipelines and Risers, Chapter 34:
Installation Design, pp. 597-636 (2005)
[2] Witgens J .F., Falepin H. and Stephan L., New Generation
Pipeline Analysis Software, Offshore Pipeline Technology,
OPT 2006
[3] Van den Abeele F. and Denis R., Numerical Modelling and
Analysis for Offshore Pipeline Design, Installation and
Operation, J ournal of Pipeline Engineering, Q4, pp. 273-
286 (2011)
[4] Van den Abeele F., Bol F. and Hill M., Fatigue Analysis
of Free Spanning Pipelines subjected to Vortex Induced
Vibrations, Proceedings of the 32
nd
International
Conference on Ocean, Offshore and Arctic Engineering,
OMAE2013-10625 (2013)
[5] Det Norske Veritas, Recommended Practice on Free
Spanning Pipelines, DNV-RP-F105 (2006)
[6] Fyrileiv O., Aronsen K. and Mork K., DNV FatFree
Verification Document, Report No. 2000-3283 (2010)
[7] Det Norske Veritas, Recommended Practice on Fatigue
Design of Offshore Steel Structures, DNV-RP-C203
(2010)
[8] Hagen O., Mork K.J ., Nielsen F.G. and Soreide T.,
Evaluation of Free Spanning Design in a Risk Based
Perspective, Proceedings of the 22
nd
Conference on
Offshore Mechanics and Arctic Engineering, OMAE-
37419 (2003)
[9] Cheuk C.Y. and White D.J ., Centrifuge Modelling of Pipe
Penetration due to Dynamic Lay Effects, Proceedings of
the 27
th
International Conference on Offshore Mechanics
and Arctic Engineering, OMAE2008-57923 (2008)
[10] Thoft-Christensen P. and Baker M.J ., Structural Reliability
and its Applications, Spinger Verlag, Berlin, Heidelberg
(1982)
[11] Ditlevsen O. and Madsen H.O., Structural Reliability
Methods, Wiley, Chichester (1996)
[12] Det Norske Veritas, Classification Notes CN30.6,
Structural Reliability Analysis of Marine Structures (1992)
[13] Rubinstein R.Y., Simulation and the Monte-Carlo Method,
Wiley, New York (1981)
[14] Schueller G.I. and Stix R., A Critical Appraisal of
Methods to Determine Failure Probabilities, Structural
Safety vol. 4, pp. 293-309 (1987)
[15] Hasofer A.M. and Lind N.C., An Exact and Invariant First
Order Reliability Format, J ournal of Engineering
Mechanics, pp. 111-121 (1974)
[16] Sigurdsson G., Mork K.J . and Fyrileiv O., Calibration of
Safety Factors for DNV-RP-F105 Free Spanning
Pipelines, OMAE2006-92099 (2006)