International Journal of Offshore and Polar Engineering (ISSN 1053-5381) http://www.isope.org/publications/publications.

htm
Copyright © by The International Society of Offshore and Polar Engineers
Vol. 25, No. 4, December 2015, pp. 241–246; http://dx.doi.org/10.17736/ijope.2015.jc633

Probabilistic Lateral Buckling Assessment

Carlos Sicilia and Philip A. Cooper
INTECSEA
Woking, Surrey, UK
Emilien Bonnet
INTECSEA
Perth, Western Australia, Australia

This paper presents a probabilistic method to assess the lateral buckling response of a pipeline. The method is based on a
Monte Carlo (MC) simulation in which the lateral buckling response is predicted through the use of a surrogate model that
employs artificial neural networks (ANNs) calibrated from nonlinear finite element (FE) analyses. The method presented
intends to improve on current industry best practice by directly considering the limit states relevant to global buckling to
produce designs with consistent levels of reliability.

INTRODUCTION The purpose of this paper is to present a probabilistic assessment

In the design of offshore pipelines, it is often convenient and
sometimes necessary to allow the formation of lateral buckles. This
relieves the axial compression but can potentially lead to severely
localized bending deformations. In such designs, the lateral buckles
are often subjected to the most onerous loading conditions and
govern the mechanical design of the pipeline.
The response of a pipeline at a lateral buckle can be modeled
in great detail through the use of sophisticated nonlinear finite
element analysis (FEA). However, such models require information
that is not available at the design stage. This information includes
the pipe-soil interaction response, which can be quantified at the
design stage but is subjected to significant uncertainty, and the
as-laid geometry of the pipe in the vertical and horizontal planes,
which cannot be determined until the pipeline has been laid. The
uncertainty in these two parameters leads to additional uncertainty
regarding where the buckles form and, more importantly, how far
apart the buckles are from one another, as this governs the level of
load (axial feed-in or expansion) that goes into each buckle location.
The uncertainty in design is normally accounted for by the use
of safety (load and resistance) factors calibrated to achieve a certain
probability of failure. In the case of the lateral buckling response of
a pipeline, this would seem impractical given the very high level of
uncertainty and the interdependence between the global response of
the pipeline, i.e., where the buckles form, and the local response at
each lateral buckle. Under these circumstances, a pipeline-specific
probabilistic assessment appears to be a suitable approach.
This type of approach was proposed by the Safebuck Joint
Industry Project (JIP) (Bruton and Carr, 2011), which developed
the Buckfast software available to the participants of the JIP. This
approach, however, is based on the concept of the characteristic
virtual anchor spacing (VAS), which requires a calibration implicit
in the Safebuck methodology/guidelines to obtain the required
probabilities of failure.
Received December 22, 2014; updated and further revised manuscript
received by the editors March 2, 2015. The original version (prior to the
final updated and revised manuscript) was presented at the Twenty-fourth
International Ocean and Polar Engineering Conference (ISOPE-2014),
Busan, Korea, June 15-20, 2014.
KEY WORDS: Pipeline, lateral buckling, probabilistic, neural network,
Safebuck, reliability, Monte Carlo (MC).
of the lateral buckling response of a pipeline that is based on
the actual limit states and therefore requires no calibration. The
proposed strategy is intended to achieve probabilities of failure in
line with those required for the rest of the design of the system
and therefore to produce a design that is better optimized.
CURRENT BEST PRACTICE AND ITS LIMITATIONS
Lateral buckling designs are currently performed following either
the DNV Recommended Practice for Global Buckling (DNV, 2007)
or the Safebuck guidelines. The former is available for public use,
whereas the latter is available for use only by the participants of
the JIP. Both documents consider limit state design (LSD) and
aim at producing pipelines with industry standard probabilities of
failure. The two documents are in the process of being merged into
a single document for general use (Collberg et al., 2011).
The Safebuck guidelines are generally considered more advanced.
At the methodological level, the main advantage is that they
contain a clear framework for performing probabilistic assessments.
This allows assessing the susceptibility of a pipeline to laterally
buckle, considering the integrity of the pipeline in the post-buckled
condition, and developing appropriate mitigation measures.
The Safebuck approach revolves around the VAS concept. The
VAS is the distance between the virtual anchor points on either side
of a buckle that do not move axially when the pipe is subjected
to operating conditions. This is a very convenient concept as the
response at each lateral buckle can be fully defined by considering
only the section of the pipeline covered by the VAS. Each buckle
can then be analyzed in isolation (VAS analysis) instead of modeling
the entire pipeline (full model analysis). This allows running a
large number of detailed simulations to assess the integrity of the
pipe at the buckle location.
In order to assess the lateral buckling response through the
use of the VAS analysis, however, the VAS of the buckle needs
to be defined, i.e., it is necessary to determine the locations of
all the lateral buckles along the pipeline. This is straightforward
when buckle triggers are introduced, and it is assumed that buckles
form at all the trigger locations (planned buckles). However, this
becomes quite complex when the reliability of the formation of the
planned buckles and/or the formation of the unplanned buckles is
taken into account.
242
The Safebuck guidelines propose a methodology to determine a
characteristic VAS from a probabilistic assessment of the overall
pipeline response. The characteristic VAS is equivalent to a charac-
teristic design load in the standard limit state design, and thus it is
defined as the VAS with a certain probability of exceedance.
The characteristic VAS along the length of a pipeline can be
determined from a Monte Carlo (MC) assessment that considers
the effective axial force (ESF) profile buildup and the buckling
force at each point along the pipeline. In this assessment, the key
probabilistic variables are the axial pipe-soil friction factor and
the buckling force. The latter is a function of the lateral pipe-soil
friction factor, the shape of the as-laid pipeline, i.e., the level of
out-of-straightness (OOS), and any intervention introduced at the
planned buckle locations.
Once the characteristic VAS has been established, the response at
a lateral buckle with a VAS length equal to the characteristic VAS
is analyzed by finite element analysis (FEA) with a VAS model,
and all the relevant limit states, i.e., local buckling, fatigue, fracture,
and plasticity limits, are checked. Alternatively, a tolerable VAS is
defined as the VAS for which the critical limit state is just satisfied.
This process is followed for the different types of lateral buckles,
i.e., buckles at the locations of buckle initiators and buckles at
unplanned locations at different points along the pipeline. The
lateral buckling design then consists of adjusting the pipeline layout
and the configuration of the buckle triggers and their locations
until the tolerable VAS is bigger than the characteristic VAS at any
point along the pipeline.
The use of the VAS concept as a proxy of the actual limit
states simplifies the problem significantly, but it also introduces an
inconsistency in the assessment. On the one hand, the characteristic
VAS is obtained from a probabilistic assessment that considers the
full range of lateral pipe-soil friction factors through the probability
function of the critical buckling force. As a result of this, the
characteristic VAS is not related to a specific lateral friction factor.
However, since it is easier for lateral buckles to form with low
lateral friction, it is to be expected that the characteristic VAS is
associated with values of lateral friction that are not close to the
upper bound. On the other hand, the most onerous conditions at
lateral buckles develop in cases with high lateral friction. As a
result of this, the tolerable VAS is determined by the highest lateral
friction at which a buckle can be triggered.
The acceptability of the solution is therefore based on comparing
VAS lengths that correspond to very different levels of lateral fric-
tion. It would be possible to calibrate the probability of occurrence
used to define the characteristic VAS in order to compensate for
this inconsistency. However, considering that the variability in
the lateral pipe-soil friction depends on the soil type and pipe
properties, a generic calibration will be either unsafe in some cases
or too onerous in other cases.
The Safebuck approach is generally implemented through the
use of the Buckfast software that was developed as part of the
JIP. The program provides the probability of buckle formation
and the characteristic VAS along the pipeline. The program also
provides the probability function of the expansions at the pipe ends.
However, the program does not provide axial displacements at
intermediate points, which are required if there are in-line tees
(ILTs) along the pipeline. Considering that axial displacements
along a pipeline can be severely affected by the location of buckles,
determining the probability function for the axial displacements
can lead to significant reductions of the displacements used for the
design of the spool pieces and ILTs.
Probabilistic Lateral Buckling Assessment
PROPOSED APPROACH
Overall Description
The main improvement in the proposed approach is to produce a
consistent limit state design in which the actual limit states for
global buckling are considered instead of a proxy concept such as
the VAS of the lateral buckle. This can be achieved by embedding
the approach in a Load and Resistance Factor Design (LRFD)
framework, such as DNV-OS-F101 (DNV, 2013), or by undertaking
a full structural reliability assessment. In both cases, the uncertainty
in the lateral buckling response is considered, and the validity of
the design is checked against actual limit states. For simplicity, this
paper will consider the LRFD option.
In the proposed approach, a probabilistic assessment of the
overall lateral buckling response of the pipeline is performed, which
takes into account the uncertainty in the lateral buckling response,
in order to derive design load effects (strains, moments, stress
ranges, etc.) at the lateral buckles. The design load effects are
then compared to the design resistances for each of the relevant
limit states for global buckling (local buckling, fatigue, fracture,
plasticity limits, and displacements).
It would be possible to embed this approach within any limit
state design code for pipelines, but this paper will consider its
application only within the framework of DNV-OS-F101 (DNV,
2013). Within that framework, the derived design load effects
are based on the definitions in Section 4 of the code, the design
resistances are calculated according to Section 5 of the code,
and thus the overall design should be compliant with the overall
probabilities of failure specified in Section 2 of the code.
Probabilistic Lateral Buckling Assessment Model
The probabilistic assessment is performed through the use of
Monte Carlo simulations of the lateral buckling response of the
whole pipeline. This allows obtaining probability distributions for
the limit state checks. The lateral buckling response of a pipeline
is typically assessed through the use of nonlinear finite element
analysis. However, this is not feasible within the Monte Carlo
framework as the analysis would take too long to run. Instead, the
Monte Carlo assessment is performed through the use of a surrogate
model for the lateral buckling response of the whole pipeline.
Global Lateral Buckling Surrogate Model
The Global Lateral Buckling Surrogate Model (GLBSM) calcu-
lates the lateral buckling response of the whole pipeline for one of
the configurations in which the pipeline could be laid once. Each
possible scenario of the as-installed pipeline can be fully defined
by the discretization of the pipeline into elements and the assign-
ment of properties to each element. This includes geometric and
material pipe properties, pipe-soil interaction data, and a parameter
describing the geometry of the as-laid pipeline at that location,
which is a measure of how much the as-laid pipeline deviates from
the nominal pipeline route, i.e., a measure of the OOS.
The defined pipeline scenario can then be solved through the use
of an analytical, one-dimensional pipeline expansion model in which
the profile of the effective axial force and axial displacement along
the pipeline is calculated for the operating conditions considered.
From the input data provided, it is possible to calculate the effective
force at which a lateral buckle would form at each element along
the pipeline. On the basis of this, the locations at which buckles
would form in the scenario considered can be determined by the
application of the operating conditions incrementally and by a
comparison of the predicted force profile with the buckling force at
each point along the pipeline. At each temperature increment and
International Journal of Offshore and Polar Engineering, Vol. 25, No. 4, December 2015, pp. 241–246 243

each pressure increment, the locations of the buckles previously

predicted can be taken into account in the analytical calculations.
The model described up to this point mostly follows the approach
considered by Safebuck and implemented by Buckfast. The main
improvement in the proposed approach is the derivation of the
mechanical conditions (the moments, strains, stress ranges, etc.)
at each lateral buckle in the considered scenario of the as-laid
pipeline. Due to this improvement, we are allowed to assess the
integrity of the pipeline at the lateral buckles by considering the
relevant limit states directly instead of following the Safebuck
approach based on the VAS concept.
The mechanical conditions at a lateral buckle are often predicted
from FE models, but this is not possible in this case as the GLBSM
has to compute quickly for use within the Monte Carlo framework.
Instead, a local surrogate model of the lateral buckling response of
a single buckle is used, which is presented in the following section.

Local Lateral Buckling Surrogate Model

The key improvements proposed for the probabilistic lateral
buckling assessment hinge on the use of a Local Lateral Buckling
Surrogate Model (LLBSM) to derive additional information from
the GLBSM that is run within the Monte Carlo framework. In
order to develop the LLBSM, an extensive parametric study of the
isolated response of a lateral buckle is performed using VAS models.
In the parametric study, the key input parameters affecting the
isolated response of a lateral buckle are varied and typically include
the lateral pipe-soil friction, initial OOS, material parameters, and
specific features introduced at planned buckles. From each run, the
required results, such as the buckling force, peak moments or strains,
stress ranges for typical operating cycles, and lateral displacement,
are extracted as a function of the axial feed-in into the buckle.
The next step requires fitting all the generated results through the
use of analytical formulations that can be computed quickly so that
they can be embedded into the GLBSM. Developing user-defined
analytical formulations can be a cumbersome and time-consuming
process that may need to be repeated for every new pipeline.
The method proposed in this paper considers artificial neural
networks (ANNs). This method is a very efficient way of fitting Fig. 1 Proposed methodology
multidimensional, highly nonlinear problems with limited user
intervention. Details on how ANNs operate can be found in classic
textbooks such as Fundamentals of Neural Networks: Architectures, buckling, fatigue, fracture, plasticity limits, and displacements.
Algorithms, and Applications (Fausett, 1994). From the probability functions, the design load effects can be
The ANN is trained and validated through the use of the results computed. The accidental load effects can also be calculated by
of the VAS parametric analysis to provide an analytical way of considering events with lower probabilities of occurrence.
predicting the response at a lateral buckle (i.e., the buckle force and The proposed probabilistic lateral buckling assessment method
mechanical conditions) as a function of the axial feed-in into the can be used to prove that uncontrolled lateral buckling is acceptable
buckle and other relevant input variables. The ANNs considered or to develop and refine lateral buckling mitigation measures until
in this paper are trained through the use of the back propagation adequate probabilities of failures are achieved. A summary of the
algorithm with architecture evolution. This process requires some overall process is illustrated in Fig. 1.
expertise in the use of artificial neural networks, but generally leads
to successful convergence with limited user intervention. CASE STUDY
In this section, a case study is presented to compare the outcome
Probabilistic Assessment Results
of the proposed approach with the outcome of the Safebuck
Once trained and validated, the ANN becomes the LLBSM and methodology. As noted in the previous sections, the proposed
is embedded in the GLBSM that is run within the MC framework. approach has significant flexibility of the parameters that are
As part of this process, some of the input variables are defined set as deterministic and those that are taken as probabilistic. In
as probabilistic and sampled randomly to produce thousands of the example presented here, the variables that are considered
possible configurations representing as-laid conditions for the probabilistic have been limited to simplify the comparison with
pipeline once the pipeline has been constructed and installed. Safebuck and the interpretation of the results. Since this paper is
Through the use of the surrogate models to compute the response intended to illustrate the advantages of the proposed approach, a
at each lateral buckle in each of the possible configurations randomly set of parameters has been assumed. Therefore, although the data
generated, it is possible to develop the probability functions for used in a real assessment are of paramount importance, they are
each of the relevant limit states for global buckling, i.e., local not part of the purpose of this paper. A summary of the parameters
244 Probabilistic Lateral Buckling Assessment

of the case study and their treatment in the probabilistic assessment

Pipe-soil interaction
is presented in Tables 1 and 2.
One of the key variables in the probabilistic assessment of lateral Axial friction Lognormal, 0.54
buckling is the distribution of OOS features along the pipeline. mean and 20% CoV
As presented by Collberg et al. (2011), the JIP that led to DNV- (coefficient of variation)
RP-F110 (Hotpipe) (DNV, 2007) could not characterize the OOS Single sampling
features from a review of the real as-laid data of existing pipelines, for whole pipeline
but Safebuck could. The Safebuck data, however, are for use only Breakout lateral Lognormal, 0.61 mean
by the JIP sponsors and cannot be presented in this paper. friction 4ŒBO
L 5 and 56% CoV curtailed at 1.70
The generic pipeline considered in the case study includes a Sampled at each element
number of buckle initiators and route curves with tight radii, where
Residual lateral Correlated at each element as
there is a significant risk of the formation of unplanned buckles. 2
friction −0025 ∗ 4ŒBO BO
L 5 + 0093 ∗ 4ŒL 5
For unplanned buckles on nominally straight sections, the critical
buckling force can be assumed to be equal to the value for the Buckle initiators buckling force [kN]
Hobbs infinite mode (Hobbs, 1984) multiplied by a coefficient Buckling force Lognormal, 360 mean
proportional to the OOS feature. This can be rearranged as a and 7% CoV
probabilistic variable multiplied by the square root of the breakout
lateral friction factor. The probabilistic variable used in this example Unplanned buckles buckling force [kN]
is presented in Table 2. Nominally Lognormal, 7000 mean
For unplanned buckles on route curves, a similar approach can be straight section and 70% CoV Sampled
taken using the product of the route curve radius and the breakout p
(to multiply ŒBO L ) at each 50 m-long
lateral resistance of the soil rather than the Hobbs force. This element
can be rearranged as a probabilistic variable multiplied by the
breakout lateral friction factor. The probabilistic variable used in Route curve with Lognormal, 3500 mean
this example is presented in Table 2. 2.5 km radius and 65% CoV Sampled
The buckle initiators considered in this case study consist of (to multiply ŒBO
L ) at each 50 m-long element
sleepers combined with snake lay. An ANN has been trained to
Table 2 Case study probabilistic input data

Field layout predict the buckling force at the buckle initiators as a function of
Pipeline length 10 km the soil friction. In the Monte Carlo simulations (in which 100,000
ILT KP 4.825 scenarios are considered), the input parameters are sampled at each
Route curves KP 1.0 to 3.0; 4.0 to 5.0; planned buckle location and used to calculate the buckling force.
8.0 to 9.0 Radius: 2.5 km The resulting buckling force distribution is presented in Table 2.
Bathymetry A flat seabed is assumed In this case study, it is assumed that fracture, i.e., the girth
weld strain capacity, is the critical limit state for both planned
Pipe properties
and unplanned buckles. Accordingly, the tolerable VAS used in
Outer diameter 0.457 m the Safebuck assessment is based only on this limit state, and the
Wall thickness 21.9 mm assessment based on the approach proposed in this paper considers
Operating submerged weight 860 N/m only the probability function of the peak mechanical tensile strain.
Modulus of elasticity 204 GPa An ANN is trained from a series of FE VAS analyses to predict
Poisson’s ratio 0.3 the peak tensile strain at planned and unplanned buckles. On the
Thermal expansion 1020 × 10−5 ž C−1 basis of the results of the VAS analyses, the peak strain is defined
Stress-strain curve at Ramberg-Osgood through as a function of the axial feed-in and lateral pipe-soil friction. The
operating temperature 422 MPa, 0.5% and ANN is used as part of the limit state check assessment. As shown
438 MPa, 1.0% in Figs. 2 and 3, the trained ANN reproduces the results from the
Operating conditions (assumed constant) FE analyses accurately.
Internal pressure 320 barg
Temperature differential 110 ž C
Lateral buckling mitigation—planned buckles
Sleeper 0.4 m high with
0.2 friction
Location KP 1.2, 2.5, 3.7, 5.0,
6.3, 7.5, 8.8
Design resistance (fracture)
Planned buckles 0.40%
Unplanned buckles 0.30%
Tolerable VAS (fracture)
Planned buckles 1500 m
Unplanned buckles 1000 m

Table 1 Case study deterministic input data Fig. 2 ANN peak tensile strain fit for planned buckles
International Journal of Offshore and Polar Engineering, Vol. 25, No. 4, December 2015, pp. 241–246 245

Fig. 3 ANN peak tensile strain fit for unplanned buckles Fig. 4 Unity ratio check: VAS vs. limit state assessment

Results comparison of this value with the probability of exceedance used

to define the design load in DNV-OS-F101 (DNV, 2013), which is
The results of the assessments performed are presented in Fig. 4
1%, it is concluded that the design considered in the case study
as a unity ratio against the pipe KP for unplanned buckles and
complies with the requirements of the code. These results show that
as discrete results for the locations of the buckle initiators. The
for this case study, the proposed methodology improvements allow
results of the Safebuck assessment are identified as VAS check,
demonstrating the validity of a design that would be considered
and the results of the assessment performed using the proposed
unacceptable under the current industry best practice approach.
methodology are identified as limit state check. In the Safebuck
The final aspect considered for the case study is the axial
approach, the unity ratio is calculated as the characteristic VAS
displacement at the ILT that is required to design the connection
over the tolerable VAS. In the limit state check approach, the unity
piece at the ILT (e.g., the spool) and the ILT itself. The axial
ratio is calculated as the characteristic value of the limit state
displacement at the ILT depends on the distance between the ILT
considered, i.e., the peak mechanical tensile strain multiplied by
and the nearest virtual anchor point, and therefore the displacement
the DNV-OS-F101 (DNV, 2013) load factors and divided by the
depends on the location of the buckles in the vicinity of the ILT.
design resistance appropriate for each buckle type (planned or
Since there is uncertainty in the location of the buckles, there is
unplanned) presented in Table 1. In both cases, the characteristic
uncertainty in the axial displacement at the ILT.
values are defined through the use of the Safebuck requirements.
Without a probabilistic assessment, it would be necessary to
For planned buckle locations, the characteristic value is defined as
consider the buckle locations that maximized the distance between
the value with a probability of exceedance of 10%. For unplanned
buckle locations, it is defined as the value with a probability of the ILT and its nearest virtual anchor point, together with the
exceedance of 1% per km of pipeline. lower bound axial friction. In the case study, this leads to an
The results of the Safebuck approach assessment show accept- axial displacement at the ILT of 2.8 m. From the probabilistic
able conditions at all planned buckle locations, but unacceptable assessment performed, it is possible to show that such a value has a
conditions at unplanned buckles forming at all the route curves negligible probability of occurrence, and therefore it would be overly
(KP 1.0 to 3.0, 4.0 to 5.0, and 8.0 to 9.0). conservative to consider it for the design of the ILT arrangement.
As noted previously, there is an inconsistency in working with the As depicted in Fig. 5, the probabilistic axial displacement at
VAS concept to assess the integrity of the pipe at buckle locations the ILT shows that the axial displacement with a probability of
as the characteristic VAS and tolerable VAS are based on different exceedance of 10−2 is 1.25 m for the design condition and the
levels of lateral friction. This is addressed in the proposed approach axial displacement with a probability of exceedance of 10−4 is
in which actual limit states are considered (in this example, tensile 1.65 m for the accidental load condition.
strains). The results of this assessment, presented as limit state
check in Fig. 4, show that acceptable conditions are predicted at
both planned and unplanned buckles. It is acknowledged that the
results presented in Fig. 4 used a characteristic strain for unplanned
buckles based on a certain probability of exceedance per km of
pipe, which is not entirely consistent with the overall reliability
targets in DNV-OS-F101 (DNV, 2013) that are defined per pipeline
rather than per km of pipeline.
A consistent assessment can be performed through the extraction
of the design load effect (i.e., the peak strain anywhere along the
pipeline with a probability of exceedance of 1%) from the Monte
Carlo run, as shown in Fig. 1. However, since the design resistance
is different for planned and unplanned buckles, this would not allow
assessing the overall integrity of the pipeline. Alternatively, the
probability of the peak strain being higher than the design resistance
at any point along the pipeline can be extracted. For the example
considered, this probability is calculated to be 0.06%. From a Fig. 5 Probabilistic axial displacement at ILT
246
ONGOING AND FUTURE DEVELOPMENTS
The approach presented in this paper is part of an ongoing
development of lateral buckling design tools and methods. The
following aspects are some of the key issues that are currently
being improved.
The proposed approach provides detailed results for the lat-
eral buckling response of the pipeline, but it is based on surro-
gate models that simplify the problem and have some restric-
tions on the results that are available. In order to confirm the
conclusions of the probabilistic assessment and derive further
information, an interface is being developed to remake selected
scenarios from the Monte Carlo assessment into nonlinear FE
models of the full pipeline. This will allow verifying the surrogate
model and provide detailed and accurate results of the critical
configurations.
The level of uncertainty in lateral buckling narrows down as the
design progresses and more data are available. The method presented
here is being developed to cope with that and allow different
levels of refinement. As part of this process, the formulations are
being adapted to work with more basic parameters, such as pipe
embedment and undrained shear strength instead of lateral friction,
so that as more information is made available, the probabilistic
predictions can be refined.
These refinements do not improve a design but increase con-
fidence in it. Moreover, for pipelines for which a wait-and-see
approach is possible (i.e., for which the lateral buckling design
can be completed with post-lay activities), refinements in analysis
can be used to minimize the extent of those interventions and to
optimize their timing. Ultimately, we intend to have probabilistic
and full 3D deterministic FE models of each pipeline that are
refined as the project progresses and can be used over the entire
design life of the asset.
CONCLUSIONS
The prediction of the lateral buckling response of a pipeline
requires detailed simulations. However, before the pipeline has
been installed, there is significant uncertainty in the data required
to perform those simulations. The best way to deal with this kind
of problem is to combine probabilistic assessments with detailed
FE models.
This paper presented a probabilistic lateral buckling assess-
ment approach that evaluates the actual limit states involved with
Probabilistic Lateral Buckling Assessment
global buckling and can therefore provide designs with consistent
probabilities of failure.
In the case study considered, the proposed approach allows
demonstrating the validity of a design that would be considered
unacceptable under the current industry best practice approach. This
shows that the proposed approach improves on the industry best
practice and should lead to more optimized designs that maintain
the required levels of safety. The improvements are expected to
be particularly significant in cases where naturally formed or
unplanned buckles are allowed to take place.
It is acknowledged that the proposed approach will require
good input data, in particular with regard to the expected OOS
features along the pipeline. It is also acknowledged that the method
presented is complex and will need to be used with caution.
However, despite these limitations, it is considered that the approach
presented provides an exciting way forward for the lateral buckling
design of subsea pipelines.
ACKNOWLEDGEMENTS
The first author would like to thank Kevin Williams for introduc-
ing him to Wooten’s laws and for years of fruitful and interesting
discussions.
REFERENCES
Bruton, D, and Carr, M (2011). “Overview of the Safebuck JIP,”
Proc Offshore Technol Conf, Houston, TX, USA, OTC-21671-
MS. http://dx.doi.org/10.4043/21671-MS.
Collberg, L, Carr, M, and Levold, E (2011). “Safebuck Design
Guideline and DNV RP F110,” Proc Offshore Technol Conf, Hous-
ton, TX, USA, OTC-21575-MS. http://dx.doi.org/10.4043/21575-
MS.
DNV (2007). Global Buckling of Submarine Pipelines, Structural
Design Due to HT/HP, Recommended Practice DNV-RP-F110,
Det Norske Veritas, Høvik, Norway.
DNV (2013). Submarine Pipeline Systems, Offshore Standard
DNV-OS-F101, Det Norske Veritas, Høvik, Norway.
Fausett, L (1994). Fundamentals of Neural Networks: Architectures,
Algorithms, and Applications, Prentice-Hall, 461 pp.
Hobbs, R (1984). “In-Service Buckling of Heated Pipelines,”
J Transp Eng, ASCE, 110(2), 175–189.
http://dx.doi.org/10.1061/(ASCE)0733-947X(1984)110:2(175).