You are on page 1of 9

Proceedings of the Twenty-ninth (2019) International Ocean and Polar Engineering Conference www.isope.

org
Honolulu, Hawaii, USA, June 16-21, 2019
Copyright © 2019 by the International Society of Offshore and Polar Engineers (ISOPE)
ISBN 978-1 880653 85-2; ISSN 1098-6189

Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE19/All-ISOPE19/1128950/isope-i-19-354.pdf by China University of Petroleum (East China) user on 11 March 2021
True Fatigue Life Calculation Using Digital Twin Concept and Operational Modal Analysis
Erik Bouvin Pedersen and Dennis Jørgensen
LICenginnering A/S
Esbjerg, Denmark

Hans Jørgen Riber


LICenginnering A/S
Svendborg, Denmark

Jonas Ballani, Sylvain Vallaghé and Benjamin Paccaud


Akselos S.A.
Lausanne, Switzerland

ABSTRACT

This paper presents a Digital Twin concept aimed at assets in the oil &
gas and wind industry, that provides an accurate estimate of the true
fatigue life of these assets in order to unlock potential fatigue life and
ultimately extend the life of assets. Data from a few strategically placed
sensors is decomposed into modal parameters by means of Operational
Modal Analysis (OMA). The modal parameters are expanded to a high-
resolution stress field solution (MDE) via a calibrated FE-model (Digital
Twin) representing the considered asset. The concept offers a compelling
and cost-effective method for offshore assets that are facing life time
extension beyond what current methodologies can provide. The concept
is currently being implemented on a platform in the UK continental shelf
of a supermajor oil company.

KEY WORDS: Digital Twin; Operational Modal Analysis; Modal


Decomposition and Expansion; Life Time Extension.
Figure 1. European oil platforms exceeding original design life assuming
INTRODUCTION an original 25-year design life.

According to the UK Health and Safety Executive (2012), approximately Low oil prices have fueled interest in extending the life of existing
50% of the fixed oil platforms on the UK continental shelf have exceeded brownfield operations beyond their original design life, instead of
their original 25-year design life. Using data from the European Marine investing in new greenfield operations. The life of existing oil platforms
Observation and Data Network (EMODnet) and assuming an original has been extended well beyond the original with improved simulation
25-year design life, approximately 38% of the all European oil platforms methods, but it is becoming increasingly difficult to extend the life
have exceeded their original design life. Over the next decade, this further using current simulation methods. This has started the Digital
number will increase to around 80%. This is illustrated in Figure 1. Twin revolution, where 1:1 digital representations of real-world assets
are able to describe the exact behavior of these assets. Four governing
parameters control the motion of any structure: mass, damping, stiffness
and load. In numerical models, there are significant uncertainties for
these parameters which often results in overly conservative prediction of
fatigue life. Digital Twins, based on accurate physics-based models

3969
augmented by real-world sensor data, drastically minimizes these can be used to evaluate how well the numerical model represents the
inherent uncertainties, leading to improved estimates of the true fatigue real-world conditions. If there is a good match between the design
life. When establishing Digital Twins for offshore assets, the nature and model and the real-world model, based on evaluation on modal
variation of modelling uncertainties to be considered are quite parameters, a Digital Twin has been established. Most likely, an initial
widespread. good match will not be achieved and the numerical model will have to
be updated to better represent real-world conditions. A general
A study (Cook, 1982) of measured accelerations of a single pile oil methodology of how model updating should be carried out can not be
platform over a three-day period, found that the total damping ratios given, as it is case-by-case specific, but important factors for offshore
varied from 1.1% to 1.5% over this period of time. The main contributor assets are top-side mass and damping. The procedure for establishing a
to the variation of the total damping ratio was found to be the soil Digital Twin is illustrated in Figure 2.
damping, which varied from 0.41% to 0.88% over the three-day period.

Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE19/All-ISOPE19/1128950/isope-i-19-354.pdf by China University of Petroleum (East China) user on 11 March 2021
Real-world
The theoretical lower and upper bound soil damping was calculated to sensor data
0.53% and 0.88% respectively. While the ranges of the theoretical and (Bandpass filtered)
measured soil damping might be close, the theoretical values do not give
any guidance of the exact damping ratio. A study (Jensen, 1990) of
measurements from 34 oil platforms (including jackets), showed typical Operational Modal
damping ratios for offshore jackets in the range of 1-3% with a Analysis
(Modal parameter
coefficient of variation of 46%. The significant variations indicate that identification)
damping is very case-by-case specific and varies over time. The same
study found an uncertainty of the mass of about 20% over the life of the
asset, leading to significant deviations between estimated and measured Modal parameters from Evaluate numerical
natural frequencies. For a specific platform, it was found that the natural numerical model model
(Natural frequency, (Natural frequency,
frequency decreased 10% over a 5-year period, indicating that the damping, Eigenmodes) damping, MAC)
behavior of offshore oil platforms is highly time dependent.

In addition to addressing uncertainties in the modal properties of the


asset properly, accurate numerical modelling of complex structural parts Numerical model
Update numerical model
has a significant influence on the predicted fatigue damage. HSE (2002) No
satisfactory?
found that inclusion of local joint flexibly (LJF), increased fatigue lives
with up to a factor of 19.3 for the Leman CD platform, compared to
Yes
traditional rigid joint modelling. Dubois (2013) found a reduction in
fatigue damage up to 22% by incorporating superelements compared to Digital Twin
LJF. Established

Figure 2. Procedure for establishing a Digital Twin.


When modelling environmental loading, governing parameters such as
hydrodynamic coefficients and marine growth thickness are often Real-time true fatigue calculation
assumed constant over the life of an asset. Sea states are assumed to be
controlled only by three parameters, significant wave height, peak period
Once the Digital Twin has been established, the true fatigue life can be
and water depth, while the reality is much more complex. These
calculated in real-time using Modal Decomposing and Expansion
simplifications are convenient in the design process where better data is
(MDE). With MDE, the acceleration measurements from a few
not available, but they can lead to overly conservative predictions of the
strategically placed sensors can be expanded to the full stress field
actual environmental loads the asset is subjected to.
solution i.e. allowing extraction of stress at any point on the asset.
This paper presents a concept of how to identify and remove the inherent
First, the sensor data is bandpass filtered using Butterworth filtering to
uncertainties in mass, damping, stiffness and load by use of real-time
remove noise. This step is crucial to avoid drift due to accumulation of
sensor data of structural response and environmental load combined with
the noise, from double integration of sensor data carried out in a later
faster and more accurate numerical modelling techniques. This results in
step. Afterwards, the accelerations are transferred to modal coordinates
a shift from design fatigue life to an accurate estimate of the true fatigue
by projection, using the eigenmodes from the OMA routine:
life of offshore assets.
+ 𝑢̈ (𝑡)
𝑞̈ (𝑡) = Φ𝑚 𝑚 (1)
DIGITAL TWIN
+ is Moore-Penrose inverse of the OMA eigenmodes.
Where: Φ𝑚
Establishing Digital Twin 𝑢̈ 𝑚 (𝑡) is the measured accelerations.
As a first step, a Digital Twin, i.e. a finite element (FE) representation Then the modal accelerations are double integrated to determine the
of the real-world asset needs to be created. An as-designed FE model modal displacements. At each integration step the data is filtered in order
will typically have discrepancies compared to the real-world asset, to avoid drift, due to accumulation of small errors (magnitude increases
which needs to be identified and mitigated in a calibration procedure. squared from integration).
The primary tool of identification is operational modal analysis
(OMA), see Brincker et al. (2015). Using OMA, the modal parameters
(natural frequencies, damping, eigenmodes) can be identified from
acceleration measurements of the real-world asset in its operational
state. The modal parameters directly relate to the matrices on the left-
hand side of the equation of motion, i.e. mass, damping, stiffness and

3970
Multiplying the modal displacements with the eigenmodes, from the
Digital Twin, results in an approximation of the full time-dependent
displacement field of the asset:

𝑢𝑝 (𝑡) = Φ𝑝 𝑞(𝑡) (2)

Where: Φ𝑝 is the FE eigenmodes.

It should be noted, that the desired resulting quantity should relate to the
corresponding modal quantity and eigenmode quantity, e.g. for

Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE19/All-ISOPE19/1128950/isope-i-19-354.pdf by China University of Petroleum (East China) user on 11 March 2021
prediction of motions, modal accelerations should be multiplied with
acceleration eigenmodes and for prediction of stress/strain modal
displacements should be multiplied with stress/strain eigenmodes.

The procedure is illustrated in Figure 3.


Real-world
sensor data
(Bandpass filtered)

Operational Modal
Analysis
(Eigenmode
identificarion)

Modal
decomposition

Double integration Figure 4. Overview of laboratory model.


(Bandpass filtering
at each step)

Eigenmodes from
Modal expansion
numerical model

Full field solutions

Figure 3. Procedure for obtaining full field solutions.

LABORATORY EXPERIMENTS Figure 5. Overview of sensor clusters.

To demonstrate the procedures of establishing a Digital Twin and The laboratory model has been excited using a solenoid with broom
carrying out MDE (illustrated in Figure 2 and Figure 3), a series of hairs. The solenoid introduces an impulse load at the edge of the
experiments has been carried out on a laboratory model. The laboratory laboratory model, thereby introducing both bending and torsional
model is a 1.68m tall lattice structure, with the structural complexity of motion. In an experiment, the laboratory model was excited by the
an offshore jacket foundation. The laboratory model is equipped with 8 solenoid for 600s while the accelerations and displacements where
accelerometers and 3 displacement sensors. The laboratory model and measured with a sampling rate of 1612 Hz and 400 Hz respectively. The
the sensors and their locations are illustrated in Figure 4 and Figure 5. measurement from one acceleration channel is illustrated in Figure 6.

3971
Comparing the eigenmodes from the RB-FEA model with those from
ARTeMIS Modal, it is found that there is an excellent match, with MAC-
values in the diagonal between 0.94 and 0.99, adjacent MAC-values are
close to zero. The MAC-matrix is illustrated in Figure 8.

Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE19/All-ISOPE19/1128950/isope-i-19-354.pdf by China University of Petroleum (East China) user on 11 March 2021
Figure 6. Acceleration measurement at location (+X,+Y) of sensor
cluster 1.

The measurements have been bandpass filtered with 5th order


Butterworth filter using a filtering interval of 2-300 Hz. The filtering Figure 8. MAC-matrix comparing FE and OMA eigenmodes.
interval has been chosen such that the relevant response is included.
From the filtered accelerations, the modal parameters have been The excellent MAC-values, indicate that the RB-FEA model is a good
identified using the ARTeMIS Modal software. A Digital Twin of the representation of the laboratory model. It has not been necessary to
laboratory model has been set up in the Akselos Integra software, which update the RB-FEA model, as there was an excellent match from the
provides accelerated high-fidelity structural analysis via the Reduced start. This is due to the exact nature of the laboratory model that has
Basis (RB-FEA) technology (Akselos 2018), see Figure 7. The model almost no uncertainties. For a real-world asset, it is very unlikely that the
with solid finite elements possesses 3.1 million FE degrees of freedom modelling would be exact from the start due to the uncertainties
and can be solved with the Akselos Integra software within a few seconds described in the introduction. In this case, the RB-FEA technology can
on a single computing core. While the solution time is comparable to a be used to automatically calibrate the model parameters using
1D beam model, the full 3D stress field is resolved in this model, which parameterized components (Akselos 2018).
is of particular importance in complex structural components like the
joints, see Figure 14. Modal decomposition of the measured accelerations is carried out with
expression 1, resulting in accelerations in modal coordinates. The
accelerations in modal coordinates are integrated twice resulting in
displacements in modal coordinates. The displacements in modal
coordinates are expanded to the full displacement field, using expression
2.

To demonstrate the accuracy of the MDE procedure, the displacements


have been determined in location (-X,+Y) of sensor cluster 1, a location
with a displacement sensor, but no accelerometer. In Figure 9, the
predicted displacements and the measured displacements are compared
for the entire experiment. In Figure 10, a five second section of the
results is illustrated to allow for closer inspection of the results.

Figure 7. Overview of Akselos Integra model of laboratory model.

3972
Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE19/All-ISOPE19/1128950/isope-i-19-354.pdf by China University of Petroleum (East China) user on 11 March 2021
Figure 9. Measured displacements and displacements predicted by MDE Figure 11. Histogram of displacement ranges at location (-X,+Y) of
at location (-X,+Y) of sensor cluster 1. sensor cluster 1.

In Figure 12, the power spectral densities (PSD) of the predicted and
measured displacements are illustrated. From the PSD, it is seen that
there is an excellent match for the three eigenmodes corresponding to
bending about y-axis, torsion and bending about x-axis and that it is these
eigenmodes that govern the majority of the displacements at this
location.

Figure 10. Five second section of the Measured displacements and


displacements predicted by MDE at location (-X,+Y) of sensor cluster 1.

From Figure 9 and Figure 10, it can be seen that in overall there is an
excellent match between the predicted displacements and the measured
displacements, but there is a general tendency of the measured
displacements being larger, in the order of 1/100th mm. In Figure 11,
Figure 12. PSD of the measured displacements and the displacements
displacement range histograms, based on the predicted displacements
predicted by MDE.
and the measured displacements, calculated using Rainflow counting,
are illustrated. From Figure 11, it can be seen that there is an excellent
match between predicted and measured displacement ranges. ACCURATE MODELLING OF COMPLEX STRUCTURAL
PARTS

Accurate modelling of complex structural parts, is important as the local


stiffness of these parts dictate the local stress distribution. Traditionally,
joints have been modelled using beam elements that intersect in the
centerline between the chord and the braces. This results in overly stiff
joints, since in reality the chord and braces intersect at the chord surface.
To mitigate this issue, parametric formulae by Buitrago (1993) that
include local joint flexibility (LJF) has been adopted by the industry.
Using these parametric formulae, springs are incorporated into the beam
model to take into account the additional flexibility of the joints. An even
more accurate approach is to directly include the joint as shell or solid
elements.

3973
The computational cost of this approach is usually mitigated by use of NEAR REAL-TIME ENVIRONMENTAL MODELLING
reduction methods such as Guyan or Craig-Bampton. The three
approaches are illustrated in Figure 13. The irregularity and the random shape of the ocean is described using
superposition of linear Airy wave components. Each wave component is
described by amplitude, frequency and phase. The amplitude and
frequency given by a wave spectrum, which is a description of a given
sea state in the frequency domain. The JONSWAP spectrum describes
fetch limited seas and is commonly used for offshore assets in the North
Sea. The controlling parameters of the wave spectra is the sea state
parameters, significant wave height (Hs) and peak period (Tp). The

Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE19/All-ISOPE19/1128950/isope-i-19-354.pdf by China University of Petroleum (East China) user on 11 March 2021
JONSWAP spectrum is suitable for sea states in the range of 3.6 <
𝑇𝑝
< 5.0 according to DNVGL-RP-C205 (2017). The duration of a sea
√𝐻𝑠
state is commonly set to 3 hours and it is assumed that the sea surface is
Figure 13. Rigid, LJF and superelement modelling of K-joint, stationary over its duration.
Schaumann (2007).

Accurate modelling of complex structural parts is primarily relevant for


local behavior. Schaumann (2007) found that there is virtually no
difference in natural frequency for the first global eigenmodes, between
modelling the joints as rigid, including LJF or using superelements. For
higher eigenmodes, exhibiting local behavior there is a significant
difference in natural frequency between the modelling approaches. There
was up to an 9.8% difference in natural frequencies, between rigid joint
modelling and superelement modelling for higher eigenmodes. For LJF
and superelement modelling, this difference was up to 4.9%. The
difference in local stiffness has a significant influence on stress
distribution and thereby fatigue damage. HSE (2002) have found that
inclusion of LJF increased fatigue lives with up to a factor of 19.3 for the
Leman CD platform. Schaumann (2007) found that including LJF
resulted in increased fatigue life up to a factor of 9. Dubois (2013) found
a reduction in fatigue damage up to 22% by incorporating superelements
compared to LJF. The industry has not widely adopted the superelement
approach, as it is significantly more computationally expensive than
beam elements, especially in post-processing where the reduced solution Figure 15. Overview of peak shape parameters, Hasselmann et al.
has to be expanded in order to obtain the full field solution. The RB-FEA (1973).
method, used in Akselos Integra, addresses both simulation accuracy and
simulation time. With the RB-FEA method, the solution is accelerated The magnitude of the spectral peak is controlled by the peak shape factor
up to 1000 times compared to conventional FEA (Akselos, 2018), which (γ) and the width on each side of the spectral speak is controlled by the
makes highly complex and large models a practical reality at virtually no spectral width parameters (σa, σb). Commonly, mean values of these
additional computational cost. As an example, a solid finite element parameters (γ=3.3, σa=0.07, σb=0.09) are used, but experiments show
representation of the laboratory model has been subjected to a load from that the peak shape parameters are significantly scattered with respect to
one side of the structure. Resulting stresses are illustrated in Figure 14. non-dimensional fetch, as illustrated in Figure 16. Hasselmann et al.
(1973) suggested that it is likely the gustiness of the wind that is a
primary cause of this scatter.

Figure 14. Resulting stresses from solid finite element with 3.1 million
FE degrees of freedom, solved with Akselos Integra software within 6
seconds using RB-FEA technology.

3974
comparison, two sea states have been identified as having the best and
worst compliance between the DFT and the JONSWAP spectrum. These
are illustrated in Figure 18.

Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE19/All-ISOPE19/1128950/isope-i-19-354.pdf by China University of Petroleum (East China) user on 11 March 2021
Figure 16. Scatter of peak shape parameters for the JONSWAP
spectrum. Mean values indicated by red line. Hasselmann et al. (1973).

To investigate how accurately the JONSWAP spectrum with mean peak


Figure 18. Best (sea state 4) and worst (sea state 7) compliance between
shape parameters describe an actual sea state, 21 hours of buoy data
the JONSWAP spectrum and DFT of buoy data.
sampled at 4 Hz from a supermajor on the UK continental shelf, has been
analyzed. The buoy data has been divided into 7 three-hour sea states,
From Figure 18, it is seen that for sea state 4 there is generally a good
which are analyzed individually. The sea states are illustrated in Figure
compliance between the DFT of the wave elevation and the JONSWAP
17.
spectrum, but there are some deviations around the spectral peak and at
lower frequencies. For sea state 7, there is a poor compliance over the
entire frequency range and the magnitude of the spectral peak is
significantly lower for the buoy data. These deviations will lead to
significant difference in fatigue damage. To determine the difference in
fatigue damage, the wave load corresponding to the JONSWAP
spectrum and the buoy data is determined.

In order to determine the wave load from the buoy data, it is decomposed
into series of sine expressions. The amplitude, frequency and phase shift
of each wave component is determined from the DFT of the buoy data.
With the calculated amplitudes, frequencies and phase shifts, every wave
component is identified. From that, the particle kinematics are calculated
by linear Airy wave theory for each wave component. Each contribution
Figure 17. Time series of water depth for identified sea states. is superposed into the complete solution. It should be noted that the wave
load is calculated at the location of the buoy and not at the asset. In order
Each sea state is band pass filtered in the range 0.01-1.00 Hz using a to make this methodology relevant in real-world scenarios,
second order Butterworth filter in order to detrend the data. From the measurements should be taken at the location of the asset or work should
filtered data, the individual wave heights have been identified using zero- be put into transformations of measurements from nearby buoys.
down crossing. The peak periods are determined from the DFT of the
wave elevation. The identified sea state parameters are listed in Table 1. To investigate the difference in response and fatigue damage, a simple
beam system has been setup. The system consists of two Timoshenko
Table 1. Identified sea states and sea state parameters from buoy data. beam elements of equal length, clamped at one end and with a 7 Mt
lumped mass at the other. This results in a system with a first natural
Sea state Significant wave Peak period (Tp) [s] frequency of 0.50 Hz, similar to that of an offshore oil and gas asset. The
height (Hs) [m] system is illustrated in Figure 19.
1 2.13 7.59
2 1.85 7.06
3 2.21 6.61
4 3.27 8.19
5 3.25 8.90
6 2.25 7.59
7 1.99 7.06

The JONSWAP spectrum is calculated for each identified sea state in


accordance with DNVGL-RP-C205 (2017). The calculated spectra are
compared with the DFT of the sea states. From review of this

3975
mean of the ratio between the fatigue damage from the JONSWAP
spectrum and the buoy data has been calculated for each sea state and is
illustrated in Figure 21.

Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE19/All-ISOPE19/1128950/isope-i-19-354.pdf by China University of Petroleum (East China) user on 11 March 2021
Figure 19. Overview of simple beam system. Figure 21. Mean ratio of damage from synthetic wave train and buoy
data for each sea state.
The wave loading is simplified such that the particle kinematic is
constant from wave crest to seabed where elevation -22.25 m wrt. MWL From Figure 21, it can be seen that the fatigue damage from the buoy
is taken as the governing elevation. The wave load is determined using data is generally, but not always, smaller than that from the JONSWAP
the Morison equation excluding relative kinematics. The system has spectrum. It is assumed that some of the differences between the
been solved using Newmark time integration. When plotting the JONSWAP spectrum and the buoy data is due to that some of the sea
displacements at the top of the model, it is difficult to discern the states are outside validity range of the JONSWAP spectrum. This is
difference between the response due to the JONSWAP spectrum and the illustrated in Figure 22, where it can be seen that there is correlation
buoy data. To create an easily discernible plot, the RMS value of the top between exceedance of the validity ranges and deviation in fatigue
displacements has been calculated for each second and it has been damage from the synthetic wave train and the buoy data. The correlation
cumulatively summarized. This is illustrated in Figure 20. is not perfect and does therefore not explain the entirety of the deviations.
Other cause of the deviations might be the gustiness of the wind, as
suggested by Hasselmann et al. (1973).

Figure 20. Cumulative sum of RMS of displacements at top.

From Figure 20, it can be seen that the development of displacement Figure 22. Exceedance of validity range of the JONSWAP spectrum.
between the JONSWAP spectrum and the buoy data is very similar for
sea state 4, but for sea state 7 the slope of the curve is much greater for CONCLUSIONS
the JONSWAP spectrum. This is in line with expectations, when taking
the spectra of Figure 18 into consideration. The fatigue damage has been Over the next decade, around 80% of all European oil platforms will
calculated for each sea state with 100 realizations, to ensure proper have exceeded their original design life. With the increasing age of these
statistical variation. For each sea state, the fatigue damage has been platforms, it is becoming increasingly difficult to warrant life time
calculated using rainflow counting, the Palmgreen-Miner rule and the extension using existing methodologies.
SN-methodology in accordance with DNVGL-RP-C203 (2016). The

3976
In numerical models, there are significant uncertainties for the REFERENCES
parameters that govern the structural motions (mass, damping, stiffness
and load), which often results in overly conservative prediction of fatigue Akselos (2018). “Component-Based Reduced Basis Simulations”.
life. https://c7d7e7n5.stackpathcdn.com/wp-
content/uploads/2018/10/White-paper-component-based-reduced-
From the literature, it has been found that there are significant deviations basis-simulations-08102018.pdf
between the design values of global mass and damping parameters Brincker, R., and Ventura, C. (2015). “Introduction to Operational Modal
leading to deviations in prediction of structural behavior. In a study, it Analysis”, Wiley.
was found that the structural damping increased from 1.1% to 1.5% over Buitrago, J, Healy, B E and Chang, T Y (1993). “Local Joint Flexibility of
a three-day period. The same study found an uncertainty of the mass of Tubular Joints” Proceedings, OMAE 1993, 12th Intl Conf on Offshore
about 20% over the life of the asset. Accurate description of local

Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE19/All-ISOPE19/1128950/isope-i-19-354.pdf by China University of Petroleum (East China) user on 11 March 2021
Mechanics & Arctic Engineering, Vol I, P405.
stiffness of complex structural parts has been found to increase fatigue Cook, F, Michael and Vandiver, Kim, J (1982). “Measured and Predicted
life up to 19.3 times compared to rigid joint modeling and up to 22% Dynamic Response of a Single Pile Platform to Random Wave
increase compared to LJF. Excitation” OTC 4285.
DNVGL-RP-C203 (2016). “DNVGL-RP-C203 - Fatigue design of
A methodology has been presented, addressing the inherent uncertainties offshore steel structures”
when modelling offshore assets using a Digital Twin concept, OMA and DNVGL-RP-C205 (2017). “DNVGL-RP-C205 - Environmental
MDE. The combination of OMA, MDE and a calibrated and highly conditions and environmental loads”
detailed structural finite element model, provided through the RB-FEA Dubois, J, Muskulus, M and Schaumann, P (2013). “Advanced
technology, allows to maintain an accurate representation of a real asset Representation of Tubular Joints in Jacket Models for Offshore Wind
throughout its entire life time. Turbine Simulation” Energy Procedia,Volume 35, 2013, Pages 234-
243
From analysis of real-world data, it has been found that there are Hasselmann, K. et al. (1973). “Measurements of wind-wave growth and
deviations in the spectral characteristics between the commonly used swell decay during the Joint North Sea Wave Project (JONSWAP)”
JONSWAP spectrum and measured data. The deviations can lead to Ergänzungsheft zur Deutschen Hydrographischen Zeitschrift Reihe A,
significant changes when estimating fatigue life. Generally, it was found Nr. 12
that basing environmental loading on measured data leads to an increase Health and Safety Executive (2002). “The effects of local joint flexibility
in fatigue life, but sufficient data has not been analyzed to give definitive on the reliability of fatigue life estimates and inspection planning”
conclusions. A method of how to calculate wave loading from real-time Offshore Technology Report 2001/056.
buoy data has been presented. Health and Safety Executive (2012). “Further information on ageing and
life extension”.
The Digital Twin concept, OMA and loading from real-time http://www.hse.gov.uk/offshore/ageing/background.htm
environmental loading will in combination significantly increase the Jensen, Jakob Laigaard (1990). “Full-Scale Measurements of Offshore
predicted fatigue life of assets compared to current methodologies. Platforms” Fracture and Dynamics, Nr. 17, Bind. R9002
Schaumann, P and Böker, C (2007). “Ermüdungsbeanspruchung aus
The concept has been validated via experiments under laboratory Seegang bei aufgelösten Offshore-Tragstrukturen” Stahlbau 76, Heft 9
conditions. The concept is currently being implemented on a platform in
the UK continental shelf of a supermajor oil company. From that, the
concept will be validated under real-world conditions in the near future.

3977

You might also like