See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/287016178

Practical fatigue calculation for offshore structures based on efficient wave

spectrum inputs

Article · January 2008

CITATION READS

1 274

3 authors, including:

Junbo Jia Tore Holmas

Aker Solutions, Bergen usfos as
101 PUBLICATIONS   130 CITATIONS    13 PUBLICATIONS   62 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Junbo Jia on 01 June 2016.

The user has requested enhancement of the downloaded file.

 Engineering Structures 33 (2011) 477–491

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Wind and structural modelling for an accurate fatigue life assessment of

tubular structures
Junbo Jia ∗
Aker Solutions, Bergen, Sandslimarka 251, 5861 Sandsli, Norway

article info abstract

Article history: Based on the joint probability with regard to wind speed and its main direction, the current paper presents
Received 4 January 2010 a practical and efficient approach for calculating wind induced fatigue of tubular structures, the effects
Received in revised form of the wind direction, across wind and wind grid size on the high cycle fatigue of the structure are
3 October 2010
addressed. In each time step of the dynamic response calculation, the large deformation effects and the
Accepted 2 November 2010
Available online 8 December 2010
wind induced drag forces due to the updated structural deformations are taken into account. It is found
that, the directional wind effects on the fatigue damage mainly depend on the orientation of the structure,
Keywords:
the location and the support condition of the selected joints, and the relative probability of occurrence
High cycle fatigue for the high winds speed in each direction, etc. Furthermore, the across wind components are proved
Along wind to be a significant contributor on the fatigue damage and cannot be ignored. The fatigue damage is also
Across wind found to be rather sensitive to the wind grid size for generating the wind fields. It is also concluded that
Wind direction vibration of each individual member interacts with the global dynamic response and the wind loading, and
Flare boom a fatigue check should therefore be against both individual member and global response. The wind fatigue
Nonlinear dynamics calculation procedure presented in the current paper has the merit of reducing uncertainties without
Non-Gaussian degrading a required safety level, this may lead to a positive economic impact with regard to construction
HHT-α method
and maintenance costs. It has been applied on quite a few industry and research projects and can be widely
Time integration step
Industry practice applied on the similar study of structures.
© 2010 Elsevier Ltd. All rights reserved.

1. Introduction
As structures tend to become lightweight and slender, the
knowledge obtained from traditional engineering practice may not
be applied directly on the slender structure, sophisticated analysis
approaches have to be adopted in the design and reassessment
process. Among those analyses, the wind induced vibration
calculation on slender tubular structures is one of the challenging
studies. The major challenge involved is the nonlinear behaviour
due to the geometric nonlinearity and fluid-structure nonlinearity.
For small and medium size structures, the effects due to geometric
nonlinearity are small and can be accounted for by safety factors
or even be neglected. Some nonlinearity effects can also be locally
taken into account in post processing procedures. However, for
high rise slender structures, these effects cannot be neglected.
Wind loading is by its nature dynamic and therefore normally
induces vibrations at the structure’s natural frequencies. These vi-
brations produce fluctuating stresses which lead to the fatigue
damage accumulation and can cause the structural failure without
exceeding the design wind actions. At the global level the lateral
∗ Tel.: +47 55224867.
E-mail addresses: jiajunbo@sina.com, junbo.jia@akersolutions.com.
0141-0296/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2010.11.004
wind load comprises in order of 10% and 25% of the total lateral load
for fixed and floating platforms, respectively. While only very few
researchers have studied the individual module contribution to the
overall wind load for the design of offshore platforms [1]. Under
extreme wind loading, the structure may reach ultimate strength
and lead to local or global structural failures. This requires sub-
stantial repairing or even replacement. Damage to the deck struc-
tures due to extreme wind loads has been reported by Kareem and
Smith [2]. Deoliya and Datta [3] studied the reliability of a 75 m tall
microwave tower under wind loads with different combinations of
mean wind speed and direction. They concluded that the probabil-
ity of strength failure for the studied steel lattice tower increases
with the increase in the standard deviation of mean wind speed,
the zero crossing rate and the correlation between the velocities in
two directions.
Quite a few pieces of research work have been reported con-
cerning wind induced fatigue. Repetto [4] proposed a method that
is applied to several different bimodal processes which are usual in
dynamic response of structures. Based on a bimodal representation
of the along wind induced stress power spectral density functions,
Repetto and Solari [5] formulated a counting method to estimate
the fatigue damage for only along wind actions on structures. Since
the wind-induced fatigue is sensitive to moderate wind velocities
for which stable or unstable atmospheric conditions can occur [6],
478 J. Jia / Engineering Structures 33 (2011) 477–491
Repetto and Solari [7] evaluated the wind-induced fatigue by tak-
ing the stable, unstable and neutral condition of the wind field into
account and applied this approach to the fatigue calculation of a
steel chimney. By including viscoelastic dampers in the modelling,
Palmeri and Ricciardelli [8] had studied the fatigue life of structural
components of tall buildings due to the buffeting response.
Since the directional effects of wind on the structural response
may give engineers a better understanding of the structural re-
sponse with regard to wind resistance, life-time extension and
maintenance, and structures’ reorientation, their influence on the
dynamic structural response has been studied by Davenport [9],
Simiun and Filliben [10], and Wen [11] using simplified structure
models. The latter two articles conclude that the worst direction
approach used for evaluating wind induced response may signifi-
cantly overestimate the response. It should be noted that for a com-
plex structure like a high rise flare boom, a simplified structural
modelling may lead to an inaccurate estimation of fatigue damage.
Because the frequency domain analysis cannot properly take
the non-linear load effects, large deformation as well as the
plasticity into account, and also because the power spectrum of the
critical stress due to the dynamic wind loading may not be narrow
banded initiated from the influence of the background components
of incident turbulence [12], crosswind induced vibrations, wind
directional effects and structural damping etc., the inappropriate
use of frequency domain based spectral methods developed for
Gaussian processes may then significantly underestimate the
fatigue damage contribution [13]. While in the mean time, the
spectral methods based on non-Gaussian process are still under
development. This is because if stress time history cherishes a
greater probability than Gaussian probability regarding taking
large stress values, this is then likely to induce a large stress range,
and may consequently cause significantly accelerated fatigue
damage [14–16]. Therefore, in recent years, the time domain
dynamic analysis of wind sensitive slender structures such as high
rise towers, flare booms, and long span bridges has attracted a
lot of research efforts [17–19]. Based on the linear finite element
calculation in the time domain, Gioffrè and his co-workers [20]
studied the wind induced response of a guyed mast for mobile
phone networks under the influence of along wind and across
winds separately. Their study proves that the across wind forces
cannot be neglected when calculating the response of the mast,
especially when dealing with severe prescriptions on the allowable
displacements. Repetto and Solari [21] proposed a mathematical
model to derive a histogram of the stress cycles, the accumulated
damage and the fatigue life of slender vertical structures (e.g.
towers, chimneys, poles and masts) in along wind vibrations. In
order to provide the information for the vibration of the Nanjing
TV tower with passive and/or active mass damper, Feng and
Zhang [22] performed a stochastic dynamic response of the tower
under turbulent wind and included both along and across wind
effects involving vortex induced vibration, drag and lift force. By
taking the joint probability density function of mean wind speed
and direction into account, Gu and his co-workers [12] presented
a method in the mixed frequency-time domain for estimating the
fatigue life of steel girders of the Shanghai Yangpu cable-stayed
Bridge due to buffeting (under along wind loading). They predict
that the fatigue life from the calculation is much longer than the
design life of the bridge and shows significant dependency on the
wind direction.
Offshore flare booms are mostly constructed with tubular com-
ponents due to their low drag coefficients and high strength to
weight ratio characteristics. However, due to the reason that tubu-
lar members give rise to significantly high stress concentrations
in the joints and a flare booms are normally slender and show
significant dynamic effects in low frequency ranges, the fatigue
life is then one of the major concerns. Wind turbulence is the
governing load with regard to fatigue for flare booms. It is ob-
vious that the failure of flare booms can lead to the substantial
damage through oil slicks as well as fire hazards on the deck. In
addition, due to its slender characteristics, the structural non-
linearities due to large deformation may significantly influence
the calculated structural response through creating a significant
additional bending moment compared to that based on small de-
formation assumptions. The procedures established in existing
standards only consider vibrations of individual members due to
vortex shedding. However, vibration of each structural member
interacts with the global structural response and the wind load-
ing. Severe vibrations have been reported on several platforms in
the North Sea, the actual vibration amplitudes were up to 20 times
higher than the ones obtained from the model testing of individ-
ual members. Cracks have also been found on several flare booms
caused by a combination of individual member and global vibra-
tion of the structure. It is then not sufficient to assess the fatigue
damage of a structure only based on the behaviour of each isolated
individual member. Due to the effects of S–N curves, the fatigue
damage is much more sensitive to the variation of loads, structural
modelling and the calculation algorithms than that of the stress,
thus a careful consideration of load and structural modelling must
be made before calculating the fatigue damage.
By discussing the wind field simulation and nonlinear finite
element analysis, the current paper first presents a practical
approach for establishing the Fatigue Limit State (FLS) model
through calculating the wind induced fatigue life of a typical flare
boom installed on an offshore oil platform. In order to include the
nonlinear structural and load effects into consideration, the current
paper adopts the direct integration method (implicit method):
the HHT-α method in the FE code USFOS [23] for calculating the
dynamic response. Dynamic analysis is performed to simulate
the response of the flare boom with 72 wind load cases/blocks,
and calculating the fatigue damages by taking the probability of
each individual wind load case/block. The high cycle fatigue has
been calculated by applying the ‘rain-flow’ counting method and
the Miner rule. Both the along and across wind components are
applied. It is found out that the directional wind effects on the
fatigue damage mainly depends on the orientation of the structure,
the location and the support condition of the selected joints, and
the relative probability of occurrence for the high wind speed
in each direction. Furthermore, the across wind components are
a significant contributor to the fatigue damage and cannot be
neglected. The fatigue damage is also found to be rather sensitive to
the wind grid size for generating the wind fields. The wind fatigue
calculation procedure presented in the current paper can be widely
adopted on the similar study on high rise tubular structures.
2. Simulation of wind field
It is assumed that the instantaneous wind speed comprises
a mean wind part and a fluctuating part as shown in Fig. 1. By
denoting the turbulent wind components u, v , and w in the along
wind (x), horizontal across wind (y) and vertical across wind (z)
direction and the mean wind component as U ′ , a wind velocity
vector at time t can then be expressed in a Cartesian coordinate
system as expressed in Eq. (1):
T
V {x, y, z , t } = U ′ + u, v, w .

(1)
2.1. Mean wind speed U ′ for along wind
The mean along wind speed at height z above the sea surface
and corresponding the mean period T (s) not more than 3600 s can
be calculated by a mean wind profile as expressed by Eq. (2) [24].
[  ]
T
U ′ (z , T ) = U (z ) · 1 − 0.41Iu (z ) · ln (2)
3600
where the one hour mean wind speed U (z ) (m/s) is given by
 J. Jia / Engineering Structures 33 (2011) 477–491 479

Z Table 1
Coefficient and separation for the 3 (h = 1, 2, 3)-dimensional coherence spectrum.
h αh rh ∆h qh ph

1 2.9 0.92 |x2 − x1 | 1.00 0.4

Utotal(z, t) =U'(z)+u(z, t) 2 45.0 0.92 |y2 − y1 | 1.00 0.4
3 13.0 0.85 |z2 − z1 | 1.25 0.5

the period of which the simulation is considered, and that spatial

homogeneity ensures similar statistical properties in all regions in
space. The along wind separation |x2 − x1 | is then neglected and
can be assumed as zero, this implies that the properties in the along
wind direction are defined as the one-point wind spectra. It should
be noted that other relevant types of coherence functions instead
of the one expressed in Eq. (5) can also be adopted.
Fig. 1. Typical wind profile composed of mean wind speed components (U ′ (z ))
In order to reduce the computation efforts as well as avoid
and fluctuating wind speed component u(z , t ) varying with height z and time t. searching for the information concerning the Suv , Suw , Svw compo-
 z  nents from the full-scale measurement, it is further assumed that
the cross covariance between u, v and w are insignificant, which

U (z ) = U0 · 1 + C · ln
10 means that the u, v and w components can be obtained indepen-
where dently and the fluctuating parts of the wind spectra of can then be
expressed by Eq. (6):
C = 5.73 · 10−2 · [1 + 0.15U0 ]0.5
S {∆, ωn } ≈ diag[Suu , Svv , Sww ]. (6)
and where the turbulence intensity Iu (z ) is given by
 z −0.22 By subdividing the frequency range into N segments (1, 2, . . . ,
Iu (z ) = 0.06 · [1 + 0.043U0 ] · N) and the flow field into M points in space grids, an M by M cross-
10 spectral density matrix can then be expressed for each sample fre-
where U0 (m/s) is the one hour mean wind speed at z = 10 m quency ωn and each flow component by Eq. (7):
above the ground or sea surface, the information of U0 and
S11 (ωn ) S12 (ωn ) S1M (ωn )
 
its probability of occurrence are normally given in the relevant ···
metocean and wind recording report.  S21 (ωn ) S22 (ωn ) ··· S2M (ωn ) 
S {ωn } =  . (7)
The fluctuating parts are simulated from wind spectra and ··· ··· ··· ··· 
coherence functions such that the spatial statistical properties are SM1 (ωn ) SM2 (ωn ) ··· SMM (ωn )
maintained in the simulated time series.
A Cholesky decomposition can be performed on the matrix equa-
tion (7) so that it can be rewritten as the product of lower triangular
2.2. Calculation of the turbulence wind components u, v , and w
matrix H (ωn ) and its transposed complex conjugate as expressed
by Eq. (8):
By subdividing the frequency range into N segments (1, 2,
. . . , N), the turbulence part of the wind field can be defined by a S (ωn ) = H {ωn } · H ∗T {ωn } (8)
3 × 3 relevant cross spectra matrix varying with the frequency ωn
as described in Eq. (3): where:
S {x, y, z , ωn } = u, v, w. H11 (ωn )
 
(3) 0 ··· 0
 H21 (ωn ) H22 (ωn ) ··· 0
It is assumed that that the cross spectra Shj {s, ωn } can be expressed H {ωn } =  .

··· ··· ··· ··· 
in terms of single-point spectra Sh and Sj (with the separation ∆) Hm1 (ωn ) Hm2 (ωn ) ··· Hmm (ωn )
with the corresponding coherence spectra as expressed by Eq. (4):
Isotropy is not a general requirement. However, it is assumed that
Shj {∆, ωn } = Sh (ωn ) · Sj (ωn ) Cohhj (ωn , ∆)eiφij {ωn }
 
(4) the velocity flow components may be described as Gaussian pro-
cesses, an assumption whose accuracy will depend on the environ-
where φhj {ωn } is the phase spectrum, for simplicity, it is assumed
ment in which it is applied (as local topography may sometimes
to be 0 in the current study. i is the imaginary unit. The coherence
spectra for along wind and across wind components are described create turbulence containing large gusts and long lulls of signifi-
in the following sub-sections. cant non-Gaussian nature) [17].
In the current study, the coherence spectrum Coh(∆, ωn ) for Based on the assumption of a Gaussian process, the simulated
both along and across wind components between any two points wind velocity components u, v , and w in space (at point m) at time
(x1 , y1 , z1 ) and (x2 , y2 , z2 ) can be expressed by Eq. (5): t can finally be expressed as Eq. (9):
[ ∑ 0.5 ] − N −1
m − √
3
− U1 · vm (t ) = |Hmk (ωn )| · 21ω
2
h=1 Ah
Coh(ωn , ∆) = e 0
(5)
k=1 n=1
√ −ph
where: Ah = αh · 2ωπn × cos [ωn t + ψmk (ωn ) + θkn ]
 rh q z1 ·z2
· ∆hh 10
. (9)
The coefficients αi ri qi pi and the separations ∆i are illustrated in where 1ω = ω
is the sample density in the frequency domain
N
Table 1.  
(0 ∼ ω); ψmk {ωn } = arctan
Im[Hmk {ωn }]
In the current study, it is assumed that the wind field is Re[Hmk {ωn }]
is the phase angle
unidirectional, stationary and homogeneous. This means that the between two points in space. If Eq. (6) is fulfilled, then ψmk {ωn } =
main flow (along wind) direction is constant in time and space, 0. θkn is a random phase angle uniformly distributed between 0 and
and that the statistical variation of the wind does not change over 2π .
480 J. Jia / Engineering Structures 33 (2011) 477–491

Statfjord Metocean wind profile Standard deviation distribution with height

140 140
Target Target
Simulated Simulated
120 120

100 100
Z (m)

Z (m)
80 80

60 60

40 40

20 20
38 39 40 41 42 43 44 45 46 47 48 4 4.5 5 5.5
U' (m/s) Std

Fig. 2. Comparison of mean wind speed (left) and standard deviation (right) varying with height between the simulated and target value.

In order to optimize the computer storage and the computation 2.3. Wind field simulation and verification
speed, Eq. (9) can be rewritten as an exponential format expressed
in Eq. (10): The time series of wind velocity are simulated in a right hand
N −1 
m −
coordinate system, x, y and z, which is defined with the x-axis
√ 
parallel to the time axis, and the y and z-axis are in the transverse
21ω · e[i(ψmk {ωn }+θkn )] · eiωn t . (10)
−
vm (t ) = |Hmk (ωn )| ·
(towards true south) and vertical directions.
k=1 n=1
A grid is defined which covers the whole structure. For each
node in the grid a time series is generated. This time series
2.2.1. One point spectra for the fluctuating part u of the along wind contains the mean wind speed and the statistical properties of the
field fluctuating part of wind components as defined by the one point
The fluctuating part u of the along wind component is assumed wind spectra and the coherence functions.
to be constant and defined as 1 point spectra S (ωn ) (m2 s−2 /Hz) In the current study, the frequency range ( 2ωπ ) for generating the
expressed in Eq. (11) [24]: wind fields is chosen between 0.0001 and 4.55 Hz, which normally
2   covers the most important eigenfrequencies of a slender structure.
z 0.45

U0
320 · · 10 In flat terrain the wind turbulence tends to be more perfectly
10
S (ωn ) =   3n5 (11) Gaussian than in complex terrain. The living quarters and other
structures on the platform may change the wind field that acts on
1 + f˜ n
the flare boom. However, since it is normally more conservative to
where n = 0.468 calculate the structure in free flow than accounting for the blocking
 ω   z 0.667  U −0.75 and speed-up effects of the nearby structures, these effects are not
n 0
f˜ = 172 · · · taken into consideration. In addition, the living quarters and other
2π 10 10 structures may make the wind non-Gaussian, which may influence
and where z is the height above the sea surface and U0 is the one the fatigue damage accumulation.
hour mean wind speed at 10 m above the sea surface. Since the non-physical cross correlation may be introduced
between wind components by decomposing the wind speeds into
2.2.2. One point spectra for the fluctuating parts v and w of the cross the new coordinate system, different directions of wind should
wind fields be addressed by a finite element modelling program that uses
For structures which are not vertical such as oblique flare generated wind fields with the along wind from a uni-direction as
booms, the cross wind turbulence may contribute significantly input. Therefore, the different directions of wind is preferred to be
to the total structural response. While a metocean report may simulated by rotating the finite element model instead of rotating
only provide relevant data for the along wind components, the the wind fields.
spectra for fluctuating wind components in y–z plane (across According to the relevant metocean report (for simulating along
wind components: v and w ) can then be specified according to wind components) and NS 3491 [25] (for simulating the across
NS3491 [25] as expressed in Eq. (12): wind components) corresponding to the sea area where the flare
boom is installed, the wind field is generated. By investigating
σε2
 
Aε · fε Fig. 2 that shows the comparison of mean wind speed and standard
S (ωn ) = · (12)
(1 + 1.5Aε · fε )1.667
ωn /2π deviations between the simulation and target value varying with
height, it can be observed that the simulated mean wind speed
where, σ is the standard deviation. ε is the symbolic index indi-
agrees rather well with the target mean wind speed. While
cating the direction in y–z plane (horizontal across wind direction
some differences between the simulated standard deviation and
( ωn )·x Lε
or vertical across wind direction). And fε = 2π ε where x Lε the target one can be clearly identified, this is mainly due to
is the integral scale length of the relevant turbulence component, the random phase angles used in the simulation, block division,
which may be determined from ESDU 86010 [26]. It is assumed detrending and Hanning windowing [28] used in the verification.
that σ = 3.5 Aε = 9.4 if there is no relevant wind recording data. At the height of 130 m, the standard deviation of the simulation
Relevant descriptions for generating the mean wind speed, the varied from 4.06 to 4.28 m/s. While the target value for it is
fluctuating parts of the wind components, the coherence function, 4.21 m/s. With the peak factor of 3.5, the difference between
and the spatial representation of the flow field can also be found in the highest and lowest wind speed is 28.4 m/s with the target
Refs. [17,27]. difference of 29.5 m/s, this gives the differences of only 1.1 m/s
 J. Jia / Engineering Structures 33 (2011) 477–491 481

Power Spectra - Series 1

obstruction, flow-interference (i.e., the fluctuating flow effects
Target from the wake of upstream body) and motion-induced forces due
Simulated to the interaction between the flow and structural motion. In the
current study, the calculation of the wind loading is mainly focused
on the buffeting effects. Each structural member is also checked
against the criteria of vortex shedding.
fs(f)

The aerodynamic forces applied on a structural component per

unit length are calculated by Eq. (13)
1 1
F = ρ · CL · vr · |vr | · A + ρ · CD · vr |vr | · d (13)
2 2
where ρ is the density of the air.
CL is the lift coefficient.
100 CD is the drag coefficient.
vr is the wind velocity relative to the member normal to the mem-
10-2 10-1 100 101 ber axis.
f (Hz)
Power Spectra - Series 2 A is the projected area on a plane normal to the oncoming wind
direction.
Target
Simulated d is the diameter of the member exposed to the air.
Note that for tubular members, lift force does not exist. The first
item of the right hand side in Eq. (13) then becomes zero. Thus
the forces on a tubular member only include the viscous effects
as expressed in the second item (drag force) 12 ρ · CD · vr |vr | · d of
Eq. (13). From this item, it also indicates a nonlinear relationship
between the forces and the wind velocity. Note that due to the
deformation or rotation of structures, the drag and lift forces may
fs(f)

100
also change during time.
In the current study, in each time step of the dynamic response
calculation, the wind induced drag forces on each structural
member were recalculated according to the updated structural
deflection.
-2
10 10-1 100 101
f (Hz)
3. Description of the HHT-α method for time integration used
Root Coherence between Series 1 and Series 2 in the nonlinear finite element program
Target
Simulated Numerically, the dynamic analysis for structures with large
0.8 degrees of freedom is mostly carried out in two steps, the first
step is to discretize the equations in space by the finite element
0.6 method, and the second step is to use a decent time stepping
method to integrate the equations in time. A system of second-
order differential equations can be used to express the dynamic
0.4
equilibrium. This system of equations is expressed here in matrix
form as Eq. (14):
0.2
[M ]{Ü } + [C ]{U̇ } + [K ]{U } = {R(t )} (14)
0 where:
d d d2
-0.2 {U̇ } = {U } and {Ü } = {U̇ } = {U }.
0 0.5 1 1.5 2 2.5 dt dt dt 2
f (Hz)
In Eq. (13), {U } is a matrix of displacements; {R(t )} is a matrix of
external forces applied on the system; [M ] is the mass matrix; [C ]
Fig. 3. Verification of wind simulation in the frequency domain between the two
points (series) in space with the separation of ∆ = 36.6 m. is the damping matrix; and [K ] is the stiffness matrix.
The second order differential equations of motions can be
and is regarded as to be acceptable from engineers’ judgement. The solved using (mainly) one of the two numerical strategies, one is
verification of the simulated wind fields can also be proved in the the direct integration method, and the other is modal superposi-
frequency domain as shown in Fig. 3. It illustrates the target and the tion method. The main difference between these two strategies is
simulated values of both wind power spectra and the coherence that in the modal superposition method, a transformation is always
function between two points (series). The separation ∆ between performed prior to the numerical integration.
the two points is 36.6 m. The simulated ones and the target ones The direct integration method uses a numerical step-by-step
also agree very well. procedure for the integration of the governing equation (14). The
most commonly used methods are the central difference method,
2.4. Wind force applied on structural members the Houbolt method, the Wilson-θ method, and the Newmark
method [29]. In the computational program USFOS, the HHT-α
The effects of fluctuating wind loading can be categorized as method is adopted to solve the second order differential equations.
buffeting due to the oncoming flow, vortex shedding due to flow This method is a one parameter multi-step implicit method and it
482 J. Jia / Engineering Structures 33 (2011) 477–491

employs time averaging of the damping, stiffness and load term
expressed by the α -parameter [23]. The advantage of this method
is that it introduces the numerical damping of higher frequency
modes without degrading the accuracy. The governing equilibrium
equation can then be expressed as Eq. (15) with the assumptions
defined by Eqs. (16) and (17):
[M ]{Ü }n+1 + (1 + α)[C ]{U̇ }n+1 − α [C ] U̇
 
n D
+ (1 + α)[K ]{U }n+1 − α[K ]{U }n
= (1 + α){R}n+1 − α{R}n (15)
where
γ tn2
{U }n+1 = {U }n + 1t {U̇ }n + (1 − 2β)Ün + 1t 2 β Ün+1 (16)
2 4.
{U̇ }n+1 = U̇ n + 1t γ Ü n + 1t γ Ü n+1
     
(17)
where n is the order number of the time step; 1t = tn+1 − tn , is
the time increment; β and γ are the coefficients that, together with
α , determine the integration accuracy and stability. The coefficient
β denotes the variation in the acceleration during the time-
incremental step. In the original Newmark-β method (α = 0), γ
is set equal to 0.5 to avoid artificial damping. Different values of β
indicate different schemes of interpolation of the acceleration over
each time-step. For example, β = 0 indicates a scheme equivalent
to the central difference method; and β = 1/6 corresponds to the
linear acceleration method. The linear acceleration approach can
also be obtained by setting θ = 1 in the Wilson-θ method, see
Ref. [29]. In the HHT-α method, unconditional stability is achieved
by satisfying the following conditions as expressed in Eqs. (18)–
(20). The default α value adopted in USFOS is 0.
1 the time increment in the calculation should not be more than
− <α<0 (18)
3 one eighth of the fundamental eigenperiod and in no case more
1 than one eighth of the peak period of the wind field.
β= (1 − α)2 (19)
4 members.
1 11.
γ = (1 − 2α) . (20)
2
For computational reasons, the damping matrix is expressed in
terms of a Caughey series as written in Eq. (21)
− k
αk [M ] [M ]−1 [K ] .

[C ] = (21)
k ply it with the corresponding probability.
The weight factors αk are calculated from modal damping of
structures [23]. This expansion reduces to the Rayleigh-damping
form when the series are truncated after the two first terms. Note
that if k is larger than 2, the damping matrix is likely to be a
full matrix, which requires more computation cost. To limit the
bandwidth of the stiffness matrix, Rayleigh damping is adopted
in the current study. The Rayleigh damping factors are calculated
by the solution of a pair of equations with the damping ratio
associated with two specific modes defined by the user.
Compared to the explicit method, the current implicit HHT-α
integration method is numerically stable for any 1t. The time
step selection is then only based on the accuracy considerations.
However, the calculation for each time step is more costly for the
implicit method than that of the explicit method, and that the
number of operations in the direct integration method is directly
proportional to the number of time-steps in the analysis. Hence, 1t
should be chosen as large as possible provided the desired accuracy
is achieved.
Both the geometric and load nonlinearities are taken into ac-
count in the FE code.
4. Analysis methodology
In order to carry out the fatigue study of the target flare boom
structure, based on the discussions above, the analysis procedure
is proposed as follows:
1. Check all members using the following criteria to avoid vortex
induced vibration (VIV) [30]:
L √
< 22.7 and < 50 · D
t D
where D, t and L are diameter, thickness and length of each
structure member.
2. Establish the structure (FE) model.
3. Perform modal analysis to identify the dynamic characteristics,
verify the structure model and guide dynamic analysis set ups.
Divide the scatter diagram of winds into reprehensible blocks:
m types of directions with n numbers of speeds according to
the relevant metocean design report at the location where the
structure is installed.
5. Calculate the probability of each block that corresponds to the
mean annual directional frequency distribution of wind speed.
6. Create m directions of structure models by rotating the original
structure model anti-clockwise every (360/m)°.
7. Generate the dynamic time history of the wind fields with n
different mean wind speed, both the along wind coming from
a uni-direction and the horizontal and vertical across wind
should be generated.
8. Include the aerodynamic drag coefficient and structural damp-
ing in the structure model.
9. Calculate the time history of forces: start from the first di-
rection of the structure model and first speed block of wind
field and step one by one to the final direction of the struc-
tural model and the final speed block of the wind field. In order
to include sufficient sample data points for fatigue calculation,
10. Input the Stress Concentration Factors (SCFs) of the structure
Choose a decent S–N curve.
12. Calculate the stress history with a duration of t hours, number
the cycles and stress range for each cycle. In order to exclude
the influence of the stress history from the initial loading, the
first few time step histories should be skipped in the stress cy-
cle counting.
13. Calculate the partial fatigue damage for each block and multi-
14. Accumulate the partial fatigue damage of all blocks, extend the
fatigue damage from t hours to one year.
15. The fatigue life can then be calculated as:
Fatigue life (years) = 1/(maximum 1-year fatigue damage).
The computer file structure to calculate the fatigue damage is
shown in Fig. 4.
5. Basis for analysis
The study will be performed in accordance with Norsok N-
001 [31], Norsok N-003 [24] and Norsok N-004 [32].
5.1. Material data
See Table 2 for an overview of the material properties that
is used for the modelling. In order to reach the target structural
weight and center of gravity of each part, the density of the
material for each part of the structure is slightly increased. For
the purpose of calculating high cycle fatigue, the strain hardening
of the material is neglected by assuming a linear elastic material
property.
 J. Jia / Engineering Structures 33 (2011) 477–491 483

0° (N)
6
60°
4 Probability of
120°
2 occurrence
180°(S) (%)
Direction 0
240°
15-20
300°
Speed
0-5 range (m/s)

Fig. 5. The mean annual directional frequency (%) distribution of wind for wind
speed within given classes.

Mean wind speed distribution varying with height

140

Height (m) from the sea surface

120

100

80

Fig. 4. Methodology flowchart for calculating the wind induced fatigue of tubular 60
structures. 40

Table 2 20
Material properties. 0
Constant 5 15 25 35 45
Mean wind speed (m/s)
Yield stress (N/mm2 ) 355
Young’s Modulus, E (N/mm2 ) 210 000 5m/s 10m/s 15m/s 20m/s
Basic density, ρ (kg/m3 ) 7850
25m/s 30m/s
Poisson’s ratio 0.3

5.2. Structural condition A typical wind profile when the wind velocity at 10 m
above sea surface is 30 m/s
140
Regular inspection has been carried out. The available inspec-
tion reports document the presence of some corrosion on the main 120
Height (m) from the sea surface

legs, leg joints and some secondary structure members such as 100
staircases. In the current fatigue calculation, corrosion is not taken
into account. 80

60
5.3. Wind data and full directional scatter diagram
40

Based on the wind data collected in the relevant sea area 20

during the period lasting for 20 years, Fig. 5 shows the mean
annual directional frequency distribution of wind for wind speed 0
29 31 33 35 37 39 41
within given classes. The distributions are based on the composite
Mean Speed (m/s)
database of the wind in the relevant sea area.
For all seasons of winds, the South–West–North directions have
Mean wind speed Mean wind speed + turbulence part
the highest probability of occurrence. The wind from the East has
the minimum frequency of occurrence. It is also noticed that the
high reference wind speed (higher than 15 m/s) from the south has
Fig. 6. The 1 h mean wind speed distribution (upper) varying with height for
a higher probability of occurrence than other directions of wind. each wind speed ranging from 5 m/s to 30 m/s and an example of wind speed
A total of 72 wind-blocks are defined based on 12 headings (30° distribution including its turbulence part (lower) at time t = 0 s.
spacing) and 6 mean wind speed blocks in the range of 0–30 m/s
(increment of 5 m/s). The upper bound of the wind speed for 5.4. Drag coefficients
each block is used. The mean wind speed distribution varying with
height for each wind speed range and an example of the wind The drag-coefficients are assigned according to the specifica-
speed including its turbulence part are shown in Fig. 6. With the tions of DnV [33], see Table 3 and Fig. 7 for details. In the cur-
increase of the reference wind speed, the increment of the mean rent study, all tubular members on the flare boom are regarded as
wind speed varying with the height becomes significant. smooth tubular members.
484 J. Jia / Engineering Structures 33 (2011) 477–491

5.8. Structural modelling

c
The model represents the complete flare boom including flare-
and vent-lines as shown in Fig. 8. Flare- and vent-line supports
k/D = 1× 10-2
1.0 are modelled with releases to allow longitudinal movement. The
projected longitudinal axis on the horizontal plane towards 23°
1× 10-3 clock wisely of the east as shown in Fig. 9.
1× 10-4
Ladders are modelled as pipes to include the contributions of
1× 10-5 wind area, weight and bending stiffness. The drag coefficient in
0.5 the radius direction of the pipes is varied to represent the varied
wind area towards each direction. Access platforms and staircases
are modelled as lumped masses to represent their weight contri-
butions.
The height of the flare boom extends from around 29 m above
0 Lowest Astronomical Tide (LAT) and upwards.
104 105 106 Re 107 The supports for tie backs and wind-struts, the connections
Re = 500,000
Re = 240,000 Re = 370,000 between wind-struts and main flare boom, and the connections
between tie backs and main flare boom are modelled as free to
Fig. 7. Drag coefficient categories of fixed circular cylinder for steady flow in
rotate around the axis perpendicular to the longitudinal vertical
a critical flow regime, for various roughness (Re: Reynolds number, Dc: Drag
coefficient). planes of each part of structures. All other degrees of freedom are
fixed on these supports.
Table 3 Based on the geometry model, a finite element model is created,
The drag coefficient specification. which consists of 1950 beam elements with 883 nodes in total.
Reynolds number (Re) Cd
To avoid the non-physical cross correlation introduced between
wind components by rotating the wind direction, only wind fields
Re < 3.7 · 105 1.2
coming from the west are generated. The finite element structure
3.7 · 105 < Re < 5.0 · 105 1.0 model is then rotated anti-clockwise 12 times from its original
Re > 5.0 · 105 0.6 location at a 30 degrees’ interval.

5.5. Design target fatigue lives 6. Dynamic analysis

The time domain simulation is performed in the computational

The design fatigue life is 50 years. The design fatigue factor
requirements are in accordance with Norsok N-001 [31] (Table 4)
and to be taken as 2. This is due to the reason that the failure of
some components among flare boom, wind struts and tie backs
may have significant consequences on other parts of the structure
and may even lead to the failure of the total structure. The resulting
target design fatigue life is then set up as 100 years.
5.6. Stress Concentration Factors (SCFs)
SCFs for tubular Joints are calculated based on Efthymiou
equations. The minimum SCFs for axial, in plane bending, and out
of plane bending are set as 1.5 for all tubular joints.
The SCFs for conical joints are calculated according to DnV-RP-
C203 [34].
5.7. S–N curve
The S–N curve for ‘tubulars in air’ (T -curve) is used for all
tubular joints according to DnV-RP-C203 [34]. The mean stress
effects are not taken into account due to the existing residual
stress in the welded joints. However, if a structure is constructed
with mean stress sensitive material such as composites, its effects
should be considered.
The maximum wall-thickness for the flare boom is 25 mm, i.e.
no thickness correction is required for the S–N curve.
program USFOS [23]. The structural response for each wind block
is calculated. The time increment for the time integration in USFOS
are defined as 0.1 s. This gives about 8 calculations per eigenperiod
and is less than 1/100 of the wind peak period. Based on a series of
sensitivity analysis, for each wind block, a half hour time duration
is adopted to calculate dynamic response. This is believed to give
a sufficiently accurate dynamic response of the structure. The
Rayleigh damping of the structure is set up by providing a relative
damping of 1% for the modes at eigenfrequency of 0.1 Hz and 10 Hz,
respectively.
7. Fatigue calculation basis
7.1. Stress range and cycle counting
The stresses include the SCFs at 8 stress points around the tube.
Both the fatigue damage in the chord side and the brace side is
calculated. The number of cycles and stress ranges are calculated
using ‘rainflow’ counting [35].
7.2. Fatigue damage
A half hour part fatigue damage is calculated with the Miner–
Palmgren method [35]. The fatigue damage of 72 wind load
cases/blocks are multiplied individually by the probability for each
wind state and accumulated to one year fatigue damage.

Table 4
Design fatigue factors.
Classification of structural components based on Not accessible for inspection and repair or in the Accessible for inspection, change or repair and
damage consequence splash zone where inspection or change is assumed

Substantial consequences 10 2
Without substantial consequences 3 1
 J. Jia / Engineering Structures 33 (2011) 477–491 485

Fig. 8. The structural geometry model of the flare boom.

Fig. 9. Orientation of the flare boom structure (23° south of true east).

8. Modal analysis
Modal analysis has been performed to verify the analysis model
and to assess the areas where high response is foreseen. The
results correspond well with data previously reported in the
relevant design report. See Table 5 and Fig. 10 for the first 10
eigenfrequencies and the corresponding mode shapes. Note that
the fundamental and 2nd eigenmode only show the flare top
flexural vibration without significant participation of the lower
part of the flare boom, wind struts, and tie backs, an almost quasi
static response of the lower part of the flare boom is then expected
in the free vibrations of the structure (see Fig. 10).
9. Results and analysis
The dynamic analysis is based on the reference input parameter
set ups as shown in Table 6.
The fatigue lives for all connections have been calculated.
Generally, the fatigue life in most of the joints is more than
500 years.
486 J. Jia / Engineering Structures 33 (2011) 477–491

Fig. 10. The selected flare boom eigenfrequencies and mode shapes.

Table 5
The flare boom eigenfrequencies and mode shapes.
Mode Eigen-frequency (Hz) Remarks

1 1.237 1st vertical flexural eigenmode at the top of the main flare boom
2 1.303 1st sway flexural eigenmode at the top of the main flare boom
3 1.474 Local vibration of the staircase
4 1.498 Local vibration of the staircase
5 1.676 The sway flexural vibration of both wind struts and tie backs combined with the local vibration of the staircase
6 1.684 The significant sway flexural vibration of both wind struts and tie backs without the local staircase vibrations
7 1.723 Vertical flexural vibration of the wind struts
8 1.7626 The second sway flexural vibration of wind struts (anti-symmetric) and tie backs combined with the local vibration of the staircase
9 1.7629 The sway flexural vibration of tie backs and torsion vibration of the wind struts
10 1.821 The significant sway flexural vibration of both wind struts and tie backs (symmetric)

10. Investigation of wind direction effects
Since the directional effects of wind on the structural response
may give engineers a better understanding of the structural
response with regard to wind resistance, life time extension and
maintenance, and structures’ reorientation, the current paper also
investigates the contribution of the wind from each individual
direction on the fatigue damage of the structure. Six joints (Fig. 11)
on the structure have been selected for the illustration of fatigue
damage.
 J. Jia / Engineering Structures 33 (2011) 477–491 487

Fig. 11. Selected typical joints for fatigue life illustration in Fig. 12 and Table 7.

Table 6 the most significant contribution on the fatigue damage, and

Reference input parameter set ups for the dynamic analysis. the winds coming from the west (270°), northwest (300°), east
Wind grid size y × z 12.15 m × 7.3 m (90°), northeast (60°), and east–southeast (120°) have the least
(reference) significant influence on the fatigue damage. This is mainly due
Wind components Both along and across wind components
to the reason the high wind speed for wind coming from the
(reference)
Time duration simulation Half hour
south cherishes a higher probability of occurrence than that from
for each wind load other directions (Fig. 5), in addition, the directional structural
case/block effect (wind perpendicular to the flare boom induces higher
Time step increment 0.1 s (1/8 of the fundamental eigenperiod) dynamic global forces than wind parallel to the flare boom)
Connection flare Hinged
also contributes to it. The trends described above are even
boom–wind-struts
Drag coefficients From DNV-RP-C205 more obvious when taking the wind directional probability of
occurrence into consideration. This is because that the probability
of occurrence of the south wind is much higher than that from
Two cases are analyzed in the current study: other directions, and the probability of occurrence of the winds
(1) Without considering the wind directional probability: only from west, northwest, east and northeast is rather low.
considering the probability of occurrence for each wind speed Furthermore, by only taking the high wind speed with the
block in each individual direction, and then assuming that reference speed of 25 m/s into account, and assuming that the
wind from each direction cherishes the same probability of wind with this speed from each individual direction has the same
occurrence, this is a special case of Fig. 5, with only a 2 probability of occurrence as 100%, the one year fatigue damage
dimensional graph (i.e. probability of occurrence–wind speed for the 6 selected joints is calculated as shown in Fig. 13. Due to
range). the directional structural effect, it is found that for most of the
joints, the winds from the directions approximately perpendicular
(2) Considering the wind directional probability: take the proba-
to the flare boom orientation direction are most effective to excite
bility of occurrence for each wind speed block and each direc-
the vibration of the structure and accumulate the fatigue damage.
tion into account, this is illustrated in Fig. 5.
While for the joint 524 located close to the tie back main flare
The one year fatigue damages for both cases are shown in boom connection, the wind parallel to the flare boom orientation
Fig. 12. It is found out that, when the wind directional probability (significantly induces the forces along the flare boom orientation)
of occurrence is not taken into account, generally, the winds therefore has the most significant contribution on the fatigue
coming from the south (180°) and south–southeast (150°) have damage of these two joints.
488 J. Jia / Engineering Structures 33 (2011) 477–491
Fig. 12. One year fatigue damage under each separate direction of wind for the selected joints.
11. Investigation of across wind and wind grid size effects
Due to the difficulties of obtaining the necessary data from
relevant reports, it is quite usual to ignore the contribution of
the across wind components in the wind induced fatigue analysis.
While it is noticed that, due to the relationship between stress
range and fatigue damage, the fatigue damage is more sensitive
to the wind loads than that of the stress amplitude, therefore,
it is interesting to investigate how significant the across wind
components are on the fatigue damage of the structure. For a
structure that is not vertical like the current oblique flare boom,
the across wind components may be significant in determining the
wind induced fatigue life of the structure. In addition, the variation
of wind grid size for generating the wind may also significantly
influence the wind load distribution on a structure and are then
investigated in the current study.
The effects of the across wind components and different wind
grid size setups are illustrated in Table 7.
It can be observed that the fatigue life without including the
across wind components is much higher than that with considering
the across wind contribution. Generally, the vertical across wind
components has a slightly more significant contribution on the
fatigue damage of the structure than that of the horizontal across
wind components.
By observing the Table 7 it is also found out that, generally, the
larger the wind grid results in a lower fatigue life. This is because
the larger wind grid size gives more correlations (apply a more
correlated wind on the grid dimension) in the wind action and
therefore leads to higher dynamic response amplitude. This trend
becomes more significant at higher locations of the flare boom.
Note that the fatigue damage is proportional to the third power
of the stress amplitude, the stress amplitude difference between
different wind grid size set ups is not as significant as the fatigue
life difference is.
It should be mentioned that Aas-Jakobsen and Strømmen [17]
present a criteria for selecting the element length in the structure
model as expressed in Eq. (22). By realizing the fact that the wind
grid size and element length are interchangeable quantities, this
criteria may provide practical recommendations for selecting a
decent wind grid size. Further studies in this respect are currently
carried out by the author.
min{Ltot /3, 2Ψ } : symmetric modes

1l ≤ (22)
min{Ltot /3, Ψ } : anti-metric modes
where 1l is the element length, Ltot is the total length of the
structure (or the active mode) and Ψ is the span-wise integral of
the coherence function at a dominating eigenfrequency [17].
12. Conclusions and future studies
The current paper presents a practical approach for calculating
the wind induced fatigue of structures based on nonlinear dynamic
analysis. A realization of spatially correlated wind fields was
generated and used to compute the dynamic response that takes
geometric and load nonlinearities into considerations. High cycle
fatigue has been calculated by applying the ‘rain-flow’ counting
method and the Miner rule. The fatigue damage of a structure
is much more sensitive than the stress response to the variation
of the load and structure modelling. By presenting a practical
approach for calculating wind induced fatigue based on the joint
probability with regard to wind speed and its main direction, the
current paper assesses the effects of the wind direction, across
wind and wind grid size on the high cycle fatigue of the structure.
It is found that, the directional wind effects on the fatigue damage
depends on the orientation of the structure, the location and the
support condition of the selected joints, and the relative probability
of occurrence for the high wind speed in each direction (the wind
from directions of which the high wind speeds cherish a high
probability of occurrence would have the most significant effects
on the fatigue damage of structures). Furthermore, the cross wind
components are a significant contributor on the fatigue damage
 J. Jia / Engineering Structures 33 (2011) 477–491 489

Fig. 13. One year fatigue damage under the high reference wind speed of 25 m/s, the probability of occurrence of 100% is assumed for each individual direction of wind
with the speed of 25 m/s.

and cannot be neglected. The fatigue damage is also found to be
rather sensitive to the wind grid size for generating the wind
fields. The wind fatigue calculation procedure presented in the
current paper can be widely adopted on a similar study on tubular
structures.
In the current study, it is found out that the winds from
directions (south and south east) perpendicular to the longitudinal
axis of the flare boom contribute most to the fatigue damage. From
the structural response point of view, this is mainly due to the
reason that wind perpendicular to the flare boom induces higher
global forces than wind parallel to the flare boom. Since in the
Norwegian Continental Shelf of North Sea, most of the flare booms
face approximately east, and the winds from the south generally
cherish a high probability of occurrence, the wind in this area may
be more effective to accumulate fatigue damage.
For the current structure, the vertical across wind component
has a slightly more significant contribution on the fatigue damage
of the structure than that of the horizontal cross wind components.
In the current study, the wind blocks (mean wind speed and
the corresponding probability of occurrence) are extracted from
the relevant metocean report with evenly spaced wind speed
between each two adjacent blocks. However, since the winds
cherishing the high wind speed or high probability of occurrence
normally contribute more to the fatigue damage, it is then highly
recommended to select decent blocks with closely spaced wind
velocities at those wind speed ranges.
It is also recommended that, based on the parameters such
as the size for a structure component and global dimension, the
dynamic characteristics of the structure, and calculation accuracy
requirement, etc., a decent wind grid size can be decided.
490 J. Jia / Engineering Structures 33 (2011) 477–491

Table 7
The fatigue life influenced by the across wind components and wind grid size.
Joint ID Brace ID Position description Fatigue life (years) with (half hour duration of simulation for each wind load case)
Wind grid (y × z ):12.15 m × 7.3 m Wind grid Wind grid
(y × z ) : (y × z ) :
17.56 × 10.56 m 26.3 × 15.8 m
Without across With horizontal With vertical With across With across With across
wind components of components of wind + along wind + along wind + along
components the across wind the across wind wind wind wind
+ along wind + along wind components components components
components components

109 246 A lower leg joint at 3 015 1 386 1 082 734 590 288
35 m above the sea
level
407 911 A joint at the south 2 518 1 711 1 309 1 009 833 367
wind strut at 62 m
above the sea level
524 1251 A north leg joint at 264 550 62 500 62 150 22 129 18 957 9302
70 m above the sea
level
540 1289 A joint at the top of 45 893 10 396 16 628 5 459 4 354 2007
the north tie back.
762 1755 A north leg joint on 24 462 8 953 7 128 3 289 2 979 1361
the flare boom at
96 m above the sea
level
850 1882 A joint on the flare 25 221 13 031 6 361 3 758 2 997 1329
boom close to the
flare tip at 111 m
above the sea level.

It is very usual to exclude the secondary structures such as
flare-vent-lines and staircases in the FE modelling. While from
the current study, they appear to have a certain influence on the
fatigue damage through the influence on the global or local flare
boom stiffness, reaction forces on the supports of the flare and
vent lines, the attraction of wind loads, the weight distribution of
the flare boom boom-secondary structural system, as well as the
strengthening of frames. Thus, it must be carefully decided if these
secondary structures should be modelled or not and how detailed
should the modelling be.
Even though in the direct time integration calculations, the
eigenanalysis are not required to perform, in order to identify
structural characteristics for verification and essential factors
contributing to the dynamic structural responses, a modal analysis
is highly recommended. The eigenfrequency analysis can also
provide suggestions on how large the time increment step length
should be in the time integration calculation.
Since non-physical cross correlation may be introduced be-
tween wind components by decomposing the wind speeds into
the new coordinate system, different directions of wind should be
addressed by a finite element structural program. Therefore, the
different main direction of wind is preferred to be simulated by ro-
tating the finite element model instead of rotating the wind com-
ponents’ main direction during the generation of the wind field.
The possible non-Gaussian structural response is mainly due
to the nonlinearity of a structure and the non-Gaussian wind
loads. The current study includes the former effects by carrying
out nonlinear finite element calculations. The nonlinearity induced
from drag loads is also included. However, the wind field
generation is based on the assumption of a Gaussian process. The
load non-normality has a great effect on both the cycle distribution
and the fatigue damage [36]. Several pieces of research work
[14,37,38] have shown that a significantly increased rate of fatigue
damage accumulation may be expected when the wind load is non-
Gaussian, this is especially the case on corner zones as well as
on other separated flow regions [39]. Thus it is worthy to study
the difference of fatigue damage under the wind fields generated
from Gaussian and non-Gaussian process. It is recommended to
adopt decent stress cycle accounting methods such as the rain flow
counting method, which does not presume the stress process as
narrow banded.
Attention should also be paid on the design fatigue factor.
It is normally presumed that all the joints on a flare boom are
accessible for inspection. However, in reality, due to the equipment
installation, harsh environment (e.g. high temperature nears the
flare tip, etc.), production restrictions, some joints may not be
accessible for inspection for a part of or even the entire life of a
structure. Therefore, it is suggested that the critical joints under
a fatigue design factor that corresponds to ‘not accessible for
inspection’ may also be listed for a better inspection plan.
In addition, the sensitivity analysis of the fatigue damage due
to the scatter effects of the environment data, material data
(elastic modulus, S–N curves), and manufacturing tolerance are
recommended to be carried out to obtain a reliability based fatigue
life calculation, which can be used for inspection purposes. The
SCFs must be very carefully assigned since the fatigue life is much
more sensitive to even a slight change of the SCFs than that of the
stress history is. The calculation method presented in the current
paper has been applied on quite a few industry and research
projects for calculating wind and/or sea wave induced fatigue for
offshore structures and wind turbine support structures.
Acknowledgements
The author would like to thank Mr. Rune Ellefsen, Mr. Gunnar
Bremer, Mr. Zhibin Jia, Dr. Tore Holmås and Mr. Hesam Soltanpour-
Namin at Aker Solutions and Dr. Ketil Aas-Jakobsen at Dr. Ing. A.
Aas-Jakobsen AS for their cooperation on the research.
References
[1] Gomathinayagam S, Vendhan CP, Shanmugasundaram J. Dynamic effects of
wind loads on offshore deck structures—a critical evaluation of provisions and
practices. J Wind Eng Ind Aerodyn 2000;84(3):345–67.
[2] Kareem A, Smith CE. Performance of offshore platforms in hurricane Andrew.
In: Proceedings of the seminar on hurricanes of 1992 held in December 1993.
1993. p. C3 I-1-10.
[3] Deoliya R, Datta TK. Reliability analysis of a microwave tower for fluctuating
mean wind with directional effects. Reliab Eng Syst Saf 2000;67(3):257–67.
[4] Repetto MP. Cycle counting methods for bi-modal stationary Gaussian
processes. Probab Eng Mech 2005;20(3):229–38.
[5] Repetto MP, Solari G. Bimodal alongwind fatigue of structures. J Struct Eng,
ASCE 2006;132(6):899–908.
 J. Jia / Engineering Structures 33 (2011) 477–491
[6] Panofsky HA, Dutton JA. Atmospheric turbulence: models and methods for
engineering applications. New York: Wiley; 1984.
[7] Repetto MP, Solari G. Wind-induced fatigue of structures under neutral
and non-neutral atmospheric conditions. J Wind Eng Ind Aerodyn 2007;95:
1364–83.
[8] Palmeri A, Ricciardelli F. Fatigue analyses of buildings with viscoelastic
dampers. J Wind Eng Ind Aerodyn 2006;94(5):377–95.
[9] Davenport AG. The prediction of risk under wind loading. In: Proceedings of
the second international conference in structural safety and reliability. 1977.
p. 511–38.
[10] Simiu E, Filliben JJ. Wind directional effect on cladding and structural loads.
J Eng Struct 1981;3:181–6.
[11] Wen YK. Wind direction and structural reliability. J Struct Eng, ASCE 1983;
109(4):1028–41.
[12] Gu M, Xu YL, Chen LZ, Xiang HF. Fatigue life estimation of steel girder of Yangpu
cable-stayed bridge due to buffeting. J Wind Eng Ind Aerodyn 1999;80(3).
[13] Sarkani S, Kihl DP, Beach JE. Fatigue of welded joints under narrowband non-
Gaussian loadings. Probab Eng Mech 1994;9:179–90.
[14] Ko N-H. Verification of correction factors for non-Gaussian effect on
fatigue damage on the side face of tall buildings. Int J Fatigue 2007;
doi:10.1016/j.ijfatigue.2007.07.006.
[15] Lutes LD, Corazao M, Hu SJ, Zimmerman J. Stochastic fatigue damage
accumulation. J Struct Eng, ASCE 1984;110(11):2585–601.
[16] Lutes LD, Sarkani S. Stochastic analysis of structural and mechanical vibrations.
Prentice Hall; 1997.
[17] Aas-Jakobsen K, Strømmen E. Time domain buffeting response calculations of
slender structures. J Wind Eng Ind Aerodyn 2001;89(5):341–64.
[18] Aas-Jakobsen K, Strømmen E. Time domain calculations of buffeting response
for wind-sensitive structures. J Wind Eng Ind Aerodyn 2001;74–76(5):687–95.
[19] Santos J, Miyata T, Yamada H. Gust response of a long span bridge by the time
domain approach. In: Proceedings of the third Asia-Pacific symposium on wind
engineering. 1993. p. 211–6.
[20] Gioffr Gioffrè M, Gusella V, Materazzi AL, Venanzi I. Removable guyed mast for
mobile phone networks: wind load modeling and structural response. J Wind
Eng Ind Aerodyn 2004;92:463–75.
491
[21] Repetto MP, Solari G. Dynamic alongwind fatigue of slender vertical structures.
Eng Struct 2001;23(12):1622–33.
[22] Feng Maria Q, Zhang Ruichong. Wind-induced vibration characteristics of
Nanjing TV tower. Internat J Non-Linear Mech 1997;32(4):693–706.
[23] Søreide Tore H, Amdahl Jørgen, Eberg Ernst, Holmås Tore, Hellan Øyvind.
USFOS—a computer program for progressive collapse analysis of steel offshore
structures. Theory manual. Report no. F88038. Trondheim: SINTEF; 1993.
[24] NORSOK standard N-003. Actions and action effects, rev. 2. October 2004.
[25] NS 3491-4:2002. Design of structures. Design actions. Part 4: wind loads.
[26] ESDU 86010. Characteristics of atmospheric turbulence near ground, part II
and part III. London: Engineering and Science Data Unit; October 1985.
[27] Sachs Peter. Wind forces in engineering. 1st ed. Oxford: Pergamon Press; 1972.
[28] Ewins DJ. Modal testing: theory and practice. New York: John Wiley & Sons;
1991.
[29] Bathe KJ. Finite element procedures in engineering analysis. Englewood Cliffs
(NJ): Prentice-Hall, Inc.; 1982.
[30] Sjursen Kurt. Avoidance criteria for wind-induced vortex shedding vibrations.
In: Proceedings of the ninth (1999), international offshore and Pohu’’ engi-
neering conference. 1999.
[31] NORSOK standard N-001. Structural design, rev. 4. February 2004.
[32] NORSOK standard N-004. Design of steel structures, rev. 2. October 2004.
[33] Det Norske Veritas. Recommended practice DNV-RP-C205—environmental
conditions and environmental loads. April 2007.
[34] Det Norske Veritas. Recommended practice DNV-RP-C203—fatigue design of
offshore steel structures. August 2005.
[35] Dowling Norman E. Mechanical behaviour of materials. London: Prentice-Hall;
1999.
[36] Benasciutti D, Tovo R. Fatigue life assessment in non-Gaussian random
loadings. Int J Fatigue 2006;28(7):733–46.
[37] Kuresh Kunmar K, Stathopoulos T. Fatigue analysis of roof cladding under
simulated wind loading. J Wind Eng Ind Aerodyn 1998;77–78(5).
[38] Wintersein SR. Non-linear vibration models for extremes and fatigue. J Eng
Mech, ASCE 1986;114:1772–90.
[39] Stathopoulos T. PDF of wind pressures on low-rise buildings. J Struct Div, ASCE
1980;106(5):973–90.