Vous êtes sur la page 1sur 6

Simulation method for joint exibility in modern oshore support

structures
Kristina Wiemann, Louis Quesnel, Jan Wagener, Holger Huhn
1
, Hans-Gerd Busmann
Fraunhofer Institute for Wind Energy and Energy System Technology (IWES)
Am Seedeich 45, 27572 Bremerhaven, Germany
1
holger.huhn@iwes.fraunhofer.de, Phone +49 471 902 629-21, Fax -19
Abstract
As oshore wind energy reaches deeper waters, use of
branched steel support structures such as jackets or
tripods increases. Stepping up from typical oil-and-gas
single-build designs to serial production requires the use
of reproducible parts. One proposed solution for jackets
is the design of cast joints to connect chords and braces.
For general dimensioning purposes, the static and dy-
namic bearing capacity of such structures is frequently
simulated initially by spatial frameworks with either sti
joints or, more realistically, exible joints. Without the
consideration of the exible joints, eigenfrequencies and
loads are overestimated, resulting in an uneconomical
design. A typical approach for welded joints is to model
the local joint exibility by dimensionless factors which
are integrated in the simulation via oset nodes and ex-
ible elements. These factors are obtained from experi-
ments or nite element calculations and apply only to
the investigated geometries and load cases.
Cast joints feature dynamic properties signicantly dif-
ferent from welded ones. The use of local joint exibility
for those joints has been studied in this work. An Ade-
quate Beam Model was designed with deformations pro-
posed in the literature, focusing on reducing the number
of beams to decrease calculation time. Benets from
ABMs include increased exibility for implementation in
common framework and aero-elastic simulation tools.
The inclusion of wave loads is also possible.
A Best-Fit tool was created to match the exibility
properties of the Adequate Beam Model with a Solid
Element Model. The analysis of the remaining defor-
mations showed signicant mismatch. The adaptation
process for local joint exibilities from welded to cast
joints was then studied more closely, and the dependence
on the number of parameters, geometry and adaptable
properties is discussed.
1 Introduction
Installation of wind energy converters in waters deeper
than 25 to 30 m requires the use of more sophisticated
bottom mounted support structures rather than tradi-
tional monopiles. The dynamic behavior of those struc-
tures must be studied carefully to ensure safe operation
and satisfactory service life and thus secure prots from
wind energy. Among those new support structure con-
cepts, jackets are becoming increasingly common, with
prototypes already installed both at sea and on land.
Joint exibility plays a major role in dynamic simula-
tion of those branched steel support structures [1]. Sev-
eral modelling methods of dierent quality and complex-
ity are available. Those include perfectly sti elements,
ideal truss joints, shell element models (FEM), or the
so-called Adequate Beam Models, which provide a more
realistic dynamic behavior than sti joints while not re-
quiring advanced software necessary to FEM simulation.
The design of Adequate Beam Models for plane welded
joints has been described by Buitrago and Healy in [2].
Cast joints will be increasingly used on support struc-
tures of oshore wind energy converters as they allow for
mass production and economies of scale. However, their
dynamic behavior diers notably from welded ones, so
that specic modelling methods have to be developed.
An eort is made to develop ABMs for double-K cast
joints, inspired by Buitrago and Healys work. First the
denition of the ABM design and the choice of a set
of parameters are presented, followed by the optimiza-
tion strategy and solutions. Results as well as future
prospects are subsequently discussed.
Proceedings of the European Oshore Wind Conference 2009 Wiemann
2 Nomenclature
ABM Adequate Beam Model
D End diameter
d Displacement
DOF Degree Of Freedom
E Youngs Modulus
FEM Finite Element Model
i Number of beam set
j Number of end
LJF Local Joint Flexibility
N Axial force
M
i p
In-plane bending moment
M
op
Out-of-plane bending moment
SEM Solid Element Model
3 Buitragos parametric approach
for welded joints
Modelling tubular joints with conventional rigid-beam
elements leads to overestimation of internal member
forces at the joint, resulting in an uneconomical design.
Therefore it is essential to include realistic joint exibili-
ties in the FE-simulation.
The substructuring technique described in [3] replaces
the joints by user-dened elements. In a rst step a
sub-model of the joint is generated with solid or shell
elements and master DOFs at the joint interfaces are
dened. Then the stiness and mass matrix are sub-
sequently derived by means of statical condensation for
the master DOFs. The matrices are nally implemented
in the beam element model of the support structure as
superelements.
The approach introduced by Buitrago and Healy in [2]
uses parametric formulae to describe the joint exibility.
Therein the Local Joint Flexibility (LJF) is dened as
the deformation of the joint at the footprint. For the
calculation of the LJFs a shell model of the joint is built
up. The chord ends are clamped for the analysis (see
Figure 1).
The brace ends are loaded one by one using single
unit loads and the corresponding deformations are cal-
culated at the brace ends. The deformations are then
substracted by the global beam deformations, and the
net deformation is obtained. The LJFs are nondimen-
sionalized using the member diameter D and the modu-
lus of elasticity E. This calculation has been carried out
for dierent geometries of planar tubular joints. With the
results, parametric formulae have been developed for a
wide range of joint congurations, diameters, angles and
Figure 1: General joint geometry, loads and degrees of
freedom, from [2]
overlap in K-joints [2].
The exibility matrix of a single brace tubular joint
should be given by a 6x6-matrix, since 3 translational
and 3 rotational DOF are present at the brace end. Ac-
cording to Buitrago and Healy the two shear forces and
torsional moment are not relevant for the determina-
tion of the LJFs. Therefore only the normal forces and
in-plane and out-of-plane bending moments have to be
considered.
As a result the matrix is reduced to a 3x3-matrix. For
a joint with two braces, a 6x6-matrix is obtained. The
exibility matrix consists of direct terms coupling the
load and the deformation at the same brace end and
cross terms coupling the load at one brace end and the
deformation at the other brace (see Figure 2).
These cross terms exist for all multi-brace joints, be-
cause the deformation regime of their footprints is in-
uenced by both chord-brace-intersections. This eect
can be considered approximately by the superposition of
dierent LJFs or by direct stress calculations.
In order to implement the LJF into the branched steel
support structure simulation, the LJFs are calculated,
taking into account the type of loading and the geom-
etry of the joints. These LJFs are represented by axial
or rotational spring elements or by short exible beam
elements with the corresponding stinesses. These el-
ements are arranged near the chord surface since the
deformations of the chord are located there. The oset
between the chord-brace axis intersection point and the
spring or ex element may be modeled with sti beam
2
Proceedings of the European Oshore Wind Conference 2009 Wiemann
elements or rigid links (see Figure 3).
Brace1 Brace2
N Mip Mop N Mip Mop
u1 #
rip1 # symmetric
Brace1
rop1 #
u2 x +
rip2 x +
Brace2
rop2 x +
#, +: direct terms, x: cross terms
Figure 2: 6x6-matrix for two-brace joints with direct and
cross terms, from [2]
Figure 3: Joint model with LJF realized as spring ele-
ments, from [4]
4 Adequate Beam Model
The Adequate Beam Model presented in this paper was
inspired by Buitrago and Healys works, however the aim
was to simplify the equivalent model further by reducing
the number of elements and nodes, omitting the exible
element and optimizing the beam properties instead to
achieve a matching dynamic behavior.
4.1 Approach
When designing an ABM intended to simplify dynamic
analysis, it is logical to try and keep the complexity of the
ABM itself as small as possible in order to streamline the
ABM adaptation process as well as reduce calculation
time. This is achieved by keeping the number of beams
as low as possible, while still adequately describing the
main dynamic properties.
The geometry of the ABM is based on the real cast
joint where the beam geometrical properties (length, po-
sition, angle) are taken from the centrelines of the joint
geometry (see Figure 5). The cross-sectional area A
and second order moments of inertia I
yy
and I
zz
of each
beam are the design variables to be optimized. By tak-
ing advantage of symmetries in the joint geometry, the
number of distinct design parameters can be reduced
further. The resulting model reproduces the correct dis-
placements at each of the joint tips, but internal forces
and displacements within the joint cannot be determined.
The correct values of the design parameters must then
be determined to match the elastic behavior of the ABM
with that of the real cast joint. This can be done by xing
the model at one end, applying unit loads successively at
each unxed joint interface and comparing the resulting
displacements at all joint ends. In doing so, the following
assumptions are made:
The joint shows a linear elastic behavior
The interfaces of the joint (tube sections) are per-
fectly sti, i.e. no ovalization of the chord and
braces takes place
Torsion moments and shear forces in the chord and
braces are small and thus negligible, reducing the
number of unit loads to be tested by a half
4.2 Case study
The aforementioned approach is implemented on a
double-K cast joint by comparing the ABM with a Solid
Element Model (SEM) within the nite element simula-
tion software ANSYS [5].
The Solid Element Model consists of 670.000
SOLID185 elements with tetrahedral option (see Figure
4). For the ABM, the chord and the four braces are rep-
resented by single beams. One main feature of this joint
is the oset between the intersection point of the braces
and the chord. It appears thus necessary to include two
extra beams linking the braces to the chord (see Figure
5). The resulting Adequate Beam Model then consists
of seven beams, which using the joint symmetry can be
reduced to four beams.
The joint is xed at the lower end of the chord, and
displacements caused by unit loadings applied at the free
chord and brace ends are compared. The design param-
eters for each beam are then varied and optimized to
match the displacements of the ABM with those of the
SEM.
3
Proceedings of the European Oshore Wind Conference 2009 Wiemann
Figure 4: Solid Element Model of a double-K cast joint
Figure 5: Adequate Beam Model of a double-K cast
joint
5 Optimization
The displacements induced by unit loadings are the
benchmark for quality of the ABM. The optimization
criterion is therefore the relative displacement error
j,i

j,i
=
d
SEM,j
d
ABM,j,i
d
SEM,j
5.1 Grid search
A rst optimization attempt consists in searching for the
set of parameters that provides the best, if not perfect,
match for all DOFs on all nodes of the joint. To this
eect, a basic grid-search is used to test dierent sets:
the parameters are varied stepwise and for each set, the
greatest value of
j,i
is taken as a sign of the overall
quality of the set. This eectively leads to frontal op-
timization on all DOFs simultaneously, and to leveling
of the model delity over all DOFs. All combinations
are exhaustively tested. This allows the problem to be
solved despite non-monotonic behavior, however it im-
plies a lenghty calculation. The best parameter set is
the one with the lowest maximal value of
j,i
.
To reduce solution times a simple break criterion is
devised and described in [6]. As soon as any
j,i
for a
given parameter set exceeds the maximum of the best
set to date, the given set is discarded and the algorithm
steps to the next one.
5.2 Simultaneous parameter optimization
The grid-search method requires high computational
costs despite a relatively big step size. Therefore other
numerical solution algorithms such as the Hooke-Jeeves
Direct search optimization method introduced in [7]
may be more appropriate to enhance the performance of
the adaptation process. The Hooke-Jeeves method can
be described by two steps:
1. The gradient is determined by changing the input
parameters of the function one after another by an
innite step
2. The new parameter set is calculated by extrapola-
tion in the direction of the largest descent
Hence, the approach is similar to the Newton-method
but does not require the analytical knowledge of the par-
tial derivates.
Due to symmetry conditions only one half of the ABM
(see Figure 5) including the chord, the two braces and
the oset beam needs to be considered. As there are
three sectional properties per beam type there exist a to-
tal number of 12 design variables (see Figure 6). Since
the parameters of the chord can be investigated sep-
arately, the number of variables may be reduced to 9.
Only direct terms according to Buitrago and Healy are
considered.
The optimization consists of the following steps:
1. Calculating the deformations d
SEM,j
of the Solid
Element Model by loading the brace ends one by
one with the axial force N, in plane moment M
i p
and out of plane moment M
op
2. Initialization with start values for ABM properties
3. Begin iteration
4
Proceedings of the European Oshore Wind Conference 2009 Wiemann
a) Updating the model with ABM properties
b) Loading the brace ends one by one individually
with the axial force N, in plane moment M
i p
and out of plane moment M
op
c) Calculating the beam end deformations d
ABM
and the relative displacement error
j,i
at the
beam end where the loads is applied
d) Computation of the objective function
e) Parameter optimization step, repeating steps 1
to 4 by incremental change of each parameter
f) Extrapolation to the new ABM properties
4. End iteration
Beam 1
Beam 2
Beam 3
Beam 4
x
4
y
4
z
4
Figure 6: Simplied Adequate Beam Model with a local
coordinate system
6 Results
6.1 Grid search
The grid search approach did not generate satisfactory
results. An error of 20% remains on some DOFs of
the best found parameter set (Figure 7). Furthermore,
several dierent parameter sets are found to give the
same maximal residual error, while the error is distributed
dierently among the DOFs for dierent sets. They are
obtained with dierent initial values of the parameters,
and compete as the optimal solution. This indicates
that likely no exact solution exists, which in turn raises
the question of the analysis of the ABM quality. It is a
complex question to determine which parameter set best
ts the original Solid Element Model.
0
0.2
0.4
0.6
0.8
1
1.2
0 5 10 15 20 25 30 35 40 45
Iteration
A
D
rip1
rip2
u1
u2
Figure 7: Development of the relative error
j,i
(AD)
A deeper understanding of joint dynamics would allow
prioritization of some DOFs over others, depending on
their contribution to global joint and structure behavior.
This would allow a sound choice between competing pa-
rameter sets and could be achieved with further nite
element simulation.
6.2 Simultaneous parameter optimization
The results are shown in Figure 8. The rst three graphs
from the top depict the development of the ABM pa-
rameters through the iteration process. The respective
cross sectional properties are given as ratios relative to
the starting values, which were initially set to those of
the adjoining pipe sections. The relative displacement
errors are shown in the last six graphs. The results show
that some relative errors are nearly independent of the
parameters.
For example, the plot in the 3
rd
row, 2
nd
column shows
the relative errors for the deformations at the beam end
of the lower brace for the axial force load case. The
relative errors for the rotations around the local x-axis
(torque), the y-axis (out-of-plane bending) and the de-
formation in z-direction (out-of-plane) remain constant
at a level of 90%.
A comparison of the results with the calculated de-
formations of the reference SEM for the same load case
shows that the in-plane deformations x, z and rotation
y are larger than the respective out-of-plane displace-
ments. It would therefore be reasonable to select only
the most signicant deformation and to determine the
ABM properties in several steps.
5
Proceedings of the European Oshore Wind Conference 2009 Wiemann
0.0
0.5
1.0
1.5
2.0
2.5
0 500 1000 1500
Iteration [-]
VAProz
VIyProz
VIzProz
0.0
0.5
1.0
1.5
2.0
2.5
0 500 1000 1500
Iteration [-]
BUAProz
BUIyProz
BUIzProz
0.0
1.0
2.0
3.0
4.0
5.0
0 500 1000 1500
Iteration [-]
BDAProz
BDIyProz
BDIzProz
0.0
0.5
1.0
1.5
2.0
0 500 1000 1500
Iteration [-]
BUDNU
x_min
BUDNU
y_min
BUDNU
z_min
BUDNR
otx_min
BUDNR
oty_min
BUDNR
otz_min
0.0
0.5
1.0
1.5
2.0
0 500 1000 1500
Iteration [-]
BDDNU
x_min
BDDNU
y_min
BDDNU
z_min
BDDNR
otx_min
BDDNR
oty_min
BDDNR
otz_min
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
0 500 1000 1500
Iteration [-]
BUMip
Ux_min
BUMip
Uy_min
BUMip
Uz_min
BUMip
Rotz_min
0.0
0.5
1.0
1.5
2.0
0 500 1000 1500
Iteration [-]
BDMip
Ux_min
BDMip
Uy_min
BDMip
Uz_min
BDMip
Rotz_min
0.0
0.5
1.0
1.5
2.0
2.5
3.0
0 500 1000 1500
Iteration [-]
BUMop
Ux_min
BUMop
Uy_min
BUMop
Roty_min
BUMop
Uz_min
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
0 500 1000 1500
Iteration [-]
BDMop
Ux_min
BDMop
Uy_min
BDMop
Uz_min
BDMop
Roty_min
Legend:
V: Offset beam
BU: Upper brace
BD: Lower brace
A, Iy,Iz: Beam properties
Proz: Percentage of adjacent
pipe sectional prop.
: Relative displacement
error
DN, Mip, Mop: Load case
ux,uy,uz: Translational deformation
Rotx,Roty,Rotz: Rotational deformation
Figure 8: Results of simultaneous parameter optimiza-
tion
7 Conclusion and outlook
As the results for both optimization approaches show,
the use of the lowest possible number of beams limitates
the adaptability of the Adequate Beam Model. Improve-
ments could be made by increasing the number of beams
and parameters with the aim of achieving a single, satis-
fying solution. However, the additional complexity inher-
ent to those extra parameters implies a trade-o between
two main objectives: quality and simplicity.
Optimization approaches could also be further im-
proved. One possibility is the use of parameterized ma-
trices. The stiness matrix from the solid element model
can be obtained by means of static reduction as de-
scribed in [3]. A 36x36 matrix describes the deforma-
bility properties for the case study presented here. The
stiness matrix from the Adequate Beam Model can in
turn be analytically determined taking advantage of a
computer algebra program, with terms containing the
optimization parameters. The sectional properties of the
ABM could then be identied by comparison of stiness
matrices of the Solid Element Model and the analytical
model.
First tests by using the least-square sum of the matrix
coecients of the dierence of both stiness matrices
as minimization criteria revealed diculties in the stabil-
ity of the solution and sensitivity to starting parameters.
A possible approach to ensure better convergence could
be use of the eigenmodes of the solid element models
stiness matrix by means of minimizing the energy devi-
ation in the non-zero energy modes or the use of realistic
beam deformations from the global structure model of
the oshore wind turbine.
References
[1] Det Norske Veritas . Design of oshore wind turbine
structures, Standard DNV-OS-J101 2007.
[2] Buitrago J., Healy B. E.. Local joint exibility of
tubular joints Oshore Mechanics & Arctic Engineer-
ing, ASME. 1993;1.
[3] Guyan R.J.. Reduction of stiness and mass matrices
AIAA journal. 1965;3:380.
[4] Ibs J. B.. Fatigue design of oshore wind tur-
bines and support structures in Proceedings of the
OWEMES Conference, Naples, Italy 2003.
[5] ANSYS Mechanical . Release 11.0 Documentation
for ANSYS. SAS IP Inc. 2007.
[6] Wiemann K., Huhn H., Busmann H.-G.. Best t
approach for joint exibility in simulation of space
frames in Proceedings of the European Oshore
Wind Conference, Berlin, Germany 2007.
[7] Hooke R., Jeeves T. A.. Direct Search solution
of numerical and statistical problems Journal of the
Association for Computing Machinery. 1961;8:212
229.
6

Vous aimerez peut-être aussi