Thin-Walled Structures 54 (2012) 140155

Contents lists available at SciVerse ScienceDirect

Thin-Walled Structures
journal homepage: www.elsevier.com/locate/tws

Experimental investigation of buckling of wind turbine tower cylindrical

shells with opening and stiffening under bending
C.A. Dimopoulos n, C.J. Gantes
Metal Structures Laboratory, School of Civil Engineering, National Technical University of Athens, Greece

a r t i c l e i n f o a b s t r a c t

Article history: An experimental and numerical study of the buckling behavior of cantilevered shells with opening and
Received 13 June 2011 stiffening is presented in this paper. Unlike previous experimental studies, the present work focuses on
Received in revised form shell slenderness as well as opening and stiffening reecting the main geometric characteristics of wind
17 February 2012
turbine towers. The specimens can be classied as medium slenderness shells affected mainly by
Accepted 19 February 2012
Available online 22 March 2012
inelastic effects and secondarily by geometric imperfections. Both loaddisplacement curves as well as
strain measurements are presented and compared with numerical predictions by nite element
Keywords: analyses, accounting for both inelastic effects and geometrical nonlinearity as well as for contact
Shells interaction between the various parts of the specimens. A good agreement between numerical and
Opening
experimental results was found in terms of loaddisplacement curves and ultimate load. Due to the
Stiffening
inuence of the shape and size of geometric imperfections, a complete match of the numerically
Buckling
Experiments obtained strains to the corresponding experimental ones was not possible. The provided stiffening was
Numerical analysis found to be able to compensate the strength loss due to the presence of the cut-out.
Wind turbine tower & 2012 Elsevier Ltd. All rights reserved.

1. Introduction As far as the bending loading is concerned, some experimental

Shells, due to their importance as structural elements in many
engineering structures, have been receiving a great attention by
scientists. Their response has been investigated both experimen-
tally and numerically, especially when it comes to cylindrical
shells under axial compression or bending.
Axial compression tests in the elastic region were conducted
by Lundquist [1], Tennyson [2], Weingarten et al. [3], Schneider
et al. [4] and Athiannan and Palaninathan [5] just to name a few.
In the inelastic region some experimental work on axially com-
pressed shells is that of Lee [6], Batterman [7], Osgood [8], Horton
et al. [9], Bardi et al. [10] and Bardi and Kyriakides [11].
It is well known nowadays that the experimental buckling
loads of very thin shells in compression, which buckle in the
elastic region, exhibit a wide scatter and may be signicantly
lower than the classical analytical predictions. This phenomenon
can be attributed mainly to the inevitable geometrical imperfec-
tions and secondarily to loading eccentricities, boundary condi-
tions and variability in thickness and material properties. As the
shell slenderness decreases the effect of those factors on the
collapse loads becomes less signicant.
n
Corresponding author.
E-mail address: dchristoforos@hotmail.com (C.A. Dimopoulos).
work on thin shells is that of Mossman and Robinson [12], Rhode
and Lundquist [13], Imperial [14], Lundquist [15,16], Donnell [17].
In the studies of Suer et al. [18] and Mathon and Limam [19], the
effect of the internal pressure was also taken into account. In the
inelastic region, experiments for the case of pure bending have
been carried out by Moore and Clark [20], Jirsa et al. [21],
Sherman [22], Tuggu and Schroeder [23], Reddy [24], Kyriakides
and Shaw [25] and Kyriakides and Ju [26]. In the work of Johns
et al. [27], Corona and Kyriakides [28] and Ju and Kyriakides [29]
the effect of external pressure on the bending strength of the
shells was also included.
The aforementioned references concerned shells without any
geometrical discontinuity on their surface. However, in some
engineering structures, such as wind turbine towers, chimneys
and tanks, it is mandatory to provide openings in order to fulll
various practical needs, such as manholes. The effect of openings
on the shell strength was investigated in a number of experi-
mental works. In some cases, the effect of stiffening of the
opening on the collapse load of shells was also considered due
to its practical importance. Typical experimental studies of axially
compressed shells with openings are those of Starnes [3033],
Tennyson [34], Toda [3538], Schulz [39], Almroth and Holmes
[40], Bennet et al. [41], Han et al. [42] and Shariati and Rokhi [43].
In the case of shells with openings under bending, experi-
mental work has been carried out by Yeh et al. [44], Poursaedi
et al. [45] and Knodel and Schulz [46]. Yeh et al. [44] studied the

0263-8231/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tws.2012.02.011
 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155 141

elasto-plastic buckling of aluminum cylindrical shells with unrei-

forced cut-outs with orthogonal or circular shape under the state
of pure bending, both experimentally and numerically. The ratio
of diameter to thickness was equal to 50 while the ratio of length
to diameter was equal to 7.9. Poursaedi et al. [45] studied the
plastic buckling of stainless steel cylindrical shell with rectangu-
lar or circular cut-outs under bending. The diameter to thickness
ratio of the specimens was 40.4 and their length to diameter ratio
was 7.94. In contrary to previous work, Knodel and Schulz [46]
studied experimentally the strength of cylindrical steel shells
with rectangular cut-outs, with chamfered angles, under bending.
The effect of stringer stiffeners of various cross-section proles on
the strength was also studied. However, the slenderness of their
specimens was, in general, signicantly larger than values corre-
sponding to wind turbine towers. Moreover, in the few cases with
lower slenderness, the angles of their openings was signicantly
larger (in the order of 1201) than the typical openings of wind Fig. 1. Basic parts of test specimens.
turbine towers (approximately 251).
More experimental work on steel cylindrical shells with
reinforced or unreinforced rectangular cutouts under bending
are those of Baehre and Knodel [47] and O ry et al. [48]. As in

the case of Knodel and Schulz experiments [46], in these two
studies, either the angles of the openings and/or the slenderness
of the specimens are generally larger than those encountered in
wind turbine towers. Fig. 2. Geometrical properties of the opening.
To the authors knowledge no previous experimental work has
been carried out that is especially designed so as to correspond to
wind turbine towers, which exhibit specic geometrical and
loading characteristics. In order to cover this gap, an experimental
study is presented in this paper which focuses on the buckling
behavior of cantilever shells with opening and stiffening that
reect the main geometric characteristics of wind turbine towers.
The specimens are loaded with a prescribed transverse displace-
ment at the end of the cantilever, thus making bending the
predominant action. Both loaddisplacement curves as well as
strain measurements at a number of characteristic positions of
the external surface of the shells are presented. These results are
compared with numerical analyses performed with the commer-
cial nite element program ABAQUS [49]. Three different rene-
ment levels of numerical modeling are presented, which are
characterized by different levels of accuracy. In the less satisfac-
tory model, the presence of bolts tying the anges is neglected Fig. 3. Arrangement of the stiffener.
and the support is considered to be fully clamped. In the most
accurate model, the interaction of bolts and anges and the
interaction between anges as well as a partially clamped support
are taken into account.

2. Geometric characteristics of specimens and experimental

set-up

A total number of six shells with the same overall geometry

were tested, among which the rst two had no opening, the next
two had an unstiffened opening near their base, while in the last
two the opening was stiffened. The shape of the opening was
rectangular with elliptical ends and the stiffening, when used,
was a frame welded around the opening. The geometrical char-
acteristics of these shells (external diameter, thickness, cut-out
dimensions and stiffening) were chosen in such a manner as to
correspond to the characteristics of wind turbine towers encoun-
tered in practice. The other parts of the specimens were dimen-
sioned with sufcient overstrength, so that they would remain in
the elastic region during the whole duration of loading.
The various parts of the test specimens are shown in Fig. 1.
Each specimen consisted of three parts. The rst part (Block 18),
used in all tests, was a shell with external diameter of 400 mm Fig. 4. Dimensions of Block 10 plate.
142 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155

and a thickness of 6 mm. Imposed transverse displacement was
applied through the load cell to the end ange of this part. This
rst part was connected to the second, replaceable part through
two 30 mm thick anges and 28 M20/10.9 bolts. This second part,
which was the basic experimental specimen scaled down from
actual wind turbine tower cylindrical shells in terms of geometric
properties, was a shell with external diameter of 400 mm and a
thickness of 4 mm. This part came in six pieces and was replaced
for each test: two without opening (Block 12), two with opening
(Block 14) and the last two with openings reinforced by a
stiffening frame (Block 16). In Block 16 the stiffeners had a width
of 35 mm and a thickness of 6 mm. The geometrical character-
istics of the openings are given in Fig. 2 while the arrangement of
the stiffeners with respect to the opening is given in Fig. 3. The
second part was connected to the third part, used in all tests,
through two 35 mm and 40 mm thick anges and 28 M20/10.9

Fig. 5. Geometrical characteristics of anges.

Fig. 6. Laboratory test frame and test conguration.

 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155 143

bolts. The third part (Block 10) consisted of a thick short cylinder curves, using the transverse displacement of the upper point
with external diameter of 400 mm, thickness of 20 mm and of the ange at the free end of the specimens as characteristic
length of 80 mm welded to a 60 mm thick plate. The geometrical deformation quantity. The measurements of the applied
characteristics of this plate are given in Fig. 4 with hole locations displacement and the developed force of the load cell pro-
dictated by the corresponding bolt holes in the testing frame. The vided the desirable loaddisplacement curves.
geometrical characteristics of the anges are shown in Fig. 5, (ii) measurements that characterise the local structural response
while the experimental set-up is illustrated in Figs. 6 and 7. of the specimens. These measurements were performed with
The cut-out of the specimens was chosen to be situated on the
compression side. This was found in [44] to be the most severe
position of the angle that leads to the lowest collapse load, while
the maximum load appeared when the cut-out was on the neutral 600
axis. On the contrary, [50] found that the most adverse case for
shells with stiffened cut-out is when the position angle of the cut-
out is 22.51 with respect to ber of maximum compression. This 500
issue is further investigated in Section 7.
The measurements performed during the tests can be classi-

True Stress (MPa)

400
ed in two categories:

(i) measurements that characterise the overall structural 300

response of the specimens in terms of loaddisplacement

200

Block 12
100 Block 14 & Shell of Block 16
Stiffener of Block 16
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Plastic Strain

Fig. 9. True stressplastic strain curves of the steel material.

Table 1
Basic properties of specimens steel.

Shell of Shell of Stiffener

block 12 block 14 and 16 of block 16

Youngs modulus (GPa) 208 206 208

Yield stressa (MPa) 270 272 269

a
Fig. 7. Experimental set-up. The yield stress is taken equal to the 0.2% proof stress.

Fig. 8. Position of strain gauges for strain measurement.

144 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155
the aid of a number of strain gauges at characteristic positions
at the outer surface of blocks 12, 14 and 16. The number and
positions of the strain gauges were the same for all the
specimens. In the case of the shell with reinforced opening
two more strain gauges were positioned at the outer surface
of the stiffeners. The exact positions of strain gauges are
shown in Fig. 8. In all cases axial strains were measured, with
the exception of positions A1, A2, A3 and A4 where the
circumferential strains were measured as well.
3. Material characteristics of specimens
Due to the overstrength of all other parts, only the material
properties of the second part have been extracted through tensile
tests. The true stressplastic strain plots for the material of the
three blocks are given in Fig. 9. This material behavior was used in
the numerical simulation described below. For the shells of parts
Block 14 and Block 16 the Youngs Modulus was equal to
206 GPa while the yield stress, taken equal to the 0.2% proof stress,
was equal to 272 MPa. The stiffeners of the part Block 16 had
Youngs Modulus equal to 208 GPa and yield stress equal to
269 MPa. Finally, the shell of the part Block 12 had Youngs
Fig. 10. Numerical model for the estimation of support exibility.
Modulus equal to 208 GPa and yield stress equal to 270 MPa. The
basic properties of the specimens steel are summarized in Table 1.
4. Description of numerical models
For the numerical computations, the commercial nite ele-
ment analysis program ABAQUS [49] has been used. In order to
maintain an acceptable level of accuracy and at the same time
build a numerical model that is computationally effective in
terms of computation resources and solution time, in the basic
numerical model the presence of the test frame and the interac-
tion of the plate of block 10 with the frame column were ignored.
This type of interaction was, however, found to be important
when it comes to an accurate representation of the initial stiffness
of the loaddisplacement curve of the cantilever response.
4.1. Model for simulation of the support structure
For an acceptable estimation of the support exibility, a
simplied numerical model was built that simulated the interac-
tion of block 10 with the column of the frame (Fig. 10). Because of
symmetry conditions only one half was simulated and appro-
priate symmetry boundary conditions were applied at the column
and block 10. Moreover, the edge sections of the column were
considered as clamped. As can been seen in Fig. 10, stiffeners
were introduced on both sides of the column web in order to
increase its stiffness and local strength.
This model was composed of: (i) Block 10 including cylinder
and end plate, (ii) frame column with HEB 500 cross-section and a
length of 1800 mm, (iii) stiffeners, (iv) M20/10.9 bolts including
head and nut. For all the parts the C3D8R continuum nite
element was used, which is an 8-node linear brick, reduced
integration, hourglass control nite element. The applied load
was a concentrated load of magnitude 27.6266 kN and a concen-
trated moment of magnitude 76.665 kNm at the center of the
edge cross-section of the thick cylinder of Block 10. The load was
transmitted to the edge of the thick cylinder via the use of an MPC
Constraint.
The contact interaction between the different parts of the
model was simulated by the approach of contact pairs [49],
according to which all possible surfaces that are susceptible to
contact are recognized and set into contact pairs. For all contact
surfaces the same contact properties were assumed. According to
this property, the normal behavior was characterized by a Hard
Fig. 11. Numerical model of the cantilever.
 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155 145

Fig. 12. Details of nite element mesh in the area of cut-out, stiffening and bolted ange connection.

30 70

60
25

50
Load cell force (kN)
Applied Load (kN)

20
40
15
30

10 Block 121 (experiment)

20
Block 122 (experiment)
ABAQUS (no bolts clamped support)
5 10 ABAQUS (bolts clamped support)
ABAQUS (bolts flexible support)
0
0
0 0.02 0.04 0.06 0.08 0.1 0.12
0 1 2 3
x 104 Load cell displacement (m)
Displacement (m)
Fig. 15. Loaddisplacement curves for the shell with no opening.
Fig. 13. Loaddisplacement curve for frame columnplate model.

80 60

70
50
60
Applied Moment (kNm)

Load cell force (kN)

40
50

40 30

30
20
Block 141 (experiment)
20 Block 142 (experiment)
10 ABAQUS (no bolts clamped support)
10 ABAQUS (bolts clamped support)
ABAQUS (bolts flexible support)
0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 0 0.02 0.04 0.06 0.08 0.1 0.12
Rotation (m) x 103 Load cell displacement (m)

Fig. 14. Momentrotation curve for frame columnplate model. Fig. 16. Loaddisplacement curves for the shell with opening.
146 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155

Contact pressure-overclosure and the default constraint enforce-
ment method, where the tangential behavior was described by
the penalty friction formulation and a friction coefcient equal to
0.5. For a detailed description of the properties of this contact
simulation, the reader should consult ABAQUS manuals. Due to
the highly nonlinear nature of the problem, the automatic
stabilization feature of ABAQUS was adopted in order to resolve
the convergence problems that were encountered during the
numerical solution.
4.2. Model for simulation of the cantilever specimen

The basic numerical model was the cantilever cylinder shown

in Fig. 11, composed of: (i) parts of Block 10 ignoring the thick
70
plate, (ii) Block 12 (shell without opening), 14 (shell with open-
ing) or 16 (shell with reinforced opening), including edge anges,
60 (iii) Block 18 including shell and end anges, (iv) M20/10.9 bolts
connecting anges, including head and nut.
50 The shells of Blocks 12, 14, 16 and 18 as well as the stiffener
Load cell force (kN)

of Block 16 were simulated via the S4R shell element, which is

40

30 70

Block 161 (experiment) 60

20
Block 162 (experiment)
ABAQUS (no bolts clamped support)
10 50
ABAQUS (bolts clamped support)
ABAQUS (bolts flexible support)
Load (kN) 40
0
0 0.02 0.04 0.06 0.08 0.1 0.12
Load cell displacement (m) 30

Fig. 17. Loaddisplacement curves for the shell with reinforced opening.
20
Block 12 (GMNA)
Table 2 10 Block 14 (GMNA)
Experimentalnumerical collapse loads for specimens. Block 16 (GMNA)
0
Experimental analyses Numerical analyses
0 0.02 0.04 0.06 0.08 0.1 0.12
Specimen 1 Specimen 2 Level 1 Level 2 Level 3 Displacement (m)

Block 12 64.492 65.130 63.528 62.289 62.077 Fig. 19. Numerical loaddisplacement curves for Block 12, Block 14 and Block 16.
Block 14 49.529 49.045 53.453 51.7485 51.772
Block 16 65.584 64.020 63.440 63.160 63.119

Level 1: GMNA analysis (no bolts-clamped support).

Level 2: GMNA analysis (bolts-clamped support).
Level 3: GMNA analysis (bolts-exible support).

70

60

50
Load cell force (kN)

40

30

20
Block 121 (experiment)
10 Block 141 (experiment)
Block 161 (experiment)
0
0 0.02 0.04 0.06 0.08 0.1 0.12
Load cell displacement (m)

Fig. 18. Experimental loaddisplacement curves for Block 12-1, Block 14-1 and
Block 16-1. Fig. 20. Experimental and numerical post-collapse modes.
 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155 147

a 4-node doubly curved thin or thick shell, reduced integration
nite element with hourglass control and capable of nite
membrane strains. The rest of the cantilever model (all
anges and the thick/short cylinder at the support) were
simulated via the previously mentioned C3D8R continuum
nite element.
For the cantilever model three numerical models of three
different levels of accuracy were built. In the rst model (no
bolts-clamped support) the presence of bolts was ignored, and
the neighboring anges were glued to each other via tie
constraints. Moreover, the support of the cantilever was con-
sidered as xed. The xity was applied at the center of the edge
cross-section and was transferred to the cross-section nodes via
an MPC constraint. In the second numerical model (bolts-
clamped support) the presence of the bolts was taken into
account so that the interaction between bolts and anges as
well as the interaction between anges was considered. How-
ever, in this model the support was considered as xed. In the
third numerical model (bolts-exible support), the presence of
bolts was taken into account as before. Moreover, the effect of
the non-rigid support was considered via the use of two springs
representing the vertical translational and horizontal rotational
stiffness of the support. The stiffness value for these two springs
was obtained by means of the model described in Section 4.1
and is given in the following.
The numerical simulation was performed in two separated
analyses. In the rst analysis, a static-general step was adopted
in which a concentrated force of magnitude 40 kN was applied,
which is lower than the collapse load. In order to reach the
limit load and calculate the post-buckling region, a second
analysis was performed using the restart option, where a
concentrated displacement was applied. As in the case of the
support structure described in the previous paragraph, the
basic characteristics of the contact simulation between the
different parts of the cantilever specimen were held the same.
In the numerical solution of the cantilever specimen the
automatic stabilization feature of the ABAQUS program was
adopted as well.
In Fig. 12 the details of the nite element mesh around the
opening as well as the stiffening and the bolted ange connection
are shown. Special provision was taken so that the nite element
mesh is ner in the region of the cut-out, in order to obtain
accurate results, accounting for the stress concentration and the
local buckling taking place in this area.
All shells analyzed are symmetrical, thus one could consider
using a half model for the numerical analyses. However, in

80 80

60 60
Load(kN)

Load(kN)

40 40
Block 121 (C1)
Block 121 (A1)
20 20 Block 122 (C1)
Block 122 (A1)
ABAQUS ABAQUS
0 0
2500 2000 1500 1000 500 0 0 200 400 600 800 1000 1200
Strain (s) Strain (s)
80 80

60 60
Load(kN)

Load(kN)

40 40
Block 121 (C2)
Block 121 (A2)
20 20 Block 122 (C2)
Block 122 (A2)
ABAQUS ABAQUS
0 0
2500 2000 1500 1000 500 0 0 500 1000
Strain (s) Strain (s)
80 80

60 60
Load(kN)

Load(kN)

40 40
Block 121 (C3)
Block 121 (A3) Block 122 (C3)
20 Block 122 (A3) 20
ABAQUS
ABAQUS
0 0
6000 4000 2000 0 0 2000 4000 6000
Strain (s) Strain (s)

Fig. 21. Strain measurements for shell without opening (points 1, 2, 3, A axial strain, C circumferential strain).
148 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155

nonlinear problems prone to buckling effects, symmetric struc-
tures subjected to symmetric loading patterns do not necessarily
buckle symmetrically. Thus, analyzing a half numerical model and
enforcing symmetrical boundary conditions on the axis of sym-
metry may or may not yield the same response (e.g., the same
characteristic loaddisplacement curves). This was indeed ver-
ied in this case, where signicant differences were observed
when analyzing a half model, particularly in the case of shells
with unstiffened cut-out. Thus, results from full models are
presented here.
5. Experimental and numerical results
5.1. Numerical estimation of support exibility
For the estimation of the support exibility, the numerical
model of Section 4.1 has been analyzed and the results obtained
are described in Figs. 13 and 14. Both geometrical and material
nonlinearities were taken into account. The concentrated vertical
force that loaded the structure was equal to one half of the
collapse load obtained from a pre-test numerical analysis of the
cantilever specimen. The applied horizontal bending moment was
calculated using the concentrated load multiplied by an appro-
priate lever arm.
In Fig. 13 the loaddisplacement curve of the loaded point is
given. It can be seen that, neglecting the initial part of the
curve, the remaining part of the curve is linear. The negligible
stiffness of the initial part is due to an almost rigid translation
(slip) in the vertical direction of the bolts, having a diameter of
20 mm, within the holes, having a diameter of 22 mm. The
slope of this curve is equal to 127625.59 kN/m and represents
the stiffness of the frame columnplate system in the vertical
translational degree of freedom. In Fig. 14 the moment
rotation curve of the loaded point is given, which is also linear.
As the parts were initially in contact in the horizontal direction,
no slip was encountered, so that a constant nite stiffness was
observed. The slope of this curve is equal to 54792.02 kNm/rad
and represents the stiffness of the frame columnplate system
in the horizontal rotation degree of freedom. These two stiff-
ness values were then introduced at the support of the canti-
lever model via linear springs.

80 80

60 60
Load(kN)

Load(kN)

40 40
Block 121 (C4)
Block 121 (A4)
20 20 Block 122 (C4)
Block 122 (A4)
ABAQUS ABAQUS
0 0
4000 3000 2000 1000 0 0 1000 2000 3000
Strain (s) Strain (s)
80 80

60 60
Load(kN)

Load(kN)

40 40

Block 121 (A5) Block 121 (A6)

20 Block 122 (A5) 20 Block 122 (A6)
ABAQUS ABAQUS
0 0
2000 1500 1000 500 0 2500 2000 1500 1000 500 0
Strain (s) Strain (s)
80 80

60 60
Load(kN)

Load(kN)

40 40
Block 121 (A7)
Block 121 (A8)
20 Block 122 (A7) 20 Block 122 (A8)
ABAQUS
ABAQUS
0 0
0 1000 2000 3000 4000 2000 1500 1000 500 0
Strain (s) Strain (s)

Fig. 22. Strain measurements for shell without opening (points 4, 5, 6, 7, 8, A axial strain, Ccircumferential strain).
 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155 149

5.2. Experimental and numerical response of the cantilever structure
In Figs. 15 to 17 characteristic loaddisplacement curves are
illustrated for both the experimental tests and the numerical
analyses. The applied displacement of the load cell at the free end
of the cantilever is plotted on the horizontal axis and the
developed reaction of the load cell due to the applied displace-
ment on the vertical axis. Figs. 1517 refer to the shells without
opening (block 12), with opening (block 14) and with reinforced
opening (block 16), respectively.
In the numerical investigation, three levels of increasing
accuracy were adopted. In the rst level no contact was taken
into account (the bolts were not modeled) and the support
was considered as clamped. In the second level, contact was
taken into account (the bolts were included in the numerical
models) and the support was considered as clamped. In the
third level, contact was taken into account (bolts were
included) and the support was exible. The exibility of the
support was taken into account by appropriate springs (see
Section 5.1). In Figs. 15, 16 and 17 as well as in Table 2, the
rst, second and third levels of accuracy are labeled as no
bolts-clamped support, bolts-clamped support and bolts-exible
support, respectively.
The numerical analyses presented in these gures accounted
for material and geometric nonlinearity, without considering
imperfections (GMNA analysis). It is observed that in the case of
shells without opening and with reinforced opening, a remarkable
agreement between numerical and experimental results is
achieved, provided that the presence of bolts and a exible
support are taken into account. When these two factors are
neglected in the numerical model, the prediction of the simula-
tion is not accurate enough. In the case of shells with unrein-
forced opening, the GMNA analysis overestimates the collapse
load but the overall response is close to the experimental one,
provided again that the support exibility and the presence of
bolts are considered. The collapse loads of all specimens obtained
from both experimental tests and numerical analyses are listed in
Table 2.
In order to evaluate graphically the effect of the opening and
the reinforcement on the overall structural response of the
specimens, the experimental loaddisplacement curves of three
out of the six specimens are plotted in Fig. 18. The corresponding
numerical curves are given in Fig. 19. From both gures it can be
observed that the presence of the opening led to a strength
reduction of about 24%. Moreover, the selected stiffening scheme
was sufcient for recovering this strength loss.

60 60

40 40
Load(kN)

Load(kN)

Block 141 (A1)

20 Block 142 (A1) 20 Block 141 (C1)
Block 142 (C1)
ABAQUS
ABAQUS
0 0
0 500 1000 1500 1000 500 0 500
Strain (s) Strain (s)
60 60

40 40
Load(kN)

Load(kN)

Block 141 (A2) Block 141 (C2)

20 20
Block 142 (A2) Block 142 (C2)
ABAQUS ABAQUS
0 0
5000 0 5000 10000 15000 1000 0 1000 2000 3000 4000
Strain (s) Strain (s)
60 60

40 40
Load(kN)

Load(kN)

Block 141 (A3)

20 20 Block 141 (C3)
Block 142 (A3)
Block 142 (C3)
ABAQUS
ABAQUS
0 0
0 500 1000 1500 1000 800 600 400 200 0 200
Strain (s) Strain (s)

Fig. 23. Strain measurements for shell with opening (points 1, 2, 3, A axial strain, C circumferential strain).
150 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155

60 60

Load(kN) 40 40

Load(kN)
Block 141 (A4) Block 141 (C4)
20 Block 142 (A4) 20
Block 142 (C4)
ABAQUS ABAQUS

0 0
5000 0 5000 10000 15000 1000 0 1000 2000 3000 4000
Strain (s) Strain (s)
60 60

40 40
Load(kN)

Load(kN)
20 Block 141 (A5) 20 Block 141 (A6)
Block 142 (A5) Block 142 (A6)
ABAQUS ABAQUS
0 0
6 4 2 0 6 4 2 0
Strain (s) x 104 Strain (s) x 104
60 60

40 40
Load(kN)

Load(kN)

20 Block 141 (A7) 20 Block 141 (A8)

Block 142 (A7) Block 142 (A8)
ABAQUS ABAQUS
0 0
1.5 1 0.5 0 0.5 1 2500 2000 1500 1000 500 0 500
Strain (s) x 104 Strain (s)

Fig. 24. Strain measurements for shell with opening (points 4, 5, 6, 7, 8, A axial strain, Ccircumferential strain).

In Fig. 20 the experimental and numerical post-collapse
modes are shown. A general observation is that the experi-
mental post-collapse modes of each of the two shells of the
same type (block 12, 14 and 16) are not the same, although
similarities can be seen. The most obvious difference is
observed in the case of shells without opening which can be
attributed to the different initial geometrical imperfections
that characterized each specimen. It is interesting to note,
however, that in spite of these differences the overall struc-
tural response, as described from the experimental load
displacement curves, as well as the experimental collapse
loads, is not altered signicantly.
In Figs. 21 to 27 the strain measurements are given for all
specimens tested. As specimens are sufciently thin, so that
local buckling develops in the post-collapse region, the match
between numerical and experimental results in that part of
the response is, in general, not good. In some cases the
experimentally measured strain was subjected to signicant
noise (e.g., axial strain at point 7 (A7) of Fig. 24) or the
laboratory device could not measure strain above a certain
value (e.g., axial strain at point 3 (A3) of Fig. 21). However, in
the majority of cases the pre-collapse strains calculated by the
numerical simulations are in a good agreement with experi-
mental measurements.
6. Effect of buckling mode imperfection on shell strength
All previous numerical analyses referred to a perfect struc-
ture (GMNA analyses). In this section the results of a numerical
study based on GMNIA analyses are presented. In these analyses
the rst buckling mode was considered as initial imperfection,
which is a common choice in nonlinear analyses as it is
computed easily with nite element codes and leads in many
cases to the lowest ultimate load. Despite these obvious advan-
tages, it is well known that this type of imperfection may not be
the most unfavorable one, particularly for shells (see e.g.,
Eurocode 3 Part 1.6), and therefore it may not be suitable for
the numerical calculation of the strength of the structures under
consideration. Nevertheless, this approach was chosen here for
obtaining at least a qualitative indication of the effect of
imperfections, particularly for shells with unreinforced or rein-
forced openings.
The rst buckling mode of the three types of structures
presented previously is shown in Fig. 28, as obtained from linear
buckling analyses. The numerical model used to obtain these
buckling modes and then to perform the GMNIA analyses with
this buckling mode as imperfection, was the simplest model
presented earlier where both bolts and springs at the support
were neglected. The most important reason for choosing this
 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155 151

80 80

60 60

Load (kN)

Load (kN)
40 40
Block 161 (A1) Block 161 (C1)
20 Block 162 (A1) 20 Block 162 (C1)
ABAQUS ABAQUS
0 0
4000 3000 2000 1000 0 1000 2000 1500 1000 500 0 500
Strain (s) Strain (s)

80 80

60 60
Load (kN)

Load (kN)
40 40
Block 161 (C2)
Block 161 (A2)
Block 162 (C2)
20 Block 162 (A2) 20
ABAQUS
ABAQUS
0 0
8000 6000 4000 2000 0 0 2000 4000 6000 8000 10000
Strain (s) Strain (s)
80 80

60 60
Load (kN)

Load (kN)

40 40
Block 161 (C3)
Block 161 (A3)
Block 162 (C3)
20 Block 162 (A3) 20
ABAQUS
ABAQUS
0 0
8000 6000 4000 2000 0 1000 0 1000 2000 3000 4000
Strain (s) Strain (s)

Fig. 25. Strain measurements for shell with reinforced opening (points 1, 2, 3, A axial strain, C circumferential strain).

model instead of the more rened ones is the fact that a buckling
analysis can not be performed with the presence of contacts
between the different parts of the structures. In Fig. 28 a
distinction is made between an inward and an outward type of
the rst buckling mode for the case of shell with opening. The
difference between the inward and the outward type lies only in
their sign (positive amplitude for the inward mode and negative
amplitude for the outward mode).
In Figs. 29 to 32 the loaddisplacement curves for different
imperfection amplitudes are plotted. Fig. 29 corresponds to the shell
without opening, Figs. 30, 31 to the shells with unreinforced opening
and an inward and outward imperfection, respectively, while Fig. 32
to the shell with reinforced opening. It can be seen that the effect of
the rst buckling mode imperfection is appreciable only in the case
of shells without opening. In the cases of shells with unreinforced or
reinforced openings this effect is negligible. The imperfection
sensitivity in terms of the strength ratio of imperfect to perfect
shell (PGMNIA/PGMNA) is listed in Table 3.
7. Effect of cut-out position on the collapse load
As mentioned above, the effect of the direction of the
bending moment with respect to the cut-out was studied by
Yeh et al. [44], where the buckling due to bending of an
elastoplastic cylindrical shell with unstiffened cut-out was
studied both experimentally and numerically. The most severe
case leading to the lowest collapse bending moment was
found to be the case where the cut-out was situated on the
compression side. The maximum collapse load was obtained
when the cut-out was on the neutral axis (of the shell without
cut-out).
On the contrary, Nuta et al. [50] found that the most adverse
case for shells with stiffened cut-out is when the angle Y
between the direction of loading and the position of the cut-
out is 22.51.
From a series of numerical analyses of our model (with contact
conditions and elastic support taken into account) the same
conclusions were drawn for unstiffened cut-outs as those by
Yeh et al., as can be seen in Fig. 33, where on the horizontal axis
the angle Y between the direction of loading and the position of
the cut-out is plotted and on the vertical axis the collapse load is
presented. For the case of stiffened cut-outs, it has been found
that the deviations of the collapse load with respect to the angle
Y are practically negligible. These results concern the specic
cross-section area of the stiffener necessary to compensate for the
strength loss due to the cut-out. A lighter stiffener would lead to
different results and conclusions.
152 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155

80 80

60 60
Load (kN)

Load (kN)
40 40
Block 161 (C4)
Block 161 (A4)
Block 162 (C4)
20 Block 162 (A4) 20
ABAQUS
ABAQUS
0 0
6000 4000 2000 0 0 2000 4000 6000 8000
Strain (s) Strain (s)
80 80

60 60
Load (kN)

Load (kN)
40 40
Block 161 (A5) Block 161 (A6)
20 Block 162 (A5) 20 Block 162 (A6)
ABAQUS ABAQUS
0 0
3000 2000 1000 0 4000 3000 2000 1000 0
Strain (s) Strain (s)
80 80

60 60
Load (kN)

Load (kN)

40 40
Block 161 (A7) Block 161 (A8)
20 Block 162 (A7) 20 Block 162 (A8)
ABAQUS ABAQUS
0 0
2 1 0 1 2500 2000 1500 1000 500 0
Strain (s) x 104 Strain (s)

Fig. 26. Strain measurements for shell with reinforced opening (points 4, 5, 6, 7, 8, A axial strain, Ccircumferential strain).

80 80

60 60
Load (kN)

Load (kN)

40 40

20 Block 161 (S1) 20 Block 161 (S2)

Block 162 (S1) Block 162 (S2)
ABAQUS ABAQUS
0 0
2000 1500 1000 500 0 500 2000 1500 1000 500 0 500
Strain (s) Strain (s)

Fig. 27. Strain measurements for shell with reinforced opening (axial strain at points S1 and S2 of stiffeners).

8. Conclusions and numerically. The geometrical parameters of the shells as

well as those of opening and stiffening were chosen in such a
The effect of opening and stiffening on the strength of manner so that they reect the main geometric characteristics
cylindrical shells has been investigated both experimentally of wind turbine towers. The shell specimens were of a
 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155 153

Fig. 28. First buckling mode for the three types of shells.

70 60

60 50

50
40
Load (kN)

Load (kN)

40
30
30 Block 12 (GMNIA,w0/t=0) Block 14 (GMNIA,w0/t=0, Inward)
Block 12 (GMNIA,w0/t=0.25) 20 Block 14 (GMNIA,w0/t=0.25, Inward)
20 Block 12 (GMNIA,w0/t=0.50) Block 14 (GMNIA,w0/t=0.50, Inward)
Block 12 (GMNIA,w0/t=0.75) 10 Block 14 (GMNIA,w0/t=0.75, Inward)
10
Block 12 (GMNIA,w0/t=1.0) Block 14 (GMNIA,w0/t=1.0, Inward)
0 0
0 0.02 0.04 0.06 0.08 0.1 0.12 0 0.02 0.04 0.06 0.08 0.1 0.12
Displacement (m) Displacement (m)

Fig. 29. Loaddisplacement curves for imperfect shell without opening. Fig. 30. Loaddisplacement curves for imperfect shell with opening (inward
imperfection).

cantilever type and the type of loading was a transverse

imposed displacement at the end of the cantilever.
The experimental study revealed that, even though the
deformed post-collapse geometry was affected by the pre-
sence of initial imperfections, the collapse load for each type
of shell (without opening, with opening or with reinforced
opening) was practically the same. The presence of the open-
ing led to a strength reduction of about 24%. Moreover, the
selected stiffening scheme was sufcient for recovering this
strength loss.
The experimental investigation was also followed by a numer-
ical simulation with the nite element method using ABAQUS.
The numerical simulation has been found to give excellent results,
in terms of the loaddisplacement curves, for the case of shell
without opening and shell with reinforced opening. In the case of
shell with unreinforced opening, the numerically estimated
collapse load was found to exceed by a little the experimentally
observed load. The numerically obtained strains were in the
majority of the cases in a less agreement with the corresponding
experimental strains which can be contributed mainly to the
presence of initial imperfections that are inevitably present in the
specimens.
154 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155

60 70

50 60

50

Collapse Load (kN)

40
Load (kN)

40
30
Block 14 (GMNIA,w0/t=0, Outward) 30
20 Block 14 (GMNIA,w0/t=0.25, Outward)
20
Block 14 (GMNIA,w0/t=0.50, Outward)
10 Block 14 (GMNIA,w0/t=0.75, Outward)
10 Shell with unstiffened cutout
Block 14 (GMNIA,w0/t=1.0, Outward) Shell with stiffened cutout
0 0
0 0.02 0.04 0.06 0.08 0.1 0.12 0 30 60 90 120 150 180
Displacement (m) ()

Fig. 31. Loaddisplacement curves for imperfect shell with opening (outward Fig. 33. Collapse load as function of the cut-out position angle (Y 01 at
imperfection). compression side and Y 1801 at tension side).

70
respect of the position of the cut-out is negligible, provided that
an appropriate cross-section area of the stiffener is used.
60

50 References
Load (kN)

40 [1] Lundquist EE. Strength tests of thin-walled duralumin cylinders in compres-

sion, NACA Report No. 473, 1933.
[2] Tennyson RC. Buckling of circular cylindrical shells in axial compression.
30 Block 16 (GMNIA,w0/t=0) AIAA Journal 1964;2:13513.
[3] Weingarten VI, Morgan EJ, Seide P. Elastic stability of thin-walled cylindrical
Block 16 (GMNIA,w0/t=0.25) and conical shells under axial compression. AIAA Journal 1965;3:5005.
20 [4] Schneider Jr. MH. Investigations of the stability of imperfect cylinders using
Block 16 (GMNIA,w0/t=0.50) structural models. Engineering Structures 1996;8:792800.
[5] Athiannan K, Palaninathan R. Experimental investigations on buckling of
Block 16 (GMNIA,w0/t=0.75) cylindrical shells under axial compression and transverse shear. Sadhana
10
Block 16 (GMNIA,w0/t=1.0) 2004;29(Part 1):93115.
[6] Lee LHN. Inelastic buckling of initially imperfect cylindrical shells subject to
0 axial compression. Journal of Aeronautical Sciences 1962;29:8795.
0 0.02 0.04 0.06 0.08 0.1 0.12 [7] Batterman SC. Plastic buckling of axially compressed cylindrical Shells. AIAA
Journal 1965;3:31625.
Displacement (m) [8] Osgood WR. The crinkling strength and the bending strength of round aircraft
tubing, NACA Report 632, 1938.
Fig. 32. Loaddisplacement curves for imperfect shell with reinforced opening. [9] Horton WH, Bailey SC, Edwards AM. Nonsymmetric buckle patterns in
progressive plastic buckling. Experimental Mechanics 1966;6:43344.
[10] Bardi FC, Yun HD, Kyriakides S. On the axisymmetric progressive crushing of
Table 3 circular tubes under axial compression. International Journal of Solids and
Imperfection sensitivity (PGMNIA/PGMNA). Structures 2003;40:313755.
[11] Bardi FC, Kyriakides S. Plastic buckling of circular tubes under axial
Type of shell compression part I: Experiments. International Journal of Mechanical
Sciences 2006;48:83041.
w0/t Block Block 14 (inward Block 14 (outward Block [12] Mossman RW, Robinson RG. Bending tests on metal monocoque fuselage
12 imperfection) imperfection) 16 construction, NACA TN 357, 1930.
[13] Rhode RV, Lundquist EE. Strength tests on paper cylinders in compression,
bending and shear, NACA TN 370, 1931.
0 1.0000 1.0000 1.0000 1.0000
[14] Imperial FF. The criterion of elastic instability of thin duralumin tubes
0.25 0.9752 1.0108 0.9960 0.9982
subjected to bending, MS Thesis, Dept. of Mechanical Engineering. University
0.50 0.9452 0.9973 0.9947 0.9928
of California, 1932.
0.75 0.9176 0.9874 0.9938 0.9851 [15] Lundquist EE. Strength tests of thin-walled duralumin cylinders in pure
1.0 0.8985 0.9792 0.9930 0.9761 Bending, NACA Report 479, 1933.
[16] Lundquist EE. Strength tests of thin-walled duralumin cylinders in combined
transverse shear and bending, NACA Report 523, 1935.
[17] Donnell LH. A new theory for the buckling of thin cylinders under axial
Beside GMNA analyses, GMNIA analyses have been conducted compression and bending. Transactions of theASME 1934;56:795806.
with an initial imperfection based on the rst buckling mode. In [18] Suer HS, Harris LA, Skene WT, Benjamin RJ. The bending stability of thin-
the case of the shell without opening, the effect of this imperfec- walled unstiffened circular cylinder including the effects of internal pressure.
Journal of Aeronautical Sciences 1958;25(5):2817.
tion is considerable. However, in the case of the other two types
[19] Mathon C, Limam A. Experimental collapse of thin cylindrical shells sub-
of shells, this effect is negligible. mitted to internal pressure and pure bending. Thin Walled Structures
Finally, it has been numerically conrmed that, for shells with 2006;44:3950.
unstiffened cut-outs, the collapse load attains its lowest value [20] Moore RL, Clark JW. Torsion, compression and bending tests of tubular
sections machined from 75S-T6 rolled round rod, NACA RM 52125, 1952.
when the cut-out is situated on the compression side. For the case [21] Jirsa JO, Lee FK, Wilhoit JC, Merwin JE. Ovaling of pipelines under pure
of stiffened cut-outs, the deviation of the collapse load with bending, OTC 1569, Proc. Offshore Tech. Conf. I, pp. 573578, 1972.
 C.A. Dimopoulos, C.J. Gantes / Thin-Walled Structures 54 (2012) 140155
[22] Sherman DR. Tests of circular steel tubes in bending, ASCE. Journal of
Structural Division 1976;102(ST1I):218195.
[23] Tuggu P, Schroeder J. Plastic deformation and stability of pipes exposed to
external coupler. International Journal of Solids and Structures 1979;15:64358.
[24] Reddy BD. An experimental study of the plastic buckling of circular cylinders in
pure bending. International Journal of Solids and Structures 1979;15:66985.
[25] Kyriakides S, Shaw PK. Inelastic buckling of tubes under cyclic bending. ASME
Journal of Pressure Vessel Technology 1987;109:16978.
[26] Kyriakides S, Ju GT. Bifurcation and localization instabilities in cylindrical
shells under bending. International Journal of Solids and Structures
1992;29(No. 9):111742.
[27] Johns TG, Mesloh RE, Winegardener R, Sorenson JE. Inelastic buckling of
pipelines under combined loads, OTC 2209, Proc. Offshore Tech. Conf., II,
pp. 635646, 1975.
[28] Corona E, Kyriakides S. On the collapse of inelastic tubes under combined
bending and pressure. International Journal of Solids and Structures
1988;24:50535.
[29] Ju GT, Kyriakides S. Bifurcation buckling vs limit load instabilities of elastic-
plastic tubes under bending and external pressure. ASME Journal of Offshore
Mechanics and Arctic Engineering 1991;113:4352.
[30] Starnes Jr. JH. The effect of a circular hole on the buckling of cylindrical shells,
PhD Thesis, California Institute of Technology, Pasadena, CA, 1970.
[31] Starnes Jr. JH. Effect of a slot on the buckling load of a cylindrical shell with a
circular cutout. AIAA Journal 1972;10(2):2279.
[32] Starnes Jr. JH. Effect of a circular hole on the buckling of cylindrical shells
loaded by axial compression. AIAA Journal 1972;10(11):146672.
[33] Starnes Jr. JH. The effect of cutouts on the buckling of thin shells. In: Fung YC,
Sechler EE, editors. Thin Shell Structures, Theory, Experiment and Design.
Englewood Cliffs, NJ: Prentice-Hall; 1974. p. 289304.
[34] Tennyson RC. The effect of unreinforced circular cutouts on the buckling of
circular cylindrical shells under axial compression. Transactions ASME,
Journal of Engineering for Industry 1968;90(4):5416.
[35] Toda S. Some considerations on the buckling of the buckling of thin
cylindrical shells with cutouts. Transactions, Japan Society for Aeronautical
and Space Sciences 1980;23(6):10412.
[36] Toda S. Buckling of cylinders with cutouts under axial compression. Experi-
mental Mechanics 1983;23(4):4147.
155
[37] Toda S. Experimental investigation on the effects of elliptic cutouts on the
buckling of cylindrical shells loaded by axial compression. Transactions,
Japan Society for Aeronautical and Space Sciences 1980;23(59):5763.
[38] Toda S. Buckling of Cylindrical Shells with Circular Cutouts under Axial Compres-
sion. Tokyo: National Aerospace Laboratory NAL TR-560; 1979. in Japanese.
[39] Schulz U. Stabilitat axial belasteter zylinderschalen mit manteloffnungen.
Der Bauingenieur 1976;51(10):38796.
[40] Almroth BO, Holmes AMC. Buckling of shells with cutouts, experiments and
analysis. International Journal of Solids and Structures 1972;8(8):105771.
[41] Bennett JG, Dove RC, Butler TA. An investigation of buckling of steel cylinders
with circular reinforced cutouts. Nuclear Engineering and Design
1982;69:22939.
[42] Han H, Cheng J, Taheri F, Neil Pegg. Numerical and experimental investiga-
tions of the response of aluminium cylinders with a cutout subject to axial
compression. Thin Walled Structures 2006;44:25470.
[43] Shariati M, Rokhi MM. Numerical and experimental investigations on buck-
ling of steel cylindrical shells with elliptical cutout subject to axial compres-
sion. Thin-Walled Structures 2008;46:125161.
[44] Yeh Meng-Kao, Lin Ming-Chyuan, Wu Wen-Tsang. Bending buckling of an
elastoplastic cylindrical shell with a cutout. Engineering Structures
1999;21:9961005.
[45] Poursaeidi E, Rahimi GH, Vafai AH. Plastic buckling of cylindrical shells with
cutouts. Asian Journal of Civil Engineering (Building and Housing) 2004;5(3
4):191207.
[46] Knodel P, Schulz U. Zur Stabilitat von Schornsteinen mit Fuchsoffnungen.
Stahlbau 1988;57(H. 1):1321.
[47] Baehre R, Knodel und Gebrauchsfahigkeit
P. Stabilitat von biegebeanspruch-
ten Stahlschornsteinen mit Ausschnitten, Gutachten fur die Fa. Mauer u.
Sohne vom Lehrstuhl fur Stahl- und Leichtbau, Universitat Karlsruhe, Februar
1986.
[48] O ry H, Ferlic N, Reimerdes HG. Groe Ausschnitte in langen Kreiszylin-
derschalen, Forsch.ber., T1863, 2. Fassung, 1987.
[49] ABAQUS/Standard and ABAQUS/Explicit Version 6.8-1, Abaqus Theory
Manual, Dassault Systems, 2008.
[50] Nuta E, Christopoulos C, Packer JA. Methodology for seismic risk assessment
for tubular steel wind turbine towers: application to Canadian seismic
environment. Canadian Journal of Civil Engineering 2011;38:293304.