Journal of Constructional Steel Research 96 (2014) 81–94

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

Experimental investigations on the stability of stiffened cylindrical shells

of steel silos
Eugeniusz Hotała, Łukasz Skotny ⁎
Wrocław University of Technology, Wybrzeże Wyspiańskiego 27, 50-370 Wrocław, Poland

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

Article history: Owing to technological reasons the silos are frequently supported in a discrete way, usually by column heads.
Received 25 December 2012 This kind of support causes a significant accumulation of compression meridional stresses in a cylindrical shell
Accepted 3 January 2014 in support regions, which may lead to local buckling. The problem of determining shell resistance over the sup-
Available online 8 February 2014
port region for different shapes of ribs constitutes a current research issue. One of highly preferred methods of
strengthening this zone is the use of short ribs interconnected with a circumferential ring. Both the results of
Keywords:
Shell stability
tests on the resistance of such shells and numerical analysis allowing to determine the course of this research
Stiffened shell are presented in this study. It has been demonstrated that the global resistance of the stiffened shells supported
Discrete support discretely is always much smaller from that in similar shells supported uniformly around the perimeter.
Steel silo © 2014 Elsevier Ltd. All rights reserved.
Experimental investigations

1. Introduction
Cylindrical walls of steel silos are often supported on columns' heads
(Figs. 1, 2) or on the ceiling beams in production buildings. In such cases
the silo walls are longitudinally compressed as cylindrical shells dis-
cretely supported on columns along the width of s0 (Figs. 2, 3).
The vertical reaction of the single discrete support N1 (Fig. 3a) with
an arbitrary width s0 causes meridional compression stress σx01 deter-
mined by formula (1) at x = 0, which is much higher than the stress
σx0 obtained for a shell uniformly supported around the perimeter
(Eq. (2)):
N1
σ x01 ¼
s0  t
n  N1
σ x0 ¼ : ð2Þ
2π  r  t
The meridional compression stresses σx01 for shell's column sup-
ports with number n ≥ 3 are several times larger than stresses σx0
along the bearing edge (x = 0) for a uniformly supported shell. The
local instability of the unstiffened cylindrical shell in the discrete sup-
port regions (Fig. 2a) is studied in interalia [1–4]. The strengthening of
the supporting zone due to buckling is often executed by short ribs
with the length L1 (Fig. 2b). The resistance of such shells has been tested
by Komman and Pasternak [5–7].
⁎ Corresponding author. Wrocław University of Technology, Ul. Skrzydlata 1/7, 54-129
Wrocław, Poland.
E-mail address: lukasz.skotny@pwr.wroc.pl (Ł Skotny).
It is often assumed that the use of transition ring over the ribs with
length L1 N r (Fig. 2c) provides a uniform distribution of meridional
stresses σx around the entire perimeter just above the ring. One might
think that the ribbed, bottom portion of the shell wall could act as a
rigid apron referred to in EN 1993-4-1:2007 + AC:2008 (p. 5.4.2(1b))
[8], and that is supposed to be able to provide a state of meridional
stress, such as in the case of the shell uniformly supported on the
circumference of the silo, although the actual support is realized by
means of columns. This provision in the standard for the role of a cir-
cumferential apron in the dispersion of stress gradient from the support
reaction N1 (Fig. 3a) is not correct, which can be shown by any simple
numerical analysis of the state of stress in this zone.
The authors have done a series of experimental tests on ribbed cylin-
ð1Þ
drical shell walls supported discretely showing clearly that short ribs
connected by a transition ring (Figs. 1, 2) are not able to provide the uni-
form distribution of meridional stresses σx in the cylindrical silo shell,
supported discretely by columns, and that compression silo shell can
buckle in the zone placed over the short ribs.
2. Methods of the analysis of stability of cylindrical steel shell walls
in the region of discrete supports
Samuelson and Eggwertz [9] have proved clearly, that the main
stream of meridional compression stress σx01 (Fig. 3a) in the cylindrical
shell supported discretely on bearings with the width s1 is not dispersed
at an angle α = 45°, as it is taken in planar members according to Saint–
Venant principle. Hence, some authors [1,4] have considered that safe
value of stresses in analyzed resistance of shells is equal to the maxi-
mum of meridional stresses σx = σx01 (Fig. 3a) occurring at the bottom

0143-974X/$ – see front matter © 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jcsr.2014.01.009
82 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94

Fig. 1. Examples of stiffened, column supported steel silos with the use of transition ring over short ribs.

edge of the shell. The design value of meridional stresses σx,Ed at the
level x = 0, determined on the basis of the modified formula (Eq. (1))
should be less than the value of design critical stresses σx,Rd, determined
in accordance with current standards.
Knödel and Ummenhoffer [3] have proposed a less conservative way
of determining the value of stresses σx,Ed at the level x corresponding
to the half of height of the buckling wave, assuming that the angle
of dispersion of the stress stream up to the half height of the wave is
α = 75° (Fig. 3a). Another approach in the analysis of the resistance
of shell walls, column supported has been applied by Hotała [2], who
proposed the determination of the critical reaction N1,Rd from the dis-
crete support, as compared with the design value of reaction N1,Ed. He
has also calibrated the buckling curve needed to determine the value
N1,Rd. Similarly, the analysis of the resistance of cylindrical, column sup-
ported shells has been proposed by Komman [5], who determined the
critical reaction N1,Rd in the case of short ribs used over the column sup-
ports (Fig. 2b). Komman and Pasternak [5–7] have clearly shown that
the use of short ribs has a beneficial influence on the resistance of
discretely supported cylindrical shell, and that the basic shape of buck-
ling for this system occurs as displacement of the upper edges of ribs

Fig. 2. Cylindrical shells of steel silos supported by columns: a) without ribs, b) with short ribs, c) with short ribs and transition ring.
 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94
Fig. 3. Distribution of stress σx in a cylindrical shell over the column support with the width so and reaction N1: a) simplified model, b) stress contours σx from FEM analysis.
towards the inside of the shell (Fig. 4b). Owing the form of buckling the
use of intermediate rings appears to be justified in order to prevent this
effect.
The strong concentration of stress σx,Ed is observed along the axes
of columns supporting the cylindrical shell. The standard EN 1993-4-
1:2007 [8] provides the procedure for a reliable determination of the
stress σx1,Ed (Fig. 5a) for a relatively smooth distribution of the stress
around the perimeter. The actual distribution of the stress in the
discussed above zone (Fig. 5b) differs greatly from that given in stan-
dard. The difference is caused by greater discrepancy between the
values of stresses directly placed over the support axis σx0,Ed, and stress-
es σx1,Ed occurring at the distance y (Eq. (3)) from the support axis
(Fig. 5b). From the standard assumptions their ratio s should satisfy
the condition (Eq. (5)):
pffiffiffiffi
y ¼ rΔθ ¼ 4 rt
σ x1;Ed
s¼
σ x0;Ed
0:3 b s b 1:0:
On the basis of own numerical analysis of the shells discretely
supported it has been found, that for the small widths s1 of bearing
(Fig. 2a), the stress ratio is usually s = 0.1 ÷ 0.2. In this situation,
when condition (5) of this standard is not satisfied, the distance y
should be chosen so that s ≈ 0.5. Therefore for narrow discrete supports
Fig. 4. Obtained from GNA analysis the shapes of buckling for a discretely supported compressed shell: a) without ribs in the supporting zone, b) with short ribs, c) with short ribs topped
by an intermediate ring (displacements enlarged 10 times).
83
the actual stress distribution is not important since arbitrarily imposed
stress ratio s = 0.5 must be used while distance y has no influence on
design. This approach raises considerable doubts.
A completely different issue is the possibility to determine the resis-
tance of discretely supported cylindrical shells by using numerical
methods, introduced in standards EN 1993-4-1:2007 [8] and EN 1993-
1-6:2009 [10]. One should stress the importance of rapid and simple
methods for the shell's resistance evaluation to verify the correctness
of numerical analyses. Simple analytical methods are always useful in
a practical design, especially when they are supported by experimental
investigations.
The standard EN 1993-4-1:2007 [8] presents rather complicated so-
lutions of discrete shells' support, but there are no references to a rela-
tively simple and very efficient constructions of supporting zones for
shells' column supports, presented in Figs. 1 and 2c. The introduced
ð3Þ test results of cylindrical models of silo shells have shown tangible ben-
efits resulting from the application of such design solutions. These tests
have also revealed that even good stiffening of the supporting zones
ð4Þ
does not provide a compression resistance of the shell comparable to
the resistance of shells uniformly supported around the perimeter.
ð5Þ
3. Characteristics of tested models of cylindrical shells
Experimental investigations of the resistance of cylindrical, discrete-
ly supported shells under the longitudinal compression have been
carried out on steel models (Fig. 6) with diameter D = 1000 mm
and length L = 1000 mm. The cylindrical shells of models have been
84 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94

Fig. 5. Distribution of compression, meridional stresses over the discrete support: a) pre-
sented in standard EN-1993-4-1, b) obtained from FEM analysis for shell with r =500 mm,
t = 1 mm, s0 = 2 mm.
Fig. 7. Measuring strip.

made out of steel with Young modulus E = 195 400 N/mm2 and yield
strength fy = 331 N/mm2 for the shell thickness t = 1 mm, and E =
204 000 N/mm2, fy = 428 N/mm2 for t = 2 mm. Both edges have been
finished by rings 20 × 20 mm (tG = tD = 20 mm), the intermediate
ring and stiffeners were made from a flat bar 20 × 10 mm (tP = tZ =
10 mm). Stiffeners and rings from steel S235 were welded to the shell.
Basic tests have been executed on models with short ribs with length
L1 within range 0.24 ≤ L1 / L ≤ 1.04 and topped by an intermediate ring
(Fig. 6c). A part of the comparative research has been associated with
unstiffened models (Fig. 6a) and with stiffened models without the in-
termediate ring (Fig. 6b). Research has been carried out for 70 cases of
loading.
Two widths of discrete support have been used: s0 = 0.0006πD
and s0 = 0.0191πD. The first width corresponds to almost a spot
supporting and the second one is the same as generally used for silo col- Fig. 8. Results of the analysis of convergence for the numerical model with parameters:
umn heads. Three or four discrete supports have been applied. The class L1 / L = 0.27, r / t = 500, s0 / 2πr = 0.0191.

Fig. 6. Geometry of tested cylindrical shells' models: a) without ribs in the region of column supports, b) with short ribs over the supports, c) with short ribs topped by an intermediate ring,
d) cross-section through the shell in support axis.
 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94 85

Fig. 9. Schemes of individual numerical models: a) model 1, b) model 2, c) model 3.

of workmanship of the tested shells has been high (the quality parame- In the described tests above the way of loading has been opposite to
ter Q = 25 according to EN 1993-1-6:2009 [10]) and evaluated on the the load pattern occurring in the real silos. The upper edge of each model
basis of detailed measurements of geometrical imperfections (Fig. 7). (Fig. 6) has been supported directly on a flat non-deformable surface

Fig. 10. Results of elastic analysis for model 1 (L1 / L = 0.27, r / t = 500, s0 / 2πr = 0.0191) loaded uniformly: a) distribution of meridional compressive stress, b) distribution of vertical
displacements.

Fig. 11. Results of elastic analyses for model 2 (L1 / L = 0.27, r / t = 500, s0 / 2πr = 0.0191) loaded by a uniform displacement Δ (Fig. 9b): a) distribution of meridional compressive stresses,
b) distribution of vertical displacements. Color scale matches the scale of Fig. 10.
86 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94
Fig. 12. Load–displacement relationship for numerical models 1–3.
around the entire perimeter, so the edge with the discrete supports has
been placed on the top. The longitudinal load has been applied to the
discrete supports by hydraulic cylinders until shell buckling. The vertical
displacement was controlled in order to stop the process of buckling in
any time, and for reading the load–displacement relationship.
4. Pre-experimental numerical analyses
Many numerical analyses of cylindrical silo shells have been carried
out before the experimental tests. The aim of this was to confirm the
predicted shape of models' collapse and to examine their resistance in
the region of the single support for the state of a single local load in com-
parison with total load applied to all supports simultaneously. The nu-
merical analyses of elastic shell models with perfect geometry have to
confirm that the critical reaction N1cr causing the local buckling of the
shell over the single support did not depend significantly on whether
on other supports similar reactions were applied at the same time or
not. This issue is important from practical point of view, because a single
shell model might be employed in tests 3 or 4 times throughout the
longitudinal loading applied only to the single discrete support. On the
contrary, the axial loading applied to all supports simultaneously
might lead to the total buckling of the model in support regions and,
consequently such a model could not be further used. It has been ex-
pected that for the tested shell the instability has a local character, and
the load-bearing capacity attained in one part of the shell over a single
support does not change the resistance of the rest of support regions.
This assumption might be valid for small number of supports, where
the interaction of individual shell's support regions is very limited. In
Fig. 13. Model with L1 / L = 0.27, r / t = 500, s0 / 2πr = 0.0191 loaded by displacement of one discrete support: a) distribution of meridional compression stresses, b) vertical displacements
in elastic range of model's behavior. Color scale matches the scale of Fig. 10.
numerical analysis of the models the 4 discrete supports have been
adopted, assuming the less favorable case.
As it has been mentioned before the test model has been reversed
with the discrete supports upwards, and the load of individual supports
has been forced by displacements Δ (Fig. 9b). Such a model has been
also adopted for numerical studies.
GNA analysis has been employed because in real models buckling
occurs in the elastic state, hence the inclusion of plastic properties of
the material was unnecessary. To validate the assumptions of the anal-
ysis and select the appropriate finite elements, the calculation for cylin-
drical shell model with welding imperfections (of type A) has been
made. The results have been introduced by Rotter [11], and they are
most commonly used during tests on shells' stability. For a finite ele-
ment S4R the obtained results corresponds perfectly to these obtained
by Rotter. It is generally believed that finite elements S4R are well suited
for analysis of stability, that is why they have been chosen for the anal-
ysis described here. A series of calculations have been made for a shell
model with parameters: L1 / L = 0.27, r / t = 500, s0 / 2πr = 0.0191
to optimize the sizes of the finite elements. Obtained results are pre-
sented in Fig. 8. The horizontal axis shows the inverse of the number
of finite elements in the model, which enables interpolation to the accu-
rate result by comparison to the number 0 and not infinity. The vertical
axis shows the result for a given size of the finite element R in compar-
ison with the result obtained from ultimately selected size of the finite
element Rcho. Finally the elements with the length of the side 0.008 m
have been chosen, which is about 0.36(r / t)1/2, allowing a relatively
quick calculation with accuracy to 2.5%.
The first computational model (Fig. 9a) has presented a shell with
parameters: L1 / L = 0.27, r / t = 500, s0 / 2πr = 0.0191. The upper
ring (at the unstiffened side) has been loaded by a uniform linear
force directed vertically downwards, acting on the entire circumference.
Distribution of meridional, compressive stresses and the vertical dis-
placements within the elastic range of the structure's behavior (before
buckling) has been presented in Fig. 10.
The second numerical model (Fig. 9b) has the same geometry as
model 1, but the model 2 has been loaded by a uniform displacement
Δ of the upper edge. Some results of the elastic analysis of model 2
have been introduced in Fig. 11 (results for displacements for which
the sum of support reactions is equal to these obtained for the model
1 presented in Fig. 10).
The third analyzed model (Fig. 9c), having the same parameters as
models 1 and 2 has concerned the reverse situation to that adopted
for the model 2: for all 4 discrete supports the same displacement Δ
has been forced, and the uniform support has been realized on the
 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94 87

Fig. 14. Load–displacement relationships for a model with L1 / L = 0.27, r / t = 500, s0 / 2πr = 0.0191, loaded sequentially at discrete supports 1/4, 2/4, 3/4, 4/4.

ring from the unstiffened side. This model corresponded therefore to
the manner of planned tests. The load–displacement relationship for
all 3 numerical models has been presented in Fig. 12. For models 2
and 3 the obtained results have been the same.
All models have lost the stability in a local manner over short ribs,
although there has been a significant difference in the distribution of
meridional compressive stresses, vertical displacements and resistances
between model 1 and models 2 and 3. It is obviously seen that the
change the way of loading (from uniformly applied forces to displace-
ments) is essential for the numerical analysis. Usually, the load is forced
in this case by displacements produced by rigid disks of the testing de-
vice. In presented tests, the models have been supported at the
unstiffened side by a ring on undeformable base, whereas the controlled
displacements Δ have been forced by a hydraulic press from the other
side of the model, like in Fig. 9c, but for the model inverted by 180°.
The cylindrical silo shells are quite high, so the meridional rigidity of
the adjacent cross sections is practically infinite. That means that for any
horizontal cross-section of the shell, the vertical displacements are the
same, while the stress might change along the shell's perimeter. The
test models correspond to the bottom portion of the entire silo shell,
therefore the uniform meridional displacements at the unstiffened
side of the model's ring is fully correct. The employment of the inversed
model in comparison to the Fig. 9c is pertinent. The next step of the
numerical analysis was the study of a model supported as the model 3
(Fig. 9c), on the ring from the unstiffened side, but for a single support
loaded (Fig. 13). This analysis has revealed a good agreement between
critical reactions of the analyzed model and the model loaded at four
supports — the difference did not exceed 1.2%.
The last step, very important for numerical computations, prior to
joining the experimental investigations has been the entire cycle of

Table 1
Results of tests for shell models from series No. 1.

L1 / r r/t s0 / 2πr Test no. RLIM Vertical weld

[–] [–] [–] [–] [kN] [–]

0.24 500 0.0191 1/4 31.10 Yes

1/3 36.57 No
3/3 33.14 Yes
4/4 33.46 No
0.0006 2/3 31.89 No
2/4 32.94 No
3/4 31.82 Yes
250 0.0191 1/3 130.90 No
2/3 120.83 Yes
3/3 121.45 No
1/4 114.92 No
2/4 109.86 Yes
0.0006 3/4 115.33 No
4/4 102.00 Yes

Table 2
Results of tests for shell models from series No. 2.

L1 / r r/t s0 / 2πr Test no. RLIM Vertical weld

[–] [–] [–] [–] [kN] [–]

0.54 500 0.0191 3/3 36.76 No

1/4 27.38 Yes
3/4 28.06 Yes
4/4 30.26 No
0.0006 1/3 34.30 Yes
2/3 30.62 No
2/4 41.26 No
250 0.0191 1/3 152.86 No
3/3 139.50 Yes
1/4 136.78 No
2/4 107.96 Yes
4/4 120.72 Yes
0.0006 2/3 132.26 No
Fig. 15. Modes of buckling of two models from series No. 1 (r / t = 500, L1 / r = 0.24,
3/4 115.54 No
s0 / 2πr = 0.0191).
88 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94

Table 3 Table 5
Results of tests for shell models from series No. 3. Results of the additional tests of shells.

L1 / r r/t s0 / 2πr Test no. RLIM Vertical weld L2 / r r/t RLIM L2 / r r/t RLIM L2 / r r/t RLIM
[–] [–] [–] [–] [kN] [–] [–] [–] [kN] [–] [–] [kN] [–] [–] [kN]

1.04 500 0.0191 1/3 48.92 No 0 500 20.14 0.54 500 22.64 0.24 500 18.42
2/3 34.28 Yes 20.62 21.64 21.60
1/4 41.30 No 20.16 21.74 22.58
2/4 35.84 Yes 19.94 17.24 1.04 500 26.34
0.0006 3/3 49.10 No 18.70 20.92 21.24
3/4 48.94 No 19.96 250 79.60 250 70.66
4/4 40.64 Yes 250 81.98 74.64 71.76
250 0.0191 1/3 133.60 No 76.14 81.76 73.72
3/3 147.22 Yes 76.50 68.88
1/4 148.34 No 81.16 72.84
2/4 173.34 Yes 75.36 74.62
0.0006 2/3 145.54 No 72.38 74.10
3/4 142.88 No
4/4 136.98 Yes

loading and disloading executed for a single model. The analysis has
been based on the subsequent loading of the single support (until shell's
buckling over this support). After the local buckling of one support
shell's region, the other, stable discrete support, has been loaded. The
local buckling deformations have been treated as model's imperfections.
For 4 analyzed discrete supports, 4 load–displacement relationships
have been obtained (Fig. 14).
A very good agreement has been reached for critical reaction of all 4
discrete supports. The difference between critical reactions in support
2/4 and 3/4 has been 0.9%, whereas the maximum difference (between
support 1/4 and 4/4) has been 1.7%. Hence, it has been considered that
the tests can be carried out similarly.
For long ribs L1 = r there have been cases, when buckling near the
short rib appeared as the first one (Fig. 21b). The models' shells have
been made of two pieces of hot rolled sheet welded along the meridian
(see e.g. Fig. 15a). For models with 3 ribs and 3 discrete supports such a
weld has occurred along the axis of one of the ribs. For models with 4
ribs such weld has occurred in two opposite ribs regions. These places
have been marked in presented results.
The experimental results are shown in Tables 1–5. Each specimen is
described by the number of test (the first digit) and the number of test-
ed model supports (the second digit). For instance “Test No. 2/3” means
that in model with 3 discrete supports the second of them has been
loaded. For one value of parameter L1 / r four models have created a se-
ries of tests: two of them with slenderness r / t = 500 and two other

5. Experimental investigations on cylindrical shells models

During each of the loading tests for subsequent discrete support, the
both support's displacement Δ and the load R have been recorded,
which allows to determine the load-bearing capacity RLIM and the R–Δ
relationship as well. On this basis the maximum load-bearing capacity
and corresponding displacement have been determined. Moreover,
the shape of local buckling has been recorded as well as its localization.
Two different modes of local instability have been observed. The basic
one concerned the unstiffened part of the shell along the extension of
the support ribs (e.g. Fig. 15), the second one appeared usually after
the primary mode of buckling in the unstiffened part between the
support ring and the intermediate ring next to short ribs (Fig. 18b).

Table 4
Results of shells' tests from series No. 4.

s0 / 2πr L1 / r r/t RLIM

[–] [–] [–] [kN]

0.0191 0.24 500 20.48

28.52
250 107.64
107.68
0.54 500 29.82
26.58
250 105.48
111.02
1.04 500 25.22
25.84
Fig. 16. Load–displacement relationships for tested models from series No. 1 (r / t = 500,
250 133.70
L1 / r = 0.24): a) models with support widths s0 / 2πr = 0.0191, b) models with support
140.38
widths s0 / 2πr = 0.0006.
 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94 89

with r / t = 250, whereby for each slenderness r / t one of the models has
had 3 short ribs, and the other one 4 short ribs. Under each short rib, the
discrete support has been arranged.
Tests have been carried out in four series. In the first one (No. 1), the
shells has had ribs satisfying the condition L1 / r = 0.24. In each of the
tests (for models with r / t = 500 and r / t = 250) the loss of stability
has taken the shape of rhomboidal indentation at the unstiffened
region, over the short rib and intermediate ring (Fig. 15). The relevant
results have been shown in Table 1, where RLIM denotes the value of re-
action at the single discrete support with width s0, for which the shell
has lost the stability. In this series of investigations, the loss of stability
has not been noticed in the region between the intermediate ring and
the support ring, even after local buckling at the unstiffened area. The
load–displacement relationship for models of series No. 1 have been
shown in Figs. 16 and 17.
The test series No. 2 has been associated with models with L1 / r =
0.54. For all models the load-bearing capacity has been accompanied
by the local instability of the unstiffened area (Fig. 18a), whereas for
models with r / t = 500, the instability has been additionally recorded
at the stiffened area (between the support ring and the intermediate
ring), already after buckling of the unstiffened zone (Fig. 18b) — this
secondary instability has appeared after the load-bearing capacity of
the model have been reached. The results of tests for models from series
No. 2 have been shown in Table 2 and in Figs. 19 and 20.
The third series of tests (No. 3) has been applied to models with
L1 / r = 1.04. For these models, the instability has been observed at
the unstiffened area and between the rings as well. Both of the modes
appeared almost for the same loads coming from discrete supports. In
shells with r / t = 500 mostly the unstiffened area has lost the stability
(Fig. 21), wherein in one test the rib has buckled (test 3/3 marked as

Fig. 18. Buckling modes for models from series No. 2: a) buckling at the unstiffened shell's
area, b) additional buckling next to the ribs.

Fig. 17. Load–displacement relationship for tested models from series No. 1 (r / t = 250,
L1 / r = 0.24): a) models with support widths s0 / 2πr = 0.0191, b) models with support
widths s0 / 2πr = 0.0006.
bold in Table 3). It has been basically assumed that the ribs are suffi-
ciently rigid and able to resist buckling along the adjacent shell. For
this reason, the ribs have been suitable stiffened in the rest of tests.
This treatment has not affected the critical values of loads RLIM, as only
the ribs buckling length was reduced. For equal terms of comparison
with previous models the enlarging of ribs' cross-section area has
been postponed.
For shells with r / t = 250 in tests 1/3 and 3/3 (marked as bold in
Table 3), the rib has buckled, and the described above treatment has
been employed as for shells with r / t = 500. Obtained load–
displacement relationships have been presented in Figs. 22 and 23.
The fourth (No. 4) series has concerned the comparative models,
where models with short ribs but without an intermediate ring have
been examined. Owing to the merely complementary and comparative
nature of these tests, the range of studies has been limited to a small
number of models with support width s0 / 2πr = 0.0191. The ribs in
the discussed models have been arranged in such a way that the vertical
meridional welds have not been in the vicinity of the load axis. As it has
been expected, the buckling has taken the shape of the rib indentation
inside the shell (Fig. 24). Results from this test series have been present-
ed in Table 4, where the number of each single test corresponds to the
particular load–displacement relationships shown in Fig. 25.
Unstiffened models with support rings, but without longitudinal ribs
have been tested as well (Fig. 26a). These tests of secondary character
should be regarded as indicative only. After the primary studies of
shell models, the strength of applied steel has been investigated on
specimens taken from portions located far from buckling deformations
on tested models. Additional tests have been executed for shells without
90 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94
Fig. 19. Load–displacement relationship for tested models from series No. 2 (r / t = 500,
L1 / r = 0.54): a) models with support width s0 / 2πr = 0.0191, b) models with support
width s0 / 2πr = 0.0006.
longitudinal ribs, but with an intermediate ring, placed at distance L2
from the support ring (Fig. 26b).
6. Summary and conclusions
The study has covered discretely supported shells (Fig. 2) of steel silos.
Models with short ribs topped with intermediate ring (e.g. Fig. 27c) de-
scribed in the summary as “type A”, and with short ribs without an inter-
mediate ring (e.g. Fig. 27b) described in the summary as “type B” have
been predominantly studied. Additional studies have been carried out
for discretely supported shells without ribs (e.g. Fig. 27a) as well. Two
support widths have been analyzed s0 = 0.0006πD (support type I) and
s0 = 0.0191πD (support type II) with different lengths of the longitudinal
ribs within the range: 0.24 ≤ L1 / L ≤ 1.04. Abbreviated description of the
models is used in charts (Figs. 28 and 29): type A with supports of type II
is described as “type A/II”, other models are described in a similar manner.
In summary charts (Figs. 28 and 29) results of the individual tests are
shown by marked points, whereas the averaged values (for one set of
parameters) are depicted by a continuous line. Point and line chart for
one model type are described in the above explained manner. As ex-
pected, the use of vertical ribs of L1 length influence the load-bearing ca-
pacity RLIM of shells loaded on discrete supports with meridional force R.
The increase of the length for ribs L1 and the use of intermediate ring
over the ribs clearly affects the growth of capacity RLIM of axially com-
pressed shells supported discretely (Figs. 28, 29).
The authors of presented herein study believe that the most impor-
tant conclusion from presented experimental tests is the negation of the
usually formulated opinion that the use of longitudinal ribs of consider-
able length (L1 / r ≥ 0.5) topped with intermediate ring in axially com-
pressed cylindrical shells protects the unstiffened section, localized
above the intermediate ring against the local instability. Presented
Fig. 20. Load–displacement relationship for tested models from series No. 2 (r / t = 250,
L1 / r = 0.54): a) models with support width s0 / 2πr = 0.0191, b) models with support
width s0 / 2πr = 0.0006.
experimental studies with simplified numerical analyses clearly show
that the part of the shell above the intermediate ring buckles over ver-
tical ribs even if used ribs are of considerable length (L1 / r ≥ 0.5). The
strengthening of zones adjacent to supports of short ribs, and interme-
diate rings increase obviously the buckling capacity of the shell, but the
differences in the capacities of uniformly supported shell around its
circumference and shells supported on discrete supports are always
very high. Experimental tests of shells supported on entire circumfer-
ence have not been carried out, so GNA numerical analysis has been
performed. In this calculation shell model with parameters: r / t =
500, s0 = 0.0006πD and the number of supports n = 4 has been ana-
lyzed for all four cases of geometry and support way shown in Fig. 27.
The analyses have enabled to obtain a critical load for a perfect elastic
shell. Resultant axial force for shell supported on entire circumference
(Fig. 27d) is labeled Po. For shells supported on n = 4 discrete supports
(Fig. 27a–c) the resultant critical force was determined as P = 4 × RLIM.
Perfect shells have been analyzed, but it is expected that planned by au-
thors GMNIA analysis of those models will lead to similar differences be-
tween global capacity P and Po as in GNA analysis.
The results from numerical analysis of the perfect shell models
(Fig. 30), clearly show that the use of vertical ribs in the support zone
of shells increases the global load capacity P (line b), in relation to
the unstiffened shell (line a). The use of an intermediate ring over
short ribs leads to additional increase in load capacity P (line c) and im-
proves the effectiveness of longer ribs. It should be strongly emphasized
that there is a very big difference between the global load-bearing
capacities of shells without ribs evenly supported around the cir-
cumference (Fig. 30 — line d), and global capacities P = n × RLIM of par-
tially ribbed shells based on discrete supports (Fig. 30 — lines a, b, c).
These differences are related to the phenomenon described in [9], as a
strong concentration of meridional stress σx above the discrete supports
 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94 91

Fig. 21. Modes of buckling for models from series No. 3 (r / t = 500, L1 / r = 1.04): a) s0 / 2πr = 0.0191, b) s0 / 2πr = 0.0006.

in axially compressed shells without longitudinal ribs (Fig. 3b). The lon-
gitudinal ribs are also conducive to the meridional stress concentration
(Figs. 10a, 11a), and in the axes of these ribs above intermediate ring the
maximum meridian stress σx,max is only slightly smaller than the corre-
sponding stress σx,max over the support in the shells without longitudi-
nal ribs (for the same support width). This issue is often overlooked in
the design of silos, leading to incorrect assessment of ribbed cylindrical
shells capacity when the traditional and simplified methods of design
are applied. For a discretely supported cylindrical silo shell to be de-
signed according to procedures given for uniformly supported shells
in codes [8,10] the obtained resistance would be too small even when
the strengthening (ribs and circumferential ring) is present. The total

Fig. 22. Load–displacement relationship for tested models from series No. 3 (r / t = 500, Fig. 23. Load displacement relationship for tested models from series No. 3 (r / t = 250,
L1 / r = 1.04): a) models with support width s0 / 2πr = 0.0191, b) models with support L1 / r = 1.04): a) models with support width s0 / 2πr = 0.0191, b) models with support
width s0 / 2πr = 0.0006. width s0 / 2πr = 0.0006.
92 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94

Fig. 24. Shapes of buckling for models from series No. 4 (r / t = 500, s0 / 2πr = 0.0191): a) L1 / r = 1.04, b) L1 / r = 0.54.

resistance of all supports P0 (lines a, b, c in Fig. 30) is always definitely
smaller than the analogical capacity P0 of the same shell supported
along the entire circumference. The usage of solution analyzed in
this paper (line c in Fig. 30) increases significantly the capacity P0 of dis-
cretely supported shell (in this case of 4 supports) in comparison to the
same shell but without ribs and circumferential ring (line a in Fig. 30),
anyhow the resulting resistance is always much smaller than the one
determined for a uniformly supported shell. The use of numerical

Fig. 25. Load–displacement relationship for models from series No. 4 (s0 / 2πr = 0.0191, Fig. 26. Buckling modes for additional tests: a) r / t = 250, s0 / 2πr = 0.0006, L2 / r = 0,
L1 / r according to Table 4): a) r / t = 500, b) r / t = 250. b) r / t = 500, s0 / 2πr = 0.0191, L2 / r = 0.54.
 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94 93

Fig. 27. Examples of analyzed shells with four discrete supports (a–c) and evenly supported shell (d).

analysis GMNIA according to [10] should provide a safe evaluation of the
analytical values of capacities for steel silo shells.
The positive influence of material horizontal pressure on silo shell on
global capacity of the shell due to stability failure is widely known. Local
unsymmetrical horizontal pressure that may appear in shells is also well
recognized [12]. This unsymmetrical pressure can appear randomly in
any section of the shell and in some cases leads to local disappearance
of material horizontal pressure. Since in case of the local buckling of dis-
cretely supported shells (above supports or ribs; Figs. 4, 26), it cannot be
excluded that disappearance of the horizontal pressure will take place

Fig. 28. Summary of results of experimental shell models of type A.

Fig. 29. Summary of results of experimental shell models of type A and type B.
94 E. Hotała, Ł Skotny / Journal of Constructional Steel Research 96 (2014) 81–94
Fig. 30. Comparison of global capacity P0 of perfect shell supported on entire circumfer-
ence (d) with capacities P = 4RLIM of shells supported on discrete supports: without ribs
(a) and with ribs (b and c) — marking according to Fig. 27.
in the region where buckling wave is forming. In authors opinion in case
of discretely supported shells positive influence of horizontal bulk ma-
terial pressure on shell capacity due to local stability failure should not
be taken into account. However if the constant horizontal pressure
would be provided in the axially compressed shell supported discretely,
shells global capacity P0 would increase in a similar degree as capacity of
an analogical shell supported uniformly. It can be assumed that Fig. 30
would have a similar course but values of global capacity P0 would be
proportionally higher. Capacity of shells due to buckling with regard
to the normal pressure can be calculated according to standards [8,10].
Acknowledgments
Calculations have been carried out in Wroclaw Centre for Network-
ing and Supercomputing (http://www.wcss.wroc.pl), grant No. 241.
Experimental and numerical studies presented in this paper
were made for PhD thesis of Łukasz Skotny “Nośność graniczna
użebrowanych cylindrycznych powłok płaszczy stalowych w strefie
podpór odcinkowych” (in Polish), Prof. Hotała was a supervisor of this
dissertation. Several figures and tables presented in this paper were
taken from this thesis or are based on figures presented in it.
References
[1] Hotała E. Stability of cylindrical steel shell under local axial compression, European
Workshop Thin-walled Steel Structures 26–27 Sept. 1996, Krzyżowa, Poland. In:
Rykaluk K, Pasternak H, editors. 1996. p. 61–8.
[2] Hotała E. Buckling of cylindrical steel shell under local meridional load. Proc. 9th Nordic
Steel Construction Conference, Helsinki (Finland), 18–20 June 2001; 2001. p. 213–20.
[3] Knödel P, Ummenhofer T. Ein Einfaches Modell zum Stabilitätsnachweis zylindrischer
Schalentragwerke auf Einzelstűtzen. Der Stahlbau, 6; 1998. p. 425–9.
[4] Pasternak H, Hotala E. Schäden an Stahlsilos - Ursache und Beispiele. Bauingenieur
1996;71:223–8.
[5] Komann S. Stabilität von diskret gestűtzen, axialbelasteten, dűnnwandigen
Kreiszylinderschalen aus Stahl, BTU Cottbus, Schriftenreicht Stahlbau, Heft 4/2005;
2005.
[6] Pasternak H, Komann S. About the role of longitudinal stiffeners in silos. Xth Interna-
tional Scientific — Technical Conference “Metal Structures — Gdansk 2001”; 2001.
p. 161–8.
[7] Pasternak H, Komann S. Diskret gelagerte, axialbelastete und über den Auflagern
längsversteifte Kreiszylinderschalen. Bauingenieur 2004;79(10):S. 443–6.
[8] EN 1993-4-1. Eurocode 3: Design of steel structures. Part 4.1: Silos, Eurocode 3 Part
4.1, CEN, Brussels; 2007.
[9] Samuelson LÅ, Eggwertz S. Shell stability handbook. London: Elsevier; 1992.
[10] EN 1993-1-6. Eurocode 3: Design of steel structures. Part 1.6: General rules –
Strength and stability of shell structures, Eurocode 3 Part 1.6, CEN, Brussels; 2007.
[11] Rotter JM, Teng JG. Elastic stability of cylindrical shells with weld depressions.
J Struct Eng 1989;5:1244–53.
[12] EN 1991-4. Eurocode 1: Actions on structures. Part 4: Actions on silos and tanks,
Eurocode 1 Part 4, CEN, Brussels; 2006.