Thin-Walled Structures 84 (2014) 226–240

Contents lists available at ScienceDirect

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

Wind buckling of tanks with conical roof considering shielding

by another tank
Carlos A. Burgos a, Rossana C. Jaca a, Jorge L. Lassig a, Luis A. Godoy b,c,n
a
Engineering School, Universidad Nacional del Comahue, Neuquén, Argentina
b
FCEFyN, Universidad Nacional de Córdoba, Córdoba, Argentina
c
Science and Technology Research Council (CONICET), Argentina

ar t ic l e i nf o a b s t r a c t

Article history: Oil storage tanks are usually arranged in groups in tank farms, and this configuration may affect their
Received 19 April 2014 buckling and postbuckling strength under wind loads. The assessment of wind action on tank structures
Received in revised form is performed in this work by means of wind tunnel experiments to evaluate the pattern of pressure
10 June 2014
distribution for a tank which is shielded by another tank under various configurations and separation
Accepted 18 June 2014
between them. The experimental results show significant changes in pressures due to shielding effects.
Available online 26 July 2014
In a second stage the structural response under the pressures previously evaluated is performed by finite
Keywords: element analysis using both linear bifurcation and geometrically nonlinear analysis. Results of two-tank
Buckling interaction are compared with those of an isolated tank. Based on the results, it is concluded that the
Group effects
changes in wind pressures due to group effects induce changes in buckling loads and in the associated
Shells
deflected patterns.
Tanks
Wind loads & 2014 Elsevier Ltd. All rights reserved.

1. Introduction This is evidenced in the American [2] and European [5] recom-
mendations for the design of aboveground tanks, in which only
Short steel tanks are usually employed in the oil industry to the behavior of isolated shells is considered in detail.
store large volumes of fluid, with aspect ratios between 0.2 oH/ The first investigation on the interaction between neighboring
Do 1.0 in which H is the height of the cylindrical part and D is the cylinders with a roof was perhaps published by Esslinger et al. [4]
diameter; however, frequently employed dimensions have aspect in Germany, in which wind tunnel tests were reported on two
ratios between 0.2 and 0.6. Tanks may have an internal floating small-scale silos with similar dimensions. The models were placed
roof and a fixed roof (either conical, flat, or dome roof). Some tanks in a line with the wind direction, with dimensions which were
do not have a fixed roof so that the floating roof is directly exposed representative of tall silos with H/D 42. For even taller structures,
to the environment. with H/D4 10, Zdravkovich [28] and Tsutsui et al. [22] studied the
Oil tanks are frequently constructed in groups in what are interaction between two aligned cylinders. More recent studies
known as tank farms. Farms include between tens and hundreds of concerning pressures in shells which are localized close to each
tanks, which may be the property of one or several oil companies. other under wind were published by Gu and Sun [6] and Orlando
Under strong winds, the structural behavior of tanks depends on [12]. However, such studies are not relevant to explain interaction
their location within the group, so that it may be possible to effects between oil storage tanks, which are short cylinders with
distinguish between tanks located in a front line from those placed relative dimensions in the order of 0.25o H/D o0.5.
in a second or third line with respect to the perimeter of the A wind tunnel investigation on silos placed very close to each
facility. other in a line perpendicular to the direction of wind was carried
Because tank farms are so common in oil facilities, it is out in Australia in the 1970s by Vickery and Ansourian [24]. The
surprising to find that most available information on the wind results have been reported in the form of an analytical expression
response of tanks concentrates on isolated tanks in flat terrain. for pressure coefficients around the circumference in the European
recommendations [18], but without reference to the dimensions
and separation between the shells.
n
Corresponding author at: FCEFyN, Universidad Nacional de Córdoba, Córdoba,
Wind tunnel studies have been made of the external pressure
Argentina. distributions on multiple circular cylinders with conical or flat
E-mail address: [email protected] (L.A. Godoy). roofs. Regarding wind tunnel studies of tanks, MacDonald et al.

http://dx.doi.org/10.1016/j.tws.2014.06.007
0263-8231/& 2014 Elsevier Ltd. All rights reserved.
 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240 227

[10] concluded that pressure distributions are independent of reductions were obtained for larger values of S. Changes were
Reynolds number provided Re41
104. Sabransky and Melbourne found not only in pressure values but also in pressure distribu-
[19] studied silo structures with aspect ratios H/D ¼0.66 and tions. Reductions in pressures on the internal walls were also
conical roof inclination angle of 271. reported in the windward region. The results highlight the
MacDonald et al. [11] performed wind tunnel testing of five expected changes in pressures for open tanks, but the results
tanks in a line in which the blocking and the target tank models cannot be directly employed for tanks having a fixed roof.
had both a flat roof; however, only those in tandem configuration The diversity of configurations which may be found in tank
are reviewed here because they can be compared with present farms, even for tanks having similar dimensions and spaced in a
results. For point pressure measurements, a configuration with regular pattern, points to the need to have more information on
S ¼0.125D was tested, where S is the wall-to-wall minimum pressure coefficients and on the structural response to such wind.
distance between tanks. Mean value pressures showed two lobes This work addresses the problem of a tank with conical roof which is
of positive pressures centered with respect to the windward obstructed by another one having the same geometry, in which the
meridian, each with a central angle of approximately 501. The angle of wind incidence is taken as a variable to investigate several
positive pressures resulted in values significantly lower than in the group configurations. Wind tunnel studies are performed to obtain
isolated tank, with pressure coefficients Cp o0.5. Peak suctions pressure coefficients, which are subsequently employed in a finite
(located at 901 from the windward meridian) were also smaller, element analysis of shell buckling. Two approaches of shell buckling
with Cp o1.0. The single case investigated does not allow under- are investigated, namely linear bifurcation analysis (LBA) and
standing effects due to tank separation. Panel measurements, on geometrically nonlinear analysis with imperfections (GNIA).
the other hand, were studied for configurations at S¼0.125D,
0.25D, and 0.5D. The results for S¼ 0.5D are shown in the paper.
Tanks located in a second line with respect to the periphery of a 2. Wind tunnel experiments
tank farm, in which the blocking tank had a flat roof and the target
tank had a conical roof, were studied by Portela and Godoy [13] 2.1. Main features of the wind tunnel facility
based on wind tunnel tests. Six configurations were tested, with
changes in the separation between tanks (S ¼0.5D and S¼ 1.0D) The wind tunnel facility at the National University of La Plata,
and in the relative height between blocking and target tank. Other which is the largest facility in its kind in Argentina, has been used
cases reported include two tanks in the first line blocking the flow in this research. The tunnel is capable of reproducing an atmo-
of a tank in a second line, with separations S¼ 0.5D and S¼1.0D. spheric boundary layer, in which high turbulence may be gener-
Contours of pressure coefficient were presented and subsequently ated together with a non-uniform wind velocity in elevation.
employed to carry out linear bifurcation analysis (LBA) and Fig. 1 shows the main components of this close-circuit wind
geometrically nonlinear analysis (GNA) on the tanks for which tunnel, with a cross section having 1.40 m (width), 1.0 m (height),
measurements were taken, always in the second line with respect and 7.5 m in length. The fan has six blades and is moved by a
to the periphery of the tank farm; results were compared with 50 HP engine. The engine has a system of velocity control which
those obtained for an isolated tank [14]. Case studies concerning allows changing the flow velocity up to a maximum of 20 m/s
six tanks in a small plant were investigated in Ref. [15] in wind measured at the center of the cross section. The access door with
tunnel to obtain pressure coefficients for one target tank under glass panels to visualize the development of the test is shown in
various wind directions; LBA buckling and GNA post-buckling Fig. 1a, whereas the section where testing is done is shown in
were next computed. Iamandi et al. [7] performed wind tunnel Fig. 1b.
testing of a four-tank configuration due to an accident in a small Air flows through a honeycomb to enforce axial symmetry and
chemical storage station in Romania but did not provide pressure through a set of horizontal obstacles (shown in Fig. 1b) which can
coefficients. rotate on their axes to generate turbulence. Changes in turbulence
Tall cylinders (H/D ¼2.56) with flat roof in tandem arrays were are obtained by means of variations of the relative location of the
studied by Said et al. [20] by means of wind tunnel tests and finite obstacles with respect to the wind direction. Once the desired
volume simulations. The flow pattern was found to be highly turbulence has been obtained, roughness is modeled by small
dependent on the separation S between both cylinders: for the parallelepiped blocks attached to the floor.
short separation S ¼1.28D, the flow accelerates on the roof of The mechanism employed can represent mean velocities in
the first cylinder and impacts on the top part of the second elevation that follow power or logarithmic laws, depending on the
cylinder, while increasing the pressure. The wake of the first needs of the study; in our case the applied power law was adopted
cylinder modifies the pressure field on the target tank and reduces with an exponent P ¼0.32. Different types of turbulence may be
the pressures on the windward region. This effect decreases as the implemented in the lower atmospheric boundary layer. During
distance increases to S¼ 5.12D, with the consequence that the testing, the turbulence intensity was 0.15.
target tank becomes subjected to a flow pattern that is similar to The tunnel is equipped with a system of NetScanner electronic
that in the isolated tank. pressure sensors with 128 channels, in which pressures are
Uematsu and coworkers [23,25] reported wind tunnel results recorded. A computer is connected for the acquisition and proces-
on open top tanks to investigate group effects in arrangements of sing of experimental data. A hot wire anemometer with telescopic
two, three and four tanks. The tanks had the same geometry with arm is employed to record reference velocity and temperature of
aspect ratios of H/D ¼0.25D, 0.5D, and 1.0D and spacings of flowing air.
0.125Do So1.0D. Zhao et al. [27] were also interested in large
open-topped tanks with low aspect ratio (H/D¼0.275), and 2.2. Prototype tanks considered
performed a comprehensive wind tunnel study considering two,
three, and four interacting tanks, all of which were instrumented. A specific geometry was chosen as a case study in this research,
Two tanks of identical geometry in tandem configuration were having H/D ¼0.52; in the prototype, the dimensions are
tested at S¼ 0.5D, 1.0D, and 1.5D. Pressures on the external wall of D¼ 30.48 m and H ¼15.75 m.
the second tank showed large changes, with peak positive pres- The separation S between tanks in a tank farm is an important
sures in the windward region for S ¼0.5D being reduced to 0.24 of parameter in the present study. Because of limitations in available
their values in the isolated tank; whereas less significant space in an oil facility, there is a trend to locate them as close as
228 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240

Fig. 1. Wind tunnel at the National University of La Plata in Argentina: (a) external view and (b) internal view of the section where tests are conducted.

Fig. 2. Groups of tanks in two tank farms located on the Northern part of Patagonia, Argentina.

possible; on the other hand, typical regulations establish a mini- is the cross sectional area of 1.4 m2. A value of 5% was adopted as
mum distance between tanks equal to the diameter of the largest maximum relation between model area and cross sectional area to
tank, measured from wall to wall. There are also limitations avoid blockage of the flow, as recommended by ASCE [3].
regarding the distance to public roads and to other installations. The models were fabricated using a 200 mm diameter PVC
A survey of aerial photographs of tank farms located in the tube, thus having a prototype-model relation of Lr ¼Dprototype/
northern region of Patagonia in Argentina [8] shows that typical Dmodel ¼152.4. Other dimensions in the model are calculated based
separations S between tanks are between 1.0D and 1.8D, as on Lr, and the dimensions adopted are shown in Table 1. Blocking
illustrated in Fig. 2. In general terms, tanks are placed in pairs, of the cross section of the tunnel for this model is of only 3.22%,
even though there may be more tanks in the plant. It was decided which is smaller than recommended values.
to perform wind tunnel testing considering two tanks, in which For a reference wind velocity of 17 m/s in the tunnel at
one is the target tank, which is instrumented to obtain pressure elevation 1.74H, the corresponding velocity at 0.92H was
coefficients, and the other one (which has not been instrumented 12.63 m/s. The elevation at 0.92H is close to the top of the cylinder
in this research) is placed to block the wind flow. Based on and was chosen to identify reference velocities used to evaluate
observations, the distances between tanks were adopted at pressure coefficients. This procedure is similar to what was
S ¼1.0D, S ¼1.5D, and S¼ 1.8D. adopted, for example, by MacDonald et al. [10].
The tank tested in the experiments is shown in Fig. 3. Pressure
2.3. Experimental model gauges were placed at three elevations in the cylinder (at 0.10H,
0.50H, 0.90H), plus three in the roof (at 1.02H, 1.09H, and 1.15H)
The dimensions of the small scale models to be tested in the and one at the top of the conical roof. In the circumferential
wind tunnel were chosen by similitude considerations with the direction the gauges were located at 22.51 spacing, with a total of
prototype discussed previously, and taking into account the main 97 points at which pressures were measured.
features of the wind tunnel facility in which they were to be There are two models in the experiments, both with the same
tested. To this effect, the most important aspect of the wind tunnel characteristics, but a bottom plate was present in the blocking
 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240 229

tank whereas no plate was placed at the bottom of the tank in flow velocity; and (p  p1) is the difference between the pressure
which measurements were taken, because all pressure gauges measured in the model and that of the unperturbed flow. The data
were conducted on the inside of such model. acquisition system employed records the term (p p1) directly in
units of pressure, with a frequency of data acquisition of 3.5 Hz.
2.4. Testing procedure The unperturbed density recorded was 1.22 kg/m3.

To simulate shielding and interaction between tanks in a tank

farm, two variables were considered: the distance S between tanks
and the direction of wind incidence. Separations between tanks 3. Wind tunnel results
S ¼1.0D, S ¼1.5D, and S¼ 1.8D were investigated, with directions
characterized by angles ranging between 01 and 901. The target 3.1. Wind pressures
model was tested in the same position, and the position of the
blocking tank was changed to obtain different configurations. Pressure contours evaluated from the tests are shown in three-
Measurements were made at a wind speed of 17 m/s, and pressure dimensional representations in Fig. 5 for the five configurations
and temperature were recorded in each case to adjust the values. investigated and for the isolated tank. Positive values indicate
The velocity inside the wind tunnel was measured using an pressures acting on the surface, while negative values indicate
anemometer at 1.27H from the surface, which is equivalent to suction.
20 m elevation in a real scale. For the tandem array C1, the peak pressures on the cylinder are
For separation between tanks S ¼1.0D, for each direction of displaced towards the roof junction and with values of Cp ¼1.15
wind incidence, the flow was stabilized and pressures were which are higher than in the single tank configuration. Pressures
measured during one minute. Five directions were considered, as on the roof are predominantly suction, with values of Cp ¼  1.4 in
follows: In Configuration C1, known as tandem array, both tanks the isolated tank and Cp ¼  1.6 in Configuration C1. There are low
are aligned with the direction of wind, so that maximum blockage suction values in C1 at the junction with the cylinder. In the other
occurs. Configurations C2, C3, C4 and C5 were tested with the Configurations C2–C5, peak pressures occur at the center of the
blocking tank displaced at angles of 22.51, 451, 67.51, and 901, cylinder (close to 0.5H) with maximum values that are lower than
measured counter-clockwise with respect to the direction of wind in C1.
incidence. In Configuration C5 the two tanks are perpendicular Because it is difficult to appreciate details and differences in a
to the wind direction to represent tanks located at a first line. three-dimensional representation, more detailed two dimensional
A configuration with just an isolated tank, named C0 has also been plots are discussed next. Plots of pressure coefficients are sum-
studied for reference. Testing for separations S ¼1.5D and S ¼1.8D marized for the cylinder at elevations 0.50H and 0.90H, consider-
was similar to the above description. Fig. 4 illustrates the scheme ing all five configurations at S ¼D in Fig. 6, together with results for
of tests with directions of incidence. the isolated tank. The results shown are not perfectly symmetric
Pressure coefficients Cp are defined in the usual way because of experimental deviations, but no averaging has been
p  p1 made in order to present the original results.
Cp ¼ ð1Þ Pressures on the roof are plotted in Fig. 7 for the intersection of
ð1=2Þρ1 V 21
the roof with a plane passing through the windward meridian.
where ρ1 is the fluid density of the unperturbed flow, V1 is the Pressures are negative with the exception of small zones with low
positive pressures at leeward.
Table 1 Pressure distributions in elevation are represented in Fig. 8 at
Geometries of prototype and model.
three meridians: windward meridian; the meridian facing the
Prototype Model blocking tank in the windward region and the meridian at 1801
from the previous one.
Diameter Cylinder Roof Diameter Cylinder Roof In most configurations the Cp values are higher than those in
D (m) height H (m) height h D0 (cm) height H0 (cm) height h0
the isolated tank, considering both positive and negative pres-
(m) (cm)
sures. The most severe changes are detected for the tandem
30.48 15.75 2.86 20 10.33 1.88 configuration C1, both in pressure patterns and values. For this
configuration the flow passing the first tank produces stagnation

Fig. 3. Model tested with pressure gauges.

230 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240

Fig. 4. (a) Test with a tank shielding a target tank and (b) configurations tested.

pressures on the target tank at an elevation closer to the joint with elevation (closer to the junction between the cylinder and the
the roof. This effect was also observed in Ref. [13]. roof) than in the isolated tank. This is a change in the flow
As the blocking tank is displaced to one side in Configurations conditions between the isolated tank (C0) and the tandem config-
C2–C4 the pressure pattern resembles the pattern in the single uration (C1), because in the former (Fig. 11b) the stream lines affect
tank. In the leeward region, pressures are not so significantly the central region at windward.
modified, except for Configuration C1 in which they are smaller. This effect is not the same as what has been previously
The results in Fig. 8c indicate that on the meridian opposite to observed by other authors in tandem configurations in which
the blocking tank, the pattern of pressures are similar to the the first tank has a fixed roof. Results by Portela and Godoy [13] for
isolated tank but with higher values of suction. a configuration with S ¼1.0D but in which the first tank has a flat
For larger separations S ¼1.5D and S ¼1.8D, the pressures are roof indicate that the flow passing the first tank tends to generate
similar between them and approach the case with S ¼1.0D, but stream lines which follow the direction of the flat roof without
with lower positive and negative pressure coefficients. This is inducing a descent between the two tanks. The first tank in this
shown in Fig. 9 at elevation 0.90H (close to the joint with the roof), case shields the flow to the second tank, thus reducing the
in which case the group effect is still different from the isolated pressures on the target tank. Comparisons with tests by Sabransky
configuration. and Melbourne [19] are not meaningful because those authors
The three-dimensional pressures for tandem (C1) and parallel considered two tanks at very short spacing (S ¼0.25D), in which
(C5) configurations have been plotted in Fig. 10 for S ¼1.8D. If the case the nature of the flow is completely different.
target tank is located further away from the blocking tank, then Although the effect shown in Fig. 11 has not been shown before
pressures decrease between 20% and 5% with respect to S ¼1.0D. in the context of oil storage tanks, this change in the position of
Positive pressures increase by at least 15% and negative pressures stream lines and flow acceleration is commonly observed when
by 25% with respect to the single tank. flow passes a mountain and induces a wave on the lee of the
mountain [21]; depending on the downstream location of a second
3.2. Flow visualization and comparison with other studies object there may be an increase in pressures, as found for some
configurations studied in the present research.
As mentioned before, the most relevant changes occur for the
tandem configuration C1, with pressures being higher than in
the isolated tank and affecting the tank in the upper zone of the 4. Bifurcation buckling of tanks under wind
cylinder, where thicknesses are smaller.
To visualize the flow pattern, smoke was used in some wind 4.1. Finite element model of the target tank
tunnel tests for Configuration C1 at S¼ 1.0D. A typical photograph
depicting the flow for a Configuration C1 at S¼1.0D is shown in The stability analysis of shells has been more or less standar-
Fig. 11a, whereas the flow for the isolated tank is represented in dized thanks to the European Recommendations of Ref. [18]. Based
Fig. 11b. on the general purpose finite element code ABAQUS [1], the
Because of the conical shape of the roof of the first tank, the structural response of the tanks has been computed in this work
stream lines become attached to the roof and then follow an using bifurcation analysis (LBA) and geometrically nonlinear
initially descendent path between the two adjacent tanks. This is analysis with imperfections, GNIA.
shown in the smoke tests in Fig. 11a by means of dotted lines The cylindrical shell was discretized by use of 8-node doubly-
which approximate the stream lines. As the flow passes the first curved elements identified as S8R5, with reduced integration,
tank, there is a wake which is characterized by low pressure and a whereas 6-node triangular elements with five degrees of freedom
vortex being generated which produces a change in the stream per node (element STRI65) were used for the roof. The mesh
lines that arrive at the second tank. For a separation S¼ 1.0D, the (some 10,000 elements) was defined by means of convergence
stream lines elevate as they approach the target tank. studies.
The vortex modifies the pressures on the windward region of The tank was designed according to API 650 [2] requirements,
the target tank, with a shift in maximum pressures to a higher considering ASTM A-36 steel (E ¼211 GPa, ν¼0.3). Values of the
 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240 231

Fig. 5. Pressure coefficients for tanks with S ¼1.0D: (a) Configuration 1; (b) Configuration 2; (c) Configuration 3, (d) Configuration 4, (e) Configuration 5 and (f) isolated tank.

resulting thicknesses are shown in Table 2. The roof is assumed as equivalence between the stiffness of a roof with rafters supporting
a self-supported shell, but with a larger equivalent thickness, equal a thin shell and a thicker shell with uniform thickness and without
to three times the thickness in the thinnest course. This simplifi- rafters. Because relative values are of interest in this work to
cation has been used in a number of investigations based on compare isolated tanks with two tank configurations, using the
232 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240

Fig. 6. Pressure coefficients around the circumference: (a) Elevation 2 (0.5H) and (b) Elevation 3 (0.9H).

Fig. 7. Pressure coefficients on the roof.

 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240 233

Fig. 8. Pressure coefficients in elevation: (a) windward meridian; (b) normal to wind direction, facing the blocking tank and (c) normal to wind direction, on the side
opposite to the blocking tank.

Fig. 9. Pressure coefficients around the circumference, at elevation 0.90H.

same simplification in both cases does not seem to produce great are obtained in pairs, due to a shift in the wave pattern. The
harm on the conclusions. eigenmode for C1 and S ¼1.0D is shown in Fig. 12b, with a shape
which is very similar to what is computed for an isolated tank. The
4.2. Linear bifurcation analysis, LBA mode deflections are restricted to the windward region in the
target tank and the rest of the tank remains unaffected at
Wind pressure coefficients measured in the tests were bifurcation buckling. The main difference is that the buckled
employed as the basic pressure patterns and were scaled by a region is displaced toward the top of the tank, where maximum
load factor λ in the LBA strategy to identify classical critical values pressures were measured in the wind tunnel tests.
λc and their associated eigenmodes. As shown in the literature In the isolated tank, the lowest eigenvalue was λc ¼2.38 kPa,
[26], buckling of the shell is entirely dependent on the region of while λc ¼ 1.65 kPa was computed in the most significant group
positive pressures at windward, and is only marginally affected by effect (Configuration C1, S¼1.0D), leading to a reduction of 30%.
suction at other locations around the circumference. Bifurcation results are only an approximation to the buckling
A summary of LBA results is shown in Fig. 12a considering the behavior of a shell and more refined GNIA studies are reported in
lowest five eigenvalues in all studied configurations. Eigenvalues the next section.
234 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240

Fig. 10. Pressure coefficients for tanks at S¼ 1.8D: (a) Configuration C1 and (b) Configuration C5.

Fig. 11. Flow visualization: (a) tandem Configuration C1 at S¼ 1.0D and (b) isolated tank.

5. Geometrically nonlinear analysis with imperfections, GNIA For imperfection amplitudes ξ ot there is a clear maximum
load in the equilibrium path, but for ξ ¼t in most cases the
5.1. Imperfections in the form of the lowest eigenvalue maximum is lost and the problem becomes one of large displace-
ments rather than buckling. As expected, the most severe drop
The same finite element models as in LBA were used in the GNIA occurs for Configuration C1, even for larger values of S. Differences
studies, in this case following the nonlinear equilibrium path by between C5 and the isolated tank are not so significant.
means of an algorithm due to Riks [16,17]. Eigenmode-affine Curves of imperfection-sensitivity (maximum load in an equili-
imperfections (i.e., geometric imperfections having the same shape brium path vs. imperfection amplitude) are shown in Fig. 14: all curves
as the eigenmodes in LBA) were used. The amplitude ξ of the have similar trends but at different values of maximum loads. Thus,
imperfection was assumed as ξ¼ 0.5t, ξ¼0.75t, and ξ¼ 1.0t, where t the sensitivity slope is seen to be the same as in the isolated case C0.
is the thinner course thickness (top course in the cylindrical shell). A summary of results for LBA and GNIA studies (for ξ¼0.75t) is
Equilibrium paths for the isolated tank and for Configurations given in Table 3 for all configurations considered. The ratio
C1, C2, and C5, and for C1 at S ¼1.5D and S¼ 1.8D, have been plotted between GNIA and LBA indicates that results with imperfections
in Fig. 13. The overall shape of the equilibrium path is the same in are lower than bifurcation loads. The most severe reductions in
all cases, but the results are different depending on the config- LBA critical loads occur for the Configuration C1, in which the tanks
uration and imperfection amplitude considered. are aligned with the direction of wind, with the reduction being
 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240 235

less pronounced as the distance S between tanks increases. A however, for large imperfection amplitudes there is no certainty
similar trend was found for GNIA studies. that higher eigenmodes do not produce ever more stringent
reductions in maximum loads.
To evaluate effects due to the shape of imperfections, the first
100 eigenmodes have been considered and results are discussed
5.2. Influence of imperfection shape on GNIA results
in this section for isolated as well as for two tank configura-
tions. Eigenmodes for the isolated tank are shown in Fig. 15: the
Eigenmode-affine imperfections investigated in the previous
largest displacements in mode 1 (M1) occur in the cylinder at
section were limited to shapes associated with the lowest eigen-
the windward region; most lower modes are similar with varia-
value in LBA study. This shape is the most detrimental imperfec-
tions in the number of circumferential waves and their extent.
tion in the vicinity of the critical state (as shown by Koiter [9]);
Fig. 15b shows the first mode with significant roof deflections,
M15.
Table 2
Thickness of the cylindrical shell. For two-tank Configuration C1, the modes are shown in Fig. 16
for S ¼1.0D. The lowest modes (Fig. 16a, b, and e) have displace-
Course Elevation (m) Thickness ments in the windward region; a roof mode is shown in Fig. 16c,
whereas mode M23 (Fig. 16d) has displacements in the leeward
(m) (in.)
region.
1 1.50 0.0191 3/4 To evaluate equilibrium paths, imperfection amplitudes
2 3.00 0.0159 5/8 ξ¼ 0.75t were considered because this is the imperfection for
3 4.50 0.0143 9/16 which the most severe reductions in maximum loads are obtained
4 6.00 0.0127 1/2
before a maximum is lost. Results for the isolated tank are given in
5 7.50 0.0127 1/2
6 9.00 0.0095 3/8 Fig. 17: only a limited number of cases are plotted in this figure.
7 10.50 0.0079 5/16 Modes have been taken in isolation or in combinations of two
8 12.00 0.0063 1/4 modes to include cylinder and roof displacements. There is a slight
9 13.50 0.0063 1/4 reduction in maximum load for imperfection shape M1 þM15
10 15.00 0.0063 1/4
11 15.75 0.0063 1/4
with respect to mode M1, as shown in Fig. 17 and Table 4;
however, the difference is of only 1.7%.

Fig. 12. LBA results: (a) critical loads for the lowest five eigenvalues and (b) first mode (λC ¼ 1.65 kN/m2), Configuration C1 at S¼ 1.0D (indicated as 1D-C1 in (a)).
236 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240

Fig. 13. Equilibrium paths: (a) isolated tank; (b) Configuration C1 at S ¼1.0D; (c) Configuration C2 at S ¼1.0D; (d) Configuration C5 at S ¼ 1.0D; (e) Configuration C1 at S ¼1.5D
and (f) Configuration C1 at S¼ 1.8D.

Because of the minor effect of mode combinations on the

maximum load in GNIA studies, it may be accepted that results
based on the lowest eigenmode are a good approximation.

6. Conclusions

The pressure patterns due to wind on two tanks in tandem or

skew arrangements have been identified in this work by means of
wind tunnel testing. The configurations studied are simplified
representations of wind tanks located on the perimeter of a tank
farm, or in a second line in which a tank is shielded by another
tank placed in the perimeter.
The results show significant differences between a tandem
Fig. 14. Imperfection-sensitivity diagram for isolated tank and five configurations configuration (both tanks aligned with the wind direction) and an
considered.
isolated tank, with peak pressures on the windward meridian
being at a higher elevation (closer to the roof), and with increases
in pressure coefficients. An increase in suctions is also found on
For two tanks in Configuration C1 separated at S ¼1.0D, results the roof. The differences with the isolated tank become less
for imperfection shapes in the form of individual eigenmodes are important as the angle of wind incidence increases, and in the
shown in Fig. 18, assuming ξ ¼0.75t. The most detrimental case is limit the two tanks that are in line perpendicular to wind have the
mode M1. lesser effect.
Combinations of modes are shown in Fig. 19, for cylinder modes The pressures accounting for two tank interaction were next
and roof modes. There are combinations that yield lower λmax used in a finite element analysis of shell buckling and post-
values but the differences with respect to mode M1 are of 1.5% (see buckling. Group effects are reflected in both, bifurcation LBA and
Table 4). geometrically nonlinear analysis with imperfections GNIA studies.
 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240 237

Table 3
Critical load factors (λc, LBA) and maximum loads (λmax, GNIA) for several configurations (ξ ¼0.75t).

Isolated tank S ¼1.0D S ¼ 1.5D S ¼1.8D

C1 C2 C3 C4 C5 C1 C1

LBA
λc [kPa] 2.38 1.65 2.29 2.30 2.24 2.28 1.89 1.93
λc =λc isolated – 0.69 0.96 0.96 0.94 0.96 0.79 0.81

GNIA
λmax [kPa] 1.76 1.20 1.61 1.81 1.81 1.70 1.43 1.48
λmax =λmax isolated 0.68 0.91 1.03 1.03 0.96 0.81 0.84

GNIA vs. LBA

λmax/λc 0.74 0.73 0.70 0.79 0.81 0.75 0.76 0.77

Fig. 15. Eigenmodes for isolated tank: (a) first mode (λC ¼ 2.38 kN/m2) and (b) Mode 15 (λC ¼6.37 kN/m2).

Maximum changes in buckling loads occur in the tandem config- arrangements, but some observations can be made at present
uration with tank separation of one diameter. The maximum which are of value to design and to understanding the safety of
reductions computed (tandem configuration) are of 30% in classi- existing plants under wind loads.
cal critical pressures (or in geometrically nonlinear analysis), with The results presented in this paper, taken in conjunction with
smaller changes (in the order of 5%) in other two-tank configura- results by other authors for other groups of tanks considered, have
tions. Reductions in buckling pressures of approximately 20% were some direct consequences for design. First, the most severe
found for tank separations of 1.5 and 1.8 times the diameter. changes in pressures due to blocking of a tank by another one
The shape of the imperfection considered in GNIA studies for occur in a tandem configuration, for which two tanks are aligned
the lowest 100 eigenvalues, indicate that the shape associated with the wind direction. From the point of view of design, this
with the lowest eigenvalue yields good estimates of most severe cannot be anticipated because wind direction changes with
imperfection sensitivity. respect to the arrangement of tanks in a tank farm. Second, there
A correlation between pressure changes and buckling changes is an influence of the configuration of the front tank on pressures
cannot be easily made, but in the present study there were 15% and buckling on the target tank. For a given tank arrangement,
changes in pressures for tandem configurations, which induced tandem configurations in which tanks in the periphery of a plant
changes of the same order in buckling loads as computed via LBA, have a flat roof or do not have a fixed roof, the state of the art
and even larger drops (30%) in maximum loads via GNIA. seems to indicate that no increase in pressures on the target tank
Because pressures and critical loads depend so heavily on the will occur and the case of an isolated tank represents a lower
geometry of the blocking tank, results obtained for long cylinders bound to buckling loads. On the other hand, if tanks in the
do not provide useful information on this problem. Even data periphery have a conical roof (and this may extend to shallow
obtained for one type of roof cannot be generalized to another dome configurations), then pressure increases at critical locations
type. (top part of the cylinder, for which the thickness is small) should
There are limitations to draw general recommendations be expected, with the consequence that the case of an isolated
oriented to design based on case studies of limited tank tank may be an upper bound to buckling loads.
238 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240

Fig. 16. Shape modes for Configuration C1, S ¼1.0D: (a) first critical mode; (b) Mode 14; (c) Mode 15; (d) Mode 23 and (e) Mode 94.
 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240 239

Fig. 17. Equilibrium paths for an isolated tank for imperfection amplitude ξ ¼0.75t Fig. 19. Equilibrium paths for Configuration C1, S ¼1.0D, for imperfection ξ¼ 0.75t
and various mode shapes. and various imperfection shapes.

Acknowledgments

Table 4 The authors thank Dr. Sebastián Delnero, Dr. Jorge Colman-
Maximum loads obtained via GNIA (with ξ ¼ 0.75t) for Configuration C1 with
Lerner, Dr. Julio Marañón-DiLeo, and Mariano García, members of
S ¼1.0D.
the Boundary Layer and Fluid Dynamics Lab at the National
Eigenmode affine Isolated tank Two tanks at S ¼1.0D, C1 University of La Plata in Argentina, for their help during the tests.
imperfection RCJ thanks the support during this research received through
max 2
in mode M λ [kN/m ] Difference λmax [kN/m2] Difference grants from the National University of Comahue (SECYT-UNCo).
[%] [%]
LAG thanks the support received through grants from the National
M1 1.7582 1.2025 University of Cordoba (SECYT-UNC) and CONICET (PIP 112-201201-
M1 þ M15 1.7283  1.70 1.2019  0.05 00126-CO).
M2 þ M15 1.1851  1.45
M1 þ M23 1.2020  0.04
M2 þ M23 1.1846  1.49
References

[1] ABAQUS. User's manuals. Rhode Island, USA: Hibbitt, Karlsson and Sorensen,
Inc.; 2006.
[2] API. Standard 650: welded steel tanks for oil storage. Washington, DC:
American Petroleum Institute; 2010.
[3] ASCE 7 Standard. Minimum design loads for building and other structures.
Reston, VA, USA: American Society of Civil Engineering; 1999.
[4] Esslinger M, Ahmed S, Schroeder H. Stationary wind loads of open topped and
roof-topped cylindrical silos. Der Stahlbau; 1971. p. 1–8.
[5] Eurocode 3. Design of steel structures – Part 4-2: tanks. Brussels: European
Committee for Standardization; 2007.
[6] Gu Z, Sun T. Classifications of flow pattern on three circular cylinders in
equilateral-triangular arrangements. J Wind Eng Ind Aerodyn 2001;89:553–68.
[7] Iamandi C, Georgescu A, Erbasu C. Experimental modeling of a four steel tanks
battery. In: Proceedings of the 11th international conference on wind
engineering. Lubbock (Texas, USA); 2003.
[8] Jaca RC. Lower bounds to shell instability in thin-walled tanks [Ph.D. thesis].
Córdoba, Argentina: National University of Córdoba; 2005 [in Spanish].
[9] Koiter WT. On the stability of elastic equilibrium [Ph.D. thesis]. Delft, Holland:
Delft Institute of Technology; 1945 [in Dutch] [English translation by NASA,
1967].
[10] MacDonald PA, Kwok KC, Holmes JD. Wind loads on circular storage bins, silos
and tanks – Part I: point pressure measurements on isolated structures. J Wind
Eng Ind Aerodyn 1988;31:165–88.
[11] MacDonald PA, Holmes JD, Kwok KC. Wind loads on circular storage bins, silos
and tanks – Part II: effect of grouping. J Wind Eng Ind Aerodyn 1990;34:77–95.
[12] Orlando M. Wind-induced interference effects on two adjacent colling towers.
Eng Struct 2001;23:979–92.
[13] Portela G, Godoy LA. Shielding effects and buckling of steel tanks in tandem
arrays under wind pressures. Wind Struct 2005;8(5):325–42.
[14] Portela G, Godoy LA. Wind pressures and buckling of aboveground steel tanks
with a conical roof. J Constr Steel Res 2005;61(6):786–807.
Fig. 18. Equilibrium paths for Configuration C1, S¼ 1.0D, for imperfection amplitude [15] Portela G, Godoy LA. Wind pressures and buckling in grouped steel tanks.
ξ ¼0.75t and various imperfection shapes. Wind Struct 2007;10(1):1–22.
240 C.A. Burgos et al. / Thin-Walled Structures 84 (2014) 226–240

[16] Riks E. An incremental approach to the solution of snapping and buckling [23] Uematsu Y, Yasunaga U. Wind loads on open-topped oil-storage tanks in
problems. Int J Solids Struct 1979;15:529–51. various arrangements. In: Proceedings of the sixth European and African
[17] Riks E. The application of Newton's method to the problem of elastic stability. J conference on wind engineering. Cambridge, UK; 2013.
Appl Mech 1972;39:1060–5. [24] Vickery BJ, Ansourian P. An investigation of the failure due to wind action of a
[18] Rotter JM, Schmidt H, editors. Buckling of steel shells: European design group of six silos at Boggabri, NSW, Report S152. Sydney: School of Civil
recommendations. 5th ed.. Mem Martins, Portugal: European Convention for Engineering, University of Sydney; 1974.
Construction Steelwork; 2008. [25] Yasunaga J, Koo C, Uematsu Y, Kondo K, Yamamoto M. Wind loads on two or
[19] Sabransky IJ, Melbourne WH. Design pressure distribution on circular silos three open-topped oil-storage tanks in various arrangements. In: Choi CK,
with conical roofs. J Wind Eng Ind Aerodyn 1987;26:65–84. editor. Proceedings of the world congress on advances in civil, environmental,
[20] Said NM, Mhiri H, Bournot H, Le Palec G. Experimental and numerical
and materials research. Seoul, Korea: Techno Press; 2012. p. 2339–52.
modelling of the three-dimensional incompressible flow behaviour in the
[26] Zhao Y, Lin Y. Buckling of cylindrical open-topped steel tanks under wind
near wake of circular cylinders. J Wind Eng Ind Aerodyn 2008;96:471–502.
load. Thin-Walled Struct 2014;79:83–94.
[21] Scorer RS. Theory of waves in the lee of mountains. Q J Meteorol Soc
[27] Zhao Y, Lin Y, Shen Y. Wind loads on large cylindrical open-topped tanks in
1949;75:41–56.
group. Thin Walled Struct 2014;78:108–20.
[22] Tsutsui T, Igarashi T, Kamemoto K. Interactive flow around two circular
[28] Zdravkovich MM. Review of flow interference between two cylinders in
cylinders of different diameters at close proximity. J Wind Eng Ind Aerodyn
1997;67:279–91. various arrangements. ASME J Fluids Eng 1977;99:618–33.