Вы находитесь на странице: 1из 18

International Journal of Solids and Structures 96 (2016) 92–109

Contents lists available at ScienceDirect

International Journal of Solids and Structures


journal homepage: www.elsevier.com/locate/ijsolstr

Imperfection sensitivity of locally supported cylindrical silos subjected


to uniform axial compression
Arne Jansseune a,∗, Wouter De Corte a,b, Jan Belis b
a
Department of Structural Engineering, Faculty of Engineering and Architecture, Ghent University, Valentin Vaerwyckweg 1, 9000 Ghent, Belgium
b
Department of Structural Engineering LMO, Faculty of Engineering and Architecture, Ghent University, Technologiepark 904, 9052 Zwijnaarde-Ghent,
Belgium

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

Article history: For the prediction of the real failure load of shell structures, such as locally supported cylindrical steel
Received 15 May 2015 silos under axial compression, it is convenient to take into account imperfections. It is assumed that such
Revised 14 April 2016
silos are very sensitive to a wide range of (even small) geometric imperfections, and that they lower the
Available online 15 June 2016
failure load significantly. Furthermore, these imperfections caused by the fabrication or the manufactur-
Keywords: ing process, are the dominant factor in the discrepancy between the theoretical/numerical predictions
Cylinder based on a perfect geometry and the experimental results of an imperfect geometry. In other words, it
Locally supported is important to make a well-considered choice for an imperfection when predicting the real failure load.
Yielding However, the imperfection sensitivity depends, among other things, on the shape of the shell, the stiff-
Buckling ening configuration, the boundary and loading conditions, etc. Before proceeding to the calculation of
Axial compression interaction curves and the development of new design rules for imperfect barrels, it is essential to per-
Imperfection sensitivity
form an extensive study to examine the influence of imperfections to the failure behaviour and to choose
a sufficiently detrimental imperfection shape.
In this study, different imperfection forms are numerically investigated: the linear bifurcation mode,
the non-linear buckling mode, several post-buckling deformed shapes of the perfect shell, and a weld
depression type A and B. Additional aspects, such as the orientation, the amplitude of the equivalent
imperfection, and the position of the influence of the weld depression are also investigated. The present
study takes into account the European normative documents and the guidelines of the recommendations
of the ECCS.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction presence of ring stiffeners (Jansseune et al., 2013a, 2013b). These


columns have a rectangular or a square cross-section, and are at-
For many branches of industries and during different stages of tached to the silo wall by welding. This relative simple type of
the manufacturing process, steel silos play an important role in the support is used for smaller silo structures (Rotter, 2001). In the
storage requirements of bulk solids. The steel barrel frequently has second configuration, the supporting columns are extended to the
a cylindrical shape and is placed in elevated position by a limited bottom of the silo wall and are concentrically placed beneath the
number of supporting columns (Doerich, 2007; Jansseune et al., centre of the silo wall (Vanlaere, 2006). Above each supporting col-
2013a, 2013b, 2015a, 2015b). As a consequence of this way of sup- umn, a longitudinal U-shaped stiffener is placed, which is attached
port, the total load exerted on the structure (mainly vertical) has to (i.e. welded) along a specific distance to the silo wall (Jansseune
be transferred to a relatively small proportion of the total circum- et al., 2013a, 2013b, 2015a, 2015b). These U-shaped stiffeners ab-
ference at the bottom of the barrel, resulting in locally high axial sorb a large part of the vertical loads, depending on the relative
stress concentrations and failure due to excessive yielding and/or stiffnesses of the stiffeners and the silo wall. In this case, two ex-
local instability. tra ring stiffeners are provided at the lower edge of the cylindrical
In this paper, the failure behaviour of locally supported shells barrel, and at the top of the stringer stiffeners. This stiffening con-
will be examined for two supporting/stiffening arrangements. The figuration is used for intermediate to large silo structures (Rotter,
first type are silos supported by engaged columns, without the 2001). Despite the difference in geometry between both configu-
rations, both have in common that the ground reaction force is

gradually introduced into the silo wall by shear, spreading the load
Corresponding author.
E-mail address: Arne.Jansseune@UGent.be (A. Jansseune).
better in circumferential direction, increasing the failure load.

http://dx.doi.org/10.1016/j.ijsolstr.2016.06.019
0020-7683/© 2016 Elsevier Ltd. All rights reserved.
A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109 93

Whereas the authors focused on the influence (Jansseune et al., surface of the silo wall, Fourier decomposition, etc.). Moreover, it
2013a, 2013b, 2015a, 2015b) and the optimisation (Jansseune et is not obvious to derive a (preferably simple) equivalent geometric
al., 2015a, 2015b) of the stiffening configuration (i.e. the engaged imperfection shape in a feasible and repeatable manner for typical
columns and the U-shaped longitudinal stiffeners) to the failure civil engineering structures (Arbocz, 1983).
behaviour for perfect silos, attention is now turning to the impact
of geometrical imperfections. Since all real structures are imperfect 2.1.3. Worst possible imperfections
and since the failure behaviour of axially compressed steel silos Searching to the very “worst possible” geometrical shape
is extremely susceptible to imperfections, it is of exceptional im- (within a specific range of tolerance) is the second approach, and
portance to take into account imperfections when predicting the is intended to provide a safe lower bound for design. This method
real failure load. Even very small imperfections can cause a sig- has been used from the beginning that imperfections were intro-
nificant reduction in the load carrying capacity. These imperfec- duced, and can in principle be applied for different shell prob-
tions, which are caused by the fabrication or the manufacturing lems. To find the most severe shape, parametric studies have been
process, are the dominant factor of the discrepancy between the done for specific problems (Greiner and Derler, 1995; Błachut and
theoretical/numerical predictions based on a perfect geometry and Jaiswal, 1999). Others used mathematical investigations to deal
the experimental results of an imperfect geometry. In other words, with this topic (Jamal et al., 2003; Koiter, 1963; Deml and Wun-
it is highly desirable to consider such (geometric) imperfections derlich, 1997). However, nowadays, these attempts are not widely
to be able to predict the real failure load. Before being able to spread in the design stage of shell structures. Furthermore, such
deduce buckling or interaction parameters of imperfect silos, this methods are difficult to apply due to several inevitable shortcom-
much-needed research is an important step to explore the failure ings: real structures generally do not necessarily have the "worst"
behaviour of locally supported imperfect silo structures subjected mode as geometric imperfection, and the "worst" mode frequently
to axial compression, in combination with U-shaped longitudinal is far from realistic (Rotter, 2004). In other words, it is doubtful
stiffeners or engaged columns. that this method provides imperfections which are close enough to
real silo structures, and consequently simulate the real imperfec-
2. Imperfection shapes tion sensitivity and the failure behaviour in practice. Furthermore,
underpredictions of the real buckling strength are not economi-
2.1. Approaches for the choice of an imperfection shape cal. In conclusion, this method is less appropriate to determine the
buckling strength by numerical simulations with the most severe
Since locally supported thin-walled steel silos subjected to ax- geometrical imperfection shape.
ial compression are highly sensitive to a wide range of imperfec-
tions, and the imperfection sensitivity depends on the amplitude 2.1.4. Simple equivalent imperfections
of the chosen imperfection, it is very difficult to single out one im- The third and last approach is the use of a relatively simple
perfection shape and amplitude, which is sufficiently disadvanta- “equivalent” geometric imperfection. Such a shape might per-
geous for one geometry (as designer during the design process) or haps not be 100% realistic nor is it the most severe possible shape,
for all geometries (as researcher). The latter is necessary if further its main purpose is to sufficiently influence the behaviour of the
progress has to be made in the development of design rules. How- silo (in an adverse way) to reduce the buckling load. Likely candi-
ever, it is not possible to select an imperfection shape from pre- dates to be used as equivalent shape are shapes which have a cer-
vious research, since the imperfection sensitivity depends on the tain degree of geometric similarity to either failure patterns (such
shape of the shell, the corresponding stiffening configuration, and as buckling or post-buckling modes) or the fabrication-caused
the boundary and loading conditions. Furthermore, several stud- shape deviations (e.g. an axisymmetric weld depression) (ECCS,
ies considered multiple defects simultaneously in the structure, 2008). These imperfections are modelled as initial shape deviations
such as different localised imperfections (Limam et al., 2011), lo- perpendicular to the middle surface of the perfect silo wall.
calised and distributed imperfections (Jamal et al., 2003), or pat- Since it is the purpose to develop design rules according to the
terned welds consisting of circumferential and/or meridional welds Eurocode (EN 1993-4-1, 2007a, EN 1993-1-6, 2007b), the last ap-
(Hubner et al., 2006; Pircher and Bridge, 2001a, 2001b), possi- proach, namely the use of equivalent geometric imperfections, was
bly causing (strong) interaction between the defects. Because of adopted in the current investigation as prescribed by the require-
the above mentioned reasons, it is an enormous challenge as a ments of the European normative documents (EN 1993-4-1, 2007a,
designer/researcher to choose one (or a combination of) specific EN 1993-4-6, 2007b). The reason for this choice is simply that, at
imperfection(s). In general, there are three main philosophies for this moment, the use of "equivalent" imperfections is by far the
choosing an imperfection, as described by Schmidt and Rotter: (1) most suitable approach to predict realistic failure loads by a nu-
the most realistic imperfection shape, (2) the worst imperfection merical analysis (ECCS, 2008). Furthermore, the present study takes
shape, and (3) an equivalent imperfection shape (Schmidt, 20 0 0; into account the guidelines and the commentary of the recommen-
Rotter, 2004; ECCS, 2008). dations of the ECCS (ECCS, 2008).

2.1.2. Realistic imperfections 2.2. Equivalent imperfection shapes


The first conceptual approach is to model geometric imperfec-
tions as “realistic” as possible based on measurements of simi- 2.2.1. Shape
lar silo structures (full-scale shells (Coleman et al., 1992; Ding et In previous work, different imperfection shapes have been sug-
al., 1991; Teng et al., 2005) or laboratory shells (Mathon and Li- gested for the use as equivalent geometric imperfection: a linear
mam, 2006; Jansseune, 2015)), while residual stresses and mate- or non-linear bifurcation buckling mode of the perfect shell (LBM
rial imperfections are frequently neglected, because of the difficul- or NBM) (Greiner and Derler, 1995; Koiter, 1963, 1945; Brendel
ties to quantify them (ECCS, 2008). Arbocz was probably the first and Ramm, 1980; Yamaki, 1984; Combescure, 1986; Speicher and
who used such measurements of imperfections in aerospace shells Saal, 1991; Wunderlich and Albertin, 20 0 0; Guggenberger et al.,
(Arbocz, 1974; Arbocz and Sechler, 1974; Arbocz and Babcock Jr, 20 0 0; Song et al., 2004; Song, 2002), a post-buckling deformed
1976). Currently, such measurements are only to a limited extent shape (PDS) (Guggenberger et al., 20 0 0; Song et al., 20 04; Song,
available on large steel silos, because of the cost and the difficul- 2002; Guggenberger, 1998; Schneider et al., 2001; Esslinger and
ties of its execution and implementation (e.g. define the best-fit Geier, 1972), or a combination of (bifurcation) buckling modes
94 A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109

(with different wavelengths) (Jamal et al., 2003). Other studies Table 1


Individual contributions to the overall effect from imperfections.
of measured imperfections pointed out that geometric imperfec-
tions are closely related to the fabrication process of the struc- Esslinger (%) Ummenhofer
ture (Arbocz, 1982). A good example are the circumferential welds and Knödel (%)
which are commonly found in silos which are made by welding Geometric imperfections 60 50
together circular strakes (Limam et al., 2011; Coleman et al., 1992; Residual stresses 10
Ding et al., 1991; Song et al., 2004; Song, 2002; Clarke and Rot- Unevenness of the supports 40 40
ter, 1988). The effect of weld depression on the buckling strength and loading imperfection
has already extensively been studied by (Pircher and Bridge, 2001a,
2001b; Ding et al., 1991; Rotter and Teng, 1989; Rotter, 1996; Rot-
ter, 1997; Berry et al., 20 0 0; Pircher et al., 20 01; Khamlichi et al., ward and outward deviations are considered, relative to the perfect
2004). In general, all of the above are good candidates, but none silo wall.
can yet be identified for "universal" application (ECCS, 2008).
Next, since all of the above mentioned shapes seem to be possi-
2.2.3. Amplitude
ble candidates as equivalent geometric imperfection, we will have
The amplitude of an equivalent geometric imperfection should
a look at the requirements of the Eurocode (ECCS, 2008; EN 1993-
be chosen in relation to the fabrication quality of the struc-
4-1, 2007a, EN 1993-1-6, 2007b). At first instance, the Eurocode
ture and should include the effects of both geometrical and non-
requires that the imperfection with the "most unfavourable effect"
geometrical imperfections. The Eurocode defines three tolerance
on the failure behaviour/load should be chosen. In other words, a
quality classes: excellent (quality class A), high (quality class B),
sufficient number of different imperfection patterns should be in-
and normal (quality class C) (EN 1993-4-1, 2007a, EN 1993-4-6
vestigated to identify the most severe one. The main reason why
2007b). The reason why the Eurocode defines several fabrication
different forms need to be considered is that imperfections of
tolerance quality classes is that the reduction of the strength of a
many different shapes and forms are found in real shell structures
perfect geometry largely depends on the amplitude of a bulge. As
(ECCS, 2008). First, the eigenmode-affine pattern (i.e. a linear bi-
a result, it is highly valuable to consider different classes of fabri-
furcation mode of the perfect shell) is proposed, unless a differ-
cation, and make the failure strength dependent on the quality of
ent unfavourable pattern could be justified. Furthermore, the Eu-
fabrication of the silo (ECCS, 2008). Indeed, Class A corresponds
rocode recommends the use of realistic unfavourable imperfection
with silos with an excellent quality (i.e. smaller imperfections)
shapes, which reflect the constructional detailing (such as axisym-
and will have the largest strength. High and normal quality silos
metric weld depressions (WD)) and boundary condition in an ad-
(i.e. medium sized and larger imperfections) are classified in re-
verse way. In contrast, shapes can be excluded from the investiga-
spectively Class B and Class C, and will have smaller strengths
tion as they are considered as unrealistic because of the method of
compared to Class A silos.
fabrication, manufacture, or erection.
Furthermore, it is important to know that the amplitude of an
In the present study, four alternative imperfection shapes are
“equivalent” geometric imperfection is about 60% larger than the
considered: i) the linear buckling mode of the perfect shell
standardised values of the dimple tolerance measure, because they
(LBM); ii) the non-linear buckling mode of the perfect shell
additionally must cover the effect of all other types of (non-) ge-
(NBM); iii) several post-buckling deformed shapes of the perfect
ometric, material, and non-measurable imperfections (ECCS, 2008),
shell (PDS); and iv) the axisymmetric weld depression (WD). The
such as deviations from the nominal geometric shape of the mid-
(dis)advantageous effect of these imperfection shapes are all eval-
dle surface, irregularities near welds, variations in nominal thick-
uated relative to the reference resistance. In other words, the fail-
ness and material properties, residual stresses (near welds), etc.
ure load of a geometric and material non-linear analysis including
(EN 1993-4-1, 2007a, EN 1993-4-6, 2007b).
imperfections (GMNIA) is always compared with the failure load
In previous research, Esslinger and Ummenhofer and Knödel es-
of a geometric and material non-linear analysis of a perfect silo
timated the individual contributions of geometrical imperfections
structure (GMNA) to assess the magnitude of the influence of the
(± 50%), residual stresses (±10%), and other imperfections ( ± 40%)
considered imperfections.
to the overall effect from imperfections (Esslinger and Geier, 1972;
Ummenhofer and Knödel, 1996) (Table 1).

2.2.2. Orientation
The Eurocode only gives a general guide on the orientation of 2.3. Description of the imperfection forms
equivalent geometric imperfections: the orientation of the maxi-
mum initial deviation (perpendicular to the middle surface of the In this section, a brief outline is given of the four imperfec-
perfect shell) should be chosen unfavourably towards the centre tion shapes which are considered here: i) the linear bifurcation
of the shell curvature (ECCS, 20 08; EN 1993-4-1, 20 07a, EN 1993- mode (LBM), ii) the non-linear bifurcation mode (NBM), iii) the
4-6, 2007b). It is expected that the inwardly oriented dents (i.e. post-buckling deformed shapes (PDS), and iv) the weld depression
the maximum amplitude is oriented towards the centre of the cur- (WD).
vature) are generally more disadvantageous, because they reduce Type 1—Linear Bifurcation Mode of the perfect shell or LBM
the curvature of the shell, resulting in a smaller (buckling) strength In previous research, an eigenmode-affine imperfection (i.e. a
(e.g. in Teng and Rotter, 1992). However, this rule of thumb is nor- linear bifurcation mode) of the perfect shell has been commonly
mally valid for shells which fail by elastic buckling, but this is not used as equivalent imperfection shape. This shape should also
always the case for thick-walled silos, which fail by plastic yield- be taken into consideration as recommended in the Eurocode
ing. In these, the yield condition is sometimes fulfilled earlier at (EN 1993-4-1, 2007a, EN 1993-4-6, 2007b), unless a different more
the centre of an outward dent: axial compression in combination severe pattern can be justified. In general, a LBM imperfection is a
with tension in circumferential direction (Schneider, 2006). rather severe form of imperfection (EN 1993-4-1, 2007a, EN 1993-
Since the choice for an inwardly oriented imperfection is not 4-6, 2007b). The advantage of an eigenmode as imperfection shape
always the most disadvantageous, it is difficult to give a generally is that it is easy to obtain from a linear bifurcation analysis of a
applicable guideline. Therefore, in this study, the influence of the perfect geometry (i.e. LBA). In most cases, the first eigenmode, cor-
orientation is investigated for all imperfection shapes, and both in- responding with the lowest eigenvalue, has been taken. However, it
A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109 95

is important to know that higher eigenmodes could be more criti- those in the adjacent strakes by half a panel width. Subsequently,
cal than the first eigenmode (ECCS, 2008). the strakes are connected with continuous circumferential welds to
Type 2—Non-linear Buckling Mode of the perfect shell or NBM form the silo. Previous research has demonstrated that, for cylin-
Compared to a LBM, a non-linear buckling mode or NBM (ei- ders under axial compression alone, the circumferential welds lead
ther the incremental mode at the limit point load or the non-linear to a larger reduction of the buckling load compared to meridional
bifurcation mode) also considers the influence of geometric non- welds (Hubner et al., 2006; Rotter and Teng, 1989). This is the rea-
linearity and thus the pre-buckling deformations. As a result, the son why in this study (pure axially compressed silos) only contin-
shape, the location, and its influence on the failure behaviour may uous circumferential welds have been considered.
be substantially different compared to a LBM. At each circumferential welded joint, a weld depression (which
The shape is obtained from a geometric non-linear analysis of a is predominantly axisymmetric) is caused by two phenomena. The
perfect geometry (i.e. GNA) using the modified Riks algorithm with first reason is the plate rolling process. During this process, flat
a severe restriction on the maximum increment size to i) be able steel plates are rolled into circular strakes, inducing plastic defor-
to accurately follow the highly non-linear path near failure and ii) mations, especially near its circumferential edges. Before welding,
avoid that no bifurcation point would stay undetected. the plate near the curved edges is often more slightly inwardly
Type 3—Post-buckling Deformed Shapes of the perfect shell or curved than the central part of the plate (Rotter and Teng, 1989).
PDS In other words, such curved panels are, in practice, not perfectly
A post-buckling deformed shape or PDS is a post-buckling de- cylindrical. Secondly, weld shrinkage occurs in the vicinity of the
formation pattern which is obtained from the decreasing post- weld joint during cooling, imposing a radial inward force on the
buckling path (i.e. after the limit load) of a geometric and ma- shell, causing radial deformations (Rotter and Teng, 1989). Rotter
terial non-linear analysis of a perfect geometry (i.e. GMNA) using and Teng proposed a well-defined expression for a weld depres-
the modified Riks algorithm. When post-buckling deformed shapes sion, as is presented in Eq. (1). In this formula, w is the devia-
are taken as imperfection shape, it is evident that the post-critical tion at a distance x from the centre of the weld depression; δ max
path/behaviour is of great importance. To avoid that no bifurca- is the maximum deviation at the centre of the weld depression;
tion point would stay undetected (as a shell often has many dif- λ is the half-wavelength; k is a shape factor and is a measure for
ferent potential bifurcation modes), these calculations were per- the moment continuity and the roundness at the weld centre) and
formed with very great care by introducing some additional mea- depends on the extent to which the deformations are prevented
sures (compared to standard analyses): i) arc-length control with during the cooling process.
very small increments and ii) calculate with two slightly different  πx  π x 
w = δmax · e−( λ ) · cos
πx
set of Riks parameters. + k · sin (1)
Because this pattern gradually changes due to geometric non- λ λ
linearity, it is difficult to define the shape in a unique and re-
Rotter and Teng proposed two extreme shapes, namely a weld de-
peatable manner (Rotter, 2004; Yamaki, 1984; Riks et al., 1996).
pression type A (k = 1) and a weld depression type B (k = 0). A
For practical considerations, the current study is restricted to ex-
type A weld depression (see Fig. 1 (a)) is completely rotationally
plore the influence of three different PDS patterns to the failure
stiff during cooling. This type is formulated according to the lin-
behaviour, each time after the limit load (similar to (Song et al.,
ear elastic shell bending theory for long thin-walled cylinders by
2004; Song, 2002)):
assuming full meridional moment continuity at the weld. Further-
• Immediately after the limit load Fmax (PDS - MAX); more, it is expected that this shape is more conservative than a
• At the lowest load Fmin of the post-buckling path after the limit type B weld depression (see Fig. 1 (b)). For the latter, the weld de-
load (PDS - MIN); pression is completely flexible (i.e. rotationally free) during cooling
• At the load Fmid at approximate halfway between the limit load and it is assumed that there is no moment continuity in merid-
Fmax and the lowest load Fmin (PDS - MID). ional direction at the weld. However, in real structures, the shape
lies in between these two extreme cases (0 < k < 1) and the flex-
Due to a lack of generality, a NBM and a PDS are less applicable
ural yielding is only partially prevented during the cooling process,
as equivalent imperfection shape for all geometries and load cases
and there is a certain moment continuity at the weld (Berry et al.,
compared to a LBM (Rotter, 2002).
20 0 0; Pircher et al., 2001).
Finally, an important remark about the choice for a linear bi-
The influence of the half-wavelength λ has already measured
furcation mode LBM, a non-linear buckling mode NBM, or a post-
on cylinders (Berry et al., 20 0 0) and extensively studied in previ-
buckling deformed shape PDS of the perfect shell as imperfec-
ous studies (Pircher and Bridge, 20 01a, 20 01b; Rotter, 20 04; Song
tion form. For geometric non-linear analyses, previous research
et al., 2004; Berry et al., 20 0 0; Muggeridge and Tennyson, 1969)
demonstrated that the reduction of the failure load is largely influ-
and mainly depends on the imperfection amplitude. In this study,
enced by the deformations before failure (Jansseune et al., 2013a,
the assumed half-wavelength λ was the linear meridional bending
2013b; Jansseune, 2015; Doerich, 2008). In short, the “flattened”
half-wavelength of a thin-walled cylinder (Eq. (2)).
silo wall has a smaller curvature as a result of which the effec-
√ √
tive yield stress (plastic yielding for thick-walled silos) and/or the λ = λb = 2 · λcl = 2.44 · R · t (2)
critical buckling stress (elastic buckling for thin-walled silos) re-
duce (Jansseune, 2015). As a consequence, the LBM or NBM or PDS An axisymmetric weld depression has several advantages com-
imperfection form not only introduces a geometric imperfection it- pared to other imperfection shapes. Firstly, it represents a com-
self, but may also hamper the flattening effect of the silo wall and mon shape in the real structures, namely a depression in the vicin-
weakens - to a greater or lesser extent - the corresponding disad- ity of a circumferential weld joint. As proposed in (EN 1993-4-
vantageous influence on the failure behaviour of the imperfect silo 1, 2007a, EN 1993-4-6, 2007b), it strongly reflects the fabrication
wall. method of the silo structure, in contrast to a LBM, NBM, or PDS.
Type 4 - Axisymmetric weld depression or WD Furthermore, previous research demonstrated that such circumfer-
In general, circular steel silos are constructed by welding to- ential welds are characteristic for circular steel silos (Rotter and
gether many circular panels. First, the panels are connected with Teng, 1989). In addition, the proposed formulation (see Eqs. (1)–
short meridional welds to form strakes. The circumferential posi- (2)) is relatively simple and it can easily be applied in numerical
tion of the meridional welds in one strake are usually offset from analyses. The shape function proposed by Rotter and Teng (Eq. (1))
96 A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109

Fig. 1. Shape of a weld depression for three fabrication tolerance classes defined in (EN 1993-1-6 2007) (R/t = 200).

Table 2
Overview of the different geometrical imperfection shapes.

Type Abbreviation Description Orientation

Type 1 LBM - INW Linear bifurcation mode of the perfect shell Inward
LBM - OUTW Linear bifurcation mode of the perfect shell Outward
Type 2 NBM - INW Non-linear buckling mode of the perfect shell Inward
NBM - OUTW Non-linear buckling mode of the perfect shell Outward
Type 3 PDS - MAX - INW Post-buckling deformed of the perfect shell, immediately after the limit load Inward
PDS - MAX - OUTW Post-buckling deformed of the perfect shell, immediately after the limit load Outward
PDS - MID - INW Post-buckling deformed of the perfect shell, at the load in halfway between the limit load and the lowest load Inward
PDS - MID - OUTW Post-buckling deformed of the perfect shell, at the load in halfway between the limit load and the lowest load Outward
PDS - MIN - INW Post-buckling deformed of the perfect shell, at the lowest load Inward
PDS - MIN - OUTW Post-buckling deformed of the perfect shell, at the lowest load Outward
Type 4 WD - A - INW Weld depression type A Inward
WD - A - OUTW Weld depression type A Outward
WD - B - INW Weld depression type B Inward
WD - B - OUTW Weld depression type B Outward

can be simplified even further to a localised triangular shape with Table 3


Recommended values for the dimple imperfection amplitude parameters (EN
a similar disadvantageous effect (Limam et al., 2011).
1993-4-1, 2007a, EN 1993-4-6, 2007b).
Although in principle the weld depression can be placed any-
where along the height of the silo wall, in this study, only those Fabrication tolerance quality class Description Un, 1 [-] Un, 2 [-]
positions are investigated in the region of the silo wall directly Class A Excellent quality 0.010 0.010
above the terminations of the U-shaped longitudinal stiffeners or Class B High quality 0.016 0.016
engaged columns, as will be discussed in Section 5.4. Class C Normal quality 0.025 0.025
In the numerical model (Abaqus), a mesh refinement has been
applied in the vicinity of the weld depression to model the shape
with sufficient accuracy (see Fig. 1 (c) and Fig. 14 (b)). Where lg is the relevant gauge length (Eq. (5)); R and t are re-
spectively the silo radius and silo wall thickness; ni is a multiplier
2.3.1. Overview imperfection shapes to achieve an appropriate tolerance level (a value of 25 is recom-
Table 2 summarises the chosen set of equivalent imperfection mended); Un, 1 and Un, 2 are the dimple imperfection amplitude
shapes. It is important to note that a LBM/NBM/PDS buckling mode parameters, and depend on the fabrication tolerance quality class,
was considered as inwardly oriented when the maximum ampli- as listed in Table 3.
tude (i.e. the maximum initial deviation perpendicular to the mid-
dle surface of the perfect shell) is oriented towards the centre of
curvature of the shell wall. 3. Geometry

2.4. Determination of the amplitude of equivalent imperfection shape 3.1. Cylindrical barrel

The initial amplitude δ max of the equivalent imperfection (in the The cylindrical barrel used in this study has a fixed cylinder
beginning of a GMNIA analysis) is the maximum deviation of the radius R of 1.0 m. The radius-to-thickness ratio R/t has been var-
imperfection and is measured perpendicularly to the middle sur- ied between 100 (i.e. a very thick-walled silo) and 10 0 0 (i.e. a
face of the perfect shell. This amplitude is defined in (EN 1993-4-1, very thin-walled silos) (See first column in Table 4) to investigate
2007a, EN 1993-4-6, 2007b) and is equal to the maximum of δ eq, 1 the influence of imperfections for different failure behaviours (i.e.
(Eq. (3)) and δ eq, 2 (Eq. (4)). plastic yielding, elastic buckling, or a combination of those phe-
nomena). This range corresponds with the range for which the au-
δeq,1 = lg · Un,1 (3) thors want to develop design rules. Furthermore, both short si-
los (i.e. h/R = 2.0) and high silos (i.e. h/R = 10.0) are considered
δeq,2 = ni · t · Un,2 (4) to be able to explore the influence of the boundary and loading
conditions on the upper edge of the cylindrical barrel to the im-
√ perfection sensitivity. In this study, the cylindrical barrel is always
lg = lgx = 4 · R·t (5) locally supported. The number of local supports nsup is equal to
A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109 97

Table 4
Variable geometrical parameters of the stiffened locally supported
silo.

R/t [-] dstif /R [-] tstif /t [-]

100 0.20; 0.30 Max.a


200 0.20; 0.30 Max.a
250 0.20; 0.30 Max.a
333.3 0.15; 0.30 Max.a
500 0.10; 0.20 Min.a
666.6 0.10; 0.20 Min.a
10 0 0 0.10; 0.20 Min.a
a
Minimum (min.), average (ave.), or maximum (max.) thickness—
Eqs. (6)–(7).

4 and the supports are distributed over the whole circumference


with equally spaced intervals.

3.2. Longitudinal stiffeners/engaged columns Fig. 2. Cross-section of the ring stiffeners.

For the first stiffening configuration, the local rigid supports are
positioned concentrically underneath the silo wall (see Fig. 3 (c)). or transition ring stiffener is used at the bottom of the cylindrical
Above each supporting column, a U-shaped longitudinal stiffener barrel with wlr /R = 0.20 and tlr /t = 1.0. Secondly, an intermediate
is placed on the exterior side of the silo wall with a limited length ring stiffener is situated at the top of the longitudinal stiffeners
sup
hsti f /R (see Fig. 3 (a)). The circumferential width of the longitudinal with wur /R = 0.10 and tur /t = 1.0. Both ring stiffeners have a rect-
stiffener dstif is equal to the circumferential width of the support angular cross-section (see Fig. 2).
dsup . The chosen values of the circumferential width dstif can be
found in the second column of Table 4 and depend on the radius-
to-thickness R/t. The ratio of the radial width to the circumferential 4. Numerical model
width wstif /dstif is equal to 0.25.
In the other case, the engaged supporting columns, which have The investigation of the influence of imperfections to the failure
a rectangular cross-section, are eccentrically placed on the exte- behaviour of the silo have been carried out with the finite element
rior side of the silo wall (see Fig. 3 (d)). The total height of the program ABAQUS (58). Rectangular doubly curved shell elements
in f
column can be split up into an "unattached" height hsti f /R (here: (S8R5) were used for the whole model representing the midsur-
sup face of each part: the cylindrical barrel, the U-shaped longitudinal
4.0) and an "attached" height hsti f /R (see Fig. 3 (b)). The values of
the circumferential width dstif are the same as for U-shaped stiff- the stiffeners/the engaged columns, and, if present, the ring stiff-
eners and are given in Table 4. Further, the radial width wstif is al- eners. When the silo is stiffened with U-shaped stringers, the local
ways equal to the circumferential width dstif . In other words, only steel supports are modelled by means of solid elements (C3D20R),
columns with a square cross-section are considered. which is visible in Fig. 4 (b). The silo wall, the longitudinal stiffen-
For both stiffening configurations, the attached column height ers and the lower ring are rigidly attached to the upper side of the
sup
hsti f /R is equal to 0.5, 1.0 for short silos (i.e. h/R = 2.0) and 1.0, support. The nodes of the lower side of the support are supported
in all directions, resulting in a rigid support.
2.0 for high silos (i.e. h/R = 10.0). The thickness of the longitu-
To save computing time, symmetry boundary conditions were
dinal stiffeners/engaged columns tstif should fulfil two conditions
exploited down to the longitudinal edges through the centre of the
(Eqs. (6)–(7)). Firstly, the thickness tstif must be at least equal to
support and the midplane between the supports. In this way, only
the silo thickness t and the thickness tstif may not be larger than
one eighth was modelled as shown in Fig. 4 (b). The results of one
five times the silo thickness t (Eq. (6)). The latter value has been
eighth of the shell were verified with results of the model of a
chosen by the author for practical considerations: because of the
complete shell (Jansseune et al., 2012). However, the figures in this
necessity to weld both parts together, the difference in thickness
paper show a complete shell model, which were obtained by mir-
may not be too large. Secondly, the longitudinal stiffeners/engaged
roring the 45 ° segment in circumferential direction.
columns should not be too thin or too thick circumferential width
The conical roof and the conical hopper of the cylindrical bar-
dstif (Eq. (7)).
rel were omitted in the numerical model and replaced by out-of-
tsti f round boundary constraints at the top and bottom edges of the
1.0 ≤ ≤ 5.0 (6)
t cylindrical barrel. When a lower ring is added at the lower edge of
the silo (for the silos with U-shaped stringer stiffeners), the above
dsti f
10 ≤ ≤ 40 (7) mentioned boundary condition is not taken into account at the
tsti f lower edge.
These equations provide a minimum and maximum allowable ra- The loading was chosen as a uniformly distributed axial load
tio of the stiffener thickness to the silo thickness tstif /t. Depend- at the upper edge of the silo, subjecting the silo wall to vertical
ing on the radius-to-thickness ratio R/t, either the minimum ratio compression (Rotter, 2004, 2009; Nielsen, 2008).
(Min.) or the maximum ratio (Max.) has been used for the para- To investigate the imperfection sensitivity of such locally sup-
metric study (see Table 4). ported steel silo, GMNA and GMNIA analyses (EN 1993-4-1, 2007a,
EN 1993-4-6, 2007b) were used, including geometric and material
3.3. Ring stiffeners non-linearity, respectively without and with (geometrical) imper-
fections. For all analyses, the modified Riks algorithm was used
When U-shaped longitudinal stiffeners are used, two extra ring to track the non-linear equilibrium path before, during and after
stiffeners are added to the stiffening configuration. Firstly, a base buckling.
98 A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109

Fig. 3. Supporting arrangements.

The material behaviour of the whole numerical model consists depression generally is an adverse imperfection shape. In the next
of a perfectly elastic-plastic material behaviour (without strain section (i.e. Section 5.4), the critical meridional position will be de-
hardening) with a Young’s modulus E of 210 GPa, a Poisson’s ratio termined of such a circumferential weld depression.
ν of 0.3, and a yield stress σ y of 235 MPa.
5.1. Influence of the imperfection orientation

5. Results and discussion Before discussing the influence of the imperfection orienta-
tion, the meaning of an inwardly oriented shape is given. A
The influence of the imperfection orientation and imperfection LBM/NBM/PDS buckling mode or a weld depression are considered
shape on the imperfection sensitivity is discussed, respectively, in as inwardly oriented when the maximum amplitude (i.e. the max-
Sections 5.1 and 5.2 of the result section. Next, the influence of the imum initial deviation perpendicular to the middle surface of the
fabrication tolerance quality class (i.e. the equivalent initial ampli- perfect shell) is oriented towards the centre of curvature of the
tude of the imperfection) will be considered in Section 5.3. From shell wall.
these sections, it will appear that an inwardly oriented type A weld The influence of the imperfection orientation is investigated by
comparing the failure load of an imperfect silo with an inward
A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109 99

Fig. 4. Model in ABAQUS.

imperfection shape Finward with the failure load of an imperfect imperfection shape, and amplitude. On the horizontal axis, either
silo with the corresponding outward imperfection shape Foutward . In the radius-to-thickness ratio R/t (Fig. 6 (a)) or the imperfection
Fig. 5, the dimensionless ratio Finward /Foutward is plotted on the ver- type (Fig. 6 (b)) is displayed.
tical axis for all imperfection shapes and for all 28 geometries, or- Before proceeding to the discussion of Fig. 6, it may be men-
dered in the horizontal axis according to the radius-to-thickness tioned that, for most imperfection shapes, it is expected that the
ratio R/t (4 cases for each thickness). ratio FGMNIA /FGMNA will be smaller than the unity, meaning that the
From this figure, two trends can be identified. First, the ratio failure load of the imperfect structure is smaller than for a perfect
Finward /Foutward is usually smaller than the unity (only a few excep- structure (i.e. FGMNIA < FGMNA ). Furthermore, the smaller the ratio
tions), indicating that inwardly oriented imperfections are indeed FGMNIA /FGMNA ( < 1) is for a specific imperfection pattern, the more
more disadvantageous than outward imperfections. This finding is sensitive the structure is for this imperfection pattern, or in other
expected as mentioned in Paragraph Orientation in Section 2.2. words, the more detrimental the imperfection pattern is for the
Secondly, a general downward trend can be noticed for the value structure.
Finward /Foutward as a function of the radius-to-thickness ratio R/t on From Fig. 6, the following findings can be derived.
the horizontal axis. In other words, thin-walled silos (which fail by • As expected, the (average) values of FGMNIA /FGMNA are always
elastic buckling) are more sensitive to the choice of the imperfec-
smaller than one, for all imperfection shapes and for all fab-
tion orientation than thick-walled silos (which fail by elasto-plastic
rication tolerance quality classes.
buckling), because inward imperfections seems to be more detri- • In general, the ratio FGMNIA /FGMNA decreases and thus the imper-
mental than outward imperfections when the influence of plastic-
fection sensitivity increases when the radius-to-thickness ratio
ity decreases and elasticity becomes more important to the failure
R/t increases. In other words, the buckling load of a thin-walled
behaviour and load.
silo (with a larger slenderness λ) is more susceptible to imper-
fections than the (elasto-)plastic failure load of a thick-walled
5.2. Influence of the imperfection shape
silo (with a smaller slenderness λ).
• The linear buckling mode (LBM) seems to be more detrimental
In this study, the influence of 7 different imperfection shapes
than the non-linear buckling mode (NBM), especially for thick-
(i.e. LBM, NBM, PDS - MAX, PDS - MID, PDS - MIN, WD type A,
walled silos.
and WD type B) is investigated on the failure load by means of
• The imperfection sensitivity is different for each post-buckling
GMNIA analyses. Furthermore, about 28 different geometries were
deformed shape (PDS). The PDS - MID is the most severe pat-
examined, taking into account the radius-to-thickness ratio R/t, the
tern, shortly followed by PDS - MIN. PDS - MAX is a less severe
silo height (high and short silos), and two stiffening configurations
pattern, and its sensitivity has the same order of magnitude as
(i.e. U-shaped stiffeners and engaged columns).
the NBM pattern (both types have a very similar shape)
In Fig. 6, the results of the exploratory imperfection sensitiv-
• The weld depression type A (WD - A) seems to be more severe
ity study are presented for inwardly oriented imperfections with
than a weld depression type B (WD - B).
fabrication quality class A (similar results are obtained for the im-
perfections with fabrication quality class B or C). Overall, the imperfections can be arranged from less to more
To be able to evaluate all results and to quantify the (negative) detrimental as follows: the non-linear buckling mode (NBM) to-
influence of an imperfection pattern, the failure load of an imper- gether with the post-buckling deformed shape just after failure
fect silo FGMNIA must always evaluated to the same reference re- (PDS - MAX), the linear buckling mode (LBM), the post-buckling
sistance, more specifically with the failure load of the perfect silo deformed shape at the lowest load (PDS - MIN), the post-buckling
wall FGMNA . This ratio FGMNIA /FGMNA is plotted on the vertical axis. deformed shape at a load equal to the average of the failure load
However, due to the large number of calculations, it is not possi- and the minimum load at the post-critical path (PDS - MID), a
ble to present the effective ratio of FGMNIA /FGMNA for each combina- weld depression type B (WD - B), and lastly a weld depression type
tion of geometry (28), imperfection shape (7), and amplitude (3). A (WD - A).
Therefore, an "averaged" value of the dimensionless failure load It is important to note that this order of importance was deter-
FGMNIA /FGMNA is plotted on the vertical axis of Fig. 6, which is ob- mined on the basis of a limited number of geometries (28) with a
tained by taking the average of all effective ratios of FGMNIA /FGMNA variable stiffening and supporting configuration (2: U-shaped stiff-
for the cases which have the same radius-to-thickness ratio R/t, eners or engaged columns), a variable radius-to-thickness ratio of
100 A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109

Fig. 5. The ratio of the failure load of an imperfect silo with an inward imperfection shape divided by the failure load of an imperfect silo with the corresponding outward
imperfection shape for all imperfection shapes and for all 28 cases, ordered according to the radius-to-thickness ratio R /t (quality class A).

the cylindrical barrel (7), and a variable silo height (2). To ver- a fabrication tolerance quality class, as defined in (EN 1993-1-6
ify whether this trend also extends to other locally supported si- 2007).
los, GMNIA calculations with variable imperfection patterns must
be done and investigated for much more stiffening configurations
(variable shape of the cross-section and height of the longitudinal
5.3.1. All imperfection shapes
stiffeners/engaged columns).
The failure load of an imperfect silo usually is less than the fail-
ure load of the failure load of a perfect silo (FGMNIA /FGMNA < 1). The
general trend is a decrease of the GMNIA failure load when the im-
5.3. Influence of the amplitude perfection amplitude δ max increases and thus the quality of the silo
decreases. Furthermore, it appears that the curve of a type A weld
Now, the influence of the initial amplitude of equivalent imper- depression (indicated with a black arrow) generally lies below the
fection shape is investigated for inwardly oriented imperfections curves representing the other imperfections. In other words, an in-
(see Fig. 7). The amplitude δ max is expressed as a dimensionless ward type A weld depression appears to be relatively disadvanta-
value by dividing it by the silo wall thickness t, and is related to geous for all amplitudes.
A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109 101

Fig. 6. The averaged ratio of the failure load of an imperfect silo with an inwardly oriented imperfection shape FGMNIA divided by the failure load of a perfect silo FGMNA for
all imperfection shapes and for all geometries (Class A).

5.3.2. Inward weld depression (type A) In contrast, when a weld depression is introduced just above
This section is devoted to the disadvantageous effect of (the the terminations of the stiffener/column, the location of failure
amplitude of) an inward weld imperfection. A weld depression shifts to the axisymmetric weld depression, for the thick-walled
largely influences the stress pattern in its vicinity, and conse- silo (see Fig. 8 (b) and (c)) as well as for the thin-walled silo (see
quently the location of failure and the failure load. Its effect first Fig. 9 (b) and (c)). Due to the deviations relative to the midsur-
is illustrated by means of (contour)plots of the axial stress σ x dis- face of the perfect shell wall in the vicinity of the weld depression,
tribution at the moment of maximum load (which is indicated in the stress field (and failure behaviour) is, besides the membrane
the corresponding load-displacement diagram) of two specific ge- stresses, influenced by bending stresses at the WD crest and WD
ometries, more particularly for a thick-walled silo (the geometry of troughs. As a result and in contrast to a perfect shell wall, the axial
Fig. 7 (a)) and a thin-walled silo (the geometry of Fig. 7 (b)). stress vary over the silo wall thickness. At the WD crest, compres-
For the perfect silo, failure occurs just above the terminations sive axial stresses arise on the exterior side (Ce) and tensile axial
of the U-shaped stiffener. The thick-walled silo fails by elasto- stresses on the interior side (Ti). In contrast, at the WD troughs,
plastic yielding (the yielding region corresponds with the dark blue tensile axial stresses arise on the exterior side (Te) and compres-
shaded area in Fig. 8 (a)), while the thin-walled silo fails by pure sive axial stresses on the interior side (Ci).
elastic buckling (the buckles are visible in Fig. 9 (a)). Furthermore, Another finding is that, in spite of the thin-walled nature of
the compressive axial stress (σ x < 0) is constant over the silo wall the cylindrical barrel (the radius-to-thickness ratio R/t is equal
thickness and the level of compressive stresses decreases to the to 10 0 0), yielding occurs at the WD centre: Ce and Ti (the dark
meridional plane halfway between the longitudinal stiffeners/the blue/red region in Fig. 9 (b) and (c)).
local supports.
102 A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109

Fig. 7. Imperfection sensitivity for inwardly oriented imperfection shapes with a variable amplitude (quality class A, B, and C) (high silo: h/R = 10.0; hsti f /R = 1.0).

Fig. 10 (a) plots a vertical section of the inward weld depression mensionless von Mises stress σ Mises /σ y (Fig. 12 and Fig. 13) at the
and gives an overview of the above found signs (compression or moment of maximum load along paths in circumferential direction.
tension) of the axial stresses in its vicinity. This stress pattern can The stress distribution is plotted along two different paths: path I
be explained by cutting the silo wall at the WD centre (Fig. 10 (b)) at the centre of the point of maximum deviation (i.e. the WD crest)
or at the (lower) WD trough (Fig. 10 (c)). In both cases, the com- and path II just below the centre of the point of maximum devia-
pressive load N (applied in meridional direction on the upper edge) tion (i.e. the WD lower trough) (see Fig. 11), and both on the exte-
must be kept in equilibrium by a membrane force N and a bending rior and interior side of the silo wall. Negative stresses correspond
moment M, which induce membrane and bending stresses in the with compression, positive values with tension.
silo wall, respectively. As will be demonstrated further, the magni- In Fig. 12, the stress distribution is investigated for the thick-
tude of this bending moment (and corresponding bending stresses) walled example silo with an inward weld depression type A with
increases with the WD amplitude δ max , making them increasingly variable quality class. Initially, if no imperfections are present, the
important and decisive for the failure behaviour. axial compressive stress is constant over the silo wall thickness
Next, the influence of the amplitude δ max (or the imperfection and the entire circumference of the silo is compressed in axial di-
quality class) on the stress level will be discussed by means of rection (in every point: σ x < 0). In contrast, when a weld depres-
plots of the dimensionless axial stress σ x /σ y (Fig. 12) and the di- sion is introduced, the axial stress vary over the silo wall thickness
A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109 103

Fig. 8. Contourplot of the axial stress at the moment of maximum load (R/t = 200; h/R = 10.0; dsti f /R = 0.30; wsti f = 0.25 · dsti f ; tstif /t = max.; hsti f /R = 1.0). (For interpretation
of the references to colour in the text, the reader is referred to the web version of this article.).

Fig. 9. Contourplot of the axial stress at the moment of maximum load (R/t = 10 0 0; h/R = 10.0; dsti f /R = 0.20; wsti f = 0.25 · dsti f ; tstif /t = min.; hsti f /R = 1.0). (For interpreta-
tion of the references to colour in the text, the reader is referred to the web version of this article.).

at the WD crest and WD troughs caused by an interaction of mem- tude of the axial compressive stress σ x increases at the exterior
brane and bending stresses. side of the WD crest (Ce1) and at the interior side of the WD
troughs (Ci1) to the effective yield stress σy . This “effective” yield
ef f
In first instance, the axial stresses are compared between an
imperfect silo wall with quality class A (membrane and bending stress σy is the critical meridional stress at which the von Mises
ef f
stresses) to those of the perfect shell wall (membrane stresses stress σ Mises reaches the yield stress σ y of 235 MPa, which depends
only). This comparison yields the following findings: the magni- according to the von Mises yield criterion on the circumferential
104 A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109

Fig. 10. Regions of compression (σ x < 0) and tension (σ x > 0) near the weld depression.

sionless axial stress σ x /σ y and the dimensionless von Mises stress


σ Mises /σ y at the moment of maximum load are plotted in Fig. 13
for five different positions across the silo wall thickness for the
thick-walled example silo with an inward weld depression type A
with quality class C.
From Fig. 13, it can be indeed seen that the axial stresses
vary strongly over the silo wall thickness at the weld crest (Ce to
Ti) and weld troughs (Te to Ci) due to the presence of bending
stresses, while the ratio of the von Mises stress to the yield stress
σ Mises /σ y at the weld centre is equal to the unity over 2/3 of the
circumferential width (θ ≤ ±28°) and over the entire thickness.
From the latter, it can be concluded that the most of the material
is exhausted and the silo wall has been failed by plastic yielding in
spite of the partially compressed silo wall.
In conclusion, the presence of a weld depression completely
Fig. 11. The stress paths near the weld depression.
changes the stress field in its vicinity. In contrast to a perfect shell
wall where only membrane stresses develop in the shell wall and
the bending stresses can be neglected, failure near the weld de-
stress σ θ . This can be explained by the region of yielding shifts pression is governed by an interaction of membrane stresses with
from the silo wall just above the stiffener (see Fig. 8 (a)) to the bending stresses. These disadvantageous bending stresses, which
weld depression (see Fig. 8 (b) and (c)). Furthermore, the compres- cause premature failure of the partially compressed shell wall, are
sive axial stresses (σ x < 0) turn into tensile axial stresses (σ x > 0) greatest at the weld crest and weld troughs (i.e. where the devi-
at the interior side of the WD crest (Ti1) and at the exterior side ations are largest relative to the midsurface of the perfect shell
of the WD troughs (Te1). wall) and increase when the fabrication tolerance quality deteri-
As the imperfection quality class decreases (A → B → C) or the orates (i.e. Class A → B → C) or the WD amplitude increases.
WD amplitude δ max increases, the region of “effective” yielding in
compression (with σMises = σy ) expands in circumferential direc- 5.4. Influence of the weld depression position
tion (Ce2 and Ci2), while the magnitude of the tensile stresses in-
creases at the exterior side of the WD troughs (Te2), and the re- In this section, the most disadvantageous position will be de-
gion of tensile stresses expands in circumferential direction (Ti2 termined for an inwardly oriented weld depression type A with
and Te2). half-wavelength equal to λb (Eq. (2)). Since the worst WD posi-
Clearly, the silo wall is no longer fully compressed over its en- tion will be used in the development of a new proposal for the
tire thickness in the vicinity of the weld depression. As the im- design rule (i.e. the purpose of the PhD of the first author as men-
perfection quality class deteriorates or the WD amplitude δ max tioned in the Introduction), it is important to verify this position
increases, the bending stresses become increasingly important, for sufficient geometries. However, to keep the number of GMNIA
which results in partial tension-compression in the silo wall (see analyses manageable, it was decided to limit the number of WD
Figs. 8– 12). As a result, failure is reached sooner at a lower com- positions to those in the region of maximum compressive stresses,
pressive load. i.e. the region just above the terminations of the U-shaped stiffen-
To illustrate the increasing influence of the bending stresses ers/engaged columns. In general, failure (i.e. plastic yielding and/or
in the WD crest (path I) and WD troughs (path II), the dimen- elastic buckling) will occur in that area.
A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109 105

Fig. 12. Distribution of the dimensionless axial stress σ x /σ y at the moment of maximum load (R/t = 200; h/R = 10.0; dsti f /R = 0.30; wsti f = 0.25 · dsti f ; tsti f /t = max.;
hsti f /R = 1.0) for an inward weld depression type A with variable quality class.

The WD position is determined by defining the distance xWD that significant. In general, a weld depression which is located fur-
(see Fig. 14 (a)), which is the distance in meridional direction be- ther away from the terminations of the stiffeners/columns is more
tween the top of the U-shaped stiffeners/engaged columns and the disadvantageous, within the range studied here.
centre of the weld depression. In this study, the values of WD The critical location of the weld depression is determined by
position xWD were chosen carefully over the entire direct vicin- means of GMNIA analyses for about 112 cases, taking into account
ity of the top of U-shaped stiffeners/columns with a fixed step the radius-to-thickness ratio R/t, the silo height (high and short si-
of 0.3 · R. A total of four WD positions xWD /R were considered: los), two stiffening configurations (i.e. U-shaped stiffeners and en-
0.3, 0.6, 0.9, and 1.2. Furthermore, those positions were eliminated gaged columns), the attached height hstif , and the width of the
where the largest deviations (i.e. the deviations which are located supporting columns in circumferential direction dsup = dsti f . In this
at a distance less than the half-wavelength λb of the WD cen- way, a suitable choice can be made about the critical WD position.
tre) are located under the top of the stiffeners/columns or above When viewing the results of all 112 cases, it was found that
the top edge of the cylindrical barrel (in the case of less high si- the critical WD location is mainly affected by the silo thickness
los). In other words, the weld depression should be completely lo- t (or the radius-to-thickness ratio R/t) and silo height h (or the
cated in the unstiffened silo wall. This restriction translates into a height-to-radius ratio h/R). Indeed, the radius-to-thickness ratio R/t
lower and an upper limit of the WD position xWD , as presented in clearly influences the shape of the depression, as shown in Fig. 16
Eq. (8). for three different radius-to-thickness ratios. Indeed, both the half-
  wavelength λ (Eqs. (1)–(2)) and the maximum deviation δ max (Eqs.
λb ≤ xW D ≤ h − hsup
sti f
− λb (8)
(3)–(5)) of the weld depression are dependent on the thickness of
In Fig. 15, the results are shown for four cases. In this figure, the the cylindrical barrel. The silo height with its boundary conditions
dimensionless failure load FGMNIA /FGMNA is plotted against the WD (preventing the out-of-roundness deformations) and loading condi-
position xWD divided by the cylinder radius R. Again, it appears tions (i.e. a uniform line load) changes the distance between these
that an inward weld depression type A strongly influences the fail- end conditions at the top edge of the cylindrical barrel and the
ure behaviour (FGMNIA /FGMNA < 1). In general, the GMNIA failure weld depression. From this, it can be concluded that both the silo
load decreases as the fabrication tolerance quality deteriorates (i.e. thickness and the silo height influence, to a limited extent, the de-
Class A → B → C). Now, if we look at the influence of the position of formations and the stress pattern at the weld depression, and con-
the weld depression xWD , one can observe that its influence is not sequently their critical location.
106 A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109

Fig. 13. Dimensionless axial stress σ x /σ y and dimensionless von Mises stress σ Mises /σ y distribution at the moment of maximum load for an inward weld depression type A
with quality class C.

Fig. 14. Position and mesh of the weld depression.

Table 5 shows the dimensionless critical WD position xWD /R Table 5


as a function of the radius-to-thickness ratio R/t and silo height Overview of the critical weld depression position xWD /R.
(high and short silo). The first finding is that the worst WD posi- R/t [-] High silo Short silo
tion is located closer to the terminations of the stiffeners/engaged
columns for thick-walled silos (R/t < 500) than for thin-walled si- 100 0.9 0.6
200 0.9 0.6
los (R/t ≥ 500). Secondly, the critical WD position moves down as
250 0.9 0.6
the top edge of the cylindrical barrel comes closer to the weld de- 333.3 0.9 0.6
pression. 500 1.2 0.6
666.6 1.2 0.9
10 0 0 1.2 0.9
A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109 107

Fig. 15. Imperfection sensitivity for an inward circumferential weld depression type A with a variable position in meridional direction (h/R = 10.0; hsti f /R = 1.0).

Fig. 16. The shape of a type A weld depression for three fabrication tolerance classes defined in (EN 1993-1-6 2007) with a variable radius-to-thickness ratio R/t.

6. Conclusions deformed shapes of the perfect shell, and a weld depression type
A and B. In addition, the amplitude (related to the Eurocode) and
In this paper, the imperfection sensitivity has been investigated the orientation (inward/outward) of these equivalent imperfections
for locally supported steel silos for a wide range of geometries: were varied to investigate their influence on the failure behaviour
both the silo geometry, the stiffening configuration, and stiffening of an imperfect structure.
dimensions were varied. In this way, well-founded conclusions can This comprehensive study demonstrates that silos are very sus-
be drawn. ceptible to a wide range of geometrical imperfections, and that
Different imperfection shapes were examined: the linear bifur- these imperfections significantly adversely influence the failure be-
cation mode, the non-linear buckling mode, three post-buckling haviour and load compared to a perfect silo. In other words, the
108 A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109

presence of imperfections is crucial for the prediction of the real ECCS, 2008. Stability of steel shells: European design recommendations—fifth ed.
failure load. The following conclusions can be drawn. ECCS—European Convention for Constructional Steelwork.
Esslinger, M., Geier, B., 1972. Gerechnete nachbeullasten als untere Grenze der ex-
perimentellen axialen Beullasten von Kreiszylindern. Deutsche Forschungs-und
• Overall, the imperfections can be arranged from less to more
Versuchsanst. für Luft-und Raumfahrt.
detrimental as follows: the non-linear buckling mode (NBM) to- Greiner, R., Derler, P., 1995. Effect of imperfections on wind-loaded cylindrical shells.
gether with the post-buckling deformed shape just after fail- Thin Walled Struct 23, 271–281.
Guggenberger, W., 1998. Proposal for design rules of axially loaded steel cylinders
ure (PDS - MAX), the linear buckling mode (LBM), the post-
on local supports. Thin Walled Struct 31, 169–185.
buckling deformed shape at the lowest load (PDS - MIN), the Guggenberger, W., Greiner, R., Rotter, J.M., 20 0 0. The behaviour of locally-supported
post-buckling deformed shape at a load equal to the average of cylindrical shells: unstiffened shells. J. Constr. Steel. Res. 56, 175–197.
the failure load and the minimum load at the post-critical path Hubner, A., Teng, J.G., Saal, H., 2006. Buckling behaviour of large steel cylinders with
patterned welds. Int. J. Pressure Vessels Pip. 83, 13–26.
(PDS - MID), a weld depression type B (WD - B), and lastly a Jamal, M., Lahlou, L., Midani, M., Zahrouni, H., Limam, A., Damil, N., Potier-Ferry, M.,
weld depression type A (WD - A). 2003. A semi-analytical buckling analysis of imperfect cylindrical shells under
• As expected, inward imperfections are generally more un- axial compression. Int. J. Solids Struct. 40, 1311–1327.
Jansseune, A., 2015. Optimisation of the stiffening configuration of axially com-
favourable than outward imperfections, because they reduce pressed steel silos on local supports. In: Department of Industrial Technol-
the curvature of the shell, resulting in a smaller (buckling) ogy and Construction, Faculty of Engineering and Architecture. Department of
strength. Structural Engineering, Faculty of Engineering and Architecture, Ghent Universi-
tyGhent, p. 462.
• In general, the larger the amplitude of the equivalent imperfec- Jansseune, A., De Corte, W., Belis, J., 2015a. Optimal stiffening configuration for
tion, the greater the adverse effect of the imperfection to the locally supported cylindrical silos: engaged columns. Constructional Steel Re-
failure behaviour/strength. search. Elsevier.
Jansseune, A., De Corte, W., Belis, J., 2015b. Elastic failure of locally supported silos
• The failure behaviour in the vicinity of a weld depression is in-
with U-shaped longitudinal stiffeners. KSCE J. Civ. Eng. 19 (4), 1041–1049.
fluenced by an interaction of membrane and bending stresses. Jansseune, A., De Corte, W., Van Impe, R., 2013a. Plastic failure of locally supported
The disadvantageous bending stresses increase with the weld silos with U-shaped longitudinal Stiffeners. In: The 2013 world congress on Ad-
vances in Structural Engineering and Mechanics, Jeju, Korea, pp. 345–360.
depression amplitude, cause premature failure in a partially
Jansseune, A., De Corte, W., Van Impe, R., 2013b. Column-supported silos: elasto–
compressed silo wall and are therefore decisive for the failure plastic failure. Thin Walled Struct 73, 158–173.
behaviour of an imperfect silo wall. Jansseune, A., De Corte, W., Vanlaere, W., Van Impe, R., 2012. Influence of the cylin-
• For the later proposal of a design rule, an inward weld depres- der height on the elasto-plastic failure of locally supported cylinders. Steel Com-
pos. Struct. 12, 291–302.
sion type A with half-wavelength equal to λb has been cho- Khamlichi, A., Bezzazi, M., Limam, A., 2004. Buckling of elastic cylindrical shells
sen as equivalent imperfection shape. At first instance, such a considering the effect of localized axisymmetric imperfections. Thin Walled
circumferential weld depression appears to be relatively detri- Struct 42, 1035–1047.
Koiter, W., 1945. On the Stability of Elastic Equilibrium. Delft University, Delft.
mental compared to the other imperfection shapes. Secondly, Koiter, W., 1963. The effect of axisymmetric imperfections on the buckling of
this shape has the advantage that it is closely related to the cylindrical shells under axial compression. In: Proc Koninklijke Nederlandse
fabrication process of a silo, and thus can be found in a real Akademie van Wetenschappen, pp. 265–279.
Limam, A., El Bahaoui, J., Khamlichi, A., El Bakkali, L., 2011. Effect of multiple local-
silo. ized geometric imperfections on stability of thin axisymmetric cylindrical shells
under axial compression. Int. J. Solids Struct. 48, 1034–1043.
References Mathon, C., Limam, A., 2006. Experimental collapse of thin cylindrical shells sub-
mitted to internal pressure and pure bending. Thin Walled Struct 44, 39–50.
Arbocz, J., The effect of initial imperfections on shell stability, in: Thin-Shell Struc- Muggeridge, D.B., Tennyson, R.C., 1969. Buckling of axisymmetric imperfect circular
tures: Theory, Experiment, and Design, 1974, pp. 205–245. cylindrical shells underaxial compression. AIAA J. 7, 2127–2131.
Arbocz, J., 1982. The imperfection data bank, a mean to obtain realistic buckling Nielsen, J., 2008. From silo phenomena to load models. In: Structures and Granular
loads. In: Buckling of Shells. Springer, pp. 535–567. Solids. Taylor & Francis, pp. 49–57.
Arbocz, J., 1983. Shell stability analysis—theory and practice. In: Collapse: The Buck- Pircher, M., Berry, P.A., Ding, X., Bridge, R.Q., 2001. The shape of circumferential
ling Of Structures In Theory And Practice, pp. 43–74. weld-induced imperfections in thin-walled steel silos and tanks. Thin Walled
Arbocz, J., Babcock Jr, C.D., 1976. Prediction of Buckling Loads Based on Experimen- Struct 39, 999–1014.
tally Measured Initial Imperfections. Springer. Pircher, M., Bridge, R., 2001a. The influence of circumferential weld-induced imper-
Arbocz, J., Sechler, E., 1974. On the buckling of axially compressed imperfect cylin- fections on the buckling of silos and tanks. J. Constr. Steel Res. 57, 569–580.
drical shells. J. Appl. Mech. 41, 737–743. Pircher, M., Bridge, R.Q., 2001b. Buckling of thin-walled silos and tanks under axial
Berry, P., Rotter, J., Bridge, R., 20 0 0. Compression tests on cylinders with circumfer- load-some new aspects. J. Struct. Eng. 127, 1129–1136.
ential weld depressions. J. Eng. Mech. 126, 405–413. Riks, E., Rankin, C.C., Brogan, F.A., 1996. On the solution of mode jumping phenom-
Błachut, J., Jaiswal, O., 1999. On the choice of initial geometric imperfections in ex- ena in thin-walled shell structures. Comput. Meth. Appl. Mech. Eng. 136, 59–92.
ternally pressurized shells. J. Pressure Vessel Technol. 121, 71–76. Rotter, J., 1996. Elastic plastic buckling and collapse in internally pressurised axi-
Brendel, B., Ramm, E., 1980. Linear and nonlinear stability analysis of cylindrical ally compressed silo cylinders with measured axisymmetric imperfections: in-
shells. Comput. Struct. 12, 549–558. teractions between imperfections, residual stresses and collapse. In: Proceedings
EN 1993-4-1, 2007a. Eurocode 3: Design Of Steel Structures Part 4-1: Silos. BIN, of the International Workshop on Imperfections in Metal Silos: Measurement,
Brussels. Characterisation and Strength Analysis. CA-Silo, Lyon, France, pp. 119–140.
EN 1993-1-6, 2007b. Eurocode 3: Design Of Steel Structures Part 1-6: Gener- Rotter, J., Teng, J., 1989. Elastic stability of cylindrical shells with weld depressions.
al—Strength And Stability Of Shell Structures. BIN, Brussels. J. Struct. Eng. 115, 1244–1263.
Clarke, M.J., Rotter, J.M., 1988. A technique for the Measurement Of Imperfections Rotter, J.M., 1997. Pressurised axially compressed cylinders. In: Proceedings Int.
In Prototype Silos And Tanks. University of Sydney, School of Civil and Mining Conf. on Carrying Capacity of Steel Shell Structures. Brno, p. 7.
Engineering. Rotter, J.M., 2001. Guide for the Economic Design of Circular Metal Silos. Spon Press.
Coleman, R., Ding, X., Rotter, J., 1992. The measurement of imperfections in full-s- Rotter, J.M., 2002. Shell Buckling and Collapse Analysis for Structural Design: The
cale steel silos. In: 4th International Conference on Bulk Materials, Storage, han- New Framework of the European Standard. Festschrift for Professor Chris Calla-
dling and Transportation Seventh International Symposium on Freight Pipelines. dine, Cambridge, pp. 1–29.
Institution of Engineers, Australia, pp. 467–472. Rotter, J.M., 2004. Cylindrical shells under axial compression. In: Teng, J.G., Rotter, J.
Combescure, A., 1986. Static and dynamic buckling of large thin shells: design pro- (Eds.), Buckling of Thin Metal Shells. Spon Press, London, pp. 42–87.
cedure, computation tools, physical understanding of the mechanisms. Nucl. Rotter, J.M., 2009. Silos and tanks in research and practice: state of the art and cur-
Eng. Des. 92, 339–354. rent challenges. In: Domingo Cabo, A. (Ed.), Symposium of the International As-
Deml, M., Wunderlich, W., 1997. Direct evaluation of the ‘worst’ imperfection shape sociation for Shell and Spatial Structures. Editorial de la Universitat Politécnica
in shell buckling. Comput. Meth. Appl. Mech. Eng. 149, 201–222. de Valencia, Valencia, pp. 65–76.
Ding, X., Coleman, R., Rotter, J., 1991. A profile measurement system and its ap- Schmidt, H., 20 0 0. Stability of steel shell structures: general report. J. Constr. Steel
plication in precise engineering surveying. Proc q/58th FIG PC Meeting and lnt Res. 55, 159–181.
Syrup. Schneider, W., 2006. Ersatzimperfektionen für den numerischen Beulsicherheit-
Doerich, C., 2007. Strength and stability of locally supported cylinders. Institute for snachweis stählerner Schalentragwerke—state of the art. Stahlbau 75, 754–760.
Infrastructure & Environment. The School of Engineering and Electronics, Uni- Schneider, W., Höhn, K., Timmel, I., Thiele, R., 2001. Quasi-collapse-affine imperfec-
versity of Edinburgh, Edinburgh. tions at slender wind-loaded cylindrical steel shells. In: Proceedings of second
Doerich, C., 2008. Effect of geometric nonlinearity of locally supported cylinders. In: European Conference on Computational Mechanics-ECCM-2001, Cracow, Poland
Structures and Granular Solids. Taylor & Francis, pp. 183–197. CD-Rom.
A. Jansseune et al. / International Journal of Solids and Structures 96 (2016) 92–109 109

Simulia, Abaqus, 2016. Teng, J.G., Lin, X., Michael Rotter, J., Ding, X.L., 2005. Analysis of geometric imper-
Song, C., 2002. Buckling of Shells Under Non-Uniform Stress States Ph.D. thesis. De- fections in full-scale welded steel silos. Eng. Struct. 27, 938–950.
partment of Civil and Structural Engineering, The Hong Kong Polytechnic Uni- Ummenhofer, T., Knödel, P., 1996. Typical imperfections of steel silo shells in civil
versity. engineering. In: Proceedings of the International Workshop on Imperfections in
Song, C.Y., Teng, J.G., Rotter, J.M., 2004. Imperfection sensitivity of thin elastic Metal Silos: Measurement, Characterisation and Strength Analysis. CA-Silo, Lyon,
cylindrical shells subject to partial axial compression. Int. J. Solids Struct. 41, France, pp. 103–118.
7155–7180. Vanlaere, W., 2006. Buckling Behaviour of Stiffened Cylinders on Local Supports.
Speicher, G., Saal, H., 1991. Numerical calculation of limit loads for shells of revo- Department of Structural Engineering, Faculty of Engineering Sciences, Ghent
lution with particular regard to the applying equivalent initial imperfection. In: University, Ghent.
Buckling of Shell Structures, on Land, in the Sea, and in the Air, pp. 466–475. Wunderlich, W., Albertin, U., 20 0 0. Analysis and load carrying behaviour of imper-
Teng, J., Rotter, J., 1992. A study of buckling in column-supported cylinders. In: fection sensitive shells. Int. J. Numer. Methods Eng. 47, 255–273.
Contact Loading and Local Effects in Thin-Walled Plated and Shell Structures. Yamaki, N., 1984. Elastic Stability of Circular Cylindrical Shells. Elsevier.
Springer, pp. 52–61.