Thin-Walled Structures 76 (2014) 2841

Contents lists available at ScienceDirect

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

Imperfection amplitudes for nonlinear analysis of open thin-walled

steel cross-sections used in rack column uprights
M.M. Pastor n, M. Casafont, J. Bonada, F. Roure
Department of Strength of Materials and Structural Engineering, Escola Tcnica Superior d'Enginyeria Industrial de Barcelona (ETSEIB), Universitat
Politcnica de Catalunya (UPC), Av. Diagonal, 647, 08028 Barcelona, Spain

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

Article history: When numerical models are used to predict behaviour of open thin-walled steel cross-sections, the
Received 15 March 2013 nonlinear analysis results can be inuenced by the magnitudes introduced as initial imperfection. At
Received in revised form present, a wide range of values is considered in research concerning this topic. This paper explores up to
31 October 2013
what point the nonlinear analysis results are sensitive to the choice of imperfection magnitudes and
Accepted 31 October 2013
attempt to rene the spectrum of magnitudes that should be used. The study focuses on rack uprights
Available online 30 November 2013
(with and without perforations) under compression and the numerical model has been validated with
Keywords: experimental tests conducted by the authors. Three different column lengths have been selected to
Initial geometrical imperfections reproduce a mainly local, distortional and global failure mode, so that coupled instabilities have not to be
Rack column uprights
considered in this case. The results show that the ultimate load and collapse mode are both sensitive to
FE analysis
imperfection amplitude, mainly in the case of distortional buckling.
& 2013 Elsevier Ltd. All rights reserved.

1. Introduction intends to show that the nonlinear analysis can give a prediction of
the behaviour of the upright near or far from reality depending on
Computational methods have become an efcient and robust the size of equivalent geometric imperfection introduced in the
tool for structural analysis. In particular, the nite element method model. Since, as seen through the literature on the subject, a wide
(FEM) is one of the most widely used methods; which is probably range of values is used by different authors, it is worth analysing the
due to its capability to include the features of the real behaviour of importance of this factor on the numerical result.
structures in the numerical modelling. In the present numerical imperfection sensitivity analysis the
The FE results are sensitive to many factors; such as the de- range of imperfection values tested covers most of those cited in
nition of the geometry and material, element type and mesh the research literature on this matter. In order not to limit the
selected, loads and boundary conditions applied, analysis type, scope of the analysis, perforated (racks) and non-perforated
and solver control; among others [1]. The aim of this paper is not sections have been considered. Results of experimental tests
focussing on the global FE model implementation, but on a specic carried out by the authors' research group have been used as a
issue: equivalent geometrical imperfection magnitudes to be used
to tackle the instability of cold-formed steel structural members
subject to compression; specically addressed to nonlinear analy-
sis (GMNIA) of uprights.
The conventional procedure in collapse modelling consists of
introducing the displacements produced by the critical buckling
mode, conveniently scaled, as initial imperfection for use in subse-
quent nonlinear analysis. An equivalent imperfection magnitude is
commonly used, which includes the effects of the material and
geometrical imperfections and residual stresses owing to manufac-
turing process as well as possible load eccentricities. This article

n
Corresponding author. Tel.: 34 93 4016532; fax: 34 93 4011034.
E-mail addresses: m.magdalena.pastor@upc.edu (M.M. Pastor),
miquel.casafont@upc.edu (M. Casafont), jordi.bonada@upc.edu (J. Bonada),
francesc.roure@upc.edu (F. Roure). Fig. 1. Denition of geometric imperfections according to [2].

0263-8231/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.tws.2013.10.025
 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841
Fig. 2. (a) Main dimensions of the cross-sections. (b) Storage rack with periodic perforations in web and anges.
Fig. 3. Compression test. Experimental setup.
reference for calibration and validation of numerical model. Con-
clusions regarding inuence and tting of imperfection magni-
tudes analysed will be derived from this study.
2. Literature review on numerical imperfections
The equivalent geometric imperfection sizes used in this eld
can be divided into two groups: those used in predominantly
sectional failure modes (local-L and distortional-D) and those used
when the failure mode is mainly global. It may be said, in
simplied form, that the rst would be suitable for the analysis
of columns of short and intermediate length, and the latter for
long column lengths.
In sectional failure modes (L, D) initial imperfection magni-
tudes are usually dened either as a function of the plate thickness
(t) or the web width (w) of the cross section. Fig. 1 [2] denes two
types of imperfection: type 1 is the maximum local imperfection
in a stiffened element, and type 2 is the maximum deviation from
straightness for a lip stiffened or unstiffened ange.
In local mode, we nd imperfection magnitudes of (i) 0.006w
[24] and w/200 [57] among those that are dened in terms of
29
the web width, and (ii) a wide spectrum of values ranging from
0.1t [812] to 1t [11] when the imperfection is dened as a
function of the plate thickness. Intermediate values in use are
0.34t [1] and 0.5t [11]. The most cited reference on this subject [2]
suggests 0.006w and EN 1993-1-5:2006/AC:2009 [5] recommends
w/200 0.005w for local buckling.
In addition to the previous ones, the expression suggested by
Walker (1975) to scale initial imperfections in FE models is
successfully used by Gardner and Nethercot [12], Yang and Han-
cock [13], Becque and Rasmussen [14] and Ashraf et al. [15]. In
[16,17] the rst local buckling mode is multiplied by a factor, 0.01
and 1 mm respectively.
In distortional mode, we nd imperfection magnitudes of
(i) f/50 [5,7] where f is the ange width of the cross-section, and
(ii) a wide spectrum of values ranging from 0.1t [810,18] to 1.5t
[19] when the imperfection is dened as a function of the plate
thickness. Intermediate values in use are 0.15t [18,20], 0.5t [19],
0.64t [20], 0.94t [1] and 1t [24,19,21]. Reference [2] suggests 1t
and EN 1993-1-5:2006/AC:2009 [5] recommends f/50 0.02f for
distortional buckling.
The imperfections specied in [5] are being used by the
authors' research group [6], although they were not dened for
cold-formed structures. These magnitudes are similar to those
suggested in [2].
In global mode, the imperfection magnitudes usually used or
tested to create the perturbed mesh range from L/5000 [12] to L/
750 [18,19,21,22] where L is the member length. Intermediate
values in use are L/2000 [12,19,23,24], L/1500 [14,19,24,26,27], L/
1250 [18], and L/1000 [3,4,7,9,12,18,19,21,2325]. EN 1090-2:2008
[22] recommends L/750 and ECCS [25] does L/1000, being the
latter the most commonly used for global buckling.
Combinations of imperfection magnitudes varying between
0.075t and 0.2t (mostly 0.1t), for local/distortional buckling, and
L/750L/2000 (mostly L/1000), for global buckling, are used in
[810,18,23,28]. As in [18], for columns experiencing distortional/
global interaction and practical purposes, pure global initial
imperfections may be taken as the most detrimental ones, in the
sense that they lead to the lowest ultimate loads. As in [19],
statistically it is not recommended to combine all imperfections to
cumulate their negative effects because of their random compen-
sation. In this paper coupled instabilities have not been analysed.
As can be realized from this literature review, quite different
values are adopted by researchers to scale the numerical models.
30 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841

L=250mm

L=600mm

L=2600mm
Fig. 4. Experimental deformed shapes. Rack (left) and non-perforated (right) members.

This paper explores up to what point the imperfection magnitude Table 1

affects the predicting ultimate load and the whole behaviour of Experimental results. Column lengths (in bold) selected.
the structural member.
Length PultExp,Rack Collapse PultExp,Solid Collapse mode Non-perf/
(mm) (N) mode (N) PERF

3. Cross-section analysed 250 143,733 L 177,718 L (SD) 1.236

400 129,407 SD 157,101 SD 1.214
Two types of cold-formed steel sections have been investi- 600 125,483 SD 138,671 SD 1.105
800 109,967 SD 131,650 SD 1.197
gated: with and without perforations. The material and geometry
1000 106,975 SD 129,401 SD 1.210
of the cross-sections are the same (Fig. 2a). The only difference is 1200 100,327 SD 127,471 AD (GFT)/SD 1.271
that one of them was manufactured without punching. 1500 93,852 AD GFT 111,582 AD GFT 1.189
The perforated cross-section is a typical storage rack produced 1800 74,500 (AD) GFT 93,797 (AD) GFT 1.259
2200 47,870 GFT 64,462 GFT 1.347
by one of the main manufacturers in Europe. Perforations consist
2600 44,037 GFT 49,413 GFT 1.122
of slots along the web of the member and circular bolt holes
along the side anges of the member (Fig. 2b). The cross-section L: Local; SD/AD: symmetric/asymmetric distortional; GFT: global exural-torsional.
chosen 80 mm web and 1.8 mm thickness is available in
three widths (80, 100 and 120 mm) and three thicknesses geometric details and real shapes of perforations are not
(1.8, 2 and 2.5 mm). It is a cross-section which behaviour is included, and only the main dimensions of the cross-section are
well-known by this research group. For condentiality reasons, here shown.
 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841 31

450000

400000

350000

Critical Load (N)

300000

250000

200000

150000

100000

50000

0
250 750 1250 1750 2250 2750
Column length (mm)

Fig. 5. Critical load using GBT.

120
SD (%) AD (%)
G (%) L (%)
100
Modal participation (%)

80

60

40

20

0
250 750 1250 1750 2250 2750

Column length (mm)

Fig. 6. Evolution of modal participation for the critical load.

Fig. 7. Sections tested and analysed.

32 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841

Fig. 8. Bilinear and multilinear material models. (a) Stressstrain curves modelled and (b) transition from elastic to plastic behaviour.

Fig. 9. FE results on rack uprights.

4. Experimental results Table 2

Experimental and numerical results.
Solid (non-perforated) and storage rack (perforated) members
Length PultExp,Rack PultNum,Rack NUM/ PultExp,Solid PultNum,Solid NUM/
of different lengths were tested according to the relevant test (mm) (N) (N) EXP (N) (N) EXP
procedures dened in Annex A of EN 15512:2009 [29]. Tensile test
coupons were extracted from the sheet and tested in order to 250 143,733 142,715 0.993 177,718 171,965 0.968
400 129,407 126,760 0.980 157,101 166,042 1.057
determine the actual strength of the material. The yield stress fy is 600 125,483 117,070 0.933 138,671 141,730 1.022
423 MPa and the ultimate stress fu is 491 MPa. The measured 800 109,967 110,672 1.006 131,650 129,515 0.984
1000 106,975 105,778 0.989 129,401 123,831 0.957
average thickness is t 1.83 mm.
1200 100,327 101,958 1.016 127,471 120,668 0.947
The experimental setup to test the upright specimens in com- 1500 93,852 99,278 1.058 111,582 110,567 0.991
pression has the following features (Fig. 3). A very stiff carbon steel 1800 74,500 82,962 1.114 93,797 97,350 1.038
2200 47,870 60,670 1.267 64,462 70,087 1.087
30 mm thickness plate is clamped at both ends of the specimen 2600 44,037 44,377 1.008 49,413 52,049 1.053
by xing the web and anges with bolts and pads to distribute
 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841 33

Fig. 10. Experimental and FE results for rack sections.

Fig. 11. Experimental and FE results for non-perforated sections.

the load uniformly (pressure) over all the section. It has been
Table 3
proved that this special plate introduces in the end section of the Amplitudes of imperfection tested.
specimen a local restraint, equivalent to the one introduced by
welding the plate to the section [30]. Both end plates are ball- Column length (mm) Mode Amplitude mm Ref.
pinned, but with torsional restraint. The ball position denes the
250 Local 0.1t 0.18 [812]
force line. The relative position of the ball to the section can be 0.15t 0.27
adjusted so the optimal position of the force line to pass through w/200 0.4 [57]
the effective centre of gravity [6,7]. Force and displacement data 0.006w 0.48 [24]
0.34t 0.612 [1]
were recorded during tests.
0.7
For each length three tests have been done, the mean of the 0.5t 0.9 [11]
three failure loads has been calculated and the shape failure 0.64t 1.152
modes (Fig. 4) have been recorded in order to validate the f/50 1.38
0.94t 1.692
numerical model. 1t 1.8 [11]
The maximum experimental loads obtained for both cross- 2.2
sections (non-perforated and perforated) are summarised in 1.5t 2.7
Table 1. As can be seen in the sixth column, the Pult,Solid-to-Pult, 600 Distortional 0.1t 0.18 [810,18]
Rack ratio is quite a lot different depending on the length. It is
0.15t 0.27 [18,20]
w/200 0.4
also observed that, for a same column length, the failure modes
0.006w 0.48
(third and fth columns) do not always coincide in both cross- 0.34t 0.612
sections. 0.7
0.5t 0.9 [19]
0.64t 1.152 [20]
f/50 1.38 [5,7]
0.94t 1.692 [1]
5. Selection of column lengths 1t 1.8 [24,19,21]
2.2
1.5t 2.7 [19]
When a nonlinear nite element analysis is carried out to
2600 Global L/7500 0.347
predict the load carrying capacity of cold formed steel rack column
L/2000 1.3 [12,19,23,24]
is necessary to introduce a geometrical imperfection. One of the goals L/1500 1.73 [14,19,24,26,27]
of this investigation is to study the effect of this geometrical imperfec- L/1250 2.08 [18]
tion magnitude in a nonlinear nite element analysis for the local, L/1000 2.6 [3,4,7,9,12,18,19,21,2325]
L/750 3.47 [18,19,21,22]
distortional and global failures. The rst step of the investigation is to L/500 5.2
determine the three lengths of study. These lengths have to match with
one of the experimental column length compression tests (250400 Thickness: t 1.8 mm, Web: w 80 mm, Flange: f 69 mm, L: length.
34 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841
Fig. 12. FE results on racks depending on imperfection. Short columns exhibiting local buckling mode.
600800100012001500180022002600 mm) in order to com-
pare the experimental and numerical results.
A linear buckling analysis has been done to choose the lengths
where the failure mode is as pure as possible (local, distortional
or global). The linear buckling analysis is done through the general-
ised beam theory (GBT), because this theory allows calculating
the modal participation of each pure mode and reproducing the
experimental boundary conditions perfectly. The section studied
by GBT does not have any perforations because the GBT classic
formulation cannot handle the perforations. However, the linear
buckling analysis without perforations is useful to determinate the
right column length because the maximum modal participation of
each pure mode occurs at similar column lengths for perforated and
non-perforated sections.
The value of the critical load for lengths between 250 and
3000 mm of the linear buckling analysis are presented in Fig. 5.
In addition, Fig. 6 shows the evolution of the local (L), symmetrical
distortional (SD), anti-symmetrical distortional (AD) and global
(G) modal participation for the critical load.
As it can be seen in Fig. 6, the inuence of pure modes is
different for each column length. For lengths among 250 and
500 mm the participation of local mode is clearly relevant
although the participation of symmetrical distortional is also
important. For lengths between 550 and 700 mm the predominant
mode is symmetric distortional buckling, where the modal parti-
cipation is around 90%. For lengths longer than 700 mm the
critical mode is a combination of anti-symmetrical distortional
mode and global (exural-torsional) mode. It can be observed that
for lengths longer than 1800 mm the participation of anti-
symmetrical distortional mode is less than 5% [31]. As a consequence
of these results the three lengths to study can be correctly selected.
To analyse the inuence of the geometrical imperfection magnitude
for the local mode, the 250 column length has been selected. This is
the shortest experimental length, so it is the length with most local
mode participation. For the distortional mode, 600 mm column
length has been chosen because the predominant failure mode is
mainly distortional. For the global mode, the 2600 column length has
been selected because it is the longer experimental column tested
 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841
Fig. 13. FE results on racks depending on imperfection. Intermediate columns exhibiting distortional buckling mode.
with less anti-symmetrical distortional mode participation (less than
1%). So in Table 1 are represented the three column lengths taken (in
bold) as to be representative of the three quasi-pure modes.
6. Finite element analysis
6.1. Finite element modelling
The nite element analysis has been carried out with the
specic software ANSYS 13.0 [32].
6.1.1. Elements and mesh
To do the discretisation of the model, a 4-node shell element (6
degrees of freedom per node) has been used for the specimen and
an 8-node solid element (3 degrees of freedom per node) for the
load plates. Shell elements are 5 mm sized. Shell and solid
elements employ a minimum of 5 and 8 integration points,
respectively. Shell elements use reduced integration scheme.
Fig. 7 shows sections tested and analysed.
35
6.1.2. Material model
The behaviour of the material has been dened by a bilinear
isotropic hardening material model, however the effects of using a
material model as multilinear to take in account the plastic
deformation have also been investigated. Fig. 8 shows a graph
where bilinear and multilinear models have been superimposed
on each other. Two multilinear material models have been
dened; one by taking the 0.2% yield strength (fy) as the transition
from elastic to plastic behaviour, and the other by taking the
proportional limit (fp) as the departure from the linear elastic
region. Proportional limit is dened as that stress at which
residual permanent strain is 0.02%.In both cases the plastic region
follows the true tensile stressstrain curve of the material.
Fig. 9 shows the compression test results with the three types
of stressstrain curve used as the input model. These simulations
have been carried out with the geometric imperfections specied
in [6,7] (w/200, f/50 and L/1000, respectively).
There is no difference between the bilinear model and the
fy-multilinear model. All three models predict the same ultimate
strength in the case of long columns, whose failure mode is global,
what means that the ultimate load is reached in the linear elastic
region of the stress- strain curve. The most important difference
36 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841
Fig. 14. FE results on racks depending on imperfection. Long columns exhibiting global buckling mode.
occurs in short columns, whose failure mode is local, where the
fp-multilinear curve gives a slightly less rigid model than the two
others, because the material is working beyond the proportional
limit even though before reaching the ultimate load. It should also
be pointed out that the bilinear and fy-multilinear models predict
a slightly higher ultimate strength for short and intermediate
length columns (with local and distortional failure modes, respec-
tively), but for long columns the loaddisplacement response is
practically the same regardless of the stressstrain curve used as
the input model.
6.1.3. Boundary conditions
The connection between plate and upright is done by coupling
the three free degrees of displacement of the upright's end nodes
with the nodes located at the same position of the load plates. In
this way the distortion of the end sections of the upright is
restrained. Two displacements of the symmetry axis of each plate
are also blocked to prevent its rotation about the longitudinal axis.
Finally, in one of the plates, the three free degrees are blocked at
node located at the effective centre of gravity, and in the other the
displacement in the plane of the plate are blocked an increasing
force in the direction of the upright's axis is imposed.
6.1.4. Solution control
As it has been previously mentioned, rst a linear buckling
analysis is performed. Different mathematical methods can be
used to quickly nd the buckling modes. Subspace is the method
used to calculate them, although the block Lanczos method has
also been used with the same results. Subsequently, a nonlinear
analysis is carried out starting from a perturbed mesh as a current
conguration, where the displacements from the previous analysis
are to be scaled as initial imperfection. To solve the nonlinear
analysis the arc length technique has been used instead of
incremental-iterative NewtonRaphson methods. The arc length
approach is useful to deal with snapping phenomena.
6.1.5. Calibration and validation of numerical model
The actual data of geometry (thickness) and material (yielding and
ultimate stress) have been used in the numerical model. So the results
can be compared directly with the mean value of the experimental
tests. The magnitudes specied in [6,7] (w/200, f/50 and L/1000,
respectively) have been used as initial geometric imperfections.
The comparison of experimental and numerical failure loads is
shown in Table 2 and Figs. 10 and 11, where it can be seen that the
FE predictions are in good agreement with the experimental tests,
M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841 37

Fig. 15. FE results on sections without perforations.

38 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841

Fig. 16. Numerical and experimental results of short column. (a) Numerical and experimental curves, and (b) Numerical curves and experimental ultimate loads.

Fig. 17. Numerical and experimental results of intermediate column. (a) Numerical and experimental curves and (b) numerical curves and experimental ultimate loads.

Fig. 18. Numerical and experimental results of long column. (a) Numerical and experimental curves and (b) numerical curves and experimental ultimate loads.

for both cross-sections (non-perforated and storage rack mem-
bers) in the full range of lengths.
6.2. Imperfection values tested
For use in nonlinear FE modelling, the rst Eigenmode (linear
buckling mode) representing the imperfect geometry of the
structural member is used in conjunction with the amplitude of
imperfection.
A wide range of amplitudes has been tested, which includes most
of geometric imperfections reported in Section 2. The maximum
amplitudes are usually taken as a percentage of plate thickness,
cross-section width or height, or specimen length in the case of
long columns. Table 3 summarises the imperfection values tested.
 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841 39

Fig. 19. Numerical and experimental results of non-perforated sections. (a) Numerical and experimental curves and (b) numerical curves and experimental ultimate loads.

As it can be seen in Table 3, some imperfection magnitudes
have been included in the study besides those mentioned in
Section 2 with the purpose of checking a similar interval of
amplitudes in local, distortional and global modes (the ratio of
the largest to smallest imperfection values is 15 for the three
column lengths analysed).
6.3. Nonlinear FE results
elastically deformed or, on the contrary, the material deforms
plastically beyond the yield stress (fy 423 MPa).
6.3.2. Solid (non-perforated) sections
The FE results on sections without perforations can be seen in
Fig. 15 in a similar way to those shown in Figs. 12b, 13b and 14b
for rack sections. The discussion of results is given in the next
Section 7.

6.4. Comparison of numerical and experimental results

The analyses have been performed in accordance with Section
6.1. The behaviour of the material has been dened as isotropic,
As it was explained in the previous Section 4, an experimental
with a bilinear stressstrain curve. The nite element model is the
campaign of compression tests was conducted by the authors on a
same in all cases, with the exception of the maximum initial
wide range of column lengths, from 250 mm to 2600 mm. Those
geometric imperfection magnitudes.
results provide valuable information for the assessment of the
initial geometric imperfections to be applied in the FE nonlinear
modelling. Figs. 16a, 17a and 18a show the numerical and experi-
6.3.1. Rack sections
mental loaddisplacement curves for the three lengths of rack
Figs. 1214 show the loaddisplacement curve obtained from
columns analysed. In Figs. 16b, 17b and 18b the ultimate load is
the FE nonlinear compression test on rack sections, for L 250, 600
depicted as a solid square.
and 2600 mm, respectively. In Figs. 12a, 13a and 14a all the curves
The numerical and experimental results in the case of solid
are superposed, showing the ultimate loads for the smallest
(without perforations) sections can be seen in Fig. 19.
imperfection (the highest curve) and the largest imperfection
(the lowest curve). In Figs. 12b, 13b and 14b only the extreme
curves and that obtained by using the values specied in [7] are 7. Discussion of results
depicted, together with the deformed shapes of the section at the
ultimate load and post buckling range. The corresponding Von Table 4 compares experimental ultimate load with predicted
Misses stress is also shown, which suggests when the material is numerically and demonstrates its sensitivity to imperfection
40 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841

amplitude. Tests were conducted according to the EN 15512: columns, both rack and solid sections, reach lower stress levels
2009 [29] and described above (Section 4). than short and intermediate columns, for similar deformation
Differences observed from graphs in Figs. 1215 are sum- levels.
marised in Table 5. The distortional buckling mode presents the As far as the ultimate load capacity is concerned, the ampli-
highest sensitivity to the size of imperfection. tudes of imperfection that better t with the experimental results
With respect to the values of equivalent Von Mises stress at are the following:
ultimate load in each case, the results and comments are sum-
marised in Table 6. As it can be seen, the long columns reach the  Short columns exhibiting local buckling mode: imperfections
ultimate load in the elastic range of the material. equal to or greater than 0.5t underestimate the load carrying
Table 7 shows the equivalent Von Mises stress in the capacity of the column for both rack and solid sections. The
post-buckling range, where column displays unstable collapse amplitudes recommended would be w/200, 0.006w and 0.34t.
with partial plastic deformation of the section. The long This conclusion agrees with [33]. The smallest imperfections
(0.1t and 0.15t) provide a post-buckling strength higher in the
case of rack columns.
Table 4  Intermediate columns exhibiting distortional buckling mode:
Comparison between test and FEM results.
imperfections less than 0.5t overestimate the load carrying
Section Column Mode Amplitude mm Pult,FEM Pult,EXP Pult, capacity of the column for both rack and solid sections, and
length FEM/ besides show the snap-through phenomenon after the max-
Pult,EXP imum load is reached. The amplitudes recommended would be
0.5t, 0.64t and f/50. Imperfections above 1t underestimate the
Rack 250 mm Local 0.1t 0.18 144,395 143,641 1.005
0.15t 0.27 144,355 1.005 maximum load.
w/200 0.4 142,715 0.994  Long columns exhibiting global buckling mode: the interval of
0.006w 0.48 141,585 0.986 imperfections recommended to be used is L/1000L/500. Spe-
0.34t 0.612 139,974 0.974 cically L/750 (Fig. 18, Racks) and L/500 (Fig. 19, Solid) are those
0.5t 0.9 137,365 0.956
0.64t 1.152 134,870 0.939
that better match the experimental loads.
f/50 1.38 133,770 0.931
1t 1.8 131,595 0.916
1.5t 2.7 127,725 0.889
Minor differences have been found between perforated
600 mm Distortional 0.1t 0.18 136,760 125,425 1.090
0.15t 0.27 134,545 1.073
(Racks) and non-perforated (Solid) sections where this matter is
w/200 0.4 132,265 1.055 concerned.
0.006w 0.48 130,295 1.039
0.5t 0.9 123,805 0.987
0.64t 1.152 119,520 0.953
Table 6
f/50 1.38 117,070 0.933
Equivalent Von Mises stress at ultimate load.
0.94t 1.692 113,872 0.908
1t 1.8 112,590 0.898
Type of section seqVM (MPa) at Pult Range of deformation
1.5t 2.7 104,495 0.833
(fy 423 MPa)
2600 mm Global L/7500 0.347 49,031 44,009 1.114
L D G
L/2000 1.3 46,556 1.058
L/1500 1.73 46,150 1.049
Rack 437456 427439 256423 Plastic (L, D). Elastic (G)
L/1250 2.08 45,270 1.029
Solid 429432 428455 238400 Plastic (L, D). Elastic (G)
L/1000 2.6 44,377 1.008
L/750 3.47 43,450 0.987
L/500 5.2 41,460 0.942

Solid 250 mm Local 0.1t 0.18 172,710 177,629 0.972

w/200 0.4 171,965 0.968
1.5t 2.7 144,120 0.811 Table 7
600 mm Distortional 0.1t 0.18 163,290 138,580 1.178 Equivalent Von Mises stress in the plastic range.
f/50 1.38 141,730 1.023
1.5t 2.7 127,265 0.918 Type of section seqVM (MPa) Displacement (mm)
2600 mm Global L/7500 0.347 56,629 49,394 1.146
L/1000 2.6 52,049 1.054 L D G L D G
L/750 3.47 50,873 1.030
L/500 5.2 48,802 0.988 Rack 545576 576552 433431 1.5 1.7 3.0
Solid 549576 590572 432431 1.5 1.7 3.0
Thickness: t 1.8 mm, Web: w 80 mm, Flange: f 69 mm, L: length.

Table 5
Inuence of amplitude on ultimate load, stiffness and post-buckling behaviour.

Type of section Pult,emin/Pult,emax Stiffness and post-buckling range

L D G

Rack 1.131 1.309 1.183 Section stiffness does not vary signicantly with imperfection values.

Solid 1.198 1.283 1.160 In distortional buckling, a sudden drop in load is experienced after attaining the ultimate load when very
small imperfections are used.
In global buckling, very small imperfections result in a non-realistic curve.

emin The smallest amplitude of imperfection.

emax The largest amplitude of imperfection.
 M.M. Pastor et al. / Thin-Walled Structures 76 (2014) 2841
8. Conclusions
 Imperfection amplitudes not only cause differences in the value
of ultimate load, but also can inuence the post-buckling
strength of the upright.
 Distortional mode is more sensitive to the magnitude of initial
imperfection. For the range of values tested, differences have
been detected up to 31% in the value of ultimate load.
 From this analysis is concluded that the imperfection values
used by the authors (w/200 (L), f/50 (D) and L/1000 (G)) predict
the ultimate load correctly. However, in the case of global
mode, L/1000 overestimates the strength capacity of the
column (see Table 2).
 For values of imperfection smaller than those, the strength
capacity of the upright may be overestimated. Likewise, values
of imperfection bigger than those may underestimate the
strength capacity of the upright.
 Found no signicant differences in behaviour, in terms of initial
geometric imperfection, between perforated and non-per-
forated sections.
References
[1] Schafer BW, Li Z, Moen CD. Computational modeling of cold-formed steel.
Thin-Walled Struct 2010;48:75262.
[2] Schafer BW, Pekz T. Computational modeling of cold-formed steel char-
acterising geometric imperfections and residual stresses. J Constr Steel Res
1998;47(3):193210.
[3] Dubina D, Ungureanu V. Effect of imperfections on numerical simulation of
instability behaviour of cold-formed steel members. Thin-Walled Struct
2002;40:23962.
[4] Szabo IF, Dubina D. Recent research advances on the ECBL approach. Part II:
Interactive buckling of perforated sections. Thin-Walled Struct 2004;42:
195210.
[5] European Standard EN 1993-1-5:2006/AC:2009. Eurocode 3 design of steel
structures Part 15: Plated structural elements. Brussels. European Commit-
tee for standardization. 2009.
[6] Roure F, Pastor MM, Casafont M, Somalo MR. Stub column tests for racking
design: experimental testing, FE analysis and EC3. Thin-Walled Struct
2011;49:16784.
[7] Pastor MM, Casafont M, Roure F, Bonada J, Noguera J. Inuence of holes on the
behaviour of cold-formed steel sections under compression. SDSS'Rio; 2010.
[8] Dinis PB, Camotim D, Silvestre N. FEM-based analysis of the local-plate/
distortional mode interaction in cold-formed steel lipped channel columns.
Comput Struct 2007;85:146174.
[9] Camotim D, Dinis PB. Coupled instabilities with distortional buckling in cold-
formed steel lipped channelcolumns. Thin-Walled Struct 2011;49:56275.
[10] Silvestre N, Camotim D, Dinis PB. Post-buckling behaviour and direct strength
design of lipped channel columns experiencing local/distortional interaction.
J Constr Steel Res 2012;73:1230.
41
[11] Chou SM, Chai GB, Ling L. Finite element technique for design of stub columns.
Thin-Walled Struct 2000;37:97112.
[12] Gardner L. Nethercot D.A, Numerical modelling of stainless steel structural
components a consistent approach, J Struct Eng ASCE 130(10), 2004, 1586
1601.
[13] Yang D, Hancock GJ. Numerical simulation of high strength steel lipped
channel columns. Adv Steel Struct Vol. I, 2005:14954.
[14] Becque J, Rasmussen KJR. A numerical investigation of local-overall interaction
buckling of stainless steel lipped channel columns. J Constr Steel Res
2009;65:168593.
[15] Ashraf M, Gardner L, Nethercot DA. Finite element modelling of structural
stainless steel cross-sections. Thin-Walled Struct 2006;44:104862.
[16] Kwon YB, Kim NG, Hancock GJ. Compression tests of welded section columns
undergoing buckling interaction. J Constr Steel Res 2007;63:1590602.
[17] Yao Z, Rasmussen KJR. Inelastic local buckling behaviour of perforated plates
and sections under compression. Thin-Walled Struct 2012;61:4970.
[18] Dinis PB, Camotim D. Post-buckling behaviour and strength of cold-formed
steel lipped channel columns experiencing distortional/global interaction.
Comput Struct 2011;89:42234.
[19] Crisan A, Ungureanu V, Dubina D. Numerical analysis and model calibration
for perforated pallet rack uprights. CST. Dubrovnik; 2012.
[20] Yap D, Hancock G. Interaction buckling and postbuckling in the distortional
mode of thin-walled sections [Research report no R870]. The University of
Sidney; April 2006. issn:1833-2781.
[21] Crisan A, Ungureanu V, Dubina D. Behaviour of cold-formed steel perforated
sections in compression: Part 2 Numerical investigations and design
considerations. Thin-Walled Struct 2012;61:97105.
[22] EN 1090-2:2008. Execution of steel structures and aluminium structures
Part 2: Technical requirements for steel structures. CEN. Brussels; 2009.
[23] Dinis PB, Batista EM, Camotim D, dos Santos ES. Local-distortional-global
interaction in lipped channel columns. Experimental results, numerical
simulations and design considerations. Thin-Walled Struct 2012;61:213.
[24] Ungureanu V, Dubina D. Erosion effect of geometrical and material imperfec-
tions on the buckling strength of thin-walled cold-formed steel members. Adv
Steel Struct 2005;I:497504.
[25] ECCS, European recommendations for the design of light gauge steel mem-
bers. ECCS. Brussels; 1987.
[26] Becque J, Lecce M, Rasmussen KJR. The direct strength method for stainless
steel compression members. J Constr Steel Res 2008;64:12318.
[27] Bjorhovde R. Deterministic and probabilistic approaches to the strength of
steel columns [PhD dissertation]. PA: Lehigh University; 1972.
[28] Dinis PB, Camotim D. Direct strength method to predict the resistance of cold-
formed steel columns against local/distortional interactive failure. CST.
Dubrovnik; 2012.
[29] European Standard EN 15512:2009. Steel static storage systems adjustable
pallet racking systems principles for structural design. Brussels: European
Committee for Standardization; 2009.
[30] Casafont M, Roure F, Pastor MM, Somalo MR. Compression tests on uprights:
checks for the effects of distortional buckling. In: Proceedings of the 2nd ERF
workshop "Tests on racking systems". Barcelona, Spain; 2010.
[31] Bonada J, Casafont M, Roure F, Pastor MM. Selection of the initial geometrical
imperfection in nonlinear FE analysis of cold-formed steel rack columns. Thin-
Walled Struct 2012;51:99111.
[32] ANSYS release 13.0 documentation. Ansys Inc.
[33] Pastor MM, Casafont M, Roure F, Ferrer M. The effect of local geometric
imperfections on the non-linear analysis of stub columns. CC2009. Madeira.