Journal of Constructional Steel Research 95 (2014) 162–171

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

Cross-section slenderness limits for columns with plastic rotations

C.M. King a,b, J.B. Davison c,⁎
a
Steel Construction Institute, Silwood Park, Ascot, Berkshire SL5 7QN, UK
b
Buckland & Taylor Ltd., 101–788 Harbourside Drive, North Vancouver, BC V7P 3R7 Canada
c
Department of Civil and Structural Engineering, University of Sheffield, Sheffield S1 3JD, UK

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

Article history: This paper reports on a study of local inelastic buckling in square hollow section columns with large plastic rota-
Received 8 March 2013 tions. The study was conducted as part of the validation of a proposed design method for discontinuous columns
Accepted 26 November 2013 in braced frames in which plastic rotations in the columns are used to limit the moments in the columns. The
Available online 5 January 2014
study included both testing of full-scale columns and a parametric study by finite element analysis. The results
demonstrate that current codes permit cross section slenderness in plastic sections which are likely to lead to
Keywords:
Discontinuous column
premature buckling in structures using plastic (inelastic) design if the rotations are large. Design limits are pro-
Tube wall slenderness posed for square hollow sections relating cross-section slenderness to column end rotations.
Ductility requirements © 2013 Elsevier Ltd. All rights reserved.
Rotation capacity
Local buckling

1. Introduction of the joints (which means including the effect of axial compression)
or specify end-plates so thick that the joint is sensibly rigid even for
A new form of braced frame has recently appeared in Britain for res- low axial compression. In a continuous frame, the bending moments in
idential construction in which the columns are discontinuous [1]. Rather the columns calculated by elastic analysis can be of such a magnitude
than passing over a number of storeys, each column is only one storey that they cause a significant reduction in the resistance to axial compres-
high and is fitted with a base and cap plate to bolt to the beams below sion. To compensate for this, larger column areas are required, increasing
and above, as shown in Fig. 1 [2]. Columns are square hollow sections the bending stiffness and attracting more bending moment. This may
with the smallest possible external size so they can be hidden in the lead to heavy columns, negating one of the attractions of the construc-
thickness of the walls. The beams are continuous and pass over the tion method which is to have small column cross-sections to allow
top of the columns thus requiring little in fabrication yet benefitting them to be hidden in walls or limit the visual impact of exposed columns.
from the efficiency of continuity. However, this continuity of the The application of traditional design methods which might be used
beams may cause some rotations to be induced at the top and bottom to design a frame with discontinuous columns is unsatisfactory for a
of the column under certain loading arrangements resulting in curva- number of reasons. For example, the use of the ‘simple construction’
ture of the column, which would reduce the resistance of the column method [3] (in which the beam is assumed to be supported by a cap
below that of a pin-ended strut. plate) is compromised by the practical effects of using end plates of
The behaviour of discontinuous columns is significantly affected by sufficient thickness to satisfy UK building regulations tying capacities.
two issues (i) the stiffness of the column-beam joint and (ii) the effect This necessitates the addition of a moment in the column arising from
of bending moments in the columns on the compression resistance. At the column end rotation induced by the beam rotation (the stiffer
the top of a building, the axial compression in the columns is small and the connection the greater the column rotation) and the resulting
if relatively thin column end-plates are used, the connections will be calculated column capacities are relatively low. If joints are assumed
flexible so the beam can rotate relative to the columns. This would result to be rigid and elastic analysis is used, the column end moments will
in higher sagging moments in the beams than would be calculated in a be large thus lowering the calculated column capacity and relatively
rigid frame analysis. At the bottom of a building, the axial compression expensive connection details will be required to be consistent with
is high and this compression clamps the columns and beams so that the analysis model. To analyze such frames rigorously taking into
very little rotation of the beam relative to the column is possible, so the account joint flexibility requires considerable effort, making them
frame resembles a continuous one. If the frame is analyzed elastically economically unattractive to design offices. Other methods of design
as a continuous frame, the designer must either determine the stiffness might include designing the frame plastically (provided it is braced
independently) and allowing plastic hinges to develop in the
⁎ Corresponding author at: Department of Civil and Structural Engineering, University columns [1,4] or allowing for the semi-rigid nature of the joint in
of Sheffield, Sir Frederick Mappin Building, Mappin Street, Sheffield, S1 3JD, UK. an approximate manner. In reviewing available methods, early

0143-974X/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jcsr.2013.11.019
 C.M. King, J.B. Davison / Journal of Constructional Steel Research 95 (2014) 162–171 163

the yielding of the column allowed such large end rotations that the
columns “shed” the moments, as shown in Fig. 2(b).
As the axial compression was increased the end moments reduced to
zero and then changed to acting in the opposite direction to some small
amount before the member failed by flexural buckling in the plane
orthogonal to the plane of the web of the column, even when the end
rotations were applied in the plane of the web. It is important to note
that the end moment was applied by the turnbuckle system — it was
not applied as an eccentric load on the column; application as an eccen-
tric load does not allow the moment to reduce as the column ends rotate
and this is the design case assumed in codified checks of resistance to
combined axial compression and bending.
Gent and Milner's tests showed that at the Ultimate Limit State in
braced frames with rigid beam-column connections plasticity in the
column reduced column stiffness thus limiting the bending moment
attracted to the columns and also permitting more severe curvature in
the columns. When moments are applied to the columns by the
beams, the reduced column stiffness allows increased rotation of the
column ends, tending to shed the applied moments provided that the
beams can resist the moments shed by the columns. Gent and Milner
[6] observed “that even under biaxial bending restrained columns
have a remarkable capacity to sustain high axial loads by shedding
end moments”. Gent [5] wrote that “By considering limiting cases in
Fig. 1. Typical column-beam connection in discontinuous construction. this way, the design of the beams and the columns could largely be
divorced”. Although the papers propose a possible approach to design,
work by Gent and Milner provides an interesting approach, which is it is not developed into a complete method. Experimental work by
briefly outlined next. Davison et al. [7,8] and Gibbons et al. [9,10] on full size semi-rigidly con-
nected braced steel frames demonstrated this same phenomenon and
1.1. Gent & Milner's column research formed the basis of a design method which assumed the columns to
be pin ended and ignored the column moments because at the ultimate
Gent published a paper in 1966 [5], followed by a second with Milner limit state the beneficial restraining effect of the attached beams
in 1968 [6], describing tests on small scale steel I-section columns outweighed the detrimental effect of the diminishing moments as the
subject to an initial end-rotation and then to increasing axial compres- column buckles [11,12].
sion while the end-rotation remained applied. In the tests, end rotations
were imposed at the two ends of the columns by moments applied 1.2. Design using plastic rotations (moment shedding)
through short cantilever beams loaded at their ends through a turn-
buckle arrangement, as shown in Fig. 2(a). Importantly, this system A new design method for discontinuous columns in braced frames
allowed the moment to reduce as the column ends rotated, just as the was proposed by King [13] using moment shedding so that the columns
end moments of a fixed-ended beam reduce if the end restraints are are designed for zero end-moment even if the connection of the
allowed to rotate. Initially the column had no axial load applied. The columns to the beams is effectively rigid. The proposed method is for
axial load was then increased and the end moment resisted was mea- square hollow section columns, assuming that the full cross-section is
sured. The experiments showed that as the axial load was increased, effective. The columns are assumed to derive no stability from the

a b
Fig. 2. Gent & Milner (a) Experimental arrangement (b) Moment shedding from increasing axial load.
164 C.M. King, J.B. Davison / Journal of Constructional Steel Research 95 (2014) 162–171

Table 1 Table 2
Test programme. Material properties.

Test no. Specimen Comment Specimen Thickness (mm) Yield (N/mm2) UTS (N/mm2) Elongation %

kc1 120 SHS × 5 mm Pilot test kc1/2 5 396 548 27.5

kc2 120 SHS × 5 mm Pilot test continuation kc3 10 375 530 35.0
kc3 120 SHS × 10 mm Stable wall thickness kc4 10 390 530 35.0
kc4 120 SHS × 10 mm Stable wall thickness kc5 6.3 428 561 25.0
kc5 120 SHS × 6.3 mm Possible sensitivity to wall slenderness kc6 5 389 540 24.0
kc6 120 SHS × 5 mm wall Expect sensitivity to wall slenderness kc7 10 Not available Not available Not available
kc7 120 SHS × 10 mm wall Stable wall thickness kc8 10 375 514 36.5
kc8 120 SHS × 10 mm wall Stable wall thickness kc9 6.3 431 549 32.0
kc9 120 SHS × 6.3 mm wall Possible sensitivity to wall slenderness kc10 5 402 533 32.0
kc10 120 SHS × 5 mm wall Expect sensitivity to wall slenderness
and the shoe at the top of the column was similar. Articulation about a
cylindrical pin ensured that the deformation of the column was in one
adjacent beams and are assumed to be in single curvature because this
predetermined plane. The use of these shoes introduced some friction,
results in the lowest resistance to axial compression. The possible rigid-
ity of the connection is used to determine the end-rotation of the col-
umn in which the column is designed to resist. The columns are
designed to resist the applied axial compression plus the coexistent mo-
ment at mid-height resulting from the greatest end-rotation of the ad-
joining beam plus the additional moment arising from a design
imperfection. The design imperfection is calculated so that the resis-
tance of the discontinuous column cannot exceed the buckling resistance
of a pin-ended strut. The design model is equivalent to a rigid-plastic
pin-ended column with a plastic hinge at mid-height and with an offset
from the line of thrust.
To verify the proposed design method, a series of tests was required
both to demonstrate the resistance to axial compression and to establish
the requirements for slenderness of the cross-section (c/t in Eurocode 3
or b/t in most other design codes). The slenderness requirements of
existing codes had been noted as a possible issue by King [13] following
calculations using an approximate closed solution to local inelastic
buckling in the presence of high longitudinal plastic strains.

2. The test programme

The test programme is shown in Table 1. All specimens were Celsius1

355 120 × 120 Square Hollow Sections (SHS). As the purpose of the
testing was to provide data for use in validating the proposed design
approach, it was desirable to have more than one test so that repeatability
could be investigated. The most common applications of discontinuous
columns tend to require thick wall columns to reduce the column section
size to the minimum hence more tests were performed on sections with
10 mm wall thickness than with thinner walls. Testing at full scale
allowed the effect of slender walls to be investigated experimentally by
using both a thick and a thin walled section of the same serial size.
The yield stress of the test columns was measured from coupons cut
from most of the specimens after the tests were completed as shown in
Table 2.

2.1. Test rig and instrumentation

Full-scale specimens were tested in a hydraulic two-post rig as

shown in Fig. 3. Although the rig is normally operated under load con-
trol rather than displacement control, with skill the displacement can
be controlled by limiting the flow of oil from the pump to the jack.
Using this control technique, very smooth curves were recorded on
the falling branch of the plot of load versus displacement.
End-fittings or “shoes” were manufactured to fit to the ends of the
specimen without fastening thus avoiding fabrication work on the spec-
imen themselves, so the plane SHS sections supplied could be used
directly. The shoe at the bottom of the test column is shown in Fig. 4

1
Celsius is the brand name for structural grade hot-finished hollow sections produced
by Tata to EN 10210 in steel grade S355J2H. Fig. 3. Test rig with column in position.
 C.M. King, J.B. Davison / Journal of Constructional Steel Research 95 (2014) 162–171 165

The instrumentation on each specimen consisted of an in-plane and

an out-of-plane inclinometer at the top and bottom of the column and,
at column mid-height, two LVDTs to measure in-plane deflection
(one on each side of the column) and a single LVDT to measure de-
flection out of plane. The LVDTs were connected by wires to a spring-
grip frame (shown in Fig. 5) shaped to avoid displacements arising
from local deformations of the walls by fitting the section at all four
corners. The inclinometers were attached directly to the specimen at
200 mm from the centres of the pins. Only in-plane rotations were
measured in the pilot tests (kc1 and kc2) and tests kc8, 9 and 10.
Axial shortening was measured by the ram movement so included the
extension of the posts of the rig. The ram displacement within the
jack was measured by an externally mounted LVDT on the lower load-
ing platen and the jack force was determined by a pressure transducer
fitted to the hydraulic circuit of the machine.

2.2. Test results

The strain rate was as slow as reasonably practical with the test
equipment used. However, it proved impossible to limit the unloading
rate to be as slow as the loading rate. This is because, after maximum
load, the distance between the end-bearings of the test-rig is decreasing
but the overall length of the column is increasing as a result of the re-
duction in compressive strain. This increase in the overall length of
the column together with the decrease in distance between the end-
bearings causes an increase in the mid-height deflection which further
reduces the compression resistance of the column. The growth of the
mid-height deflection is a dynamic effect. Therefore some dynamic
effects were expected in the results measured after maximum load. To
give an indication of the scale of the dynamic effect, the duration of
the tests is listed in Table 3.
The test results proved to be broadly as predicted with all the tests
showing a long falling branch after maximum resistance. The general
Fig. 4. Shoe at bottom of test column, in-plane inclinometer and LVDT. form of the results was very similar for all the tests except for the loss
of strength due to local inelastic buckling in the thinnest wall sections
at end rotations greater than 40 mrad. Different aspects of the results
which was evaluated through unload/re-load cycles, and added 75 mm are discussed below.
to each end of the column i.e. the overall length of the strut was
2650 mm from centre of roller to centre of roller. This length is repre- 2.2.1. The unloading/re-loading cycle
sentative of domestic construction and the longest piece that could be The test curves confirm the shape of the falling branch as predicted
fitted within the available test-rig. by analysis. However, it was necessary to choose an arbitrary static
The shoes were adjusted to give an eccentricity of L/750 in the point on the falling branch of the test curve to allow a precise compari-
column to simulate the geometric imperfections that might be expe- son with the analyses which assumed static equilibrium. The static point
rienced in practical construction. The initial imperfections were mea- was established by unloading the test. The test was then continued to
sured and the maximum value recorded was L/1600. complete the test curve to the desired end-rotations.

2.2.2. In-plane displacements

Plots of load versus mid-height displacement in the plane of buckling
are shown in Fig. 6.
The mid-height displacements differ between tests during the initial
loading up to the maximum load. This is shown more clearly in the inset
plot in Fig. 6 which shows the displacements at a larger scale. This
shows that the tests cover a wide range of imperfections as might be
expected in practical construction.

2.2.3. Out-of-plane displacements

It was intended that the out-of-plane displacements should be limited
to small values relative to the in-plane displacement by the design of the
end fittings. The plots in Fig. 7, which show the out-of-plane displacement
plotted against the in-plane displacement for each test, shows that this
was achieved even when large in-plane displacements occurred along
the falling load-displacement branch seen in Fig. 6.

2.2.4. In-plane end rotations

Plots of axial load versus mean end-rotation (i.e. average of the top
Fig. 5. Spring grip frame for LVDT connections. and bottom values) are shown in Fig. 8. This shows that the load/end-
166 C.M. King, J.B. Davison / Journal of Constructional Steel Research 95 (2014) 162–171

Table 3
Duration of tests and peak recorded loads.

Test Time to reach maximum load (mins) Time from maximum load to end (mins) Total time (mins) Maximum test load (kN)

kc3 9.4 3.3 12.7 1328

kc4 13.8 5.0 18.8 1458
kc5 10.5 4.3 14.8 915
kc6 10.5 3.6 14.1 695
kc7 11.7 14.7 26.4 1290
kc8 9.9 7.4 17.3 1298
kc9 10.2 6.7 16.9 971
kc10 7.7 2.4 10.1 777

rotation from all the tests has long falling branches as expected. The 3. Effects of breadth to thickness ratios of wall
figure also shows as horizontal lines the design values of resistance of
the three sections when calculated to EN 1993-1-1 [14] as pin-ended It is well known that the stability of elements in compression
struts. depends upon the magnitude of the compressive stress and the slender-
The plots for the two tests on specimens with 5 mm wall thickness ness of the element. In design codes, the stability is verified by calculat-
(kc6 and kc10) show marked reductions in resistance below the general ing the breadth to thickness ratio (b/t) of the components of each cross-
trend due to local inelastic wall buckling, which was expected. It is in- section and comparing them with a limiting value. Currently, design
teresting to note that kc10 has a higher maximum resistance than kc6 codes rely on a single value of limiting breadth to thickness ratio for
but has a lower rotation capacity before the dramatic loss of resistance cross sections allowed to sustain plastic rotations. However, in members
at about 50 mrad. This suggests that the lower yield of kc6 produced a allowed to sustain plastic rotations, the curvature may be severe when
more uniform curvature which was lower than the curvature of kc10. loaded. In these circumstances, the local stability is dominated by the
Plots with the maximum load normalised are shown in Fig. 9. These out-of-plane force on the element caused by the axial compressive
plots show how similar the behaviour is up to the rotation at which local stresses and the curvature. Hence, a single limiting value cannot ensure
inelastic buckling of the wall precipitates a drastic loss of resistance. It is the local stability of the cross-section for all rotations. For cross-sections
difficult to identify the individual tests in Fig. 9 but the important point sustaining plastic rotations, design codes should either give a limit to
is that all the tests are similar except that the 5 mm wall tests (kc6 and the curvature for which the single value of b/t for plastic behaviour is
kc10) drop significantly from about 50 mrad. safe or they should give limiting values for different curvatures.
The effect of breadth to thickness ratios of the walls of hollow
sections was investigated by King [13] using a closed solution for local
2.3. Comparison of test data with the proposed design model inelastic buckling including the effects of plastic flow caused by longitu-
dinal strains exceeding yield strain. This investigation included a
The test results show that the moment shedding predicted by King 140 × 140 square hollow section, 3 m long, in single curvature with
[13] occurs in full-scale columns. Fig. 10 shows the results from tests end rotations of 40 mrad. This end rotation is possible at Ultimate
of two columns with the thickest walls compared with graphs derived Limit State (ULS) for beams with Permanent (Dead) Load that is about
from the simplified design method by King by calculating the design im- twice the Variable (Live) Load, using common deflection limits for the
perfection from the maximum compression resisted in the test instead Serviceability Limit State and accounting for some plasticity in the
of the compression resistance calculated from a design code. beam at ULS.

Fig. 6. Load v mid-height displacement (in-plane).

 C.M. King, J.B. Davison / Journal of Constructional Steel Research 95 (2014) 162–171 167

Fig. 7. Out-of-plane mid-height displacement v In-plane mid-height displacement. Fig. 9. Normalised axial load v mean end-rotation.

The investigation demonstrated that b/t must be increased for in- thickness. For members with yield stress of 355 MPa, this gives the
creased curvature of the member if the walls are to remain stable. This limiting b/t = 0.94√(200,000/355) = 22.3, where b is the internal
cast doubt on the reliability of a single value of b/t for plastic design, flat face.
as commonly used in modern design codes.

3.2. Test results

3.1. Slenderness limits in Codes of Practice
The range of wall thicknesses of the test columns was chosen to
3.1.1. Requirements of Eurocode 3
study the effect of the breath to thickness ratio as noted in Table 1.
For a hot-finished square hollow section designed to Eurocode 3
Taking the predictions by King [13] for a 140 × 140 square hollow
[14], the slenderness limits for sections is expressed as c/t = 33ε for
section 3 m long and multiplying the wall thickness by the ratio of the
Class 1, where c is the width of the internal flat face and ε = √(235/
member breadths, 120/140, the walls of the 120 × 120 × 6.3 SHS in
yield). According to Annex A3 of EN 10210–2 [15], the internal corner
tests kc5 and kc9 were expected to be stable at end-rotations greater
radius may be taken as 1.0 times the wall thickness. For members
than 40 mrad but the walls of the 120 × 120 × 5.0 SHS in tests kc6
with yield stress of 355 MPa, ε = √(235/355) = 0.814. Therefore, for
and kc10 were expected to be unstable at 40 mrad.
the flat face, c/t = 33 × 0.814 = 26.9 for Class 1.
The two columns with 5 mm wall thickness, kc6 and kc10, show a
clear drop in resistance to axial load arising from local inelastic buckling
3.1.2. Requirements of AISC 360–10 from about 55 and 40 mrad. This is seen in terms of end-rotations in
AISC 360–10 [16] has limits for b/t ratios in two sections. Chapter B, Fig. 11. The two columns with 6.3 mm wall thickness, kc5 and kc9,
Design Requirements, Table B4.1b, gives the compact/non-compact show only a very slight reduction in axial resistance in the tests. The
limit of 1.12√(E/Fy) for flanges of beams of rectangular HSS and boxes tests kc5 and kc9 show a small increase in the downward slope of the
of uniform thickness. For members with yield stress of 355 MPa, this axial load versus end-rotations curve from about 70 mrad as seen in
gives the limiting b/t = 1.12√(200,000/355) = 26.6, where b is the Fig. 11, indicating a reduction in resistance due to local inelastic
internal flat face. This is similar to the limit given in EN 1993-1-1 for buckling.
Class 1 members in general. For plastic design of columns, Appendix 1, There was no drop of resistance due to local inelastic buckling for the
Design by Inelastic Analysis, section 1.2.2b [16], gives the limit of 10 mm wall columns, even at the maximum test rotation.
0.94√(E/Fy) for flanges of rectangular HSS and boxes of uniform The onset of local inelastic buckling with end rotation is shown for
different b/t ratios in Table 4. Here the wall slenderness is calculated

Fig. 10. Comparison of experimental and simplified rigid-plastic design method axial load
Fig. 8. Axial load v mean end rotation. v mean end rotation.
168 C.M. King, J.B. Davison / Journal of Constructional Steel Research 95 (2014) 162–171

Fig. 11. Axial load v mean end rotation for 5 mm and 6.3 mm wall thicknesses.

as the internal flat width between internal radii (taken equal to the
wall thickness) divided by the wall thickness, so, for a 120 × 120 × 5
SHS, b/t = (120 − 4 ∗ 5)/5 = 20. The end rotations in the tests are
large at the point of local inelastic buckling, but the columns were in
single curvature. In real structures, columns are commonly in double
curvature, so the magnitude of the curvature can be expected to be
doubled. This reduces the stability of the wall for a given b/t ratio.

3.3. Parametric study for inelastic wall buckling

The design model proposed by King [13] assumes that the column is
in single curvature to simplify the calculation of overall buckling resis-
tance. However, many columns will be in double curvature, which re-
sults in higher curvatures than that in a column in single curvature
with the same magnitude of end rotation. Because the stability of the
wall depends on the out of plane force on the wall (which is the product
of the longitudinal force in the wall and the curvature of the wall) and
plastic flow, the effect of higher curvatures needed investigation to
find b/t ratios at which the walls of the section would be stable.
This effect was investigated by King [13] by a parametric study using
finite element analysis. Finite element models were made and used to
simulate the results of the tests on the 120 × 120 SHS sections. Abaqus
was used to conduct geometric non-linear and material non-linear
finite element analyses. The model (shown in Fig. 12) used shell ele-
ments with the nodes and the mid-thickness of the elements in the
plane of the centre-line of the walls of the column. This results in a
model with square corners, which is slightly different from the tight
radius corners of hot-formed SHS sections. The ends of the model are
connected to a “spider” of rigid-body elements, whose legs radiate to
the point of intersection of the centre-line of the column and the
plane of the end of the column. Most analyses were conducted using a
model of the entire column although the effect of mesh refinement
was checked by using a half-model that comprised a column cut longitu-
dinally along the centre-line of two opposite sides [13]. The magnitude
of the residual stress was not measured due to financial constraints
but was taken as 10% of the minimum specified yield stress of the sec-
tion, which is appropriate for hot-finished hollow sections, and followed
a bi-triangular pattern as illustrated in Fig. 13. The analysis assumed
elastic/perfectly-plastic non-linear material behaviour. Fig. 12. Abaqus FE model (shell elements).

Table 4
End rotation achieved before the onset of local inelastic buckling.

Test columns in single curvature:

Column length/external breadth, L/B = 22

b/t = internal flat width between radii/thickness

Test b/t Rotation achieved before local inelastic buckling Comment

kc6, kc10 20 0.040 rad Start of severe local buckling
kc5, kc9 15 0.070 rad Start of loss of stiffness just noticeable
kc3, kc4, kc7, kc8 8 0.095 rad No sign of loss of resistance before end of test
 C.M. King, J.B. Davison / Journal of Constructional Steel Research 95 (2014) 162–171 169

120 × 120 SHS sections because the performance had been shown to
be satisfactory by comparison with the full-scale tests. The same mesh
was used, but scaled in length and width to suit the desired length/
breadth ratios. The size of sections modelled was 140 × 140 SHS
sections because these had been identified as the largest sections used
in the frames recently constructed in the UK. Different thicknesses and
different member lengths were used to develop the greater curvatures
Fig. 13. Assumed residual stress pattern (same on all four faces). experienced in members in double curvature. The effects with end-
rotations in a rectangular plane were investigated for columns in single
The analyses were made using enforced shortening of pin-ended curvature of lengths 750 mm and 1500 mm to reproduce the behaviour
columns with the initial imperfection in the plane of the end-rotations. of 140 × 140 SHS columns in double curvature of 1500 mm and
For the simulation of the tests on the 120 × 120 SHS sections, the 3000 mm length. The results are shown in Fig. 14; it can be seen that
yield stress from the test coupons were used for each test. The only the wall stability is more demanding for 750 mm (the solid lines)
change to the geometry of the models was to adjust the eccentricity of than for 1500 mm (the broken lines). For example, for a 1500 mm
the application of load so that the end-rotations in the elastic range of long member with wall thickness 6.3 mm, the resistance starts to fall
the loading cycle were the same as those measured on test. This ensured away from the stable line (shown by the 12 mm wall thickness section)
that the model had the same effective eccentricity as the test specimen. at 20 mrad end rotation whereas for a 750 mm long member with
The resulting curves from the finite element analysis were generally 6.3 mm wall thickness, the resistance starts to fall away from the stable
close to the curves from the tests, especially in the range of higher line (shown by the 10 mm wall thickness member) at 30 mrad end
end-rotations, which is the region of interest for inelastic buckling of rotation.
the walls. The only exception was test kc10 which is discussed in Similar analyses were conducted for end-rotation in a plane at 45° to
more detail below. the rectangular planes as shown in Fig. 15. Comparison of Figs. 14 and
For the parametric study of inelastic wall buckling, the same finite 15 shows that the local inelastic buckling requirements for end-
element model was used as for the simulation of the tests on the rotations at 45° are almost identical to those for end-rotations in a

Fig. 14. Effect of in-plane end rotation on axial load resistance for a range of wall thicknesses.

Fig. 15. Effect of end rotation on a 45° plane on axial resistance for a range of wall thicknesses.
170 C.M. King, J.B. Davison / Journal of Constructional Steel Research 95 (2014) 162–171

Fig. 16. Design limits for 140 × 140SHS in 355 MPa steel in single curvature.

rectangular plane. The breadth to thickness ratio above which local the point where there is a radii fall in end rotation from 40 mrad. This
inelastic buckling impairs the column resistance is shown in Fig. 16 suggests that the finite element analysis model used may overestimate
plotted against end rotation. the end-rotations at which the walls remain stable for the thinner
walled sections. It is possible that the difference was caused by variations
in the wall thickness of the test specimen in contrast with the uniform
3.4. Comparison of test results with finite element analysis thickness assumed in the finite element model because specifications
for hollow sections allow more generous thickness tolerances than are
For studying the inelastic buckling of the wall, the most important commonly allowed for open sections.
part of the load-rotation curve is the region of highest end rotations. The close correlation between the load-rotation curves for tests and
In this region, the curves from the finite element analysis are generally analysis for the six tests with wall thicknesses of 10 mm and 6.3 mm
close to the curves from the tests. At the end of the test range, that is gives confidence that the finite element model performs well for the
at the highest end rotations, all the tests gave a slightly higher rotations region of high end-rotations. The difference between the two tests
for a given load except test kc10 which had a 5 mm wall thickness. with 5 mm walls suggests that the inelastic buckling of lower wall
Fig. 17 shows the two tests with 5 mm wall thickness, kc6 and kc10. It thicknesses is more variable than that for thicker walls.
can be seen that test kc6 gave higher end-rotations at a given load
than the finite element analysis, but kc10 gave lower end-rotations at
3.5. Proposed design limits

The proposed design limits are defined by the curve “b/t from FEA”
shown in Fig. 16. These limits were derived for 140 × 140 square
hollow sections of length 1500 mm in double curvature i.e. a length/
breadth ratio of 10.7. The limits for greater length/breadth ratios
would be less onerous because for a longer member the change of
angle between the ends is spread over a greater length thus reducing
the severity of the curvature and lowering the out-of-plane forces on
the cross-section walls. The numerical values are given in Table 5.
Given the difference between the finite element analysis and the test
curves for test kc10, the authors recommend reducing the limiting end
rotations shown in Table 5 by around 20% for b/t values greater than 20;
the revised values are shown in brackets in Table 5.

4. Conclusions

Tests have been performed on full-scale columns with different wall

thicknesses, using 120 × 120 SHS sections, which were the largest that
could be tested in the available test rig, with the load applied to maxi-
mum resistance and then the hydraulic ram extension was continued
to measure the load versus end-rotation beyond maximum resistance
of the column. The failure mode was single curvature flexural buckling.

Table 5
Proposed b/t limits for increasing end rotations.

b/t Limiting end rotation

26.0 10 (8)
20.2 20 (16)
15.5 36
12.0 58
Fig. 17. Comparison of experimental and FEA axial compression v end rotation for (a) Test
9.7 83
kc6 (b) Test kc10.
 C.M. King, J.B. Davison / Journal of Constructional Steel Research 95 (2014) 162–171 171

Sections with 10 mm wall thickness showed no local inelastic buckling References

over the full range of the tests, whereas sections with thinner walls
[1] British Constructional Steelwork Association/Steel Construction Institute. The use of dis-
displayed local inelastic buckling. Finite element analyses were made continuous columns in simple construction. AD281 New Steel Construction; February
of the 120 × 120 SHS sections in the rig using a non-linear geometry 2005.
solution and elastic–plastic material properties. The curves of load versus [2] British Constructional Steelwork Association/Steel Construction Institute. Discontin-
uous columns in simple construction: Beam-column connections. AD288 New Steel
end-rotation from the finite element analysis are generally close to the Construction; July 2005.
curves measured in the test rig. [3] British Constructional Steelwork Association/Steel Construction Institute. The use of
A parametric study of inelastic buckling related to wall slenderness discontinuous columns in simple construction. AD283 New Steel Construction;
March 2005.
was then conducted using the same finite element model scaled to in- [4] Wood RH. A new approach to column design. Building Research Establishment
vestigate the rotation capacity of members with ratios of length/breadth Report. Department of the Environment, HMSO; 1974.
as low as 10.7 failing in double curvature. The study shows that the [5] Gent AR. Elastic–plastic column stability and the design of no-sway frames. Proc Inst
Civil Eng 1966;34:129–52.
single values of limiting b/t in modern codes are not sufficient to avoid
[6] Gent AR, Milner HR. The ultimate load capacity of elastically restrained H-columns
local inelastic buckling when plastic (or inelastic) analysis is used if under biaxial bending. Proc Inst Civil Eng 1968;43:685–704.
the curvature of the members is high due to plasticity. The parametric [7] Davison JB, Kirby PA, Nethercot DA. Column behaviour in PR construction —
study was confined to square hollow sections, but the same issue of experimental behaviour. ASCE J Struct Eng 1987;13(9):2032–50.
[8] Davison JB. Strength of beam-columns in flexibly connected steel frames. Ph.D
decreased stability with increased local curvature will apply to all Thesis University of Sheffield; 1987.
cross-sections. [9] Gibbons C, Nethercot DA, Kirby PA, Wang YC. An appraisal of partially restrained
With the increasing use of material non-linear behaviour of struc- columns behaviour in non-sway steel frames. Proc Inst Civ Eng Struct Build
1993;99:15–28.
tural steel in industry, made possible by the wider use of finite element [10] Gibbons C, Kirby PA. Nethercot, DA. Experimental behaviour of partially restrained
analysis, design codes should either give a limit to the curvature for steel columns. Proc Inst Civ Eng Struct Build 1993;99:29–42.
which the single value of b/t for plastic behaviour is safe or they should [11] Gibbons C. The strength of biaxially loaded beam-columns in flexibly connected
steel frames. Ph.D. Thesis University of Sheffield; 1990.
give limiting values for different curvatures. [12] Carr JF. A simplified approach to the design of semi-rigidly connected columns in multi-
storey non-sway steel framed buildings. M. Phil Thesis University of Sheffield; 1993.
[13] King CM. Column design for axial compression and end rotation. PhD Thesis
University of Sheffield; June 2010.
Acknowledgements
[14] EN 1993–1–1. Eurocode 3: design of steel structures — Part 1.1: general rules and
rules for buildings. CEN (European Committee for Standardization); 2005.
The authors wish to thank the Steel Construction Institute for [15] EN 10210–2. Hot finished structural hollow sections of non-alloy fine grain steels.
Part 2: tolerances, dimensions and sectional properties. CEN (European Committee
funding the fees of the first author for the first two years of his PhD
for Standardization); 2006.
study and Corus (now Tata) for financial support of the experimental [16] AISC. Specification for structural steel buildings. ANSI/AISC 360–10. American
work. Institute of Steel Construction; 2010.