Structures 4 (2015) 58–68

Contents lists available at ScienceDirect

Structures

journal homepage: http://www.elsevier.com/locate/structures

Structural modeling of cold-formed steel portal frames

Xi Zhang ⁎, Kim J.R. Rasmussen, Hao Zhang
School of Civil Engineering, the University of Sydney, Australia

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

Article history: The paper describes the modeling of pitched roof cold-formed steel portal frames with slender cross-sections.
Received 15 July 2015 Two types of finite element models are introduced: a shell finite element model and a modified beam finite
Received in revised form 19 October 2015 element model. The shell element model involves explicit modeling of each structural member and accounts
Accepted 19 October 2015
for the semi-rigid behavior of apex and eave joints by incorporating spring-like elements. The beam element
Available online 26 October 2015
model utilizes a reduced tangent rigidity method to account for cross-sectional instability. Both models are com-
Keywords:
pared with results of experimentally tested portal frames, and good agreement is demonstrated.
Steel portal frames © 2015 The Institution of Structural Engineers. Published by Elsevier Ltd. All rights reserved.
Shell and beam element models
Local and distortional buckling
Semi-rigid joints

1. Introduction can be used to study the interaction of local/distortional buckling and

Conventional beam finite element analysis assumes the cross-
section and remains unchanged and so cannot consider the effect of
cross-sectional instability such as local and/or distortional buckling. In
contrast, the Generalised Beam Theory (GBT) [1] was developed as a
beam-element-based analysis method capable of accounting for local
instability of the cross-section. The theory has been proven capable of
predicting interaction between local and overall buckling modes using
beam elements [2–4]. The main drawback of GBT is its complexity and
associated reluctant take-up by practicing engineers.
In this paper an alternative method [5] is used to include local/
distortional buckling effects in beam element models, where local/
distortional buckling deformations are considered by simply reducing
the rigidities of the section. The reduction of the axial rigidity (EA), the
flexural rigidities (EIz, EIy), and the warping rigidity (EIw), as well as
other rigidities, are determined by means of a priori finite element anal-
yses of short lengths of members, which produce average values of
reduced tangent rigidities for nominated lengths of section [5].
The purpose of this paper is to demonstrate the application of the
stiffness-based beam element method to portal frames. This is achieved
by comparing results obtained using the method with experiments and
predictions obtained using full shell element discretization. In the anal-
yses, material properties, loading, boundary conditions, and initial geo-
metric imperfections were the same as those in the experiments.
Representative load–deflection curves obtained from the analysis re-
sults are shown to agree closely with the tests. The calibrated models
⁎ Corresponding author.
E-mail addresses: xzha8132@uni.sydney.edu.au (X. Zhang),
kim.rasmussen@sydney.edu.au (K.J.R. Rasmussen), hao.zhang@sydney.edu.au (H. Zhang).
frame sway buckling failure.
2. Tests of portal frames
Portal frame tests [6] were carried out to investigate the effect of
cross-sectional instability on the two-dimensional frame behavior
(stiffness, sway deflection, ultimate capacity, etc). Thus, the frames
were designed to ensure that local/distortional buckling developed at
an early stage, and that large sway displacements occurred in the
tests. Besides, through discrete lateral and torsional restraints, the
frames were restrained to deform in-plane.
Three pitched roof portal frames with the same nominal geometry
were assembled for testing, and were named as Frames 1, 2 and 3.
Frames 1 and 2 were subjected to nominal vertical loading only whereas
Frame 3 was subjected to a combination of horizontal and vertical loads.
The column and rafter sections are shown in Fig. 1(a) and (b), respec-
tively. The cross-section of the column was relatively slender, with a
web depth-to-thickness ratio of 183, and a flange width-to-thickness
ratio of 67. The latter ratio was slightly larger than the maximum ratio
(60) specified for flange elements in Section 2.1.3 of AS/NZS 4600 [7].
The cross-section of the rafter was stockier than that of the column,
which caused local buckling to develop in the column and not the rafter.
The apex and eave joints which were similar to those tested by Lim
and Nethercot [8] and Chung and Lau [9] featured brackets bolted to
both the webs and flanges of the back-to-back channel sections forming
the rafters and columns. Grade 350 mild steel 6 mm plates were used as
brackets. The apex and eave joint brackets were cut at a 152° angle and
104° angle respectively. The nominal diameters of the bolts and bolt
holes were 16 mm and 18 mm for the webs, and 8 mm and 10 mm
for the flanges. The bolt edge distances were 50 mm for the apex joint

http://dx.doi.org/10.1016/j.istruc.2015.10.010
2352-0124/© 2015 The Institution of Structural Engineers. Published by Elsevier Ltd. All rights reserved.
 X. Zhang et al. / Structures 4 (2015) 58–68 59

t = 1.2 t = 1.9 Based on the theory presented in [10], the stiffness matrix (Kl) and
20 internal force vector (p) in the local system are defined as,
30
Z Z
Kl ¼ NTδd2 GNδd2 dx þ NTδd2 NTδd1 St Nδd1 Nδd2 dx ð1Þ
220 220 L0 L0

Z
p¼ NTδd2 NTδd1 D dx: ð2Þ
L0

80 76
The matrix St, which accounts for the reduction of tangent rigidities
(a) Nominal dimensions of (b) Nominal dimensions of during the analysis, is termed as the tangential rigidity matrix and is
column section rafter section defined as,
(Dimensions in mm) 2     3
ðEAÞt ðESz Þt ES EI ðESw Þt 0
6 ðES Þ  y t  p t
6  z t ðEI Þ EIyz t EIpz t ðEIzw Þt 0 7 7
Fig. 1. Column and rafter cross-sections. 6  z t       7
6 ES EI EI EI EI 0 7
St ¼ 6  y t  yz t  y t  py t  yw t 7 ð3Þ
6 EI p t EI pz t EI EI EIpw t 0 7
6  py t  pp t 7
and 65 mm for the eave joints. Grade 8.8 M16 and Grade 8.8 M8 bolts 4 ðESw Þ ðEIzw Þt EIyw t EIpw t ðEIw Þt 0 5
t
were used for the webs and flanges, respectively. Figs. 2 and 3 show 0 0 0 0 0 ðGJ Þt
the details of the apex and eave joints respectively.
Fig. 4 shows a general layout of the test frames. Each frame had a where the tangent rigidity terms ((EA)t, (EIz)t, (EIy)t, etc.) are reduced
span of 8 m and a total height of 5 m from the base plate to the centre when local and/or distortional buckling develops. Therefore, the funda-
of the apex joint. The height from the base plate to the centre of the mental approach of this method is to find the appropriate reduction
eave joints was 4 m, and the frame had a 14° pitch. A pinned base was factors (τg) for each tangent rigidity term, so that
used and point loads were applied in the tests. Further details of the 2          3
tests can be found in [6]. τ g EA EA τg ESz τg ESy τg EI p τg ESw 0
6   ESz  ESy  EIp  ESw 7
6 τ g ESz ESz τ g EIz EI z τ g EIyz EI yz τ g EIpz EI pz τ g EIzw EI zw 0 7
6          7
6 τg ESy τg EI yz τg EI y τg EI py τg EI yw 0 7
3. Beam element modeling of frames subject to local/distortional 6
St ¼ 6  ESy  EIyz  EIy  EIpy  EIyw 7
7
buckling 6 τ g EIp EI p τ g EIpz EI pz τ g EIpy EI py τ g EIpp EI pp τ g EIpw EI pw 0 7
6          7
6 τ τ g EIzw EI zw τ g EIyw EI yw τ g EIpw EI pw τ g EIw EI w 7
4 g ESw ESw 0
  5
The basic approach of the beam-element-based method involves 0 0 0 0 0 τ g GJ GJ
considering the primary effect of local/distortional buckling as the ð4Þ
reduction of member stiffness against overall compression, bending
and torsion. Consequently, the overall behavior of the structure was where the unreduced rigidities (EA, EIz, EIy, etc.) can be calculated based
achieved by using the stiffness of the locally/distortionally buckled on the geometry of the undistorted cross-section. The calculation of the
cross-section rather than the stiffness of the undistorted cross-section. reduction factors (τg) is described in [11].

152°
Bolt-group size 540 540
360x120 A
226

936

A
63 M8 bolt
30 Bracket (6mm)

M16 bolt 60

60
Channel section

Section A-A

Fig. 2. Geometry of apex joint.

60 X. Zhang et al. / Structures 4 (2015) 58–68

700

Bolt-group size 63 M8 bolt

300x120 Bracket (6mm)
30
700

0
82
A A
M16 bolt 60
Bolt-group size 60
400x120
Channel section

Section A-A

Fig. 3. Geometry of eave joint.

In the analysis, the geometry, boundary conditions, and material
properties of the models were the same as those used in the experi-
ments. Local and distortional buckling imperfections were incorporated
in the analysis by implementing an imperfection in the shape of the first
buckling mode in the a priori post-buckling analysis used to determine
the reduction factors (τg), see [10] for details. However, an amplitude
of 0.01 t (0.012 mm for the columns) was considered in the post-
buckling analysis which was smaller than the measured imperfections
[6].
In modeling the apex and eave joints, the semi-rigid behavior was
considered by inserting rotational spring elements at the centre of the
joints. Each rotational spring was of zero length and connected to two
coincident nodes, with one node belonging to each member. The rota-
tional stiffness of the springs in the apex and the eave joints was
adopted from the tests on these joints, as shown in Table 1 where the
parameters are defined in Fig. 5.
The effect of material yielding was considered while calculating
reduced tangent rigidities (EI)t and stress resultants in the a priori
post-local/distortional buckling analysis. In this material modeling,
predefined combinations of axial strain and major axis curvature were
assumed. Specifically, a range of pre-set axial strain to major axis curva-
ture ratios was selected, and for each ratio, the applied strain was
gradually increased and the tangent rigidities were determined at
each increment of strain. Thus the axial strain and major axis curvature
were assumed to increase proportionally in calculating the tangent
rigidities. Since the sequence of assumed combinations of axial strain
and major axis curvature was likely to differ from the actual sequence
experienced in the experiments, and since the tangent rigidities are

North South
Apex joint
800 Eave joint Horizontal load
1000 (only for Frame 3)
Lateral
restraint
Load bracket Pin connection

Main horizontal
loading beam
Strong
4000 Gravity load column
simulator

Bolt-group
size 200x120 Pin end Dead
bearing weight
Strong floor
150
75 8000
Dimensions in mm

Fig. 4. Portal frame test set-up.

 X. Zhang et al. / Structures 4 (2015) 58–68 61

45

40

35

30

Moment (kNm) 25 K3
20
K2
15

10

5 K1
ϕ1 ϕ2 ϕ3
0
0 0.01 0.02 0.03 0.04
Rotation (rad)

Fig. 5. Parameters representing the stiffness of eave and apex joints.

dependent on the strain history, it was implicitly assumed that the
tangent rigidities were not strongly strain path dependent. The tangent
rigidity values (EI)t and stress resultants were read from file and stored
as multi-dimensional arrays in the nonlinear beam-element analysis.
Previous mesh convergence analyses have shown that a number of six
elements per member were sufficient to produce accurate results
while also achieving computational efficiency. Therefore a mesh of six
elements per column and six elements per rafter was selected, resulting
in a total of 24 elements.
4. Shell element modeling
4.1. Contact of back-to-back channels
The commercial software Abaqus [12] was used to create shell ele-
ment models and perform geometric and material nonlinear shell finite
element analyses.
When modeling back-to-back channels, the connection between the
webs needs to be considered, as otherwise one channel may protrude
into the other and the two webs would slide independently of each
other which was contrary to the experimental tests, in which the
webs were connected intermittently using screws or bolts. In the shell
element model, discrete fasteners were not explicitly modeled. Rather,
surface-to-surface contact was defined between the outside surfaces
of the two webs. The normal behavior of the contact was defined as
“hard”, which allowed no penetration between two surfaces, and tan-
gential behavior was defined as “rough” to indicate that the friction be-
tween two surfaces was infinitely large, i.e. friction was used to model
the effect of fasteners. Because longitudinal slip was prevented, full
composite action of the two channels was assumed over the contact
region.
The surface-to-surface contact was only applied in regions where
there was no gap between the webs. In the joint regions, brackets
inserted between the webs and fasteners were modeled as shell ele-
ments, where “tie” constraints were applied between each surface
of the brackets and the webs. For regions with less than 6 mm gap,
point-to-point connector elements were used to connect corresponding
nodes of the two webs at fastener points. The connection method thus
described is shown in Fig. 6.
4.2. Restraints, loading and boundary conditions
In the tests, end plates and stiffeners were used at column bases and
loading points to spread the reaction forces and jack loads. These plates
were not explicitly modeled in the shell element model. Instead, “Rigid
Body” constraints were used to limit the motion of a group of nodes to
the rigid body motion defined by a reference node, and to restrict all
the other degrees of freedom of the nodes. By using this approach,
boundary conditions were applied to the reference nodes at the centre
of the column bases to simulate pinned ends and all other nodes at
the ends of the columns were constrained by the reference nodes. Sim-
ilarly, vertical loads were applied concentrically to reference nodes,
where the reference nodes were located at the centre of pins to repre-
sent loads from the jack and all the nodes located within 100 mm
from the reference nodes were rigidly constrained. The multi-point con-
straints used at the column bases and load point areas are shown in
Fig. 7.
The brackets used for the apex and eave joints were modeled on
their nominal dimensions. The bolt holes and bolts, however, were not
modeled explicitly. Instead, the semi-rigid behavior of the joints was
considered by incorporating rotational springs located at the intersec-
tion of the column and rafter centrelines. The bracket was divided into
Bracket Tie
Connector element
Channel section
Surface-to-surface
contact
Fig. 6. Method of connecting back-to-back C-sections.
62 X. Zhang et al. / Structures 4 (2015) 58–68

Reference node

Reference node

Fig. 7. Multi-point constraints at column base and loading area.

Table 1
Simplified moment–rotation relationship for apex and eave joints.

K1 (kN.m/rad) K2 (kN.m/rad) K3 (kN.m/rad) φ1 (rad) φ2 (rad) φ3 (rad)

Apex joint 6283 1320 −2249 0.0025 0.02 0.035

Eave joint 5945 483 −1754 0.0044 0.012 0.018

two parts, and each was coupled with a reference node using multi-
point restraints. The two coincident reference nodes were then connect-
ed by spring-like connector elements, which allowed equal in-plane
translation degrees of freedom of the two points to occur freely but
added springs to the in-plane rotational degree of freedom. The spring
characteristics were based on the moment–rotation curves obtained
from tests of apex and eave joints, as shown in Table 1. Details of the
apex and eave joint models are shown in Fig. 8.
4.3. Initial geometric imperfections
Recognizing that geometric imperfections in the shape of the critical
elastic buckling modes usually have the most detrimental effect on the
ultimate strength, a common procedure for including imperfections is
to superimpose factored buckling modes onto the perfect geometry.
Three types of imperfections dominant in the tested frames were mea-
sured: the out-of-plumb (global) imperfections of the frame as well as
the local and distortional buckling imperfections of the column section.
Out-of-plumb imperfections were determined according to the mea-
sured geometry of the test frames as described in [13]. The local and
distortional buckling imperfections varied along the length of the
member, and were generally largest at mid-span. Hence, the critical
local and distortional buckling modes obtained from eigenvalue
analyses were factored so that the maximum amplitudes were equal
to the average measured imperfections at mid-span as per [6]. The
member imperfection was not measured for reasons explained in [6].
The measured global, local and distortional imperfections of the two
columns of each frame (referred to as “North” and “South”) for all tests
are listed in Table 2. The out-of-plumb (global) imperfection refers to
the top of the column at the intersection between the centrelines of
the column and rafter. All three frames were initially leaning in the
southern direction and failed in this direction when the most highly
loaded southern column reached its ultimate capacity under combined
bending and compression.
In order to produce clear local, distortional, and global buckling
modes for incorporating imperfections, and to avoid any coupled eigen-
modes, buckling analyses were performed on frames with different col-
umn section thickness. In the first step, a large shell element thickness
(typically 10 mm) was chosen to ensure that global sway buckling
was the first buckling eigenmode without the influence of local or
distortional modes. In the second step, the shell thickness was reduced
to a medium value (such as 5 mm) so that distortional buckling was the
first buckling eigenmode. The thickness was further reduced to 1.2 mm
in the third step to ensure that local buckling was the first buckling ei-
genmode. The buckling displacements of each of the three analyses
were then scaled using the scaling factors shown in Table 2 and

Two reference nodes

with same coordinates

Fig. 8. Modeling of apex and eave joints.

 X. Zhang et al. / Structures 4 (2015) 58–68 63

Table 2 between the channel sections at the connection points. The entire
Imperfection magnitudes of global, local and distortional buckling modes. model consisted of 31,474 elements.
Frame 1 Frame 2 Frame 3

North South North South North South

5. Analysis results

Global 12 (S) 13 (S) 8 (S) 8 (S) 7 (S) 8 (S)

5.1. Comparisons of analytical and experimental results
Local 0.33 0.58 0.35 0.82 0.45 0.43
Distortional 0.66 0.81 0.45 1.35 1.42 0.98
Abaqus shell finite element analyses were performed as were
Note: All dimensions are in mm. S represents that the column out-of-plumb is to the south.
OpenSees beam finite element analyses using the models described in
Sections 3 and 4. The analysis results are compared with those of the
portal frame tests summarized in Section 2. Three cases related to initial
Table 3 imperfections were considered in the shell element analyses. In Case 1,
Local and distortional buckling loads obtained from tests and analyses. only initial distortional buckling imperfections and initial out-of-plumb
Frame 1 Frame 2 Frame 3 imperfections were considered; in Case 2, initial local and distortional
(kN) (kN) (kN) buckling imperfections as well as out-of-plumb imperfections were
Test local buckling 46 45 41 considered, whereas in Case 3, only initial out-of-plumb imperfections
Shell element local buckling–case 1 38 44 40 were considered. In the beam element analyses, only initial out-of-
Shell element local buckling–case 2 38 44 40 plumb imperfections were modeled while a local buckling imperfection
Shell element local buckling–case 3 38 47 40
with magnitude 0.01 t, where t is the thickness of the cross-section
Test distortional buckling 68 63 76
Shell element distortional buckling–case 1 70 60 76 modeled, was incorporated in the a priori nonlinear analysis conducted
Shell element distortional buckling–case 2 67 61 76 to determine tangential rigidities.
Shell element distortional buckling–case 3 70 77 76 Table 3 summarizes the local and distortional buckling loads derived
from the results of the shell element analysis, and compares the loads to
the experimental results. In all shell element analyses, local buckling
superimposed onto the geometry of the geometrically perfect frame. It formed at an early stage, at loads comparable to those observed in the
should be noted that imperfections were incorporated for all columns tests, whereas distortional buckling developed when the applied load
and rafters in the analyses. was close to the ultimate load. For Frames 1 and 3, which featured rela-
tively small initial local and distortional buckling imperfections in the
critical south column, the local and distortional buckling loads were
4.4. Element types and mesh similar for the three imperfection cases in the shell element analyses
and the tests, whereas for Frame 2, which had large initial distortional
The element mesh used in the shell finite element analysis should be buckling imperfections in the critical south column, the results for im-
fine enough to accurately model the local/distortional buckling defor- perfection Case 3 overestimated the local and distortional buckling
mations and spread of plasticity. A preliminary convergence analysis loads because the model ignored initial geometric imperfections in
indicated that a mesh of 8 elements per web, 3 elements per flange these modes.
and 1 element per lip was sufficiently fine while also computationally In both experiments and shell element analyses, local buckling
efficient, featuring a maximum aspect ratio of 1.4 for the elements. deformations occurred over the full length of the frame, whereas distor-
The S4R element was selected from the Abaqus element library tional buckling deformations only occurred in the top half of the column
for the analyses. It is a 4-node shell element with reduced integration, members which were predominantly subject to bending. The failure of
featuring three translational and three rotational degrees of freedom all frames in the shell element analyses was due to the formation of a
at each node. It accounts for finite membrane strains and arbitrarily spatial plastic mechanism which was similar to the failure mode
large rotations. It is suitable for large strain analyses and geometrically observed in the tests. Figs. 9 and 10 show the local and distortional
nonlinear problems. The numbers of elements in the webs, flanges, buckling deformations as well as the spatial plastic mechanism that
and lip stiffeners were 9952, 3619, and 1357, respectively. This implies developed in the test and the finite element model for Frame 2,
an element width of approximately 25 mm for the flat parts of the cross- respectively. It can be seen that the deformations predicted by the
sections. The length of the elements was approximately 25 as well, finite element analysis are in close agreement with those observed
aiming for an aspect ratio of unity. This mesh resulted in a total of experimentally.
7488 elements for each column and 7440 elements for each rafter. Table 4 compares the ultimate loads obtained from the analyses with
The element size was approximately 35 mm for brackets inserted the experimental results. Good agreement with maximum error of 5% is

Local buckling Distortional buckling Spatial plastic

mechanism

(a) (b) (c)

Fig. 9. Local buckling, distortional buckling and spatial plastic mechanism in the test of Frame 2.
64 X. Zhang et al. / Structures 4 (2015) 58–68
Local buckling
Distortional buckling
(a)
Fig. 10. Local buckling, distortional buckling and spatial plastic mechanism in the finite element analysis of Frame 2.
observed between the tests and shell element analyses for Frames 1 and
3 for all imperfection cases, and for Frame 2 for imperfection Cases 1 and
2. For Frame 2 the discrepancy increases to 11% for imperfection Case 3
because of the large distortional imperfection present in the critical
column of this frame, which was not modeled in imperfection Case 3.
Thus, ignoring the initial local and distortional buckling imperfections
causes an overestimation of the ultimate load for Case 3 for Frame 2.
The ultimate loads predicted by the beam element analysis were consis-
tently higher than the test strengths and shell element strengths. The
difference relative to the total strength was 2%, 16% and 5% for Frames
1, 2 and 3 respectively. These differences are attributed to the smaller
magnitude (0.012 mm) of distortional buckling imperfection assumed
in the a priori shell element analysis used to determine tangent rigidities
((EA)t, (EIz)t, (EIy)t, (EIw)t) compared to the measured values of imper-
fection (Table 2). This applies particularly to Frame 2.
The applied load versus apex horizontal and vertical deflections for
the shell and beam element analyses are compared with the test results
in Figs. 11–16. Among the shell element results, the horizontal and ver-
tical responses are almost identical between Case 1 and Case 2. Results
from Case 3 show good agreement with Cases 1 and 2 for Frames 1
and 3, whereas for Frame 2, small discrepancies in the horizontal re-
sponse are observed at loads close to the ultimate load. On the basis of
these results, it is concluded that the impact of initial distortional buck-
ling imperfections on the results of the analysis is not significant when
the imperfections are small, but can potentially not be ignored when
they are of a certain magnitude, say greater than the thickness of the
profile. The initial local buckling imperfections can be ignored as the
effect of such is negligible.
The slopes of the horizontal and vertical deflection curves are similar
between the results of the beam element and shell element analyses,
which indicates that the determination of tangent rigidities in the
beam element analysis models was accurate. The ultimate loads of
the beam element analyses, however, are slightly higher than those of
the shell element analyses. Load–deflection curves extending well into
the post-ultimate range were not achieved for the beam element analy-
ses because the pre-determined tangent rigidities were difficult to ob-
tain when large scale yielding of the section had occurred, causing
convergence failure of the shell element post-local/distortional buckling
Table 4
Ultimate loads of tests and analyses.
Frame 1 (kN) Frame 2 (kN) Frame 3 (kN)
Test 87 76 76
Shell element–case 1 83 (0.95) 78 (1.03) 76 (1.00)
Shell element–case 2 83 (0.95) 78 (1.03) 76 (1.00)
Shell element–case 3 83 (0.95) 84 (1.11) 77 (1.01)
Beam element 89 (1.02) 88 (1.16) 80 (1.05)
(*) represents ultimate capacity relative to test strength.
Spatial plastic
mechanism
(b) (c)
analysis. Frequently, convergence failure occurred when large localized
deformations developed, signaling the formation of a spatial plastic
mechanism. At this point, the section capacity was reached and the
values of many tangent rigidities dropped to close to zero. Accordingly,
the beam element analysis could predict the pre-ultimate behavior and
the ultimate capacity of the member but only a small part of the post-
ultimate behavior. It follows that the load–displacement curves for the
beam element analysis shown in Figs. 11–16 do reach the ultimate
capacity although this may not be apparent from the curves which
appear to be increasing at the ultimate load.
The experimental load versus horizontal displacement curves for
Frames 1 and 2 exhibits a rapid growth in sway deflection after distor-
tional buckling occurred, whereas a more moderate growth is observed
in the analyses.
Figs. 11 and 13 indicate a significant difference in horizontal stiffness
between the analysis and experimental results for Frames 1 and 2. This
can be explained by the difference in stiffness of the two eave joints in
the tests. As per Section 5.2, in the tests, the effects of out-of-plumb
and different eave joint stiffness were to cause overall displacements
of the apex to occur in opposite directions and thus lead to a nearly
vertical slope of the horizontal displacement curve. However, in the
analyses, the stiffness of the two eave joints was assumed to be equal.
Since no horizontal load was applied to Frames 1 and 2, a difference in
joint stiffness may have caused obvious deviations between the experi-
mental and numerical results.
Fig. 12 indicates good agreement between the numerically predicted
and experimental load versus vertical apex displacement curves for
Frame 1 until the load reaches its ultimate value. The FE models exhibit
slightly lower stiffness at the initiation of loading but slightly higher
stiffness near the ultimate load compared to the test, while the
100
80
Applied load (kN)
60
Test
Shell element - Case 1
40 Shell element - Case 2
Shell element - Case 3
Beam element
20
0
0 5 10 15 20
Displacement (mm)
Fig. 11. Load vs apex horizontal deflection curves for Frame 1.
 X. Zhang et al. / Structures 4 (2015) 58–68 65

100 100

80 80

Vertical load (kN)

Vertical load (kN)
60 60

40 Test 40
Shell element - Case 1
Test
Shell element - Case 2
Shell element - Case 1
Shell element - Case 3
Shell element - Case 2 Beam element
Shell element - Case 3 20 20
Beam element

0 0
-60 -50 -40 -30 -20 -10 0 -40 -35 -30 -25 -20 -15 -10 -5 0
Displacement (mm) Displacement (mm)

Fig. 12. Load vs apex vertical deflection curves for Frame 1. Fig. 14. Load vs apex vertical deflection curves for Frame 2.

experimental and numerical vertical displacements near the ultimate
load are similar. Fig. 14 also shows good agreement between the vertical
displacements before reaching the ultimate load for Frame 2. Again the
FE model displayed more significant nonlinearity compared to the test
frame.
Good agreement is observed between test and analyses for the load
versus horizontal and load versus vertical apex displacement curves for
Frame 3, as shown in Figs. 15 and 16. The initial horizontal stiffness of
the experiment is higher but the horizontal deflection then increases,
and is close to that of the FE models when the applied load is more
than 20 kN. The ultimate loads of the shell element models and the
test frame are essentially equal, and good agreement in horizontal stiff-
ness is also achieved in the post-ultimate range. The ultimate load of the
beam element model, however, is 5% larger than that of the test frame.
5.2. Effect of unequal eave joint stiffness
In the portal frame tests, the stiffness of the two eave joints was not
identical because the brackets connecting the joints had slightly differ-
ent dimensions, as explained in [6]. Of the two eave joints, one was
the same as that used in the experiments conducted to determine the
joint stiffness, while the second was manufactured afterwards, and no
experiment was conducted to determine the stiffness of this joint. Dur-
ing the assembly of the test frames it was observed that the fit between
the second bracket and the adjoining members was substantially tighter
than the fit of the first bracket. This suggested that the rigidity of the
eave joint assembled with the second bracket was larger than that
measured in the eave joint tests.
Fig. 17a shows the linear deformations of a portal frame for which
the stiffness of the south eave joint is larger than that of the north
joint (KS N KN), as in the case of the test frames. The figure assumes
the out-of-plumb is zero. It follows that the apex displaces in the north-
ern direction. Fig. 17b shows the linear deformations of a portal frame
with an out-of-plumb in the southern direction, as in the case of the
test frames. In this figure, the joint stiffness of the northern and south-
ern eaves is assumed to be equal. As shown in Fig. 17b, the out-of-
plumb causes the apex to displace in the southern direction. Fig. 17c
shows the linear deformations of a frame with unequal eave stiffness
and out-of-plumb, i.e. a superposition of the deformations shown in
Fig. 15a and b, from which it follows that the horizontal apex displace-
ment is negligible initially.
In an attempt to determine the real stiffness of the south eave joint of
the test frames (all constructed using the second bracket), the stiffness
of this joint was increased by increasing the stiffness K1 (indicated in
Fig. 5) of the corresponding joint. The stiffnesses K2 and K3 did not
need to be changed because it was verified in the portal frame analyses
that the joint always fell within the load range of the first (elastic) slope
of the moment–rotation curve. It was therefore decided to systematical-
ly increase the stiffness K1 of the south eave joint for all frames and
compare the results with the tests to find the most appropriate K1
value. Table 5 shows the final moment–rotation relationship of the
south eave joint thus obtained, where the stiffness K1 of the south
eave joint was increased by 34%, while K2 and K3 remained constant.
Figs. 18–20 compare the horizontal deflections predicted by shell
element models with the experiments. It is obvious that overall, com-
pared to the shell element models with equal eave joint stiffness, the

100 100

80 80
Applied load (kN)

Vertical load (kN)

60 60
Test
Test Shell element - Case 1
40 Shell element - Case 1 40 Shell element - Case 2
Shell element - Case 2 Shell element - Case 3
Shell element - Case 3 Beam element
Beam element
20 20

0 0
-2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 -5 0 5 10 15 20 25 30 35
Displacement (mm) Displacement (mm)

Fig. 13. Load vs apex horizontal deflection curves for Frame 2. Fig. 15. Load vs apex horizontal deflection curves for Frame 3.
66 X. Zhang et al. / Structures 4 (2015) 58–68

100 Table 5
Moment–rotation relations of eave joints.

80 K1 K2 K3 φ1 φ2 φ3
(kN.m/rad) (kN.m/rad) (kN.m/rad) (rad) (rad) (rad)

Vertical load (kN)

North eave joint 5945 483 −1754 0.0044 0.012 0.018
60 (measured)
South eave joint 7945 483 −1754 0.0032 0.012 0.018
(adjusted)
40
Test
Shell element - Case 1
Shell element - Case 2 20
Shell element - Case 3
Beam element

0
-40 -35 -30 -25 -20 -15 -10 -5 0
Displacement (mm)

Fig. 16. Load vs apex vertical deflection curves for Frame 3.

horizontal apex displacement obtained using the shell element model

with unequal stiffness agrees better with the experiments.
For Frame 1, good agreement is obtained between the shell element
model with unequal stiffness and the experiment when the applied load
is less than 60 kN. However there is a slight difference at higher loads
where the experiment underwent larger horizontal displacements
than the FE model.
Similarly, for Frame 2, the displacement of the shell element model
with unequal stiffness agrees with the experiment when the applied
load is less than 50 kN, but afterwards the stiffness of the test frame
decreases significantly as a result of distortional buckling, leading to a
rapid increase in horizontal deflection. In the analysis, the applied load
increases to 70 kN before a rapid change in horizontal deflection occurs.
For Frame 3, the initial stiffness of the horizontal deflection curve for
unequal joint stiffness shows good agreement with the test, but a slight
difference is apparent after reaching a load of 40 kN. The ultimate load is
almost identical to that reached in the test.
The difference in initial stiffness between the shell element analyses
with unequal and equal eaves joint stiffness is pronounced for Frames 1
and 2, which indicates that the effect of changing the stiffness of an eave
joint for frames with no applied horizontal load is significant.
Fig. 18. Load vs apex horizontal deflection curves for Frame 1 with different eave stiffness.
Figs. 21–23 show the vertical apex displacements predicted by the
shell element models with equal and unequal eave joint stiffness as
well as the experimental curve. All frames indicate good agreement be-
tween the shell element models with unequal stiffness and experi-
ments, apart from the slight difference in vertical stiffness for Frame 2,
for which the test frame displays larger initial stiffness. Overall, it is
apparent that modeling eave joints with different stiffness does not
change the vertical displacement and ultimate load significantly.
6. Conclusions
Shell and beam finite element analysis methods for portal frame
structures constructed from thin-walled members subject to local and

Δ Δ

KN KS KN KS

N S N S

(a) KS > KN Δ≈ 0 (b) Out-of-plumb/ KS = KN

KN KS

N S

(c) Out-of-plumb/ K S > KN

Fig. 17. Deformations of portal frame under vertical load.
 X. Zhang et al. / Structures 4 (2015) 58–68 67

100

80

Vertical load (kN)

60

40
Test
ABAQUS - equal-stiffness-Case 1
ABAQUS - unequal-stiffness-Case 1 20

0
-40 -30 -20 -10 0
Displacement (mm)

Fig. 19. Load vs apex horizontal deflection curves for Frame 2 with different eave stiffness. Fig. 22. Load vs apex vertical deflection curves for Frame 2 with different eave stiffness.

100 determined from a separate prior analysis which incorporated local/

distortional imperfections.
80 Both methods accurately predicted the ultimate load and closely
traced the load–deflection characteristics. In particular, the beam ele-
Vertical load (kN)

ment model is capable of accurately analyzing frames with slender sec-

60 tions which are subject to local and/or distortional buckling. The beam
element analysis can be further modified to produce a tool for practicing
40 engineers that features simple expressions for the reduction of the
section rigidities caused by local/distortional buckling in a similar way
Test
ABAQUS - equal-stiffness-Case 1 to the τb-factors currently specified in the AISC-360 Specification to
20 ABAQUS - unequal-stiffness-Case 1 account for the reduction in rigidity caused by yielding. The shell finite
element analysis method can be used to assist in the development and
validation of this approach.
0
-5 0 5 10 15 20 25 The limitations of the presented beam element method can be seen
Displacement (mm) from Figs. 11 to 16. The method tends to be stiffer and the ultimate load
slightly higher compared to the test and shell element results. Also, the
Fig. 20. Load vs apex horizontal deflection curves for Frame 3 with different eave stiffness. method does not easily predict the post-ultimate behavior because of
numerical difficulties encountered when reaching the ultimate load or
distortional buckling prior to frame failure were assessed by a compar- slightly after reaching the ultimate load. However, it presents a compu-
ison with experimental results for three portal frame tests. The shell el- tationally attractive alternative to full shell element discretization and
ement analyses were performed using Abaqus, while the beam element is sufficiently simple to be readily adapted by practicing structural
analyses were performed in OpenSees using the method presented in engineers.
[8]. The semi-rigid behavior of apex and eave joints was accounted for
in both methods. Initial out-of-plumb, and local and distortional buck- Acknowledgment
ling imperfections were considered in the shell element analyses,
whereas only initial out-of-plumb was explicitly considered in This research is supported by Australian Research Council
the beam element analyses. The latter implicitly considered first- under Discovery Project Grant DP0989030. This support is gratefully
critical buckling mode imperfections by the use of tangent rigidities acknowledged.

100 100

80 80
Vertical load (kN)

Vertical load (kN)

60 60

40 40
Test Test
ABAQUS - equal-stiffness-Case 1 ABAQUS - equal-stiffness-Case 1
ABAQUS - unequal-stiffness-Case 1 20 20
ABAQUS - unequal-stiffness-Case 1

0 0
-60 -50 -40 -30 -20 -10 0 -35 -30 -25 -20 -15 -10 -5 0
Displacement (mm) Displacement (mm)

Fig. 21. Load vs apex vertical deflection curves for Frame 1 with different eave stiffnesses. Fig. 23. Load vs apex vertical deflection curves for Frame 3 with different eave stiffness.
68 X. Zhang et al. / Structures 4 (2015) 58–68
References
[1] Schardt R. Verallgemeinerte technische biegetheorie. Berlin, (in German): Springer-
Verlag; 1989.
[2] Davies JM, Leach P, Heinz D. Second-order generalised beam theory. J Constr Steel
Res 1994;31(2–3):221–41.
[3] Basaglia C, Camotim D, Silvestre N. Global buckling analysis of plane and space thin-
walled frames in the context of GBT. Thin-Walled Struct 2008;46(1):79–101.
[4] Basaglia C, Camotim DD, Silvestre NN. Post-buckling analysis of thin-walled steel
frames using generalised beam theory (GBT). Thin-Walled Struct 2013;62:229–42.
[5] Rasmussen KJR. Bifurcation of locally buckled members. Thin-Walled Struct 1997;
28(2):117–54.
[6] Zhang X. Steel Portal Frames with Locally Unstable Members. the University of
Sydney; 2014[PhD thesis].
[7] AS/NZS4600. Cold-formed Steel Structures, AS/NZS 4600. Sydney: Standards
Australia; 2005.
[8] Lim JBP, Nethercot DA. Ultimate strength of bolted moment-connections between
cold-formed steel members. Thin-Walled Struct 2003;41(11):1019–39.
[9] Chung KF, Lau L. Experimental investigation on bolted moment connections among
cold formed steel members. Eng Struct 1999;21(10):898–911.
[10] Zhang X, Rasmussen KJR, Zhang H. Beam-element-based analysis of locally and
distortionally buckled members: theory. [submitted for publication] Thin-Walled
Struct 2015.
[11] Zhang X, Rasmussen KJR. Analysis of locally/distortionally buckled beams. SSRC
Annual Stability Conference. St. Louis, MO; 2013.
[12] ABAQUS, Inc. ABAQUS. Ver. 6.8. analysis user's manual; 2009.
[13] Zhang X, Rasmussen KJR, Zhang H. Experimental investigation of locally and
distortionally buckled portal frames. [submitted for publication] J Constr Steel Res
2015.