Journal of Constructional Steel Research 79 (2012) 9–21

Contents lists available at SciVerse ScienceDirect

Journal of Constructional Steel Research

Design rules for out-of-plane stability of roller bent steel arches with FEM
R.C. Spoorenberg a,⁎, H.H. Snijder a, J.C.D. Hoenderkamp a, D. Beg b
a
Eindhoven University of Technology, Faculty of Built Environment, P.O. Box 513, 5600 MB Eindhoven, The Netherlands
b
University of Ljubljana, Faculty of Civil and Geodetic Engineering, Jamova 2, 1000 Ljubljana, Slovenia

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

Article history: This paper describes a numerical investigation into the out-of-plane buckling behavior of freestanding roller
Received 14 May 2012 bent steel arches. As roller bent arches have structural imperfections which differ considerably from those of
Accepted 25 July 2012 hot-rolled or welded sections, specific attention is paid to their inclusion in the numerical model. Sensitivity
Available online xxxx
analyses are performed to assess the influence of the imperfections due to roller bending on the out-of-plane
buckling response. The accuracy of the finite element model is checked by comparing the results with earlier
Keywords:
Roller bent arch
performed experiments as presented in a related paper. The finite element model is able to replicate the
Out-of-plane buckling structural behavior displayed by the experiments with good accuracy. A database is created with elastic–
Finite element modeling plastic buckling loads for a large number of freestanding roller bent arches. The numerical data is analyzed
Design rules and presented in a so-called imperfection parameter diagram from which imperfection parameter curves
are derived. The imperfection parameter curves are substituted into the European column curve formulation,
leaving the original column curve formulation unaffected but extending its applicability to the out-of-plane
buckling response of roller bent arches. The column curve with proposed imperfection parameter expres-
sions can be used to check the out-of-plane buckling response of a roller bent steel arch with known
non-dimensional slenderness.
© 2012 Elsevier Ltd. All rights reserved.

1. Introduction 1.1. Previous studies on out-of-plane arch buckling

The application of roller bent steel has seen a steady increase in The earliest theoretical studies on out-of-plane arch buckling only
the construction industry over the past decades. Ease of manufactur- considered elastic buckling where material non-linearities and im-
ing make roller bending a suitable method for achieving curved struc- perfections were ignored. Valuable contributions were published by
tures. Roller bent steel is often applied in circular arch structures Timoshenko and Gere [2] and Vlasov [3] who provided formulae to
where its primary function lies in carrying the acting loads to the approximate the elastic out-of-plane buckling load of freestanding
abutments. The loads are resisted by means of a combination of com- arches. Further refinements to calculation procedures for approxi-
pression and bending, making the member susceptible to buckling. mating the elastic buckling load were proposed by Vacharajittiphan
When local buckling is not considered, arch instability can be and Trahair [4], Yoo [5] and Rajasekaran and Padmanabhan [6].
subdivided into three different categories: snap-through buckling The necessity to include material non‐linearities and imperfec-
(Fig. 1(a)), in-plane buckling (Fig. 1(b)) and out-of-plane buckling tions to obtain an accurate representation of out-of-plane buckling
(Fig. 1(c)). The latter occurs when an arch has no lateral bracing behavior of arches was recognized in Japan by the end of the 1970s.
and is considered ‘freestanding’. This paper presents a study of the Research studies included experiments conducted on arches with
structural performance of freestanding circular roller bent steel square hollow sections by Sakimoto et al. [7] and welded I-sections
arches by means of the finite element method, for which out-of- by Sakata and Sakimoto [8] supplemented with finite element analy-
plane buckling is the governing failure mode. The performance of ses by Komatsu and Sakimoto [9], Sakimoto and Komatsu [10] and
the finite element model is verified through comparison with exper- Sakimoto and Komatsu [11]. These Japanese research studies culmi-
imental results as reported in a related paper, La Poutré et al. [1]. nated in design rules. For the calculation of the slenderness, the
arch was treated as a straight column under uniform compression
with identical cross-section, where the arch length corresponded with
the column length. Column curves were proposed by Sakimoto et al.
[12] and Sakimoto and Sakata [13] to allow a check of the out-of-plane
arch stability. Their applicability was limited to arches with square hol-
⁎ Corresponding author at: Eindhoven University of Technology, Faculty of Built
Environment, Den Dolech 2, P.O. Box 513, 5600 MB Eindhoven, The Netherlands.
low sections and rise-to-span ratios between 0.1 and 0.2.
Tel.: + 31 40 247 2948; fax: + 31 40 245 0328. As the Japanese design provisions treated the out-of-plane arch
E-mail address: r.c.spoorenberg@bwk.tue.nl (R.C. Spoorenberg). buckling case identically to that of a column, the rise-to-span ratio

0143-974X/$ – see front matter © 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jcsr.2012.07.027
10 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21

(a) Snap-through (b) In-plane buckling (c) Out-of-plane buckling

F F F
F

Fig. 1. Global instability phenomena for arches.

of the arch was considered to be of minor importance. However, process on the structural properties of wide flange sections. Using
earlier theoretical studies revealed that the rise-to-span ratio can the existing design rules to check the out-of-plane buckling response
have a significant effect on arch buckling. This was recognized by of freestanding arches without taking into account the influence of
Papangelis and Trahair [14] who performed experiments on arch the roller bending process can lead to either conservative or
buckling. These experiments were used to validate an in-house finite unconservative designs. The main goal of this paper is two-fold: pro-
element code developed by Pi and Trahair [15] from which design viding numerical modeling techniques for roller bent steel members
rules for arch buckling were developed and proposed. Pi and Trahair and suggesting design rules for out-of-plane buckling of arches with
[16] stated that for pin-supported arches subjected to radial loading finite element analyses. The European column curve formulation as
and simply-supported arches subjected to uniform bending, the described in EN 1993-1-1 [19] (Eurocode3) will be adapted to include
Australian column curves were considered suitable for the design of the out-of-plane buckling of arches. The results from 3 different finite
arches failing by out-of-plane buckling. For out-of-plane fixed arches element analysis types will be used to express the numerical data in
a similar approach was used; Pi and Bradford [17]. According to Pi the column curve diagram to arrive at a design rule. Arches can be
and Trahair [16] and Pi and Bradford [17] the arch slenderness was subjected to a wide range of loading types. In general a distinction
defined as the square root of the ratio between the plastic capacity is made between symmetric and unsymmetric loads. The present fi-
and out-of-plane elastic buckling load, taking implicitly into account nite element study is limited to the former one. Design rules are pro-
the geometric properties of the arch. For arches subjected to vertical posed for a total of four load cases as shown in Fig. 2. The arch
loading, interaction formulae were proposed to check the out-of- geometry is shown in Fig. 2(a), where S is the arch length, R the radi-
plane stability. These interaction formulae are analogous to those of us, 2γ is the subtended angle, L is the span and f is the rise. Arches
a beam-column failing by elastic–plastic buckling. The interaction with two opposite end moments (Fig. 2(a)) and a radially distributed
formulae were valid for out-of-plane simply supported arches, Pi load (Fig. 2(b)) are rather academic load cases. These load cases serve
and Trahair [18] and out-of-plane fixed arches, Pi and Bradford [17]. for comparison with other load cases, as the internal forces are limit-
ed to uniform bending or uniform compression for an arch under two
1.2. Scope and aims opposite end moments or a radially distributed load, respectively.
Arches with a central point load (Fig. 2(c)) or uniformly distributed
It is clear that the out-of-plane buckling behavior of steel arches load (Fig. 2(d)) display a combination of internal bending moments
has received large attention, comprising analytical, numerical and and compressive forces. During the loading phase, the loads will un-
experimental studies. However, investigations involving material dergo no directional change and are hence termed gravity loading.
non-linearities and imperfections were limited to either welded The present study is limited to circular I-section arches made from
box-sections or wide-flange sections for which the influence of the either steel grade S235 or steel grade S355 bent about their major
roller bending process on the structural properties of the arch was axis through the roller bending process.
not taken into account. Earlier experiments and finite element analy- With the exception of arches subject to two opposite end mo-
ses by the authors have shown the influence of the roller bending ments, all arches are pin-supported, preventing outward spreading

(a) Two opposite end (b) Radially distributed load - Uniform

moments - Uniform bending compression
S q

M R M f
2γ

L
(c) Central point load (d) Uniformly distributed load
F
q

Fig. 2. Load cases under investigation.

 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21 11

of the support and introducing compressive action in the arch-rib. In the present finite element model Multi-Point-Constraint elements
Arches under uniform bending need to be simply supported, as (MPC184 in ANSYS) were used to ensure a smooth introduction of
pin-ended support conditions would induce a small normal force. the support reactions. The supports were applied at the mid-height
No in-plane fixed boundary conditions are considered. For all arches node of the web. The MPC elements applied over the cross-section
the torsional degree of freedom is fixed at the support (fork support). ensured proper transfer of the displacements and rotations of the
From preliminary finite element analyses it was found that arches mid-height node to the adjacent nodes. Restrained warping at the
with out-of-plane pin-ended conditions provided little resistance to support is included when applying MPCs over the web and flanges,
out-of-plane buckling. In order to compensate for this and hence thereby preventing separate rotation of the flanges (Fig. 4).
consider arches which can sustain sufficient loads before failing by The number of MPC elements is related to the shell element distri-
out-of-plane buckling all arches are fixed out-of-plane at the supports bution over the height of the web and width of the flanges. Loads are
and warping deformations are prevented (restrained warping). The applied at the centroid of the cross-section. The element mesh shown
behavior of the supports with respect to the local coordinate system in Fig. 3 and Fig. 4 is based on mesh refinement studies as presented
of the arch is outlined in Section 2.3. in Section 2.8.

2. Finite element modeling 2.4. Imperfections

The general purpose finite element package ANSYS v.11.0 is used 2.4.1. Residual stresses
to perform the numerical computations. All analyses are performed The residual stress model proposed by Spoorenberg et al. [20] is
in the implicit environment. used to define the initial stress state in the roller bent steel member
prior to loading (Fig. 5(a)). The compressive residual stresses and ten-
2.1. Types of analysis sile residual stresses are denoted negative and positive respectively. As
this residual stress model is presented in non-dimensional form, the
For each arch configuration three types of analysis are necessary nominal yield stress according to material specifications is used to ob-
to obtain the buckling response and to plot the finite element results tain stress values (fy = 235 N/mm2 for S235 and fy = 355 N/mm2 for
in the column curve diagram: Linear Buckling Analysis (LBA), Materi- S355). The residual stress values are applied at the integration points
al Non-linear Analysis (MNA) and Geometrical Material Non-linear of the shell element which coincide with the element centroid. As the
Imperfect Analysis (GMNIA). An additional Geometrical Non-linear residual stress value in the integration point defines the stress state
Imperfect Analysis (GNIA) was performed to obtain the load corre- for the whole element, a step-wise pattern is obtained for the finite
sponding to the onset of yielding. It should be noted that a LBA element model (Fig. 5(b)).
produces a simplified representation of buckling behavior of arches,
while GNIA and GMNIA are able to account for more complex buck- 2.4.2. Geometric imperfections
ling behavior of arches. The loads will be represented by the so-called La Poutré [21] measured the out-of-straightness of 12 full-scale
load amplifiers α. The load amplifier is the load divided by the applied roller bent arches prior to testing. Lateral crookedness, radial crook-
loading, resulting in a load parameter independent of the type of edness and twist imperfections were measured on HE 100A arches
loading (point load, uniformly distributed load or end moments). with an arch length S of 6 m (Fig. 6). A ‘substitute’ out-of-straightness
For example, for a GMNIA the ultimate load amplifier for an arch sub- pattern was derived that represents the measured geometric deviations
ject to a central point load can be obtained as follows: in the arch with nearly identical out-of-plane elastic–plastic buckling
behavior. The substitute out-of-straightness is characterized by the
α ult ¼ F ult =F: ð1Þ lowest global eigenmode from a LBA and variable amplitude. The lowest
eigenmode was selected as it is regarded as the most detrimental shape
Where: to define the out-of-straightness because it shows strong coherence
with the arch deflections at elastic–plastic buckling.
αult is the ultimate load amplifier obtained by GMNIA; In a finite element parametric study the amplitude is modified to
Fult is the maximum central point load obtained from the achieve a best fit to the measured out-of-straightness for each of the
load-deflection characteristics; 12 arches. This produced a total of 12 different amplitudes Spoorenberg
F is the central point load applied on the arch. [22]. From this finite element study it was concluded that the lowest
As F is always equal to unity, Fult is equal to αult. Analogous to αult, global eigenmode from a LBA with amplitude of S/1000 provides
αcr and αpl can be computed from Fcr and Fpl, respectively. a good substitute to define the out-of-straightness of a roller bent
arch. It is mentioned that the above suggested ‘substitute’ out-of-
straightness is only applicable to freestanding arches subject to
2.2. Model out-of-plane buckling failure.

The arch is meshed with shell elements. The selected SHELL181 2.4.3. Material modeling
element is a four-node shell element which permits inclusion of During roller bending the stress–strain curve of the material prior
residual stresses and is capable of handling large rotations and large to forming changes from a single bi-linear curve to several different
strains. A reduced integration scheme has been adopted, featuring a non-linear curves over the cross-section. From the experimental ob-
single integration point over the surface and 5 integration points servations a prediction model was developed that defines generally
over the shell thickness to capture growth of plastic zones. applicable material models for a roller bent wide flange section;
Spoorenberg et al. [23]. The prediction model consists of two different
2.3. Boundary conditions and loading components: the estimation of salient mechanical properties of roller
bent steel and the development of curves defining the stress–strain
The boundary conditions and loading types comply with those de- characteristic. Depending on the steel grade (S235 or S355), bending
scribed in Section 1.2. Fig. 3 shows the finite element model for a ratio (R/h) (ratio between arch radius R and nominal section height h)
pin-supported arch. Finite element models comprising shell elements and yield stress of the material prior to forming (fy;s), the prediction
need specific attention with respect to their supports or boundary model produces the 7 different stress–strain curves for 9 zones of
conditions, i.e. enforcing zero displacements and/or zero rotations. the roller bent arch (Fig. 7).
12 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21

Wide flange section bent

y about the major axis
Support
x u=v=w=ζ=θ=0
Arch rise f Restraine dwarping
deformations
z Shell-elements y

Support Arch spa

R nL
u=v=w=ζ=θ=0
Restrained warping deformations
Multi point constraint elements (MPC184) z η x
v
y
z
w x
γ γ θ u
ζ
Coordinate system with corresponding
deformations and rotations

Fig. 3. Finite element model with local coordinate system.

From tensile tests conducted on coupons taken from roller bent imply that the difference between nominal yield stress and measured
arches it was observed that the flanges exhibit gradual yielding yield stress in roller bent steel is completely due to the roller bending
behavior. Therefore, the modified Ramberg–Osgood model, which is process. This, of course, is not correct. Therefore, the prediction model
used for stainless steel and aluminum members, was selected to de- was calibrated on the difference between yield stress values obtained
fine the stress–strain curve from the onset of loading up to the ulti- from measurements of coupons taken from straight sections, serving
mate tensile stress in the flanges. A bi-linear material curve was as reference sections, and coupons taken from roller bent sections.
selected to approximate the material characteristics in the web. The Although calibration in this way gives a better representation of the
roller bent stress–strain curves are employed for both compression change in material properties due to roller bending, it imposes
and tension, i.e. the tensile and compression regime have an identical some drawbacks on the use of the prediction model. In most cases
stress–strain curve. For the GMNIA the stress–strain curves from this measured values of the yield stress prior to forming are absent and
prediction model were converted to true-stress, true-strain curves to the nominal yield stress must be utilized to define the stress–strain
account for large strain effects. The material models were entered in curves with the prediction model. Using the nominal yield stress
the finite element model by discrete stress–strain values, resulting can lead to conservative assumptions of the stress–strain curves, es-
in a piece-wise curve for each of the 7 zones (Fig. 8). pecially for steel grade S235. The best accuracy with the prediction
The materials' response to plastic straining is described by the Von model is achieved when measured values for the yield stress of the
Mises yield criterion, Prandtl-Reuss flow rule and an isotropic hard- straight steel are used.
ening law. The difference between the nominal yield stress according
to material specifications and the measured yield stress in roller bent 2.5. Solution procedure
steel is caused by two different phenomena. Firstly, the material is
plastically strained during roller bending resulting in hardening ef- All analyses are load-controlled. As the GNIA, MNA and GMNIA are
fects thereby changing the yield stress values. Secondly, prior to roller dominated by non-linear behavior the load is applied gradually in
bending, there exists already a difference between the nominal yield these analyses. As the primary objective is to find the ultimate load
stress (fy) and the measured yield stress (fy;s), where measured rather than post-buckling behavior the Newton–Raphson procedure
values usually exceed nominal ones. Differences between fy and fy;s was selected in preference to the arc-length method to solve the
were significant for steel grade S235 whereas for steel grade S355 non-linear equilibrium equations. The convergence criterion for the
this difference less pronounced. Including the difference between fy out-of-balance load vector is equal 0.5%. In addition to the force
and fy;s in the calibration process of the prediction model would norm check a convergence criterion of 0.5% was employed for the dis-
placement increments. For the LBA the Block–Lanczos method is used
for the eigenvalue extraction.
Arch MPC184 elements
along flange width 2.6. Load-deflection curves
4×

For all GMNIA the ultimate load αult is defined as the load level
8×
4×
4×

Support corresponding to the last converged substep. Typical load-deflection

curves from GNIA and GMNIA are shown in Fig. 9 for HE 100A arches
with a radially distributed load with imperfections as given in
4×

Number of Section 2.4. The acting load α is normalized with respect to αcr as
MPC184 elements found from a LBA executed on the same arch. The non-linear material
MPC184 elements model used is based on the prediction model for steel grade with
over web height fy;s = 235 N/mm 2. In order to correctly express the GMNIA results in
the column curve diagram the plastic multiplier αpl must be obtained
Fig. 4. Modeling of restraints. from MNA using the same non-linear material models. In addition, as
 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21 13

(a) Residual stress model (b) Implementation of residual

with element mest stress model in FE-code

Fig. 5. Residual stress model and finite element implementation.

geometrical non-linear effects are ignored the arch continues to resist increase of η and a decrease of χ.The load amplifiers obtained from
loads during the MNA as plastic straining progresses over the a MNA and LBA, αpl and αcr, respectively, can be used to define the
cross-section. In order to make a proper estimate of the plastic multi- non-dimensional slenderness for the arch:
plier, the plastic collapse load is defined as the intersection between qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
two tangents to the load-deflection curve. In Fig. 10 the load- ¼
λ α pl =α cr : ð5Þ
deflection curves are shown for MNA performed on the same arches
as selected for Fig. 9, where the acting loads are normalized with The load amplifier αult can, together with αpl, be used to plot the
respect to αcr. The intersection between the two tangents defines reduction factor of the arch:
the normalized plastic collapse load (αpl/αcr).
χ ¼ α ult =α pl : ð6Þ
2.7. Column curve and imperfection parameter curve representation
Eqs. (5) and(6) allow the finite element results to be plotted in a
As mentioned in Section 2.1 each arch configuration requires 3 dif- column curve diagram. Another approach to devise a design rule
ferent analyses to be performed, allowing the ultimate load found based on column curves is by plotting the same finite element results
from the GMNIA to be plotted in a column curve diagram. The column in an imperfection parameter diagram. When isolating the expression
curve expression as presented in EN 1993-1-1 [19] (ECCS column for η from the column curve formulation by rewriting Eqs. (2) and (3)
curve) relates the normalized flexural buckling strength of columns a different expression of η can be obtained, which is only depending

χ to their non-dimensional slenderness λ: 
on χ and λ:
1  
χ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi but χ≤1:0: ð2Þ 1 2:
 2 −1−λ
2 η¼χ þ λ ð7Þ
Φ Φ2 −λ χ2

Where: Eq. (7) can be used in conjunction with Eqs. (5) and (6) to allow
  the finite element results to be expressed in an imperfection parame-
2 :
Φ ¼ 0:5 1 þ η þ λ ð3Þ ter diagram:
 2 
α ult α pl α pl α pl
The generalized imperfection or imperfection parameter is given ηnum ¼ þ −1− : ð8Þ
by the following equation: α pl α ult α cr α cr

 
 λ
η ¼ α λ−  ≥0 ð4Þ Plotting finite element results in the imperfection parameter dia-
0
gram, finding an expression for the imperfection parameter and
where the factor α is based on the typical cross-section. Depending on replacing Eq. (4) by the newly found expression in the column
the column curve used, a value for α is given in Table 1. The parame- curve formulation results in an accurate expression for the column
 is set at 0.2, irrespective of the column curve. For identical
ter λ curve. Moreover, it leaves the original column curve expression
0
non-dimensional slenderness values, higher values of α result in an intact.

(a) top view (b) elevation (c) side view

S ζimp
vimp wimp
R f
L L

Lateral crookedness Radial crookedness Twist imperfections

Fig. 6. Geometric imperfections measured by La Poutré.

14 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21

the GMNIA a bi-linear material model and no residual stresses are

implemented to rule out any additional discretization effects with
1 respect to the zonal material distribution and step-wise residual
stress model. The geometric imperfections for the GMNIA are based
2 on Section 2.4.2. The results of the mesh refinement study are
shown in Table 2. The difference in Ncr and Nult for a specific mesh
1 5 and mesh no.4 is given as well.
It can be seen that in general an increase in mesh density produces
and increases in both the elastic buckling force and ultimate force. It
6 is noted that most mesh refinement studies are characterized by a de-
3
crease in elastic buckling load and ultimate load as the mesh is further
refined. The small difference between mesh no.3 and mesh no.4 re-
7 veals that the discretization error has almost vanished. Mesh no.3
will therefore be used for all computations. The adopted finite
4 element mesh is shown in Fig. 3.
3
3. Validation

Experiments as published in the related paper are selected for the

Fig. 7. Zonal distribution of mechanical properties.
validation study. Full-scale tests on circular roller bent steel arches
were conducted as described in a related paper by La Poutré et al.
The difference between plotting the finite element results in the [1]. Load was applied by means of a wire hanger, which meant that
column curve diagram using Eqs. (5) and (6) or in the imperfection it remained directed to the center of the baseline of the arch through-
parameter diagram using Eq. (8) is illustrated in Fig. 11 with the finite out the entire loading stage (this type of load is also known as a
element data from Fig. 9 and Fig. 10 in addition to ECCS column curve directed load). It is assumed that close coherence between the exper-
“a” and “d”. It can be seen that plotting the finite element data in the iments by La Poutré and the finite element model for the directed load
column curve suggests that ECCS column curve “d” might be appro- implies that the model is also able to correctly produce the out-of-
priate. However, looking at the imperfection parameter diagram it plane elastic–plastic buckling response for arches under gravity load-
appears that the linear imperfection parameter expression of column ing. A full overview of the experimental plan by La Poutré is shown in
curve “d” has no strong coherence with finite element results for Table 3. Only the experimental results from the full-scale arches were
arches with higher slenderness values. Hence, it is assumed that used for the evaluation of the finite element model.
using the approach the imperfection parameter will result in a more The stress–strain curve obtained from coupons taken from the
accurate expression for column curves to predict the elastic–plastic roller bent arches and the geometric imperfections measured prior
buckling response of steel arches. to testing by La Poutré in addition to the residual stress model
shown in Section 2.4.1 were used to define the initial state of the roll-
2.8. Mesh refinement study er bent arch. The arch was equipped with a tension rod by means of a
3D spar element (LINK8) to represent the tension rod (directed load-
A mesh refinement study was performed such that the finer mesh ing). The maximum vertical displacement imposed on the tension rod
is part of the larger mesh. A HE 100A with a subtended angle of 90°, during testing was also applied in the finite element model. After
arch length S of 3 m and out-of-plane fixed supports with restrained reaching this maximum, the imposed displacement was released,
warping is subjected to a radially directed uniformly distributed load. allowing the arch to deflect back. A comparison between the experi-
The arch is meshed with four different element distributions, where mentally and numerically obtained load-deflection curves is shown
the number of elements over the flange width, web height and over in Fig. 12 for two tests. The maximum loads (Fult), the corresponding
the arch length is varied. For each mesh a LBA and GMNIA are in-plane displacements (wult), out-of-plane displacements (vult) and
performed to evaluate the normal force at the abutment at elastic twists (ζult) as obtained from the experiments are tabulated in
buckling (Ncr) and at elastic–plastic buckling (Nult), respectively. For Table 4, and compared to finite element results. It can be seen that

600 600
fy;s=235N/mm2, R/h=39.79 fy;s=235N/mm2, R/h=39.79
500 Zone 1 500 Zone 1
True stress [N/mm2]

True stress [N/mm2]

400 400

300 300

200 200

100 100

0 0
0 0.005 0.01 0.015 0.02 0 0.05 0.1 0.15 0.2 0.25
True strain [-] True strain [-]
Prediction model Numerical input

Fig. 8. Example stress–strain curves for roller bent steel, initial stage (left) full stress–strain curve (right).
 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21 15

1.2 1.2
αcr=1.0 αcr=1.0
1 1

0.8 0.8

α/αcr [-]
α/αcr [-]
αult /αcr=0.70
αult /αcr=0.60
0.6 α 0.6 α

0.4 0.4

0.2 0.2
HE 100A S=6m 2γ=180 deg. HE 100A S=4m2 γ=120 deg.
R=1.91 m R=1.91 m
0 0
0 30 60 90 120 0 10 20 30 40
Central arch deflection v [mm] Central arch deflection v [mm]

1.2 1.2
αcr=1.0 αcr=1.0
1 1
α α
0.8 0.8

α/αcr [-]
α/αcr [-]

0.6 0.6
αult /αcr=0.48
0.4 0.4
αult /αcr=0.29
0.2 0.2
HE 100A S=3m 2γ=90 deg. HE 100A S=2m 2γ=60 deg.
R=1.91 m R=1.91 m
0 0
0 10 20 30 40 0 10 20 30 40
Central arch deflection v [mm] Central arch deflection v [mm]
GNIA GMNIA

Fig. 9. Load-deflection curves from GNIA and GMNIA normalized with respect to the elastic buckling load from LBA.

20 8

αpl /αcr=14.98
15 6 α
α/αcr [-]

α/αcr [-]

α
10 4
αpl /αcr=2.82

5 2
αcr=1.0
HE 100A S=6m 2γ=180 deg. HE 100A S=4m 2γ=120 deg.
αcr=1.0 R=1.91 m R=1.91 m
0 0
0 25 50 75 100 0 25 50 75 100
Vertical deflection u [mm] Vertical deflection u [mm]

2 2

α
1.5 1.5
αpl /αcr=1.10
α/αcr [-]
α/αcr [-]

αcr=1.0 αcr=1.0
1 1
α

0.5 0.5 αpl /αcr=0.38

HE 100A S=3m 2γ=90 deg. HE 100A S=2m 2γ=60 deg.

R=1.91 m R=1.91 m
0 0
0 25 50 75 100 0 25 50 75 100
Vertical deflection u [mm] Vertical deflection u [mm]

Fig. 10. Load deflection curves from MNA normalized with respect to the elastic buckling load from LBA.
16 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21

Table 1 Table 2
Imperfection factor α for column curves according to EN 1993-1-1 [19]. Mesh refinement.

Column curve a0 a b c d Mesh no. No. of elements Elastic buckling Ultimate force
force
Imperfection value α 0.13 0.21 0.34 0.49 0.76
Flanges/ Arch length Total Ncr [N] Diff [%] Nult [N] Diff [%]
web
the finite element model is able to replicate the experimental 1 2 24 144 426808 −15.28 317537 −6.34
load-deflection behavior with moderate to good accuracy. Overall, 2 4 48 576 488955 −2.94 335417 −1.07
the finite element model underestimates the experimentally obtained 3 8 96 2304 501591 −0.43 339160 + 0.03
ultimate loads. 4 16 192 9216 503772 – 339049 –

4. Sensitivity analyses
load-deflection curves from a single arch geometry with different am-
Sensitivity analyses were conducted on HE 100A and HE 360B plitudes are presented in Fig. 14. A change in imperfection amplitude
arches with a radially uniformly distributed load, which resulted in changes the load-deflection curves from the onset of loading and af-
uniform compression in the arch-rib. An initial residual stress distri- fects the buckling response to a considerable extent. The results of
bution from Fig. 5, a geometric imperfection using the lowest global all GMNIA are presented in the column curve diagram. The influence
buckling mode together with an amplitude of S/1000 and a bi-linear of a change in amplitude is most significant within the range of
material model with nominal yield stress were used to define the 
0:6≤λ≤2:5.
baseline of the initial state of the roller bent member from which
variation in imperfections was applied. 4.3. Material modeling

4.1. Residual stress Arches are equipped with different material models to monitor
their influence on the elastic–plastic buckling load. Each arch config-
The group of arches covering a non-dimensional slenderness uration is analyzed with different material properties: (1) nominal

range of 0:6≤λ≤3:8 was used for the sensitivity analyses. For each steel properties (fy = 235 N/mm 2 for steel grade S235 and fy =
arch a GMNIA was performed with two different residual stress 355 N/mm 2 for steel grade S355) in conjunction with a bi-linear ma-
models and one GMNIA was performed without inclusion of residual terial law, resembling the material properties for straight hot-rolled
stresses, in addition to a GNIA. A typical hot-rolled residual stress pat- steel. The second material model (2) is featured by substituting a
tern was used and the roller bending residual stress model as shown yield stress of the material prior to forming equal to nominal values
in Fig. 5. The various load deflection curves for a single arch geometry (fy;s = fy) in prediction model to generate 7 different stress–strain
are shown in Fig. 13. The hot-rolled residual stress model is more curves for steel grade S235 and for steel grade S355. This case
detrimental to the elastic–plastic buckling load than the roller-bent would resemble the situation when measured yield stresses for
residual stress model. The results for all arches are presented in the straight steel are absent and one has to resort to nominal values.
column curve diagram to show the sensitivity of residual stresses with The third material model (3) is identical as (2) but with a yield stress
respect to the arch non-dimensional slenderness. The influence of resid- for the material prior to forming of fy;s = 290 N/mm 2 and fy;s =
ual stresses on the elastic–plastic buckling load is most pronounced 370 N/mm 2 for steel grade S235 and S355 respectively. These values
for arches with non-dimensional slenderness smaller than 1.5. give a better approximation of the measured values of the yield stress
compared to nominal values in straight steel prior to forming. The
4.2. Geometric imperfections results from the analyses are shown in Fig. 15.
From the comparison between the material models (2) and (3) it
The same set of arches as employed in Section 4.1 was selected to can be seen that the yield stress value of the material prior to roller
investigate the sensitivity of a change in imperfection amplitude on bending is of significant influence on the results. When nominal
the elastic–plastic buckling response. Three different levels of ampli- yield stress values are substituted in the prediction model instead of
tude together with the lowest eigenmode from the LBA were used yield stress values resembling measured properties from straight
to define the out-of-straightness of the member: S/100, S/1000 and steel, the calculated failure load is conservative. This effect is most
S/10000. For each arch geometry with a given amplitude, two pronounced for steel grade S235. There exists a large difference in re-
different analysis types were performed: GNIA and GMNIA. The duction factors χ obtained with the prediction model for steel grade

1.2 7.5
imperfection parameter [-]

1
ECCS column curve a 6
reduction factor [-]

0.8
4.5
0.6
3 ECCS column curve d
0.4

0.2 1.5
ECCS column curve a
ECCS column curve d
0 0
0 1 2 3 4 0 1 2 3 4
non-dimensional slenderness [-] non-dimensional slenderness [-]

Fig. 11. Finite element results in column curve graph (left) and imperfection parameter diagram (right).
 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21 17

Table 3 Table 4
Experimental program La Poutré et al. [1] of full-scale arches. Load-deflection results obtained from experiments and numerical analyses.

Test No. of Radius R Angle Arch Rise f Span L Rise-to-span La Poutré (Exp.) Ratio [−]
no. tests [mm] 2γ [°] length [mm] [mm] ratio f/L [−] F ult ðexp:Þ W ult ðexp:Þ V ult ðexp:Þ ζ ult ðexp:Þ
Test Fult wult vult ζult W ult ðFEMÞ W ult ðFEMÞ V ult ðFEMÞ ζ ult ðFEMÞ
S [mm]
no. [kN] [mm] [mm] [º]
1A, 1B, 1C 3 1910 180 6000 1910 3820 0.50
1A 104.7 39.5 89.0 6.59 1.09 1.03 1.01 0.72
2A, 2B 2 2149 160 6000 1775 4231 0.42
1B 103.2 38.8 103.3 10.3 1.08 1.01 1.18 1.12
3A, 3B 2 2546 135 6000 1572 4705 0.33
1C 104.8 41.6 93.6 6.35 1.11 1.08 1.07 0.70
4A, 4B 2 3125 110 6000 1333 5120 0.26
2A 104.9 35.0 65.7 6.10 1.11 0.91 0.95 0.79
5A, 5B, 5 C 3 3820 90 6000 1119 5402 0.21
2B 104.3 36.1 69.3 9.22 1.11 0.96 1.03 1.24
3A 100.0 34.7 68.1 8.21 1.09 0.96 1.12 1.07
3B 99.2 33.5 58.3 7.68 1.07 0.94 1.06 1.09
S235 and for steel grade S355. The degree of roundness of the stress–
4A 99.4 33.3 36.6 4.60 1.10 1.02 0.75 0.72
strain curve for S235 is higher than that of S355, resulting in prema- 4B 96.3 30.7 40.0 4.47 1.05 0.87 0.77 0.62
ture loss of stiffness and hence lower χ-values. 5A 97.3 36.5 19.8 3.98 1.18 1.17 0.54 0.77
Furthermore, it should be noted that for an incidental case a value 5B 95.2 35.4 26.9 5.55 1.15 1.12 0.77 1.19
of χ greater than 1.0 is obtained. This can be explained by the plastic 5C 95.0 32.0 41.2 5.00 1.17 1.02 1.05 0.89
multiplier αpl being determined from finite element analyses with a
bi-linear material model with nominal yield stress. Arches with low
non-dimensional slenderness values will have higher elastic–plastic

120000 120000
wult, Fult (experimental)
wult, Fult (experimental)
100000 100000
Central point load F [N]

Central point load F [N]

wult, Fult (numerical)
80000 80000
wult, Fult (numerical)

60000 60000

40000 40000

20000 20000
Test 1A Test 5A
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60 70 80
In-plane deflection w [mm] In-plane deflection w [mm]

120000 120000
vult, Fult (experimental)
vult, Fult (experimental)
100000 100000
Central point load F [N]

Central point load F [N]

vult, Fult (numerical)

80000 80000
vult, Fult (numerical)

60000 60000

40000 40000

20000 20000
Test 1A Test 5A
0 0
0 30 60 90 120 150 0 30 60 90 120 150
Out-of-plane deflection v [mm] Out-of-plane deflection v [mm]

120000 120000
ζult, Fult (experimental)
ζult, Fult (experimental)
100000 100000
Central point load F [N]

Central point load F [N]

ζult, Fult (numerical)

80000 80000
ζult, Fult (numerical)

60000 60000

40000 40000

20000 20000
Test 1A Test 5A
0 0
0 5 10 15 20 0 5 10 15 20
Twist ζ [º] Twist ζ [º]

Fig. 12. Load-deflection characteristics for tests 1A (left) and 5A (right).

18 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21

1.2 1.2
GNIA GMNIA (roller bent r.s.)
GMNIA (no r.s.) GMNIA (hot -rolled r.s.)
1 1
ECCS column curve a

reduction factor [-]

0.8 0.8 no residual stresses

α/αcr [-]
roller bent residual stresses
0.6 0.6
hot-rolled residual stresses

0.4 0.4
ECCS column curve b
0.2 0.2
HE 100A S=2.4m 2γ=60 deg.
Out-of-plane fixed and restrained warping
0 0
0 4 8 12 16 20 0 1 2 3 4
Central arch deflection v [mm] non-dimensional slenderness [-]

Fig. 13. Influence of residual stresses: load-lateral deflection for single HE 100A (left) and results plotted in column curve diagram (right).

buckling loads than plastic collapse loads, as the former is obtained for the regression analyses. The steepness of the calibrated curves is
with finite element models with stress strain curves from the predic- different for arches under uniform bending or uniform compression
tion model. For the subsequent parametric studies identical material and also depends on the steel grade.
models will be used for the MNA and GMNIA. Finite element results for arches with a central point load or a uni-
formly distributed load and for steel grade S235 or steel grade S355
5. Proposed design rule 
are shown in Fig. 17 for 0≤λ≤4:0 and 0 ≤ η ≤ 1.5. With the exception
of an arch subjected to a central point load and steel grade S355, all im-
Based on the procedure described in Section 2.7, imperfection perfection parameters show a linear increase with non-dimensional
parameter expressions are derived to best represent the numerical slenderness. Hence a linear expression selected for these arches to
data. A sample of 174 arches configurations was selected to compute arrive at an imperfection parameter expression. When looking at the
the elastic–plastic buckling load, plastic collapse load and elastic imperfection parameter diagram for arches with steel grade S355,
buckling load, for two steel grades (S235 and S355), four load cases subjected to a central point load, it can be seen that for low slenderness
(Fig. 2) and two cross-sections (HE 100A and HE 360B). The values the imperfection parameter is fairly constant at 0.3 b η b 0.5 How-
non-dimensional slenderness of these arches is in the range from ever, for higher non-dimensional slenderness values the imperfection
0.52 to 4.75 and the subtended angle in the range from 30° to 180°. parameter drops suddenly. For these arches a polynomial of the 2nd
The imperfections are based on Section 2.4. The material properties order was selected to arrive at a closed-form expression of the imper-
are defined by the prediction model (Section 2.4.3) with a yield stress fection parameter.
prior to forming of fy;s = 235 N/mm 2 and fy;s = 355 N/mm 2 for steel The description of the imperfection parameters are given in
grade S235 and steel grade S355, respectively. Table 5. For each load case and steel grade an imperfection parameter
In Fig. 16 the numerical results are shown in the imperfection pa- expression is presented. These expressions can be substituted into
rameter diagram for arches under uniform bending or uniform com- Eq. (3) to obtain the column curve formulation for roller bent arches
pression and steel grade S235 or S355 for 0≤λ≤4:0 and 0 ≤ η ≤ 2.0. subject to out-of-plane elastic plastic buckling failure.
The imperfection parameter expression of ECCS column curve “a”
and “d” is shown in addition to the calibrated imperfection parameter 6. Discussion
expression, denoted with the dashed gray lines and solid black line,
respectively. The calibrated imperfection parameter expression is Using the design rule requires the availability of two intrinsic arch
obtained with linear regression analyses. It can be clearly seen that properties: the elastic buckling load and the in-plane plastic capacity
for these arches, the imperfection parameter increases linearly with to compute the arch non-dimensional slenderness λ  and the reduc-
non-dimensional slenderness. Hence linear imperfection parameter tion factor χ. In the parametric studies, the elastic buckling load and
expressions were selected in preference to higher order expressions in-plane plastic capacity were determined with the same finite

1.2 1.2
S/10000
S/10000
1 1 ECCS column curve a0
S/1000
reduction factor [-]

S/1000
0.8 0.8 S/100
α/αcr [-]

ECCS column curve a

0.6 S/100 0.6

0.4 0.4
GNIA GMNIA
0.2 0.2
HE 360B S=18 m 2γ=60 deg.
Out-of-plane fixed and restrained warping ECCS column curve d
0 0
0 200 400 600 800 1000 0 1 2 3 4
Central arch deflection v [mm] non-dimensional slenderness [-]

Fig. 14. Influence of amplitude: load-lateral deflection for single HE 100A (left) and results collated in column curve diagram (right).
 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21 19

1.2 1.2
Bi-linear Bi-linear

1 Pred. Mod.: fy;s=235 N/mm2 1 Pred. Mod.: fy;s=355 N/mm2

Pred. Mod.: fy;s=290 N/mm2 2 Pred. Mod.: fy;s=370 N/mm2

reduction factor [-]

reduction factor [-]
0.8 0.8
ECCS column curve a
0.6 ECCS column curve a 0.6

S235 S355
0.4 0.4
R/h=19.89 R/h=19.89
0.2 ECCS column curve d 0.2 ECCS column curve d

0 0
0 1 2 3 4 0 1 2 3 4
non-dimensional slenderness [-] non-dimensional slenderness [-]

Fig. 15. Influence of material models: Steel grade S235 (left) and steel grade S355 (right).

element model as from which the ultimate load was obtained. This roller bent steel arches under symmetric loads. Earlier published
hampers the feasibility of the design rule as it obliges the engineer models for the residual stresses and mechanical properties typical
to make finite element calculations to compute both αcr and αpl to for roller bent steel were used to define the initial state of the finite
check the out-of-plane buckling response. element model prior to loading. Sensitivity analyses were performed,
Design graphs to approximate the in-plane plastic collapse load for to investigate the influence of imperfections on the out-of-plane elas-
arches subjected to a single force applied at the crown and a uniformly tic plastic buckling response, from which the following conclusions
distributed load are given in Spoorenberg et al. [24]. In order to obtain can be drawn.
the elastic buckling load research contributions by Namita [25], Usami
and Koh [26] for arches under a uniformly distributed load and Pi et • Residual stresses in roller bent steel are less detrimental to elastic–
al. [27] for an arch under a central point load can be used. plastic buckling of steel arches when compared to residual stress
patterns in hot-rolled steel.
7. Conclusions • The results from finite element models with gradual yielding typical
for roller bent steel show a significant decrease in elastic–plastic buck-
A design rule was proposed based on finite element analyses, to ling load in comparison to finite element computations performed on
check the out-of-plane buckling response of freestanding circular arches with bi-linear material models resembling hot-rolled steels.

2 2
M M M M
imperfection parameter [-]

imperfection parameter [-]

1.5 1.5
ECCS column curve d
ECCS column curve d

1 1
Imperfection parameter curve ECCS column curve a

0.5 0.5
ECCS column curve a Imperfection parameter
curve
S235 S355
0 0
0 1 2 3 4 0 1 2 3 4
non-dimensional slenderness [-] non-dimensional slenderness [-]

2 q 2 q
Imperfection parameter curve
imperfection parameter [-]

imperfection parameter [-]

1.5 1.5
ECCS column curve d
ECCS column curve d
Imperfection parameter
curve
1 1
ECCS column curve a

0.5 0.5
ECCS column curve a

S235 S355
0 0
0 1 2 3 4 0 1 2 3 4
non-dimensional slenderness [-] non-dimensional slenderness [-]

Fig. 16. Imperfection parameter diagram for arches under uniform bending and uniform compression for steel grade S235 and steel grade S355.
20 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21

2 2
S235
F
S355

imperfection parameter [-]

imperfection parameter [-]

ECCS Column curve d
1.5 F 1.5
ECCS Column curve d

1 1
Imperfection parameter ECCS Column curve a
curve

0.5 0.5
Imperfection parameter
ECCS Column curve a curve

0 0
0 1 2 3 4 0 1 2 3 4
non-dimensional slenderness [-] non-dimensional slenderness [-]

2 2
S235 q S355
imperfection parameter [-]

imperfection parameter [-]

Imperfection Imperfection
parameter curve parameter curve
1.5 1.5
ECCS Column curve d
ECCS Column curve d

1 1 q

0.5 0.5
ECCS Column curve a
ECCS Column curve a

0 0
0 1 2 3 4 0 1 2 3 4
non-dimensional slenderness [-] non-dimensional slenderness [-]

Fig. 17. Imperfection parameter diagram for arches with a central point load and a uniformly distributed load for steel grade S235 and steel grade S355.

From the numerical analyses, a large database was created University) and Dr. Maljaars (TNO). The authors would like to ac-
containing the elastic–plastic buckling response for a wide range of knowledge their input and valuable comments.
arch configurations. The solution was plotted in generalized imper-
fection parameter diagrams and generalized imperfection parameter
expressions were derived through curve fitting techniques. The References
imperfection expressions can be substituted into the column curve
[1] La Poutré DB, Spoorenberg RC, Snijder HH, and Hoenderkamp JCD. Experimental
formulation to render the European column curve applicable for roll- investigation of the out-of-plane stability of roller bent steel arches. Journal of
er bent arches, leaving the original column curve formulation Constructional Steel Research submitted for publication.
unaltered. It was found that the imperfection curve is highly depen- [2] Timoshenko SP, Gere JM. Theory of elastic stability. Tokyo: McGraw-Hill; 1961.
[3] Vlasov VZ. Thin-walled elastic beams. Jerusalem: Israel Program for Scienftic
dent on the steel grade and load case which resulted in a total of Translations Ltd; 1961.
eight different imperfection parameter expressions and hence eight [4] Vacharajittiphan P, Trahair NS. Flexural-torsional buckling of curved members.
different column curves to predict the buckling response of roller J Struct Div 1975;101(6):1223-38.
[5] Yoo CH. Flexural-torsional stability of curved beams. J Eng Mech Div 1982;108(6):
bent arches for four different load cases with two different steel 1351-69.
grades. Provided the arch slenderness is known, the design rule can [6] Rajasekaran S, Padmanabhan S. Equations of curved beams. J Eng Mech ASCE
be a useful tool to check the out-of-plane buckling response of free- 1989;115(5):1094-111.
[7] Sakimoto T, Yamao T, Komatsu S. Experimental study on the ultimate strength of
standing roller bent steel arches.
steel arches. Proc Japan Soc Civ Eng 1979;286:139-49.
[8] Sakata T, Sakimoto T. Experimental study on the out-of-plane buckling strength of
steel arches with open cross-section. Proc Japan Soc Civ Eng 1990;416:101-12.
8. Acknowledgements [9] Komatsu S, Sakimoto T. Ultimate load carrying capacity of steel arches. J Struct Div
1977;103(12):2323-36.
[10] Sakimoto T, Komatsu S. Ultimate strength formula for central-arch-girder bridges.
This paper is based on the Ph.D. thesis of the first author. In addi- Proc Japan Soc Civ Eng 1983;333:183-6.
tion to the co-authors, the following members of the Ph.D. committee [11] Sakimoto T, Komatsu S. Ultimate strength formula for steel arches. J Struct Eng
ASCE 1983;109:613-27.
contributed to the completion of the thesis and (indirectly) to this [12] Sakimoto T, Sakata T, Tsuruta E. Elasto-plastic out-of-plane buckling strength of
paper: Prof. Soetens (Eindhoven University of Technology), Prof. through type and half-through type arch bridges. Proc Japan Soc Civ Eng
Sluys (Delft University of Technology), Prof. van Bogaert (Gent 1989;410:137-48.
[13] Sakimoto T, Sakata T. The out-of-plane buckling strength of through-type arch
bridges. J Constr Steel Res 1990;16:307-18.
[14] Papangelis JP, Trahair NS. Flexural-torsional buckling tests on arches. J Struct Eng
Table 5 ASCE 1987;113(7):1433-43.
Imperfection parameter η. [15] Pi Y-L, Trahair NS. Nonlinear inelastic analysis of steel beam-columns. I: theory.
J Struct Eng ASCE 1994;121(9):2041-61.
Load case Steel grade S235 Steel grade S355
[16] Pi Y-L, Trahair NS. Out-of-plane inelastic buckling and strength of steel arches.
Central load 
0:63λ−0:11 −0:30λ 
 2 þ 0:90λ−0:17 J Struct Eng ASCE 1998;124(2):174-83.
Uniformly distributed load 
0:95λ−0:38 
0:46λ−0:14 [17] Pi Y-L, Bradford MA. Out-of-plane strength design of fixed steel I-section arches.
Uniform bending 
0:69λ−0:31 
0:17λ J Struct Eng ASCE 2005;131(4):560-8.
  [18] Pi Y-L, Trahair NS. Inelastic lateral buckling strength and design of steel arches.
Uniform compression 1:82λ−1:22 0:71λ−0:31
Eng Struct 2000;22(8):993–1005.
 R.C. Spoorenberg et al. / Journal of Constructional Steel Research 79 (2012) 9–21 21

[19] EN 1993-1-1. Eurocode 3. Design of steel structures, general rules and rules for [24] Spoorenberg RC, Snijder HH, Hoenderkamp JCD. A theoretical method for calcu-
buildings. Brussels: CEN European Committee for Standardization; 2004. lating the collapse load of steel circular arches. Eng Struct 2012;38:89–103.
[20] Spoorenberg RC, Snijder HH, Hoenderkamp JCD. Proposed residual stress model [25] Namita Y. Die Theorie II. Ordnung von krummen Stablen und ihre Anwendung auf
for roller bent wide flange steel sections. J Constr Steel Res 2011;67(6):992–1000. das Kipp-Problem des Bogentragers. Trans Japan Soc Civ Eng 1968;155:32-41.
[21] La Poutré DB. Inelastic spatial stability of circular wide flange steel arches. Ph.D. [26] Usami T, Koh SY. Large displacement theory of thin-walled curved members and
thesis. Eindhoven: Eindhoven University of Technology; 2004. its application to lateral-torsional buckling analysis of circular arches. Int J Solids
[22] Spoorenberg RC. Structural properties and out-of-plane stability of roller bent Struct 1980;16:71-95.
steel arches. Ph.D. thesis. Eindhoven: Eindhoven University of Technology; 2011. [27] Pi Y-L, Bradford MA, Tong G-S. Elastic lateral-torsional buckling of circular arches
[23] Spoorenberg RC, Snijder HH, Hoenderkamp JCD. Mechanical properties of roller bent subjected to a central concentrated load. Int J Mech Sci 2010;52(6):847-62.
wide flange sections — part 2: prediction model. J Constr Steel Res 2012;68(1):
63-77.