Thin–Walled Structures xxx (xxxx) xxx

Contents lists available at ScienceDirect

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

Full length article

Improving the Direct Strength Method prediction of column

flexural-torsional failure loads
Pedro B. Dinis a, Dinar Camotim a, *, Alexandre Landesmann b, Andr�e Dias Martins a
a
CERIS, ICIST, DECivil, Instituto Superior T�ecnico, Universidade de Lisboa, Portugal
b
Civil Engineering Program, COPPE/UFRJ, Federal University of Rio de Janeiro, Brazil

A R T I C L E I N F O A B S T R A C T

Keywords: This paper reports the latest results of an ongoing investigation on the accuracy of the codified Direct Strength
Cold-formed steel columns Method (DSM) global strength curve in predicting the ultimate strength of cold-formed steel columns failing in
Fixed-ended and pin-ended columns flexural-torsional modes. The first part of the paper is devoted exclusively to fixed-ended columns and continues
Flexural-torsional post-buckling behaviour
recent work by the authors [1, 2] on the improvement of the flexural-torsional failure load estimation, in the
Flexural-torsional failure
moderate or high slenderness ranges it is shown that the use of a novel set of strength curves, dependent on a
Shell finite element analyses
Generalised Beam Theory (GBT) analyses cross-section normalised geometric parameter (involving the area, major and minor moments of inertia, and
Direct Strength Method (DSM) design warping constant), leads to excellent failure load predictions, eliminating the large scatter stemming from the
codified design curve and improving the DSM-based design approach proposed in [2]. The second part of the
paper aims at extending the investigation to columns with other support conditions, namely three types of pinned
supports, all fixed with respect to torsion and having warping prevented: end cross-sections attached to rigid
plates resting on spherical or cylindrical hinges (i.e., pinned with respect to major and/or minor-axis flexure).
Initially, a parametric study is performed, aimed at gathering failure loads concerning columns (i) with the cross-
section shapes considered earlier (plain channels, unstiffened and stiffened lipped channels, return lipped
channels, hat-sections and rack-sections), (ii) with various geometries (cross-section dimensions and lengths) and
(iii) covering a wide slenderness range. These failure load data are then used to assess the quality of their es­
timates provided by the codified global DSM strength curve and by the strength curve set developed for fixed-
ended columns. It is found that neither of them yields consistently good failure load estimates, which prompts
(i) an in-depth comparative study on the elastic post-buckling strength of fixed-ended and pin-ended columns,
and (ii) the proposal of modifications that lead to an efficient failure load prediction for the pin-ended col­
umns although the failure load set obtained in this work is necessarily limited, the fact that their predictions by
the proposed DSM global design curves (i) exhibit very high quality and (ii) clearly outperform those yielded by
the current design curve provides strong encouragement to search for further experimental and numerical
validation.

1. Introduction The Direct Strength Method (DSM e.g. Refs. [3,4]) was first pro­
posed by Schafer & Peko €z [5], based on an original idea of Hancock et al.
Most cold-formed steel members display very slender thin-walled [6], and is nowadays the most rational approach for the design of
open cross-sections, a feature making them highly susceptible to cold-formed steel columns. The currently codified design/strength
several instability phenomena, namely local, distortional and global curves are able to handle local, distortional, global and local-global
(flexural or flexural-torsional) buckling Fig. 1(a)-(d) show cross- interactive failures. In the context of this investigation, the relevant
section deformed configurations of lipped channel columns buckling nominal strength is the global one (fnG), which is given by
in these modes. Depending on the member length and cross-section
shape and/or dimensions, any of the above buckling modes may be
critical.

* Corresponding author.
E-mail addresses: [email protected] (P.B. Dinis), [email protected] (D. Camotim), [email protected] (A. Landesmann), [email protected]
(A.D. Martins).

https://doi.org/10.1016/j.tws.2019.106461
Received 22 July 2019; Received in revised form 1 October 2019; Accepted 12 October 2019
0263-8231/© 2019 Elsevier Ltd. All rights reserved.

Please cite this article as: Pedro B. Dinis, Thin–Walled Structures, https://doi.org/10.1016/j.tws.2019.106461
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 1. Cross-section deformed configurations of lipped channel columns buckling in (a) local, (b) distortional, (c) flexural-torsional and (d) flexural modes.

� �
8
f 0:658λG
2
if λG � 1:5 sffiffiffiffiffiffiffi cylindrical (i.e., pinned with respect to major or minor-axis
flexure PCM and PCm columns, respectively).1 At this stage, it should
>
< y fy
fnG ¼ � � with λG ¼ ; (1) be mentioned that the geometries (cross-section dimensions and
>
: f 0:877 f crG
y 2
λG
if λG > 1:5 lengths) of all the columns considered in this investigation, including
those dealt with in Refs. [1,2], were carefully chosen to ensure buckling
where fcrG and λG are the column global critical buckling stress and in flexural-torsional modes this was achieved by making sure that the
global slenderness, and fy is the material (steel) yield stress. This design critical flexural-torsional buckling loads of the selected columns are well
curve, combining an exponential expression [7] with the (lowered) below their local and/or distortional counterparts.
Euler curve, was first included in the cold-formed steel design manual in Following a brief overview of the most relevant findings recently
1996 [8], due to the work of Peko €z and Sümer [9], who showed that the reported concerning the DSM-based FMT failure load estimation for
above design curve, already codified in the context of hot-rolled steel fixed-ended (F) columns [1,2], the paper shows that it is possible to
members used in buildings [10], provided better quality estimates than further improve this estimation, by developing a novel set of strength
that adopted at that time by the cold-formed steel community [11]. curves, dependent on a cross-section normalised geometric parameter
These authors based their findings on 214 test results dealing with that involves the area, major and minor moments of inertia, and warping
concentrically loaded cold-formed steel columns with various constant the strength curve set proposed in Refs. [1,2] depended only
cross-sections (lipped channels, hat-sections, box-sections and I-sections on the ratio between the major and minor moments of inertia. Indeed, by
formed by back-to-back plain channels), all exhibiting low-to-moderate using this new parameter it becomes possible to handle F columns with
global slenderness values (λG � 1.75). arbitrary cross-sections (including those with one-wall and two-wall end
Recently, the authors [1,2] reported numerical investigations stiffeners) by means of a single set of DSM-based strength curves. The
intended to assess the accuracy of the current DSM column global merits of the proposed DSM-based design approach are assessed through
strength curve [12] in predicting the failure loads of fixed-ended the comparison with numerical (shell finite element) FMT failure load
cold-formed steel columns collapsing in flexural-torsional or flexural data previously gathered in Ref. [2], concerning F columns with various
modes. While the latter were found to be quite well predicted (there was (i) cross-section shapes (plain channels, lipped channels, return lipped
only room for slight improvements in the low and intermediate slen­ channels, web-stiffened lipped channels, web/flange-stiffened lipped
derness ranges), it was shown that the former are often considerably channels, hat-sections and rack-sections) and/or dimensions, and (ii)
underestimated by the current design curve in the moderate and high yield stresses, in order to cover a wide slenderness range also included
slenderness ranges. It is worth noting that this underestimation was in the comparison are the experimental failure loads reported in
already perceptible in the results reported by Peko €z & Sümer [9] and Ref. [13]. In particular, the reliability assessment prescribed in Section
was also pointed out about six years ago by Bandula Heva and Mahen­ K2.1.1 of [12] (see Section 1.1 of this paper) shows that the LRFD
dran [13]. Indeed, the ultimate-to-yield stress ratio fu/fy values of (i) the resistance factors associated with the modified DSM-based design
five most slender specimens addressed in Fig. 2 of reference [9], con­ approach proposed are higher than 0.90, i.e., visibly above the value
cerning lipped channel and hat-section columns with λG > 1.6, and (ii) currently recommended for compression members ϕc ¼ 0.85.
three flexural-torsional buckling tests (at room temperature) reported in Then, the paper addresses the DSM design of the aforementioned
Ref. [13], concerning lipped channel columns with λG > 1.7, are visibly three types of pin-ended cold-formed steel columns collapsing in FMT
underestimated by the current DSM global design curve. This fact led
Dinis et al. [1,2] to propose a new DSM-based set of strength curves
dependent on the ratio between the major and minor moments of 1
Note that pin-ended columns with free warping at the end cross-sections are
inertia. Although this geometric parameter enabled a fairly good esti­
extremely difficult to achieve experimentally, even if a good attempt has
mation of the numerical failure loads obtained, it was also found that recently been made by Santos et al. [14]. This is much more easily done in
distinct strength curve sets were needed for columns with cross-sections numerical analyses, particularly in linear buckling ones moreover, due to the
exhibiting one-wall and two-wall end stiffeners the latter are nature of the analytical functions employed, such end support conditions are
rack-section and double-fold return lipped channels. automatically modelled in virtually all the semi-analytical finite strip analyses.
This work continues the research effort described in the previous In shell finite element non-linear analyses (as those carried out in this work to
paragraph and has two main objectives. The first one consists of obtain the column failure loads), modelling pinned end cross-sections with free
improving the prediction of the fixed-ended column major-axis flexural- warping requires preventing the (i) global transversal translations, (ii) torsional
torsional (FMT) failure loads in particular, a single set of DSM-based rotation and (iii) local transverse membrane and flexural displacements, while
keeping the warping/axial displacements free. Such support condition
strength curves able to handle columns with arbitrary cross-sections is
arrangement tends to “artificially stiffen” the column response as the dis­
sought. The second objective consists of extending the investigation to
placements and rotations cease to be small, making it impossible to obtain
columns with other support conditions, namely three types of pinned
reliable failure loads indeed, because the warping displacements become
supports, all fixed with respect to torsion and having warping fully progressively more (partially) restrained, the column strength increases and,
prevented. They correspond to end cross-sections rigidly attached to therefore, its failure load is larger (obviously, this effect is not captured by the
rigid plates resting on hinges that may be either spherical (i.e., pinned DSM design approach, since the critical buckling stress is calculated for
with respect to major and minor-axis flexure PS columns) or completely free warping displacements).

2
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

modes. Initially, in order to gather representative sets of PCM, PCm and The failure load data are obtained by means of shell finite element
PS column FMT failure loads, a parametric study is performed the geometrically and materially non-linear analyses of columns containing
columns analysed exhibit the same seven cross-section shapes as the F initial geometrical imperfections (the so-called GMNIA) carried out in
columns addressed previously and various geometries and yield stresses, the software package ABAQUS [17]. All the columns analysed contain
selected to cover a wide slenderness range. Then, it is shown that the critical-mode (flexural-torsional) global initial imperfections with
quality of the predictions provided by the current DSM global design L/1000 amplitude, a value that is in line with experimental observations
curve [12], for the above failure loads, varies considerably with the (e.g. Ref. [18]) and is often prescribed in specifications dealing with steel
column support conditions. This rather unexpected finding prompts the members and structures, regardless of the end support conditions in
performance of a numerical comparative study on the structural re­ this particular case (columns buckling in FMT modes), the L/1000 value
sponses of fixed-ended and pin-ended columns, carried out by means of corresponds to the mid-span flange-stiffener corner vertical displace­
Generalised Beam Theory (GBT) and shell finite element geometrically ment, which combines contributions from a clockwise torsional rotation
non-linear analyses, which is intended to shed new light on the different and a downward vertical translation. The shell finite element model
mechanics involved in the elastic buckling and post-buckling behaviours employed, previously used by the authors, (i) is based on an
of F, PCM, PCm and PS columns. Finally, on the basis of the knowledge elastic-perfectly plastic steel material behaviour (Prandtl-Reuss’s model:
acquired from this numerical comparative study and failure load data von Mises yield criterion and associated flow rule), (ii) discretises the
obtained, DSM-based strength curves, concerning flexural-torsional columns into fine meshes of 4-node isoparametric elements (length-­
failures, are proposed for the three types of pin-ended columns consid­ to-width ratio roughly equal to 1), (iii) disregards rounded corner and
ered in this work, and their merits and reliability are assessed. residual stress/strain effects, since various researchers have shown that
The paper closes with a summary of the various DSM-based design these two effects practically cancel each other (e.g. Ref. [19]), and (iv)
curves proposed in this work, which includes a discussion on their simulates the various support conditions by attaching rigid plates to the
benefits and drawbacks, and the need for further validation/research. column end sections, thus automatically precluding warping these
rigid plates have prevented torsional rotations and flexural displace­
1.1. Load and Resistance Factor Design (LRFD) ments, whereas the two flexural rotations are either fully prevented or
completely free (the four possible combinations correspond to the F,
According to Section K2.1.1 of reference [12], the LFRD resistance PCM, PCm and PS columns).
factor ϕ is given by
� � � � 2. Overview of the available DSM-Based design approaches for F
pffiffiffi2ffiffiffiffiffiffiffi2ffiffiffiffiffiffiffiffiffiffi2ffiffiffiffiffiffi2ffiffi
1 m columns
(2)
β V M þV F þCP V P þV Q
ϕ ¼ Cϕ Mm Fm Pm e 0 with CP ¼ 1 þ ;
n m 2
This section summarises the main findings that were unveiled in recent
where (i) Cϕ is a calibration coefficient (Cϕ ¼ 1.52 for LRFD), (ii) Mm numerical investigations concerning the strength and DSM design of F
¼ 1.10 and Fm¼ 1.00 are the mean values of the material and fabrication columns failing in FMT modes [1,2]. These investigations dealt with col­
factors, respectively, (iii) β0 is the target reliability index (β0 ¼ 2.5 for umns exhibiting a wide variety of cross-section shapes: plain channels (U),
structural members in LRFD), (iv) VM ¼ 0.10, VF ¼ 0.05 and VQ ¼ 0.21 unstiffened (C), return lip (RLC), web-stiffened (WSC) and web/flange-
are the coefficients of variation of the material factor, fabrication factor stiffened (WFSC) lipped channels, hat-sections (H) and rack-sections
and load effect, respectively, (v) CP is a correction factor depending on (R) all cross-section dimensions and lengths can be found in Annexes
the numbers of tests (n) and degrees of freedom (m ¼ n 1), and (vi) Pm A to G of reference [20]. A total of 1710 columns were analysed, all con­
and VP are the mean and the coefficient of variation of the “exact”-to- taining critical-mode initial geometrical imperfections with L/1000
predicted failure load ratios. The value recommended for compression amplitude and covering wide slenderness ranges (fy¼ 75, 150, 300, 450,
members is ϕc ¼ 0.85, regardless of the column failure mode nature. 600 MPa). Fig. 2(a)–(g) compare the codified DSM global design curve
(fnG) with the obtained column fu/fy ratios. Fig. 3(a)–(g) plot fu/fnG against
1.2. Numerical models employed in this work λG for the seven column sets. It is observed that the fu/fy values (i) are well
aligned along the DSM global strength curve (marginally above and with
This work requires the performance of column (i) elastic linear small vertical dispersion) in the low-to-moderate slenderness ranges
buckling/bifurcation and non-linear (post-buckling) and (ii) elastic- (λG � 1.5) and (ii) lie well above that curve and are much more scattered
plastic non-linear (post-buckling) numerical analyses the former are (the vertical dispersion grows with λG) in the moderate and high slen­
carried out by means of GBT-based beam finite element formulations.2 derness ranges (λG > 1.5) moreover, the fu/fy and fu/fnG “clouds” are
The bifurcation analyses are carried out in the code GBTUL [15], quite similar for all the seven column sets.
including only conventional deformation modes, namely the global, Following the above observation, it was decided to improve the
distortional and most relevant local modes. The cross-section and lon­ current DSM column global design curve, in order to achieve a higher
gitudinal discretisations involve (i) 3 intermediate nodes per web and failure load prediction quality in the moderate and high slenderness
flange, and 1 intermediate node per stiffener wall, and (ii) 10 ranges (λG > 1.5 this means that the exponential expression is kept in
equal-length beam finite elements, respectively. The elastic non-linear the low-to-moderate slenderness range). The search for such improve­
analyses are based on the GBT formulation developed and numerically ment led to the proposal of a new DSM-based approach (fnFT) involving a
implemented by Martins et al. [16], and include conventional, shear and set of strength curves dependent on a geometric parameter β ¼ II/III,
(linear and quadratic) transverse extension deformation modes. Only relating the major (II) and minor (III) moments of inertia. For λG (�λFT)
plain channel columns are analysed (see Section 5.1.1), with >1.5, the strength curve becomes β-dependent and is defined by a
cross-section and longitudinal discretisations involving (i) nine inter­ general “Euler-type” expression similar to that appearing in the current
mediate nodes (3 per wall) and (ii) 20 equal-length beam finite ele­ DSM global strength curve (see Eq. (1))3 this strength curve set is
ments, respectively this cross-section discretisation leads to 51 defined by the expressions
deformation modes, namely 15 conventional (4 global and 11 local), 12
shear, 12 linear transverse and 12 quadratic transverse modes.

3
In order to make a clear distinction between the current DSM global design
2
For validation purposes, elastic non-linear results obtained from shell finite curve and the proposed DSM-based strength curves, concerning exclusively
element analyses are also presented. columns failing in FMT modes, λG is replaced by λFT in the latter.

3
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 2. Plots fu/fy vs. λG for (a) U, (b) C, (c) H, (d) R, (e) RLC, (f) WSC, (g) WFSC columns.

� �
8
f 0:658λFT
2
if λFT � 1:5 sffiffiffiffiffiffiffiffi (accurate and reliable) estimates of the available fixed-ended column
>
< y fy numerical FMT failure loads, and clearly outperform those yielded by the
fnFT ¼ � � with λFT ¼ (3) current global design curve, they still exhibit one non-negligible draw­
>
: f a fcrFT
y
λbFT
if λFT > 1:5 back: distinct strength curve sets are needed for column cross-sections
with one-wall and two-wall end stiffeners. The next section addresses
where the β-dependence is felt through parameters a and b, which are the development and merit assessment of another DSM-based design
given by approach for F columns buckling and failing in FMT modes, which (i)
shares the positive performance indicators exhibited by the strength
a ¼ 0:39 � 1:5b (4) curves defined by Eqs. (3)–(6) and (ii) is able to handle columns with
� arbitrary cross-sections (including those with one-wall and two-wall end
b¼
0:10β þ 0:85 if β < 11:5
(5) stiffeners) this is because it is founded on a structural reasoning more
2 if β � 11:5
closely linked to the column flexural-torsional buckling mechanics.

for all the U, C, H, WSC and WFSC columns analysed in Ref. [2]. Note
3. New DSM-Based design approach for fixed-ended columns
that (i) these a and b expressions were obtained by means of a “tri­
al-and-error curve-fitting procedure”, and (ii) Eq. (1) is recovered for
On the basis of the observation made in the previous section and with
β � 11.5, value beyond which the current design curve performance is
the objective of reaching a more inclusive and mechanically sound
quite good one has then a ¼ 0.877 and b ¼ 2.0. However, in the case of
estimation of the column flexural-torsional ultimate strength (fnFT), it
the R and RLC columns (both with “two-wall” end stiffeners), it was
was decided to modify the strength curve sets defined by Eqs. (3)–(6),
found that the same high-quality FMT failure load prediction can only be
valid for λFT > 1.5 (moderate and high slenderness ranges) naturally,
obtained if the DSM-based strength curve set proposed is mod­
this means that the exponential expression is still kept in the low-to-
ified/lowered. The modification proposed for those columns involved
moderate slenderness range (λFT � 1.5). The modification consists of
exclusively parameter b(β), which becomes
considering a cross-section geometric parameter that involves not only
�
0:20 β þ 0:90 if β < 5:5 the major (II) and minor (III) moments of inertia, but also the warping
b¼ (6)
2 if β � 5:5 constant Iw note that, nowadays, this constant can be easily calculated
numerically, by means of freely available codes like GBTUL [15] or CUFSM
Note that parameter a is still given by Eq. (4) and the new flexural-
[21]. The new parameter, termed βFT, is given by
torsional strength curves still coincide with Eq. (1) for columns with
high β values (i.e., a ¼ 0.877 and b ¼ 2). II þ Iw =A
βFT ¼ (7)
The prediction quality achieved by the two sets of strength curves III
can be observed in Fig. 4(a)-(b), which plot, against λFT, the fu/fnFT
values obtained (i) with Eq. (5), for all but the R and RLC column sets where A is the gross cross-section area, included in the expression to
considered (“all but R þ RLC columns”), and (ii) with Eq. (6), exclusively keep its non-dimensional nature.
for the R and RLC column sets the associated averages, standard de­ Following exactly the same strategy adopted previously in Refs. [1,
viations, maximum/minimum values, as well as the ϕ values they 2], a new set of βFT-dependent strength curves was sought to predict, as
originate, are also included in these figures. It is clearly observed that efficiently as possible, the column FMT failure loads. After grouping all
the two strength curve sets yield accurate and mostly safe predictions of the column sets considered according to their βFT values, a “tri­
the numerical FMT failure loads obtained in Ref. [2] the fu/fnFT aver­ al-and-error curve-fitting procedure” was employed to reach a new
ages and standard deviations are equal to 1.070/0.056 (all but R þ RLC expression for parameter b(βFT) parameter a is still given by Eq. (4)
columns) and 1.059/0.064 (R þ RLC columns), and lead to LRFD and the new flexural-torsional strength curves still coincide with Eq (1)
resistance factors visibly superior to that recommended by AISI [12] for for columns with high βFT values (i.e., a ¼ 0.877 and b ¼ 2) since, as
compression members ϕc ¼ 0.85. mentioned before, the current design curve performs quite well for such
Although the above DSM-based design approaches provide efficient columns. The output of the procedure just described is a b expression

4
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 3. Plots fu/fnG vs. λG for (a) U, (b) C, (c) H, (d) R, (e) RLC, (f) WSC, (g) WFSC columns.

Fig. 4. Plots fu/fnFT vs. λFT for the (a) “all but R þ RLC columns” (Eqs. (3)þ(5)) and (b) “R þ RLC columns” (Eqs. (3)þ(6)).

defined by deviations and maximum/minimum values the LRFD resistance fac­

8 tors (ϕ) obtained from the failure-to-predicted ultimate strength ratios of
< 0:06βFT þ 0:71 if βFT < 21:5
all the columns analysed are also included. As for Fig. 7(a)-(b), they plot,
b¼ (8)
: also against λFT, the failure-to-predicted fu/fnFT ratios of (i) all numerical
2 if βFT � 21:5
failure loads reported in Ref. [2] and (ii) the six experimental failure
In order to illustrate the prediction quality achieved by the new loads reported in Ref. [13], concerning lipped channel columns these
strength curve set, Fig. 5(a)-(b) plot, against λFT, the fu/fy values con­ figures also include the fu/fnFT means, standard deviations and max­
cerning pairs of C and R columns exhibiting similar β values imum/minimum values, as well as the LRFD resistance factor values
(β ¼ 1.55–2.83 and β ¼ 1.57–2.53, respectively), but different βFT ones they lead to. The observation of all these results prompts the following
(βFT ¼ 5.47–6.14 and βFT ¼ 6.53–11.96). Their comparison with the remarks:
current DSM column global design curve and the strength curve set
obtained with Eq. (8) (solid and dashed lines, respectively), clearly (i) This new global strength curve set yields quite accurate numeri­
shows that the new DSM strength curves follow the fu/fy trends quite cal ultimate strength predictions in the λFT > 1.5 slenderness
well, providing rather accurate underestimations of the numerical fail­ range for the seven column sets considered, having cross-sections
ure loads of columns with cross-sections exhibiting both one-wall and with both one-wall and two-wall end stiffeners, as their fu/fnFT
two-wall end stiffeners. statistical indicators (averages, standard deviations, maximum/
The tables A.1 to A.7 of the supplementary data file described in the minimum values) are similarly good: they read 1.079-0.082-
Annex provide representative samples of the new failure-to-predicted 1.207-0.845 (U), 1.054-0.061-1.234-0.875 (C), 1.113-0.066-
strength ratios fu/fnFT respectively for the U, C, H, R, RLC, WSC and 1.308-0.906 (H), 1.064-0.056-1.225-0.888 (R), 1.049-0.041-
WFSC columns analysed in Ref. [2], together with relevant quantities 1.147-0.965 (RLC), 1.063-0.054-1.157-0.901 (WSC) and 1.069-
involved in their calculation, namely the cross-section warping constant 0.077-1.288-0.860 (WFSC) note that the statistical indicators
Iw and the βFT value the full set of results can be found in Ref. [20]. In given in Fig. 6(a)-(g) concern the whole slenderness range.
order to assess the performance and merits of the newly proposed (ii) In view of the content of the previous item, it seems fair to argue
strength curve set, Fig. 6(a)-(g) plot, against λFT, the fu/fnFT values that column flexural-torsional post-buckling characteristics are
concerning the seven column sets. Moreover, Table 1 provides, for each adequately captured by the cross-section geometric parameter
column set, the failure load numbers (n) and fu/fnFT averages, standard βFT. Indeed, the adoption of the βFT-dependent design curve set

5
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 5. Comparison between the current DSM design curve (fnG), the newly proposed DSM-based strength curves (fnFT) and the fu/fy values concerning (a) C columns
with βFT ¼ 5.47–6.14 and (b) R columns with βFT ¼ 6.53–11.96.

Fig. 6. Plots fu/fnFT vs. λFT for the (a) U, (b) C, (c) H, (d) R, (e) RLC, (f) WSC, (g) WFSC columns.

Table 1
Means, standard deviations, maximum/minimum values and LRFD resistance factors of the numerical failure-to-predicted ultimate strength ratios provided by the
newly proposed DSM-based strength curve set.
U columns C columns H columns R columns RLC columns WSC columns WFSC columns All columns

λG �1.5 >1.5 �1.5 >1.5 �1.5 >1.5 �1.5 >1.5 �1.5 >1.5 �1.5 >1.5 �1.5 >1.5 �1.5 >1.5 All

n 29 61 128 142 134 136 96 174 96 174 108 162 107 163 698 1012 1710
Mean 1.065 1.079 1.058 1.054 1.074 1.113 1.036 1.064 1.024 1.049 1.054 1.063 1.046 1.069 1.051 1.068 1.061
S. Dev.
0.028 0.082 0.035 0.061 0.037 0.066 0.025 0.056 0.027 0.041 0.019 0.054 0.024 0.077 0.033 0.064 0.054
Max
1.126 1.207 1.155 1.234 1.170 1.308 1.092 1.225 1.086 1.147 1.105 1.157 1.108 1.288 1.170 1.308 1.308
Min
1.014 0.845 0.958 0.875 0.093 0.906 0.953 0.888 0.939 0.965 1.010 0.901 0.982 0.860 0.939 0.845 0.845
ϕ – – – – – – – – – – – – – – 0.96 0.97 0.97

overcomes the limitation of the DSM, i.e., the fact that it does not Specification [12]. They read ϕ ¼ 0.96 (n ¼ 698, Pm ¼ 1.051,
take into account the variation in post-buckling strength (it is VP ¼ 0.031) and ϕ ¼ 0.97 (n ¼ 1012, Pm ¼ 1.068, VP ¼ 0.060),
based on critical stresses), which has been found to correlate respectively.
quite well with the βFT value. (iv) Because only three of the C columns tested by Bandula Heva and
(iii) The LRFD resistance factors for λFT � 1.5 and λFT > 1.5 are Mahendran [13] exhibit global slenderness values higher than
virtually identical and well above ϕc ¼ 0.85 value recom­ 1.5, further experimental studies involving slender column
mended for compression members by the North American specimens are required in order to provide adequate

6
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 7. Plots fu/fnFT vs. λFT for the available (a) numerical and (b) experimental column failure loads.

experimental validation for the proposed strength curves. (i) Each Pcr vs. L curve exhibits two distinct zones, one associated
Nevertheless, the six-value (three below λFT ¼ 1.5 and three with local buckling in modes with several half-waves (p5 is
above λFT ¼ 1.5) fu/fnFT statistical indicators obtained are quite dominant) and the other with single half-wave global buck­
good: average, standard deviation, maximum/minimum values ling naturally, the length corresponding to the transition be­
equal to 0.999-0.075-1.114-0.915 the associated LRFD resis­ tween local and global buckling depends on the column end
tance factor is also above 0.85. support conditions: LT ¼ 320 cm (PCM and PS columns) and
LT ¼ 358 cm (F and PCm columns).
4. Pin-ended column flexural-torsional failure load data (ii) Depending on the L value, the curve (global) descending branch
may be associated with two buckling mode natures: major-axis
4.1. Linear buckling behaviour – column geometry selection flexural-torsional (modes 2 þ 4) or minor-axis flexural (mode 3).
(iii) The relative participations of modes 2 and 4 in the column critical
The signature curves depicted in Fig. 8(a) provide the variation of the FMT buckling mode vary with the column end supports higher
critical buckling load Pcr with the length L (logarithmic scale) for steel participations from mode 2 in the PCM and PS columns.
(E ¼ 210 GPa, ν ¼ 0.3) U columns with bw ¼ bf ¼ 100 mm, t ¼ 3 mm and (iv) The FMT buckling length range (LFT) is quite large and the critical
four end support conditions: fixed, spherically-hinged and two flexural-torsional (Pcr) and lowest flexural (Pb.fm) buckling loads
cylindrically-hinged supports (axis along the cross-section major or are quite far apart. The exception are the PCm columns: the FMT
minor principal axes) F, PS, PCM and PCm columns/curves, respec­ buckling length range is fairly small (505�LFT>358 cm) and the
tively. These curves were obtained through GBT linear buckling ana­ values of Pcr and Pb.fm are very close e.g., for L� LT their ratios
lyses, carried out with code GBTUL [15] for the discretisation described read Pb.fm/Pcr ¼ 5.71 (PCM), 4.55 (F), 1.43 (PS) and 1.14 (PCm).
in Section 1.2 9 deformation modes are included: 4 global (1–4) and 5
local (5–9). As for Fig. 8(b), it shows the associated modal participation
diagrams, providing the contributions of each deformation mode 4.2. Failure load data
(pi modes 2–5 are shown in Fig. 8(c)) to the column critical buckling
modes. Finally, Fig. 8(c) displays the in-plane buckled shapes of the The next task of this ongoing research effort is to address the DSM
mid-span cross-sections of PS columns with lengths L ¼ 200, 600, design of cold-formed steel pin-ended columns buckling and failing in
1000 cm. These buckling results prompt the following remarks: FMT modes this study involves PCM, PCm and PS columns with the
same seven cross-sections shape considered before, for the F columns,
which are depicted in Fig. 9: U, C, H, RLC, R, WSC and WFSC columns.
The first step consists of assessing the merits (accuracy and safety) of the

Fig. 8. (a) Pcr vs. L signature curves and (b) GBT modal participation diagrams of U columns (bw ¼ bf ¼ 100 mm, t ¼ 3 mm) with four support conditions, and (c) 4
GBT deformation mode in-plane shapes and 3 PS column buckled cross-sections.

7
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

current DSM global strength curve in predicting their failure loads, (i) The seven column sets exhibit similar fu/fy “clouds” for each
which must be preceded by the acquisition of significant numerical support condition: they are well aligned along the DSM global
failure load data. This is done by means of a parametric study strength curve, with relative small vertical dispersion. Indeed, for
comprising ABAQUS non-linear elastic-plastic SFEA, adopting a model all columns analysed the fu/fnG statistical indicators (averages,
identical to that employed to analyse the F columns [1,2], of pin-ended standard deviations, maximum/minimum values) read 1.181-
columns with the cross-section dimensions and lengths given in Table 2, 0.191-1.922-0.949 (PCM), 0.891-0.074-1.091-0.803 (PCm),
selected to ensure buckling in FMT modes for the three end support 0.973-0.030-1.080-0.904 (PS).
conditions. As before, for each geometry the columns analysed have (ii) Although it may appear, at first sight, that the current DSM design
yield stresses fy selected to cover wide critical slenderness ranges, curve provides fairly good ultimate strength estimates for the
namely 75, 150, 300, 450, 600 MPa,4 and contain critical-mode initial whole set of columns, a more careful observation shows that this
geometrical imperfections with amplitude L/1000. For each pin-ended assertion is only true for the PS columns (fu/fnG values with
support condition, the numerical failure load data concern (i) 80 U average equal to 1.0 and very low scatter). On the other hand, the
columns, such that 1.00�bw/bf � 0.83, (ii) 120C and H column pairs fu/fy values of the PCm and PCM columns lie (ii1) a bit below the
sharing the same cross-section dimensions, such that 1.14�bw/bf � 0.83 design curve for λG > 1.0, with little scatter (average and
and 14.00�bf/bs � 5.00, (iii) 120 R and RLC column pairs also sharing maximum/minimum values: 0.865-1.060-0.803) and (ii2) above
the same cross-section dimensions, such that 1.20�bw/bf � 1.00 and that curve for λG > 1.5, with visible scatter (average, standard
10.00�bf/bs � 5.00, and (iv) 120 WSC and WFSC column pairs (same deviation, maximum/minimum values: 1.262-0.181-1.922-
cross-section dimensions), such that 1.13�bw/bf � 1.00 and 0.966) note, however, that the latter ultimate strength under­
10.67�bf/bs � 8.00 in the last two column sets, the “v-shaped” stiff­ estimation is lower than that observed for the F columns
eners dimensions are such that d1 ¼ d3 ¼ 10 mm and d2 ¼ d4 ¼ 20 mm. (compare Figs. 2(a)–(g) and 11(a1)-(a7)).
Therefore, a total of 800 columns are analysed for each pin-ended sup­ (iii) In view of what was mentioned in the previous items, it can be
port condition. Representative samples of their critical buckling stresses concluded that the DSM design quality of the PCm and PCM col­
fcr and ultimate strengths fu are given in Tables B.1-B.7, C.1-C.7 and umns can be improved. The search for such improvements is
D.1-D.7 of the data file described in the Annex, each of them including addressed in the next section of the paper. However, this can only
results of columns with one type of pin-ended support conditions the be done after some mechanical insight on the differences between
results not shown here can be found in Ref. [20].5 Fig. 10(a1)-(a2), the elastic post-buckling strengths of F, PCM, PCm and PS columns
concerning U1(L4) PCM and PS columns with various yield stresses, has been acquired, which is the objective of the limited numerical
illustrate the column equilibrium paths that it is necessary to determine study presented next.
in order to obtain the numerical failure loads given in the above three
sets of tables. Moreover, Fig. 10(b1)-(b2) display the failure modes
(deformed configurations at collapse) of the columns with 5.1.1. Elastic post-buckling strength of fixed-ended and pin-ended
fy/fcr � 1.5 note that the two column failure modes under consider­ columns comparative study
ation are practically the same: the collapse occurs after the full yielding The columns analysed in this section (i) exhibit the U4(L1) geometry
of the web-top flange corners at the column central region. (see Table 2) and (ii) contain FMT initial geometrical imperfections with
L/1000 amplitude. At this stage, it is worth recalling that the closeness
5. DSM-based design approaches for pin-ended columns between this column major-axis flexural-torsional (critical) and minor-
axis flexural buckling loads varies considerably with the support con­
The failure loads just obtained are now used to assess the merits of ditions indeed, the Pb.fm/Pcr ratio is equal to 5.95 (F), 2.35 (PCM), 1.49
the current DSM global strength curve in predicting them and, if (PCm) or 1.87 (PS). In order to acquire in-depth fresh insight on the
necessary, also to propose modifications/improvements to this design differences between the mechanics associated with the elastic post-
curve. buckling behaviours of the F, PCM, PCm and PS columns, geometri­
cally non-linear GBT analyses are employed [16].
5.1. Current DSM ultimate strength predictions Fig. 13(a)-(b) show four post-buckling equilibrium paths (i) P/Pcr vs.
θ þ θ0 (θ is the mid-span web chord rigid-body rotation and θ0 the
Tables B.1-B.7, C.1-C.7 and D.1-D.7 of the data file described in the corresponding initial value) and (ii) P/Pcr vs. dm/t (dm is the mid-web
Annex provide representative samples or all the fnG estimates, corre­ transverse displacement), which concern the F, PCM, PCm and PS col­
sponding numerical-to-predicted ratios fu/fnG and global slenderness umns, and were obtained by means of geometrically non-linear GBT
values λG (�λFT) for the U (Tables B.1þD.1), C (Tables B.2þD.2), H analyses [16] – these figures also compare the GBT-based results with
(Tables B.3þD.3), RLC (Tables B.4þD.4), R (Tables B.5þD.5), WSC values obtained through ABAQUS SFEA an excellent agreement was
(Tables B.6þD.6) and WFSC (Tables B.7þD.7) columns analysed. Fig. 11 found in the moderate rotation range (up to around θ þ θ0 ¼ 0.4 rad).
(a1)-(c7) compare the DSM global design curve with the fu/fy values of The U4 cross-section is discretized with nine intermediate nodes (three
each column set with PCM (Fig. 11(a1)-(a7)), PCm (Fig. 11(b1)-(b7)) and per wall), which leads to 51 deformation modes: (i) 15 conventional (4
PS (Fig. 11(c1)-(c7)) end support conditions. As for Fig. 12(a1)-(c7), they global and 11 local – 1–15), (ii) 12 shear (16–27), (iii) 12 linear
plot, against λG, the fu/fnG values of the U, C, H, RLC, R, WSC and WFSC transverse extension (28–39), and (iv) 12 quadratic transverse extension
columns with the three support conditions Table 3 provides, for each (40–51) modes. Finally, Fig. 13(c) displays modal participation dia­
column set, the failure load numbers (n) and fu/fnG averages, standard grams, based on the modal amplitude functions, for the four columns
deviations and maximum/minimum values. The observation of all these analysed. As for Fig. 14, it shows again the column post-buckling equi­
results prompts the following remarks: librium paths P/Pcr vs. θ þ θ0 that were already presented in Fig. 13(a),
together with the associated mid-span cross-section deformed configu­
rations at four equilibrium states, namely those corresponding to
θ þ θ0 ¼ 0.1; 0.2; 0.3; 0.4 rad. The joint observation of all these column
post-buckling results prompts the following remarks:
4
In order to achieve low slenderness values, for the sake of completion, some
of these fy values are unrealistically low. (i) It is clear that the four equilibrium paths depicted in Fig. 13(a)–
5
Since no WSC and WFSC columns were analysed in Ref. [20], all the cor­ (b) reflect two mechanically distinct post-buckling behaviours.
responding results are given in Annexes B to D. The F and PCM column paths exhibit the stable nature and

8
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 9. Pin-ended column cross-sections considered: (a) U, (b) C, (c), H, (d) R, (e) RLC, (f) WSC and (g) WFSC.

Table 2
Geometries of the PCM, PCm and PS columns: bw, bf, bs, bl, t, A, II, III, Iw, and L values (mm, mm,2 mm4 and mm6).
Column bw bf bs bl t A bw/bf bf/bs II ( � 104) III ( � 104) Iw ( � 106) βFT L1 L2 L3 L4

U1 100 100 – – 3.0 900 1.00 – 175.0 100.0 1790.2 3.74 3600 3800 4000 4200
U2 100 110 – – 3.0 960 0.91 – 190.0 129.0 2326.4 3.35 4200 4700 5200 5700
U3 100 120 – – 3.0 1020 0.83 – 205.1 162.7 2957.9 3.04 4500 5000 5500 6000
U4 100 120 – – 2.0 680 0.83 – 136.7 108,4 1969.1 3.93 8200 8400 8600 8800

C1 60 55 11 – 1.2 230.4 1.09 5.00 15.7 10.1 91.0 5.47 2300 2500 3000 35000
C2 80 70 10 – 1.0 240 1.14 7.00 29.1 16.1 219.7 7.48 3000 3500 4000 4500
C3 100 100 10 – 2.0 640 1.00 10.00 124.8 83.3 1676.3 4.64 4000 4500 5000 5500
C4 100 120 10 – 2.0 720 0.83 12.00 144.8 131.2 2655.3 3.91 5000 5500 6000 6500
C5 130 120 10 – 2.5 975 1.08 12.00 317.3 179.1 5805.8 5.10 5000 5500 6000 6500
C6 150 140 10 – 3.0 1350 1.07 14.00 586.4 331.9 14086.1 4.91 6800 7200 7600 8000

H1 60 55 11 – 1.2 230.4 1.09 5.00 17.4 10.1 51.4 3.93 2300 2500 3000 35000
H2 80 70 10 – 1.0 240 1.14 7.00 30.7 16.1 159.2 6.03 3000 3500 4000 4500
H3 100 100 10 – 2.0 640 1.00 10.00 128.8 83.3 1365.5 4.11 4000 4500 5000 5500
H4 100 120 10 – 2.0 720 0.83 12.00 148.8 131.2 2209.8 3.47 5000 5500 6000 6500
H5 130 120 10 – 2.5 975 1.08 12.00 323.8 179.1 5068.5 4.71 5000 5500 6000 6500
H6 150 140 10 – 3.0 1350 1.07 14.00 595.4 331.9 12689.9 4.63 6800 7200 7600 8000

R1 60 55 10 10 1.0 210 1.09 5.50 13.8 10.9 126.9 6.84 2000 25000 3000 3300
R2 100 100 20 15 1.5 555 1.00 5.00 101.4 91.8 3304,6 7.59 5000 5500 6000 6500
R3 110 100 20 15 1.5 570 1.10 5.00 125.3 95.0 3808.5 8.35 4000 4500 5000 5500
R4 130 130 20 20 1.5 705 1.00 6.50 222.7 194.0 9975.9 8.44 6000 6500 7000 7500
R5 150 130 20 15 2.5 1200 1.15 6.50 501.2 318.9 18992.6 6.53 5000 5500 6000 6500
R6 180 150 15 15 2.0 1080 1.20 10.00 657.9 361.2 25732.3 8.42 8000 8300 8600 8900

RLC1 60 55 10 10 1.0 210 1.09 5.50 13.8 9.7 114.6 7.08 2000 25000 3000 3300
RLC2 100 100 20 15 1.5 555 1.00 5.00 101.4 84.5 3094.4 7.80 5000 5500 6000 6500
RLC3 110 100 20 15 1.5 570 1.10 5.00 125.3 87.5 3552.5 8.55 4000 4500 5000 5500
RLC4 130 130 20 20 1.5 705 1.00 6.50 222.7 176.8 9150.8 8.60 6000 6500 7000 7500
RLC5 150 130 20 15 2.5 1200 1.15 6.50 501.2 301.9 17952.8 6.62 5000 5500 6000 6500
RLC6 180 150 15 15 2.0 1080 1.20 10.00 567.9 344.7 24387.1 8.20 8000 8300 8600 8900

WSC1 170 150 15 – 2.0 1017 1.13 10.00 551.6 295.3 16652.2 7.41 8500 8750 9000 9250
WSC2 170 160 15 – 2.5 1321 1.06 10.67 725,7 433.5 24460.9 5.95 8000 8500 9000 9500
WSC3 180 160 15 – 3.0 1615 1.13 10.67 985.0 530.1 33183.4 5.73 7500 8000 8500 9000
WSC4 180 180 20 – 3.0 1765 1.00 10.00 1098.0 750.1 49789.6 5.22 7500 8000 8500 9000
WSC5 150 140 15 – 2.0 937 1.07 9.33 398.8 238.9 10758.0 6.48 7000 7500 8000 8500
WSC6 130 120 15 – 2.0 817 1.08 8.00 259.4 155.8 5465.4 5.96 5500 6000 6500 7000

WFSC1 170 150 15 – 2.0 1017 1.13 10.00 566,3 297.0 16171.6 7.10 8000 8250 8500 8750
WFSC2 170 160 15 – 2.5 1321 1.06 10.67 744.0 435.6 23793.4 5.72 7500 8000 8500 9000
WFSC3 180 160 15 – 3.0 1615 1.13 10.67 1010.5 532.9 32335.2 5.54 7000 7500 8000 8500
WFSC4 180 180 20 – 3.0 1765 1.00 10.00 1123.5 752.7 48592.8 5.05 7500 8000 8500 9000
WFSC5 150 140 15 – 2.0 937 1.07 9.33 409.3 240.1 10379.2 6.16 7000 7500 8000 8500
WFSC6 130 120 15 – 2.0 817 1.08 8.00 266.5 156.7 5208.9 5.61 5000 5500 6000 6500

Fig. 10. (a) Column elastic-plastic P/Pcr vs. v/t equilibrium paths (fy/fcr � 1.0, 1.5, 3.0, ∞), and (b) failure modes and plastic strains at collapse of the U1(L4) columns
with fy/fcr � 1.5 and (1) PCM or (2) PS support conditions.

9
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 11. Plots fu/fy vs. λG for (a) PCM, (b) PCm and (c) PS columns with (1) U, (2) C, (3) H, (4) RLC, (5) R, (6) WSC and (7) WFSC cross-sections. (this figure continues in
the next page)

10
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 11. (continued).

sizeable post-critical strength that it would be logical to expect (iv) The PS and PCm column modal participation diagrams, shown in
from a flexural-torsional post-buckling behaviour moreover, Fig. 13(c3)þ(c4), are again qualitatively similar and differ mostly
the post-critical strengths of these two columns are very similar on the relative participations of modes 2 and 4 (see also the
(at least up to θ þ θ0 ¼ 0.4 rad). On the other hand, the PCm and critical buckling mode modal decomposition depicted in Fig. 8
PS column equilibrium paths are associated with much smaller (b)). In the PS column, p2 equals 30.4% at the early loading stages
post-critical strengths and, most of all, exhibit limit points at and decreases, almost linearly, until 16.1% at P/Pcr ¼ 0.825↓ and
fairly low applied loading levels and moderate rotations: they θ þ θ0 ¼ 0.4 rad) similarly, p4 reduces from 69.2% to 52.6% in
correspond to Pmax/Pcr ¼ 0.724 þ (θ þ θ0)max ¼ 0.234 rad, for the the same loading interval. In the PCm column, p2 is much lower: it
PCm column, and Pmax/Pcr ¼ 0.847 þ (θ þ θ0)max ¼ 0.324 rad, for varies from 6.7% to 2.4%, while p4 changes from 79.1% to 55.8%.
the PS column. (v) The major behavioural difference between the post-buckling be­
(ii) The occurrence of the above limit points stems from the combi­ haviours of the column pairs F þ PCM and PS þ PCm is the much
nation of two effects, namely (ii1) the closeness between Pb.fm and more significant participation of mode 3 (minor-axis flexure),
Pcr, and (ii2) the absence of reactive end moments (mostly the particularly at the late loading stages. Indeed, unlike their F þ
minor-axis ones). This explains why a limit point never occurs in PCM counterparts, the PS þ PCm columns contain contributions
the F column and only occurs at fairly advanced post-buckling from mode 3 that grow quite rapidly along the equilibrium path
stages (not shown in Fig. 13(a)–(b)) in the PCM column. It is and cannot be explained solely on the basis of the effective
worth noting that these two columns exhibit high Pb.fm/Pcr ratios centroid shift effects – recall that such contributions are not in the
and minor-axis reactive end moments. critical-mode initial geometrical imperfection considered. The
(iii) The modal participation diagrams shown in Fig. 13(c1)þ(c2), feature responsible for the much more pronounced presence of
concerning the F and PCM columns, are qualitatively similar up to minor-axis flexure along the PS þ PCm column equilibrium paths
θ þ θ0 ¼ 0.4 rad. Indeed, the major difference between them, is the closeness between the critical (major-axis flexural-
already visible in the critical buckling mode modal decomposi­ torsional) and minor-axis flexural buckling loads, thus leading
tion displayed in Fig. 8(b), lies in the ratio between the contri­ to a fairly high amplification of the mode 3 deformations caused
butions of torsion and major-axis flexure to the flexural-torsional by the effective centroid shift effects – in other words, these two
deformations, which is considerably higher along the F column columns experience interaction between two global buckling
equilibrium path naturally, since the major-axis flexure support modes (major-axis flexural-torsional and minor-axis flexural),
conditions are fixed, instead of pinned. This ratio is quantified by which may be termed a “global-global interaction”.6 Fig. 13(c3)þ
p4 and p2, which vary within the ranges (iii1) 82.3%–74.2% and (c4) show that p3 reaches 25.9% (PS column) and 35.4% (PCm
7.0%–7.3%, for the F column, and (iii2) 64.1%–54.0% and column) at θ þ θ0 ¼ 0.4 rad, which is clearly perceptible in the
28.3%–33.8%, for the PCM column. In addition, Fig. 13(c1)þ(c2) mid-span deformed configurations displayed in Fig. 14.
evidence of the emergence of mode 3 (minor-axis flexure) at the
early post-buckling stages, which stems from effective centroid
shift effects caused by the cross-section single symmetry. The 6
Note that even the GBT modal features are unable to separate the mode 3
participation of this deformation mode reaches moderate levels: contributions stemming from the (i) centroid shift and (ii) “global-global
10.1% and 5.6% at θ þ θ0 ¼ 0.4 rad, for the F and PCM columns, interaction” effects. A similar finding was reported in Ref. [22], in the context
respectively. of the distortional-global (minor-axis flexural) interaction in lipped channel
columns containing critical-mode (distortional) initial geometrical
imperfections.

11
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 12. Plots fu/fnG vs. λG for (a) PCM, (b) PCm and (c) PS columns with (1) U, (2) C, (3) H, (4) RLC, (5) R, (6) WSC and (7) WFSC cross-sections. (this figure continues in
the next page)

12
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 12. (continued).

Table 3
Means, standard deviations, maximum/minimum values of the numerical-to-predicted ultimate strength ratios provided by the current DSM global design curve for the
PCM, PCm and PS columns.
PCM columns PCm columns PS columns

λG �1.5 >1.5 All �1.0 >1.0 All �1.5 >1.5 All

n 155 545 800 127 673 800 255 545 800

Mean 1.006 1.262 1.181 1.026 0.865 0.891 0.987 0.967 0.973
S. Dev.
0.025 0.181 0.191 0.029 0.046 0.074 0.037 0.024 0.030
Max
1.083 1.922 1.922 1.091 1.060 1.091 1.080 1.013 1.080
Min
0.949 0.966 0.949 0.942 0.803 0.803 0.904 0.909 0.904

(vi) The modal participation diagrams displayed in Fig. 13(c) include possible to shed further light on this issue after an in-depth investigation
tiny contributions from the local (5–15) and shear (16–27) on the “global-global interaction” mentioned above such an investi­
modes, whereas the participation from the linear transverse gation is currently under way and the authors hope to report about it in
extension modes increases gradually along the column equilib­ the near future.
rium paths, varying almost linearly with θ þ θ0. It is still worth
noting that the contribution from mode 4 (torsion) is always
dominant, regardless of whether the column is (PS þ PCm) or is 5.2. Proposal of DSM-Based design approaches for pin-ended columns
not (F þ PCM) susceptible to the occurrence of global-global
interaction see also the cross-section deformed configurations As shown in Section 5.1, the PS column FMT failure loads are
displayed in Fig. 14. adequately (safely and accurately) predicted by the current DSM
strength curve, regardless of the slenderness range this is not sur­
At this stage, it should be mentioned that, on the basis of this very prising if one realizes that this strength curve was developed almost
limited investigation and regardless of the cross-section shape, it is exclusively on the basis of experimental and numerical failure loads
virtually impossible to select column geometries that preclude the concerning columns (mostly hot-rolled) with these end support
closeness between the major-axis flexural-torsional (critical) and minor- conditions.
axis flexural buckling loads in the presence of end supports pinned with Conversely, it was also shown in Section 5.1 that the PCM and PCm
respect to minor-axis flexure, namely in the PS and PCm columns (the Pb. column FMT failure loads are only adequately estimated by the current
fm/Pb.FT ratios tend to be lower in the latter). Conversely, in columns DSM strength curve in the low-to-moderate slenderness range this has
with end supports fixed with respect to minor-axis flexure (F and PCM been shown to be true also for the F columns [1,2] (see Section 2).
columns), it is possible to enforce or preclude such buckling load Indeed, (i) the PCM column FMT failure loads are underestimated for λG
closeness through an adequate column geometry selection. In view of (�λFT)>1.5 (but by smaller amounts than the F columns) and (ii) the
these post-buckling strength differences exhibited by the F, PCM, PCm PCm column FMT failure loads are slightly overestimated for λG
and PS columns, obviously not properly reflected by the associated FMT (�λFT) > 1.0. Proposals of DSM-based approaches for the design of PCM
buckling loads, the need to adopt different design/strength curves for and PCm columns are presented in the next sections they are aimed at
columns with distinct end support conditions (see the next sections of improving the failure load predictions provided by the current DSM
this paper) is not at all surprising. Moreover, note also that it will be design curve.

13
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 13. U4(L1) F, PCm, PCM and PS column elastic equilibrium paths (a) P/Pcr vs. θ þ θ0 and (b) P/Pcr vs. dm/t, and (c) modal participation diagrams.

Fig. 14. U4(L1) F, PCm, PCM and PS column elastic equilibrium paths P/Pcr vs. θ þ θ0 and associated mid-span cross-section deformed configuration evolutions (at 0.1,
0.2, 0.3 and 0.4 rad).

5.2.1. PCM columns 8

Naturally, the first idea that sprung to mind to improve the quality of < 0:06βFT þ 1:25 if βFT < 12:5
b¼ (9)
the PCM column failure load predictions was to use the strength curve set :
2 if βFT � 12:5
developed for F columns (fnFT), which depends on the cross-section
geometric parameter βFT (see Section 3) recall that the exponential
while parameter a is still given by Eq. (4) the new flexural-torsional
expression is kept in the low-to-moderate slenderness range (λFT � 1.5).
strength curves still coincide with Eq. (1) for high βFT values (i.e.,
However, this strength curve set leads to excessively unsafe failure load
a ¼ 0.877 and b ¼ 2). Tables B.1-B.7 of the data file described in the
predictions indeed, for λFT > 1.5 the associated fu/fnFT average, stan­
Annex provide the fnFT estimates and associated numerical-to-predicted
dard deviation, maximum and minimum values read 0.839-0.099-
fu/fnFT ratios for a representative sample of the PCM columns analysed in
1.038-0.639. Therefore, it is necessary to develop new flexural-
this work. In order to assess the merits of the new strength curve set,
torsional strength curves for PCM columns in the moderate and high
Fig. 15(a)-(g) plot fu/fnFT against λFT for the U, C, H, RLC, R, WSC and
slenderness ranges.
WFSC columns, showing also the fu/fnFT statistical indicators. It is found
Following the same strategy already adopted for the F columns in
that:
Section 3, a new strength curve set, also dependent on βFT, is sought for
the PCM columns. After grouping the columns according to their βFT
(i) The proposed strength curve set leads to a fairly uniform failure
values, a “trial-and-error curve-fitting procedure” leads to a new
load prediction quality along the whole slenderness range (see
expression for parameter b(βFT),
Fig. 15(a)–(g)). For the various column sets, the fu/fnFT statistical
indicators read 1.028-0.032-1.108-0.969 (U columns), 1.016-

14
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

0.042-1.117-0.874 (C columns), 1.076-0.064-1.216-0.961 (H columns these figures also show the fu/fnFT statistical indicators. These
columns), 1.028-0.028-1.100-0.949 (RLC columns), 1.030-0.026- results prompt the following remarks:
1.096-0.956 (R columns), 1.017-0.035-1.084-0.933 (WSC col­
umns) and 1.026-0.033-1.093-0.959 (WFSC columns). (i) The fu/fnFT ratios provide ample evidence of the prediction
(ii) Considering now all the PCM columns analysed, the improve­ quality improvement achieved with the proposed design
ment, with respect to the predictions provided by the current curve the fu/fnFT averages, standard deviations and maximum/
DSM design curve, is also very impressive: for λFT > 1.5, the sta­ minimum values are 1.132-0.044-1.203-1.045 (U columns),
tistical indicators read 1.044-0.045-1.216-0.874 (fu/fnFT) vs. 1.108-0.051-1.185-0.980 (C columns), 1.135-0.045-1.204-1.019
1.262-0.181-1.922-0.966 (fu/fnG). Moreover, the failure load (H columns), 1.097-0.054-1.169-0.967 (RLC columns), 1.098-
prediction quality is now the same across the whole slenderness 0.054-1.162-0.968 (R columns), 1.083-0.046-1.146-0.942 (WSC
range the associated LRFD resistance factors, for λFT � 1.5 and columns) and 1.092-0.044-1.150-0.970 (WFSC columns).
λFT > 1.5, are ϕ¼ 0.93 (n ¼ 255, Pm ¼ 1.006, VP ¼ 0.025) and (ii) The improvement, with respect to the current global design curve
ϕ¼ 0.95 (n ¼ 545, Pm ¼ 1.044, VP ¼ 0.043), respectively. predictions, is very significant again for λFT > 1.0. The statistical
indicators concerning all the columns analysed read 1.120-0.040-
5.2.2. PCm columns 1.204-0.952 (fu/fnFT) vs. 0.865-0.046-1.060-0.803 (fu/fnG), which
The proposal to improve the FMT failure load prediction quality of means that the failure load prediction quality is now fairly similar
the PCm U, C, H, RLC, R, WSC and WFSC columns consists of a single across the whole slenderness range the associated LRFD resis­
DSM-based strength curve (fnFT), which only differs from the current one tance factors, for λFT � 1.0 and λFT > 1.0, are ϕ¼ 0.94 (n ¼ 127,
for λFT > 1.0 recall, from Section 5.1, that all the unsafe ultimate Pm ¼ 1.026, VP ¼ 0.028) and ϕ¼ 1.03 (n ¼ 673, Pm ¼ 1.120,
strength predictions provided by the current DSM design curve concern VP ¼ 0.036), respectively.
columns in this slenderness range (see Fig. 12(b1)-(b7)). It was found
that perfectly adequate FMT failure load predictions are achieved by 6. Summary of the proposed DSM-Based strength curves
merely (i) considering the current exponential expression only up to
λFT ¼ 1.0 and (ii) lowering the Euler curve to match its end value. The proposed DSM-based strength approaches for the design of cold-
Therefore, the proposed strength/design curve is defined by formed steel F and PCM columns collapsing in FMT modes can be cast in a
8
� � unified form, by means of the expressions
λ2
> fy 0:658 FT if λFT � 1:0 sffiffiffiffiffiffiffiffi � �
< fy 8 2

� � (10) > f 0:658λFT if λFT � 1:5 sffiffiffiffiffiffiffiffi

fnFT ¼
> 0:658
with λFT ¼ < y fy
f
: f
y 2
if λFT > 1:0
crFT
fnFT ¼ � � with λFT ¼ (11)
λFT >
: a fcrFT
fy if λ FT > 1:5
λbFT
Fig. 16(a)-(b) illustrate the comparison between the fu/fy values of
the C and R columns and the current (dashed fnG) and proposed where a ¼ 0.39 � 1.5b and b is given by either
(solid fnFT) DSM strength curves similar plots can be obtained for the 8
remaining U, H, RLC, WSC and WFSC columns. It is readily noticed that < 0:06βFT þ 0:71 if βFT < 21:5
the fu/fy values of the columns with λFT > 1.0 are much better predicted b¼ (12)
:
by the proposed strength curve. 2 if βFT � 21:5
Tables C.1-C.7 of the data file described in the Annex provide the fnFT
estimates and corresponding numerical-to-predicted ratios fu/fnFT for a or
representative sample of the PCm columns analysed. In order to assess 8
< 0:06 βFT þ 1:25 if βFT < 12:5
the merits of the proposed strength curve, Fig. 17(a)-(g) plot, against λFT, b¼ (13)
the fu/fnFT values of all the U, C, H, RLC, R, WSC and WFSC :
2 if βFT � 12:5

Fig. 15. PCM columns: plots fu/fnFT vs. λFT for the (a) U, (b) C, (c), H, (d) RLC (e) R, (f) WSC and (g) WFSC columns.

15
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Fig. 16. Comparison of the PCm (a) C and (b) R column fu/fy values with the current (fnG) and proposed (fnFT) strength curves.

Fig. 17. PCm columns: plots fu/fnFT vs. λFT for (a) U, (b) C, (c), H, (d) RLC, (e) R, (f) WSC and (g) WFSC columns.

respectively for the F and PCM columns recall that the cross-section columns), all well above the value recommended by the North American
geometric parameter βFT reads Specification [12] for compression members ϕc¼ 0.85.

βFT ¼
II þ Iw =A
(14) 7. Conclusion
III
This work reported the results of a fairly extensive numerical
where (i) II and III are the major and minor moments of inertia, (ii) Iw the investigation aimed at assessing the accuracy of the currently codified
warping constant and (iii) A the gross cross-section area. As for the PCm Direct Strength Method (DSM) global strength curve in predicting the
columns, the proposed strength/design curve is defined by ultimate strength of cold-formed steel columns failing in flexural-
8
� 2
� torsional modes. The first part of the paper was devoted exclusively to
fy 0:658λFT if λFT � 1:0 sffiffiffiffiffiffiffiffi
>
< fy fixed-ended columns (F columns) and continued recent work by the
fnFT ¼ � � with λFT ¼ (15) authors [1,2] on improving the column flexural-torsional (FMT) failure
: f 0:658
> fcrFT
y 2
i λFT > 1:0 load estimation in the moderate and high slenderness ranges (λFT > 1.5).
λFT
The aim was overcoming a non-negligible drawback of the strength
Table 4 shows the values involved in the calculation of the LRFD curve set proposed in the above works, namely the fact that distinct
resistance factor ϕ concerning the proposed DSM-based global strength strength curve sets are needed for columns with cross-sections exhibiting
curves to estimate all the column failure load sets considered in this one-wall and two-wall end stiffeners.
work a distinction is made between the slenderness ranges unaffected The second part of the paper extended the above investigation to
and affected by the strength curve proposal (the current DSM design columns with three types of pinned end supports, all fixed with respect
curve is kept in the former). The table also includes the values obtained to torsion and having warping prevented: end cross-sections attached to
with the current DSM strength curve for the PS columns. It is readily rigid plates resting on either spherical (PS columns) or cylindrical
observed that the proposed DSM-based strength curves lead to a high- hinges the latter may be pinned with respect to major-axis or minor-
quality FMT failure load prediction along the whole slenderness ranges. axis bending (PCM or PCm columns, respectively). After addressing the
Indeed, the corresponding LRFD resistance factors are ϕ¼ 0.97 (F col­ column geometry selection, a parametric study was carried out to gather
umns), ϕ¼ 0.95 (PCM columns), ϕ¼ 1.03 (PCm columns) and ϕ¼ 0.89 (PS flexural-torsional failure loads of columns (i) exhibiting different cross-

16
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

Table 4
Means, standard deviations, maximum/minimum values and LRFD resistance factors (φ) of the numerical-to-predicted ultimate strength ratios for F, PCM, PCm and PS
columns failing in FTM modes.
F columns PCM columns PCm columns PS columns

λG �1.5 >1.5 All �1.5 >1.5 All �1.0 >1.0 All �1.5 >1.5 All

n 698 1012 1710 255 545 460 127 673 800 255 545 800
Mean 1.051 1.068 1.061 1.006 1.044 1.049 1.026 1.120 1.105 0.987 0.967 0.973
S. Dev.
0.033 0.064 0.054 0.025 0.045 0.049 0.029 0.040 0.052 0.037 0.024 0.030
Max
1.170 1.308 1.308 1.083 1.216 1.216 1.091 1.204 1.204 1.080 1.013 1.080
Min
0.939 0.845 0.845 0.949 0.874 0.874 0.942 0.952 0.942 0.904 0.909 0.904
ϕ – – 0.97 – – 0.95 – – 1.03 – – 0.89

section shapes (plain channels, lipped channels, return lipped channels, to LRFD resistance factors higher than 0.90, i.e., considerably
hat-sections, rack-sections, web-stiffened lipped channels and web- above the value currently recommended by the North American
flange-stiffened lipped channels), (ii) various geometries (cross-section Specification [12] for compression members (φc ¼ 0.85), thus
dimensions and lengths) and (iii) covering wide slenderness ranges. providing strong encouragement to search for further validation.
Then, after showing that the codified DSM design curve is only able to
predict adequately the flexural-torsional failure loads of the PS columns, It should also be pointed out that further research is needed in order
a GBT-based numerical investigation, aimed at comparing the structural to acquire in-depth structural insight on the differences between the
responses of fixed-ended and pin-ended plain channel columns, was post-buckling mechanics of the F, PCM, PCm and PS columns in
presented and discussed. Next, the failure load data obtained were used particular, it is indispensable to address the roles played by (i) the
to propose and assess the merits of newly proposed DSM-based strength relative importance of the major-axis flexure and torsion contributions
curves to improve the failure load prediction quality of the PCM and PCm to the column critical buckling mode and (ii) the minor-axis flexural
columns with moderate and high slenderness. displacements stemming from effective centroid shift effects and/or
Among the various findings of this investigation, the following interaction with minor-axis flexural buckling (“global-global interac­
deserve to be specially mentioned: tion”). The authors are currently investigating these issues by means of
shell finite element and GBT non-linear analyses the modal nature of
(i) The new strength curve set proposed for F columns, dependent on the latter is ideally suited to shed fresh light on the above post-buckling
a cross-section normalised geometric parameter βFT (involving behavioural differences. The output of this investigation will be reported
the area, major and minor moments of inertia, and warping in the near future.
constant), leads to excellent numerical failure load predictions for Finally, one last word to mention that (i) additional numerical and
all the columns analysed (with either one-wall or two-wall end (mostly) (ii) experimental validation is indispensable before the DSM-
stiffeners) with λFT > 1.5 the current DSM design curve pro­ based column global design curves proposed in this work (possibly
vides adequate failure load estimation for λFT � 1.5. Moreover, modified/improved) can be considered for codification in particular,
the available six lipped channel column experimental failure it is essential to assess how efficiently they predict the experimental
loads are also quite well estimated by the new strength curves. flexural-torsional failure loads of columns in the moderate and high
Nevertheless, further test results are needed before adequate slenderness range. In addition, it is also indispensable to assess the
experimental validation can be claimed. merits of the proposed DSM-based strength curves in prediction of
(ii) The PS column flexural-torsional failure loads are efficiently flexural-torsional failure loads of columns with partially restrained end
(safely and accurately) predicted by the current DSM design flexural rotations.
curve, regardless of the slenderness range. This is not surprising,
since this strength curve was developed on the basis of experi­ Appendix A. Supplementary data
mental and numerical failure loads concerning columns (mostly
hot-rolled) with these end support conditions, namely end cross- Supplementary data to this article can be found online at https://doi.
sections pinned with respect to minor-axis and major-axis org/10.1016/j.tws.2019.106461.
bending, fixed with respect to torsion and having warping fully
prevented (i.e., rigidly attached to rigid plates resting on spher­ References
ical hinges).
(iii) The PCM and PCm column flexural-torsional failure loads are only [1] P.B. Dinis, D. Camotim, A.D. Martins, On the accuracy of the current Direct
Strength Method (DSM) design curve for columns failing in global modes, in:
adequately estimated by the current DSM design curve in the low- Website Proceedings of Structural Stability Research Council (SSRC) Annual
to-moderate slenderness range. Indeed, the PCM column failure Stability Conference, 2018 (Baltimore, 10-13/4), (72 pages).
loads are underestimated for λFT > 1.5 (but by smaller amounts [2] P.B. Dinis, D. Camotim, A. Landesmann, A.D. Martins, On the Direct Strength
Method design of columns against global failure, Thin-Walled Struct. 139 (June)
than the F columns), and the PCm column failure loads are (2019) 242–270.
slightly overestimated for λFT > 1.0. [3] B.W. Schafer, Review: the direct strength method of cold-formed steel member
(iv) A modification of the βFT-dependent strength curve set developed design, J. Constr. Steel Res. 64 (7–8) (2008) 766–778.
[4] D. Camotim, P.B. Dinis, A.D. Martins, Direct Strength Method (DSM) a general
for the F columns was proposed and shown to provide a high- approach for the design of cold-formed steel structures, in: C. Yu (Ed.), Recent
quality estimation of the PCM column flexural-torsional failure Trends in Cold-Formed Steel Construction, Woodhead Publishing (Amsterdam
loads analysed in this work. Series in Civil and Structural Engineering), 2016, pp. 69–105.
[5] B.W. Schafer, T. Pek€ oz, Direct strength prediction of cold-formed steel members
(v) A single DSM-based strength curve was proposed to estimate the
using numerical elastic buckling solutions, in: W.W. Yu, R. LaBoube (Eds.),
PCm column FMT failure loads. It only differs from the current one Proceedings of 14th International Specialty Conference on Cold-Formed Steel Structures
for λFT > 1.0 and provides high-quality failure load estimates, (St. Louis, 15-16/10), 1998, pp. 69–76.
which clearly outperform those yielded by the current design [6] G.J. Hancock, Y.B. Kwon, E.S. Bernard, Strength design curves for thin-walled
sections undergoing distortional buckling, J. Constr. Steel Res. 31 (2–3) (1994)
curve. 169–186.
(vi) Although the failure load data considered in this work was [7] R. Ziemian (Ed.), Guide to Stability Design Criteria for Metal Structures, sixth ed.,
necessarily limited, the proposed DSM-based strength curves lead John Wiley & Sons, Hoboken, 2010.

17
P.B. Dinis et al. Thin-Walled Structures xxx (xxxx) xxx

[8] AISI (American Iron and Steel Institute), Cold-Formed Steel Design Manual, 1996 [16] A.D. Martins, D. Camotim, R. Gonçalves, P.B. Dinis, Enhanced geometrically non-
(Washington DC). linear Generalized Beam Theory (GBT) formulation: derivation, numerical
[9] T. Pek€oz, O. Sümer, Design Provisions for Cold-Formed Steel Columns and Beam- implementation and illustration, J. Eng. Mech. 144 (6) (2018) paper 04018036 (20
Columns, Research Report RP92-1, American Iron and Steel Institute, Washington pages).
DC, 1992. [17] Simulia Inc, ABAQUS Standard, 2009, version 6.9-3.
[10] AISC (American Institute of Steel Construction), Load and Resistance Factor Design [18] D. Popovic, G.J. Hancock, K.J.R. Rasmussen, Axial compression tests of cold-
Specification for Structural Steel Buildings, AISC, Chicago IL, 1986, 328:1986. formed angles, J. Struct. Eng. 125 (5) (1999) 515–523.
[11] AISI (American Iron and Steel Institute), Cold-Formed Steel Design Manual, 1986 [19] E. Ellobody, B. Young, Behavior of cold-formed steel plain angle columns, J. Struct.
(Washington DC). Eng. 131 (3) (2005) 469–478.
[12] AISI (American Iron and Steel Institute), North American Specification (NAS) for the [20] P.B. Dinis, D. Camotim, A. Landesmann, A.D. Martins, Accurate Direct Strength
Design of Cold-Formed Steel Structural Members (AISI-S100-16) þ Respective Method (DSM) prediction of column flexural-torsional failure loads, in: Website
Commentary, 2016 (Washington DC). Proceedings of Structural Stability Research Council (SSRC) Annual Stability
[13] Y. Bandula Heva, M. Mahendran, Flexural-torsional buckling tests of cold-formed Conference (St. Louis, 2-5/4), 2019.
steel compression members at elevated temperatures, Steel Compos. Struct. 14 (3) [21] Z. Li, J.C.B. Abreu, J. Leng, S. Ad�
� any, B.W. Schafer, Review: constrained finite strip
(2012) 205–227. method developments and applications in cold-formed steel design, Thin-Walled
[14] W.S. Santos, A. Landesmann, D. Camotim, Distortional strength of end-bolted CFS Struct. 81 (August) (2014) 2–18.
lipped channel columns: experimental investigation, numerical simulations and [22] A.D. Martins, D. Camotim, R. Gonçalves, P.B. Dinis, On the mechanics of
DSM design, Thin-Walled Struct. (2019) accepted for publication, 2019. distortional-global interaction in fixed-ended columns, Thin-Walled Struct. 123
[15] R. Bebiano, D. Camotim, R. Gonçalves, GBTul 2.0 – a second-generation code for (February) (2018) 162–184.
the GBT-based buckling and vibration analysis of thin-walled members, Thin-
Walled Struct. 124 (2018) 235–257. March.

18