See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/361218280

Local and global instabilities of rolled T-section columns under axial

compression

Article  in  Thin-Walled Structures · September 2022

DOI: 10.1016/j.tws.2022.109517

CITATIONS READS
2 168

3 authors:

Liya Li Mario Fafard

Laval University Laval University
8 PUBLICATIONS   31 CITATIONS    284 PUBLICATIONS   2,539 CITATIONS   

SEE PROFILE SEE PROFILE

Nicolas Boissonnade
Laval University
108 PUBLICATIONS   629 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Discrete Element Method Simulation of Carbon Anode Paste View project

Lightmechanics View project

All content following this page was uploaded by Liya Li on 02 September 2022.

The user has requested enhancement of the downloaded file.

 Local and global instabilities of rolled T-section columns under axial

compression

Author names

Liya Li, Ph.D., Laval University, Québec, Canada

Mario Fafard, Ph.D., Laval University, Québec, Canada

Nicolas Boissonnade, Ph.D., Laval University, Québec, Canada

Abstract

The resistance capacity of rolled T-section columns comprising slender webs is investigated in this

paper. Both (i) cross-sectional behavior influenced by local buckling and (ii) member behavior

influenced by global buckling are investigated through extensive numerical analyses. Regarding cross-

section resistance, current design rules rely on the traditional cross-section classification system and

the Effective Width Method (EWM), while local-global coupled instabilities in members with slender

plate elements are usually addressed through the EWM combined with flexural-torsional buckling

curves, which leads to conservative and scattered predictions for T-section members, mainly because

the effects of torsional buckling are considered twice: through the EWM and through the member

buckling curves. To overcome these shortcomings, an Overall Interaction Concept (OIC) approach is

proposed in this paper, providing a more economic and simple design method for both T-section

members. The load resistances predicted by the OIC, Eurocode 3, the American Specifications and the

Australian Standards are compared with numerical and experimental results. Overall, the results show

that the OIC-based approach provides more accurate and consistent predictions than current design

recommendations and is suitable to be included in current steel structural design standards.

1 
 
1 Introduction

As manufacturing technology advances, steel structural components comprising high strength steel

thin-walled plate elements are becoming more widespread. Such elements indeed provide an improved

performance-to-weight ratio, yet they show more prone to local buckling. The latter phenomenon may

also further influence the member’s response and interact with flexural buckling or lateral torsional

buckling, leading to so-called “coupled instabilities”. Dating back to 1969, Van der Neut [1]

established a theoretical framework to characterize the response of members affected by such coupled

instabilities. After that, many researchers (see for example [2-10]) have investigated various

particularities of mode interactions, such as highly unstable post-buckling behavior or high sensitivity

to imperfections.

This paper focuses on the local, global and local-global interactive buckling of T-section members

under axial compression. T-section members, which are often used as connection members between I-

sections and other members, as chord members in steel roof trusses or as tie beams in precast concrete

desk slabs [11,12] (see Fig. 1), can be easily obtained by connecting two angle sections back-to-back,

by plate welding or by cutting hot-rolled I-sections into half as shown in Fig. 2. The latter ones,

considered in this study, are ideal substitutes for double-angle sections since they are easier to

manufacture and allow for simpler anti-corrosion treatments and coatings.

Fig. 1. Composite Steel Tee Concrete Deck [11].

2 
 
 r 0 r 0

a) b) c)
Fig. 2. Manufacturing T-section members from – a) Double angles back-to-back – b) Plate welding – c) Tee cut from a
hot-rolled I-section.

Although steel T-section members have been widely used in construction, research on the local and

local-global coupled instabilities of T-section members has received much less attention than I-shaped

and angle sections, and only limited experimental research has been carried out. An early attempt can

be found dating back to 1972 by Kennedy [13], who tested 27 Tee struts and proposed the member

capacity to be determined from the minimum theoretical buckling loads, i.e., among the local buckling

load, the flexural buckling load and the flexural-torsional buckling load. Although these test results

were in reasonable agreement with his proposal, the possibility of interactions between local and global

buckling was ignored. Later on, similar tests were performed in 1986 by Kitipornchai [14] who carried

out a series of experimental tests followed by theoretical analyses on the inelastic buckling of T-section

columns under axial compression. The results showed that the failure mode of T-section columns with

wide flanges was flexural buckling, while the failure type of other members was flexural-torsional

buckling. In more recent years, Luo [15] analyzed the overall stability capacity of eight long T-section

columns loaded through their centroid, shear center and a mid-point between the centroid and the shear

center. It was found that lower capacities were achieved by the columns where compression was

applied through the shear center, eccentric loading increasing the compression stresses in flange plates.

In contrast, Chen’s study [16] revealed that for T-section members with aspect ratios h / b higher than

1.25, higher resistances could be obtained when compression acts through the shear center since

flexural-torsional buckling is not dominant and only flexural buckling can happen.

3 
 
 b
Shear Centre

tf Centroid

h
z tw

a) b) c)
Fig. 3. Buckling deformation of T-section columns – a) Section dimensions and notations – b) Local buckling
mode – c) Global buckling mode (flexural-torsional buckling).

Further, the similarity between plate local buckling and member torsional buckling of T-section

columns as shown in Fig. 3 has received special attention [17-22]. In particular for a thin-walled T-

section column with identical flange and web elements, i.e., tf = tw and h / b = 0.5, the interaction

between elements can be ignored and the torsional buckling critical stress is identical to the plate

critical stress [17]. As for T-section column with a stocky flange and slender web elements, which is

the case studied in this paper, the web plate may deflect besides torsional twist because the stocky

flange can provide a better support to the web plate. As an illustration, Fig. 4 presents the elastic

buckling behavior of a T-section column computed from software GBTUL [23] in which Ncr is the

critical load and L is the length of the member. With an increase in L, the value of Ncr first fluctuates

mildly, in a typical local plate buckling pattern, and then drops suddenly. More specifically, two

separate stages can be observed: short columns – L < 3800 mm in this example – buckle in plate local-

torsional mixed mode whereas long columns buckle in member flexural-torsional mixed mode. In both

stages, the torsional buckling mode is dominant (see Fig. 4b modal participation), despite half-

wavelengths being different.

4 
 
 Fig. 4. Buckling behavior of T-section column with b = h = 300 mm, tf =10 mm and tw = 7.5 mm – a) Signature
curve – b) Modal participation.

In addition to the elastic buckling behavior, Trahair [18] found that the local and torsional post-

buckling behavior are also very similar, causing resistance predictions to be over-conservative when

both buckling effects are considered. Other researchers [19-22] have reached similar conclusions.

Fig. 5. Comparison among column buckling curves for T-section columns.

This particular buckling behavior makes it difficult to develop accurate design rules for T-section

members in current standards. Regarding cross-section resistance, current code-oriented design relies

on the traditional cross-section classification system and the Effective Width Method (EWM), whose

deficiencies were discussed in many papers, e.g., [24-27]. For members with slender plate elements,

local-global coupled instabilities are usually addressed by the EWM (1st step) combined with member

buckling curves (2nd step) in current design standards, to consider the detrimental effects of local

buckling on member instabilities. The design procedures of Eurocode 3 (EC 3) [28,29], of the

5 
 
American Specifications (AISC) [30] and of the Australian Standards (AS) [31] for T-section columns

have conveniently been summarized by Cardoso [20,21] and Taras [19] in detail. Three aspects are

worth noticing herein:

 Fig. 5 shows a comparison among European, American and Australian column buckling curves

(curves "a0” to “d”) for non-slender T-shaped section columns. Note that Eurocode 3 proposes up

to 5 column buckling curves and that only curve “c” is used for the design of T-section columns.

The global slenderness λ̅ G used in Fig. 5 is defined in Eq. (1), where Npl represents the plastic

capacity of the T-section column under axial compression and Ncr,G the (global) critical buckling

load. According to the classical theory of elastic stability [22], Ncr,G for T-section members shall be

kept as the minimum of (i) the two flexural buckling loads Ncr,y and Ncr,z as defined in Eq. (2) and

(3), where Iy and Iz designate the second moments of area about the principal axes and E is elastic

modulus, and (ii) the flexural-torsional buckling load Ncr,FT as defined in Eq. (4) and (5), in which

G is the shear modulus, Iω the warping constant, r0 the polar radius of gyration and zc the distance

between centroid and shear centre. In particular for T-section column with Iy > Iz, Ncr,FT is always

lower than Ncr,y and Ncr,z [20,21]. For these sections, both Eurocode 3 and the American

Specifications examine the torsional buckling mode twice – in the local buckling check and in the

global buckling check through flexural-torsional column buckling curves. In contrast, the Australian

Standards consider the torsional buckling mode in a 1st step, followed by relying on flexural

buckling curves in a 2nd step;

G  N pl Ncr,G (1)

Ncr,y  π 2 EI y L2 (2)

Ncr,z  π 2 EI z L2 (3)

GI t   2 EI ω L2
N cr,T  (4)
r02  zc2

6 
 
 N cr,z  N cr,T   ( N cr,z  N cr,T ) 2  4 N cr,z N cr,T r02 r
0
2
 zc2 
N cr,FT  (5)
2r02 r0
2
 zc2 

 Because the mid-thickness lines of the web and flange of T-section members intersect at the shear

center, the primary warping constant II of T-section members is zero. Besides, since the secondary

warping constant III is small and only affects very short columns [20,21], Eurocode 3 and the

American Specifications neglect the secondary warping of T-section members and assume

Itot = II + III ≈ 0. Therefore, with a decrease in L, the member slenderness ̅λG,FT cannot approach

zero for T-section members whose flexural-torsional buckling mode is dominant. Accordingly, the

buckling curves cannot be appropriate in this case;

 Eurocode 3 considers an extra bending moment arising from the possible shift eN from the centroid

of the gross section area A to the centroid of the effective section area Aeff calculated by the EWM

Interaction equations for combined loading design check should therefore be applied for both cross-

sectional and member design, which might lead to more conservative predictions [26], further to

more complicated calculations.

Besides, various new methods such as the Erosion of Critical Bifurcation Load (ECBL) approach

[4,32,33], the Direct Strength Method (DSM) [8,34-36] or the Overall Interaction Concept (OIC)

[10,27] also provide design guidelines to address interactive buckling in steel members. In this paper,

an OIC approach is proposed, with the main objective to provide a more economic and simple design

method for both T-shapes sections and members. The OIC, which is based on the well-established

resistance-instability interaction thanks to a definition of generalized relative slenderness, abandons

the cross-section classification concept as well as the EWM, and deals with all cross-section shapes in

a similar way, both at the section and member levels, for simple or combined load cases. More

precisely, the OIC uses the section plastic capacity as a reference and considers penalty factors χ (one

for cross-sectional resistance L and another for member resistanceG) to account for buckling

detrimental effects, imperfections and their interactions. This approach is therefore capable of

7 
 
capturing all types of sections’ responses, from stocky to slender ones. Fig. 6 presents the general OIC

flow chart with the steps that need to be followed to check the ultimate resistance according to the OIC

methodology. More details on the practical use of the proposed approach are given below:

 Step 1: calculate the plastic capacity under axial force of the section Npl;

 Step 2: calculate the local (L) and global (G) critical loads (Ncr,L and Ncr,G), which are relative to

either local or global instabilities, without considering any imperfections. Critical loads can be

obtained from approximate formulas [28-31,37], or through specific tools [27];

 Step 3: take the balance between the influences of plasticity and instability (i.e., material vs

geometrical non-linearities) by defining a generalized local relative slenderness L  Npl Ncr,L

and a global relative slenderness G  N pl Ncr,G ;

 Step 4: determine separate reduction coefficients χL (local) and χG (global) which consider all

resistance-stability interactions and the effects of imperfections;

 Step 5: determine the ultimate compression resistance of T-section columns Nu. In this last step,

load reduction coefficients χi’s are also multiplied by (i) a local-global coupling factor fL/G and by

(ii) the plastic resistance Npl to obtain the ultimate resistance of cross-sections or members.

Assuming a partial safety factor equal to unity, when only local buckling behavior is considered,

Nu,L = χL ꞏ Npl; when only global buckling behavior is considered, Nu,G = χG ꞏ Npl; when considering

local-global coupled instabilities, Nu,L+G = χL ꞏ χG ꞏ fL/G ꞏ Npl where fL/G is a local-global coupling factor

which accounts for the interaction between local and global buckling [10]. Partial safety factors ϕ or

γM shall eventually be added to address reliability aspects of the proposed design equations.

8 
 
 Cross-sectional local behaviour (L) Member global behaviour (G)

Ncr,L Npl = fy ꞏ A Ncr,G

3 key factors: calculated by advanced tools (or by formulae)

N pl N pl
L  G 
N cr,L N cr,G

L G
Cross-section buckling curve Member buckling curve
fL/G
L Plastic resistance 1L G Plastic resistance
Critical load Critical load
Cross-section local Member global
buckling curve 1 buckling curve
0 G

L  L+G   L   G  f L G G

Design compression resistance Nb,L+G: Note:

subscript “L” – only consider the local buckling effect;
N u,L+G   L+G  N pl  M (Eurocode 3 format)
subscript “G” – only consider the global buckling effect;
N u,L+G   L+G  N pl   (AISC format) subscript “L+G” – both local and global buckling effects are considered.

Fig. 6. OIC design flow chart for cross-section and member design.

As a particular point and as discussed before, one may note that a torsional twist of the section about

the shear centre may occur in both local and global buckling modes, causing the local and global

buckling behaviors to appear quite similar. Hence, considering the torsional buckling mode in both χL

and χG may lead to overconservative results [18-22]. In the present paper, the member buckling curve

only intends at taking into account the effect of flexural buckling, such as in the Australian Standard.

Therefore, Ncr,G (Step 3) is taken as the minimum value of Ncr,y and Ncr,z, and the value of χG (Step 4)

excludes the effect of torsional buckling.

A few years ago, Boissonnade et al. [27] first described the mechanical background, principles and

application steps of the OIC Consecutively, OIC-based design approaches have been developed for

square and rectangular hollow sections [25,38,39], I-sections [10,26,40,41] and mono-symmetric

sections [42,43]. Along all these studies, OIC expressions were proved to be more precise and

consistent than current design proposals.

9 
 
This paper aims to extend the application of the OIC to the design of T-section members. Both cross-

sectional behavior as influenced by local buckling and member behavior affected by global buckling

are investigated through extensive numerical analyses. The modeling procedure and relevant

validation results are provided in Section 2. Based on validated numerical models, a parametric study

was carried out (Section 3) to analyze the possible effects of section dimensions, section/member

slenderness and steel grade. Then, OIC-based design formulae for T-section columns are proposed.

The resistances predicted by OIC, Eurocode 3, the American Specifications and the Australian

Standards are compared with numerical and experimental results in Section 4, where a reliability

analysis based on EN 1990 [44] is also carried out.

2 Numerical investigations

2.1 Basic features and assumptions

Numerical models for T-section columns under axial compression were built by using non-linear finite

element (FE) software ABAQUS [45]. The general-purpose shell element S4R, which has been applied

in many previous numerical studies on mono-symmetric and asymmetric sections [42,46-48], was used

in subsequent Geometrically and Materially Non-linear with Imperfections Analyses (GMNIA). In

order to balance out computation time and accuracy, the mesh size was selected as 1 / 20th of the web

height (see Fig. 7a). Extra hollow beam section and spring elements were added in web-to-flange zones

to simulate real geometries in the fillet areas [26,40,42,43,49], see Fig. 7b. A quad-linear stress-strain

material relationship [50], which was converted into true stress and logarithmic plastic strain, was

adopted in ABAQUS for both validation and further parametric studies.

10 
 
 a) b)
Fig. 7. Numerical modeling – a) Typical member deformation at peak load (magnified) – b) Radius area of T-section
members.

As shown in Fig. 8, all the T-section columns were assumed to be under simply supported, fork-type

boundary conditions, in which both end sections were restrained against torsional displacements and

against out-of-plane displacements about both principal axes, i.e., uy = uz = x = 0. At one end, the

compression load was applied at a reference point which was coupled to the end section through rigid

body conditions, while the longitudinal displacement ux of the opposite reference point was prevented.

Similar numerical boundary conditions were adopted in [47,51,52]. Note that, in the validation process,

a non-zero longitudinal distance between the reference point and the end-sections ex (Fig. 8) has been

considered in order to replicate the real boundary conditions, while this distance was set to zero in

parametric studies.

Fig. 8. Boundary conditions in numerical models.

The manufacturing process of T-section members, which are usually obtained by cutting hot-rolled I-

sections in half, triggers more complicated distributions of residual stresses. Unfortunately, to the

authors’ knowledge, no experimental measurements have been focused on residual stresses patterns of

these T-section members. An idealized residual stresses pattern adopted in other numerical research

11 
 
works [19-21] assumes that no extra heat is created during the cutting process and ignores a potential

straightening process, which could be conservative [19]. In the current study, a triangular residual

stresses pattern suggested in [53] (see Fig. 9a) and proved reliable by other researchers [15,54-57],

was adopted in the numerical models. Note that the distribution of residual stresses in each plate fulfils

self-equilibrium.

Fig. 9. Definition of – a) Residual stresses – b) Local geometrical imperfections – c) Global geometrical imperfections.

As for geometrical initial imperfections, no detailed physical measurements could be found in the

literature for hot-rolled T-section members. Since it is unthinkable to measure the actual distribution

of geometrical imperfection for each steel cross-section or member in daily practice, many numerical

analyses rely on “standardized” imperfection patterns (i.e., safe-sided yet suitable patterns). As

examples, close-formed analytical solutions based on energy methods to determine local buckling

modes may be found in [58-60]. In this paper, both local and global geometrical imperfections were

introduced in the FE shell models by modifying nodes’ coordinates through sine-wave functions, as

suggested in [61]. Identical approaches have been considered in previous studies for hollow sections

as well as I-sections [62-64], and this approach has been proved to be accurate enough compared with

test results. As depicted in Fig. 9b and Fig. 9c, the half-wavelength for global buckling is assumed to

be L, while the half-wavelength aL for local buckling is assumed to be associated with the flange plate

buckling length af (so-called “flat length”) and the web plate buckling length aw, where af = b – tw – 2 r

and aw = 2 (h – tf – r). In order to keep the middle section as the weakest one, odd numbers of half-

12 
 
waves have been contemplated. The amplitudes of local imperfections in the web and flange plates are

defined as alocal,w and alocal,f respectively, while aglobal represents the amplitude of both major-axis and

minor-axis global imperfections. These geometric imperfections were complemented by residual

stresses patterns, and no reductions with respect to amplitudes have been considered, such as the

suggestion in EC3 Part 1.5 Annex C [29] for bridge (welded) girders. More discussions about the

influence of local and global imperfection patterns on ultimate resistance of T-section members are

provided in Section 2.2.

2.2 Validation of finite element models

The FE models were firstly validated against experimental results reported by Kitipornchai [14] and

Luo [15], and then used in parametric studies on a wider range of section dimensions and member

slenderness. For validation purposes, the dimensions of specimens and material properties measured

through coupon tests as summarized in Table 1 were adopted in the numerical models. Key section

properties such as h / b or Iy / Iz are also presented in Table 1. Note that the h / b ratio is usually

positively correlated to the value of Iy / Iz, since larger web plates usually provide larger Iy value when

flange dimensions and web thickness are fixed. Table 1 also provides failure modes observed in

experiments as well as the ratios between the flexural buckling load Ncr,y or the flexural-torsional

buckling load Ncr,FT to the local buckling load Ncr,L for the seven specimens under axial compression.

Note that some failure modes could not clearly be identified and that some failure shapes at peak load

do not correspond to their respective buckling modes (with the lowest critical loads), owing to (i) the

similarity between local and torsional buckling behavior and to (ii) the influence of the actual

(unknown) distribution of geometrical imperfections [14].

An overall view of two experimental tests set-up is recalled in Fig. 10. For the six tests reported in

[14], all six columns were loaded concentrically through spherical and supports, which allowed

bending rotations but restrained torsional twist in the end sections. The distance between each sphere

13 
 
center and end-section ex was 92.5 mm. As for the tests carried out in [15], the experimental program

comprised three loading situations, i.e., the columns were loaded through the centroid, shear center or

the midpoint between them. The end sections were attached to knife-edge supports, which only allowed

bending rotations about the z-axis, but restrained rotations around the y-axis i.e., y = 0. The height of

each knife-edge support was 180 mm, i.e. ex = 180 mm.

a) b)
Fig. 10. Overall views of experimental test set-ups – a) Test reported in [14] – b) Test reported in [15].

14 
 
1

2 Table 1. Section geometries, material properties and test results of T-section columns.

Ratios of critical Failure

Section geometries Material properties Ultimate loads
buckling loads modes
Specimen Loading point
b h tf tw L h/b Iy / Iz E fy Ncr,y / Ncr,L Ncr,FT / Ncr,L Nu,exp.
[mm] [mm] [mm] [mm] [mm] [-] [-] [MPa] [MPa] [-] [kN]
T1 [14] 75.6 75.0 10.0 6.4 500.0 0.99 1.47 214000 289 1.22 0.83 FT 315.0
T2 [14] 75.6 75.0 10.0 6.4 600.0 0.99 1.47 214000 289 0.93 0.63 FT 320.0
T3 [14] 133.5 99.5 7.8 5.8 1600.0 0.75 0.81 214000 311 0.78 0.96 F 371.0
T4 [14] Centroid 133.5 99.5 7.8 5.8 1800.0 0.75 0.81 214000 311 0.63 0.78 F 385.0
T5 [14] 147.5 127.0 9.8 5.9 1600.0 0.86 1.03 214000 303 1.65 1.60 FT 519.0
T6 [14] 147.5 127.0 9.8 5.9 1800.0 0.86 1.03 214000 303 1.33 1.30 FT 513.0
ZZ [15] 149.0 148.0 9.2 6.8 3271.2 0.99 1.81 201000 306 0.77 0.42 F 324.4

JZ [15] Shear centre 149.0 147.0 8.8 6.7 3265.5 0.99 1.80 201000 306 - - FT 268.6
Midpoint
MZ [15] between centroid 148.0 151.0 9.0 6.7 3269.9 1.02 1.95 201000 306 - - FT 330.4
and shear centre
3
4 Note: F = flexural buckling mode, FT = flexural-torsional buckling mode

15 
 
 5 Fig. 11 displays the comparison between test and numerical results with various local and global

6 geometrical imperfection types. Twenty-seven sets of imperfection types were investigated, including

7 (i) three local imperfection half-wavelengths, i.e., aL,1 = (aw + af) / 2, aL,2 = af and aL,3 = aw, (ii) three

8 local imperfection amplitudes a / 100, a / 200 and a / 400, and (iii) three global imperfection

9 amplitudes L / 1000, L / 1500 and L / 3000. Note that a = aw for web plates and a = af for flange plates.

10 In Fig. 11, the vertical axes report normalized Nu,FE / Nu,exp. ratios of numerical-to-experimental

11 ultimate capacities. When Nu,FE / Nu,exp. < 1.0, the numerical predictions are on the safe side and vice-

12 versa. Clear tendencies can be observed from Fig. 11: as expected, lower resistances are obtained when

13 applying higher local or global imperfection amplitudes. More specifically and as expected, short to

14 middle length members are more sensitive to local imperfections, yet the latter become almost

15 negligible when it comes to long columns ZZ, JZ and MZ. Likewise, only middle length and longer

16 members are obviously affected by global imperfection. Especially for middle length columns T3 to

17 T6, which are extremely sensitive to imperfections, very detrimental results with Nu,FE / Nu,exp. ≈ 0.6

18 are reported when using the combination of a / 100 & L / 1000. In comparison, a shift in half-

19 wavelength aL is less influential than a change in imperfection amplitude – less than 10% difference

20 is reported. Note that according to previous experimental measurements and numerical analyses on

21 open and hollow sections [25,62,65-68], the a / 200 value shall be seen as reasonable for local

22 imperfection amplitudes. However, the resistance of middle length columns T3 to T6 is unexpectedly

23 underestimated when using a / 200, which might be because (i) the residual stresses pattern introduced

24 in the FE models was slightly overconservative, and because (ii) T-section members are more prone

25 to experience flexural-torsional buckling when considering more severe local imperfection amplitudes

26 as a result of the similarity between plate local buckling and member torsional buckling of T-section

27 members. Therefore, since there is a lack of information on measured geometrical imperfection

28 amplitudes and measured residual stresses patterns, a / 400 values were adopted as local imperfection

29 amplitudes for T-section members. In conclusion, it is shown that when considering the half-

16 
 
30 wavelengths aL,1 = (aw + af) / 2, local imperfection amplitudes alocal = a / 400 and global imperfection

31 amplitudes aglobal = L / 1000, the most accurate and consistent numerical predictions are obtained – the

32 corresponding average Nu,FE / Nu,exp. ratio is 0.98 associated with a quite low Coefficient of Variation

33 (COV) equal to 0.05. These geometrical imperfections were therefore adopted for further parametric

34 studies.

1.4
a/100 & aL,1 a/200 & aL,1 a/400 & aL,1
1.3 a/100 & aL,2 a/200 & aL,2 a/400 & aL,2
1.2 Unsafe a/100 & aL,3 a/200 & aL,3 a/400 & aL,3
Nu, F.E. / Nu, Exp. [-]

1.1
1.0
0.9
Safe
0.8
0.7
0.6
0.5
0.4
T1 T2 T3 T4 T5 T6 ZZ JZ MZ
35
36 a)

1.4
a/100 & aL,1 a/200 & aL,1 a/400 & aL,1
1.3
a/100 & aL,2 a/200 & aL,2 a/400 & aL,2
1.2 Unsafe a/100 & aL,3 a/200 & aL,3 a/400 & aL,3
Nu, F.E. / Nu, Exp. [-]

1.1
1.0
0.9
Safe
0.8
0.7
0.6
0.5
0.4
T1 T2 T3 T4 T5 T6 ZZ JZ MZ
37
38 b)

17 
 
 1.4 a/100 & aL,1 a/200 & aL,1 a/400 & aL,1
1.3 a/100 & aL,2 a/200 & aL,2 a/400 & aL,2
1.2 Unsafe a/100 & aL,3 a/200 & aL,3 a/400 & aL,3

Nu, F.E. / Nu, Exp. [-]

1.1
1.0
0.9
Safe
0.8
0.7
0.6
0.5
0.4
T1 T2 T3 T4 T5 T6 ZZ JZ MZ
39
40 c)
41 Fig. 11. Comparison of test and FE results with various local and global geometrical imperfection
42 types – a) aglobal = L/1000 – b) aglobal = L/1500 – c) aglobal = L/3000.

43 For such imperfection sets, comparison between experimental and FE load-displacement curves for

44 specimen ZZ are further displayed in Fig. 12. Good agreement levels are observed with respect to

45 initial stiffness, ultimate resistance and general load-displacement history. Note that the numerical

46 results exhibit a loss in stiffness after reaching 140 kN, attributed to a likely more severe residual

47 stresses pattern introduced in the shell models. Fig. 12 also reports ultimate resistances predicted by

48 Eurocode 3, the Australian Standards and the American Specifications. The latter provides an unsafe

49 prediction (10% higher than Nu,exp.), while over-conservative results are obtained from the other two

50 design approaches (around 20% lower than Nu,exp.). More discussions about the performances of

51 current design approaches are provided in Section 4. Besides, as shown in Fig. 13, the numerical failure

52 mode closely matches with the experimental one for specimen JZ. Therefore, it is concluded that the

53 developed numerical models are capable of providing accurate predictions for the buckling resistance

54 of T-section members and are eligible to perform further parametric studies.

18 
 
55
56 Fig. 12. Experimental and FE load-displacement curves for specimen ZZ.

57
58 Fig. 13. Comparison of test and numerical failure modes for specimen JZ.

59 2.3 Parametric studies

60 Based on the validated numerical models, extensive parametric studies were carried out to analyze the

61 possible effects of steel grade, section dimension and section/member slenderness on the cross-section

62 and member resistance of T-section columns. Three steel grades, S235, S355 and S460 with nominal

63 yield strengths fy = 235 MPa, 355 MPa, 460 MPa were considered in parametric studies for both short

19 
 
64 columns – in which local buckling prevails, and for long members – in which flexural or flexural-

65 torsional buckling is dominant.

66 A total of 138 various geometries of T-section short columns, which were assumed to be obtained

67 through cutting standardized hot-rolled I-sections [69,70] into half, were considered for studying the

68 cross-sectional local behavior. The height h of T-section short columns was varied between 50 and

69 559 mm, while width b spanned from 55 to 424 mm, web thickness tw from 4.1 to 45.5 mm and flange

70 thickness tf from 5.2 to 82 mm. Therefore, the section aspect ratio h / b of these T-section geometries

71 ranges from 0.5 to 1.7, the web plate slenderness h / tw ranges from 10.0 to 31.5 and flange plate

72 slenderness b / (2 tf) varies from 2.5 to 10.8. In order to exclude any possible influence of global

73 buckling, the length of short columns L was set as three half-waves, i.e., L = 3 ꞏ (af + aw) / 2, to ensure

74 that (i) the weakest section remains in the middle and (ii) the length is small enough to neglect flexural

75 buckling but long enough to ignore the influence of boundary conditions on ultimate capacities [49].

76 Besides, the lateral displacements uy of shear centers (web-to-flange nodes) are constrained all along

77 the member in order to prevent minor-axis global buckling.

78 For analyzing the resistance of T-section columns as influenced by local-global coupled instabilities,

79 27 different sections were selected from the previous 138 geometries, in which h was varied between

80 60 and 495 mm, while b spanned from 64 to 400 mm, and for each section geometry, six relative

81 member lengths (or member slenderness ̅λG) were considered, leading to the global-to-local critical

82 load ratio Ncr,G / Ncr,L spanning from 0.02 to 3.18. Accordingly, 414 computations have been carried

83 out to obtain ultimate cross-sectional resistance characterized by the local reduction factor χL only,

84 while 486 numerical results were obtained for the ultimate load of columns characterized by their

85 global reduction factor through χL+G.

86 In order to isolate pure global buckling behavior in which the influence of local buckling is ignored,

87 the same number of FE models, i.e., 486, was considered to obtain χG reduction factors. In order to

20 
 
 88 prevent any possible plate buckling, the vertical displacements uz of flange nodes were restrained to

89 the middle of flange plates and the horizontal displacements uy of web nodes were constrained to the

90 centroid point as illustrated in Fig. 14a. The same constraints were adopted previously in [10] for I-

91 section members. Besides, since it is difficult to distinguish between local and torsional buckling

92 modes in T-section members, considering them both may lead to overconservative results [18-22].

93 Therefore, when attempting to isolating the pure flexural buckling resistance χG, torsional twist about

94 the shear center was prevented. Consequently, in these particular simulations, member global behavior

95 was only influenced by member flexural buckling as shown in Fig. 14b and Fig. 14c, and no local plate

96 flexural nor torsional buckling could occur.

97
98 a) b) c)
99 Fig. 14. FE models for member global resistance χG – a) Extra constrains in shell models – b) Member flexural buckling
100 about z-axis – c) Member flexural buckling about y-axis.

101 3 OIC-based design rules

102 3.1 Cross-section and member buckling curves

103 Based on the numerical results collected along the parametric studies described in Section 3, both local

104 and global buckling design curves following the OIC framework as presented in Fig. 6 could be

105 developed. Although the influence of many parameters such as the steel grade or the section’s

106 geometry are already “naturally” embedded in the OIC ̅λL – L format, further investigations are still

107 needed to identify key factors responsible for the observed scatter in the results; such factors may be

108 used to derive multiple, parameter-dependent buckling curves. Firstly, the influence of steel grade on

109 cross-section and member resistance is analyzed in Fig. 15, which are in an OIC λ̅ – χ format and where

21 
 
110 the horizontal axes represent the section local slenderness ̅λL or member global slenderness ̅λG, while

111 vertical axes refer to the local buckling reduction coefficient χL or the global buckling reduction

112 coefficient χG. The solid reference line χ = 1.0 represents the attainment of plastic resistance and the

113 dashed curve χ = 1 / ̅λ2 represents the elastic plate buckling or elastic column buckling limits.

114 For cross-section resistance (Fig. 15a), increasing cross-sectional aspect ratio h / b or Iy / Iz usually

115 leads to higher section slenderness and thus lower relative resistance levels, since (i) an increase in h

116 may lead to a more slender web plate and thus to a more slender section, (ii) the flange with shorter

117 width is less likely to provide enough support to the web plate and prevent it from local buckling and

118 (iii) according to the EWM, for a T-section column with a slender web plate, the effective width of the

119 web plate is reduced, so the centroidal axis of the effective cross-section shifts upward, resulting in an

120 extra major-axis bending moment ∆M = N ꞏ eN, where the eN is the distance between the centroid of

121 the effective area and the centroid of the gross area. Additionally, the most slender T-shaped cross-

122 sections (see green ellipse in Fig. 15a) can be shown to exhibit post-buckling effects which leads to

123 higher L factors than the stability “reference curve” in 1 / ̅λL2. In general, for a given ̅λL, the values

124 of L are very close to each other so that applying only one single buckling curve seems sufficient, i.e.,

125 the influence of steel grade has been sufficiently accounted for through the OIC approach.

126
127 a) b)
128 Fig. 15. Influence of steel grade on ultimate capacity – a) Cross-section resistance – b) Member resistance.

22 
 
129 For member resistance, as expected, the member response of long columns ( ̅λG > 1.5) is more

130 controlled by global instability and less influenced by geometrical imperfection and residual stresses.

131 For short to intermediate length columns, where the effects of local buckling are more prevalent, a

132 limited scatter (see orange ellipse in Fig. 15b) is observed and can be shown associated with the various

133 section dimensions being considered in the current study. In particular, with a decrease in Iy / Iz, the

134 minor-axis switches from being the z-axis to the y-axis. Therefore, a slightly lower resistance is

135 achieved in such situations as torsional buckling modes are excluded, since more compressive stresses

136 are applied in the flange and upper web plate due to the development of additional bending moments

137 ∆M. Especially for members with intermediate global slenderness (0.7 < ̅λG < 1.0), which are more

138 affected by imperfections and residual stresses, around 10% differences among the different steel

139 grades considered are observed. As for long members, where the effects of residual stresses and local

140 imperfections become negligible, there is almost no difference between results from different steel

141 grades. Therefore, although slightly higher nominal resistance χL is observed for middle-length

142 members with S460 steel grade, a single buckling curve for each case is adopted for simplicity.

143 The proposed OIC design equations for both local buckling and member buckling curves are

144 summarized in Table 2. Note that the OIC approach extends the Ayrton-Perry format [71,72] well-

145 known for member buckling to cross-section resistance predictions. In this format, λ̅ 0 captures the

146 length of the “plateau” where χL or χG = 1.0, αL and αG take account of the influence of local and global

147 imperfections, respectively, and δ considers the benefits of post-buckling effects on cross-section

148 resistance. As for global buckling, note that the OIC global buckling curve is the same as recommended

149 by Eurocode 3 (buckling curve b), which is lower than the buckling curve proposed in the American

150 Specifications, but higher than European buckling curve c (see Fig. 5). The corresponding buckling

151 curves are compared with the reference numerical results in Fig. 15, revealing that both cross-section

152 and member buckling cases are represented accurately.

23 
 
153 Table 2. Design procedure and key parameters for cross-section and member buckling curves.

Buckling resistance Local buckling Global buckling

1 1
L  1 G  1
Reduction coefficient L    L
2
L

G  G2  G2
(“buckling
coefficient”) Where L  0.5   1           Where G  0.5   1   G  G  0   G2 
 L L 0 L  

Key parameters 0  0.3

0  0.2
(Ayrton-Perry  L  0.05
format) G  0.34
  1.5
154

155 3.2 Local-global coupling factor fL/G

156 The local-global coupled instability design of T-section members was accounted for by resorting to

157 the coupling factor fL/G that was introduced in Section 1 and Fig. 6. After obtaining the cross-section

158 and member buckling reduction factors L and G separately (i.e., without any local-global coupling),

159 a coupling factor fL/G shall be considered in the final step to account for the interaction between local

160 and member buckling (𝜒 ). The relationship between fL/G and λ̅ G is illustrated in Fig. 16 – note that

161 fL/G = χL+G / (L ꞏ G). It is shown that the impact of local/global interactions changes with global

162 slenderness λ̅ G, since the column relative resistance χL+G is mostly influenced by local buckling when

163 it comes to short members, i.e., χL+G ≈ χL, but the effect of local buckling diminishes for long members

164 and only global buckling shall be taken into account, i.e., χL+G ≈ χG. Therefore, for short columns

165 whose χG ≈ 1.0 and χL+G ≈ χL, fL/G is close to unity; yet, for some cases where the local/global

166 interaction is very pronounced, fL/G can be slightly lower than unity (see blue ellipse in Fig. 16). As for

167 long columns, global buckling is dominant, which leads to fL/G ≈ 1/ χL in order to “eliminate” the

168 influence of local buckling. For columns with intermediate global slenderness, both local and global

169 buckling should be considered, resulting in the following repationship: 1.0 < fL/G <1 / χL. More

170 information about fL/G has been provided in [10].

24 
 
171
172 Fig. 16. Local-global coupling factor fL/G.

173 In accordance with the above, a trilinear design formulation was proposed for fL/G as presented by

174 Eq. (6). The suggested fL/G design formula was compared to numerical results as shown in

175 Fig. 16 – note that only three upper bounds of fL/G design curves are represented. As a particular point,

176 the end point of the second segment of the proposed design equations, i.e., for intermediate ̅λG values,

177 is seen increasing with a decrease in χL, since the OIC adopts flexural column buckling curves and

178 only considers the effects of torsional buckling within a local buckling curve. Indeed, as discussed in

179 Section 3.1, more slender T-shaped cross-section usually possess higher h / b or Iy / Iz ratios and the

180 associated T-section members are more prone to experience flexural torsional buckling, so that the

181 corresponding second segments shall be relatively longer, in order to fully account for the detrimental

182 effects of torsional buckling. Note that for the sections with the lower χL values (black points in

183 Fig. 16), the member slenderness ̅λG never exceeds 1.5, since extreme cases, such as T-section columns

184 with extremely slender flange plates, were not considered in this study. Overall, the proposed fL/G

185 expressions are applicable to all hot-rolled cut T-section columns studied in this paper. The general

186 performance of this OIC-based design approach for T-shaped cross-sections and members is assessed

187 in Section 4.

25 
 
 1 G  0.5


   0.5
188 fL G  G
 1.4  L  0.5
1  L  1  1 0.5  G  1.4  L (6)

1  G  1.4  L
 L

189 4 Assessment of the OIC-based design approach

190 Detailed comparisons between the numerical results and four analytical predictions involving the

191 American Specifications, Eurocode 3, the Australian Standards and the proposed OIC-based design

192 approach for T-shaped cross-sections and members are presented in Figs. 17 to 21. Among those

193 figures, Figs. 17 to 19 present the results for T-shaped cross-sections with various steel grades while

194 Figs. 20 to 21 divide the results as a function of the Iy / Iz ratio that serves as a way to separate the most

195 slender section shapes – large Iy / Iz ratios are associated with the most sender sections. The

196 χL,Ref. / χL,FE and χL+G,Ref. / χL+G,FE ratios refer to the ultimate resistance predicted by any of the four

197 reference design rules over the FE results. If the ratio is larger than one, the design predictions are on

198 the unsafe side, and vice-versa. “a figures” in Figs. 17 to 19 plot χL,Ref. / χL,FE or χL+G,Ref. / χL+G,FE ratios

199 as a function of cross-sectional slenderness L or member slenderness G. “b figures” in Figs. 17 to 19

200 indicates relevant frequency histograms. In addition, Table 3 provides some statistical results related

201 to χL,Ref. / χL,FE and χL+G,Ref. / χL+G,FE ratios, which include mean values, COV, maximum and minimum

202 values, and the percentage of corresponding predictions greater than 3%, 10% and 20% on the

203 unconservative side. These data quantify the proportion of slightly unconservative and significantly

204 unsafe predictions, which might or might not be compensated by usual values of safety factors.

205 4.1 Cross-section resistance

206 As shown in Figs. 17 to 19 and Table 3, the performances of both the American Specifications and the

207 OIC-based approach for T-shaped cross-sections under axial compression are hardly affected by the

208 steel grade. For the OIC-based approach, the mean value of the χL,OIC / χL,FE ratio is up to 0.99 with an

209 excellent C.O.V of 0.02 and a maximum value remaining within 4% on the unconservative side. In

26 
 
210 contrast, Eurocode 3 provides much more conservative predictions, especially for slender cross-

211 sections with ̅λL larger than 0.7. Since sections become more slender with an increase in yield stress,

212 many more low values of χL, EC 3 / χL,FE are obtained for cross-sections with fy = 355 MPa and 460 MPa.

213 For all three steel grades, the mean value of the χL,EC 3 / χL,FE ratio is 0.86 with a relatively large COV

214 value of 0.13 and a minimum value is as low as 0.68, indicating an important level of over-

215 conservatism when it comes to more slender sections. This shall primarily be associated with

216 Eurocode 3 considering the extra bending moment resulting from the shift of the centroid of the gross

217 section area A to the centroid of the effective section area Aeff for slender cross-sections, requiring

218 combined loading design checks and therefore leading to quite conservative predictions [24,26].

219 Oppositely, the Australian Standards exhibit more unsafe predictions for sections with higher yield

220 strengths. As shown in Fig. 19, more than 20% of the predictions from the Australian Standards are on

221 the unsafe side, indicating that its design formulae are less applicable to the design of T-shaped cross-

222 sections with fy = 460 MPa. As for sections with fy = 235 MPa (see Fig. 17), its predictions are overall

223 relatively more conservative than the American Specifications and the OIC proposal but more accurate

224 than Eurocode 3 for sections with ̅λL between 0.7 to 1.0. Yet, overall, the OIC-based approach is seen

225 to provide accurate predictions for T-shaped cross-sections with various yield strengths.

226
227 a) b)
228 Fig. 17. Design rules vs. FE results for T-shaped cross-sections in compression with fy = 235 MPa – a) Accuracy of
229 resistance predictions as a function of ̅λL – b) Frequency distributions.

27 
 
230
231 a) b)
232 Fig. 18. Design rules vs. FE results for T-shaped cross-sections in compression with fy = 355 MPa – a) Accuracy of
233 resistance predictions as a function of ̅λL – b) Frequency distributions.

234
235 a) b)
236 Fig. 19. Design rules vs. FE results for T-shaped cross-sections in compression with fy = 460 MPa – a) Accuracy of
237 resistance predictions as a function of ̅λL – b) Frequency distributions.

238

28 
 
239 Table 3. Statistical results for χL,Ref. / χL,FE and χL+G,Ref. / χL+G,FE ratios.

Number of cases >1.03 >1.10 >1.20

Proposal Mean COV Max. Min.
considered [%] [%] [%]
OIC 0.99 0.02 1.04 0.91 2.2 0.0 0.0
EC 3 0.86 0.13 1.03 0.68 0.2 0.0 0.0
Steel grades 414
AISC 0.94 0.03 1.08 0.86 1.7 0.0 0.0
AS 1.02 0.09 1.48 0.89 18.8 10.9 5.8
OIC 1.00 0.01 1.04 0.95 2.9 0.0 0.0
EC 3 0.89 0.13 1.02 0.69 0.0 0.0 0.0
fy =235 MPa 138
AISC 0.95 0.01 0.98 0.91 0.0 0.0 0.0
Cross- AS 0.88 0.06 1.00 0.79 0.0 0.0 0.0
sections OIC 0.99 0.02 1.03 0.94 1.4 0.0 0.0
EC 3 0.85 0.13 1.02 0.68 0.0 0.0 0.0
fy = 355 MPa 138
AISC 0.94 0.03 1.03 0.88 0.0 0.0 0.0
AS 0.90 0.08 1.11 0.77 7.2 0.7 0.0
OIC 0.98 0.02 1.03 0.91 2.2 0.0 0.0
EC 3 0.83 0.12 1.03 0.68 0.7 0.0 0.0
fy = 460 MPa 138
AISC 0.94 0.05 1.08 0.86 5.1 0.0 0.0
AS 0.96 0.12 1.34 0.80 19.6 13.0 8.7
OIC 0.96 0.06 1.06 0.78 10.5 0.0 0.0
EC 3 0.81 0.15 1.00 0.49 0.0 0.0 0.0
All cases 486
AISC 1.03 0.07 1.21 0.85 48.6 18.2 1.8
AS 0.89 0.11 1.29 0.69 4.5 1.6 0.6
OIC 0.99 0.03 1.04 0.91 16.4 0.0 0.0
EC 3 0.90 0.06 1.00 0.73 0.0 0.0 0.0
Members Iy / Iz < 1.0 234
AISC 1.06 0.07 1.21 0.88 68.9 32.4 3.7
AS 0.84 0.07 1.01 0.74 0.0 0.0 0.0
OIC 0.94 0.06 1.06 0.78 4.8 0.0 0.0
EC 3 0.73 0.15 0.96 0.49 0.0 0.0 0.0
Iy / Iz ≥ 1.0 252
AISC 1.00 0.05 1.19 0.85 28.8 4.4 0.0
AS 0.79 0.12 1.17 0.64 2.0 0.8 0.0
240

241 4.2 Member resistance

242 Figs. 20 to 21 and Table 3 present the results for all T-section columns considered in the current study.

243 In general, more scattered outcomes are observed since the inaccuracies reported for the design of

244 cross-sections worsen when it comes to members, as a consequence of the additional presence of

245 column buckling. As shown in Fig. 20, for sections with Iy / Iz < 1.0, the Australian Standards provide

246 the most conservative resistance predictions – the mean value of the χL+G,AS / χL+G,FE ratio is 0.84 with

247 a minimum value as low as 0.74, as a consequence of a lower column buckling curve being proposed

248 by the Australian Standards (see Fig. 5). In contrast, for the American Specifications, which suggest

249 the most favorable buckling curve, around 70% of the χL+G,AISC / χL+G,FE ratios lie on the unsafe side,

250 indicating an overall inaccurate design approach. For Eurocode 3, many conservative predictions are

251 provided for short columns (see blue ellipse in Fig. 20) and that conservatism gradually reduces with

29 
 
252 an increase in global slenderness ̅λG. Eurocode 3 indeed provides over-conservative predictions for

253 cross-section resistance and that conservatism can be compensated to some extent for long members

254 by considering a global slenderness λ̅ L+G, modified to account for local buckling through an effective

255 area Aeff. (see Eq. (7)).

256 L+G  f y Aeff N cr,G (7)

257 As shown in Fig. 21, for T-section columns with Iy / Iz ≥ 1.0 which are more prone to flexural-torsional

258 buckling, Eurocode 3 offers the most conservative and scattered predictions. The mean value of the

259 χL+G,EC 3 / χL+G,FE ratio is 0.73 with the highest COV value (0.15) and the minimum value is as low as

260 0.49 – i.e., the ultimate capacity of T-section columns is reduced by 50%, compared to its “true” level,

261 mainly because the global critical buckling load Ncr,G in Eq. (7) is taken as the flexural-torsional

262 buckling load Ncr,FT. As mentioned before, torsional buckling effects in T-section members are

263 considered redundantly and it can be shown that this results in quite conservative and scattered

264 resistance predictions.

265 In comparison, the American Specifications provide many unsafe predictions for members with ̅λG

266 comprised between 1.0 and 1.5, mostly because of a seemingly too favorable buckling curve. The

267 maximum value of χL+G,AISC / χL+G,FE is around 1.2 with more than 4.4% of the results above the 10%

268 safety limit. The Australian Standards, which only consider torsional buckling at the cross-sectional

269 level, exhibit an overall better performance than Eurocode 3 but there are still 0.8% of the results

270 remaining over the 10% unsafe side limit, mostly due to the Australian Standards providing inaccurate

271 predictions for short columns with fy = 460 MPa.

272 Compared to these code design approaches, the OIC proposal exhibits better accuracy and higher

273 consistency for both T-section columns with Iy / Iz < 1.0 and Iy / Iz ≥ 1.0. The mean value of

274 χL+G,OIC / χL+G,FE ratios for all cases remain at 0.96, with a COV as low as 0.06.

275 Further to these purely numerical analyses, a comparison between seven test results and the four design

276 approaches predictions for the seven T-section columns under axial compression is summarized in

30 
 
277 Table 4, where Nu,Ref. = χL+G,Ref. ꞏ Npl. Similarly, Eurocode 3 and the Australian Standards provide

278 relatively conservative and scattered predictions, especially for middle length members T3 to T6 – the

279 mean value of Nu,EC 3 / Nu,exp. and Nu,AS / Nu,exp. ratios are 0.78 and 0.74, respectively, while the American

280 Specifications provide an unsafe result for specimen ZZ with Nu,AISC / Nu,exp. = 1.10. In contrast, the

281 mean value of Nu,OIC / Nu,exp. is 0.94 with a COV as low as 0.05, indicating that the OIC yields

282 significantly more accurate results than the other three proposals. Therefore, owing to the various

283 member geometries considered and the complex local-global coupled instabilities analyzed, the OIC

284 proposal can be seen as an appropriate design approach in terms of accuracy, consistency and safety.

285
286 a) b)
287 Fig. 20. Design rules vs. FE results for T-section columns in compression with Iy / Iz < 1.0 – a) Accuracy of resistance
288 predictions as a function of ̅λG – b) Frequency distributions.

289
290 a) b)
291 Fig. 21. Design rules vs. FE results for T-section columns in compression with Iy / Iz ≥ 1.0 – a) Accuracy of resistance
292 predictions as a function of ̅λG – b) Frequency distributions.

31 
 
293 Table 4. Comparison of test results and resistance predictions from different proposals.

Nu,OIC / Nu,exp. Nu,EC 3 / Nu,exp. Nu,AISC / Nu,exp. Nu,AS / Nu,exp.

Specimens
[–] [–] [–] [–]
T1 0.96 0.91 0.97 0.82
T2 0.92 0.86 0.93 0.77
T3 0.95 0.78 1.01 0.71
T4 0.85 0.69 0.93 0.63
T5 0.98 0.68 0.98 0.75
T6 0.95 0.67 0.97 0.72
ZZ 0.95 0.84 1.10 0.77
Mean 0.94 0.78 0.99 0.74
COV 0.05 0.12 0.06 0.08
Min. 0.85 0.67 0.93 0.63
Max. 0.98 0.91 1.10 0.82
294

295 4.3 Reliability analysis

296 Within the Eurocode framework, Annex D of EN 1990 [44] introduces an evaluation method to obtain

297 the design model partial safety factor γM associated with a given design proposal (see Fig. 6), which

298 covers all uncertainties derived from materials, section/member geometries and design model

299 approximations or inaccuracies. According to the provisions of EN 1990 and a simplified approach

300 suggested in [73], a statistical analysis was carried out to investigate the reliability of the American

301 Specifications, Eurocode 3, the Australian Standard, and the OIC proposals for the design of T-section

302 members.

303
304 Fig. 22. Example of a tail approximation for T-section members.

305 Note that a large proportion of over-conservative results could result in unduly high γM factors.

306 Therefore, the Tail Approximation (T.A.) technique [65,73] has been applied to select more

32 
 
307 representative subsets for the different proposals. An example of a such a T.A. is illustrated in Fig. 22,

308 together with a linear regression line for the worst 124 cases proposed by Eurocode 3. In this figure,

309 the horizontal axis Nu,L+G,FE(exp.) / Nu,L+G,EC 3 represents the ultimate numerical and experimental results

310 over the ultimate resistances predicted by Eurocode 3, and the vertical axis is normalized to “standard

311 distribution”, meaning that all the results are assumed to be scattered with a normal distribution. It is

312 shown that when all datasets are considered, their general distribution does not align with a normal

313 distribution, which opposes to the assumption of the statistical procedure in Annex E of EN 1993-1-1

314 [28], and therefore leads to inaccurate and overconservative γM0 results. Therefore, in resorting to the

315 tail approximation, the error derived from over-safe results can be corrected by approximating the

316 dataset through a fitted normal distribution about the “tail end”.

317 Reliability analysis results and relevant key statistical parameters are presented in Table 5, in which n

318 is the number of numerical and experimental results in each subset that have been selected through the

319 tail approximation, kd,n is the design fractile factor, b is the least squares estimator of the regression

320 slope, Vδ is the COV of the numerical and experimental resistances relative to the design modes, and

321 Vr, which stands as the combined COV considering the uncertainties derived from both the design

322 model and basic variables, can be calculated through Eq. (8) and (9) [73], where 𝑉mat is the COV of

323 material strength and 𝑉geom is the COV of geometric properties. The value of 𝑉mat was selected based

324 on Annex E of EN 1993-1-1 [28] and 𝑉geom was taken as 0.03, following the recommendations in [74].

325 Vr2  Vδ2  Vrt2 (8)

326 Vrt2  Vmat

2
 Vgeom
2
(9)

327 Overall, based on the results reported in Table 5, it is shown that for cross-section resistance, γM values

328 associated with the American Specifications, Eurocode 3 and the OIC proposals were found to be

329 either 1.0 or very close to unity, while the partial safety factor γM for the Australian Standards is larger

330 than 1.20, further confirming that the latter provide inaccurate and also unsafe predictions for sections

33 
 
331 with fy = 460 MPa. As for member resistance, the γM factor corresponding to Eurocode 3 design rules

332 is lower than unity, indicating that the γM value suggested in EN 1993-1-1 is satisfactory. However,

333 the γM value for the American Specifications is larger than 1.10, which could lead to many unduly

334 penalizing cases where a much lower safety level is needed. In general, the OIC, which provides

335 reasonable γM values compared to the other three proposals, can be deemed as a reliable and accurate

336 design approach.

337 Table 5. Reliability analysis results after applying tail approximations (T.A.).

Design cases Proposals n kd,n b Vδ Vr γM

OIC 92 3.199 0.986 0.011 0.056 1.01
EC 3 83 3.212 0.993 0.006 0.056 1.00
Cross-sections
AISC 66 3.245 0.996 0.029 0.060 1.03
AS 57 3.271 0.933 0.081 0.096 1.23
OIC 120 3.173 0.965 0.015 0.060 1.04
EC 3 124 3.171 1.065 0.025 0.062 0.95
Members
AISC 132 3.166 0.898 0.031 0.063 1.14
AS 82 3.213 1.048 0.051 0.075 1.01
338

339 5 Conclusions

340 The results of extensive numerical analyses towards the local and global buckling behaviors of T-

341 section columns with slender web plates have been presented in this paper. The developed non-linear

342 shell FE models were validated against existing experimental tests and consecutively used in a series

343 of parametric studies carried out to analyze the effects of steel grade, section dimension and

344 section/member slenderness on cross-section and member resistance. Based on the numerical results,

345 an OIC-based design approach was proposed for T-section columns, which included (i) design at the

346 cross-section level through a local buckling curve, (ii) an original flexural buckling curve and (iii) a

347 local-global coupling factor fL/G to capture the interaction between local buckling and member buckling

348 of T-section members. The accuracy of this OIC approach and of other codified design methods was

349 assessed against the reference FE results. It was observed that Eurocode 3 provided the most

350 conservative predictions for both T-shaped cross-sections and members; resistance predictions based

351 on the American Specifications were found to be unsafe and scattered, especially for members with

34 
 
352 intermediate global slenderness. As for the Australian Standard, inaccurate predictions for columns

353 with fy = 460 MPa were observed at both local and global levels. In comparison, the OIC-based

354 proposal provides better accuracy and higher consistency than the other recommendations. Eventually,

355 reliability analyses based on EN 1990 proved that the OIC proposed approach could be safely and

356 suitably used for the design of T-shaped cross-sections and members.

357 References

358 [1]Van der Neut, A. "The interaction of local buckling and column failure of thin-walled compression

359 members". In Applied Mechanics, pp. 389-399, Springer, Berlin, Heidelberg, 1969.

360 [2]Sridharan, S. and Ali, M.A. "Interactive buckling in thin-walled beam-columns". Journal of

361 engineering mechanics, Vol. 111, No. 12, pp.1470-1486, 1985.

362 [3]Kolakowski, Z. "A semi-analytical method for the analysis of the interactive buckling of thin-walled

363 elastic structures in the second order approximation". International journal of solids and structures,

364 Vol. 33, No. 25, pp.3779-3790, 1996.

365 [4]Dubina, D. and Ungureanu, V. "Effect of imperfections on numerical simulation of instability

366 behaviour of cold-formed steel members". Thin-walled structures, Vol. 40, No. 3, pp.239-262,

367 2002.

368 [5]Becque, J. "Local-overall interaction buckling of inelastic columns: A numerical study of the

369 inelastic Van der Neut column". Thin-walled structures, Vol. 81, pp.101-107, 2014.

370 [6]Bai, L., Wang, F., Wadee, M.A. and Yang, J. "Nonlinear mode interaction in equal-leg angle struts

371 susceptible to cellular buckling". Proceedings of The Royal Society A: Mathematical, Physical and

372 Engineering Sciences, Vol. 473, No. 2207, pp.20170583, 2017.

373 [7]Shen, J. and Wadee, M.A. "Length effects on interactive buckling in thin-walled rectangular hollow

374 section struts". Thin-Walled Structures, Vol. 128, pp.152-170, 2018

375 [8]Shen, J. and Wadee, M.A. "Imperfection sensitivity of thin-walled rectangular hollow section struts

376 susceptible to interactive buckling". International Journal of Non-Linear Mechanics, Vol. 99,

35 
 
377 pp.112-130, 2018.

378 [9]Bai, L., Yang, J. and Wadee, M.A. "Cellular buckling from nonlinear mode interaction in unequal-

379 leg angle struts". Thin-Walled Structures, Vol. 132, pp.316-331, 2018.

380 [10]Li, L. and Boissonnade, N. “Local/global coupled instabilities of slender I-sections under

381 compression”. Thin-Walled Structures, Vol. 172, pp.108842, 2022.

382 [11]Naito, C.J., Hendricks, R., Sause, R. and Cercone, C. "Composite steel tee concrete deck bridge

383 system: Design, fabrication, and full-scale verification". Journal of Bridge Engineering, Vol. 26,

384 no.1, pp.04020109, 2021.

385 [12]Way, A.G., Cosgrove, T.C. and Brettle, M.E. "Precast concrete floors in steel framed buildings".

386 Steel Construction Institute, 2007.

387 [13]Kennedy, J.B. and Murty, M.K. "Buckling of steel angle and tee struts". Journal of the Structural

388 Division, Vol. 98, no. 11, pp. 2507-2522,1972.

389 [14]Kitipornchai, S. and Lee, H.W. "Inelastic experiments on angle and tee struts". Journal of

390 Constructional Steel Research, Vol. 6, no. 3, pp. 219-236, 1986.

391 [15]Luo, G. " Study on overall stability behavior of large slenderness ratio hot-rolled cut T-section

392 struts", Master thesis, Xi'an University of Architecture and Technology, 2012. (In Chinese)

393 [16]Chen, S. "Buckling of T-strut subject to compressive force on its shear center". Journal of

394 Constructional Steel Research, Vol. 63, no. 3, pp.332-336, 2007

395 [17]Chen, S. "Interactive local buckling and limiting slenderness of T-strut stem". Journal of

396 Constructional Steel Research, Vol. 63, no. 6, pp.833-837, 2007.

397 [18]Trahair, N.S. "Post-buckling strength of steel tee columns". Engineering structures, Vol. 56,

398 pp.1800-1807, 2013.

399 [19]Taras, A., Kugler, P. and Unterweger, H. "On the behaviour and Eurocode design of T-section

400 columns, beams and beam-columns with slender webs". Journal of Constructional Steel Research,

401 Vol. 129, pp.250-262, 2017.

36 
 
402 [20]Cardoso, F.S. and Rasmussen, K.J. "Behavior and design of concentrically loaded T-section steel

403 columns". Journal of Structural Engineering, Vol. 140, no. 7, pp.04014039, 2014.

404 [21]Cardoso, F.S. and Rasmussen, K.J. "The behaviour and design of concentrically loaded T-section

405 steel columns", RESEARCH REPORT R933, 2013.

406 [22]Dinis, P.B., Camotim, D. and Silvestre, N. "On the local and global buckling behaviour of angle,

407 T-section and cruciform thin-walled members". Thin-Walled Structures, Vol. 48, pp.786-797,

408 2010.

409 [23]Bebiano, R., P. Pina, N. Silvestre, and D. Camotim. "GBTUL–Buckling and Vibration Analysis

410 of Thin-Walled Members (software)." DECivil/IST, University of Lisbon (http://www. civil. ist.

411 utl. pt/gbt), 2008.

412 [24]Chen, Y., Cheng, X. and Nethercot, D.A. “An overview study on cross-section classification of

413 steel H-sections”. Journal of Constructional Steel Research, Vol. 80, pp. 386-393, 2013.

414 [25]Nseir, J. “Development of a new design method for the cross-section capacity of steel hollow

415 sections”. PhD thesis, University of Applied Sciences of Western Switzerland - Fribourg,

416 University of Liège, Saint-Joseph University Beirut, 2015.

417 [26]Li, L., Gérard, L., Kettler, M. and Boissonnade, N. “The O.I.C. for the design of hot-rolled and

418 welded I-sections under combined loading”. Thin-Walled Structures, Vol. 172, pp.108623, 2022.

419 [27]Boissonnade, N., Hayeck, M., Saloumi, E. and Nseir, J. “An Overall Interaction Concept for an

420 alternative approach to steel members design”. Journal of Constructional Steel Research, Vol. 135,

421 pp. 199-212, 2017.

422 [28]EN 1993-1-1, Eurocode 3: Design of steel Structures – Part 1–1: General rules and rules for

423 Buildings, European Committee for Standardization, 2020.

424 [29]EN 1993-1-5, Eurocode 3: Design of Steel Structures, Part 1-5: Plated structural elements,

425 European Committee for Standardisation, 2020.

426 [30]American Institute of Steel Construction, ANSI/AISC 360-10, An American National Standard,

37 
 
427 Specification for Structural Steel Buildings, Chicago: American Institute of Steel Construction,

428 2016.

429 [31]AS 4100-2020 steel structures, Sydney: Standards Australia, 2020.

430 [32]Dubina, D., 2001. The ECBL approach for interactive buckling of thin-walled steel members. Steel

431 and Composite Structures, An International Journal, 1(1), pp.75-96.

432 [33]Dubina, D. and Ungureanu, V., 2014. Instability mode interaction: From Van Der Neut model to

433 ECBL approach. Thin-Walled Structures, 81, pp.39-49.

434 [34]Shen, J. and Wadee, M.A. "Local–global mode interaction in thin-walled inelastic rectangular

435 hollow section struts part 2: Assessment of existing design guidance and new recommendations".

436 Thin-Walled Structures, Vol. 145, p.106184, 2019.

437 [35]Yuan, H.X., Wang, Y.Q., Gardner, L. and Shi, Y.J., 2014. Local–overall interactive buckling of

438 welded stainless steel box section compression members. Engineering Structures, 67, pp.62-76.

439 [36]Martins, A.D., Camotim, D. and Dinis, P.B. Distortional-global interaction in lipped channel and

440 zed-section beams: Strength, relevance and DSM design. Thin-Walled Structures, 129, pp.289-308,

441 2018.

442 [37]Timoshenko, S., 1970. Theory of elastic stability 2e. Tata McGraw-Hill Education.

443 [38]Hayeck, M. “Development of a new design method for steel hollow section members resistance”.

444 PhD thesis, University of Applied Sciences of Western Switzerland - Fribourg, University of Liège,

445 Saint-Joseph University Beirut, 2016.

446 [39]Li, L., Gérard, L., Hayeck, M. and Boissonnade, N. “Stability of Slender Section H.S.S. Columns”.

447 Proceedings of the Annual Stability Conference, Structural Stability Research Council, San

448 Antonio, Texas, 2012.

449 [40]Gérard, L., Li, L., Kettler, M. and Boissonnade, N. “Steel I-sections resistance under compression

450 or bending by the Overall Interaction Concept”. Journal of Constructional Steel Research, Vol.

451 182, pp. 106644, 2021.

38 
 
452 [41]Gagné, A.S., Gérard, L. and Boissonnade, N. “Design of stainless steel cross-sections for simple

453 load cases with the OIC”. Journal of Constructional Steel Research, Vol. 168, pp. 105936, 2020.

454 [42]Li, L., Gérard, L., Langlois, S. and Boissonnade, N. “The O.I.C. for the design of mono-symmetric

455 I-sections under simple loading cases”. Thin-Walled Structures (accepted)

456 [43]Beyer, A. “On the design of steel members with open cross-sections subjected to combined axial

457 force, bending and torsion”. PhD thesis, University of Lorraine, 2017.

458 [44]EN 1990. Basis of structural design. Brussels: European Committee for Standardization, 2002.

459 [45]Abaqus. Reference manual, version 6.13. Simulia, Dassault Systèmes, France, 2013.

460 [46]Kang, S.B., Yang, B., Zhang, Y., Elchalakani, M. and Xiong, G. "Global buckling of laterally-

461 unrestrained Q460GJ beams with singly symmetric I-sections". Journal of Constructional Steel

462 Research, 145, pp.341-351, 2018

463 [47]Zhao, O. and Gardner, L. "The continuous strength method for the design of mono-symmetric and

464 asymmetric stainless steel cross-sections in bending". Journal of Constructional Steel Research,

465 150, pp.141-152, 2018.

466 [48]Liang, Y., Jeyapragasam, V.V.K., Zhang, L. and Zhao, O. "Flexural-torsional buckling behaviour

467 of fixed-ended hot-rolled austenitic stainless steel equal-leg angle section columns". Journal of

468 Constructional Steel Research, 154, pp.43-54, 2019

469 [49]Gérard, L. “Contribution to the design of steel I and H-sections members by means of the Overall

470 Interaction Concept”. PhD thesis, Laval University, 2019.

471 [50]Yun, X. and Gardner, L. “Stress-strain curves for hot-rolled steels”. Journal of Constructional Steel

472 Research. Vol 133, pp. 36-46, 2017.

473 [51]Yun, X., Gardner, L. and Boissonnade, N. "The continuous strength method for the design of hot-

474 rolled steel cross-sections". Engineering Structures, Vol. 157, pp.179-191, 2018.

475 [52]Anbarasu, M. and Ashraf, M. "Behaviour and design of cold-formed lean duplex stainless steel

476 lipped channel columns". Thin-walled structures, Vol. 104, pp.106-115, 2016.

39 
 
477 [53]Management Bureau of Code for Design of Steel Structures. "Notice of uniform mode of residual

478 stress about compression member". Bejing: Management Bureau of Code for Design of Steel

479 Structures, 1983.

480 [54]Xiong, X.L., Jin, L.B. and Wang, H. "Design Capacity of T-Strut Subject to Compressive Force".

481 Advanced Materials Research, Vol. 163, pp. 550-556, 2011.

482 [55]Xiong, X.L. and Li, X.Y. "Discussion on the stability coefficient φb for flexural-torsional buckling

483 of hot-rolled cut-T section struts". Building Structure, Vol. 47, no. 21, 2017. (In Chinese)

484 [56]Xiong, X. and Pang, R. "Study on overall instability of hot-rolled cut-T section struts". Building

485 Structure, Vol. 43, no. 13, 2013. (In Chinese)

486 [57]Peng, X.T., Gu, Q. and Zhao Y.S. "Web Local Buckling and Width-thickness Limit of Cut-T Struts

487 Under Compression". Building Structure, Vol. 35, no. 2, 2005. (In Chinese)

488 [58]Smith, T.G. "The effect of initial imperfections on the strength of thin-walled box columns".

489 International Journal of Mechanical Sciences, Vol. 13, No. 11, pp.911-925, 1971.

490 [59]Hui, D. "Design of beneficial geometric imperfections for elastic collapse of thin-walled box

491 columns". International journal of mechanical sciences, Vol. 28, No. 3, pp.163-172, 1986.

492 [60]Shen, J. and Wadee, M.A. "Sensitivity of elastic thin-walled rectangular hollow section struts to

493 manufacturing tolerance level imperfections". Engineering Structures, Vol. 170, pp.146-166, 2018.

494 [61]Johansson, B., Maquoi, R., Sedlacek, G., Müller, C. and Beg, D. "Commentary and worked

495 examples to EN 1993-1-5 Plated structural elements", JRC scientific and technical reports, 2007.

496 [62]Gérard, L., Li, L., Kettler, M. and Boissonnade, N. “Recommendations on the geometrical

497 imperfections definition for the resistance of I-sections”. Journal of Constructional Steel Research,

498 Vol. 162, pp. 105716, 2019.

499 [63]Boissonnade, N. and Somja, H. “Influence of imperfections in FEM modeling of lateral torsional

500 buckling”. Proceedings of the Annual Stability Conference, Structural Stability Research Council,

501 Grapevine, Texas, 2012.

40 
 
502 [64]Couto, C. and Real, P. V. “Numerical investigation on the influence of imperfections in the local

503 buckling of thin-walled I-shaped sections”. Thin-Walled Structures, Vol. 135, pp. 89-108, 2019.

504 [65]Kettler M. “Elastic-Plastic Cross-Sectional Resistance of Semi-Compact H-and Hollow Sections”.

505 PhD Thesis, Graz University of Technology, 2008.

506 [66]Johansson B., Maquoi R., Sedlacek G., Müller C. and Beg D. “Commentary and worked examples

507 to EN 1993-1-5: Plated structural elements”. JRC scientific and technical reports, Oct 2007.

508 [67]Nseir, J., Hayeck, M., Saloumi, E. and Boissonnade, N. “Influence of imperfections on the local

509 buckling response of hollow structural shapes”. Structural Stability Research Council, Orlando,

510 Florida, April 12-15, 2016.

511 [68]Beg, D., Kuhlmann U., and Davaine L. “ECCS Eurocode Design Manual Design of Plated

512 Structures”. 2010.

513 [69]AISC. Steel construction manual. 15th ed. Chicago: AISC, 2017.

514 [70]Handbook of structural steelwork: Eurocode edition, The British Constructional Steelwork

515 Association Ltd and The Steel Construction Institute, 2013.

516 [71]Ayrton, W. E. and Perry, J. “On struts”. The Engineer, vol. 62, 1886.

517 [72]Maquoi, R. and Rondal, J. “Mise en équation des nouvelles courbes européennes de flambement”.

518 Construction métallique, Vol. 15, no. 1, pp. 17-30, 1978.

519 [73]Taras, A., Dehan, V., da Silva L. S., Marques, L. and Tankova, T. “SAFEBRICTILE:

520 Standardization of Safety Assessment Procedures across Brittle to Ductile Failure Modes”. 2016.

521 [74]Byfield, M. P. and Nethercot, D. A. “Material and geometric properties of structural steel for use

522 in design”. Structural Engineer, Vol. 75, No. 21, 1997.

41 
 
View publication stats