Engineering Structures 106 (2016) 332–347

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Performance of practical beam-to-SHS column connections against

progressive collapse
Wei Wang a,b, Cheng Fang b,⇑, Xi Qin b, Yiyi Chen a,b, Ling Li b
a
State Key Laboratory of Disaster Reduction in Civil Engineering, Tongji University, Shanghai 200092, China
b
Department of Structural Engineering, School of Civil Engineering, Tongji University, Shanghai 200092, China

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

Article history: This paper comprehensively discussed the performance of practical beam-to-SHS column connections
Received 30 July 2015 against progressive collapse. Three full scale tests were performed on double-span systems subjected
Revised 25 October 2015 to the typical column loss scenario, where three connection types were covered in the test programme,
Accepted 27 October 2015
namely, welded flange-weld web connection with internal diaphragms (specimen I-W), welded flange-
Available online 11 November 2015
bolted web connection with internal diaphragms (specimen I-WB), and welded flange-bolted web
connection with short through diaphragms (specimen ST-WB). The test results, including failure
Keywords:
modes/sequence, load–deformation responses and stress conditions, were presented in detail, and the
Square hollow sections (SHSs)
Beam-to-SHS column connections
load transfer mechanism over the entire loading process was thoroughly discussed. Comprehensive finite
Robustness element (FE) models were then established to enable further discussion of the test results. Both experi-
Progressive collapse mental and FE results showed satisfactory ductility supply and load resistance of the three specimens,
Ductility although specimen ST-WB exhibited the lowest ductility against initial fracture due to evident non-
symmetrical stress distribution over the width of the beam flange. For the post-peak response, specimens
I-WB and ST-WB showed reasonable regaining of the resisting load at later catenary stage due to a suf-
ficient redistribution of the stress at the bolted shear tab connections, but limited post-peak load resis-
tance was shown for specimen I-W which underwent quick propagation of the crack over the entire beam
section. A simplified energy balance analysis showed that when a sudden column loss scenario is consid-
ered, progressive collapse may be triggered upon the initial fracture of the bottom beam flange of all the
specimens.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction to the column via the diaphragms using either bolted or welded
connections (or a combination of the two approaches). SHS/RHS
Hollow sections are widely considered for structural column may hold the benefit of their flat surfaces, where a more straight-
members due to their inherent architectural and structural advan- forward connection detailing may be employed. In practice, the
tages. The commonly used types of hollow sections include square, beam flange is normally connected to the column surface via weld-
rectangular, and circular hollow sections (abbreviated as SHS, RHS, ing, whilst the web of the beam is connected to the column
and CHS, respectively), whilst elliptical hollow sections (EHS) have through either direct welding or shear tab bolting. For both cases,
also recently emerged as a structurally and aesthetically appealing as illustrated in Fig. 1(a) and (b), internal diaphragms (not shown
solution [1]. Whilst the sound structural efficiency with good tor- in the figure), which are placed inside the column at the same ele-
sional stiffness make these hollow sections ideal for column mem- vations as those of the beam flanges, should be employed for
bers, some difficulties may exist associated with the practical strengthening purposes. It should be noted that this detailing leads
detailing of the beam-to-column connections due to limited acces- to double-sided welding at the column wall (i.e. internal welding
sibility to the internal space of the closed section columns. for the diaphragm plus external welding for the beam flange),
For hollow section columns with non-flat surfaces (e.g. CHS and which may influence the local performance of the SHS/RHS col-
EHS), outer diaphragms are usually welded around the column, umn, especially when the wall thickness is small. If the internal
and the adjacent open section (e.g. H-section) beams are connected welding is not sufficiently strong (which may happen due to qual-
ity issues), separation between the internal diaphragm and column
wall can occur when the connection rotates, and the tension flange
⇑ Corresponding author. Tel.: +86 (0)21 65982926.
of the beam can cause local fracture of the column wall [2]. In
E-mail address: [email protected] (C. Fang).

http://dx.doi.org/10.1016/j.engstruct.2015.10.040
0141-0296/Ó 2015 Elsevier Ltd. All rights reserved.
 W. Wang et al. / Engineering Structures 106 (2016) 332–347 333

Column Column
Weld Weld Column Weld
Shear plate Shear
plate

High-strength bolt High-strength bolt

(a) I-W (b) I-WB (c) ST-WB
Beam Through
diaphragm

Fig. 1. Practical beam-to-SHS column connections.

addition, to facilitate internal welding (for the internal dia- However, most of the existing studies on structural progressive
phragms), a stub column segment is usually prepared first, and collapse have focused on steel frames with open section beam-to-
after completing the weld work, the short column then needs to open section column connections, whereas the robustness perfor-
be welded to the remaining part of the column. This can further mance of those with practical beam-to-tubular column connec-
complicates the detailing of the connection, and importantly, tions is still not well understood. Compared with the case of
causes more complex initial stress state of the column. An alterna- open section columns where bolted beam-to-column connections
tive solution is to use short ‘through diaphragms’ (i.e. continuous can be readily used, the behaviour of beam-to-hollow section col-
plates ‘cutting’ through the column), as shown in Fig. 1(c). In this umn connections may be less easily predictable due to more com-
case, the beam flange is directly welded to the edge of the dia- plex connection detailing. With increasing popularity of the
phragm and the beam web can be bolted to the column with a nor- application of hollow section columns in modern construction
mal shear tab connection. The use of the through diaphragms can [27], sufficient attention needs to be paid in terms of the robust-
be convenient for fabrication, where in practice, the steel tube is ness of such structural types to ensure public safety. This paper
first cut into three pieces and then they are welded together with sheds considerable light on the fundamental progressive collapse
the two through diaphragms. performance of steel frames with beam-to-SHS column connec-
From the load transfer point of view, these beam-to-column tions, where three full-scale sub-frame tests, covering the three
connections are mainly subjected to bending and shear under nor- practical connections types as shown in Fig. 1, are reported. The
mal conditions. When appropriately designed and fabricated, all test results, including failure modes/sequences, load–deformation
the three solutions shown in Fig. 1 can be readily employed in responses and stress conditions, are presented, and the load trans-
practice and it seems that they have been serving the construction fer mechanism along the entire loading process is thoroughly dis-
industry reasonably well over the past decade. However, there is cussed. Comprehensive finite element (FE) models are then
sparse investigation into their performance when subjected to established to enable further interpretation of the test results,
more complex loading states (e.g. combined axial force, bending, and design comments are finally outlined based on both experi-
and shear) in conjunction with high ductility demand, and the lack mental and FE results.
of information may pose risk of unexpected failure when extreme
loads occur. In particular, progressive collapse has been recognised
as an essential design consideration, following several major acci- 2. Experimental programme
dents in the last century (e.g. Ronan Point apartment block and
Murrah Federal building) [3]. The collapse of World Trade Centre 2.1. Test specimens
in 2001 [4] brought the issue of structural robustness design back
to the fore, and a global attention was then boosted. The design Three test specimens, corresponding to the three typical con-
requirement against progressive collapse started to be introduced nection configurations shown in Fig. 1, were considered in the cur-
in design specifications and guidelines [5,6], and concurrently, sig- rent experimental programme. As the main variation was the
nificant research efforts have been devoted towards more compre- connection detailing, the specimens were named as I-W, I-WB,
hensive understanding of the progressive collapse mechanism for and ST-WB for easy identification. These designations start with
structures. For the performance of steel structures, which is the either ‘I’, which stands for internal diaphragm, or ‘ST’, which stands
focus of the current study, intensive experimental work has been for short through diaphragm, and they end with either ‘W’ or ‘WB’,
carried out at both sub-frame and full-frame levels [7–9], where representing fully welded connection and welded flange-bolted
idealised boundary conditions were normally considered for the web connection, respectively. Each specimen was mainly com-
former case to reflect the influence of the adjacent unaffected prised of two H-section beams (H300
150
6
8 mm) and one
structures. Furthermore, as it has been well recognised that short SHS column (250
14 mm) located at the mid-span, such
beam-to-column connections play a critical role in mitigating pro- that a typical sub-frame subjected to column loss scenario could
gressive collapse potentials, some studies focused on the connec- be reflected. The geometric details of the test specimens are shown
tion performances under combined loading scenarios, where both in Fig. 2(a). The overall length l0 of the double-span sub-frame was
rigid [10–14] and semi-rigid [15–19] connections have been 4.5 m, leading to a span-to-depth ratio of l0/H = 15.0. The two ends
studied in detail. Numerical investigations have also been per- of the system were vertically and horizontally constrained but
formed to further examine the failure mechanism of structures/ were free to rotate (i.e. pin-supported). This boundary condition
sub-structures [20,21], and in either the experimental or numerical was considered to reasonably reflect the points of contra-flexures
work, the most widely used strategy is the alternative path method in real moment frames, as illustrated in Fig. 2(b).
using the ‘column loss’ scenarios to simulate the consequence of Complete penetration groove welds were used to connect the
blast, impact or fire [22,23]. In parallel, analytical models and prac- beams and diaphragms to the column walls. The thickness of the
tical design frameworks [24–26] have been emerging aiming to internal diaphragms for specimens I-W and I-WB was the same
refine the existing design regulations. as that for the beam flange (i.e. 8 mm), whilst for specimen
334 W. Wang et al. / Engineering Structures 106 (2016) 332–347

I-W I-WB ST-WB

(a) Details of specimens

(b) Test setup

Fig. 2. Test arrangements.

ST-WB, 12 mm-thick through diaphragms were employed. For the Fig. 2(b). A quasi-static point load was applied by the actuator to
bolted shear tab connections, four M20 Grade 10.9 frictional high- the top of the centre column with displacement control. During
strength bolts were employed with a pre-tightening force of the loading process, the upper part of the column was fixed to
155 kN [28]. The contact surfaces between the shear tab and the the loading head, and the lower part was guided to move vertically
beam web were pre-treated with sand blasting. Grade S355 steel through the sliding support, such that no rotational movement of
was considered for the specimens, and the detailed material prop- the column was allowed. This led to a generally symmetrical per-
erties were obtained via coupon tests according to the relevant formance of the system at the two sides of the column (i.e. ‘east’
specification [29], as detailed in Table 1. Each set of results in the and ‘west’ sides for ease of discussion), which is true when the col-
table was based on the average values of three coupons for each umn immediately above the affected floor can offer sufficient rota-
part. tional restraints. In fact, minor gaps existed between the column
and the sliding support, and therefore slight rotation can still be
2.2. Test setup and procedure induced. The tests were terminated when either complete fracture
of the connection occurred or the limiting displacement Dmax of
The test setup, consisting of a servo actuator (capac- the test frame was reached (Dmax was approximately 400 mm,
ity = 2000 kN), a column base sliding support, and self-balanced corresponding to a beam chord rotation of 0.178 rad). For speci-
horizontal and vertical support frames, is schematically shown in men I-W, however, malfunction of the column base sliding support
 W. Wang et al. / Engineering Structures 106 (2016) 332–347 335

Table 1
Material properties from coupon tests.

Material Yield strength (MPa) Ultimate strength (MPa) Rupture strain (%) (A0  Afracture)/A0
Beam flange (8 mm) 430 577 24 0.46
Beam web (6 mm) 417 514 27 0.48
Shear tab plate (6 mm) 409 505 29 0.61
Diaphragm (12 mm) 450 574 18 0.48
Column wall (14 mm) 482 545 24 0.68

Note: A0 = original cross-sectional area of coupon, Afracture = necked cross-sectional area after fracture.

occurred during the test, and a certain level of column rotation was the occurrence of local buckling, and a peak load of 220 kN (point
induced. In order to prevent damage to the actuator, the test A1 in Fig. 4) was achieved at D = 142 mm (h = 0.0 63 rad), at which
stopped at a relatively early stage for this specimen when the dis- moment fracture of the bottom flange at the W3 section occurred,
placement reached 200 mm. The two ends of the system were leading to abrupt decreasing of the resisting load. The crack was
pin-supported via two hinges such that a full axial restraint was initiated near the weld access hole, and then quickly propagated
maintained. This was used to reflect the condition of relatively along the flange and web of the beam near the welds. In-plane
strong/rigid neighbouring sub-structures providing a sufficient rotation of the SHS column was then induced due to the malfunc-
level of axial restraints, which can be true for most typical steel tion of the column base sliding support, as discussed previously.
frames due to the strong axial restraints offered by the adjacent The incline of the column caused a concentration of fracture at
beams in conjunction with the ‘diaphragm effect’ of the floor sys- the W3 section, whereas no fracture was induced at the opposite
tem [30]. The case of flexible adjacent structures, which leads to side. With a further increase of the deflection, the column contin-
insufficient axial restraints to maintain catenary action of the ued to incline, and the resisting load was maintained below 60 kN.
affected double-span system, is not within the scope of the current To avoid damage to the loading system, the test was terminated at
discussion, and this may be worth future studies. D = 200 mm, where the crack had propagated to 1/3 depth of the
beam web.
When the weld along the beam web was replaced by a shear tab
2.3. Instrumentations
bolted connection, as was the case for specimen I-WB, the failure
behaviour could be evidently changed. As shown in Fig. 5, buckling
The applied point load was automatically recorded by the actu-
of the top flange of the west side beam occurred at D = 75 mm
ator system, and the deflection of the double-span system was
(h = 0.0 33 rad), and initial cracking was induced at the bottom
monitored through placing a series of displacement transducers
flange of the W3 section near the weld access hole. When D
along the beam length with certain intervals, as shown in Fig. 3
achieved 166 mm (h = 0.0 75 rad), the crack propagated over the
(a). Strain gauges were employed to monitor the strain distribu-
entire bottom flange, causing abrupt drop of the resisting load from
tions over critical beam sections, as generally shown in Fig. 3(b).
219 kN to 146 kN. After that, the resisting load started to pick up
For each specimen, the strains over the E1 and W1 sections
but the trend was quickly interrupted due to the subsequent frac-
(615 mm from the east and west beam end hinges respectively)
ture at the E3 section, where the resisting load dramatically
were recorded to deduce the axial force development within the
dropped from 219 kN to 77 kN. With further increase of the deflec-
sub-frame, as discussed in detail in Section 4. The strains over
tion, the resisting load was slightly regained until the occurrence of
the critical sections, i.e. sections E2/W2 and E3/W3, as illustrated
fracture at the west side and east side shear tabs at D = 320 mm
in Fig. 3(b), were also recorded to examine the strain development
and D = 387 mm, respectively, during which stage the load–
conditions near the joint fractural zone.
displacement curve showed evident fluctuation due to successive
fractures along the bolt line. The W3 section experienced complete
3. Test results fracture at D = 400 mm (h = 0.178 rad), and then the test was
terminated.
3.1. General behaviour and failure modes For specimen ST-WB which employed short through dia-
phragms instead of internal diaphragms, top flange buckling was
In general, all the specimens exhibited satisfactory load resist- developed when the displacement reached 71 mm, and meanwhile
ing capacity and ductility supply. No weld failure was observed, minor cracking was induced at the bottom flange of the beam near
which indicated good weld quality for the specimens. The load– section W3. The crack was initiated near the edge of the bottom
deflection responses of the three specimens, with the associated flange at section W3, and it gradually propagated until complete
key failure stages identified during the loading process, are shown fracture of the bottom flange when D reached 98 mm, and as a
in Figs. 4–6. It should be noted that both the applied load F and the result the resisting load decreased from 154 kN to 100 kN. This
normalised load F/FP are given in the figures, where FP is the theo- was followed by an increase of the resisting load, which was
retical vertical resisting load corresponding to the formation of full accompanied by significant bearing deformation of the bolt holes
plastic hinges of the beam at the critical sections W3/E3, under a of the shear tabs. Complete fracture of the bottom flange at section
flexural bending mechanism. In addition, both the displacement E3 occurred when D = 141 mm, and as a result the resisting load
D and the beam chord rotation h are given in the figures. dropped abruptly from 186 kN to 48 kN. Subsequently, gradual
For specimen I-W, where the beam was fully welded to the col- fracture of the shear tabs at both sides occurred whilst the resisting
umn wall, linear load–displacement response was shown when the load kept increasing. Finally at D = 382 mm (h = 0.170 rad), com-
centre column displacement D was less than 30 mm. This indi- plete fracture of section W3 was induced, and the specimen com-
cated elastic flexural bending performance in the initial stage. FP pletely lost its load resistance. It is worth mentioning that the peak
was achieved when D exceeded 60 mm (h = 0.0 27 rad), where resisting load immediately before the complete fracture of section
local buckling was induced at the top flange of the beam, which W3 (i.e. point C5) exceeded the initial peaks (i.e. C1 and C2). This
may be attributed to combined effects of flexural bending and implies that at later stages, where the connections had already
minor compressive arching. The resisting load kept increasing after been severely damaged, a remarkable level of resisting load could
336 W. Wang et al. / Engineering Structures 106 (2016) 332–347

(a) Displacement transducers

Top
S1(S2) S3(S4)
West East
S9(S10)
Below
S11(S12)
S13(S14)

S5(S6) S7(S8)

S17(S18)S19(S20) T15(T16) T17(T18)

S15(S16) S21(S22) T13(T14) T19(T20)
S31(S32)
S33(S34)
S35(S36)
S37(S38)
S39(S40)
S23(S24) S29(S30) T21(T22) T27(T28)
S25(S26)S27(S28) T23(T24) T25(T26)

S15(S16) S17(S18) S33(S34) S35(S36)

S23(S24)
S25(S26)
S27(S28)
S29(S30)
S31(S32)

S19(S20) S21(S22) S37(S38) S39(S40)

S15(S16) S17(S18) S33(S34) S35(S36)

S23(S24)
S25(S26)
S27(S28)
S29(S30)
S31(S32)

S19(S20) S21(S22) S37(S38) S39(S40)

(b) Strain gauges

Fig. 3. Test instrumentations.
 W. Wang et al. / Engineering Structures 106 (2016) 332–347 337

θ (rad)
0.00 0.05 0.10 0.15 0.20
400
I-W Test 2.0
a1 FE-symmetric
300 FE-non-symmetric
1.5

F (kN)
a2

F /F P
200 A1
FP 1.0

100
0.5

0 0.0
0 100 200 300 400 500
Δ (mm)

Test

No test data

Bottom flange and beam web fractured at

Section W3 (A1)
FE

Bottom flange and beam web fractured at

Top flange fractured at Section W3/E3 (a2)
Section W3/E3 (a1)

Fig. 4. Test and FE results for specimen I-W.

still be maintained, which was probably attributed to the catenary system was mainly dependent on the performance of these con-
action. This can be later confirmed by the axial force calculated nections under complex internal forces.
from the strain distributions at sections E1/W1, as will be dis-
cussed in Section 4. 3.2. Strain gauge readings
The load–deflection response of the specimens can be further
depicted through the deflection profiles obtained from the dis- Typical strain gauge readings of the specimens are shown in
placement transducers placed along the length of the beams, as Fig. 7(b). At the sections which were relatively far away from the
shown in Fig. 7(a). For specimens I-WB and ST-WB, because of connection zone (e.g. W1 section), typical bending action with
the presence of the column base sliding support, the performance top flange under compression and bottom flange under tension
at the two sides of the column was generally symmetrical. For was shown at initial loading stages. With increasing deflection,
specimen I-W, however, the malfunction of the sliding support the negative (compressive) strains started to convert to positive
caused column rotation, and thus larger deflection was observed values, which indicated that the cross sections gradually turned
at the west side of the column due to localised development of from bending action to axial tension (i.e. catenary action). Some
cracking. Importantly, it can be seen that at initial loading stages, sudden changes of the strain gauge readings corresponded to
i.e. D < 30 mm, a typical curved deformation shape was exhibited, the fracture of the connection zone (which caused decrease of
which indicated that the resisting mechanism was mainly gov- the resisting load). No yielding was observed in the E1/W1 sections
erned by flexural bending. With increasing deformations, the of the specimens, and therefore these strain gauge readings could
deformed configuration was gradually changed to that exhibiting be further used to calculate the axial force and bending moment
two straight lines intersecting at the location of centre column, of these sections. Much higher strain levels were observed at the
indicating that plastic hinges were developed near the connection critical sections in the connection zone (e.g. E2/W2 sections) due
zone. This reveals that the ductility demand of the specimen sub- to considerable plastic deformation. Typical bending type strain
jected to column removal was mainly accommodated by the con- distributions were shown in the initial loading stage, but the
nection zone, and the progressive collapse resistance of the strains increased significantly when D exceeded 30 mm. The
338 W. Wang et al. / Engineering Structures 106 (2016) 332–347

θ (rad)
0.00 0.05 0.10 0.15 0.20
300
I-WB b1
b3 1.5
B1 B2
b2 B4
200 B5
1.0
FP

F (kN)

F / FP
B3
100
0.5
Test
FE
0 0.0
0 100 200 300 400 500
Δ (mm)

Test

Fracture of bottom flange at W3 (B1) Fracture of bottom flange at E3 (B2) Fracture of shear plate near W3 (B3)

Fracture of shear plate near E3 (B4) Fracture of top flange at W3 (B5)

FE

Fracture of bottom flange (b1) Fracture of shear plate (b2) Fracture of top flange (b3)

Fig. 5. Test and FE results for specimen I-WB.

strains at these sections showed less regular distribution patterns of the bottom flange of the beam, provided that no premature weld
compared with those at E1/W1 sections, which was due to the failure occurred. Therefore, it was believed that the similar ulti-
more complex behaviour of the connection zones (e.g. local buck- mate load resistance of the three specimens was attributed to
ling and fracture of the beam). Most of the strain gauge readings the similar connecting method for the beam flanges to the column,
over the height of the beam cross section exceeded the yield strain, i.e. by welding. It is of interest to find that although a similar level
which indicated that a plastic hinge was formed. At later loading of resisting load upon initial fracture was observed, the fracture for
stages, a transition trend from bending to bending–tension com- specimen ST-WB was induced earlier than that for the other two
bined response was also shown, as signified by the gradual specimens. This may be caused by the difference of the connection
decrease of the negative strain. detailing near the bottom flange of the beam, noting that the beam
flanges were directly welded to the column face for specimens I-W
4. Discussions of test results and I-WB, but they were welded to the edge of the short dia-
phragm for specimen ST-WB. The different configurations of the
4.1. Influence of connection detailing weld access holes may also attribute to the varying deflections
where fracture of the beam bottom flange occurred.
The test results generally showed a similar level of ultimate After the initial fracture, the subsequent load resisting response
load resistance (as listed in Table 2) for the three specimens. As differed evidently with different connection detailing. For speci-
it was observed that the initial fracture was developed at the bot- men I-W, where the web of the beam was welded to the column
tom flange of the beam for all the specimens, the resisting load face, the resisting load dropped significantly after the first load
against initial fracture should be governed by the fractural strength peak to a very low level with no sign of picking up until the end
 W. Wang et al. / Engineering Structures 106 (2016) 332–347 339

θ (rad)
0.00 0.05 0.10 0.15 0.20
300
ST-WB c3
c1 1.5
c2
C2
200 C1 C5
1.0

F (kN)
FP

F / FP
C4
100 C3 0.5

Test
FE
0 0.0
0 100 200 300 400 500
Δ (mm)

Test

Fracture of bottom flange at W3 (C1) Fracture of bottom flange at E3 (C2) Fracture of shear plate near W3 (C3)

Fracture of beam web near E3 (C4)

Fracture of top flange at W3 (C5)
FE

Fracture of bottom flange (c1) Fracture of shear plate (c2) Fracture of top flange (c3)

Fig. 6. Test and FE results for specimen ST-WB.

of the test, where 1/3 height of the beam web had been fractured. 4.2. Load resistance mechanism
The low resisting load could be due to limited internal force
redistribution capability of the specimen once the crack propa- The load transfer mechanism of the specimens can be further
gated to the web of the beam. Further numerical studies may be interpreted through the internal forces developed within the sub-
required to trace the load resistance conditions of the specimen frame. As mentioned previously, the strain gauge readings in sec-
at later stages. On the other hand, the internal force could be effec- tion E1/W1 can be employed to calculate the internal forces at
tively transferred to the shear tab connections for specimens I-WB the section, as given by:
and ST-WB, and as a result a considerable portion of the loss of the
resisting load could be regained during the redistribution of the Re De M1
N1 ¼ EA ; M1 ¼ EI ; V 1 ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð1Þ
force within the shear tab. This process could allow a further n Dh 2
l1 þ d21
deflection of approximately 200 mm for the system with a reason-
able level of resisting load sustained after initial beam flange frac-
ture. During this process, the fractural strength of the shear tab where N1, M1 and V1 are the axial force, bending moment and shear
was gradually exhausted, which finally led to complete fracture force at section E1/W1, respectively; E, A and I are the Young’s mod-
of the shear tab along the bolt line. This was quickly followed by ulus, cross-sectional area and second moment of area of the beam,
the fracture of the top flange of the beam, leading to complete fail- respectively; Re/n is the average strain over the section; De/Dh rep-
ure of the systems. resents the curvature of the section; l1 is the horizontal distance
between section E1/W1 to the pin-support; and d1 is the vertical
340 W. Wang et al. / Engineering Structures 106 (2016) 332–347

0 0 25mm 0 25mm
25mm 50mm
-50 50mm
50mm 100mm -50
100mm
Deflection (mm)

Deflection (mm)

Deflection (mm)
-50 75mm -100 -100
150mm 150mm
D17 100mm D16 -150 D17 D16
200mm -150 D17 D16
-100 -200 200mm
125mm 250mm -200
D15 150mm D14 -250 D15 300mm D14 D15 250mm
-250 D14
-150 175mm -300 350mm 300mm
D13 200mm D12 -350 D13 400mm D12 -300 D13 350mm D12
-200 -400 -350 382mm
D11 D10 D10
I-W D11 D10 -450 I-WB -400 D11
(D1+D2)/2 (D1+D2)/2 ST-WB (D1+D2)/2
-250 -500 -450
-2000 -1500 -1000 -500 0 500 1000 1500 2000 -2000 -1500 -1000 -500 0 500 1000 1500 2000 -2000 -1500 -1000 -500 0 500 1000 1500 2000
Distance from column axis (mm) Distance from column axis (mm) Distance from column axis (mm)

(a) Deflection responses

θ (rad) θ (rad) θ (rad)
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
2000 2000 2000
I-W:W1 I-WB:W1 ST-WB:W1
1500 1500 1500

1000 1000 1000

ε (×110 − 6 )
ε (×10−66 )

ε (×10− 6)

500 500 500

0 0 0
(S1+S3)/2 (S1+S3)/2 (S1+S3)/2
-500 S9 S11 S13 -500 S9 S11 S13 -500 S9 S11 S13
(S5+S7)/2 (S5+S7)/2 (S5+S7)/2
-1000 -1000 -1000
0 100 200 300 400 500 0 100 200 300 400 500 0 100 200 300 400 500
Δ (mm) Δ (mm) Δ (mm)

θ (rad) θ (rad) θ (rad)

0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
20000 20000
I-W:W2 I-WB:W2 ST-WB:W2
15000 20000
10000
10000
ε (×110−6)

ε (×110− 6)
0
ε (×10−6)

5000
0
0
-20000
(S15+S17+S19+S21)/4 (S15+S17)/2 (S15+S17)/2
-5000 S31 S33 S35 -10000
S23 S25 S27 S23 S25 S27
-10000 S37 S39 -40000 S29 S31 S29 S31
(S23+S25+S27+S29)/4 (S19+S21)/2 (S19+S21)/2
-15000 -20000
0 100 200 300 400 500 0 100 200 300 400 500 0 100 200 300 400 500
Δ (mm) Δ (mm) Δ (mm)
(b) Strain gauge readings
Fig. 7. Deflection and strain gauge readings.

Table 2
Results of experimental and FE studies.

Specimens Test results FE results Connection type

FIF (kN) D/h at FIF FFF (kN) D/h at FFF FIF (kN) D/h at FIF FFF (kN) D/h at FFF
(mm/rad) (mm/rad) (mm/rad) (mm/rad)
I-W 220 142/0.063 – – 273 176/0.078 155 375/0.167 Welded flange-welded web
I-WB 219 166/0.075 202 387/0.172 244 177/0.079 237 372/0.165 Welded flange-bolted web
ST-WB 184 84/0.038 218 382/0.170 220 118/0.052 242 399/0.177 Welded flange (via short diaphragm)-bolted web

Note: FIF = load at initial fracture (initial peak), FFF = load prior to final fracture.

deflection of the beam at section E1/W1 (according to the readings that the calculation of Mi is based on the assumption of non-
from the displacement transducers), as illustrated in Fig. 8(a). Based fractured section, where the centroid is at the mid-depth of the
on the internal forces at section E1/W1, the vertical and horizontal beam. When fracture occurs, this calculated moment is inaccurate
reactions (VR and HR, respectively) of the pin support can be over the fractured section, and thus the result can only be consid-
obtained: ered as an indicative (or ‘virtual’) moment.
Based on the above equilibrium equations, Fig. 9(a) shows the
V R ¼ V 1 cos h1 þ N1 sin h1 ; HR ¼ N 1 cos h1  V 1 sin h1 ð2Þ calculation of the axial force at the critical section E3/W3 (where
fracture occurred). It was observed that minor compressive force
where tan h1 = d1/l1. Considering the equilibrium of the beam seg-
was induced at the very beginning, which was probably due to
ment, as shown in Fig. 8(a), the internal forces at any section can
the compressive arching effect. As the double-span system was
be obtained using the following expressions:
not overly ‘stocky’ with a reasonable l0/H ratio of 15.0, the com-
Ni ¼ HR cos hi þ V R sinhi ; V i ¼ V R cos hi  HR sinhi ; M i ¼ V R li  HR di pressive force quickly decreased and tensile force was subse-
quently induced. The initial crack of the bottom flange of the
ð3Þ
beam led to decrease of the axial force, which echoes the
where i indicates the corresponding values (i.e. axial force, shear load–deformation responses discussed previously. The axial force
force or bending moment) at any section Ei/Wi. It should be noted developed upon initial beam flange fracture could achieve 0.15Np
 W. Wang et al. / Engineering Structures 106 (2016) 332–347 341

beam section, the maximum axial force could increase beyond

0.5Np. This confirmed that significant catenary action was devel-
oped at later stages of the loading process. For specimen I-W, the
evolution of axial force could not be traced after the fracture of
the beam flange due to early finish of the test, but it could be
observed that no significant catenary action had been developed
yet when the test was terminated.
Whilst the axial force continued to increase after fracture of the
(a) Internal force distributions beam bottom flange, the flexural bending action showed an oppo-
site trend, as shown in Fig. 9(b). At initial loading stages, the bend-
ing moment was developed following a typical flexural bending
FV
N1-W V 1-W V 1-E N1-E action, and the short plateau prior to the fracture of the beam
flange was due to the significant plastic deformation induced at
the critical sections, where plastic hinges could be formed (or
nearly formed). After the occurrence of fracture, the bending
moment decreased significantly, which indicated that the load
(b) Contribution of internal forces to resisting load resistance mechanism started to convert from flexural bending
action into catenary action.
Fig. 8. Free body diagram of test systems.
The load resistance mechanism can be further elaborated by
illustrating the contributions from flexural bending action and
(Np = fyA). Importantly, it can be clearly seen that after the incep- catenary action to the overall resisting load. Based on the free body
tion of fracture, the axial force could still keep increasing for both diagram shown in Fig. 8(b), the portion of the resisting load con-
specimens I-WB and ST-WB during the successive fracture of the tributed to by the catenary action FR-C can be expressed as:
shear tab connections, and prior to complete fracture of the entire F R-C ¼ N1-E sin h1 þ N1-W sin h1 ð4Þ

θ (rad) θ (rad)
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
C5
1000 I-W: W3 C5 0.6 1000 I-W: E3 B5 0.6
B5
I-WB: W3 I-WB: E3
800 ST-WB: W3 800 ST-WB: E3
0.4 0.4
N (kN)

N / NP

600
N (kN)

600 B2

N / NP
400 B1 400 C2
A1 0.2 A1 0.2
200 C1 200

0 0.0 0 0.0
0 100 200 300 400 500 0 100 200 300 400 500
Δ (mm) Δ (mm)
(a) Development of axial forces
θ (rad) θ (rad)
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
C1 A1 B1 A1
200 200 B2
1.0 1.0

100 0.5 100 0.5

C2
M (kN⋅m)

M (kN⋅m)

M / MP
M / MP

0 0.0 0 0.0
I-W: W3 SI-W: E3
-100 I-WB: W3 -0.5 -100 SI-WB: E3 -0.5
ST-WB: W3 SST-WB: E3
-200 -1.0 -200 -1.0
0 100 200 300 400 500 0 100 200 300 400 500
Δ (mm) Δ (mm)
(b) Development of bending moment
θ (rad) θ (rad) θ (rad)
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
ST-WB
300 I-W 300 I-WB B5 300
1.5 1.5 C5 1.5
A1
200 200 B1 200 C1
1.0 1.0 C2 1.0
B2
F (kN)
F (kN)

F (kN)

F / FP
F / FP

F / FP

100 0.5 100 0.5 100 0.5

0 0.0 0 0.0 0 0.0

FR-C FR-C FR-C
-100 FR-F -0.5 -100 FR-F -0.5 -100 FR-F -0.5

0 100 200 300 400 500 0 100 200 300 400 500 0 100 200 300 400 500
Δ (mm) Δ (mm) Δ (mm)
(c) Contributions of internal forces to load resistance
Fig. 9. Load resistance mechanism.
342 W. Wang et al. / Engineering Structures 106 (2016) 332–347

and that contributed to by the flexural bending action FR-F can be over the initial peak load, is taken as 0.2. Following the design
easily obtained by deducting FR-C from the overall resisting load. guidelines, the load vs. beam chord rotation responses of the three
Fig. 9(c) shows the contributions of the two actions to the resisting specimens can be idealised by multi-linear curves as shown in
load, and the responses were in line with the previously discussed Fig. 10(a). The plastic rotations can be obtained by deducting the
evolutions of the axial force and bending moment. The results con- yield rotation hy from the overall beam chord rotation. In general,
firmed that the load resistance of the specimens significantly relied all the three specimens showed sound ductility when subjected
on catenary action at large deflections, especially when D exceeded to mid-column removal. The plastic rotation angle a (i.e. that prior
200 mm. to initial fracture) ranged from 0.031 to 0.065 rad, which signifi-
cantly exceeded the acceptance criterion proposed by DoD [6]. This
indicates that all the three practical beam-to-SHS column connec-
4.3. Ductility supply
tions can accommodate significant plastic rotation without loss of
resisting load. For the post-peak response, the plastic rotation
In practical design of building structures against progressive
angles b for specimens I-WB and ST-WB were evidently larger than
collapse, it is normally considered that the progressive collapse
those stipulated in DoD [6]. The minimum residual strength ratio c
potential of a structure is governed by the ductility supply of the
of the specimens was 0.251, which exceeded the acceptance crite-
connections. The major design guidelines (e.g. DoD) stipulated a
rion of 0.2. The results generally showed that the recommended
series of acceptable plastic rotation angles (Acceptance Criteria)
acceptance criteria for ductility performance of fully restrained
for various connection types for nonlinear modelling of steel con-
moment connections are on the conservative side for the current
nections, and failure may be considered to occur if the plastic rota-
specimens.
tion angles of the connection exceeds the acceptable ones. In
particular, if a connection exhibits an idealised multi-linear curve
type as shown in Fig. 10(a) (which is the case for the current spec- 4.4. Prediction of dynamic response
imens), detailed acceptable plastic rotation angles (i.e. limiting val-
ues of ‘a’ and ‘b’) are listed [6]: for welded unreinforced The focus of the current experimental study was given to the
flange-bolted web connections (i.e. specimens I-WB and ST-WB), static performance of the double-span systems subjected to centre
the acceptable plastic rotation angles (in radian) are column loss, but when actual extreme events occur (e.g. blast),
a = 0.021  0.0118d = 0.017 and b = 0.050  0.0236d = 0.043, sudden column removal, which is associated with dynamic
where d is the beam depth in metre. For welded unreinforced response, can be a more rational assumption that realistically
flange-welded web connections (i.e. specimens I-W), the values reflects the consequence of an extreme event. Sudden removal of
may be taken as a = 0.0284  0.0157d = 0.024 and b = 0.043  a column is in effect close to suddenly applying the gravity load
0.0236d = 0.036. In addition, the acceptable residual strength ratio on the same structure in the absence of the column at the begin-
c, which is expressed as the ratio of the post-fracture resisting load ning, especially when significant displacements can be sustained

θ (rad) θ (rad) θ (rad)

0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
400 300 300
I-W I-WB ST-WB
Test 2.0 Test 1.5 Test 1.5
Idealised model Idealised model Idealised model
300
1.5 200 200
1.0 1.0
F (kN)

F (kN)
F / FP
F (kN)

F/ FP
F/ FP

200
1.0 a=
100 0.5 100 0.031 0.5
100 0.5 a = 0.065
a = 0.055 b = 0.166 c= b = 0.159 c=
c = 0.254 0.350 0.251
0 0.0 0 0.0 0 0.0
0 100 200 300 400 500 0 100 200 300 400 500 0 100 200 300 400 500
Δ (mm) Δ (mm) Δ (mm)
(a) Ductility assessment

θ (rad) θ (rad) θ (rad)

0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
300 300 300
I-W I-WB ST-WB Static response 1.5
Static response 1.5 Static response 1.5
Pseudo-static response Pseudo-static response Pseudo-static response

200 200 200

1.0 1.0 1.0
F (kN)
F (kN)
F (kN)

F / FP

F / FP
F / FP

100 0.5 100 0.5 100 0.5

0 0.0 0 0.0 0 0.0

0 100 200 300 400 500 0 100 200 300 400 500 0 100 200 300 400 500
Δ (mm) Δ (mm) Δ (mm)
(b) Dynamic response assessment
Fig. 10. Further discussion of test results.
 W. Wang et al. / Engineering Structures 106 (2016) 332–347 343

by the structure as a result [24]. Based on the available static development of numerical modelling can offer an efficient comple-
response, such as the static load–deflection response obtained mentary insight into its progressive collapse performance at later
from the current study, a dynamic response reflecting sudden loading stages. The numerical study can also effectively reveal
column loss can be deduced using an energy balance approach the complex stress distributions within the structural components,
proposed by Izzuddin et al. [24]. An important benefit of this which may help explain the fracture initiation and propagation
method is its convenience of using the principle of energy balance phenomena of the three specimens. The general nonlinear finite
to perform simplified dynamic assessment instead of directly con- element (FE) analysis package ABAQUS [31] was used for this pur-
ducting complex nonlinear dynamic analysis. As illustrated in pose. As the current numerical study involves complex material
Fig. 10(b), if the static load–deformation response is known, the behaviour (e.g. fracture) and a large number of contact pairs, con-
allowable dynamic (i.e. sudden column loss) load resistance at vergence difficulty could be a critical issue. This can be addressed
any ductility demand can be obtained by achieving an equivalence by employing the dynamic explicit analysis which is able to well
between external work and internal energy, i.e. when the two predict the quasi-static response of structures with negligible
hatched areas become identical. By considering different levels of dynamic effect, if the kinetic energy of the deformed structure is
ductility demand and employing the energy balance principle, a sufficiently low compared with its internal energy. Therefore, by
dynamic (pseudo-static) curve, which represents the relationship ensuring that the kinetic energy was kept within 10% of the inter-
between the sudden applied load and the associated dynamic nal energy throughout the entire simulation process, as suggested
deformation, can be constructed. by [32], explicit solver was used for the current numerical analyse.
Fig. 10(b) shows the pseudo-static load vs. central column dis- The C3D8R elements, which are 8-node linear brick elements
placement responses based on the available static test results. It with reduced integration and hourglass control, were employed
is generally observed that at the same applied load, the ductility for all the structural components, including the SHS column, H
demand of the system subjected to sudden column loss is evidently beam, diaphragm plates, shear tab, and high-strength bolts. The
higher than that with static column removal. In other words, with general meshing size for the column and beams was approximately
the same ductility demand, the allowable resisting load to prevent 5 mm, but a refined meshing size of around 1 mm was employed
progressive collapse is decreased if the system experiences sudden for the areas adjacent to the connection zone, as typically shown
column loss. A gradual decrease of the pseudo-static curve is in Fig. 11. ‘Hard contact’ with no penetration in the normal direc-
shown after reaching the maximum load capacity, which indicates tion was considered for all contact pairs, and a coefficient of fric-
that if the applied load exceeds the maximum dynamic load capac- tion of 0.45 was used corresponding to the actual treatment of
ity marked in the figure, the internal energy stored within the the steel surface. As no weld failure was observed during the tests,
deformed system cannot balance the external work done by the the ‘tie’ interactions were employed to simulate all the complete
applied load again (even though significant regain of the static penetration groove welds. The boundary conditions of the models
resisting load is exhibited), and progressive collapse can hence be were applied to reflect the actual conditions of the test setup,
triggered as a result. Despite this, the regain of the static resisting where idealised pin-supports were considered at the two external
load after initial fracture still tends to slow the decreasing rate of ends of the beam and the centre column was only allowed to move
the pseudo-static curve, and it is believed that if a further pick vertically with no in-plane rotation. For specimen I-W, however,
up of the static resisting load can be developed beyond that two boundary conditions for the centre column were considered:
observed in the current specimens, the strain energy stored over (1) the column was laterally restrained along the length, which is
this regaining process can finally re-balance the external work, consistent with the boundary condition for the other two speci-
and the equivalent state will be achieved at a much larger defor- mens, and (2) the constraints at the lower part of the column
mation with larger load resisting capacity. Since no such significant was removed, which was used to consider the actual case where
regain of the static resisting load was developed for the three spec- the column base sliding guide did not work properly for this spec-
imens tested in this study, it may be preliminarily concluded that, imen. For the latter case, the material fractural strain (as discussed
when a sudden column loss scenario is considered, progressive col- below) of the beam at one side of the column (east side) was
lapse may be triggered upon the initial fracture of the bottom slightly increased in order to trigger non-symmetrical performance
flange of the beam. This can be preliminarily considered as a sim- (e.g. fractural development) of the symmetrical FE model.
plified and conservative failure criterion for such connections in The basic nonlinear material property of steel was simulated
practical design when sudden column loss strategy is incorporated. using the isotropic hardening model with the von Mises yield cri-
terion. The fundamental material properties, including the modu-
5. Numerical study and further discussion lus of elasticity, yield strength, and ultimate strength, were
obtained from the tensile coupon test results. The engineering
5.1. Modelling strategy stress rEng and strain eEng were then converted to true stress rT
and strain eT in ABAQUS, as expressed by:
As the early finish of the test for specimen I-W prohibited a
complete understanding of its load resistance mechanism, the rT ¼ rEng ð1 þ eEng Þ; eT ¼ lnð1 þ eEng Þ ð5Þ

Fig. 11. Typical finite element model and meshing scheme.

344 W. Wang et al. / Engineering Structures 106 (2016) 332–347

and the true plastic strain ep for ABAQUS input can be obtained by: observed between the FE predictions and test results, especially
rT in terms of the trend of load dropping and regaining responses as
ep ¼ eT  ð6Þ well as the fractural patterns. The minor discrepancy may be
E
caused by some test uncertainties which are difficult to be fully
To simulate fracture of steel, the progressive ductile damage reflected by the FE model. For instance, in the actual test, it was dif-
model offered by ABAQUS was employed. The damage model ficult to fully fix the pin-supports in the axial direction due to
allows for ductile fracture of steel that experiences extensive plas- slight unavoidable gaps in the hinges as well as in the adjacent
tic deformation in the necking phase prior to fracture, and after the connectors; however, the beam ends of the FE model were consid-
initiation of fracture, a user-defined degradation response of the ered as pin-supported in an idealised manner with no movement
material stiffness is enabled, where the fully damaged elements allowed in the axial direction. This idealised pin support condition
will be deleted from the mesh. Therefore, in order to capture the in the FE model can cause more significant compressive arching
progress of material damage, two sets of parameters need to be effect in the initial stage, which may explain the slightly higher ini-
incorporated, namely, a damage initiation criterion and a damage tial resisting load of the FE predictions compared with the test
evolution response. The damage initiation criterion describes the results for some specimens. Possible variations of material proper-
maximum equivalent plastic strain which initiates damage. The ties could also lead to the discrepancy. Nevertheless, the proposed
damage evolution law describes the condition of degradation of FE modelling strategy can well capture all the key stages of the
the material stiffness once the corresponding initiation criterion specimens from initial column removal until final collapse. Impor-
has been reached, and a linear damage evolution law was consid- tantly, the late-stage load resisting mechanism of specimen I-W,
ered in the current study. which was not recorded by the test, is revealed by the FE
To simulate the fracture of steel under a uniaxial state, the fail- predictions.
ure true stress rf and strain ef (at which fracture is triggered) can It is observed that the two boundary conditions considered for
be obtained from the coupon test results and by using the follow- the centre column of model I-W leads to similar load–deflection
ing equations: development trends, but different fractural performances/
  sequences are developed. When the lower part of the centre col-
F fracture A0
rf ¼ ; ef ¼ ln ð7Þ umn was laterally constrained, leading to symmetrical perfor-
Afracture Afracture
mance of the model at the two sides of the column, the
where Ffracture and Afracture are the tensile load and the necked load–deflection response exhibits two distinct load peaks, corre-
(reduced) cross-sectional area of the coupons where fracture sponding to the initial fracture at the bottom flange of the beam
occurred, and A0 is the original cross-sectional area of the coupons. and the final fracture of the whole section. It is noted that with lim-
When the material is subjected to multi-axial stress states, the ited regain of the resistance, the resisting load at the second peak is
influence of stress triaxiality on the material fractural behaviour significantly lower than that in the first peak. When the constraint
may need to be considered. However, as detailed triaxial material at the lower part of the centre column is removed, which is closer
testing is normally not considered as a standard procedure in com- to the actual testing conditions, the bottom flange of the beam at
mon experimental programmes, and metal fracture itself is a com- the west side fractures first, and it is until this crack propagates
plicated issue, inconsistent approaches were normally employed in significantly into the web of beam that the initiation of fracture
various investigations [2,10,32–34]. Some studies adopted an itera- is induced at the opposite side of the column, as shown in Fig. 12
tive procedure to determine the input parameters for fracture sim- (a). This phenomenon is generally in line with the test observa-
ulation [2,10]. In this study, a simplified approach based on tions. Both the test result and FE prediction show limited regaining
‘Rupture Index (RI)’ [34] was used to reflect stress triaxiality, as of the resisting load of specimen I-W, which indicates that the
given by: welded flange-welded web solution for beam-to-SHS column con-
nections have unsatisfactory resistance picking up mechanism at
PEEQ
RI ¼ ð8Þ catenary stage once initial fracture occurs.
e1:5T
For specimens I-WB and ST-WB, the FE results confirm that the
where PEEQ is the equivalent plastic strain, T is the stress triaxiality resisting load can regain evidently after initial fracture of the bot-
which is the ratio of the hydrostatic stress over the von Mises stress. tom flange of the beam, and the regained load can exceed the ini-
Rupture Index is an indicator reflecting the potential for ductile tial peak for specimen ST-WB. The FE predictions also confirm that
fracture (i.e. the higher the PEEQ, the higher the potential for frac- the initial fracture of ST-WB occurs earlier than that of the other
ture), and the value at fracture for a specific material may be two specimens. In order to explain this phenomenon, Fig. 12(b)
obtained by using Eq. (8) with the uniaxial coupon test results, gives the PEEQ contour of the three specimens at a series of deflec-
where T = 0.33 for the uniaxial loading state. Employing a consistent tions. At D = 103 mm, model ST-WB shows high strain levels at
Rupture Index for the same material, the plastic fracture strain (the multiple locations, including the external edge of the bottom
PEEQ value that causes fracture) at various stress triaxiality condi- flange of the beam, the toe of the weld access hole, and the bottom
tions (i.e. various T values) can be obtained. These plastic fracture corner of the shear tab near the weld line, which indicates that
strain and stress triaxiality (T) pairs were then directly input into these areas are more prone to initial fracture. Similar PEEQ contour
ABAQUS for simulation of steel fracture at various stress states. It patterns are found for the other two models, but with slightly
should be noted that this is only a convenient and simplified way lower strain levels. When D increases to 150 mm, fracture has been
of simulating the fractural phenomenon of steel with the consider- induced for specimen ST-WB, and the crack is initiated from one
ation of stress triaxiality, whereas the actual triaxial performance external edge of the bottom flange of the beam. For the other
also depends on other factors such initial material imperfections two specimens, however, no fracture has been induced at this
and the direction of rolling [34]. deflection, although high strain levels have been developed at
those critical regions. When D further increases to 200 mm, com-
5.2. FE results and discussions plete fracture of the bottom flange of the beam is induced for
model ST-WB, whereas for model I-WB, two cracks are simultane-
The load–deflection responses and fractural phenomena of the ously developed near the weld access hole and the edge of the
specimens predicted by the FE models are compared with the test beam flange. For model I-W, which is a fully welded connection,
results as shown in Figs. 4–6. Good agreements are generally initial fracture is developed near the weld access hole and then a
 W. Wang et al. / Engineering Structures 106 (2016) 332–347 345

Single-sided
Fracture

(a) Local fracture of model I-W

Fracture

Fracture Fracture Fracture

(b) Initiation and development of fracture

Out-of-plane Fracture
deformation

(c) Fracture pattern for specimen ST-WB

Fig. 12. Discussion of FE predictions.

second crack is induced at the beam web. All these fractural beha- non-symmetrical development of strain is probably due to the
viour predicted by the FE models agrees very well with test single-sided connection type between the beam web and the shear
observations. tab, which also causes out-of-plane deformation of the beam web,
It is of importance to find that for model ST-WB, the PEEQ pat- as shown in Fig. 12(c). This ‘web twisting’ effect can be more signif-
tern is distributed over the width of the beam flange in a non- icant for model ST-WB than that for model I-WB because the
symmetrical manner, and this could attribute to the initiation of ‘coped length’ (i.e. the length required for weld-access hole) of
fracture at only one edge of the bottom flange of the beam. The the steel beam is longer for the former case (in order to
346 W. Wang et al. / Engineering Structures 106 (2016) 332–347

accommodate more space for the short through diaphragms). As a of the fully yielded axial resistance of the beam section. The stress
result, the stress tends to be concentrated at one edge of the beam demand is well redistributed to the bolted shear tab connection
flange for model ST-WB, which causes earlier fracture at that area and the top flange of the beam, where the successive fracture along
for the model. It is noted that as the web of model I-W is directly the bolt line enabled further deformability of the specimens with
welded to the column wall with no out-of-plane eccentricity, a significant picking up of the resisting load.
symmetrical strain pattern is developed over width of the beam From a sudden column removal point of view, however, it was
flange. In this case, the strain demand is symmetrically shared by deduced from the energy balance principle that the regained static
the beam flange, so the fracture development is also symmetrical. load resistance at the catenary stage could not promote a second
Therefore, from the perspective of initial fracture control, double- equilibrium of the system. In other words, the internal energy
sided connections between the beam web and the column stored within the deformed system could not balance the external
(e.g. double shear tabs) may be preferred, but further test evidence work done by the applied load if it caused initial fracture of
is desirable to confirm this. the bottom flange of the beam. Towards a simplified and conserva-
Generally speaking, the rationality and effectiveness the FE tive failure criterion for such connections in practical design, it
modelling strategy, especially the way of addressing the complex may be preliminarily concluded that, when a sudden column loss
issue of fracture simulation of the steel components, is verified. scenario is considered, progressive collapse can be triggered upon
Whilst parametric studies are not within the scope of the current the initial fracture of the bottom flange of the beam.
paper, the validated numerical model can form an important basis
for the future studies towards more detailed design regulations of Acknowledgements
such beam-to-SHS column connections (or similar connections
types). The research presented in this paper is supported by the
Natural Science Foundation of China (NSFC) through Grant Nos.
51008220 and 51378380. Any opinions, findings, conclusions,
6. Summary and conclusions and recommendations expressed in this paper are those of the
writers and do not necessarily reflect the views of the sponsors.
This paper has comprehensively discussed the progressive col-
lapse resistance of practical beam-to-SHS column connections that References
are widely used in modern construction. The study commenced
with three full scale tests on double-span systems subjected to a [1] Chan TM, Gardner L. Compressive resistance of hot-rolled elliptical hollow
sections. Eng Struct 2008;30(2):522–32.
typical column loss scenario. The three specimens included welded
[2] Li L, Wang W, Chen YY, Lu Y. Effect of beam web bolt arrangement on catenary
flange-weld web connection with internal diaphragms (specimen behaviour of moment connections. J Constr Steel Res 2015;104:22–36.
I-W), welded flange-bolted web connection with internal dia- [3] Ellingwood BR. Mitigating risk from abnormal loads and progressive collapse. J
phragms (specimen I-WB), and welded flange-bolted web connec- Perform Constr Facil 2006;20(4):315–23.
[4] National Institute of Science and Technology (NIST). Final report on the
tion with short through diaphragms (specimen ST-WB). The test collapse of the world trade center towers. NCSTAR 1, US Department of
results, including failure modes/sequences, load–deformation Commerce, Gaithersburg, Maryland; 2005.
responses and stress conditions, were presented in detail, and [5] General Service Administration (GSA). Progressive collapse analysis and design
guidelines for new federal office buildings and major modernization projects,
the load transfer mechanism along the entire loading process Washington, DC; 2003.
was thoroughly discussed. Comprehensive finite element (FE) [6] Department of Defense (DoD). Design of buildings to resist progressive
models were then established to enable further discussion of the collapse, unified facilities criteria, 4-023-03; 2009.
[7] Tsitos A. Experimental and numerical investigation of the progressive collapse
test results. Good agreements were generally observed between of steel frames. State University of New York; 2010.
the FE predictions and test results, especially in terms of the trend [8] Astaneh-Asl A, Jones B, Zhao Y. Progressive collapse resistance of steel building
of load dropping and regaining responses as well as the fractural floors, UCB/CEE-STEEL-2001/03. University of California at Berkeley; 2001.
[9] Song BI, Giriunas KA, Sezen H. Progressive collapse testing and analysis of a
patterns. steel frame building. J Constr Steel Res 2014;94:76–83.
Based on both experimental and numerical results, it can be [10] Sadek F, Main JA, Lew HS, et al. Testing and analysis of steel and concrete
mainly observed that all the three specimens exhibited satisfactory beam-column assemblies under a column removal scenario. J Struct Eng
2011;137(9):881–92.
ductility supply, and they shared a similar level of maximum load
[11] Karns J, Houghton D, Hong J, Kim, J. Behaviour of varied steel frame connection
resistance due to a similar mechanism of fracture initiation. The types subjected to air blast, debris impact, and/or post-blast progressive
first peak load was achieved mainly in the flexural bending stage collapse load conditions. In: Proceedings of the 2009 structures congress, USA;
(with minor catenary action developed), where the load resistance 2009.
[12] Li L, Wang W, Chen YY, Lu Y. Experimental investigation of beam-to-tubular
could reach a level which corresponded to the full (or almost full) column moment connections under column removal scenario. J Constr Steel
plastic bending resistance of the beam section. Specimen I-WB, Res 2013;88:244–55.
which employed a typical welded flange-bolted web solution, [13] Lew H, Main J, Robert S, et al. Performance of steel moment connections under
a column removal scenario. I: Experiments. J Struct Eng 2012;139(1):98–107.
showed the highest deformability against initial fracture at the [14] Lee C, Kim S, Lee K. Parallel axial-flexural hinge model for nonlinear dynamic
bottom flange of the beam. When short through diaphragms were progressive collapse analysis of welded steel moment frames. J Struct Eng
employed (i.e. specimen ST-WB), however, the ductility supply 2010;136(2):165–73.
[15] Yang B, Tan KH. Experimental tests of different types of bolted steel beam–
against initial fracture could be evidently decreased due to high column joints under a central-column-removal scenario. Eng Struct
stress levels developed at one edge of the beam flange, causing ear- 2013;54:112–30.
lier fracture at this area. It was deduced that employing double- [16] Yang B, Tan KH. Robustness of bolted-angle connections against progressive
collapse: mechanical modelling of bolted-angle connections under tension.
sided connections between the beam web and the column (e.g. Eng Struct 2013;57:153–68.
double shear tabs) may improve the ductility supply. After initial [17] Kuhlmann IU, Rölle L. Robust structures by joint ductility: COST C-26
fracture, limited load resistance was shown for specimen I-W Workshop, Prague; 2007.
[18] Rölle L, Kuhlmann U. Partial-strength and highly ductile steel and composite
which adopted fully welded connections, and marginal catenary
joints as robustness measure: Nordic Steel Construction Conference 2009,
action was developed due to quick propagation of the crack over Malmo; 2009.
the entire beam section. For the remaining two specimens adopt- [19] Huang ZH, Shi SQ, Cai LP. Experimental analysis on strength and failure modes
ing welded flange-bolted web connections, remarkable catenary of wood beam–column connections. Front Struct Civ Eng 2014;8(3):260–9.
[20] Yu M, Zha XX, Ye JQ. The influence of joints and composite floor slabs on
action was developed after initial fracture, where the maximum effective tying of steel structures in preventing progressive collapse. J Constr
axial force developed in the system could achieve more than half Steel Res 2009;66(3):442–51.
 W. Wang et al. / Engineering Structures 106 (2016) 332–347 347

[21] Alashker Y, El-Tawil S, Sadek F. Progressive collapse resistance of steel- [27] Fadden M, McCormick J. Finite element model of the cyclic bending behavior
concrete composite floors. J Struct Eng 2010;136(10):1187–96. of hollow structural sections. J Constr Steel Res 2014;94:64–75.
[22] Fang C, Izzuddin BA, Elghazouli AY, Nethercot DA. Robustness of steel- [28] Ministry of Construction of China: Code for design of steel structures (50017-
composite building structures subject to localised fire. Fire Saf J 2011;46 2003), Beijing; 2003.
(6):348–63. [29] American Society for Testing and Materials (ASTM). Standard test methods and
[23] Fang C, Izzuddin BA, Elghazouli AY, Nethercot DA. Simplified energy-based definitions for mechanical testing of steel products, ASTM Standard A370-02,
robustness assessment for steel-composite car parks under vehicle fire. Eng Philadelphia, PA; 2002.
Struct 2013;49:719–32. [30] Wang KQ, Fang C, Li GQ, Izzuddin BA. An analytical model for evaluating axial
[24] Izzuddin BA, Vlassis AG, Elghazouli AY, Nethercot DA. Progressive collapse of boundary stiffness of steel beam restrained by neighboring floor systems. Int J
multi-storey buildings due to sudden column loss – Part I: Simplified Steel Struct 2011;11(4):457–66.
assessment framework. Eng Struct 2008;30(5):1308–18. [31] ABAQUS Analysis user’s manual, ABAQUS Standard, Version 6.11; 2011.
[25] Vlassis AG, Izzuddin BA, Elghazouli AY, Nethercot DA. Progressive collapse of [32] Yang B, Tan KH. Numerical analyses of steel beam–column joints subjected to
multi-storey buildings due to sudden column loss – Part II: Application. Eng catenary action. J Constr Steel Res 2012;70:1–11.
Struct 2008;30(5):1424–38. [33] Khandelwal K, El-Tawil S. Collapse behavior of steel special moment resisting
[26] Luu NNH. Structural response of steel and composite building frames further frame connections. J Struct Eng 2007;133(5):646–55.
to an impact leading to the loss of a column. PhD Thesis. Department of Civil [34] El-Tawil S, Vidarsson E, Mikesell T, Kunnath SK. Inelastic behavior and design
Engineering, University of liege, Belgium; 2008. of steel panel zones. J Struct Eng 1999;125(2):183–93.