Journal of Constructional Steel Research 139 (2017) 69–82

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

Experimental study of an innovative modular steel building connection

Zhihua Chen a,b, Jiadi Liu b,c, Yujie Yu b,⁎, Chenhua Zhou d, Rengjing Yan d
a
State Key Laboratory of Hydraulic Engineering Simulation and Safety, Tianjin, China
b
Department of Civil Engineering, Tianjin University, Tianjin, China
c
Tianjin University Research Institute of Architechtural Design & Urban Planning, Tianjin, China
d
China Construction Steel Structure Corp. Ltd，Shenzhen, China

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

Article history: In modular steel building (MSB) built by unit-prefabricated on-site assembled construction method, connections
Received 13 September 2016 are critical parts that can strongly influence the overall structural stability and robustness of the MSB. Previous
Received in revised form 6 February 2017 MSB connections mainly use intermediate connecting plates, which may pose practical difficulties to certain
Accepted 4 September 2017
modular arrangements or installations. Thus, this paper proposed an innovative MSB connection design with
Available online xxxx
an intermediate plug-in device and a beam-to-beam bolt system as the horizontal and vertical connections,
Keywords:
respectively. This connection design can ensure convenient installation, eliminating on-site welding. Two static
Modular steel building connection uniaxial loading tests and four quasi-static cyclic loading tests were conducted on the T-shaped MSB connection
Plug-in device to explore its load transfer capacity and aseismic behavior. Results showed that gaps would form between the
Beam-to-beam bolt connection upper and the lower columns because of its two-unit-joint structure. This gap can influence the deformation
Static uniaxial loading test patterns and bending demand distributions in each unit joint. The weld quality in the unit joints was critical to
Quasi-static cyclic loading test ensure overall safety. Stiffeners can effectively strengthen the stiffness and load-bearing capacity. The deforma-
tion capacity of the connection was significantly influenced by the stiffness of floor beam–column joint and
ceiling beam–column joint as well as the relative magnitudes between them.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction
Modular construction comprises prefabricated room-sized volumetric
units constructed in a factory and installed on-site as load-bearing
“building blocks” [1]. The discrete modular units usually form a self-
supporting structure, and they are generally built in the factory to reach
a fully fitted state, complete with floors, lighting, plumbing, and heating,
before they are delivered to the site for installation [2]. The assembly-
line construction mode in the factory can effectively reduce the time
cost and wastage as well as ensure high quality control and construction
accuracy. Modular construction is adopted in North America, Japan, and in
some parts of Europe [3], mainly for low-rise houses and several high-rise
buildings, such as apartment buildings, student residences, and hotels [4].
Modular steel building (MSB) systems differ significantly from
traditional on-site counterparts in terms of detailing requirements and
construction method [5]. The “unit-prefabricated on-site assembled”
construction method have different structural requirements compared
with traditional structures, particularly in the design of connections
[6,7]. A traditional frame structure generally features a continuous
single column to which one or multiple beams are connected with
moment connections or joint connections [8]. By contrast, in MSBs,
⁎ Corresponding author at: School of Civil Engineering, Tianjin University,
Tianjin, China.
E-mail address: yujietju@tju.edu.cn (Y. Yu).
the entire structure has many sub-structures, and each structural unit
has its own frame system. In the connecting region, numerous small
beams and columns meet together, and this configuration poses new
challenges to structural design. In Fig. 1, the corner joint has 2 columns
and 4 beams, the side joints have 4 columns and 8 beams, and the inner
joint has 8 columns and 16 beams. For this structural system, each
modular unit member has to be properly connected to ensure the
transfer of lateral loads, axial forces, and bending moments, which are
generated from external loads.
MSB connections are structurally important as they can strongly in-
fluence the overall structural stability and robustness of MSBs. Annan
et al. [5] proposed a horizontally and vertically separated connection,
in which the lower and upper modular columns are partially welded
to form the vertical connection and the shop-welded angles are
field-bolted to the floor beams to form the horizontal connection.
Subsequently, they also discussed the effect of direct welding between
the strings and the floor beams on the structural design of modular
structures [2]. Park et al. [9] introduced a unit connection and an em-
bedded steel column-to-foundation connection for a modular structural
system; the unit connection utilizes a cross-shaped plate installed on a
column flange and then bolted to the beam web. Fathieh and O. Mercan
[10] once performed a seismic evaluation of MSBs. In the evaluation, the
inter-modular connections were simulated by adding an additional
vertical short column and horizontal beam. Bae et al. [11] optimized
the parameters of beam–column joints in a unit modular system

https://doi.org/10.1016/j.jcsr.2017.09.008
0143-974X/© 2017 Elsevier Ltd. All rights reserved.
70 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82
Fig. 1. Connections in modular steel building (MSB).
through basic experiments and theoretical analysis for the development
of a high-rise frame unit modular system.
Current MSB connections are horizontal or vertical connecting
plates placed at the two ends of module columns of. Access to these
connections has to be made externally to the modules, thereby pos-
ing practical difficulties for certain module arrangements. This study
aims to solve this problem by proposing a new T-shaped beam-to-
beam modular connection. Its performance was analyzed through a
series of static tests and quasi-static tests. The test results could be
used as guide for the design and actual construction of connections
for MSBs.
Fig. 2. Two types of beam-to-beam inter-modular connections.
(a) Entire structural model in MIDAS
Fig. 3. Structural model and modular unit components of the prototype office building.
2. Details of the new beam-to-beam MSB connection
The components of the new beam-to-beam MSB connection are
presented in Fig. 2. The connection has two effective connecting parts,
namely, the plug-in device to transfer the horizontal forces and the
long stay-bolt system at the beam ends to tie the upper and lower mod-
ular beams together. The modular beams and columns are all made of
cold formed rectangular steel tubes, and the small beams and columns
are connected by welding. The cover plate is welded to the upper mod-
ular floor beam. A bottom plate and an intermediate plate are welded to
both flanges of the ceiling beam in the lower module. The cover plates
protect the beam tubes from local buckling under the tension forces of
the stay bolts. This connection is suitable for a frame module, in which
the internal forces are transmitted through the unit corners. The joints
inside each modular unit are all welded. The modular frame units are
prefabricated in the factory; thus, the welding of the beams and
columns and the welding of the cover plate are accomplished in the
factory. In addition, the column ends are left open at the end of the
modular unit construction.
In this MSB connection, no additional welding process is required at
the construction site. Each well-built unit is transported to the work site
and lifted to the position above the designed location. The four plug-in
devices are placed at the upper ends of the lower modular columns.
Subsequently, the lower ends of the upper modular columns are aligned
and inserted to the four plug-in devices. The plug-in unit is made of cast
steel with two square tubes connected to both sides of an intermediate
plate. The outer diameters of the tubes are the same as the inner
diameters of the modular columns. The tubes have shrunken ends to
facilitate the alignment and provide an allowance for installation
error. Then, the upper and lower units are clasped together along the
vertical direction. Having no vertical linking mechanism, the plug-in
unit itself cannot resist the pull-up force. The long stay-bolt system at
the beam ends compensates for this deficiency. In traditional inter-
(b) Single modular unit
 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82 71

1900
360

15

50
50 100
430

80
170 25 78
20

1000

1000

20 40
16

150
100 12817

80

16
50
50 430

162
170

1900 20 78 20
15 17 128 5 5
(a) Upper modular part (b) Lower modular part (c) Cast plug-in device

Fig. 4. Dimension of test specimens.

modular connections, side plates are often welded or bolted to the beam
or column sides; thus, additional construction space is required be-
tween the adjacent modular units. However, the intermediate space
should neither be excessively large nor allowed due to architectural
and structural considerations, thereby inducing difficulties for modular
installations. By contrast, in the proposed connection design, the long
bolt can be installed from inside of the modular units, and only a
plate-thick gap exists between the upper and the lower beams. The mo-
ment is resisted only by the welded beam section of the unit connection.
Thus, a pair of diagonal stiffeners are added to strength this region. In
this study, both stiffened and unstiffened connections were studied.
3. Experimental study
3.1. Test specimens and test setup
The objective of the experiments was to study the working mecha-
nism, load bearing, and seismic behavior of the proposed MSB connec-
tion. The connection design was based on an actual modular office
building in Sino-Singapore Eco-City in Tianjin Binhai New Area. The of-
fice building was a four-story composite modular structure with a steel
frame structure on the first floor. The upper floors were entirely made of
modular units (Fig. 3(a)). The modular unit had prefabricated concrete
slabs for the floor and lightweight composite boards for the unit ceiling,
enclosures, and partitions. The section height of the floor beam was
larger than that of the ceiling beams. The concrete slab and ceiling
boards were all connected to the inside face of the floor beam through
welded angles, exerting a slight moment-bearing effect on the frame
unit. The effective structural skeleton unit is presented in Fig. 3(b).
The specimens were of the same steel grade components as the
prototype structure. The unit columns, beams, cover plates, and stay
bolts were made of Q345B steel in accordance with the Chinese
standards, whereas the plug-in devices were made of ZG35 cast steel.
The prototype connection was based on the most heavily loaded parts
of the actual building, and the test specimens were 2/3-scale versions
of the prototype connections so they can be tested in a laboratory. The
parameters of the stay bolts were determined on the basis of the
potential maximum tension in the vertical junction, and three 24 mm
diameter long snug-tight bolts were adopted. Key information about
the test specimens is given in Fig. 4 and Table 1.
The T-shape connection design was first studied. The directly down-
scaled specimens each had a 250 mm-height steel tube as the floor
beam and a 150 mm-height rectangular tube as the ceiling beam with
diagonal stiffeners. Parametric experimental tests were conducted to
explore the working mechanism and the effect of different composite
parts. The key investigated factors are as follows: (a) the loading
method, static and quasi-static performances; (b) effect of the diagonal
stiffeners; (c) dimensions of the floor beam or modular unit stiffness;
and (d) axial compression ratio. A total of two monotonic loading and
four quasi-static loading tests were performed. The specimen informa-
tion is presented in Table 1. All the specimens were made of Q345B
steel. The welds were formed by groove fusion welding. The diagonal
stiffeners were connected by fillet welding with wire ER50-6. In accor-
dance with the actual practice, only appearance inspections were
performed to ensure weld quality. Before the loading test, material
tests were performed to determine the material properties. Tensile
coupons were cut from the modular column tube, beam tube, and
diagonal stiffeners. Three specimens were prepared for each thickness
of steel sheet. The averaged data are listed in Table 2. fy is the yield
strength and the fu is the ultimate strength of steel. Material test data
indicated that the strengths of the used steel members satisfied the
design requirements.
During the structural design process, the moment-bearing strength
was determined using the theoretical yield strength of the modular
columns, that is, when the peripheral face reaches the material yielding
state. First, the two monotonic static loading tests (S1–S2) were per-
formed to verify the safety of the design method and to understand
the practical load-bearing capacity and lateral deformation capacity.
Two specimens were tested: one was the base test, which had the
modular beam welded directly to the modular column, and the other
was the reinforced connection, which had two pairs of stiffeners
added to the inner side of the upper and lower unit joints (Fig. 4).
Other member dimensions are the same for both two specimens.
Second, four quasi-static cyclic loading tests (QS1–QS4) were per-
formed to understand the seismic performance and failure mode of
the proposed modular connection under cyclic loading. In the cyclic
tests, the predicted yield strength was determined using the previously
obtained monotonic data.
All the experiments were performed in the structural laboratory at
Tianjin University. The test setup and measurement design presented
in Fig. 5. The column end loading method was adopted in the experi-
ments; the beam ends and lower column ends were pin-constrained
to simulate the inflection point boundary. The actuator with a 200 t
hydraulic jack to apply the axial load was placed at the upper end of

Table 1
Specimen information (The bold marks indicate different varying parameters.).

Specimen no. Ceiling beam/mm Floor beam/mm Modular column/mm Stiffener thickness/mm Stay bolt/mm Axial force ratio Loading method

S1 150 × 150 × 8 150 × 250 × 8 150 × 150 × 8 None 24 0.2 Static

S2 150 × 150 × 8 150 × 250 × 8 150 × 150 × 8 10 24 0.2 Static
QS1 150 × 150 × 8 150 × 250 × 8 150 × 150 × 8 None 24 0.2 Quasi-static
QS2 150 × 150 × 8 150 × 150 × 8 150 × 150 × 8 10 24 0.2 Quasi-static
QS3 150 × 150 × 8 150 × 250 × 8 150 × 150 × 8 10 24 0.2 Quasi-static
QS4 150 × 150 × 8 150 × 250 × 8 150 × 150 × 8 10 24 0.1 Quasi-static
72 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82
Table 2
Material properties of steel specimens.
Type Thickness fy fu
(mm) (MPa) (MPa)
Modular column, beam plate 8 425 575
Diagonal stiffener 16 350 510
Cast plug-in device – 330 350
the upper modular column. With its large slenderness ratio, the modu-
lar unit column was designed with stability control, and the axial force
ratio n (the ratio of compressive force N in the actual building to the
design compressive strength Nu) was generally lower than 0.3. The de-
sign compressive strength was calculated on the basis of the all-section
yielding state and was determined to be approximately 1.43 × 106 kN.
An axial force ratio of 0.2 was adopted in the test (286.272 kN axial
force), whereas QS4 took a half axial force (143.136 kN axial force)
with 0.1 axial force ratio for comparison. The top of the loading jack
was bearing against the rollers to allow for horizontal displacement.
In addition, the bottom of the jack allows for slight rotation. The upper-
most restraint was assumed a free-end constraint condition. The axial
load was applied to the designed level before the lateral loading and
maintained at the same level throughout the lateral loading.
Lateral displacement was ensured with a horizontally placed
100 t jack. During the static loading tests, the specimens were sub-
jected to monotonic loading (horizontal direction). During the
quasi-static loading tests, the specimens were subjected to cyclic
loading (horizontal direction). A special design was needed because
of the double-beam structure to ensure that the beam ends are
hinged. As shown in Fig. 5(b), the beam end was constrained with
a capping system. The rollers were placed at the top of the upper
beam, between the floor and the ceiling beams, and at the bottom
of the lower beam to allow for separated end rotation and movement
in the two modular beams.
(b) Roller at the beam end
Fig. 5. Test setup and measurement design.
Fig. 5(c) presents the displacement measurement plan, in which the
displacement meters were all horizontally mounted at the beam and
Elongation column ends. W1 measured the horizontal displacement at upper
(%) column end, which was also the loading end. W2 and W3 measured
30 the horizontal displacement at the ends of the upper floor beam
26 and lower ceiling beam, respectively. The displacement and rotation
22.5 components can be indirectly derived from the displacement measure-
ments. In the uniaxial loading test, the lateral displacement was applied
away from the reaction wall. The loading protocol in the quasi-static
loading test was drawn from the “Regulations of seismic test method
(JGJ101-96)” in China [12]. The lateral loading method was controlled
by load before the specimen yielded, with a load increment of 1/2 of
the predicted yield strength. When the specimen yielded, the lateral
loading method will be changed to displacement controlled, with the
lateral displacement increment the same as the practical yield displace-
ment, Dy. Each displacement cycle was repeated twice, and the test was
terminated when the axial load could not be maintained or the lateral
force decreased below 85% of the maximum load.
3.2. Results of static uniaxial loading test
Specimen S1 was the unstiffened connection. As the lateral load
reached 89.8 kN and the lateral displacement reached 45 mm, gap
between the lower column and the plug-in device became visible, and
a slight outward deformation was observed in the connecting region
of the lower column (Fig. 6(a)). As the lateral load further increased,
the gap widened. When the lateral displacement at the upper column
end reached 78.7 mm (lateral load was 111.5 kN), a loud sound was
heard, and the test was terminated. A weld fracture was observed in
the welded joint between the ceiling beam and the lower column. The
fracture started from the bottom face and propagated upward by
approximately 60 mm through the two side welds (Fig. 6(b)). In this
study, yielding was determined to have occurred when an obvious
tangent stiffness change occurred in the lateral force–displacement
(a) Test setup
(c) Measurement locations
 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82
(a) Gap in the plug-in device
Fig. 6. Failure mode of S1.
curves, and the lateral force at this turning point was called the yield
strength of the connection. Specimen S1 did not demonstrate apparent
yielding behavior throughout the loading period, and when the
specimen exhibited brittle failure, the lateral load was still increasing.
Therefore, for this unstiffened connection, the weld in the unit
beam–column joint was the critical part that should be given attention
to ensure the weld quality and connection capability.
Unlike S1, specimen S2 had diagonal stiffeners to strengthen the
unit joint, as shown in Fig. 7. When the lateral displacement reached
31.4 mm (lateral load was 128.8 kN), the turning point appeared in the
load–displacement curve, and the specimen entered the elastoplastic
stage. When the lateral displacement reached 51.9 mm, the lateral load
reached 162.6 kN, and a gap was observed between the upper column
and the plug-in device, but no strength reduction was presented.
When the lateral load reached 199.4 kN, the weld between the diagonal
stiffener and the cover plate exhibited brittle failure accompanied by a
loud sound and a sudden drop in lateral strength (Fig. 7).
Fig. 8 presents the lateral load–displacement relationships for S1 and
S2. As shown in Fig. 8, the lateral load continuously increased before
brittle failure occurred in both specimens. The design lateral force
strengths when the lower modular column (the larger moment arm
from the column end to the junction surface leads to a smaller lateral
force strength at the lower modular column) reached the edge yield
bending moment (dotted lines) and all-section yield bending moment
(dashed lines) were marked in the plot. In the plot, the lateral force
strength represented by pink lines were calculated with the design
yield strength of 315 MPa, whereas those represented by blue lines
were calculated with the material yield strength presented in Table 2.
For the unstiffened connection S1, when the fracture occurred, the
lower column could barely reach the practical edge yielding state, and
upper column was still within the elastic stage. The design material
Fig. 7. Failure mode of S2.
73
(b)Weld fracture failure
strength was conservative; thus, the unstiffened connection could
reach the design yield strength, but the safety margin was minimal. By
contrast, with the test material strength, the connection strength
would be overestimated; consequently, plasticity and ductility could
hardly be completely developed before the fracture occurred. For
the stiffened connection S2, the lateral force surpassed the practical
all-section yield strength (125.4 kN) of the lower column, indicating
that plasticity was developed in the modular columns. Therefore, the
vertical stiffener can effectively strengthen the connection and protect
the unit joint, and the column bending controlled design method is
appropriate for the stiffened connection. Both connections failed with
a weld fracture, which is an undesirable failure mode in aseismic design.
For this connection type, the weld quality of the unit beam–column
joint is important.
Fig. 9 presents the lateral displacement measurements at the beam
and column ends, and the ultimate deformation state of each connec-
tion. The displacement difference between W1 and W2 (W1 − W2)
represents the inclination of the upper column, whereas W3 represents
the inclination of the lower column. In a traditional frame connection
test, the lateral displacements of the upper and lower columns are
similar, and W1 − W2 should be approximate to or larger than W3.
However, the test connections consisted of two joint units, and the
test results presented different deformation degrees between these
joint units. For both connections, the lateral difference between W1
and W2 (W1 − W2) was dramatically smaller than W3, indicating a
larger drift angle in the lower joint than that in the upper joint. This con-
clusion could be supported by the ultimate states of both specimens,
that is, the lower joint had a larger bending rotation than the upper
joint had. This deformation pattern could be attributed to the gap
between the two joint units. The stiffened joint S2 had higher unit
joint bending stiffness; as a result, a larger connection rotation demand
74 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82

descending branch, and the gap between the modular column and the
200 plug-in device widened. During the subsequent reverse loading process,
the gap could not be completely eliminated; the tearing extent of the
With test strength lower weld continued to worsen; and the lateral load dropped back
to 87.97 kN, which is approximately 83% of the maximum load (the
160
Lateral load (kN)

125.4kN
120 106.11kN
92.94kN
80
78.64kN
With design strength
40
0 beam–column joint, and the P–Δ relationship began showing a decreas-
0 20 40 60 80
Lateral Displacement (mm)
Fig. 8. Load–displacement relationship. (For interpretation of the references to color in this
figure, the reader is referred to the web version of this article.)
was taken by the gap between the column and the plug-in device, and
consequently, the degree of inequivalent joint rotation was higher
than that of the unstiffened one. This lateral deformation pattern
could lead to a gradual decrease in the bending demand from the
lower to the upper floors in an MSB under lateral loads.
3.3. Results of the quasi-static loading test
3.3.1. Specimen performance
Initially, the P–Δ relationship exhibited a nearly linear variation
trend, which indicated the elastic state of the specimen. When the
load reached 59.41 kN, the specimen exhibited plasticity; accordingly,
an obvious turning point was evident in the load–displacement curve
(Dy = 25.76 mm). Subsequently, the loading method changed to
displacement controlled, with a displacement load increment equal to
Dy (i.e., 25.76 mm). During the first cycle of 2Dy, the lower column
and plug-in device opened, and a gap was induced (Fig. 10(a)).
A micro-crack appeared in the root weld at the top of the ceiling
beam–column joint (Fig. 10(b)); henceforth, the lateral strength
increase decelerated (Fig. 10(d)). When the lateral load direction
reversed, the gap and crack closed. In the first cycle of 3Dy, the lower
weld in the ceiling beam–lower column joint was torn, and the lateral
strength and overall stiffness considerably decreased (Fig. 10(c)).
As the loading proceeded, the P–Δ relationship began exhibiting a
maximum lateral force during the hysteresis loading). During the sec-
ond cycle of 3Dy, the column flange in the upper part of the ceiling
beam–column joint was completely torn, and a sunken deformation
was produced at the bottom of the column flange. The test was termi-
nated at the end of 3Dy because of the remarkable strength reduction.
Specimen QS2 had a yield strength of 100.39 kN at a lateral displace-
ment, Dy, of 43.54 mm. Then, the displacement-controlled loading
method was applied. During the second cycle of 2Dy, the vertical weld
in one diagonal stiffener in the floor beam–column joint fractured
(Fig. 11(a)). In addition, during the second cycle of 2.5Dy, the other stiff-
S1 ener in the floor beam–upper column joint exhibited brittle failure as
well (Fig. 11(b)); as a result, the lateral strength suddenly dropped.
S2 A slight weld tearing was observed in the upper portion of the floor
100 ing tendency (Fig. 11(d)). During the first cycle of 3Dy, the lateral load
dropped to 88.67 kN, approximately 75.07% of the maximum load. The
weld tearing became more apparent (Fig. 11(c)), and the connection
strength and lateral stiffness significantly decreased. In the subsequent
reverse loading, the floor beam–column joint was completely torn;
thus, the test was terminated.
For specimen QS3, the yield strength and corresponding lateral
displacement were 80 kN and 22 mm (Dy), respectively. During the
first cycle of 4Dy, slight local buckling occurred in the upper stiffener
(Fig. 12(a)), and the upper weld in the ceiling beam–column joint
fractured (Fig. 12(b)). During the latter reverse loading process, the
cracks closed, and inward deformation and local buckling were
occurred in the lower modular column flange. During the first cycle of
5Dy, the specimen strength and lateral stiffness obviously decreased,
and the load–displacement relationship entered a decreasing stage
(Fig. 12(d)). At the end of this cycle, the column end load dropped to
134 kN, which is approximately 85% of the maximum load. The final
load in the second cycle of 5Dy was 163 kN, which is approximately
86% of the maximum load. The test was terminated because of the
severe buckling in the column flange (Fig. 12(c)).
The load–displacement relationship was consistently linear for spec-
imen QS4 before the load reached 80 kN (Dy = 23 mm). Subsequently,
the specimen yielded, and the loading process changed to displacement
controlled. During the first cycle of 4Dy, a slight local buckling occurred
in the lower stiffener between the ceiling beam and the modular
column (Fig. 13(a)). An inward deformation was also observed in the
lower column flange (Fig. 13(b)). During the first cycle of 5Dy, the
local buckle in the column flange increased in severity, and the column
flange exhibited tearing failure (Fig. 13(c)). The load–displacement
relationship showed a declining trend when the lateral load dropped

100 100
W1 W1
Lateral Displa cement (mm)

W2 W2
Lateral Displa cement (mm)

80 W3 80 W3

60 60

40 40

20 20

S1 S2
0 0
0 30 60 90 120 150 180 210 0 30 60 90 120 150 180 210
Lateral load (kN) Lateral load (kN)

(a) S1 (b) S2
Fig. 9. Lateral displacement pattern.
 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82 75

(a) Micro-crack and gap (b) Weld tearing

160 Tear at column flange and widened gap
120 Micro-crack and gap
Yield

lateral load (kN)

80
40
0
-40
-80
-120
-160 Weld tearing
QS1
-200
-120 -90 -60 -30 0 30 60 90 120
Displacement (mm)
(c)Tear in the column flange and widened gap (d) Lateral force–displacement curves

Fig. 10. Testing process of specimen QS1.

to 137 kN (approximately 78.73% of the 174 kN maximum load). The
test was terminated due to severe failure.
3.3.2. Failure mode
The applied lateral displacement was spread into different parts of
the connection given the assembled construction characteristic of the
modular connection. As shown in Fig. 14, the deformation and joint
rotation demand consisted of three main parts, namely, the rotation
deformation α in the upper modular unit joint, the rotation deformation
β in the lower modular unit joint, and the rotation γ created with the
gap between the upper/lower units. This deformation component can
be indirectly reflected by the inclinations of the upper and lower
modular columns and their difference. During the cyclic loading tests,
the lateral displacements were measured along the vertical columns

(a) Facture in an upper stiffener (b) Facture in the other upper stiffener
160 Crack at column flange
Facture at one upper stiffener
120
80 Yield
Lateral load (kN)

40
0
-40
-80
-120 Facture at the other upper
stiffener and strengh drop
-160
QS2
-200
-120 -90 -60 -30 0 30 60 90 120
Displacement (mm)
(c) Crack in the column flange (d) Lateral force–displacement curves

Fig. 11. Testing process of specimen QS2.

76 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82

(a) Local buckling in the upper stiffener (b) Facture in the lower joint
160 Facture,Upper stiffenfer bucking
Yield
120

Lateral load (kN)

80
40
0
-40 Local bucking
-80 at bottom column
-120
-160
QS3
-200
-120 -90 -60 -30 0 30 60 90 120
Displacement (mm)
(c) Local buckling in the lower column (d) Lateral force–displacement curves

Fig. 12. Testing process of specimen QS3.

(W1 at the column top had a height of 2 m; the heights of W2 and W3
were 1.5 and 0.5 m, respectively). Subsequently, the inclination degrees
of the upper and lower columns could be derived from the lateral dis-
placement measurements. Specimen QS1 had no reinforcing stiffeners.
For this specimen, the floor beam section was larger than the lower ceil-
ing beam; the bending stiffness was low for both unit joints, but the
lower joint was less than the upper joint. Moreover, the unit joint
would have a large bending deformation degree under a small lateral
load (Fig. 10(d)). During the pushing process, a gap formed between
the upper/lower column and the plug-in device; as a result, a rotation
difference, γ, and an inconsistent bending demand between the upper
and lower unit joints were yielded. In QS1 in Fig. 15(a) for example,
the lower column had a larger inclination than the upper one had dur-
ing the pushing process of 2Dy. This difference indicated a large defor-
mation demand, which leads to the early weld fracture in the lower
unit joint.

(a) Inward deformation in the column (b) Local buckling in the lower column

160 Slight local buckling

and inward deformation
120
80 Yield
lateral load (kN)

40
0
-40 Tearing failure Severe local buckling
-80
-120
-160
QS4
-200
-120 -90 -60 -30 0 30 60 90 120
Displacement (mm)

(c) Tearing failure (d) Lateral force–displacement curves

Fig. 13. Testing process of specimen QS4.

 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82 77

Fig. 14. Deformation demand in each part.

Specimens QS2–QS4 all had diagonal stiffeners reinforcing the unit
joint, and the overall bending stiffness values of these specimens were
generally higher than that of QS1 (Figs. 10–14). Specimen QS2 had a
small floor beam with a section the same as those of the ceiling beam
and modular columns. Compared with QS1, QS2 had the same bending
stiffness values for the upper and lower unit joints; thus, the inclination
degrees of the two columns were similar, as shown in Fig. 15(b).
QS3 also had reinforcing stiffeners, but its floor beam is larger than
that of QS2; as a result, for QS3, the joint stiffness in the upper floor
beam joint was high, and the bending rotation demand on the
lower ceiling beam joint was large (unit joint had low bending
stiffness). This inequivalent deformation pattern is obviously
reflected in Fig. 15(c). During the pushing process, the upper and
lower columns initially had similar inclinations. However, with the
development of plasticity and the local buckling in the unit joint, the
bending rotation was mainly taken by the gap in the intermediate
plug-in device and the joint deformation in the lower joint; as a result,
a large inclination in the lower column and a nearly vertical state for
the upper column occurred in 5Dy. During the pull loading process,
the intermediate gap and its influence were weak due to the retention
effect of the beams, and then the lateral deformation demand was
shared by the two unit joints, as reflected in the nearly similar

Fig. 15. Deformation demand in each part.

78 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82
Fig. 16. P–Δ hysteretic relationship.
inclination degrees between the two columns. QS3 and QS4 had the
same connection structure; thus, their failure modes and deformation
patterns were similar. For both specimens, with the occurrence of
local inward buckling in the lower column flange, crack formed in the
lower joint weld, and column inclination degrees were inequivalent
during the pushing process.
3.3.3. Hysteretic performance and skeleton curve
Fig. 16 presents the hysteresis loop of each inter-modular connec-
tion. Fig. 17 presents the envelop curves and the skeleton curve normal-
ized to the yield load. Table 3 lists the mechanical parameters of each
specimen. In the table, Py and Δy are the yield strength and the
corresponding lateral displacement, respectively. These parameters,
which were derived from the skeleton curves by the energy equivalent
area method, were different from the observed yield strength and yield
strain during the tests. Furthermore, Pmax and Δmax denote the maxi-
mum lateral force during the hysteresis process and its corresponding
lateral displacement, respectively; Pu and Δu are the ultimate load or
the lateral force upon failure and its corresponding lateral displacement,
respectively.
The skeleton curves show that the unstiffened connection QS1 had
the least bending stiffness and load-bearing capacity, but it had a
relatively good connection ductility. QS1 and QS2 had similar connec-
tion stiffness values, but the latter had a larger load-bearing capacity,
the strength degradation in QS2 was also apparent. QS3 had higher con-
nection stiffness, better load-bearing capacity, and energy dissipation
performance than QS1 and QS2. A comparison of QS1 and QS3 indicated
that the diagonal stiffeners helped improve the connection stiffness
and load-bearing capacity. Although a large floor beam structure also
strengthened the connection, it could also reduce connection deforma-
tion capacity and ductility. QS3 and QS4 had similar hysteretic perfor-
mances, indicating that a certain magnitude of axial load variation
may not have much influence on lateral strength but may cause early
failure. A large axial load could help reduce the deformation extent
Fig. 17. Skeleton curves.
under the same lateral load, as reflected in the slight rise in the load–
displacement curve. However, this increase effect was weak. QS1–QS4
all entered the plastic stage and presented obvious strength reduction
periods, even for those that fractured.
Ductility ratio of the specimens refers to the ratio of ultimate
displacement to the yielding displacement and the non-elastic deforma-
tion capacity without a significant reduction in carrying capacity. This
parameter can be calculated by a hysteresis curve. The displacement
ductility factor is expressed as μ = Δu / Δy. Experimental results
showed that the ductility factor were all located between 2.09 and
2.79 (all higher than 2), indicating the reasonable deformation capacity
after yielding of the specimens. During the pushing process, a gap
formed between the upper and lower columns, and this gap partly
contributed to the connection rotation and, consequently, to the reduc-
tion in the lateral bearing capacity. As a result, the ultimate load/yield
load ratios in the reverse direction (pulling) were all larger than those
in the positive direction (pushing). With the adoption of fillet welding
and insufficient weld quality control measures, QS1–QS4 all failed via
weld fracture or tearing. However, even under this kind of failure, the
Pmax/Py ratios were all between 1.19 and 1.43, indicating that plasticity
was already developed in the connections. Some connections even
failed in their increasing strength state. Without weld fractures or
with good-quality butt welds, the connections may have achieved
higher lateral bearing capacity. Thus, that a certain margin of safety
may still be ensured under large story drift conditions.
3.3.4. Energy dissipation
The inelastic deformation of a connection facilitates energy dissipa-
tion in the event of an earthquake through hysteretic behavior, thereby
reducing the transmitted energy to other structural elements. The
energy absorbed by the deformation of the specimen can be measured
as the areas enclosed by the force–displacement hysteresis loop. The
force–displacement hysteresis loop, which is the total plastic work per-
formed by the specimen, is an important indicator of energy dissipation.
 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82 79

Table 3
Mechanical parameters of each specimen.

Specimen Loading direction Py (kN) Δy (mm) Pmax (kN) Δmax (mm) Pu (kN) Δu (mm) Pmax/Py μ

QS1 (+) 64.74 31.01 82.15 56.84 77.04 80.4 1.27 2.60
(−) 72.28 29.28 103.13 61.48 89.38 81.93 1.43 2.79
QS2 (+) 92.92 48.54 117.94 74.06 89.908 115.06 1.27 2.37
(−) 100.39 43.54 138.04 84.95 134.51 110.44 1.37 2.52
QS3 (+) 134.01 41.51 159.72 80.32 133.45 100.17 1.19 2.41
(−) 134.34 41.06 184.93 80.13 176.82 100.37 1.38 2.44
QS4 (+) 110.72 39.09 142.62 60.45 140.50 80.87 1.29 2.09
(−) 120.43 44.05 169.24 80.26 141.03 93.60 1.41 2.13

The energy dissipation factor, E, and equivalent damping coefficient, he,
in the last hysteresis loop were used to describe the energy dissipation
performance. E can be calculated as the ratio of the practical dissipated
energy to the elastic potential energy at the maximum load amplitude.
The detailed calculation method is provided in [13], and a large E or he
value indicates a good energy dissipation capacity and good aseismic
performance. The energy dissipation capacity of the specimens stably
increased with increasing displacement amplitudes in the hysteresis
loops up to the failure point, and the factors in the last loop are given
in Table 4. The equivalent damping coefficients, he, of all the specimens
exceeded 0.2. A comparison of the data in Table 4 and those of concrete
and steel-reinforced concrete connections [14] shows that the energy
dissipation capacity of each of the assembled modular connections is
approximate to that of steel-reinforced concrete connection, whose
equivalent damping coefficient is approximately 0.3, and more than
twice as high as that of the concrete connection, whose damping
coefficient is only approximately 0.1. Thus, the proposed connection
design can offer sufficient energy dissipation capacity. The factors of
QS2 and QS4 were relatively low due to the early crack failure.
3.3.5. Stiffness degradation
Stiffness (K) is defined as the ratio of the averaged maximum lateral
load to the corresponding averaged lateral displacement. Therefore,
stiffness represents the averaged secant stiffness of the specimen at
different displacement levels. The stiffness degradation factor (ξ) is
the ratio of the secant stiffness at each loading amplitude of the initial
stiffness (K0). Fig. 18 indicates the stiffness degradation process of
each specimen during the cyclic loading. Stiffness degradation mainly
occurred after the specimen began to yield, and the drop in the stiffness
was mainly attributed to the cumulative plastic deformation. The con-
nections presented different stiffness degradation patterns between
the two directions. The overall stiffness of the connections decreased
slowly in the early stage of pushing process, and the degradation rate
accelerated during the large displacement loading steps. During the
pulling process, the specimens presented a rapid stiffness reduction
after yielding, but as the lateral displacement loads increased, the reduc-
tion decelerated. These different performances were mainly attributed
modular members, intermediate connecting device, and long stay-bolt
systems. Contact CONTA174 and TARGE170 elements were set at the
surface-to-surface contacts between the plug-in device and the modular
tubes to simulate the assembled relations and between the long stay
bolts and the cover plates to explore the bolting and sliding conditions.
Three linear kinematic material models were adopted for the steel ma-
terials, with the critical points defining the stress–strain relationship de-
termined using the measured data in Table 2.
A comparison of the lateral load–column tip displacement hysteresis
curves is presented in Fig. 19. The elastic stiffness in each loading
cycle was slightly higher in the FE model than in the experiments.
Nevertheless, the strength development and high stress distribution
and development obtained using the numerical model were consistent
with the test results. The FE model also demonstrated the asymmetric
strength development between the pulling and the pushing conditions.
The von Mises stress distributions under 5Dy lateral displacement
loading are presented in Fig. 20. The measured strain data and variation
in each load level of all the four cyclic loading tests are plotted in Fig. 21.
The deformation pattern and stress distribution in Fig. 20 indicated that
under large lateral displacement loading, high stress demands were
mainly concentrated in the lower joint region, especially in the diagonal
stiffeners and their junction with the lower modular columns. Most
deformation occurred in the lower column and local buckling was
presented; this feature corresponded to the failure mode of QS3 in
Fig. 12. The stress level in the beam end joints stayed in a low level,
indicating the protective effect of the diagonal stiffeners. The relatively
separate components and gap between the upper and the lower
modular parts relieved the bending deformation transfer to the upper
modular unit. The intermediate gap was also large during the pushing
process and small during the pulling conditions, thereby partly contrib-
uting to the asymmetry. Moreover, the protective effect of the cover
plates and the strengthening function of the stay bolt resulted in the
formation of a region in the beam-to-beam connecting region and
control of the stress states within a low level.
Stiffness degradationfactor ( )

to the gap formed in the inter-floor joint region.

1.0
4. Numerical analysis
0.8
A finite element (FE) model was established for specimen QS3 using
ANSYS software to understand the internal stress distribution in the
MSB connection components. For a refined and accurate simulation, 0.6
SOLID95 solid elements were used for all the steel parts, including the
QS1
QS2
Table 4
0.4
QS3
Energy dissipation factors and equivalent damping coefficients.
QS4
Specimen Energy dissipation factor, E Equivalent damping coefficient, he 0.2
QS1 1.79 0.29 -120 -90 -60 -30 0 30 60 90 120
QS2 1.29 0.21 Displacement (mm)
QS3 1.64 0.26
QS4 1.24 0.20
Fig. 18. Stiffness degradation performances of the four specimens.
80 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82
Fig. 19. Von Mises stress distribution in QS3.
Fig. 21 presents the strain variations of the different specimens at the
same scale. Gauges 1–6 recorded the strains in the outer column flanges.
The strain values remained low compared with those of QS3. Gauges
7–8 corresponded to the lower column sides. These gauges recorded
dramatic high strain development, which corresponded to the high
stress concentration on the lower column sides in Fig. 20 and indicated
the considerable internal force shared between the diagonal stiffeners
and transmitted to the column side. The specimens all showed a low
strain levels at gauges 1–4 mainly because of the disconnected design
between the upper and lower columns. For QS1, which did not have
stiffeners, gauges 5–6 presented high strain values because of the bend-
ing effect in the lower column. Gauges 7–8 presented low strain values
in the early stage because they were located far from the ceiling beam.
The sudden strain increases during the 3Dy loading process (QS1) indi-
cated the high stress demand caused by the weld fracture failure. By
contrast, the strain values of QS2 and QS4 at gauges 7–8 were lower
than those of QS3, indicating the sharing of bending deformation from
beam bending or large gap magnitude and less bending demand on
the lower column.
Fig. 21(d) shows the internal force conditions at the beam ends and
in the joint region. QS1 had high strain demand on both sides of the
beam ends, presenting a typical bending stress distribution in moment
joints. As a result of the smaller section in the ceiling beam, the internal
stress in the joint region of the lower column was higher than that of
upper column. QS3 and QS4 had the similar strain distributions: high
Fig. 20. Von Mises stress distribution in QS3.
strain demand on the diagonal stiffeners and comparatively low strains
at the beam ends. QS1, QS3, and QS4 all had large floor beam sections;
as a result, high strains concentrated in the joint region of the lower
column, particularly at the positions of gauges 17 and 18. By contrast,
the intermediate region (gauges 16–17) of QS2 had relatively high strain
values, which indicated the vulnerability of this region. QS2 had a small
floor beam section; as a result, a low stiffness leads to a high deformation
demand. Gauge 13 of QS2 also displayed a high strain, particularly during
the pull loading process (minus displacement). All these strain results
implied that modular beams (floor and ceiling beams) and the corre-
sponding unit joints all presented independent and individual bending
behavior. Furthermore, in designing modular connections, one must
consider not only the safety factor of each unit joint but also the relative
stiffness between the adjacent unit joints and their coupling effect.
5. Conclusion
In this study, a new MSB connection design is proposed. The connec-
tion has a plug-in device to transfer the horizontal loads and a beam-to-
beam bolt as the vertical connection. This MSB connection does not re-
quire an additional welding process at the construction site, thereby
providing excellent installation convenience. Two static uniaxial loading
tests and four quasi-static loading tests were conducted under cyclic
loading to understand the mechanical properties and aseismic perfor-
mance of the T-shaped MSB connection. The following conclusions are
drawn from the test results:
(1) The two static-loading specimens failed with fracture occurring in
the unit joint welds. The weld in unit beam–column joint was the
critical part that should be given attention to ensure weld quality
and connection capability. The diagonal stiffeners effectively
strengthened the connection and protected the unit joint. The
T-shaped MSB connection consisted of two joint units. Under
lateral loading, a gap formed in the intermediate plug-in device,
and this gap resulted in different bending rotation degrees
between the upper and the lower unit joints. In MSBs, this lateral
deformation pattern could lead to a gradual decrease in the
bending demand on the unit joints from the lower to the upper
floors.
(2) The T-shaped MSB connection showed different deformation pat-
terns between the pushing and pulling loads because of the possi-
ble gap between the upper and the lower columns. Different
deformation and bending rotation demands were also induced.
All the four hysteretic specimens failed with fractures, but they
still presented reasonable energy dissipation capacity, post-
yielding deformation capacity, and connection ductility.
Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82 81

Fig. 21. Strain variation during cyclic loading.

82 Z. Chen et al. / Journal of Constructional Steel Research 139 (2017) 69–82
(3) The strain distributions revealed that the unit beams and joints
all presented independent and individual bending behavior
instead of ideally working together as a composite large beam
or united connection. The deformation capacity and mechanical
properties of the proposed MSB connection were significantly
influenced by the relative stiffness between columns and
between the floor beam and ceiling beams. Thus, stiffness should
be in connection design.
6. Future work
In this study, the desirable features of the proposed MSB connection
have been presented through loading tests. However, the present tests
only explored the behavior of a corner connection under unidirectional
bending. Thus, in future studies, the biaxial bending performances or
behavior of interior connections consisting of two- or three-direction
unit joints should be investigated.
Acknowledgements
This research was sponsored by the National Natural Science
Foundation of China (Grant No. 51708402), Project of Tianjin Urban &
Rural Construction Commission (2013E3-0028) and China Postdoctoral
Science Foundation (2016M590202).
References
[1] P. Lawson, M. Byfield, S. Popo-Ola, P. Grubb, Robustness of light steel frames and
modular construction, Proc. Inst. Civ. Eng. Struct. Build. 161 (1) (2008) 3–16.
[2] C.D. Annan, M.A. Youssef, M.H. El-Naggar, Effect of directly welded stringer-to-beam
connections on the analysis and design of modular steel building floors, Adv. Struct.
Eng. 12 (3) (2009) 373–383.
[3] A.D. Gerald Staib, Markus Rosenthal, Components and Systems: Modular
Construction Design, Structure, New Technologies, Birkhauser Architecture,
Berlin, 2008.
[4] R.M. Lawson, R.G. Ogden, R. Bergin, Application of modular construction in high-rise
buildings, J. Archit. Eng. 18 (2) (2012) 148–154.
[5] C.D. Annan, M.A. Youssef, M.H. El Naggar, Seismic vulnerability assessment of
modular steel buildings, J. Earthq. Eng. 13 (8) (2009) 1065–1088.
[6] R.O. Mark Lawson, Chris Goodier, Design in Modular Construction, CRC Press, 2014.
[7] R.M. Lawson, J. Richards, Modular design for high-rise buildings, Proc. Inst. Civ. Eng.
Struct. Build. 163 (3) (2010) 151–164.
[8] Y. Wang, X. Yan, T. Zhou, Seismic behavior of SCFT column-H section steel beam
plane frame, Tianjin Daxue Xuebao (Ziran Kexue yu Gongcheng Jishu Ban)/Journal
of Tianjin University Science and Technology 48 (2015) 9–16.
[9] K.-S. Park, J. Moon, S.-S. Lee, K.-W. Bae, C.W. Roeder, Embedded steel column-to-
foundation connection for a modular structural system, Eng. Struct. 110 (2016)
244–257.
[10] A. Fathieh, O. Mercan, Seismic evaluation of modular steel buildings, Eng. Struct. 122
(2016) 83–92.
[11] Sang Sup Lee, K.W. Bae, Keum Sung Park, Sung Yub Hong, An experimental
evaluation of structural performance for the beam to column joints in unit modular
system, J. Korean Soc. Steel Constr. 25 (2) (2013) 255–265.
[12] (JGJ101-96), Regulations of Seismic Test Method, China Building Industry Press,
Beijing, 1996.
[13] Y. Qin, Z. Chen, Q. Yang, K. Shang, Experimental seismic behavior of through-
diaphragm connections to concrete-filled rectangular steel tubular columns,
J. Constr. Steel Res. 93 (2014) 32–43.
[14] T.H. Zhou, B.K. He, G.J. Chen, C.W. Wei, Y.M. Shan, Experimental studies on seismic
behavior of concrete-filled steel tubular column and steel beam joints under cyclic
loading, J. Building Struct. 25 (1) (2004) 9–16.