Engineering Structures 175 (2018) 776–789

Contents lists available at ScienceDirect

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

Experimental investigations of vertical post-tensioned connection for T

modular steel structures
⁎
Rafaela Sanchesa, , Oya Mercana, Brent Robertsb
a
Department of Civil Engineering, University of Toronto, Toronto, ON M5S 1A4, Canada
b
Jerol Technologies Inc., Aurora, ON L4G 4B6, Canada

A R T I C LE I N FO A B S T R A C T

Keywords: Modular construction is an off-site construction technique. In this method, structural volumetric modular
Modular steel structure components are produced in a factory and assembled on-site to form a larger, permanent building. Typical
Post-tensioned connection vertical connections of modular steel buildings (MSBs) are provided by on-site welding. Welding may interfere
Cyclic displacement experiments with the finishing of the modules and also when several modules are placed together at a given floor level
Hysteretic behavior
complete access for welding is compromised. As an alternative to on-site welding, the present paper proposes a
Energy dissipation
Lateral stiffness
new vertical post-tensioned (PT) connection for MSBs. This connection is comprised of a post-tensioned threaded
Strain distribution rod installed inside hollow structural sections (HSS) columns and a steel box placed between two modules. In
order to evaluate the general and the seismic performance of the proposed connection, eight quasi-static cyclic
loading tests were performed in T-shaped subassemblies. A combination of three different steel boxes and three
initial post-tensioning loads levels were considered. Additionally, two quasi-static cyclic loading tests were
performed using standard welded connections. No local buckling was observed in any of the specimens and no
welding fractures occurred up to 3% drift demand. Results indicated that in comparison to the welded con-
nection the proposed PT connection have similar lateral stiffness and strain distribution, and a higher cumulative
energy dissipation capability. Therefore, the proposed connection has the potential to eliminate on-site welding
in the assembly of the modules while providing the lateral resistance required.

1. Introduction relatively recent. Annan et al. [1] performed experiments to analyse a

As an alternative to traditional on-site construction, in modular
construction volumetric (3-dimensional) modules are pre-fabricated at
a factory. They are then transported to the site and assembled together
to form a permanent structure. Square hollow structural sections (HSS)
and wide flange (W) sections are commonly used for columns and
beams of the modules, respectively [1,2]. Due to transportation re-
strictions, the dimensions of the modules are usually in the range of
3.5–5 m in width, 12–18 m in length, and 3–4 m in height [3]. Modular
construction is ideal for buildings with repeatable units which can be
modularized more efficiently such as hotels, multifamily dwellings,
schools, hospitals, offices, prison and military facilities. Since the
modules are generally finished in a factory with a controlled environ-
ment, modular construction is known to improve accuracy and quality;
and it has the potential to reduce waste material and on-site con-
struction time [4,5].
Although there has been an increased attention to modular steel
buildings (MSBs) [6,7], studies addressing their dynamic behavior are
MSB single frame under cyclic loading and compared the results with a
regular steel frame. Both frames developed stable and ductile behavior
up to high drift levels. In a numerical study, Annan et al. [8] performed
nonlinear pushover analyses for MSBs with 2,4 and 6-stories, where a
considerable overstrength was observed. More recently, Fathieh and
Mercan [2] conducted numerical dynamic analyses for a 4-story MSB
and reported a high base shear capacity. This is attributed to the larger
number of columns that typically exist in modular structures in com-
parison to regular steel buildings. All the above-mentioned studies
considered typical welded vertical connections. Gunawardena [6] nu-
merically analyzed the seismic behavior of a 10-story MSB with bolted
connection and with some modules stiffened by concrete walls. Results
showed that column hinge formation is a concern under severe ground
motions.
Typically, the vertical intermodular connection is provided by on-
site welding. However, once one module is placed next to the other, the
access to weld the columns is reduced, which may cause independent
rotation between modules and may lead to an undesirable soft-storey

⁎
Corresponding author at: Department of Civil Engineering, University of Toronto, 35 St. George Street, Toronto, ON M5S 1A4, Canada.
E-mail address: rafaela.sanches@mail.utoronto.ca (R. Sanches).

https://doi.org/10.1016/j.engstruct.2018.08.049
Received 27 March 2018; Received in revised form 7 July 2018; Accepted 17 August 2018
0141-0296/ © 2018 Elsevier Ltd. All rights reserved.
R. Sanches et al.
response [1]. Alternatives to typical on-site welded connections have
been investigated for both modular steel structures and regular steel
moment resisting frames (MRFs). Horizontal post-tensioned connection
for regular steel MRFs, initially proposed by Ricles et al. [9,10] and
further studied by Garlock et al. [11,12], is obtained by post-tensioning
the beams to the columns using high strength steel strands. This con-
nection eliminates welding between the beam and the column. It was
concluded that the damage in the beams and residual drifts after an
earthquake event were reduced. The initial stiffness of the connection
was found to be similar to a welded connection. Lin et al. [13] in-
vestigated seismic behavior of a new structural system that consists of
high strength built up columns and bolted beam-column connections.
Cyclic displacement tests were performed in four full-scale specimens. It
was concluded that local distortion was avoided in the specimens with
stiffened connections, and the column remained elastic. Vertical post-
tensioned connection was also proposed to improve structural seismic
performance [14]. In this study, the cyclic response of a post-tensioned
column base connection of steel frame was investigated through a series
of quasi-static tests. The post-tensioned connection was able to with-
stand high drift levels without damage to the column; it presented a
stable hysteretic behavior and negligible residual rotation at the
column base connection. More recently, Chen et al. [15] proposed
vertical pre-tensioned intermodular connection for MSB with composite
columns (steel-concrete). Quasi-static tests were performed in two full-
scaled stacked modules to investigate the seismic behavior of the pro-
posed connection. The structure presented adequate load bearing
stiffness and ductility. Similarly, monotonic and quasi-static tests were
performed in a modular structure subassembly where the HSS columns
were vertically connected by a steel plug and long bolts connected floor
and ceiling rectangular HSS beams [16,17]. The results indicated that
the connection eliminates the need for on-site welding while providing
the required lateral capacity and energy dissipation.
The realization of the full potential of the modular construction
method requires more efficient and faster assembly techniques and a
clear understanding of the behavior of the MSB structure. To address
this need, the current paper introduces a new post-tensioned (PT)
connection for the vertical intermodular connection of MSBs; and
evaluates its seismic behavior and lateral load bearing capacity through
a series of quasi-static cycling loading tests. The proposed connection
does not require on-site welding and it is suitable for typical HSS steel
columns. In Section 2 of this paper, the new PT connection is described.
The dimensions of the specimens, test set-up, test matrix and the in-
strumentation layout are provided in Section 3. Section 4 shows the
general behavior of specimens with the welded connection and the ones
with the PT connection. The deformation pattern, hysteretic behavior,
energy dissipation, lateral stiffness and strain distribution for both
welded and PT connections are also described in this section. Finally,
the summary and main conclusions are reported in Section 5.
2. Proposed vertical post-tensioned connections for MSB
In order to vertically connect the columns of the bottom and top
modules, the proposed connection uses a post-tensioned threaded rod
and a steel box as shown in Fig. 1a. The threaded rod goes through the
columns full height to establish vertical connectivity between the
modules and is anchored against end plates at columns ends. A hollow
steel box with sloped sides at the top and at the bottom is inserted and
secured inside the typical hollow section column of the lower module
half way through its height. Its sloped sides help to guide both the
threaded rod through the middle plate of the box and the upper module
column as it is lowered during the assembly. The horizontal con-
nectivity of the modules is provided by the steel box that sits over the
full height of the two layers of beams at each floor level. Beams are
made of W sections and are connected to the columns by off-site
welding. Couplers can be used to connect threaded rods to cover longer
heights and enable several modules to be tied together. With the
777
Engineering Structures 175 (2018) 776–789
proposed connection on-site welding process would not be required to
vertically connect the modules and horizontal inter-modular connec-
tions are typically bolted (Fig. 1b). The recent work of Lacey et al. [7]
provides a state-of-the-art review of modular building structures and
other types of connections.
In an earthquake event, the columns may experience a variation in
the axial load demand, especially the ones located at the perimeter of
the building. Tensile forces that may occur in exterior columns are re-
sisted by the threaded rod and the friction force provided by the contact
between the steel box and the inner surface of the HSS column. In order
to avoid excessive compression load in the columns, the post-tensioning
load should be considered together with the gravity loads in the design
process.
3. Experimental study
3.1. Test specimens, test set-up and material properties
The subassemblies tested in this study represent a connection from a
MSB previously designed by Fathieh and Mercan [2] (Fig. 2a) with
minor adjustments. The thickness of the columns was increased to avoid
buckling during the tests, the floor beams and the clearance between
floor and ceiling beams were modified to account for the current
practices in modular construction. This gap between beams is often
required to facilitate easy assembly of the modules (to provide space for
the removal of the temporary supports and lifting slings that go un-
derneath the floor beams of the module as its being lifted during the
assembly) and allow electrical and mechanical systems to run through
the building during its service life [2]. The length of the columns and
the beams of the subassemblies were determined based on the inflection
points of the MSB (Fig. 2b). Centerline distances of the subassembly are
indicated in Fig. 2c as well as the direction of the displacements applied
– positive displacement (+Δ) and negative displacement (−Δ). Con-
sidering the double layer of beams that exist at each floor level, the
inflection point for the columns are not exactly at mid-height, thus the
bottom column is shorter than the upper column. Dimensions of the
modular parts and the steel box are indicated in Fig. 3 and photos of
these elements are provided in Fig. 4. Beam-column welded connection
was fabricated by the modular construction builder, NRB Inc. who is a
collaborator in this project. Fillet welding process was considered, as
shown in Fig. 3a and b. The steel box total height is determined such
that it covers the height of the two layers of beam at each story level.
The floor beam is slightly deeper than the ceiling beam; therefore, the
steel box is longer above the middle plate (Fig. 3c).
A general view of the test setup is provided in Fig. 5a. A hydraulic
actuator with maximum stroke of ± 127 mm and maximum load ca-
pacity of 350 kN was mounted at the top of the upper column. The
actuator imposed quasi-static cyclic displacement according to the AISC
protocol [18] indicated in Fig. 5b with slow rates varying from
0.35 mm/s to 1.5 mm/s to avoid any dynamic response effects [8]. A
constant axial load of 100 kN – which corresponds to 17% of the
nominal compressive resistance of the column – was applied in all tests
by a hydraulic jack mounted at the base of the bottom column, where a
load cell was also installed. This load represents the vertical loads
acting in the column of the first floor of the building and it was com-
puted assuming a typical floor system comprised of concrete floor and
steel deck (2.0 kN/m2) and insulation (0.25 kN/m2). Superimposed
dead loads for floors, roof, and ceilings are 0.75, 0.32, and 0.7 kN/m2,
respectively. Live load of 1.9 kN/m2 is considered in every floor and the
snow load of 1.1 kN/m2 is considered for the roof. For the specimens
with the proposed PT connection, a 25.4 mm of diameter threaded rod
was considered and the initial post-tensioning (PT0) load varied from
0 kN to 80 kN. The threaded rod was anchored against plates at each
end of the column. The PT load was applied as shown in Fig. 6a before
the 100 kN constant axial load was applied. A removable support is
attached to the top plate and a hydraulic jack is used to apply the PT
R. Sanches et al. Engineering Structures 175 (2018) 776–789

Fig. 1. (a) Proposed vertical post-tensioned connection; (b) Typical intermodular horizontal (bolted) and vertical (welded) connections for MSBs [7].

Fig. 2. (a) Four storey MSB; (b) Inflection points; (c) Subassembly centerline dimensions in millimetres.

load. The top end of the upper column is assumed a free end by com-
bining a spherical support and rollers (Fig. 6b). To facilitate the as-
sembly and disassembly of the test set up, the upper column was cou-
pled to the top plate, but it was not welded. The bottom column was
pinned at the base with a spherical support (Fig. 6c), which allowed the
base to rotate but prevented translational displacements. Bottom and
upper sleeves were attached to the columns to cover the external seg-
ment of the rod and to allow for the assembly of the supports. At the
end of the beams a roller boundary condition was provided by attaching
each beam to a pin that slides inside a slotted hole, allowing for in-
plane rotation and lateral displacements (Fig. 6d). Shim plates (Fig. 6e)
were used to fill the gap between the outer face of the box and the inner
face of the columns. However, the outer dimension of the steel box can
also be selected to fit tightly inside the column and thus eliminating the
shimming process.
Table 1 indicates the cross sections of the structural elements. Col-
umns and beams are categorized as Class 1 (compact sections), ac-
cording to CSA [19]. Three different thicknesses for the steel box were

778
R. Sanches et al.
Fig. 3. Specimen dimensions in millimetres: (a) Upper module; (b) Bottom module; (c) Steel box.
considered. The material properties were obtained from tensile tests
according to ASTM-A370 [20] and the values indicated for yield stress
(fy), ultimate stress (fu), yield strain (εy) and Young’s modulus (E) are
the average of three tensile tests per batch. In the W beams, three
coupons were extracted from the web and from the flanges. The cou-
pons of the column were cut from the flat surfaces and did not present a
flat plateau in the stress vs. strain curve, hence the yield point was
obtained by 0.2% offset method. The yield point of the remaining
coupons was obtained from the autographic diagram method according
to ASTM-A370 [20]. The Young’s modulus was defined as the slope of
elastic part of the stress-strain curve obtained from the tensile tests. It is
worth mentioning that the average elastic limit strain (proportional or
linear limit strain) for the column coupons was 0.07%. Elastic limit
strain corresponds to maximum strain in the elastic range of the stress-
strain curve. All the coupons were taken from the longitudinal direc-
tion. The steel beams were produced from ASTM A572 Gr50, and the
columns and boxes were from CSA G40.21-350 W Class C, respectively.
Table 2 shows the test matrix where the main variables are the steel
box thickness and initial post-tensioning load (PT0). Specimens R2W
and R4W were tested with partial and all around welding connection,
respectively. Eight tests were carried out in specimens with the PT
connection, with combination of different steel boxes (B1, B2 and B3)
and three different initial post-tensioning loads (0 kN, 60 kN and
80 kN). The thickness of the box is proportionate to the thickness of the
column: B1 has the same thickness of the column, B2 is 1.5 times
thicker than the column, and B3 is twice as thick as the column. The
maximum value for the PT load was defined such that the total axial
load applied in the column remained below the buckling load. PT0 of
60 kN and 80 kN correspond to 10% and 14% of the nominal ultimate
tensile strength of the rod, respectively. Two specimens with PT con-
nection were tested twice with increased PT loads (B1PT0 was retested
as B1PT60, and B3PT60 was retested as B3PT80). Specimen R2W with
partial welded connection was retested as R4W. In total, 10 different
test cases were considered. To clarify the distinction between the
Fig. 4. Photos of (a) upper module part, (b) bottom module part and (c) steel box.
779
Engineering Structures 175 (2018) 776–789
specimens tested with the PT connection they were named as follows:
the first two characters indicate the box size and the number following
“PT” indicates the initial post-tensioning load. For instance, the spe-
cimen B1PT60 was assembled with the steel box B1 and PT0 was 60 kN.
In order to increase the PT load, the upper column boundary system
was disassembled, the jack was attached to the rod, and the PT load was
increased incrementally (from 0 kN to 60 kN, and from 60 kN to 80 kN).
Then, the boundary connection was reassembled, the 100 kN constant
axial load was reapplied, and the specimen was tested again. The
threaded rod did not exhibit any yielding before the second set of tests.
After the test of specimen R2W no damage nor yielding was observed.
Small cracks were noticed in the weld of the beam-column connection
after testing specimen B1PT0, but no yield was observed. Specimen
B3PT60 exhibited some weld fractures and yield strain was measured in
the beam-column connection area. General behavior and damage level
experienced by each connection are described in Section 4. Two spe-
cimens (B1PT0 and B3PT0) were tested with no initial PT load, but the
100 kN constant axial load was maintained. This way, the performance
of the connection is evaluated in a case of an unexpected loss of the
post-tensioning load. Additionally, one test was performed with a
coupler connecting two segments of the threaded bar (B1PT80C) to
assess the connection performance when longer rods are needed to
connect multiple modules vertically. Both the threaded rod and the
coupler were purchased from McMaster-Carr [21]. According to the
manufacturer’s documentation both the rod and the coupler are com-
posed of high strength steel Grade 8.
Initially, all PT specimens were intended to be tested up to 3.7%
drift (Δ = ± 125.44 mm), considering the maximum stroke of the ac-
tuator. However, the lack of PT load in the first test (B1PT0) allowed a
gap opening between the top plate and the upper column which caused
an unexpected large drift in the spherical head of the upper column
support (Fig. 7). Therefore, for safety reasons, specimen B1PT0 was
tested up to 3% drift in both directions. Similarly, B1PT60 were tested
up to 3% drift in the negative direction. The issue did not occur for the
R. Sanches et al. Engineering Structures 175 (2018) 776–789

Set Drift Number of cycles

1 0.375% 6
2 0.50% 6
3 0.75% 6
4 1.00% 4
5 1.50% 2
6 2.00% 2
7 3.00% 2
8 3.70% 2

(a) (b)
Fig. 5. (a) Test setup; (b) Lateral displacement protocol.

other tests because of higher PT loads and thicker steel boxes. ceiling beam strain gauges 25, 26, 29 and 30 were not installed in the
specimen with welded connection. A total of ten Linear Variable Dif-
ferential Transformers (LVDTs) were arranged along the column and at
3.2. Instrumentation
the middle of the beams to measure in plane and out of plane dis-
placements according to Fig. 8c. Two LVDTs were positioned in the
Strain gauges were installed on the specimens in the connection
bottom column and four were installed in the upper column. LVDTs
region where higher levels of strain are expected. Fig. 8a and b indicate
BCH1 and BCH2 recorded lateral displacements of the bottom column
the location of the strain gauges in the upper module and bottom
at approximately the mid-length and at the center line of the ceiling
module, respectively. Eleven strain gauges and one rosette were in-
beam, respectively. Similarly, UCH1 and UCH2 measured lateral dis-
stalled in each column and six strain gauges were installed in each
placements of the upper column at the mid-length and at the center line
beam. However, because of the limited space between floor beam and

Fig. 6. (a) Anchorage and PT load application; (b) Boundary condition of upper column (c) bottom column and (d) beams; (e) Close-up view of shim plates.

780
R. Sanches et al. Engineering Structures 175 (2018) 776–789

Table 1 4.2. General behavior of post-tensioned connection

Cross section and material properties.
Element Cross section fy (MPa) fu (MPa) E (MPa) εy (%) Eight specimens with post-tensioned connections were tested with
three different steel boxes and three different post-tensioning loads
Column HSS 127 × 127 × 6.4 372 450 193,350 0.39 combinations to evaluate the proposed vertical connection. In addition,
Threaded rod Ø 25.4 986 1081 202,086 0.69
in order to investigate the effects of using couplers, one test was per-
Floor W 150 × 18 (Flange) 351 510 193,652 0.20
bea-
formed with a coupler connecting two segments of the threaded bar
m (B1PT80C). Similar to the welded specimens, no local buckling was
W 150 × 18 (Web) 390 532 197,463 0.19 observed in post-tensioned connections. Welding fractures occurred in
Ceiling W 100 × 19 (Flange) 331 478 202,247 0.19 the external corners of the floor beam flanges (Fig. 9b), except specimen
bea-
B2PT80 where no cracks were noticed, as will be discussed later. No
m
W 100 × 19 (Web) 371 504 196,213 0.19 welding fracture was observed in the ceiling beam in any of the post-
Steel (B1) HSS 102 × 102 × 6.4 406 469 209,738 0.40 tensioned specimens.
box As mentioned in Section 3.1, some specimens were tested twice
(B2) HSS 102 × 102 × 9.5 411 460 185,706 0.42
under increased PT load. Specimen B1PT0, with the thinner box B1 and
(B3) HSS 102 × 102 × 12.7 432 475 185,824 0.44
zero PT0 load, was tested up to one cycle of 3% drift
(Δ = ± 100.35 mm). As a result of the large drift in the spherical head
of the floor beam. In addition, LVDTs UCV1 and UCV2 were positioned of the upper column boundary condition this test was terminated before
perpendicularly to the upper column to indicate out-of-plane deflec- applying larger drifts. At this stage, small welding cracks were observed
tions. In-plane and out-of-plane displacements in the floor beam were in the upper and bottom flanges of the floor beam. Then, the PT load
obtained from LVDTs FBH and FBV, respectively. For the ceiling beam, was increased to 60 kN and the specimen was renamed as B1PT60.
these measurements were obtained by LVDTs CBH and CBV. The in- Because of the increased PT load, the specimen could be tested up to
ternal LVDT of the actuator was used to measure the displacements 3.7% drift in the positive direction, but for the negative direction the
applied at the upper column. In addition, four strain gauges were in- issue with the boundary condition appeared and the specimen was
stalled on the threaded bar as indicated in Fig. 8d. A picture of the tested up to 3% drift in this direction. Although existing welding cracks
LVDTs is provided in Fig. 8e. were enlarged, no new cracks were developed. By considering a larger
PT load (PT0 = 80 kN) the drift in the spherical head was better con-
strained and specimen B1PT80 could be tested up to 3.7% drift in both
4. Test results and analysis directions. Specimen B3PT0, assembled with the thicker box B3 and
zero PT0 load, was also tested up to 3.7% drift even though no PT load
4.1. General behavior of the welded specimens was applied, indicating that the thicker steel box contributed to the
stability of the upper boundary condition. Similarly, specimen B3PT60
The first tests were performed with the welded connection speci- was tested up to 3.7% drift in the positive and negative directions. Two
mens. When one module is placed next to another one access to all sides welding fractures could be seen in the floor beam-column connection.
of the column for welding may not be possible. Due to inherent ac- Next, the PT load was increased to 80 kN and the subassembly was
cessibility limitation during the assembly of the modular construction, renamed as B3PT80. This specimen was also tested up to maximum
partial welding is expected in the column connection. Thus, a fully range of displacements (Δ = ± 125.44 mm). The existing welding
welded connection specimen (designated as R4W with all four sides fractures were greatly enlarged (Fig. 9c) and other two cracks were
welded), and a partially welded connection specimen, (designated as formed also in the floor beam-column connection. To assess the con-
R2W with only two sides welded) were tested. In specimen R2W the nection behavior with an intermediary thickness box (B2), specimen
upper face and the side located between the beams were welded. The B2PT80 was tested up to 3.7% drift in both directions with PT0 of
test was conducted by applying lateral displacement of ± 66.9 mm (2% 80 kN. No fractures were observed. It is worth mentioning that besides
drift). No damage was observed in the specimen and the maximum B2PT80 specimens B1PT80, B1PT80C and B3PT0 were also tested only
strain was 0.3% measured in the strain gauge 10, which is located in the once. The results from these tests indicate that the welding fractures are
beam-column connection region where large strains are expected. No not related to the total number of cycles in loading that each specimen
yielding was observed. After this test, the other two sides of the column experienced. Instead, as will be discussed later in detail, it could be seen
were welded, and the specimen was designated as R4W. This specimen as an evidence that a connection with an intermediary thick box pre-
was tested up to Δ = ± 125.4 mm (3.7% drift). Although no local sents a better performance.
buckling was noticed, a small crack was observed in the welding of the For all tests with PT connection, at around 2% drift
floor beam upper flange at 3% drift (Fig. 9a). No cracks were observed (Δ = −66.9 mm), a noise was heard from the connection region in-
in the column welds and the maximum strain of 0.8% was obtained in dicating the friction between the shim plates, the steel box and the
the strain gauge 10.

Table 2
Test matrix.
Specimen Steel box PT0 (kN) Δ max (mm) Drift max (%) Observation

R2W – – ± 66.90 2.0 2 sides welded

R4W – – ± 125.44 3.7 4 sides welded
B1PT0 B1 0 ± 100.35 3.0
B1PT60 B1 60 +125.44/−100.35 +3.7/−3.0
B1PT80 B1 80 ± 125.44 3.7
B1PT80C B1 80 ± 125.44 3.7 With a coupler
B2PT80 B2 80 ± 125.44 3.7
B3PT0 B3 0 ± 125.44 3.7
B3PT60 B3 60 ± 125.44 3.7
B3PT80 B3 80 ± 125.44 3.7

781
R. Sanches et al. Engineering Structures 175 (2018) 776–789

Fig. 7. Upper column boundary condition: (a) Gap opening between top plate and upper column; (b) Drift in the spherical support.

column. At displacement of −100.35 mm (−3% drift), a gap between
the upper column and the middle plate of the steel box started to form.
As expected, for higher PT loads the gap was smaller and for specimens
B1PT0 and B3PT0 the gap could be seen even at −1% drift. At dis-
placements larger than −3% drift, it was noticed that the upper column
exhibited rigid body rotation about the middle plate of the steel box,
and the gap widened up to 2.5 mm for the specimens with zero PT0 load
(Fig. 9d). For the specimens with PT0 of 60 kN and 80 kN the maximum
gap was 1.5 mm and 1 mm, respectively. Greater deformation in the
upper column could be observed, especially for the specimens with
higher PT0, where the gap between columns was smaller, inducing
greater flexural demand in the upper column. The gap occurred only for
negative displacement (−Δ), and a similar behavior was reported by
Chen et al. [16]. At peak negative displacement the gap reaches its
maximum opening. When the displacement is reversed from negative to
positive direction, the force is initially transferred to the bottom column
while the gap is still open due to friction provided by the steel box. On a
second stage, the gap closes and finally the force completely is trans-
ferred to the bottom column.

Fig. 8. Strain gauges location at (a) upper module and (b) bottom module; (c) LVDTs location; (d) Threaded rod strain gauges; (e) Picture of the LVDTs.

782
R. Sanches et al. Engineering Structures 175 (2018) 776–789

Fig. 9. Weld fracture in floor beam of specimens (a) R4W, (b) B1PT60 and (c) B3PT80; (d) Gap opening at −3.7% drift (B1PT0).

4.3. Deformation patterns
The displacements along the length of the column corresponding to
the peak lateral displacement Δ in each set of cycles are indicated in
Fig. 10 for three different specimens. The lateral displacements were
obtained from LVDTs BCH1 (middle of the bottom column), BCH2
(centerline of the ceiling beam), UCH2 (centerline of floor beam), UCH1
(middle of the upper column) and the internal LVDT of the actuator (Δ).
It is observed that the deformation pattern of specimens with higher
PT0 resembles the welded specimen R4W. Both specimens with welded
connection presented similar deformation patterns. Specimen tested
with the coupler (B1PT80C) did not show any difference in the de-
formation pattern in comparison to specimen B1PT80. At the same level
of PT load, the deformation pattern is similar for all specimens re-
gardless of the thickness of the steel box. It is worth noticing that in the
last set of cycles of the specimen R4W the LVDT located at the mid-
length of the upper column (UCH1) stopped working properly. Thus,
the kink shown in the middle of the deformed configuration of the
upper column at Δ = −125.44 mm (Fig. 10c) did not actually happen
during the test, as later confirmed by video recordings. No out-of-plane
deflections were observed in any of the specimens.
As mentioned in Section 4.2, with increasing lateral displacement Δ

Fig. 10. Deformation pattern of columns with steel box B3 and (a) PT0 = 0 kN, (b) PT0 = 80 kN; and (c) and welded specimen R4W.

783
R. Sanches et al. Engineering Structures 175 (2018) 776–789

2.6 B1PT0 B1PT80

B1PT0 B1PT80 4
B3PT0 B3PT60 B3PT0 B3PT60
2.2 B3PT80 B2PT80 B3PT80 B2PT80
3
R4W R4W

(%)
1.8 2

1.4 1

1 0
-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150
(mm) (mm)
(a) (b) (c)
Fig. 11. (a) Schematic drawing of angles α1, α2 and α3; (b) Upper column flexural deformation (μ) and (c) relative rotation between upper and bottom columns (θ)
versus lateral displacement (Δ).

20 20 20
R4W B3PT0 B3PT80
Lateral load (kN)

10 10 10

0 0 0

-10 -10 -10

Skeleton
-20 Hysteresis -20 -20
-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150
Δ (mm) Δ (mm) Δ (mm)

Fig. 12. Hysteretic curves.

Fig. 13. Skeleton curves.

and as the gap opening is prevented by the application of the PT load
and the flexural stiffness of the steel box, the upper column experiences
large flexural deformations while the bottom column is barely de-
formed. This flexural deformation in the upper column can be evaluated
by using the bending factor (μ), defined as the ratio α1/α2 (Fig. 11a). α1
is the angle between a vertical line parallel to the undeformed column
and a line connecting the LVDT UCH2 and the actuator locations on the
deformed column, and α2 is the rotation of the upper column relative to
the undeformed configuration. The angle α1 is computed by taking the
difference between the displacement imposed by the actuator and the
displacement obtained in the LVDT UCH2 and dividing it by their se-
paration distance (h1). The angle α2 is computed by the difference of
displacement from LVDTs UCH1 and UCH2 divide by their distance
(h2). When α1 is equal to α2 (μ = 1) it shows that there is no bending in
the upper column, and that the upper column essentially experiences
rigid body rotations. On the other hand, μ > 1 is an indication of
bending deformation. It can be seen in Fig. 11b that the flexural de-
formation in the upper column tends to be smaller in PT specimens
when compared to R4W for displacements in the negative direction
(−Δ). In average, μ in PT specimens is 15% smaller than R4W for ne-
gative displacements. In addition, specimens with higher PT0 load tend
to develop similar deformation, regardless of the thickness of the steel
box. However, for zero PT0 load, thicker steel box tends to increase
bending deformations.
In general, the relative rotation between upper and bottom columns
decreases with increasing PT0 load. This relative rotation (θ) can be
evaluated by computing the difference between the rotation of the
upper column (α2) and the rotation of the bottom column (α3). α3 is
calculated by dividing the lateral displacement measure by the LVDT
BCH2 to its distance to the pin support at the base (h3), as indicated in
Fig. 11a. As shown in Fig. 11c, PT specimens tend to develop a relative
rotation pattern very similar to that of R4W up to 1% drift

784
R. Sanches et al.
Fig. 14. Energy dissipated per cycle.
Fig. 15. Cumulative energy dissipation.
(Δ = ± 33.45 mm). However, θ increases significantly for negative
displacements as the PT0 load decreases. This is due to the gap that
forms between the two columns, which were larger for zero PT0 load
cases. For the sake of clarity, only the results from seven specimens are
plotted in Fig. 11b and c. Specimen with the coupler (B1PT80C) pre-
sented similar relative rotation and flexural demand characteristics to
those of the specimen B1PT80.
4.4. Hysteretic behavior and skeleton curve
The hysteretic curves are provided in Fig. 12. These curves are si-
milar between welded and PT specimens, and with increasing level of
PT load the hysteretic curve becomes narrower. Skeleton curves are the
backbone of the hysteretic loops and they indicate the peak load at each
set of cycles together with the corresponding lateral displacement ap-
plied. Fig. 13a compares the skeleton curve for the welded specimens,
R2W and R4W. It can be seen that both have similar load-bearing ca-
pacity up to 2% drift. Fig. 13b shows the skeleton curves of specimens
with box B1 and different PT loads. It also includes the curve for the
specimen with a coupler (B1PT80C). It can be observed that the coupler
has no impact in the load capacity of the specimen. Amongst the spe-
cimens tested with box B1, B1PT0 had the most similar lateral bearing
capacity to the welded specimen up to 2% drift in positive and negative
directions. Fig. 13c shows the skeleton curves of specimens with boxes
B3 and B2. It needs to be pointed out that B3 is a thicker box than B2. It
can be seen that specimens with thicker box (B3) presented the least
bearing capacity. However, the specimen B2PT80 had very similar load
capacity to the R4W up to 3.7% drift, which suggests that there is a
limit to be considered for the box thickness.
785
Engineering Structures 175 (2018) 776–789
4.5. Energy dissipation
Seismic performance of a structural element can be evaluated by the
energy dissipated during cyclic loading, which is defined as the area
enclosed by the force-displacement hysteresis loop [8,17]. In this paper,
both the energy dissipated per cycle (using the average of energy dis-
sipated in a set of cycles) and the cumulative energy dissipation (ob-
tained by summing up the energy dissipated in every cycle) were
computed. These are then used to assess and compare the seismic
performance of the welded specimens and the specimens with the PT
connection. For specimens with steel box B1 and for different levels of
PT load, the energy dissipated per cycle and the cumulative energy
dissipated versus the drift imposed by the actuator are shown in
Figs. 14a and 15a, respectively. Fig. 14a also includes the specimen
tested with a coupler. Similarly, Figs. 14b and 15b show dissipated
energy for specimens with thicker steel boxes and different PT0 loads.
All graphs include the results from specimen R4W for comparison. It is
observed that up to 2% drift, all the specimens dissipated a similar
amount of energy per cycle. However, the variation in the energy dis-
sipation is more pronounced for drifts larger than 3%, especially for
thinner steel box (B1). At 3.7% drift, all PT specimens dissipated more
energy per cycle than the welded specimen.
For the same box thickness, as the PT0 increases the cumulative
energy dissipation tends to decrease slightly. This can be attributed to
the smaller gap openings at higher PT loads. For the same PT0 load,
when the box thickness increases, the cumulative energy dissipation
tends to decrease as a result of the increased stiffness in the connection
and smaller relative rotation between the columns. Because the spe-
cimen B1PT0 was subjected to only one cycle of 3% drift (the set
contains two full cycles), the cumulative energy for this specimen was
R. Sanches et al.
Fig. 16. (a) Secant stiffness at 2% drift (B3PT0); (b) Initial secant stiffness (Ki) and secant stiffness at 3% drift (K3%); (c) (d) Stiffness degradation factor vs. lateral
displacement; (e) (f) Normalized stiffness vs. lateral displacement.
considered only up to 2% drift; and the cumulative energy for the
specimen B3PT60 was considered up to 3% drift only. The energy
dissipation of R2W was very similar to that of R4W. For the sake clarity
the results for the specimen R2W are not included in the plots provided
above.
4.6. Stiffness degradation
When subjected to reversed cyclic loading structural elements may
gradually lose lateral stiffness. This loss is often referred to as stiffness
degradation. Secant stiffness (Fig. 16a) is defined as the ratio of the
lateral load to the maximum displacement at each set of cycles and is
used to evaluate the stiffness degradation of the specimens. The
786
Engineering Structures 175 (2018) 776–789
stiffness degradation factor (η) is defined here as the ratio of the secant
stiffness at each set of cycles to the secant stiffness of the first cycle (Ki),
when Δ = 12.54 mm (Fig. 16b). Fig. 16c and d compare the stiffness
degradation factors of PT specimens, with box B1; and B3 and B2 re-
spectively, considering different PT0 loads as well as the welded spe-
cimen, R4W. Specimens with PT connection tend to degrade faster than
the welded one when drifts are higher than 1%. They show non-sym-
metrical degradation rates, more pronounced in specimen B3PT80.
Because of the gap opening, the lateral stiffness tends to decrease faster
for negative displacements. Specimen B2PT80 had stiffness degradation
very similar to R4W and presented more symmetrical degradation be-
havior in comparison with specimens with thicker box (B3), which
suggests that very thick steel boxes may not be the best option to
R. Sanches et al. Engineering Structures 175 (2018) 776–789

Table 3 yielding at −3.7% drift. The second larger strain values were obtained
Initial stiffness, lateral load and final stiffness at 3% drift, maximum strain. in either strain gauge 2 or strain gauge 9 with a value of 0.29%, in
Specimen Ki (kN/mm) Load3% (kN) K3% (kN/mm) εmax (%) average. For the remaining strain gauges, the average maximum strain
obtained was 0.22%. Strain data suggests that the yielding was confined
* *
R2W 0.34 13.5 0.20 0.27 to the beam-column connection area and registered almost exclusively
R4W 0.31 16.9 0.16 0.79
by strain gauge 10. In specimen B3PT60, yielding was also indicated by
B1PT0 0.40 13.2 0.13 0.38
B1PT60 0.40 11.6 0.11 0.53
strain gauges 2 and 9.
B1PT80 0.28 16.8 0.17 0.70 In order to represent the strain distribution in the specimen due to
B1PT80C 0.39 14.4 0.14 1.30 cyclic displacements, Fig. 18 shows the maximum absolute strain
B2PT80 0.34 13.6 0.14 0.95 measured in each one of the strain gauges for the welded specimen R4W
B3PT0 0.29 14.5 0.14 0.39
and PT specimens with different boxes and PT0 loads. Strain gauges 1 to
B3PT60 0.31 13.8 0.14 0.85
B3PT80 0.46 12.1 0.12 0.77 11 are installed in the upper column, 12 to 22 are in the bottom
column, 23 to 28 in the floor beam and 29 to 34 in the ceiling beam.
* Values for 2% drift. Strain gauges 35 to 37 and 38 to 40 are the rosettes of the upper and
bottom columns, respectively. Peak shear strains obtained from the
decrease degradation rates, and the thickness of the box should be rosettes were 0.23%, in average. Maximum shear strain was 0.29%
limited to 50% greater than the column thickness. measured in the upper column of specimen B1PT80C. The strains
In order to better compare the change in stiffness between welded measured in the threaded rod (strain gauges 41 to 44) are shown in
and PT specimens, Fig. 16e and f provide the normalized stiffness (λ) as Fig. 19 for specimens with the box B1 and different post-tensioning
the ratio of the secant stiffness at each set of cycles to the first cycle loads. For zero PT0 load, the threaded rod remained in compression due
secant stiffness of the welded specimen R4W. Most of the PT specimens to axial load applied by the jack during the entire test while for spe-
developed a higher initial secant stiffness than the welded specimen. cimens with 60 kN and 80 kN of PT0 load, the bar remained in tension
Especially specimens B1PT0 and B1PT60 presented a secant stiffness in although, in average, 20% of the initial post-tensioning load was lost.
the first cycle that is 30% higher than that of R4W. Table 3 indicates the This post-tension force loss is mainly due to anchorage seating, and it
maximum values for initial secant stiffness (Ki), lateral load (Load3%) at was computed based on the difference between the strain measured in
3% drift and the secant stiffness at 3% drift (K3%) (Fig. 16b). the bar at the beginning and at the end of the cyclic test. The relaxation
occurred throughout the cyclic tests. It needs to be noted that the strain
4.7. Strains gauge 42 was damaged during the testing of specimens B1PT60 and
B1PT80. The rod in specimens B1PT80 and B1PT80C presented similar
For each specimen, the maximum compressive strains due to the behavior indicating that the coupler does not have significant impact in
combined axial and PT0 loads were measured in different locations, but the bar strain distribution. Both beams and the threaded rod remained
their numerical values were around 0.05%. For B3PT0, Fig. 17 presents in the elastic range.
the peak strain in the floor beam flanges (black markers) and ceiling
beams flanges (white markers) for positive and negative displacements 5. Summary, conclusions and future work
in each set of cycles. It can be seen that the measurements agree with
the imposed displacement patterns. For positive displacements (+Δ), This paper investigates a new vertical post-tensioned connection for
upper flange is in tension (positive strain) and bottom flange is in the assembly of the modules in modular steel building construction. The
compression. Furthermore, floor beam experiences higher level of proposed connection uses a post-tensioned threaded rod installed inside
strain due to gap opening and it also experiences a higher flexural de- the HSS columns of the modules that are placed on top of each other.
mand; both of which lead to weld fracture in the beam-column con- The tensioned threaded rod connects the columns vertically and the
nection. Similar behavior was observed for all PT specimens. As ex- resulting connection behaves similar to a typical welded connection.
pected, high strains occurred in the beam-column connection region. Additionally, a steel box is inserted between the columns of the upper
For all specimens, the largest strains were recorded in strain gauge 10 and lower modules to transfer lateral loads to the bottom column. These
when cyclic displacements were applied. Table 3 shows the maximum boxes also serve as a guide for the upper module and the rod during the
strain (εmax) obtained for each specimen. Data obtained from strain assembly process. The proposed connection does not require on-site
gauge 10 indicates that most specimens surpassed the linear limit strain welding which is desirable as on-site welding often interferes with the
at 0.75% drift (Δ = ± 25.09 mm). In addition, the yield strain (εy) was structural finishing of the modules and slows down the assembly pro-
reached at −3% drift (Δ = −100.35) mm for all specimens, except for cess. Eight quasi-static cyclic loading tests were performed to evaluate
B1PT0 where no yielding took place, and for B3PT0 that reached the hysteretic behavior of the proposed connection. Additionally, two

Fig. 17. Flanges strain of B3PT0 (a) positive displacement (+Δ) and (b) negative displacement (−Δ).

787
R. Sanches et al.
Fig. 18. Maximum absolute strain per strain gauge.
Fig. 19. Threaded rod strain vs. lateral displacement.
tests were performed with typical welded connection: one specimen
with partial welding that represent the realistic situation when all four
sides of the column are not accessible during the assembly and another
with all around welding connection. The following are the main ob-
servations and conclusions.
(1) Up to 2% drift, both welded specimens presented similar general
behavior, lateral bearing capacity and energy dissipation.
Additionally, all specimens presented welding fractures in the
beam-column connection at 3% drift or higher, except for B2PT80
where no welding fracture was observed.
(2) Deformation patterns along the length of the columns reveal that
higher PT0 loads decrease the relative rotation between modules,
especially in the negative direction of the imposed displacement
(−Δ). On the other hand, higher PT0 loads induce higher flexural
demand in the upper column. It is also observed that the use of
thicker steel box results in an increase in the flexural demand in the
upper column regardless of the magnitude of the initial post-ten-
sioning load. Due to the gap that forms between the upper column
and the middle plate of the steel box during negative displacement
steps, PT specimens presented unsymmetrical behavior. Both the
relative rotations and flexural column deformations were higher for
negative direction.
(3) Specimens with thicker steel box had the least lateral bearing ca-
pacity in most cases, especially for negative displacements. The
lateral stiffness at 3% drift for specimens B1PT80 and B3PT0 is 6%
and 12% higher, respectively, than the welded specimen R4W.
However, the lateral stiffness of B1PT0 and B3PT80, at same drift
demand, is 18% and 25% smaller. This indicates that thinner steel
boxes (B1) generally performs better than thicker ones (B3).
Specimen B2PT80 had a stiffness that is 12% smaller at 3% drift and
9% higher at initial drifts when compared to R4W, which suggests
that intermediary thickness for the box may improve the general
788
Engineering Structures 175 (2018) 776–789
behavior of the connection.
(4) Energy dissipation per cycle and the cumulative energy dissipation
performances were similar for all specimens up to 2% drift. At 3.7%
drift, the values of the energy dissipated per cycle of the specimens
with PT connection were significantly higher than R4W; and spe-
cimens with a thicker box appeared to dissipate more energy per
cycle. Also, all PT subassemblies had higher cumulative energy
dissipation values in comparison with the welded specimen R4W.
(5) Specimens with a PT connection presented higher stiffness de-
gradation ratios than welded specimens. However, majority of the
post-tensioned specimens developed an initial secant stiffness
higher than those specimens with welded connection. Especially,
B1PT0 and B1PT60 presented a stiffness value that is 30% higher
than R4W in the first set of cycles. Specimen B2PT80 presented
similar stiffness degradation to the specimen R4W. In general,
specimens with thinner steel box had more symmetrical degrada-
tion behavior whereas specimens with thicker box showed more
asymmetry and faster stiffness degradation, especially for the ne-
gative displacement direction.
(6) Although approximately 20% of the initial post-tensioning load was
lost during the cyclic displacement loading, the threaded rod re-
mained in tension for specimens with PT0 of 60 kN and 80 kN. For
the specimens with no PT0 load, the rod remained in compression
due to the axial force applied. Even in the cases where no post-
tensioning load was applied the specimens presented high lateral
bearing capacity for thinner steel box at small drift demands, as
pointed out in (5) above.
(7) The strain distribution was similar for all specimens with PT con-
nection in comparison to the welded one (R4W), except for the floor
beam connection (strain gauges 9 and 10) where the maximum
strain varied from 0.38% to 1.3%. The occurrence of gap opening
between upper column and steel box mid-plate and the increased
thickness in the connection region because of the steel box led to
R. Sanches et al.
this variation. In general, floor beam experienced higher strains due
to increased flexural demand in the upper column, especially for
subassemblies with a thicker box. The experimental study in this
paper revealed that the thickness of the steel box should be limited
to 50% greater than the thickness of the column to avoid excessive
strains in the beam-column connection and fast stiffness degrada-
tion.
(8) The use of a coupler to connect the threaded rod did not affect the
behavior of the specimen. Specimen B1PT80C (with a coupler) and
specimen B1PT80 (without a coupler) presented a similar de-
formation pattern, lateral stiffness and energy dissipation.
Results of the experimental analysis discussed in this paper indicate
that the proposed post-tensioned intermodular connection is able to
withstand the applied lateral force as well as a typical welded con-
nection, but with the advantage of eliminating the need for on-site
welding. A numerical model for the proposed connection is currently
being calibrated with the experimental results and will be reported
later. The numerical model will be able to simulate stiffness, stress
distribution, strength and energy dissipation capabilities of the pro-
posed connection. The calibrated model will contribute to the research
to understand the system level behavior of MSBs with the proposed
connection. For instance, for tall buildings, the effects of the cumulative
drift due to the gap opening at different story levels can be investigated
using the numerical model that also considers the relaxation effect in
the rods and other possible limitations. In future studies, beam-column
connection can be further improved in order to avoid welding fracture.
Acknowledgements
This paper is based upon work funded by the Natural Sciences and
Engineering Research Council of Canada (NSERC) Discovery (2016-
06290) and Engage (EGP 506929) Programs. The first author ac-
knowledges the scholarship program founded by National Council for
Scientific and Technological Development (CNPq, 234888/2014-8).
The authors would like to express their appreciation to NRB Inc who
was a collaborator in this project. Any opinions, findings, conclusions
and recommendations expressed here are those of the authors and do
789
Engineering Structures 175 (2018) 776–789
not necessarily reflect the views of the sponsors.
References
[1] Annan CD, Youssef MA, El Naggar MH. Seismic overstrength in braced frames of
modular steel buildings. J Earthquake Eng 2009;13:1–21.
[2] Fathieh A, Mercan O. Seismic evaluation of modular steel buildings. Eng Struct
2016;122:83–92.
[3] Ramaji IJ, Memari AM. Product architecture model for multistory modular build-
ings. J Constr Eng Manage 2016;142(10). https://doi.org/10.1061/(ASCE)CO.
1943-7862.0001159. 04016047:1-14.
[4] Lawson RM, Ogden RG, Bergin R. Application of modular construction in high-rise
buildings. J Archit Eng 2012;18:148–54.
[5] Jellen AC, Memari AM. The state-of-the-art application of modular construction to
multi-story residential buildings. In: 1st residential building design & construction
conference, Bethlehem, PA; 2013.
[6] Gunawardena T. Behavior of prefabricated modular buildings subjected to lateral
loads. Ph.D. thesis. Department of Infrastructure Engineering. University of
Melbourne; 2016.
[7] Lacey AW, Chen W, Hao H, Bi K. Structural response of modular buildings – an
overview. J Build Eng 2018;16:45–56.
[8] Annan CD, Youssef MA, El Naggar MH. Experimental evaluation of the seismic
performance of modular steel-braced frames. Eng Struct 2009;31:1435–46.
[9] Ricles JM, Sause R, Garlock M, Zhao C. Posttensioned seismic-resistant connections
for steel frames. J Struct Eng 2001;127(2):113–21.
[10] Ricles JM, Sause R, Peng SW, Lu LW. Experimental evaluation of earthquake re-
sistant posttensioned steel connections. J Struct Eng 2002;128(7):850–9.
[11] Garlock MM, Ricles JM, Sause R. Experimental studies of full-scale posttensioned
steel connections. J Struct Eng 2005;131(3):438–48.
[12] Garlock MM, Sause R, Ricles JM. Behavior and design of posttensioned steel frame
systems. J Struct Eng 2007;133(3):389–99.
[13] Lin X, Okazaki T, Nakashima M. Bolted beam-to-column connections for built-up
columns constructed of H-SA700 steel. J Constr Steel Res 2014;101:469–81.
[14] Chi H, Liu J. Seismic behavior of post-tensioned column base for steel self-centering
moment resisting frame. J Constr Steel Res 2012;78:117–30.
[15] Chen Z, Li H, Chen A, Yu Y, Wang H. Research on pretensioned modular frame test
and simulations. Eng Struct 2017;151:774–87.
[16] Chen Z, Liu J, Yu Y, Zhou C, Yan R. Experimental study of an innovative modular
steel building connection. J Constr Steel Res 2017;139:69–82.
[17] Chen Z, Liu J, Yu Y. Experimental study on interior connections in modular steel
buildings. Eng Struct 2017;147:625–38.
[18] AISC. Seismic provisions for structural steel buildings: ANSI/AISC 341-05, Chicago,
Illinois; 2005.
[19] CSA. Design of steel structures: CSA S16-14. Canadian Institue of Steel
Construction, Toronto, Ontario; 2014.
[20] ASTM. Standard test methods and definitions for mechanical testing of steel pro-
ducts: ASTM-A370-15, USA; 2015.
[21] McMaster-Carr. < https://www.mcmaster.com > [accessed on June 23, 2018].