See

discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/267268480

Compatibility Torsion in Spandrel Beams Using Modified

Compression Field Theory
ARTICLE in ACI STRUCTURAL JOURNAL MAY 2006
Impact Factor: 1.09

READS

416

1 AUTHOR:
Khaldoun Rahal
Kuwait University
34 PUBLICATIONS 311 CITATIONS
SEE PROFILE

Available from: Khaldoun Rahal

Retrieved on: 11 December 2015

ACI STRUCTURAL JOURNAL

TECHNICAL PAPER

Title no. 103-S34

Compatibility Torsion in Spandrel Beams Using Modified

Compression Field Theory
by Khaldoun N. Rahal and Michael P. Collins
This paper describes a procedure that can be used to calculate the
compatibility torsion in spandrel beams and other indeterminate
structures. The procedure relies on the ability of the Modified
Compression Field Theory (MCFT) to calculate cracked torsional
and flexural stiffness for sections subjected to various combinations of
stress resultants. The calculated ultimate loads are compared with the
experimentally observed ultimate loads of 27 statically indeterminate
floor beam-spandrel beam frames and very good agreement is
observed. The comparison with the detailed experimental results
shows that the MCFT is capable of accurately calculating the
response of reinforced concrete sections subjected to combined
shear, torsion, bending moment, and axial compression. The
calculations of the proposed procedure also show that the strength
of the frames is significantly larger than that calculated using the
equations of the ACI and CSA codes due to the longitudinal
restraint of the beams.
Keywords: beams; cracking; reinforced concrete; shear; spandrel beams;
torsion.

INTRODUCTION
Many structural elements such as beams curved in plan,
eccentrically loaded bridge girders, and spandrel beams in
buildings are subjected to significant torsional moments. To
design such members, it is essential to recognize whether the
torsional moments are required to maintain equilibrium or
compatibility. These two cases are generally referred to1-4 as
equilibrium torsion and compatibility torsion, respectively.
Torsion in determinate structures is always equilibrium
torsion, while that in indeterminate structures can be of
either type.1
Figure 1 shows examples of the two cases of torsion. The
traffic load on the statically determinate multi-cell bridge in
Fig. 1(a) is eccentrically applied, and torsion on the transverse
cross-section is required to maintain equilibrium. Redistribution
of this torsional moment is not possible, and the total
moment calculated using statics must be designed for. A
similar situation arises if the bridge is curved in plan.
Figure 1(b) shows torsion in a spandrel beam of a building
caused by loading from a cast-in-place slab. This torsional
moment is due to imposed rotations from the slab, and is
equal to the negative bending moment in the slab. The
compatibility torsion due to the required twist in this statically
indeterminate structure depends on the torsional stiffness of
the spandrel beam relative to the flexural stiffness of the slab.
A similar phenomenon takes place if a floor beam frames
into the spandrel beam, as shown in Fig. 1(c).
In most of the cases, torsional moments encountered in
reinforced concrete buildings are of the compatibility type,
and their calculation poses a challenge. Once the spandrel
beam cracks in torsion, its torsional stiffness reduces
substantially. The reduction causes a significant redistribution
of the torque to the framing elements. Assuming a gross
328

Fig. 1Examples of torsion in reinforced concrete: (a) equilibrium torsion; and (b) and (c) compatibility torsion.
uncracked stiffness in the structural analysis greatly overestimates the torsion for the specific level of twist.1,2,5-8 The
redistribution of the torsional moment to positive flexural
moment near midspan of the floor beam leads to a more
economic design because the flexural capacity of a typically
reinforced concrete section is significantly larger than its
torsional capacity. In addition, the cost of manufacturing and
placing the longitudinal steel in the floor beam or slab is less
than that for the transverse steel in the spandrel beam.
ACI Structural Journal, V. 103, No. 3, May-June 2006.
MS No. 03-486 received February 22, 2005, and reviewed under Institute publication
policies. Copyright 2006, American Concrete Institute. All rights reserved, including
the making of copies unless permission is obtained from the copyright proprietors. Pertinent
discussion including authors closure, if any, will be published in the March-April
2007 ACI Structural Journal if the discussion is received by November 1, 2006.

ACI Structural Journal/May-June 2006

ACI member Khaldoun N. Rahal is an associate professor in the Department of Civil

Engineering at Kuwait University, Kuwait. He is a member of the torsion subcommittee of
Joint ACI-ASCE Committee 445, Shear and Torsion, and is Past President of the ACI
Kuwait Chapter.
Michael P. Collins, FACI, is University Professor and Bahen-Tanenbaum Professor of
Civil Engineering at the University of Toronto, Toronto, Ontario, Canada. He is a member
of ACI Committees 318, Structural Concrete Building Code; 318-E, Shear and Torsion;
318-G, Precast and Prestressed Concrete; and Joint ACI-ASCE Committee 445, Shear and
Torsion.

Beams subjected to shear, bending, and/or torsion, experience

significant elongation after cracking. Spandrel and floor
beams in reinforced concrete buildings are subjected to this
complex combination of stress resultants. In addition, they
are restrained in the longitudinal direction. Consequently,
they are subjected to a significant axial compressive force,
which affects their strength and stiffness and makes the study
of the redistribution of torsional moment more challenging.
Design for compatibility torsion: background
Reinforced concrete buildings were designed by neglecting
the torsion that takes place in the spandrel beams. In 1963 for
example, the ACI code9 contained only a few lines on the
detailing of the torsional reinforcement in spandrel beams.
The use of computers in the 1960s made the calculation of
the torsional moments in space frame structures practical.
This left designers with a puzzling question: How to design
for torsion? Between 1963 and 1973, a significant amount of
research10 focused on the strength of individual members
subjected to torsion, and led to the introduction of approximately 16 sections of torsion design provisions in the 1971
ACI code.11 The code, however, did not include special
provisions for compatibility torsion, implying the use of
gross uncracked torsional stiffness of the spandrel beams.
In 1971, Collins and Lampert1 defined the two different
types of torsion (compatibility torsion and equilibrium
torsion). Ozerdinc et al.5 reported the test results of six floor
beam-spandrel beam assemblies shown in Fig. 2. The study
investigated the difference in behavior between frames
designed using uncracked gross stiffness (GKgross) (Design
Method A) and zero stiffness (GK = 0) (Design Method B).
See also Table 1 for brief details on these tests. The study
also investigated the effects of variable relative stiffness and
length of the spandrel and floor beams. The results proved
the feasibility of designing the frame based on redistribution
of the torsional moment if the spandrel beam is provided
with a minimum amount of properly detailed reinforcement.
In 1972, Onsongo2 tested three frames similar to those
shown in Fig. 2, but the spandrel beams were completely
restrained in the longitudinal direction. He concluded that
restrained beams designed assuming (GKgross) or assuming
zero torsional stiffness (GK = 0) performed in a satisfactory
manner, and that the restraining force has secondary effects.
In 1974, Hsu and Burton6 tested 10 floor beam-spandrel
beam assemblies. They investigated an alternative design
approach where the spandrel is designed for a torsional shear
stress of 4fc psi (0.33fc MPa), while the remaining
torsional moment is redistributed to the framing elements
(Design Method C). This approach was found satisfactory,
and was later adopted by the 1983 ACI12 code.
In 1977, Hsu and Hwang7 tested four floor beam-spandrel
beam assemblies and compared the behavior of frames
designed assuming (GK = 0) (Collins and Lamperts
approach or Design Method B) with frames designed using
ACI Structural Journal/May-June 2006

Fig. 2Test setup by Ozerdinc et al.5

Design Method C. The floor beam load was distributed along
the length, and an additional load was applied at the joint as
shown in Table 1. The joint load increased the shear force in
the spandrel, representing the high torsion and shear and
bending near the support of the spandrel. Hsu and Hwang
concluded that both design methods were satisfactory, and
recommended an expression for the minimum amount of
transverse reinforcement required in the spandrel beam.
In 1978, Abul Mansur and Rangan8 tested seven floor
beam-spandrel beam assemblies. Three of the specimens had
spandrel beam load as shown in Table 1. They compared the
behavior of frames designed assuming GK = 0 (Collins and
Lamperts approach was Method B) and Hsu and Burtons
design approach described above with frames designed
using GKgross. They concluded that both design approaches
are satisfactory, and that Hsu and Burtons approach
(Method C) was slightly more economical.
In 1991, Mo and Hsu13 developed a theoretical model for
analysis of compatibility torsion based on the use of trilinear
torque-twist and bending-curvature relationships. Their
model captured the effects of the two redistributions that
take place after torsional cracking in the spandrel and
yielding of the longitudinal reinforcement in the floor beam,
and good correlation with the experimental results was
observed. However, the model did not take into consideration
the effects of longitudinal restraint, the effect of shear and
bending on the torsional stiffness of the spandrel, and the
effect of shear on the flexural stiffness of the floor beams.
The model was applicable to 11 of the specimens discussed
above, and its results were in good agreement with the
experimental results.
Design for compatibility torsion in current
ACI and CSA codes
The current ACI3 and CSA4 codes allow partial redistribution
of the torsional moment in the spandrel beam to flexural
moment in the framing slab or beam. The ACI3 design
torque in the spandrel beam is the cracking torque given by
Eq. (1) while the remaining moment is redistributed to the
framing elements. A similar procedure is allowed in the
Canadian code4 with the design torque equal to 67% of the
cracking torque. The CSA design torque is given in Eq. (2).
329

Table 1Comparison of observed and calculated ultimate loads

v fy spandrel,
Specimen size* and loading
MPa
arrangement
Design method

Ozerdinc et al.5

Hsu and Burton6

Abul Mansur and

Rangan8

Hsu and Hwang7

Ps /Pf

Pf-exp, kN

Pf-calc, kN Pf-exp /Pf -calc

S1

1.490

164.6

254.0

240.2

1.057

S2
S3

B
A

0.507
0.705

191.3
231.3

0
0

l242.4
294.9

226.8
267.0

1.069
1.104

S4

0.507

191.3

234.6

204.6

1.147

S5

0.507

133.4

174.8

178.0

0.982

S6

0.507

209.1

304.7

258

1.181

A1

3.30

135.2

163.7

151.2

1.082

A2
A3

C
C

1.22
1.22

127.7
127.7

0
0

126.8
149.5

142.3
149.0

0.891
1.003

A4

0.00

124.5

101

A5

0.00

124.5

B1

3.28

113.9

86.7
116.1

111.2

1.044

B2

1.14

113.9

125.4

129.0

0.972

B3

0.00

113.9

77.4

B4

B-C

0.41

113.9

117.9

135

0.88

B5

B-C

0.41

113.9

109.0

SA-1

1.71

120

157.1

122
136

0.89
1.155

SA-2
SA-3

C
C

1.28
1.01

116
116

0
0

142.1
138.1

133
133

1.068
1.038

SA-4

1.08

116

142.1

140

1.015

SB-1

1.83

116

0.5

158.0

141

1.120

SB-2

C
B

1.51
1.51

116
116

0.5
0.5

166.0
154.0

150
145

1.107
1.062

MTV1

2.18

171.2

240.2

211

1.14

MTV2

0.74

160

231.7

218

1.062

MTV3

0.74

171.2

250.4

231.3

1.083

C1

1.80

113.9

144.6

139

1.040

C2
C3

C
B

1.80
1.86

113.9
113.9

0.55
1.50

142.3
111.2

135
108

1.054
1.030

C4

0.98

113.9

0.81

108.5
Average

100

1.085
1.05

SB-3

Onsongo2

Pf-des, kN

Coefficient of variation, %

7.40

Beam lengths drawn to scale, cross section dimensions of spandrel and floor beam in mm.
Design Method A: gross stiffness; Design Method B: zero stiffness (Collins and Lampert)1; and Design Method C: Hsu and Burton.6
torsional reinforcement in spandrel beam.
Inadequate torsional reinforcement.13

No

A
T ACI = 0.33 f c -----c
pc

(1)

A
T CSA = 0.268 c f c -----c
pc

(2)

If U.S. customary units are used, the factors 0.33 and 0.268
are replaced with 4 and 3.23, respectively. Proper detailing
330

of the transverse reinforcement ensures ductility of the spandrel

and its ability to provide the twist required for redistribution.
While the code approach is simple to use, it does not take
into consideration various significant factors such as the
amounts of transverse and longitudinal reinforcement in the
beams, the restraining effects in the longitudinal direction,
and the spans of the beams. Consequently, the code approach
does not take into account the actual torsional and flexural
stiffness of the spandrel and the floor beams. The satisfactory
performance of spandrel beams designed according to the
ACI Structural Journal/May-June 2006

code provisions proves that these provisions are conservative. In

the case of analysis of existing structures, however, more
rational models are required to obtain accurate results.
Calculation of stiffness in reinforced concrete beams
Following the publication of the experimental results on
compatibility torsion in the 1970s,1,2,6-8 there has been a
significant advancement in the analysis methods of reinforced
concrete beams subjected to complex loading. One such
behavioral model, the Modified Compression Field Theory
(MCFT) is capable of calculating the full response of reinforced
and prestressed concrete beams subjected to shear, bending,
and axial load.14 A further development15,16 included the
effects of torsion and enabled the calculation of the response
of rectangular sections subjected to all six possible stress
resultants on a beam cross section. A computer program
named COMBINED was developed based on this model,
and was shown to give accurate results on steel and concrete
stresses and strains, and overall deformations such as twist,
elongation, and curvatures for combined shear and torsion15
and combined torsion and bending.16 Such a program can
hence be used to calculate the torsional and flexural stiffness
in rectangular beam cross sections subjected to complex loading
in general, and in spandrel beams in particular. It is to be noted
that the MCFT has not been thoroughly tested for the
combination of torsion, shear, bending, and axial compression.
This paper describes the use of the MCFT and a standard
linear-elastic structural analysis program to calculate the
compatibility torsion in spandrel beams and other indeterminate
structures. The comparison with the test results will also
enable checking the ability of the MCFT to calculate the
response of sections subjected to the complex combination
of shear, torsion, bending, and axial compression.
RESEARCH SIGNIFICANCE
The ACI and CSA provisions dealing with compatibility
torsion such as that encountered in spandrel beams are simple
and conservative for design. There is a lack of analytical
models, however, that can account for the influence of the
variable factors that affect the performance of spandrel
beams in buildings, and of members subjected to compatibility torsion in general. The MCFT is used to calculate
more realistic cracked torsional and flexural stiffness that
can be used in the structural analysis procedure for more
accurate results. The procedure applies to the practical
case where the beams contain properly detailed transverse and
longitudinal reinforcement.
PROPOSED ANALYSIS PROCEDURE
Figure 3 shows a flow chart of the proposed procedure. A
typical structural analysis of the indeterminate frame is
performed based on assumed flexural and torsional stiffness
of the structural members. These assumed values can be the
uncracked gross values, or smaller values if cracking is
expected at the level of the load applied. Generally, the
stiffness varies considerably along the length of the structural
members such as beams. Hence, modeling a beam or a
column using only one element does not lead to an accurate
representation of the variation in stiffness. Hence, similar to
nonprismatic sections, structural members need to be
modeled using numerous elements, with lengths recommended
not to exceed twice the depth.
The average values of the shearing force V, axial load N,
and bending and torsional moments (M and T, respectively)
ACI Structural Journal/May-June 2006

Fig. 3Flow chart of proposed procedure.

in each of the elements obtained from the structural analysis
output. These values, along with the geometrical and
reinforcement details of the elements cross sections, are then
used in COMBINED, the sectional analysis computer program,
to calculate the state of strains and stresses in the materials and
the overall sectional deformations (twist and flexural
curvature ). Refined values of the assumed effective torsional
and flexural stiffness (GK)eff and (EI)eff are calculated from the
twist and flexural curvature using the following equations
(GK)eff = T/

(3)

(EI)eff = M/

(4)

where E is the modulus of elasticity of concrete and G is the

shear modulus of concrete, which may be taken as 0.5E.
Refined values of effective K and I are then calculated for
each of the elements and used as input in the structural analysis.
Shear deformations can also be included, but are neglected
for the type of frames considered in this study.
Members subjected to shear, bending, and torsion, acting
individually or in combination, experience significant elongation
after cracking. In addition, members subjected to combined
shear and torsion experience lateral curvature.15,17 The floor
331

Fig. 5Variation in torsional stiffness of spandrel beams

and flexural stiffness of floor beams at P = 178 kN (40 kips).

Fig. 4Cross section details of specimens and idealization

of structure (Specimens S1, S2, and MTV2).2,5
beam in Fig. 2 elongates, while the spandrel beam experiences
both elongation and lateral curvature due to the combination
of shear and torsion. The elongation and lateral curvature
calculated by the computer program COMBINED can be
included in the structural analysis as initial deformations
in the elements.
The structural analysis is repeated for the structure using
the refined values of the stiffness, elongation, and curvature
until convergence is achieved.
CORRELATION WITH EXPERIMENTAL RESULTS
The experimental results from the tests listed in Table 1
are used for the verification of both the proposed procedure
and the MFCT in the case of complex loading. The ultimate
load capacities of the frames are compared with the calculated
capacities. In addition, three of the frames are selected for
detailed comparison of the response.
Indeterminate frames similar to that shown in Fig. 2 have
been tested at the University of Toronto.1,2,5 Three of these
frames, S1, S2, and MTV2, were selected for the verification.
Figure 4 gives some of the details of the reinforcement in the
three frames. More detailed results can be obtained from
332

Reference 1 and 5 for Specimens S1 and S2, and from

Reference 2 for Specimen MTV2.
The frames had 2895 mm (9 ft, 6 in.) long spandrel beams
and 4572 mm (15 ft) long floor beams. Specimen S1 was
designed for a floor beam load of 164.6 kN (37 kips) based
on the uncracked gross torsional stiffness GKgross of the
spandrel beam (Method A). Specimen S2 was designed for a
load of 191.3 kN (43 kips) based on a zero torsional stiffness
assumption (Method B), and hence allowing redistribution
of the compatibility torsion to flexure in the floor beam. It
contained minimal transverse reinforcement as shown in
Table 1. Specimen MTV2 was designed for 160 kN (36 kips)
floor beam load based on a zero torsional stiffness assumption.
Hence, it was similar to Specimen S2 (that is, GK = 0), except
that the spandrel beam was restrained in the longitudinal
direction to simulate the effects of the columns and the
spandrels in adjacent bays. The actual case in a spandrel
beam lies within these three extreme cases, but with additional
longitudinal restraint in the floor beam. Additional analyses
of Specimen MTV2 with longitudinal restraint in the floor
beam are described later.
The frames were modeled as a series of 20 elements, 10 in
each beam, as shown in Fig. 4. To maintain clarity in
presenting the results, the structural modeling did not take
advantage of the symmetry in the spandrel along the axis of the
floor beam. The analysis was performed for different values of
the applied load, and the analytical results are checked against
the experimental results in the following sections.
The transverse curvature in the spandrel beam due to the
combination of shear and torsion calculated by COMBINED
does not bear an effect on the results of the structural analysis
and hence was not included. Moreover, a compressive force
is applied to the spandrel to maintain a negligible average
longitudinal strain in the 10 elements. For the simple frame,
this was more suitable than including the longitudinal strain
of each element as an initial strain in structural analysis
input. In addition, the frame shown in Table 1 and Fig. 2 is
indeterminate to the first degree, and hence it was sufficient
to check for convergence in the value of the compatibility
torsion in the spandrel beams.
Variation in torsional and flexural stiffness
Figure 5 shows the calculated torsional stiffness in the
spandrel beams and the flexural stiffness in the floor beams
at a load of 178 kN (40 kips). The calculated gross uncracked
stiffness is also shown. The variation in the stiffness along
the length of the beams is considerable and shows the significant
effect of the variable bending moment on the value of GK in
ACI Structural Journal/May-June 2006

the spandrel. This shows the need of the numerous elements

to model the beams, especially in the spandrel beams after
torsional cracking. The same was found to be true for the
floor beams after yielding of the longitudinal steel. The
longitudinal restraint limited the spread of torsional
cracking, and increased the torsional stiffness in the
uncracked regions of the spandrel beam significantly beyond
the gross uncracked values.
Load-compatibility torque diagrams
Figure 6(a) shows the compatibility torsional moment in
the spandrel of Specimen S1 at different levels of the load P.
It also shows the ACI and the CSA design torsional moments
calculated using Eq. (1) and (2) based on unit resistance
factors and the response based on the gross uncracked
stiffness, and on the cracked stiffness proposed by Collins
and Lampert.1 Before torsional cracking, the behavior was
similar to what can be obtained from a linear elastic analysis
based on gross and uncracked sectional properties. The
proposed procedure accurately calculated this precracking
response, but slightly underestimated the cracking loads.
Two distinct changes affected the behavior of the specimen;
at torsional cracking of the concrete in the spandrel and at
flexural yielding of steel in the floor beam. Upon torsional
cracking, the torsional stiffness of the spandrel decreased
dramatically, allowing redistribution of the torque into the
floor beam. Additional loading caused a relatively smaller
increase in the torsional moment and more significant
increase in the flexural moment in the floor beam. This was
accurately captured by the proposed method. When the
longitudinal reinforcement in the floor beam yielded, the
demand on the rotation in the joint increased dramatically.
The spandrel was designed to resist larger torque, and hence
the torque increased at a larger rate. In other words, the direction
of the redistribution reversed. The figure shows that the
model accurately captured both changes in behavior,
including the compatibility torsion at design load, and the
yielding in the bottom steel in both beams and in the stirrups
of the spandrel. The calculated ultimate load was attained
when the section at and near midspan (under the load)
reached its flexural and shear capacity. The ultimate load
was underestimated by less than 6%.
Figure 6(b) shows a similar behavior in Specimen S2 until
the occurrence of yielding in the bottom longitudinal
reinforcement of the floor beam. The spandrel contained
relatively small amounts of transverse steel, and hence it was
capable of twisting considerably without an increase in the
torsional moment. The demand on the twist increased when
the bottom reinforcement of the floor beam yielded, but the
transverse steel of the spandrel yielded at the same load
level, which did not allow for reverse redistribution and
increase in load. A good correlation is found in the comparison
between the calculated and observed response. The model
captured the precracking response, the change in behavior
upon torsional cracking in the spandrel, and the occurrence
of yielding in the longitudinal and transverse steel of the
spandrel and the bottom longitudinal steel in the floor beam.
The model accurately predicted the compatibility torsion
at design load, and underestimated the ultimate load by
approximately 7%.
Figure 6(c) shows a similar diagram for Specimen MTV2,
where the longitudinal spandrel was restrained in the
longitudinal direction. The response was relatively softer
than expected before cracking. After cracking, however, and
ACI Structural Journal/May-June 2006

Fig. 6Torque-applied floor load diagrams for spandrel

beams.
as was shown in Fig. 5, the longitudinal restraint caused an
increase in the torsional stiffness and strength of the spandrel.
This is reflected on the relatively stiff post-cracking behavior
of the specimen, and then on the increase in torsion after
yielding in the bottom steel of the floor beam. The ultimate load
was not significantly affected because it was dominated by
the flexural strength of the floor beam.
Figure 6(c) shows that the model is capable of calculating
the full response of the specimen, including the change in
behavior caused by the longitudinal restraint and yielding in
the bottom reinforcement of the floor beam. The model
333

of the spandrel of Specimens S1 and S2. The twist in the

different elements was variable due to the presence of the
variable bending moment. To be consistent with the
experimental results used in Fig. 7, the calculated values
reported are the average of the twist from Elements 12 to 15
and the part of Element 11 that falls outside the joint. Figure 7(a)
shows a very good agreement between the observed and
calculated response, including the occurrence of torsional
cracking and the precracking and post-cracking response.
The values of the torsional stiffness at torques corresponding
to design and ultimate loads were 37 and 14%, respectively,
of the measured uncracked stiffness.
Figure 7(a) also shows the torque-twist response of the
various elements in the spandrel beams. The elements subjected
to higher bending moment near the joint showed softer response
and hence smaller torsional stiffness. At design and calculated
ultimate loads, respectively, the effective torsional stiffness
values in the element with highest bending moment (M/V =
0.45Lf in Element 11) were 50 and 84% of those in the elements
with lowest bending moment (M/V = 0.05Lf in Element 15).
This shows the significant effect of the bending moment on the
torsional stiffness, especially at design load.
The response of Specimen S2 shown in Fig. 7(b) was
slightly different because of the relatively low amount of
transverse reinforcement. The torque remained virtually
unchanged after cracking, allowing a relatively ductile
behavior and redistribution of the torque as the load increased.
The proposed model accurately captured the response of
Specimen S2, including the precracking stiffness, the cracking
torque, and the post-cracking stiffness. The values of the
torsional stiffness at torques corresponding to design and
ultimate loads were respectively 15 and 10% of the measured
uncracked stiffness. These values are significantly smaller
than Specimen S1 designed based on gross stiffness.
Figure 7(c) shows a similar diagram for Specimen MTV2,
except that the twist was measured over a 1.22 m (4 ft) length
of the specimen and the calculated values reported in the
figure corresponds to the twist averaged over this length.
Similar to the trend observed in the load-torque diagrams, the
longitudinal restraint caused a significant stiffening in the
response. The torsional stiffness at torques corresponding to
design and ultimate loads were 43 and 33% of the measured
uncracked stiffness, respectively, which are higher those
observed in Specimens S1 and S2. Figure 7(c) shows that the
proposed method accurately calculated the precracking and
the post-cracking stiffness, and the change in behavior due to
the longitudinal restraint. Hence, the MCFT is capable of
predicting the response of members subjected to combined
shear, torsion, bending moment, and axial compression.
Fig. 7Average torque-twist diagrams for spandrel beams.
underestimated the ultimate load by approximately 6%.
Comparing the results of the three specimens, it is found that
the behavior and strength are significantly affected by the
longitudinal restraint and the amount of reinforcement in the
spandrel beam.
Torque-twist diagrams
The twist along the length of the spandrel beams in
Specimens S1 and S2 was measured using a Metrisite
extending from the center the joint to the clamping truss at
the edge of the beams as shown in Fig. 2. Figure 7(a) and (b)
compares the observed and calculated torque-twist response
334

Restraining axial force in spandrel beam

The spandrel beam in Specimen MTV2 was restrained from
expanding longitudinally using a 100 ton jack bearing on end
plates and 1-1/4 in. Dywidag bar positioned in the center of a
100 mm (4 in.) hole running at the center of the beam. This bar
was prestressed at the different levels of the loading to ensure
that the longitudinal strains in the spandrel beam were negligible. The restraining force was measured during the test. As
observed from Fig. 6 and 7, the restraining force had a significant effect on the load-torque and torque twist response.
Figure 8 shows the calculated and the experimentally
measured restraint axial force in the spandrel of Specimen
MTV2 at different levels of the applied floor load P. The
restraint force was negligible until the occurrence of torsional
ACI Structural Journal/May-June 2006

Fig. 8Restraining axial compressive force in Specimen

MTV2.
cracking, and increased almost linearly after that due to the
elongation in the spandrel beam. A good correlation is observed
between the measured and the calculated response at different
levels of loading, including the design load.
Load-deflection diagrams
Figure 9 compares the observed and calculated loaddeflection curve in the floor beam for the three specimens.
The response showed considerable softening when the
longitudinal steel in the floor beam yielded. The longitudinal
restraint in the spandrel beam of Specimen MVT2 increased
the torque and hence reduced the flexure and consequently
the vertical deflection in the floor beam. The proposed
method accurately predicted the general trends in the
response, but overestimated the stiffness.
Crack width calculation
One of the major concerns with allowing redistribution
was the suspicion that using less transverse steel in the spandrel
beams would lead to less satisfactory crack width
behavior.1,2,5,6,8 Specimen S2 designed assuming GK = 0
contained approximately 1/3 the transverse steel provided in
Specimen S1 designed assuming GKgross. The observed
load-twist diagrams in Specimens S1 and S2 were, however,
very similar,1,5 and the recorded maximum crack widths
were similar up to design load. Estimates of the crack width
are necessary to satisfy the design requirements, though it
should be realized that crack width calculations do not
usually lead to accurate results.
Average and maximum diagonal crack widths were
recorded in the three specimens. Reference 2 reports that
these widths were measured near mid-height. The computer
program COMBINED calculates average crack widths at
three different levels along the height of the cross section,
and those widths vary because of the longitudinal strain
variation across the height16 from Element 11. The
maximum crack width occurs near the bottom of the section
where the effect of the bending is significant, but no
measurements of crack widths are reported at that level. The
comparison is hence checked for the maximum width at
midheight. Due to the wide variation in crack spacing, there
will be a similar variation in the crack width. The maximum
crack width is taken as 1.7 times the calculated average
ACI Structural Journal/May-June 2006

Fig. 9Applied floor load-midspan deflection in floor beam.

width in accordance with the CEB-FIP code.18 The ACI
code does not set a limit on the width of diagonal cracks, but
it bases the serviceability performance of flexural members
on a 0.3 mm crack width limit.
Figure 10 shows the observed and calculated load versus
maximum crack width diagrams near midheight of the
section for the three specimens and the 3 mm limit. The
reported calculated crack width was that from Element 11,
shown in Fig. 4. The proposed procedure seems to give
adequate results, especially at design loads. The measured
crack widths are highest in members designed using Method B,
but were smaller than 0.3 mm at service load (assumed to be
70% of the design load). Longitudinal restraint and design
335

Fig. 10Maximum width of diagonal cracks near middepth of spandrel beams.

using Method A (providing larger amounts of transverse
reinforcement in the spandrel) significantly reduce the width
of the diagonal cracks.
Strains in transverse reinforcement
The strains in the stirrups on the inside face of the spandrel
beams were measured using 200 mm (8 in) targets as shown
in Fig. 2. These strains are critical because the stirrups in the
spandrel are designed to resist the combined shearing
stresses from torsion and shear. Figure 11 compares the
observed and the calculated strains. The theoretical values
336

Fig. 11Average strains in stirrups of spandrel beams.

reported are the average of the three values calculated along
the height of the section16 from Element 11. Figure 11(a) shows
that the stirrup strains in Specimen S1 were negligible before
torsional cracking, and increased upon cracking and reached
yielding levels. The proposed procedure accurately calculated
this behavior including the occurrence of yielding before the
ultimate conditions were reached.
Figure 11(b) shows a flat post-cracking behavior in
Specimen S2, which twisted as the load was increased,
allowing redistribution of the torsional moment. The model
captured the precracking and the post-cracking behavior, and
ACI Structural Journal/May-June 2006

the occurrence of yielding in the stirrups. Figure 11(c) shows

a similar correlation between the calculated and observed
torque-stirrups strain diagrams. The figure also shows that
the longitudinal restraint reduced the transverse strains.
Correlation with other test results
The proposed procedure was also used to calculate the
ultimate floor loads in 27 other test specimens. Table 1
reports some of the main characteristics of the test specimens
and the results of the comparison. The main variables studied
in these tests are: 1) the design method; 2) the cross section
size and reinforcement; 3) the relative length of the spandrel
and floor beams; 4) the longitudinal restraint in the spandrel
beam; 5) the concentrated load at the joint; and 6) the
uniform load on the floor beam.
It is to be noted that frames without joint loads simulate
floor beams framing near midspan of the spandrels, where
the spandrel shearing stresses are mainly due to the torsional
moment, with smaller contribution from the shearing force.
To simulate the more critical conditions near the supports of
the spandrels, Hsu and Hwang7 and Abul Mansour and
Rangan8 tested specimens with floor loading in addition to a
concentrated load at the joint. This joint load simulated the
difference in column load near the support, and increased the
shearing force in the spandrel.7 This is similar to the conditions
near the support regions of the spandrel.
Three out of the 30 frames listed in Table 1 did not contain
transverse reinforcement in the spandrels, and hence were
omitted from the comparison. For the remaining 27 specimens,
the average of the ratio Pf-exp /Pf-calc was 1.05 and the coefficient
of variation was 7.4%. Table 1 shows that the proposed
procedure gave accurate results even when the transverse steel in
the spandrels was close to minimum shear reinforcement values.
Mo and Hsu13 studied only 20 tests, six by Ozerdinc et al.,5
10 by Hsu and Burton,6 and four out of the eight by Abul
Mansur and Rangan.8 Their method was applicable only to
11 out of the 20 tests. For these specimens, the average and
coefficient of variation of Pf-exp/Pf-calc were reported13 to be
1.27 and the 10.2%, respectively. For the same 11 specimens,
the procedure proposed in this study gives an average of 1.04
and a coefficient of variation of 4.9%.
Effect of longitudinal restraint
To check the impact of the longitudinal restraint on the
behavior, Specimen MTV2 was reanalyzed twice, the first
without the longitudinal restraint in the spandrel (NR), and
the second with longitudinal restraint in both the spandrel
and the floor beams (FSR). The results of these analyses are
compared with the results of Specimen MTV2 in Fig. 5 to 11.
Figure 5 shows that the restraint in FSR increased the flexural stiffness of the floor beams. The average increase along
the length was 55%. This increase was smaller than the
increase in the torsional stiffness in the spandrel, which averaged
219%. This indicates that the increase in stiffness due to
longitudinal restraint was proportional to the decrease in
stiffness due to cracking.
Figure 6(c) shows that releasing the longitudinal restraint
from the spandrel of Specimen MTV2 (analysis NR) resulted
in a softer response. The ultimate floor load was not significantly
affected because the flexural capacity of the floor beam
was critical in determining the failure load. Adding
restraint in the floor beam (FSR analysis) increased its
flexural stiffness and its flexural (and shearing) capacities. The
ACI Structural Journal/May-June 2006

calculated ultimate load for FSR was 50% larger than that of
Specimen MTV2, and 106% larger than the design load.
Figure 7(c) shows that releasing the force (NR analysis)
softened the response and significantly reduced the torsional
ultimate capacity. Restraining the floor beam (FSR) allowed the
increase in the ultimate capacity of the frame and caused an
increase in the post-cracking stiffness and torsional capacity.
Figure 8 compares the axial compressive forces in the
restrained beams of analysis FSR with that from Specimen
MTV2. To maintain clarity of the results of Specimen
MTV2, the FSR results are shown only for P of approximately
230 kN and higher. The response was nearly linear past the
cracking in the beams. At ultimate load, the restraint in the
floor and the spandrel beams caused compressive forces of
560 and 590 kN, respectively, which are equivalent to a
stress of approximately P/Agross = 0.3fc .
Figure 9(c) shows the vertical deflections from the NR and
FSR analyses. Releasing the restraint in the spandrel beam
NR resulted in a softer response. Adding the restraint in the
flexurally critical floor beam (FSR) allowed to maintain the
stiff response of the frame up to failure load, and did not
display the softening typically associated with yielding in the
bottom longitudinal steel in the floor beam.
Figure 10(c) shows that releasing the restraint in the spandrel
(NR) caused an increase in the diagonal cracks, even though
at a specific level of load resulted in a smaller torque. Adding
the longitudinal restraint in the floor beam (FSR) allowed to
maintain the stiff response of the frame up to failure load.
Figure 10 shows that, while frames designed using Method
B can develop relatively larger cracks, these cracks remain
below 0.3 mm at estimated service load, especially when
longitudinal restraint is taken into consideration.
CONCLUSIONS
Based on the research, the following conclusions can
be made:
1. The MCFT is capable of accurately predicting the full
response of beams subjected to combined shear, torsion,
flexure, and axial compression;
2. The proposed procedure is capable of predicting the
response of indeterminate frames where the effects of
compatibility torsion are dominant;
3. The effective torsional stiffness in spandrel beams
depends mostly on the axial compressive force caused by the
longitudinal restraint provided by the columns and by the
adjacent bays. It is also affected by the amount of transverse
reinforcement and by the magnitude of the bending
moment and shearing force acting in combination with the
compatibility torsion;
4. The torsional (and flexural) stiffness varies considerably
between torsionally (and flexurally) cracked and uncracked
regions. To capture the variations in stiffness, relatively
smaller elements should be used in the structural analysis
modeling in areas where cracked elements are adjacent to
uncracked elements. A maximum length equal to twice
depth of the beam is recommended;
5. The axial compressive force caused by the longitudinal
restraint significantly increases the effective torsional stiffness in
the spandrel, and consequently increases the magnitude of the
compatibility torsion. Hence, the current ACI and CSA
equations might underestimate the torque in the spandrel beams.
The code procedures remain conservative for design purposes,
however, because the compressive force also increases the
337

torsional strength, which enables the beams to resist the

larger torque;
6. The longitudinal restraint in the spandrel and floor
beams increases the resistance of the frames significantly
beyond the design transverse loads. The calculated increase
in one of the frames exceeded 100% of the design load;
7. Design based on limit state and redistribution of the
torque leads to relatively larger diagonal crack widths in the
spandrels. The longitudinal restraints present in the beams of
actual structures, however, reduce the widths to acceptable
limits; and
8. There is a lack of a simple procedure to estimate the
effective torsional stiffness due to the numerous factors that
affect it. Experience with the use of the current ACI and CSA
code provisions show that they are conservative for design.
For the calculation of the ultimate strength of existing structures,
however, the code provisions are unduly conservative, and
more accurate calculations can be obtained using rational
procedures such as that presented in this paper.
ACKNOWLEDGMENTS
The research reported in this paper was carried out at the Department of
Civil Engineering at Kuwait University. Support, and making the computer
facilities available, is gratefully acknowledged. The development of the
MCFT has been made possible by a series of grants from the National
Science and Engineering Research Council of Canada. This support is
gratefully acknowledged.

NOTATION
Ac =
Agross =
E
=
=
fc
=
fy
G
=
I
=
K
=
Pf
=
Pf-calc=
Pf-des =
Pf-exp =
Ps
=
=
pc
M
=
N
=
T
=
TACI =
TCSA =

=
=
c
=
v

338

area enclosed by outside perimeter of concrete cross section

gross area of concrete cross section
modulus of elasticity of concrete
specified concrete strength of concrete
yield stress of reinforcing steel
shear modulus of concrete
moment of inertia of floor beam cross section
torsional stiffness of spandrel beam cross section
floor beam total load
floor beam calculated total load
floor beam design total load
floor beam observed total load
spandrel beam midspan load
outside perimeter of concrete cross section
bending moment
axial force
torsional moment
design compatibility torsion (ACI code)
design compatibility torsion (CSA code)
strength reduction factor (ACI code)
material reduction factor (CSA code)
ratio of transverse reinforcement

=
=

twist of cross section due to torsional moment

bending curvature in cross section

REFERENCES
1. Collins, M. P., and Lampert, P., Redistribution of Moments at
CrackingThe Key to Simpler Torsion Design, Analysis of Structural
Systems for Torsion, SP-35, American Concrete Institute, Farmington Hills,
Mich., 1973, pp. 343-383.
2. Onsongo, W. M., Longitudinally Restrained Beams in Torsion,
MASc thesis, Department of Civil Engineering, University of Toronto,
Toronto, Ontario, Canada, 1972, 43 pp.
3. ACI Committee 318, Building Code Requirements for Structural
Concrete (ACI 318-02) and Commentary (318R-02), American Concrete
Institute, Farmington Hills, Mich., 2002, 443 pp.
4. CSA Standard, Design of Concrete Structures (A23.3-94), Canadian
Standards Association, Rexdale, Ontario, Canada, 1994, 199 pp.
5. Ozerdinc, M. E.; Lampert, P.; and Collins, M. P., Torsion-BendingRedistribution at CrackingAn Experimental Investigation, Civil
Engineering Publication No. 72-73, University of Toronto, Toronto,
Ontario, Canada, Dec. 1972, 23 pp.
6. Hsu, T. T. C., and Burton, K. T., Design of Reinforced Concrete
Spandrel Beams, Journal of the Structural Division, V. 100, No. ST1, Jan.
1974, pp. 209-229.
7. Hsu, T. T. C., and Hwang, C., Torsional Limit Design of Spandrel
Beams, ACI JOURNAL, Proceedings V. 74, No. 2, Feb. 1977, pp. 71-79.
8. Abul Mansur, M., and Rangan, V., Torsion in Spandrel Beams, Journal
of the Structural Division, V. 104, No. ST7, July 1978, pp. 1061-1075.
9. ACI Committee 318, Building Code Requirements for Reinforced
Concrete (ACI 318-63), American Concrete Institute, Farmington Hills,
Mich., June 1963, 144 pp.
10. Torsion of Structural Concrete, SP-18, American Concrete Institute,
Farmington Hills, Mich., 1968, 505 pp.
11. ACI Committee 318, Building Code Requirements for Reinforced
Concrete (ACI 318-71), American Concrete Institute, Farmington Hills,
Mich., 1971, 78 pp.
12. ACI Committee 318, Building Code Requirements for Reinforced
Concrete (ACI 318-83), American Concrete Institute, Farmington Hills,
Mich., 1983, 111 pp.
13. Mo, Y. L., and Hsu, T. T. C., Redistribution of Moments in Spandrel
Beams, ACI Structural Journal, V. 88, No. 1, Jan.-Feb. 1991, pp. 22-30.
14. Vecchio, F. J., and Collins, M. P., Predicting the Response of
Reinforced Concrete Beams Subjected to Shear Using the Modified
Compression Field Theory, ACI Structural Journal, V. 85, No. 3, MayJune 1988, pp. 258-268.
15. Rahal, K. N., and Collins, M. P., Analysis of Sections Subjected to
Combined Shear and TorsionA Theoretical Model, ACI Structural Journal,
V. 92, No. 4, July-Aug. 1995, pp. 459-469.
16. Rahal, K. N., and Collins, M. P., Combined Torsion and Bending in
Reinforced Concrete Beams, ACI Structural Journal, V. 100, No. 2, Mar.Apr. 2003, pp. 157-165.
17. Rahal, K. N., and Collins, M. P., Effect of Thickness of Concrete
Cover on Shear and Torsion InteractionAn Experimental Investigation,
ACI Structural Journal, V. 92, No. 3, May-June 1995, pp. 334-342.
18. CEB-FIP, Model Code for Concrete Structures: CEB-FIP International
Recommendations, 3rd Edition, Comit Euro-International du Bton,
Paris, 1978, 348 pp.

ACI Structural Journal/May-June 2006