This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elseviers archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy
Engineering Structures 33 (2011) 118126
Contents lists available at ScienceDirect
Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct
Accurate time-dependent analysis of concrete bridges considering concrete
creep, concrete shrinkage and cable relaxation
Francis T.K. Au

, X.T. Si
Department of Civil Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong
a r t i c l e i n f o
Article history:
Received 12 January 2010
Received in revised form
20 September 2010
Accepted 26 September 2010
Available online 25 October 2010
Keywords:
Concrete bridges
Creep
Relaxation
Shrinkage
Stress relaxation model
Time-dependent behaviour
Time integration method
a b s t r a c t
This paper proposes a new relaxation model for steel tendons based on the equivalent creep coefficient
to enable the accurate estimation of losses of cable forces. The equivalent creep coefficient works
not only in the case of intrinsic relaxation but also under various boundary conditions. Based on the
proposed relaxation model, an accurate finite element analysis of the time-dependent behavior of
concrete bridges considering concrete creep, concrete shrinkage and cable relaxation is devised based
on the time integration method. Concrete members are modeled by beam elements while tendons are
modeled by truss elements with nodes connected to the beam axis by perpendicular rigid arms. Then the
individual and combined effects of concrete creep, concrete shrinkage and cable relaxation on the long-
term performance of concrete structures are investigated. It is found that the proposed relaxation model
and time integration method can provide a reliable method for time-dependent analysis. The numerical
results obtained indicate that the interactions among these factors should be considered carefully in
analyzing the long-term performance of concrete bridges.
2010 Elsevier Ltd. All rights reserved.
1. Introduction
A long time has elapsed since the first observation of concrete
shrinkage and the discovery of concrete creep [1]. Since then, en-
gineers and researchers have been aware of the time-dependent
behavior of concrete structures [24], and steelconcrete com-
posite structures [5,6] because the creep and shrinkage of
concrete and the relaxation of cables interact with one another
during the construction and service stages, resulting in additional
deflections, cracking, reduction of prestress and redistribution of
internal forces, which in turn affect the long-term structural per-
formance. Therefore the time-dependent effects are significant
considerations for serviceability limit states under which deflec-
tions, stresses and crack widths should be limited [710].
A reliable method for time-dependent analysis of concrete
structures is the finite element method combined with time
integration [11]. In the method, the concrete members are usually
modeled by frame elements and the tendons are treated as
truss elements connected to the structural nodes with rigid arms
[1215]. In the time-dependent analysis of prestressed concrete
structures, the interactions among concrete creep, concrete
shrinkage and cable relaxation are often considered approximately
by introducing relaxation reduction coefficients taken from charts
or tables [1618]. It is therefore desirable to develop better

Corresponding author. Tel.: +852 2859 2650; fax: +852 2559 5337.
E-mail address: francis.au@hku.hk (F.T.K. Au).
methods for modeling the interaction between cable relaxation
and other sources of time-dependent deformations.
This paper first introduces a convenient technique to model
curved tendons using rigid arms. Then a new equivalent stress
relaxation model for steel tendons is developed. An accurate finite
element method is then devised for accurate time-dependent
analysis of concrete structures with time integration taking into
account all three time-varying factors. Numerical investigations
are then carried out to demonstrate the versatility of the proposed
method.
2. Modeling of tendons using rigid arms connected to a beam
The common assumptions for analysis of concrete girder
bridges are made. The concrete, steel reinforcement and steel
cables are modeled separately, ignoring their interaction within
the element. It is also assumed that plane sections remain plane
after bending. The present analysis is mainly for the time-
dependent behavior under working conditions before any cracks
are formed. The tendons are assumed to be perfectly bonded to
the structural concrete members. Eachtendonof anarbitrary shape
is modeled by a series of truss elements with nodes connected to
the beam axis by perpendicular rigid arms [12,14]. However, to
avoid an unduly large number of degrees of freedom (DOFs) for
curved tendons, the rigid arms are modeled by the masterslave
transformation technique rather than using physical nodes.
Fig. 1 shows a spatial beam elements 12 of length L
c
aligned
with the beam axis and a spatial tendon element ab of true
0141-0296/$ see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2010.09.024
Author's personal copy
F.T.K. Au, X.T. Si / Engineering Structures 33 (2011) 118126 119
Fig. 1. Beam element with tendon element.
length l
s
and projected length L
c
on the beam. Note that the beam
element and tendon element need not be in the same plane. A
right-handed local xyz-coordinate axis system is chosen with the
x-axis coincident with the member axis having the origin at node
1. Each node of the beam has three translational DOFs u, v and w
along the local x-, y- and z-axes respectively, and three rotational
DOFs
x
,
y
and
z
about the local x-, y- and z-axes respectively. In
the global matrices, these DOFs are appended by the node numbers
as appropriate. Similarly each end node of tendon element ab has
translational DOFs u, v and walong the local x-, y- and z-axes of the
beam element. Nodes a and b of the tendon element are treated
as slave nodes connected to master nodes 1 and 2 of the beam
element by rigid links a-1 and b-2 perpendicular to the x-axis of
the beam.
The nodal displacement vectors {
ab
} of the tendon element ab
and {
12
} of the beam element 12 can be expressed respectively
as Eqs. (1) and (2) are given in Box I.
The displacement vector {
ab
} is related to {
12
} by the transfer
matrix [H] [19] by
{
ab
} =
_
A 0
0 B
_
{
12
} = [H] {
12
} (3)
where the sub-matrices [A] and [B] are given explicitly by
[A] =
_
1 0 0 0 z
a
z
1
y
a
+ y
1
0 1 0 z
a
+ z
1
0 0
0 0 1 y
a
y
1
0 0
_
(4)
[B] =
_
1 0 0 0 z
b
z
2
y
b
+ y
2
0 1 0 z
b
+ z
2
0 0
0 0 1 y
b
y
2
0 0
_
(5)
in terms of the coordinates (x
1
, y
1
, z
1
), (x
2
, y
2
, z
2
), (x
a
, y
a
, z
a
) and
(x
b
, y
b
, z
b
) of nodes 1, 2, a and b respectively in accordance with
the local x-, y- and z-axes of the beam element.
The stiffness matrix [k]
t
of the tendon element with respect to
the DOFs of the beam element can therefore be expressed as
[k]
t
= [H]
T
[T]
T
[k]
s
[T] [H] (6)
in terms of the local stiffness matrix [k]
s
of the tendon and the
transformation matrix [T] which comprises the direction cosines
of element ab in the local xyz-coordinate axis system. The load
vector {F}
t
of the tendon element with respect to the DOFs of the
beamelement can be similarly expressed in terms of the local load
vector {F}
s
as
{F}
t
= [H]
T
[T]
T
{F}
s
. (7)
By adopting the masterslave technique, the total number of DOFs
will be kept to a reasonable minimum.
3. A relaxation model for steel tendon
3.1. Intrinsic stress relaxation functions
When a tendon is stretched and fixed at two ends at a constant
distance apart while keeping the temperature unchanged, the loss
of the tensioninthe tendonis referred to as the intrinsic relaxation.
For convenience in analysis of prestressing losses, it is desirable to
express the stress relaxation
pr
as a function of time t (hours)
since stressing and initial stress
p0
. An equation to evaluate the
stress relaxation for stress-relieved strands as proposed by Magura
et al. [20] is

pr

p0
=
log(t)
10
_

p0
f
py
0.55
_
(8)
where f
py
is taken as the yield stress, which is defined arbitrarily
as the stress at a strain of 0.01. For low relaxation strands, the
equation commonly used in North America [21] is

pr

p0
=
log(24t)
40
_

p0
f
py
0.55
_
. (9)
Note that both Eqs. (8) and (9) are on the assumption that
relaxation is negligible for initial prestressing ratio
p0
/f
py
not
exceeding 55%.
In a prestressed concrete member, creep and shrinkage tend
to reduce the member length thereby reducing the tendon stress
and relaxation. To deal with it, most engineers have resorted to
approximate methods suchas relaxationfactors of BS8110[22] and
relaxation reduction coefficients [11]. It is therefore desirable to
develop a more rational method to analyze the time-dependent
behavior of concrete structures considering the interactions
among concrete creep, concrete shrinkage and cable relaxation.
3.2. Equivalent creep coefficients of steel tendons
When a tendon is stretched under a constant stress
s
(t
0
)
applied at time t
0
, the total strain
s
(t) at time t, including the
instantaneous andcreepstrain, is expressedinterms of the Youngs
modulus of elasticity E
s
and creep coefficient
s
(t, t
0
) as

s
(t) =

s
(t
0
)
E
s
[1 +
s
(t, t
0
)] . (10)
The creep coefficient
s
(t, t
0
) also depends on the initial pre-
stressing ratio
s
(t
0
)/f
py
strictly speaking, but it is omitted for
convenience. With the applied stress changing with time, the to-
tal strain of steel tendon
s
(t) is obtained by summation of re-
sponses of d
s
() applied at time based on the principle of
superposition

s
(t) =

s
(t
0
)
E
s
[1 +
s
(t, t
0
)] +

s
(t)
0
1 +
s
(t, )
E
s
d
s
() (11)
where
s
(t) is the stress increment from time t
0
to t. If the creep
behavior of tendon is taken to be independent of age, which is
a reasonable assumption, one can introduce the creep coefficient

s
(t t
0
) =
s
(t, t
0
), so that Eq. (11) can be rewritten as

s
(t) =

s
(t
0
)
E
s
[1 +
s
(t t
0
)]
+

s
(t)
0
1 +
s
(t )
E
s
d
s
() . (12)
Author's personal copy
120 F.T.K. Au, X.T. Si / Engineering Structures 33 (2011) 118126
{
ab
} =
_

b
_
=
_
u
a
v
a
w
a
u
b
v
b
w
b
_
T
(1)
{
12
} =
_

2
_
=
_
u
1
v
1
w
1

x1

y1

z1
u
2
v
2
w
2

x2

y2

z2
_
T
(2)
Box I.
Using Eq. (12) to write down
s
(t + t) explicitly and getting the
difference, the incremental strain
s
(t) can be expressed in terms
of the incremental stress
s
(t), meanmodulus of elasticity

E
s
(t)
and incremental creep strain
s
(t) as

s
(t) =

s
(t)

E
s
(t)
+
s
(t) (13)
where

s
(t) =
s
(t + t)
s
(t)

s
(t) =
s
(t + t)
s
(t)

E
s
(t) =
E
s
1 +
s
(t)/2

s
=

s
(t
0
)
E
s
[
s
(t + t t
0
)
s
(t t
0
)]
+

s
(t)

s
(t
0
)

s
(t + t )
s
(t )
E
s
d
s
()
d
d.
For the case of intrinsic relaxation at constant strain, there is
no change in strain with time, i.e.
s
(t) = 0. Let t
0
denote the
time of initial stressing and t be the time interval to define the
subsequent instants t
1
= t
0
+ t, . . . , t
n
= t
0
+ nt. Applying
Eq. (13) to time t = t
0
and noting that
s
(t) = 0, one gets upon
rearrangement

s
(t) =

s
(t
0
)

s
(t
0
) +
s
(t
0
)/2
. (14)
Repeating the same procedure, one can derive the creep coefficient

s
[(n + 1)t] at time t
n
in terms of the previous values of creep
coefficient as

s
[(n + 1)t] =

s
(t
0
)
s
(nt)

s
(t
0
) +
s
(t
0
)/2

s
(t
0
+ nt) [1 +
s
(t)/2]
s
(t
0
)
s
[(n 1)t] /2

s
(t
0
) +
s
(t
0
)/2

i=2

s
[t
0
+(i1)t]
2
{
s
[(n i + 2)t]
s
[(n i)t]}

s
(t
0
) +
s
(t
0
)/2
. (15)
In other words, the equivalent creep coefficient can be computed
based on the intrinsic stress relaxation function.
4. Time-dependent analysis of prestressed concrete bridges
In order to predict the time-dependent behavior of prestressed
concrete bridges, the proposed stress relaxation model for tendons
together with the creep and shrinkage models for concrete can be
implemented by the finite element method using time integration.
The concrete members are modeled by beam elements while
the tendons are modeled by tendon elements connected to the
beam axis by perpendicular rigid arms. The method also provides
benchmark solutions for other simplified methods.
To facilitate subsequent formulation for finite element analysis
using time integration, Eq. (13) for the tendon element for the time
interval from t to (t + t) is rewritten to give the incremental
stress
s
(t) in terms of the mean modulus of elasticity

E
s
(t),
the incremental steel strain
s
(t) and incremental creep strain of
the tendon
s
(t) as

s
(t) =

E
s
(t)
_

s
(t)
s
(t)
_
. (16)
The strain vector {}
s
of the tendon element can be related to
the local displacement vector {u}
s
by the strain matrix [B]
s
as
{}
s
=
_
du
dx
_
s
= [B]
s
{u}
s
=
_
1/l
s
1/l
s
_
_
u
1
u
2
_
s
(17)
where l
s
is the length of tendon element. Following the con-
ventional formulation of finite element method and neglect-
ing body forces, the incremental nodal force vector {q
e
}
s
=
_
f
1
f
2
_
T
s
of the tendon element can be obtained by integrat-
ing over the volume V
e
and expressed as
_
q
e
_
s
=

V
e
[B]
T
s
{}
s
dV
=
_
V
e
[B]
T
s
_

E
s
(t)
_
[B]
s
dV
_
{u}
s

V
e
[B]
T
s
_

E
s
(t)
_

s
dV
_
q
e
_
s
= [

k]
s
{u}
s
+ {f }
s
(18)
in terms of the stiffness matrix [

k]
s
, incremental displacement
vector {u}
s
and incremental load vector due to tendon creep
{f }
s
. Note that the variable t has been omitted for brevity
hereafter. The stiffness matrix [

k]
s
is given by
[

k]
s
=

E
s
(t)A
s
l
s
_
1 1
1 1
_
(19)
where A
s
is cross sectional area. The incremental load vector due
to cable relaxation {f }
s
is given by
{f }
s
=

E
s
(t)
E
s
_
[
s
(t + t t
0
)
s
(t t
0
)]
_

N
s
(t
0
)

N
s
(t
0
)
_
+

t
t
0
[
s
(t + t )
s
(t )]
_

N
s
()

N
s
()
_
d
_
(20)
where

N
s
(t
0
) is the axial force of the element at time t
0
and

N
s
()
is the derivative of the axial force with respect to the dummy time
variable .
Similarly neglecting body forces, the incremental load vector of
a concrete beam element {q
e
}
c
=
_
f
1
f
2
f
12
_
T
c
can
be obtained as
_
q
e
_
c
= [

k]
c
{}
c
+ {f }
c
+ {f }
cs
(21)
in terms of the beam stiffness matrix [

k]
c
, incremental displace-
ment vector {}
c
, incremental load vector due to concrete creep
{f }
c
and incremental load vector due to concrete shrinkage
Author's personal copy
F.T.K. Au, X.T. Si / Engineering Structures 33 (2011) 118126 121
[

k]
c
=
_
[

k
11
]
c
[

k
12
]
c
[

k
21
]
c
[

k
22
]
c
_
(22a)
[

k
11
]
c
=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
A
c

E
c
L
c
0 0 0 0 0
0
12

E
c
I
cz
L
3
c
0 0 0
6

E
c
I
cz
L
2
c
0 0
12

E
c
I
cy
L
3
c
0
6

E
c
I
cy
L
2
c
0
0 0 0

G
c
J
c
L
c
0 0
0 0
6

E
c
I
cy
L
2
c
0
4

E
c
I
cy
L
c
0
0
6

E
c
I
cz
L
2
c
0 0 0
4

E
c
I
cz
L
c
_

_
(22b)
[

k
21
]
c
= [

k
12
]
T
c
=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_

A
c

E
c
L
c
0 0 0 0 0
0
12

E
c
I
cz
L
3
c
0 0 0
6

E
c
I
cz
L
2
c
0 0
12

E
c
I
cy
L
3
c
0
6

E
c
I
cy
L
2
c
0
0 0 0

G
c
J
c
L
c
0 0
0 0
6

E
c
I
cy
L
2
c
0
2

E
c
I
cy
L
c
0
0
6

E
c
I
cz
L
2
c
0 0 0
2

E
c
I
cz
L
c
_

_
(22c)
[

k
22
]
c
=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
A
c

E
c
L
c
0 0 0 0 0
0
12

E
c
I
cz
L
3
c
0 0 0
6

E
c
I
cz
L
2
c
0 0
12

E
c
I
cy
L
3
c
0
6

E
c
I
cy
L
2
c
0
0 0 0

G
c
J
c
L
c
0 0
0 0
6

E
c
I
cy
L
2
c
0
4

E
c
I
cy
L
c
0
0
6

E
c
I
cz
L
2
c
0 0 0
4

E
c
I
cz
L
c
_

_
(22d)
Box II.
{f }
cs
. The stiffness matrix [

k]
c
is given in Box II, where

E
c
is the
mean modulus of elasticity of concrete in the time step concerned
given as

E
c
(t) =
[E
c
(t) + E
c
(t + t)] /2
1 +
c
(t + t, t)/2
. (23)

G
c
is the mean shear modulus given in terms of the Poissons ratio

c
as

G
c
(t) =

E
c
(t)
2(1 +
c
)
. (24)
L
c
is the length of element, A
c
is the cross sectional area, I
cy
and I
cz
are the second moments of area about the y- and z-axes
respectively, and J
c
is the torsional constant. In accordance with
the approach adopted by Au et al. [15], the axial force and bending
moments in a member with member loads are decomposed
into time-dependent components (governed by overall time-
dependent deformations) and time-independent components
(governed by member loading). The incremental load vector due
to creep {f }
c
is given by
{f }
c
=
_
{f
1
}
c
{f
2
}
c
_
(25a)
Author's personal copy
122 F.T.K. Au, X.T. Si / Engineering Structures 33 (2011) 118126
{f
1
}
c
=

E
c
(t)
_

c
(t + t, t
0
)
c
(t, t
0
)
E
c
(t
0
)
_

_
_
_
_
_
_
_
_
_

N
c
(t
0
)
_

M
c1
(t
0
) +

M
c2
(t
0
)
_
/L
c
_

M
c3
(t
0
) +

M
c4
(t
0
)
_
/L
c

T
c
(t
0
)

M
c3
(t
0
)

M
c1
(t
0
)
_

_
+

1
0
_

_
N
c0
( x)
6(2 x 1)M
c0
( x)/L
c
6(2 x 1)M
c0
( x)/L
c
T
c0
( x)
2(3 x 2)M
c0
( x)/L
c
2(3 x 2)M
c0
( x)/L
c
_

_
d x
_
_
_
_
_
_
_
_

E
c
(t)

t
t
0

c
(t + t, )
c
(t, )
E
c
()

N
c
()
_

M
c1
() +

M
c2
()
_
/L
c
_

M
c3
() +

M
c4
()
_
/L
c

T
c
()

M
c3
()

M
c1
()
_

_
d (25b)
{f
2
}
c
=

E
c
(t)
_

c
(t + t, t
0
)
c
(t, t
0
)
E
c
(t
0
)
_

_
_
_
_
_
_
_
_
_

N
c
(t
0
)
_

M
c1
(t
0
)

M
c2
(t
0
)
_
/L
c
_

M
c3
(t
0
)

M
c4
(t
0
)
_
/L
c

T
c
(t
0
)

M
c4
(t
0
)

M
c2
(t
0
)
_

_
+

1
0
_

_
N
c0
( x)
6(1 2 x)M
c0
( x)/L
c
6(1 2 x)M
c0
( x)/L
c
T
c0
( x)
2(3 x 1)M
c0
( x)/L
c
2(3 x 1)M
c0
( x)/L
c
_

_
d x
_
_
_
_
_
_
_
_

E
c
(t)

t
t
0

c
(t + t, )
c
(t, )
E
c
()

N
c
()
_

M
c1
()

M
c2
()
_
/L
c
_

M
c3
()

M
c4
()
_
/L
c

T
c
()

M
c4
()

M
c2
()
_

_
d (25c)
where the location of a point in the element is defined by x =
x/L
c
,

N
c
(t
0
) is the axial force at time t
0
, T
c
(t
0
) is the torsional
moment at time t
0
,

M
c1
(t
0
) and

M
c2
(t
0
) are the bending moments
about the z-axis of element at time t
0
(t
0
t) at the ends x = 0
and x = 1 respectively, and

M
c3
(t
0
) and

M
c4
(t
0
) are the correspon-
ding end moments about the y-axis, and the over-dot denotes
differentiation with respect to the dummy time variable . The
time-independent forces N
c0
( x) and T
c0
( x) can be calculated from
applied loads assuming both ends of the element to be fixed,
while the time-independent moment M
c0
( x) can be calculated
from applied loads assuming the element to be simply supported.
If all loads act at the nodes only, the integral in Eq. (25) involving
the time-independent terms will disappear. The incremental load
vector due to concrete shrinkage is given by
{f }
cs
=

E
c
(t)A
c
[
cs
(t + t, t
0
)
cs
(t, t
0
)]

_
1 0 0 0 0 0 1 0 0 0 0 0
_
T
(26)
where
cs
is the concrete shrinkage strain from time t
0
to the
specified time.
In each time interval, the incremental load vector for a beam
element with a tendon {q
e
}
bt
=
_
f
1
f
2
f
12
_
T
bt
can
be written as
_
q
e
_
bt
=
_
[

k]
c
+ [

k]
t
_
{}
c
+ {f }
c
+ {f }
s
+ [N]
T
[H]
T
[T]
T
{f }
s
(27)
where [

k]
t
is the stiffness matrix of the tendon element based on
the mean elastic modulus

E
s
(t), namely
[

k]
t
= [H]
T
[T]
T
[

k]
s
[T] [H] . (28)
Before assembling these local stiffness matrices and load
vectors into the corresponding global matrices and vectors for
solution, they must be transformed to the global coordinate system
in the usual manner. Sufficiently fine time steps have to be used
in the time integration in order to obtain accurate results. Elastic
analysis for time t
0
is first performed to get the initial deformations
and internal forces. The procedures for the subsequent time-
dependent analysis using time integration for each of the time
intervals from t
0
to the time of interest are summarized below.
(a) Estimate or update the mean moduli of the concrete and
tendon elements.
(b) Based on the updated mean moduli, calculate the local stiffness
matrices and incremental load vector due to concrete creep,
concrete shrinkage and cable relaxation.
(c) Transformand assemble the local stiffness matrices and the lo-
cal incremental load vectors to formthe global stiffness matrix
_

K
_
and global incremental load vector {Q} respectively.
(d) Solve the equation
_

K
_
{} = {Q} to obtain the incremental
displacement vector {}.
(e) Fromthe incremental element load vectors estimated fromthe
incremental displacement vector {}, calculate the total load
vector and nodal displacements.
(f) Repeat Steps (a)(e) until the time of interest.
5. Case studies
5.1. Simulation of creep coefficients of steel tendons
Define the initial prestressing ratio for a prestressed tendon in
terms of initial prestress
s
(t
0
) and yield stress of the tendon f
py
as
R =
s
(t
0
)/f
py
. (29)
Consider a prestressing tendon of unit length fixed at two ends.
Based on the stress relaxation functions shown in Eqs. (8) and (9),
the creep coefficients and stress relaxation of a stress-relieved and
a low relaxation tendon under various initial prestressing ratios R
are simulated respectively for 10 years. The time interval adopted
is 8.76 h, which is one thousandth of one year, while the yield
stress is 1670 MPa.
It is observed that the variations of the creep coefficient shown
in Fig. 2(a) are largely similar to the variations of percentage
stress relaxation shown in Fig. 2(b). Both the creep coefficient
and cable relaxation are higher for higher initial prestressing
ratios. Both of them increase rapidly at first and tend to stabilize
Author's personal copy
F.T.K. Au, X.T. Si / Engineering Structures 33 (2011) 118126 123
(a) Equivalent creep coefficients. (b) Simulated stress relaxation (markers) and assumed stress
relaxation (curves).
Fig. 2. Computed properties of a typical stress relieved tendon.
Fig. 3. Comparison of computed properties of stress relieved and low relaxation
tendons with R = 0.8.
l
s
Fig. 4. A prestressing tendon connected to a perfectly elastic spring.
after a few years. To verify that the equivalent creep coefficient
works, the time-dependent behavior of the tendon of unit length
fixed at two ends is solved by time integration based on the
equivalent creep coefficients. There is perfect agreement between
the simulated percentage stress relaxation shown as markers in
Fig. 2(b) and the assumed relaxation functions shown as curves.
The equivalent creep coefficients and percentage stress relaxation
for low relaxation tendons and stress-relieved tendons for initial
prestressing ratio R = 0.8 are similarly worked out and shown
in Fig. 3. In particular, the simulated stress relaxation results are
shown as markers while the assumed stress relaxation functions
are shown as curves. Apart from verifying again the validity of the
equivalent creep coefficient, the superiority of the low relaxation
tendons is clearly seen.
5.2. A prestressing tendon connected to a perfectly elastic spring
In order to evaluate whether a time-dependent structural
model works in structural analysis, a common test problem is
a model with partial fixity [23]. Fig. 4 shows a stress-relieved
prestressing tendon of length l
s
fixed at the left end and connected
to a perfectly elastic spring of stiffness K at the right end. In
particular, the tendon-spring assembly is fixed at the left end first
and then tensioned at the right end. When the initial prestressing
ratio reaches R = 0.7, the right end of the spring is fixed. The
relevant data of the tendon include: sectional area A
s
= 924
10
6
m
2
, Youngs modulus E
s
= 210 GPa, yield stress f
py
=
1670 MPa and length l
s
= 1 m. The time span investigated is 10
years and the time interval for analysis is 8.76 h. The stiffness ratio
R
K
is defined as the ratio of spring stiffness K to that of the tendon
K
t
as
R
K
=
K
K
t
=
Kl
s
E
s
A
s
. (30)
The cases studied cover various values of stiffness ratio, i.e. R
K
= 0,
0.5, 1, 10, 100 and .
Fig. 5(a) shows the percentage increase in tendon strain while
Fig. 5(b) shows that percentage stress relaxation in the tendon. For
R
K
= 0 which corresponds to an infinitely soft spring, the curve in
Fig. 5(a) is identical to the creep coefficient for the case R = 0.7
shown in Fig. 2(a), while there is no stress relaxation as shown
in Fig. 5(b). Fig. 5(b) also shows that the stress relaxation in the
tendon generally increases with the spring stiffness. For R
K
=
which means there is no change in strain, the curve in Fig. 5(b) is
identical to the stress relaxation for the case R = 0.7 as shown in
Fig. 2(b). The results show that this relaxation model works under
various boundary conditions.
5.3. A simply supported prestressed concrete beam with a parabolic
tendon
Fig. 6 shows a prestressed concrete beam of negligible weight
having a section of breadth b = 0.6 mand depth h = 1.2 m, which
is simply supported over a span of 40 m. The characteristic
compressive strength of concrete is f
ck
= 32 MPa. Wet curing is
carried out until T
s
= 3 days after which shrinkage begins. The
relative humidity is takenas 80%throughout. At the age of T
c0
= 28
days, the beam is post-tensioned to an initial prestressing force
of P
0
= 1108 kN with a stress-relieved prestressing tendon of
cross sectional area A
s
= 924 10
6
m
2
and Youngs modulus
E
s
= 200 GPa. The initial prestressing ratio is R = 0.8, which is
slightly on the high side for the study of relaxation. The tendon
profile is parabolic with a maximum eccentricity of 0.3 m at mid-
span and zero end eccentricities. The parameters of CEB-FIP Model
Code 1990[24] are adoptedinthe calculation. The beamis modeled
by 20 identical beam elements with tendon elements connected
to the beam axis using the masterslave technique. Analysis of
the time-dependent behavior is carried out by time integration
taking into account the interaction among concrete creep, concrete
Author's personal copy
124 F.T.K. Au, X.T. Si / Engineering Structures 33 (2011) 118126
(a) Percentage increase in tendon strain. (b) Percentage stress relaxation.
Fig. 5. Time-dependent behavior of a tendon connected to a perfectly elastic spring.
Fig. 6. A simply supported beam with a parabolic tendon (not to scale).
Fig. 7. Time-dependent deformation of simply supported beam.
shrinkage and cable relaxation for the period from t
0
= 28 days to
t = 365days by 2000equal time steps. The self-weight of the beam
is ignored in the present analysis.
To investigate the effects of various factors on the time-
dependent behavior, hypothetical estimates of the axial shortening
along the beam axis and the upward deflection mid-span are
worked out for different scenarios when selected parameters are
set to zero. For convenience in the present study, it is assumed
that the losses of prestress due to friction and anchorage draw-
in are negligible. Fig. 7 shows the hypothetical estimates of axial
shortening and vertical deflection due to concrete creep, concrete
shrinkage with and without cable relaxation. It is observed from
the results after one year that cable relaxation reduces the
axial shortening and vertical deflection due to concrete creep
and shrinkage by 8.0% and 17.5% respectively. The hypothetical
estimates of loss of prestress at mid-span due to cable relaxation
withand without concrete creepand shrinkage are worked out and
shown in Fig. 8. Also shown are hypothetical estimates of stress
at bottom fibre at mid-span due to concrete creep and shrinkage
with and without cable relaxation. Concrete creep and shrinkage
therefore increase the loss of prestress due to cable relaxation by
27.2% while cable relaxation reduces the stress at bottom fibre by
8.9%. So the interaction among concrete creep, concrete shrinkage
and cable relaxation should be considered carefully for accurate
time-dependent analysis.
For estimation of the total long-term loss of prestress
ps
due to concrete creep, concrete shrinkage and steel relaxation for
Fig. 8. Loss of prestress and stress at bottomfibre at mid-span of simply supported
beam.
sections having a single layer of prestressing steel without non-
prestressed steel, CEB-FIP Model Code 1990 [24] gives the equation

ps
=

ps

c
(t, t
0
) f
cgp
+ E
s

cs
+ 0.8
pr
1 +
ps
A
s
A
c
_
1 +
A
c
y
2
ps
I
c
_
[1 +
c
(t, t
0
)]
(31)
where
ps
= E
s
/E
c
(t
0
) is the ratio of the modulus of elasticity of
prestressing steel to that of concrete, f
cgp
is the concrete stress at
centroid of prestressing steel at transfer, A
s
is the cross sectional
area of prestressing steel, y
ps
is the y-coordinate of prestressing
steel measured fromcentroidal axis of beamsection, A
c
is the cross
sectional area of beam, is the ageing coefficient that may be taken
as 0.8 [21], and the other variables have been defined before. This
formula for the evaluation of loss of prestress is one of the most
comprehensive compared with various codes currently in use.
Consider a concrete beam with a much larger cross sectional area
compared with the tendon (i.e. A
s
/A
c
0) made with concrete
with negligible creep (i.e.
c
(t, t
0
) 0) and shrinkage (i.e.
cs

0). Eq. (31) gives
ps

= 0.8
pr
rather than the intrinsic stress
relaxation
pr
, which means that Eq. (31) cannot be degenerated
to this extreme case and it tends to under-estimate the loss of
prestress in this case.
Predictions of the loss of prestress at mid-span of the beam
calculated using Eq. (31) from day 28 to day 365 due to creep,
shrinkage with and without relaxation are compared to those
obtained by the proposed method in Fig. 9. It shows that the CEB-
FIP Model Code 1990 [24] is conservative in predicting loss of
prestress, with substantial over-estimation in the presence of cable
relaxation.
5.4. A concrete cantilever with a stay cable
The long-term behavior of the hypothetical concrete cantilever
with a stay cable [11] as shown in Fig. 10 is investigated with
a more complete set of parameters using the proposed time
Author's personal copy
F.T.K. Au, X.T. Si / Engineering Structures 33 (2011) 118126 125
Fig. 9. Loss of prestress at mid-span of simply supported beam calculated by
various methods based on different assumptions.
Fig. 10. A concrete cantilever with a stay cable (not to scale).
integration method. The cantilever is 10 m in length with a square
cross section of 1 m 1 m and weight density of 25 kN/m
3
. The
characteristic compressive strength of concrete is f
ck
= 36 MPa.
Wet curing is carried out until T
s
= 3 days after which shrinkage
begins. The relative humidity is taken as 80% throughout. The stay
cable is a stress-relieved tendon with a cross sectional area A
s
=
250 mm
2
, Youngs modulus E
s
= 200 GPa and initial tension P
0
=
200 kN applied at time t
0
= 28 days with reference to the age of
cantilever. The initial prestressing ratio is R = 0.8. For simplicity,
the beam is assumed to be supported on props along the length
initially. As the stay cable is tensioned, the props are gradually
lowered to transfer part of the weight of beam to the stay cable.
The parameters of CEB-FIP Model Code 1990 [24] are adopted
in the calculation. The beam is modeled by four identical beam
elements while the cable is modeled by a truss element. In the time
integration method, the period from t
0
= 28 days to t = 365 days
is divided into 2000 equal time steps.
The sag effect of the stay cable can be taken into account by
replacing the modulus of elasticity E
s
with the Ernst modulus
E
eq
[25] given by
E
eq
=
E
s
1 +
2
L
2
E
s
/12
3
s
(32)
where is the weight per unit volume of cable, L is the horizontal
projected length of the cable, and
s
is the tensile stress in the
cable. The Ernst modulus of the cable in this example is calculated
as 199.996 GPa assuming a bare steel cable, which indicates that
the sag effect has negligible influence and can be neglected during
the time-dependent analysis in this case. For longer stay cables in
which the sag effect is more significant, the Ernst modulus can be
updated for relaxation in each time step for accurate analysis.
The total axial shortening along the centroidal axis of the
beam and the vertical deflection at point A are predicted for
different scenarios. Fig. 11 shows the hypothetical estimates of
axial beam shortening and vertical deflection at A due to concrete
creep and shrinkage with and without cable relaxation. Comparing
the hypothetical estimates with and without cable relaxation
Fig. 11. Time-dependent deformation at point A of concrete cantilever.
Fig. 12. Loss of prestress in cable-stayed beam.
shows that, one year after casting the cantilever, cable relaxation
reduces the axial shortening by 1.4% but increases the vertical
deflection at A by 136%. Similarly Fig. 12 shows the hypothetical
estimates of loss of prestress in the cable due to cable relaxation
with and without concrete creep and shrinkage. Comparing these
hypothetical estimates shows that concrete creep and shrinkage
tend to reduce the loss of prestress in cable slightly. It should be
mentioned that the rather high initial prestressing ratio of R = 0.8
is for illustrational purposes. The much lower stress levels adopted
in practical stay cables implies that relaxation should not be as
significant as compared to tendons inprestressed concrete bridges.
Nevertheless it shows that the interactions among concrete creep,
concrete shrinkage and cable relaxation should be considered
carefully.
6. Conclusions
Based on the equivalent creep coefficient, a new relaxation
model has been developed for steel tendons to enable the accurate
estimation of losses of cable forces. The numerical investigations
show that the equivalent creep coefficient works not only in
the case of intrinsic relaxation but also under various boundary
conditions. Accurate analysis of time-dependent behavior of
concrete bridges considering concrete creep, concrete shrinkage
and cable relaxation is made possible by the implementation
of the finite element method using time integration with the
proposed relaxation model for cables. Apart from demonstrating
the versatility of the proposed method, the case studies also prove
the need for accounting for interactions among various factors in
the accurate assessment of time-dependent behavior.
Acknowledgement
The work described in this paper has been supported by
the Research Grants Council (RGC) of the Hong Kong Special
Administrative Region, China (RGC Project No. HKU 7102/08E).
Author's personal copy
126 F.T.K. Au, X.T. Si / Engineering Structures 33 (2011) 118126
References
[1] Baant ZP. Prediction of concrete creep and shrinkage: past, present and
future. Nucl Eng Des 2001;203(1):2738.
[2] Gilbert IR. Time effects in concrete structures. New York: Elsevier; 1988.
[3] Yang IH. Uncertainty and sensitivity analysis of time-dependent effects in
concrete structures. Eng Struct 2007;29(7):136674.
[4] Mar AR, Bairn JM, Duarte N. Long-term deflections in cracked reinforced
concrete flexural members. Eng Struct 2010;32(3):82942.
[5] Dezi L, Leoni G, Tarantino AM. Creep and shrinkage analysis of composite
beams. Progr Struct Eng Mater 1998;1(4):38491.
[6] Bradford MA. Generic modelling of composite steelconcrete slabs subjected
to shrinkage, creep and thermal strains including partial interaction. Eng
Struct 2010;32(5):145965.
[7] Ariyawardena N. Prestressed concrete with internal or external tendons:
behavior and analysis. Ph.D. thesis. Calgary (Alberta): University of Calgary;
2000.
[8] Benboudjema F, Meftah F, Torrenti JM. Interaction between drying, shrink-
age, creep and cracking phenomena in concrete. Eng Struct 2005;27(2):
239250.
[9] Yang IH. Prediction of time-dependent effects in concrete structures using
early measurement data. Eng Struct 2007;29(10):270110.
[10] Malm R, Sundquist H. Time-dependent analyses of segmentally constructed
balanced cantilever bridges. Eng Struct 2010;32(4):103845.
[11] Ghali A, Favre R, Elbadry M. Concrete structures: stresses and deformations.
3rd ed. London: Spon Press; 2002.
[12] Aalami BO. Time-dependent analysis of concrete structures. Progr Struct Eng
Mater 1998;1(4):38491.
[13] Elbadry MM, Ghali A. Analysis of time-dependent effects inconcrete structures
using conventional linear computer programs. Canad J Civ Eng 2001;28(2):
190200.
[14] Ariyawardena N, Ghali A. Prestressing with unbonded internal or external
tendon: analysis and computer model. J Struct Eng 2002;128(12):1493501.
[15] Au FTK, Liu CH, Lee PKK. Creep and shrinkage analysis of reinforced concrete
frames by history-adjusted and shrinkage-adjusted elasticity moduli. Struct
Des Tall Special Build 2009;18:1335.
[16] Tadros MK, Ghali A, Dilger WH. Time-dependent prestress loss and deflection
in prestressed concrete members. J Prestr Concr Inst 1975;20(3):8698.
[17] Ghali A, Trevino J. Relaxation of steel in prestressed concrete. J Prestr Concr
Inst 1985;30(5):8294.
[18] Sharif A, Taher SF, Basu PK. Time-dependent losses in prestressed, continuous
composite beams. J Struct Eng 1993;119(11):315168.
[19] Ghali A, Neville AM. Structural analysis: a unified classical and matrix
approach. 3rd ed. New York: Chapman and Hall; 1989.
[20] Magura DD, Sozen MS, Siess CP. A study of stress relaxation in prestressing
reinforcements. J Prestr Concr Inst 1964;9(2):1357.
[21] Youakim SA, Ghali A, Hida SE, Karbhari VM. Prediction of long-term prestress
losses. J Prestr Concr Inst 2007;52(2):11630.
[22] British Standards Institution. BS8110-1: 1997 structural use of concretepart
1: code of practice for design and construction. London; 1997.
[23] Liu CH, Au FTK, Lee PKK. Estimation of shrinkage effects on reinforced concrete
podiums. HKIE Trans 2006;13(4):3343.
[24] Comit Euro-International du Bton. CEB-FIP model code 1990. London:
Thomas Telford; 1993.
[25] Curley NC, Shepherd R. Analysis of concrete cable-stayed bridges for creep,
shrinkage and relaxation effects. Comput Struct 1996;58(2):33750.