You are on page 1of 17

EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2005; 34:595611


Published online 3 March 2005 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/eqe.434
Non-linear viscoelastic modelling of earthquake-induced
structural pounding
Robert Jankowski
;
Faculty of Civil and Environmental Engineering; Gda nsk University of Technology;
ul. Narutowicza 11=12; 80-952 Gda nsk; Poland
SUMMARY
Past severe earthquakes indicate that structural pounding may cause considerable damage or even lead
to collapse of colliding structures if the separation distance between them is not sucient. Because of
its complexity, modelling of impact is an extremely dicult task, however, the precise numerical model
of pounding is essential if an accurate structural response is to be simulated. The aim of this paper is
to analyse a non-linear viscoelastic model of collisions which allows more precise simulation of the
structural pounding during earthquakes. The eectiveness of the model is veried by comparing the
results of numerical analyses with the results of experiments conducted on pounding between dierent
types of structures. The results of the study indicate that, compared to other models, the proposed
non-linear viscoelastic model is the most precise one in simulating the pounding-involved structural
response. Copyright ? 2005 John Wiley & Sons, Ltd.
KEY WORDS: structural pounding; earthquakes; impact force; non-linear model, Hertz contact law
INTRODUCTION
During severe earthquakes, pounding between neighbouring, inadequately separated build-
ing structures or bridge segments has been repeatedly observed. Rosenblueth and Meli [1]
reported that in the Mexico City earthquake of 1985 about 40% of the damaged
structures experienced some level of pounding, 15% of them leading to structural collapse.
Anagnostopoulos [2] re-assessed that statement to say that evidence of pounding was found
in 15% of buildings with major damage or collapse and in 2030% of these cases pounding
could have been a signicant factor in the structural damage. During the 1989 Loma Prieta
earthquake, over 200 pounding occurrences involving more than 500 buildings were observed

Correspondence to: Robert Jankowski, Faculty of Civil and Environmental Engineering, Gda nsk University of
Technology, ul. Narutowicza 11=12, 80-952 Gda nsk, Poland.

E-mail: jankowr@pg.gda.pl
Contract=grant sponsor: European Community FP5 Programme; contract=grant number: EVK4-CT-2002-80005
Received 16 May 2004
Revised 26 August 2004 and 11 October 2004
Copyright
?
2005 John Wiley & Sons, Ltd. Accepted 11 October 2004
596 R. JANKOWSKI
at sites located over 90km from the epicentre [3]. Signicant pounding damage was observed
at expansion hinges and abutments of standing portions of a number of bridges during the
1994 Northridge earthquake at the Interstate 5 and State Road 14 interchange [4]. The reports
of damage to highway bridges during the Kobe earthquake of 1995 have identied pounding
due to fracture of bearing supports as a reason for local damage and a potential contribution
to the falling down of bridge decks [5]. The pounding-involved structural damage during other
past earthquakes has been also reported [6].
The main factor recognized as a reason for the pounding of buildings is usually the dier-
ence in dynamic characteristics of adjacent structures [79]. This is due to the fact, that the
dierence in mass or stiness induces out-of-phase vibrations which may lead to structural
interaction during the time of earthquake. On the other hand, for the longer bridge structures,
it is often the seismic wave propagation eect that is considered to be a dominant factor lead-
ing to pounding of neighbouring superstructure segments [1012]. This eect, due to time
lag and spatial variation of seismic wave, results in dierent seismic input acting on supports
along the structure [13].
Structural pounding is a complex phenomenon involving plastic deformations at contact
points, local cracking or crushing, fracturing due to impact, friction, etc. Forces created by
collisions are applied and removed during a short interval of time initiating stress waves which
travel away from the region of contact. The process of energy transfer during impact is highly
complicated which makes the mathematical analysis of this type of problem dicult.
Structural pounding during earthquakes has been intensively studied recently by applying
various structural models and using dierent models of collisions. The fundamental study
on pounding between buildings in series using a linear viscoelastic model of collisions, has
been conducted by Anagnostopoulos [14]. Jankowski et al. [10] used the same model to study
pounding of superstructure segments in bridges. Further analyses were carried out by applying
discrete multi-degree-of-freedom structural models and using FEM, though often incorporating
simplied models of collisions (see, for example, References [8, 9, 11, 15]). In order to model
the impact forcedeformation relation more realistically, a non-linear elastic model following
the Hertz law of contact has been adopted by a number of researchers [1618].
Intensive study has also been carried out on mitigation of pounding hazards. One of the ob-
jectives is to develop procedures for evaluating adequate separation distance between structures
in order to prevent contact during earthquakes [19]. The minimum seismic gap is specied in
most recent seismic-resistant design codes. Another aim is to enhance the seismic performance
of existing structures without sucient in-between space and to develop proper approaches for
reducing the pounding eects on structural members. Westermo suggested linking buildings
with beams which can transmit the forces between them eliminating dynamic contacts [20].
The idea of lling the separation gap by energy absorbing material, providing bumpers or
strong collision walls protecting parts of the structures, has been studied [2]. Jankowski et al.
[15] considered also the use of dampers, crushable devices and shock transmission units which
provide sti linking of structural members during earthquakes and do not impose undesired
forces resulting from thermal elongation, creep or shrinkage eects.
The aim of this paper is to analyse a non-linear viscoelastic model of collisions which
allows more precise simulation of the structural pounding. The model can be applied also
for the assessment of the maximum pounding force value which is required for the design
purposes of dierent types of links, bumpers or crushable devices placed between structures
or structural members. The eectiveness of the model is veried by comparing the results of
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
MODELLING OF EARTHQUAKE-INDUCED STRUCTURAL POUNDING 597
numerical analyses with the results of experiments conducted for pounding between dierent
types of structures.
CLASSICAL THEORY OF IMPACT
There are generally two dierent approaches to modelling of structural pounding. The rst
one applies the classical theory of impact, called stereomechanics, which is based on the
laws of conservation of energy and momentum and does not consider transient stresses and
deformations in the impacting bodies [2123]. The theory focuses on determination of post-
impact velocities of colliding bodies based on the approaching velocities prior to contact
and a coecient of restitution which accounts for the energy dissipation during impact in-
corporating response non-linearities. The formulae for the post-impact velocities v

1
and v

2
of
two non-rotating bodies with masses m
1
and m
2
in the case of the central impact are
given by [21]:
v

1
=v
1
(1 + e)
m
2
v
1
m
2
v
2
m
1
+ m
2
v

2
=v
2
+ (1 + e)
m
1
v
1
m
1
v
2
m
1
+ m
2
(1)
where v
1
and v
2
are approaching velocities and e is a coecient of restitution which can be
obtained from the equation
e =
v

2
v

1
v
1
v
2
(2)
A value of e =1 deals with the case of a fully elastic collision, and a value of e =0 with a
fully plastic one. The value of the coecient of restitution can be determined experimentally
by dropping a sphere on a massive plane plate of the same material from a height h and
observing the rebound height h

. Then, the following formula is used [21]:


e
2
=
h

h
(3)
It has been assessed that the coecient of restitution used to simulate real collisions between
structures ranges usually from 0.5 to 0.75 (see Reference [7]). Based on the experimental
results, Azevedo and Bento [24] suggested that e =0:65 should be used for typical concrete
structures. In fact, this value has been used by a number of researchers in the analysis of
pounding between dierent types of structures (see, for example, References [9, 10, 14]). Some
of the studies indicate, however, that collisions between structural members can be more plastic
in some cases. Zhu et al. [25], for example, obtained the value of e =0:4 based on the results
of the experiments conducted on a steel bridge girder model.
Owing to its macroscopic approach, the classical theory of impact is rather not recommended
when a precise pounding-involved structural response is required, especially in the case of
multiple impacts. Moreover, since it does not trace the structural response during contact,
assuming that it lasts a negligibly short time, its application is usually limited to the analysis
of pounding between two structures modelled as single-degree-of-freedom systems [23]. In the
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
598 R. JANKOWSKI
case when the structures are simulated by multi-degree-of-freedom models or when the study
on the pounding of buildings in series or between several segments of a bridge is conducted,
the structural response during the time when contact takes place is essential. This is due to the
fact, that when the structural members rebound after collision they might come into contact
with other members. Moreover it may also happen that at the time of contact between two
given structural members other members may collide with each other.
MODELLING OF POUNDING FORCE
The second approach to modelling of pounding is to simulate directly the pounding force
during impact. The experimental results [21, 26, 27] have shown that pounding force history
depends substantially on a number of factors, such as masses of colliding structures, their rel-
ative velocity before impact, structural material properties, contact surface geometry and pre-
vious impact history. The examples of experimentally obtained loadtime diagrams of impact
between concrete elements [26] are shown in Figure 1. Similar shapes of pounding force histo-
ries have been also observed for steel-to-concrete as well as steel-to-steel impacts for dierent
contact surface geometries, mass values and impact velocities of colliding bodies [21, 27].
The pounding force time history during impact consists of two sub-intervals. The approach
period extends from the beginning of contact up to the maximum deformation. It is followed
by a restitution period lasting until the separation. The beginning of the approach period
is attributed to elastic material behaviour but soon plastic deformations, local cracking or
crushing usually take place. In the restitution period, the accumulated elastic strain energy
is released without considerable plastic eects. In addition, during the whole time of impact,
friction between the colliding members takes place and this eect is especially important in
the case of rough surfaces. It has been shown that most of the energy which is dissipated
during impact is lost during the approach period of collision and a comparably small amount
of energy is lost during the restitution period due to friction [21]. The experimental results
0
20
40
60
80
100
120
0 2 4 6 8 10 12
test 1
test 2
P
o
u
n
d
i
n
g

f
o
r
c
e

(
k
N
)
Time (ms)
0
50
100
150
200
250
300
350
0 2 4 6 8 10 12
test 1
test 2
P
o
u
n
d
i
n
g

f
o
r
c
e
(
k
N
)
Time (ms)
(b) (a)
Figure 1. Examples of experimentally obtained pounding force time histories for impact between concrete
elements [26]: (a) impact velocity 0:5 m=s; and (b) impact velocity 2:5 m=s.
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
MODELLING OF EARTHQUAKE-INDUCED STRUCTURAL POUNDING 599
(see Figure 1) indicate that during the approach period, the relatively rapid increase in the
pounding force is observed. On the other hand, during the restitution period, the force de-
creases with lower unloading rate which is reduced even more just before separation.
The pounding force between colliding structures is usually modelled by the use of elastic or
viscoelastic impact elements which become active when contact is detected. Several types of
such elements, e.g., a linear spring, a bi-linear spring, a linear spring-damper and a non-linear
spring, have been used for the simulation purposes. The simplest impact element consists
of a linear elastic spring and does not account for the energy dissipation during collision
[8, 11, 28, 29]. The bi-linear spring element with dierent approaching and separation stiness
has been described by Valles and Reinhorn [30]. This model includes some energy dissipation
due to the hysteretic behaviour.
Linear viscoelastic model
The most frequently used type of an impact element is a linear spring-damper (KelvinVoigt
model) [7, 10, 14, 30]. The pounding force during impact, F(t), for this linear viscoelastic
model is expressed as
F(t) =k(t) + c

(t) (4)
where (t) describes the deformation of colliding structural members,

(t) denotes the relative
velocity between them, k is the impact elements stiness simulating the local stiness at
the contact point and c is the impact elements damping which can be obtained from the
formula [14]:
c =2

k
m
1
m
2
m
1
+ m
2
(5)
where m
1
and m
2
are masses of structural members and is a damping ratio correlated with
a coecient of restitution, e, by the equation [14]:
=
ln e

2
+ (ln e)
2
(6)
The disadvantage of the linear spring-damper element is that its viscous component is
active with the same damping coecient during the whole time of collision. This results in
the uniform dissipation of energy during the approach and restitution periods which is not
fully consistent with the reality [21, 30]. As was mentioned earlier, most of the energy is
dissipated during the approach period and the restitution period is mainly attributed to elastic
behaviour where the accumulated elastic strain energy is released with minor energy loss.
Nevertheless, owing to its simplicity, the linear viscoelastic model has been widely used for
the simulation purposes of structural pounding.
Non-linear elastic model
In order to model the pounding forcedeformation relation more realistically, a non-linear
elastic spring following the Hertz law of contact has been adopted by a number of researchers
[1618, 21]. The pounding force during impact, F(t), for this type of impact element is
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
600 R. JANKOWSKI
expressed by the formula
F(t) =
3
2
(t) (7)
where is the impact stiness parameter which depends on material properties and the geom-
etry of colliding bodies. Results of the experiments indicate that for impacts between concrete
elements, it ranges typically from 40 to 80 kN=mm
3=2
(1:2 10
9
2:6 10
9
N=m
3=2
) depending
mainly on the contact surface geometry [26].
The impact stiness parameter for steel-to-steel impacts takes usually higher values [21, 31].
The formulae to calculate values of for certain special impact cases, such as impacts be-
tween two spheres or between a sphere and a massive plane surface, have been given by
Goldsmith [21].
The disadvantage of the Hertz contact law model is that it is fully elastic and does not
account for the energy dissipation during contact due to plastic deformations, local cracking,
friction, etc.
Non-linear viscoelastic model
The aim of this paper is to analyse a non-linear viscoelastic impact element model which
overcomes the disadvantages of the linear viscoelastic and the non-linear elastic models and
thus can be used to simulate structural pounding more precisely. In the model proposed, a non-
linear spring following the Hertz law of contact is applied. Additionally, a non-linear damper
is activated during the approach period of the collision in order to simulate the process of
energy dissipation which takes place mainly during that period. For the simplicity reasons,
the minor energy loss during the restitution period is neglected in the model. The pounding
force during impact, F(t), for this type of impact element is expressed as
F(t) =

3
2
(t) + c(t)

(t) for

(t)0 (approach period)
F(t) =

3
2
(t) for

(t)60 (restitution period)
(8)
where

is the impact stiness parameter and c(t) is the impact elements damping which at
any instant of time can be obtained from the formula
c(t) =2

(t)
m
1
m
2
m
1
+ m
2
(9)
where

denotes a damping ratio correlated with a coecient of restitution, e.
Expressing the impact elements damping, c(t), by Equation (9), which can be considered
as the extension of Equation (5) to the non-linear case, allows us to dene the non-linear
viscoelastic model by two constant parameters,

and

, where

is independent from

,
similarly to the case of the linear viscoelastic model. Values of both parameters should be
determined based on the results of experiments. First, the damping ratio,

, should be obtained
for a given value of coecient of restitution, e, which is to be determined experimentally with
the help of Equation (2) or Equation (3). Since, however, both the spring and the damper
are non-linear in the model, the relation between

and e cannot be expressed by a relatively
simple formula (see Equation (6)) derived for the linear viscoelastic model by equating the
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
MODELLING OF EARTHQUAKE-INDUCED STRUCTURAL POUNDING 601
energy losses during impact [32]. It is suggested, therefore, that the appropriate value of

for a specied value of e should be obtained numerically through iterative simulations in


order to satisfy the relation between the post-impact and the prior-impact velocities dened
by Equation (2). It should be noted, that because the damper in the proposed non-linear
viscoelastic model is activated only during the approach period of collision, the damping
ratio,

, will take a higher value for the same coecient of restitution, e, than the damping
ratio, , in the linear viscoelastic model. After assessing the value of

, the impact stiness
parameter,

, can be determined numerically through iterative simulations which tend to t
the experimentally obtained pounding force time histories.
EXPERIMENTAL VERIFICATION OF POUNDING FORCE MODELS
In order to verify the accuracy of the linear viscoelastic, the non-linear elastic and the proposed
non-linear viscoelastic models of structural pounding, the results of the numerical analysis
have been compared with the results of the experiments conducted for dierent types of
structures with various material and contact surface geometry properties. The values of the
impact stiness parameters dening the models used in the numerical analysis have been
determined through iterative procedure so as to equalize the maximum pounding force value
obtained from the simulations with the maximum pounding force value measured during
the experiment. The dierence between the results of the experiment and the results of the
numerical analysis has been assessed by calculating the normalized error
E =
||F

F||
||F||
100% (10)
where F is a response time history obtained from the experiment,

F is a response time history
obtained from the numerical analysis, ||F

F|| is a Euclidean norm of F

F, which in the
case of the time history given in a discrete form can be calculated from the formula [33]:
||F

F|| =

i=1
(F
i


F
i
)
2
(11)
where n is a number of values in the time history record. In the rst two examples presented
below, the normalized errors have been calculated for the pounding force time histories. In the
case of the third and the fourth examples, the errors have been assessed for the displacement
and velocity time histories of structural response, respectively.
Steel-to-steel impact
In the rst example, the results of the numerical analysis, applying three dierent pounding
force models, are compared with the results of the experiment conducted by Goland et al.
[27]. In the experiment, steel balls of dierent diameters were dropped from various height
levels on the top of a steel hemisphere mounted on the upper surface of a rigid beam with
a leaf spring. A barium titanate wafer inserted between the hemisphere and the beam was
used to measure the force histories of collisions. In the numerical analysis, the model of
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
602 R. JANKOWSKI
m
h
y(t )
Figure 2. Model of a ball falling on a stationary rigid surface.
pounding between the falling ball and a stationary rigid surface, shown in Figure 2, is used.
The dynamic equation of motion for such a model can be written in the form
m y(t) + F(t) =mg (12)
where m is the mass of a ball, y(t) its vertical acceleration, g stands for the acceleration of
gravity and F(t) is the pounding force which is equal to zero when y(t)6h (h is a drop
height) and is dened by Equations (4), (7) or (8) when y(t)h, where deformation (t) is
expressed as
(t) =y(t) h (13)
The numerical analysis has been conducted for impacts of steel balls with dierent diameters
dropped from various height levels applying three models of pounding force. In this paper,
however, the results for a 5=32inch diameter ball and 2inch drop height are presented. In the
analysis, the following values of parameters dening the dierent pounding force models have
been used: k = 2:0710
7
N=m, =0:14 (e =0:65) for the linear viscoelastic model, =4:66
10
9
N=m
3=2
for the non-linear elastic model and

=1:03 10
10
N=m
3=2
,

= 0:35 (e =0:65)
for the non-linear viscoelastic model. The above values of the impact stiness parameters: k,
and

have been determined through an iterative procedure in order to attain the maximum
pounding force of F
max
= 80:7 N since such a value was measured as the maximum one
during the experiment [27]. A time-stepping integration procedure with constant time step
t =1 10
7
s has been applied to solve the equation of motion (12) numerically. The
pounding force time history measured during the experiment and the histories received from
the numerical analysis for the rst impact are presented in Figure 3. Using Equation (10),
the simulation errors for pounding force histories have been calculated as equal to: 15.8%
for the linear viscoelastic model, 78.0% for the non-linear elastic model and 21.7% for the
non-linear viscoelastic model.
Concrete-to-concrete impact
In this example, the results of the numerical analysis are compared with the results of the
experiment conducted by van Mier et al. [26] on impacts between a pendulum concrete striker
and a xed prestressed concrete pile for dierent contact surface geometries, striker mass and
impact velocity values. Neglecting the inuence of the xed pile, the numerical analysis can
be conducted using the model of one-sided pounding of a pendulum striker with a stationary
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
MODELLING OF EARTHQUAKE-INDUCED STRUCTURAL POUNDING 603
-20
0
20
40
60
80
100
0 5 10 15 20
experiment [27]
linear viscoelastic model
non-linear elastic model
non-linear viscoelastic model
P
o
u
n
d
i
n
g

f
o
r
c
e

(
N
)
Time (s)
Figure 3. Pounding force time histories during impact between falling steel ball
and steel hemisphere mounted on a beam.
m
x(0)
l
Figure 4. Model of a pendulum which strikes a stationary rigid barrier.
rigid barrier, as shown in Figure 4. The dynamic equation of motion for such a model can
be expressed as
m x(t) +
mg
l
x(t) + F(t) =0 (14)
where x(t), x(t), m, and l are the horizontal displacement, acceleration, mass and length of the
pendulum striker, respectively, F(t) is the pounding force and g stands for the acceleration
of gravity. In order to ensure a specied velocity at the rst impact, the pendulum striker has
to be initially displaced by an appropriate distance x(0) =x
0
, (x
0
0). When contact between
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
604 R. JANKOWSKI
-2 10
4
0
2 10
4
4 10
4
6 10
4
8 10
4
1 10
5
1.2 10
5
0 2 4 6 8 10 12
experiment, test 1 [26]
experiment, test 2 [26]
linear viscoelastic model
non-linear elastic model
non-linear viscoelastic model
P
o
u
n
d
i
n
g

f
o
r
c
e

(
N
)
Time (ms)
Figure 5. Pounding force time histories during impact between concrete
pendulum striker and reinforced concrete pile.
the pendulum and a rigid barrier is detected, i.e. when x(t)0, the pounding force, F(t), in
Equation (14) is dened using Equations (4), (7) or (8), in which the deformation (t) is
expressed as
(t) =x(t) (15)
On the other hand, when x(t)60 (no contact), then the value of the pounding force is equal
to zero.
The numerical simulations have been conducted for various values of striker mass and
impact velocity applying three dierent models of pounding force. In order to solve the
equation of motion (14) numerically, a time-stepping integration procedure with constant
time step t =0:0001 s has been used. An example of the results from the numerical anal-
ysis and the experiment [26] for impact of a spherical concrete pendulum striker of mass
570 kg (concrete strength 38:2 N=mm
2
) with impact velocity 0:5 m=s is shown in Figure 5. In
the numerical analysis, the following values of parameters dening dierent pounding force
models have been used: k =9:35 10
7
N=m, = 0:14 (e =0:65) for the linear viscoelas-
tic model, =1:13 10
9
N=m
3=2
for the non-linear elastic model and

=2:75 10
9
N=m
3=2
,

=0:35 (e =0:65) for the non-linear viscoelastic model. The above values of the impact
stiness parameters, k, and

, have been determined through an iterative procedure in
order to attain the maximum pounding force of F
max
=102:5 kN since such a value was mea-
sured as the maximum one during the experiment [26]. The simulation errors for pounding
force histories presented in Figure 5 have been calculated as equal to 23.8% for the linear
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
MODELLING OF EARTHQUAKE-INDUCED STRUCTURAL POUNDING 605
d
C
K
m
x(t )
Figure 6. Model of a bridge girder which interacts with an abutment.
viscoelastic model, 97.6% for the non-linear elastic model and 22.9% for the non-linear vis-
coelastic model.
Pounding between a bridge girder and an abutment
In the third example, the results of the numerical analysis are compared with the results of
the experiment conducted by Zhu et al. [25] on pounding between a bridge girder and an
abutment. In the experiment, the model of a bridge girder with mass of 2 kg placed with a
0:5 cm gap close to the xed rigid barrier has been tested on the shaking table under a sine
wave. A video camera has been employed to monitor the longitudinal displacement of the
model girder using image-processing techniques. In the numerical analysis, the bridge girder is
modelled as a single-degree-of-freedom system as shown in Figure 6. The dynamic equation
of motion which simulates the pounding-involved response for such a model can be written
in the form
m x(t) + C x(t) + Kx(t) + F(t) = p(t) (16)
where x(t), x(t), and x(t) are the horizontal displacement, velocity and acceleration of a
bridge girder, respectively, m stands for the girders mass, C and K are damping and stiness
coecients and p(t) is an external excitation (sine wave). Moreover, in Equation (16), F(t)
denotes the pounding force which is equal to zero when x(t)6d (d is an initial separation
gap) and is dened by Equations (4), (7) or (8) when x(t)d, where deformation (t) is
expressed as
(t) =x(t) d (17)
The numerical analysis has been conducted for the structural model dened by Equa-
tion (16) for: m=2 kg, C =4:1 kg=s, K =210:125 N=m (see Reference [25]), applying three
dierent models of pounding force. In the analysis, the following values of models parameters
have been used: k =1:5474 10
5
N=m, =0:28 (e =0:40) for the linear viscoelastic model
(values determined by Zhu et al. [25] based on experiments), =4:15 10
6
N=m
3=2
for the
non-linear elastic model and

=2:47 10
6
N=m
3=2
,

=0:99 (e =0:40) for the non-linear
viscoelastic model. The equation of motion (16) has been solved numerically using a time-
stepping integration procedure with constant time step t =0:001 s. The displacement time
history of the bridge girder obtained from the experiment and the histories received from the
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
606 R. JANKOWSKI
-0.025
-0.02
-0.015
-0.01
-0.005
0
0.005
0.01
0 1 2 3 4 5
experiment [25]
linear viscoelastic model
non-linear elastic model
non-linear viscoelastic model
D
i
s
p
l
a
c
e
m
e
n
t

(
m
)
Time (s)
Figure 7. Displacement time histories of a bridge girder for pounding of a
girder and an abutment induced by sine wave.
numerical analysis are shown in Figure 7. Using Equation (10), the simulation errors for the
displacement time histories have been calculated as equal to 24.1% for the linear viscoelas-
tic model, 61.8% for the non-linear elastic model and 17.6% for the non-linear viscoelastic
model.
Pounding of steel towers
The fourth example deals with the comparison between the results of the numerical simulations
applying three dierent pounding force models and the results of the experiment conducted by
Chau et al. [31]. The experiment was focused on pounding of two steel towers of equal height
of 2m and dierent masses excited horizontally on a shaking table under the NS component
of the El Centro earthquake (18 May 1940). Each tower was constructed from four columns of
rectangular hollow section with a massive plate placed on the top. To ensure that all columns
deect simultaneously and consistently to the horizontal shaking, four horizontal tie bars were
hinged at about the mid-height. The mass of the top plate, m
i
, the natural frequency, f
i
, and
the damping ratio,
i
, for Tower 1 and Tower 2 were: m
1
=204:0kg, f
1
=2:31Hz,
1
=0:014,
m
2
=146:5 kg, f
2
=2:90 Hz, and
2
=0:016, respectively [31]. The separation gap between
the towers, d, was set to be equal to 11 mm.
In the numerical analysis, the model in which both towers are simulated as single-degree-
of-freedom systems as shown in Figure 8, has been used. The dynamic equation of motion
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
MODELLING OF EARTHQUAKE-INDUCED STRUCTURAL POUNDING 607
C
2
K
2
m
2
d
x
2
(t )
K
1
m
1
x
1
(t )
C
1
Figure 8. Model of interacting steel towers.
for such a model can be expressed as

m
1
0
0 m
2

x
1
(t)
x
2
(t)

C
1
0
0 C
2

x
1
(t)
x
2
(t)

K
1
0
0 K
2

x
1
(t)
x
2
(t)

F(t)
F(t)

m
1
0
0 m
2

x
g
(t)
x
g
(t)

(18)
where x
i
(t), x
i
(t), x
i
(t), C
i
and K
i
are the horizontal displacement, velocity, acceleration,
damping coecient and stiness coecient for Tower i (i =1; 2), respectively, x
g
(t) stands
for the acceleration input ground motion and F(t) is the pounding force which is equal to
zero when (t)60 and is dened by Equations (4), (7) or (8) when (t)0, where (t) is
dened as
(t) =x
1
(t) x
2
(t) d (19)
The numerical analysis has been conducted by applying three dierent models of pounding
force. In the analysis, the following values of model parameters have been used: k =1:40
10
9
N=m, =0:14 (e =0:65) for the linear viscoelastic model, =2:36 10
10
N=m
3=2
for the
non-linear elastic model (value determined by Chau et al. [31]) and

=9:90 10
10
N=m
3=2
,

=0:35 (e =0:65) for the non-linear viscoelastic model. A time-stepping integration procedure
with constant time step t =0:0001 s has been applied to solve the equation of motion (18)
numerically. The velocity time history of Tower 1 obtained from the experiment and the
histories received from the numerical analysis are shown in Figure 9. The simulation errors
for these histories have been calculated as equal to 39.0% for the linear viscoelastic model,
30.4% for the non-linear elastic model and 28.3% for the non-linear viscoelastic model.
Results of experimental verication of pounding force models
It can be seen from Figures 3 and 5 that the linear viscoelastic and the non-linear viscoelastic
models are the most precise ones in simulating the pounding force time histories during
impact. Both models allow simulation of the relatively rapid increase in the pounding force
during the approach period and the decrease in the force with lower unloading rate during the
restitution period. However, it is only the non-linear viscoelastic model which allows the force
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
608 R. JANKOWSKI
(d) ) s ( e m Ti
s
)
/
m

(
t
y
l
o
c
i
V
e
0 0 5 4 0 0 3 2 0 1 0
1
5 0.
0
-0.5
-1
(c) ) s ( e m Ti
s
)
/
m

(
t
y
l
o
c
i
V
e
0 0 5 4 0 0 3 2 0 1 0
1
5 0.
0
-0.5
-1
(b) ) s ( e m Ti
s
)
/

(
m
y
t
i
c
o
l
V
e
0 0 5 4 0 0 3 2 0 1 0
1
5 0.
0
-0.5
-1
(a) ) s ( e m Ti
)
/
s
m
c
i
t
y

(
o
l
e
V
0 0 5 0 4 0 3 2 0 1 0
1
5 0.
0
5 -0.
-1
Figure 9. Velocity time histories of Tower 1 for pounding of steel towers under the El
Centro earthquake: (a) experiment [31]; (b) linear viscoelastic model; (c) non-linear
elastic model; and (d) non-linear viscoelastic model.
to be even further reduced just before separation. On the other hand, the activated dashpot
during the restitution period of collision in the case of the linear viscoelastic model results in
the negative impact force just before the end of the contact which does not have a physical
explanation. Figures 3 and 5 conrm also the disadvantage of the non-linear elastic model
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
MODELLING OF EARTHQUAKE-INDUCED STRUCTURAL POUNDING 609
for which the simulation errors are the largest. In this case, owing to elastic behaviour, the
pounding force history during the approach and the restitution periods is symmetric resulting
in the longer time of contact.
The pounding force time histories for the non-linear viscoelastic model presented in Fig-
ures 3 and 5 show a change in curvature after passing the peak force value. This eect is
related to the disengagement of the viscous damper in the restitution period of impact. Smooth-
ing of the impact force diagram at this point is possible but it would require redening the
formula (8) for the transition zone between the approach and restitution period. That would
make the model more complicated and therefore less practical for numerical simulations.
The results of the study presented in Figures 7 and 9 show that the application of the
proposed non-linear viscoelastic model gives the smallest simulation errors in the response
(displacement and velocity) time histories of the analysed examples of structural pounding. It
is worth noting that in the case of the analysis of pounding between a bridge girder and an
abutment (see Figure 7), the use of the non-linear elastic model gives very poor results. This
is due to the fact that the model does not account for the dissipation of energy during contact
which is of great importance in this case since the relatively low value of the coecient of
restitution (e =0:4) has been determined based on experiments [25]. There is no doubt that
in such cases only the models that take into consideration the dissipation of energy should be
used for the simulation purposes. On the other hand, the example of pounding of steel towers
(see Figure 9) is quite a dierent case. Now, the impacts are much more elastic and the
application of the non-linear elastic model (used also by Chau et al. [31]) gives very similar
simulation errors as in the case of the non-linear viscoelastic model. It should be stressed also
that the relatively high values of simulation errors obtained for all models in this example
result mainly from the fact that a less accurate single-degree-of-freedom structural model has
been used in the numerical analysis (used also by Chau et al. [31]). This fact, however, does
not prevent us from drawing the comparative conclusions on the eectiveness of dierent
pounding force models.
CONCLUSIONS
In this paper, a non-linear viscoelastic model of pounding force, which is intended to en-
hance the accuracy of the modelling of structural pounding during earthquakes, has been
analysed. The comparison between the model proposed and two other commonly used pound-
ing force models, i.e. the linear viscoelastic model and the non-linear elastic model following
the Hertz contact law, has been conducted. In order to verify the accuracy of the mod-
els, the results of the numerical analysis have been compared with the results of impact
experiments.
The results of the study show that the linear viscoelastic and the non-linear viscoelastic
models give the smallest simulation errors in the pounding force time histories during impact.
In the case of the linear viscoelastic model, however, the negative impact force just before
separation, which does not have a physical explanation, has been observed. Further analysis
has shown that the application of the proposed non-linear viscoelastic model results in the
smallest simulation errors in the response time histories of the analysed examples of structural
pounding. The results have also indicated that the non-linear elastic model should not be used
for modelling of impacts with low values of coecient of restitution. However, it can be
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
610 R. JANKOWSKI
more accurate than the linear viscoelastic model when the loss of energy during collision is
not so signicant.
Among the considered pounding force models, the linear viscoelastic model is the simplest
one, it can be easily applied in computer programs and it does not require numerical iterations
to obtain an appropriate value of the impact damping ratio as in the case of the non-linear
viscoelastic model. It seems that the major shortcoming of the linear viscoelastic model, i.e.
the negative impact force just before separation, can be eliminated by neglecting the negative
force value when it occurs. This would require, however, the re-assessment of the relation
between the impact damping ratio and the coecient of restitution (see Equation (6)) in order
to achieve the appropriate post-impact velocities during numerical simulations. Moreover, such
an improvement may be dicult to implement in the general-purpose computer programs.
The practical application of the proposed non-linear viscoelastic model for simulation pur-
poses requires the knowledge of the models parameters. In this paper, the values of these
parameters have been determined based on examples of results of experiments conducted on
structural pounding. It seems, however, that more extensive experimental studies are required
in order to assess the range of the models parameters more precisely for dierent types of
structures with various material and contact surface geometry properties.
ACKNOWLEDGEMENTS
This study was supported by the European Community under the FP5 Programme, key-action City
of Tomorrow and Cultural Heritage (Contract No. EVK4-CT-2002-80005). This support is greatly
acknowledged.
REFERENCES
1. Rosenblueth E, Meli R. The 1985 earthquake: causes and eects in Mexico City. Concrete International 1986;
8:2334.
2. Anagnostopoulos SA. Building pounding re-examined: how serious a problem is it? Eleventh World Conference
on Earthquake Engineering, Acapulco, Mexico, 2328 June 1996, paper no. 2108.
3. Kasai K, Maison BF. Building pounding damage during the 1989 Loma Prieta earthquake. Engineering
Structures 1997; 19:195207.
4. Earthquake Engineering Research Institute, Northridge Earthquake of 17 January 1994Reconnaissance Report.
EERI Report 95-03, Vol. 1, Oakland, USA, 1995.
5. Otsuka H, Unjoh S, Terayama T, Hoshikuma J, Kosa K. Damage to highway bridges by the 1995 Hyogoken
Nanbu earthquake and the retrot of highway bridges in Japan. Third U.S.Japan Workshop on Seismic Retrot
of Bridges, Osaka, Japan, 1011 December 1996.
6. Anagnostopoulos SA. Earthquake induced pounding: State of the art. Proceedings of 10th European Conference
on Earthquake Engineering, Vienna, Austria, September 1994, 897905.
7. Anagnostopoulos SA, Spiliopoulos KV. An investigation of earthquake induced pounding between adjacent
buildings. Earthquake Engineering and Structural Dynamics 1992; 21:289302.
8. Maison BF, Kasai K. Dynamics of pounding when two buildings collide. Earthquake Engineering and Structural
Dynamics 1992; 21:771786.
9. Papadrakakis M, Mouzakis H, Plevris N, Bitzarakis S. A Lagrange multiplier solution method for pounding of
buildings during earthquakes. Earthquake Engineering and Structural Dynamics 1991; 20:981998.
10. Jankowski R, Wilde K, Fujino Y. Pounding of superstructure segments in isolated elevated bridge during
earthquakes. Earthquake Engineering and Structural Dynamics 1998; 27:487502.
11. Zanardo G, Hao H, Modena C. Seismic response of multi-span simply supported bridges to a spatially varying
earthquake ground motion. Earthquake Engineering and Structural Dynamics 2002; 31:13251345.
12. Jeng V, Kasai K. Spectral relative motion of two structures due to seismic travel waves. Journal of Structural
Engineering (ASCE) 1996; 122:11281135.
13. Jankowski R, Wilde K. A simple method of conditional random eld simulation of ground motions for long
structures. Engineering Structures 2000; 22:552561.
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611
MODELLING OF EARTHQUAKE-INDUCED STRUCTURAL POUNDING 611
14. Anagnostopoulos SA. Pounding of buildings in series during earthquakes. Earthquake Engineering and
Structural Dynamics 1988;16:443456.
15. Jankowski R, Wilde K, Fujino Y. Reduction of pounding eects in elevated bridges during earthquakes.
Earthquake Engineering and Structural Dynamics 2000; 29:195212.
16. Jing H-S, Young M. Impact interactions between two vibration systems under random excitation. Earthquake
Engineering and Structural Dynamics 1991; 20:667681.
17. Pantelides CP, Ma X. Linear and nonlinear pounding of structural systems. Computers and Structures 1998;
66:7992.
18. Chau KT, Wei XX. Pounding of structures modelled as non-linear impacts of two oscillators. Earthquake
Engineering and Structural Dynamics 2001; 30:633651.
19. Penzien J. Evaluation of building separation distance required to prevent pounding during strong earthquakes.
Earthquake Engineering and Structural Dynamics 1997; 26:849858.
20. Westermo BD. The dynamics of interstructural connection to prevent pounding. Earthquake Engineering and
Structural Dynamics 1989; 18:687699.
21. Goldsmith W. Impact: The Theory and Physical Behaviour of Colliding Solids, 1st edn, Edward Arnold:
London, U.K., 1960.
22. DesRoches R, Muthukumar S. Eect of pounding and restrainers on seismic response of multiple-frame bridges.
Journal of Structural Engineering (ASCE) 2002; 128:860869.
23. Ruangrassamee A, Kawashima K. Relative displacement response spectra with pounding eect. Earthquake
Engineering and Structural Dynamics 2001; 30:15111538.
24. Azevedo J, Bento R. Design criteria for buildings subjected to pounding. Eleventh World Conference on
Earthquake Engineering, Acapulco, Mexico, 2328 June 1996, paper no. 1063.
25. Zhu P, Abe M, Fujino Y. Modelling three-dimensional non-linear seismic performance of elevated bridges with
emphasis on pounding of girders. Earthquake Engineering and Structural Dynamics 2002; 31:18911913.
26. van Mier JGM, Pruijssers AF, Reinhardt HW, Monnier T. Load-time response of colliding concrete bodies.
Journal of Structural Engineering (ASCE) 1991; 117:354374.
27. Goland M, Wickersham PD, Dengler MA. Propagation of elastic impact in beams in bending. Journal of Applied
Mechanics 1955; 22:17.
28. Filiatrault A, Wagner P, Cherry S. Analytical prediction of experimental building pounding. Earthquake
Engineering and Structural Dynamics 1995; 24:11311154.
29. Kim S-H, Shinozuka M. Eects of seismically induced pounding at expansion joints of concrete bridges. Journal
of Engineering Mechanics (ASCE) 2003; 129:12251234.
30. Valles RE, Reinhorn AM. Evaluation, prevention and mitigation of pounding eects in building structures.
Technical Report NCEER-97-0001, National Center for Earthquake Engineering Research, State University of
New York at Bualo, U.S.A., 1997.
31. Chau KT, Wei XX, Guo X, Shen CY. Experimental and theoretical simulations of seismic poundings
between two adjacent structures. Earthquake Engineering and Structural Dynamics 2003; 32:537554. DOI:
10.1002=eqe.231.
32. Anagnostopoulos SA. Equivalent viscous damping for modeling inelastic impacts in earthquake pounding
problems. Earthquake Engineering and Structural Dynamics 2004; 33:897902. DOI: 10.1002=eqe.377.
33. Bendat JS, Piersol AG. Random Data: Analysis and Measurement Procedures; Wiley-Interscience: New York,
U.S.A., 1971.
Copyright ? 2005 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:595611

You might also like