Earthquake Engineering & Structural Dynamics

EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS
Earthquake Engng Struct. Dyn. 2005; 34:120

Published online 25 October 2004 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/eqe.398
Earthquake-induced interaction between adjacent reinforced
concrete structures with non-equal heights
Chris G. Karayannis
,
and Maria J. Favvata
Reinforced Concrete Lab., Department of Civil Engineering, Democritus University of Thrace,
Xanthi, 67100 Greece
SUMMARY
The inuence of the structural pounding on the ductility requirements and the seismic behaviour of
reinforced concrete structures designed to EC2 and EC8 with non-equal heights is investigated. Special
purpose elements of distributed plasticity are employed for the study of the columns. Two distinct types
of the problem are identied: Type A, where collisions may occur only between storey masses; and
Type B, where the slabs of the rst structure hit the columns of the other (72 Type A and 36 Type B
pounding cases are examined). Type A cases yielded critical ductility requirements for the columns in
the pounding area mainly for the cases where the structures were in contact from the beginning of the
excitation. In both pounding types the ductility requirements of the columns of the taller building are
substantially increased for the oors above the highest contact storey level probably due to a whiplash
behaviour. The most important issue in the pounding type B is the local response of the column of
the tall structure that suers the hit of the upper oor slab of the adjacent shorter structure. In all the
examined cases this column was in a critical condition due to shear action and in the cases where the
structures were in contact from the beginning of the excitation, this column was also critical due to high
ductility demands. It can be summarized that in situations of potential pounding, neglecting its possible
eects leads to non-conservative building design or evaluation that may become critical in some cases.
Copyright ? 2004 John Wiley & Sons, Ltd.
KEY WORDS: structural pounding; reinforced concrete structures; ductility requirements; non-linear
seismic analysis
1. INTRODUCTION
The interaction of adjacent buildings and the collisions that occur between them during an
earthquake have been repeatedly identied as a usual cause of damage. This problem between
structural systems that are in contact or in close proximity to each other is commonly referred
to as structural pounding.
The literature provides many examples about the seismic hazard that the structural pound-
ing poses [13]. Based on the substantial knowledge that has been acquired so far through
Correspondence to: Chris G. Karayannis, Reinforced Concrete Lab., Department of Civil Engineering, Democritus
University of Thrace, Xanthi, 67100 Greece.
E-mail: [email protected]
Received 11 June 2003
Revised 28 April 2004
Copyright
?
2004 John Wiley & Sons, Ltd. Accepted 29 April 2004
TLFeBOOK
2 C. G. KARAYANNIS AND M. J. FAVVATA
numerous eld observations after damaging earthquakes it can be concluded that pounding
is frequently observed when strong earthquakes strike big cities and densely populated urban
areas. In these events the interaction between adjacent buildings is a usual cause of damage
and there are cases reported in the literature where pounding has been identied as a primary
cause for the initiation of collapse. In the earthquake that struck Mexico City in 1985 the
rst assessment, which subsequently was revised, had attributed to pounding a big part of the
observed damage and identied many cases where pounding led to collapse [2, 4]. Although
in this respect the earthquake of Mexico City is unique in terms of damage and collapse cases
attributed to pounding and the phenomenon has been overstated and exaggerated concerning
the damage, it is a fact that in all major earthquakes of the last decades structural pounding
was always present [1, 3, 5].
The typical measure that modern codes specify against structural pounding is the provision
for sucient separation between adjacent buildings in order to preclude pounding [68]. There
are however some factors that make these code provisions not always eective or applicable.
First, common code interpretations and code applications lead sometimes to building separa-
tions that are inadequate and inconsistent with the philosophy of modern codes that imply
large deformations can occur during major earthquakes due to inelastic response. Further, the
high cost of land in densely populated metropolitan cities and the small lot sizes make the
seismic separation requirements not always easy to apply. In addition, there is the argument
that weak buildings in contact with stronger ones in city blocks may actually benet from
that contact, provided that the pounding will not cause any serious local damage from which
failure could be initiated [3, 9].
As alternatives to seismic separation a few measures against structural pounding have been
proposed. In order to reduce the eects of pounding in the case of existing small separation
distances Anagnostopoulos [10] proposed the lling of the gap between the adjacent build-
ings with shock absorbing material. Westermo [11] studied the use of permanent connections
between the adjacent buildings in order to eliminate pounding. The new Hellenic Code for
seismic design of structures and Eurocode 8 propose the use of strong stiening shear walls
in order to prevent pounding.
Many analytical works on structural pounding have been reported in the last two decades.
In the beginning these studies were based on the response of pairs or sets of colliding single
degree of freedom systems in earthquake excitations. Anagnostopoulos [10] examined the
case of pounding of several adjacent single degree of freedom systems in a row. In this
study elastic and inelastic systems were studied using ve real earthquake motions and many
problem parameters were examined. The results indicated that in the case of alike systems
exterior systems tend to suer more due to the pounding eect than do the interior ones,
the latter often experiencing reductions in their response. Further, Athanasiadou et al. [9],
working in the same direction, examined the inuence of a constant phase dierence in the
base motion of each system in an attempt to approximate the travelling wave eect.
Anagnostopoulos and Spiliopoulos [12] examined the pounding eect in multi-degree-of-
freedom systems. They idealized the buildings as lumped mass, shear beam type, multi-
degree-of-freedom systems with bilinear forcedeformation characteristics. They reported the
results of collisions on the response of a 5-storey building in congurations of 2, 3 and 4
buildings in contact. They also examined the pounding eect in several cases of two buildings
with dierent heights. In situations like these, according to the authors, pounding can be
catastrophic [12].
Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:120
TLFeBOOK
EARTHQUAKE-INDUCED INTERACTION BETWEEN ADJACENT STRUCTURES 3
Numerical formulations for the pounding of two structures focusing primarily on advanced
solution techniques have also been reported during the past decade [1315]. Maison and Kasai
[16, 17] proposed the formulation and the solution of the multiple degree of freedom equations
of motion for a type of structural pounding between two buildings and presented the pounding
between a tall 15-storey structure and a shorter 8-storey stier and more massive building.
The formulation and results are based on elastic dynamic analysis.
Karayannis and Fotopoulou [18] have examined various cases of structural pounding be-
tween multistorey reinforced concrete structures designed according to the Eurocodes 2 [7]
and 8 [8]. The work was based on non-linear dynamic step-by-step analysis and its purpose
was to present initial results for the inuence of some critical pounding parameters on the
ductility requirements of the columns and to examine the possibility of taking into account the
pounding eect during the design process according to EC2 to EC8. In the examined cases
the storey levels of the two colliding structures were always the same. The eect of soil
exibility on the inelastic seismic response of a particular case of adjacent 12- and 6-storey
reinforced concrete moment-resisting frames has been examined by Rahman et al. [19].
It is emphasized that all the previously mentioned papers examine pounding problems with
buildings that have storeys with equal inter-storey heights and consequently the pounding
takes place always between the oor masses of the colliding structures.
2. SCOPE OF THE STUDY
This study is based on the thoughts that: (a) pounding is an important cause of structural
damage that under certain conditions can lead to collapse initiation; (b) according to the new
design codes (Eurocodes 2 and 8) exible frame structures can be designed; (c) adequate
seismic separation is not always easy to apply; and (d) most of the existing analytical studies
have yielded conclusions not directly applicable to the design of multistorey buildings po-
tentially under pounding. In this paper an attempt to study and quantify the inuence of the
structural pounding on the ductility requirements and the overall seismic response of rein-
forced concrete frame structures with unequal heights designed according to the codes EC2
and EC8 is presented.
The study of the pounding of adjacent multistorey reinforced concrete buildings with un-
equal total heights and dierent storey heights is also a main objective of this work. This
is probably the most critical case of interaction between adjacent buildings and although it
is a common case in practice it has not been studied before in the literature as far as the
authors are aware. Furthermore, the inuence of the size of the gap distance between adjacent
structures on the eects of the pounding is also investigated. Non-linear dynamic step-by-step
analysis and special purpose elements are employed for the needs of this study.
3. KEY ASSUMPTIONS
3.1. Model idealization of structural pounding
In this work the interaction of two adjacent structures with dierent total heights is studied.
Each structure responds dynamically and vibrates independently. It is considered that one
TLFeBOOK
3
3

2
1
2

8

7

6

5

4

1

Figure 1. Actual condition and model idealization of the pounding problem. Structures with
unequal total heights and the heights of the storey levels of the two structures are not equal.
Pounding at the columns (inter-storey pounding).
exible multistorey building is in contact or in close proximity to one less-exible shorter
structure. If there is a gap distance between the structures collisions occur when the lateral
displacements of the structures exceed the pre-dened gap distance (d
g
). The inuence of the
gap size on the pounding eects is parametrically investigated. Two distinct types of structural
pounding are identied:
(a) Pounding case type A. The storey levels of the two structures have the same height so
that collisions may occur between the storey diaphragms and consequently between the
storey masses.
(b) Pounding case type B. The heights of the storey levels of the two structures are not equal
(Figure 1). In this very common case the slabs of the diaphragms of each structure hit the
columns of the other structure at a point within the deformable height. This phenomenon
is especially intense at the contact point of the upper storey level of the short stier
structure with the corresponding column of the tall building. The actual condition and
the model idealization of this pounding case are shown in Figure 1. Contact points are
taken into account at the levels of the oor slabs of the short structure. Nevertheless,
from the analyses of the examined pounding cases it has been found that the response
of the interacting structures is inuenced only by the position and the characteristics of
the contact point at the shorter structures top oor. The inuence of the other contact
points on the results proved to be negligible in the examined cases. The same conclusion
also holds, more or less, for the examined cases with zero distance gap. This is mainly
attributed to the signicant height dierence of the interacting structures in the studied
cases. Thus, in the following analyses and results, only the inuence of the pounding
eect through the top oor contact point on the whole behaviour of the structures and
on the response and ductility requirements of the columns is examined. Analyses have
been performed using time steps of the order of 1}5000 to 1}10000 in order to achieve
TLFeBOOK
numerical stability and to adequately reproduce higher mode response excited by the
short-duration impacts.
3.2. Contact element
Collisions are simulated using special purpose contact elements that become active when
the corresponding nodes come into contact. This idealization is consistent with the building
model used and appears adequate for studying the eects of pounding on the overall structural
response for the pounding cases under examination. Local eects such as inelastic exural
deformations, yield of the exural reinforcement and ductility requirements of the columns in
the pounding area are taken into account through the special purpose elements employed for
the modelling of the columns.
The response of the contact elements has three parts. First, the negative direction of the
X-axis that represents the condition that the buildings move away from each other. In the
positive direction of the X-axis there are two parts in order to simulate the actual behaviour of
the structures in case there is a small gap distance (d
g
) between them. It is possible that the
structures move one towards the other but the displacements are small and the existing gap is
not covered. In this case the contact element remains non-active and the buildings continue
to vibrate independently. In the case where the structures move one towards the other and
the displacements bridge the existing gap or the structures are in contact from the beginning
then the contact element responds as a spring with almost innite stiness.
The damage at the contact area is expected to be concentrated from the beginning at the
column (pounding case B) that suers the impact. Thus, considering that the damage of the
building materials and the damage of the slabs of the shorter structure are not signicant,
an elastic contact element has been used. Moreover, analyses of interaction cases (pounding
cases B) using contact elements that can account for damping have been performed as well.
From comparisons between the results of these analyses with the results of the analyses using
elastic contact elements, it can be obtained that the observed dierences are negligible.
3.3. Beamcolumn elements
The frame structural systems consist of beams and columns whereas the dual (frame-wall)
systems have in addition two reinforced concrete walls. Each structure is modelled as a 2D
assemblage of non-linear elements connected at nodes. The mass is lumped at the nodes and
each node has three degrees of freedom.
Each structure responds dynamically and vibrates independently. Collision occurs when the
lateral displacements of the structures at the oor levels exceed the pre-dened gap distance
(d
g
) between the two structures. The actual conditions and the model idealization of the
pounding problem examined in this study are shown in Figure 1.
The computer program used in this work is the program package DRAIN-2DX [20]. The
nite element mesh used here for the modelling of each structure uses a one-dimensional
element for each structural member. Two types of one-dimensional beamcolumn elements
were used: (a) one special purpose element that is employed for the modelling of the columns
of the 8-storey structure; and (b) an element for the modelling of the beams of this structure
and for all members of the second shorter structure. The latter element is a common lumped
plasticity beamcolumn model that considers the inelastic behaviour concentrated in zero-
length plastic hinges at the elements ends.
TLFeBOOK
Figure 2. Analysis model for the pounding area (type B).
The special purpose element employed for the columns is one of distributed plasticity type
accounting for the spread of inelastic behaviour both over the cross-sections and along the
deformable region of the member length. This element performs numerical integration of the
virtual work along the length of the member using data deduced from cross-section analysis
at pre-selected locations. Thus, the deformable part of the element is divided into a number
of segments (Figure 2) and the behaviour of each segment is monitored at its centre cross-
section (control section). The cross-section analysis that is performed at the control sections
is based on the bre model. This bre model accounts rationally for axialmoment (PM)
interaction.
In order to accurately model the actual behaviour of the columns in the area that pounding
takes place the deformable height of each column is divided into four segments (see Figure 2).
The lengths of the segments have been determined as follows; the length of the inner two equal
segments have been taken as equal to 0.47h whereas the length of the segments at the ends
TLFeBOOK
of the element are equal to 0.03h, where h is the deformable height of the column (Figure 2).
This partition of the columns deformable height can reasonably take into account without
excessive increase of the computational eort the following important structural parameters for
the behaviour of a reinforced concrete column. (a) The actual distribution of the quantity of
the longitudinal reinforcement along the column length. (b) The variation of the connement
degree of concrete over the cross-sections and along the length of the column since a higher
degree of connement is usually applied near the elements ends. (c) This partition of the
columns length also allows for the setting of control sections near the elements ends very
close to the face of the joints. These parts are considered to be critical zones because they
are areas of potential formation of plastic hinges.
The connement degree in the middle part of the internal columns is rather low and the
connement coecient ranged from K =1.023 to 1.041, while in the end parts of the same
columns it ranged from K =1.213 to 1.305. However, according to the Eurocodes 2 and 8
the connement rules for the critical regions of columns are applied for the entire length of
the external columns. Thus, for the external columns of the examined cases, the degree of
connement was the same for the entire length and the connement coecient ranged from
K =1.213 to 1.309.
Furthermore, in this work special attention has been given to the study of the local response
of the column that suers the direct hit of the upper slab of the shorter and stier structure
in the case of pounding type B. In this direction, two special purpose elements of distributed
plasticity type are employed for this column. Each element is divided into four unequal
segments in the way that is shown in Figure 2. Thus, there are eight control cross-sections
along the height of the critical column. This partition of the columns deformable height
can reasonably take into account the actual distribution of reinforcement and the connement
degree of concrete and further it allows for the setting of the control cross-sections near the
elements critical points.
4. DESIGN OF STRUCTURES
4.1. Eight-storey structures
Two 8-storey frame structures were designed according to Eurocodes 2 and 8 [7, 8], the
rst one meeting the Ductility Capacity Medium (DCM) criteria and the latter one meeting
the Ductility Capacity High (DCH) criteria of the codes. Behaviour factors for DCM and
DCH frames were q =3.75 and 5.00, respectively. The mass of the structures is taken as
M =(G + 0.3Q)}q (where G=gravity loads and Q=live loads [7]) and the design base
shear force was taken as J =(0.3q}q)M (where q = the behaviour factor of the structure
[8]). Reduced values of member moments of inertia (I
ef
) were considered in the design to
account for the cracking; for beams I
ef
=0.5I
g
(where I
g
=the moment of inertia of the gross
section) and for the columns I
ef
=0.9I
g
. The code provision [7, 8], in most cases, proved
to be critical for the dimensioning of the columns, regarding the axial load ratio limitation
v
d
60.65 and v
d
60.55 for DCM and DCH frames, respectively, and in a few cases (columns
of the upper storeys) the code minimum dimensions requirements. The structure geometry
and reinforcement of the columns of the 8-storey frames are shown in Figure 3.
TLFeBOOK
Figure 3. Structural system and column reinforcement of the 8-storey frames designed to EC2 and EC8.
4.2. Four-storey and two-storey structures
Two 4-storey and two 2-storey structures were designed according to the codes EC2 and
EC8, meeting the DCM design criteria [7, 8]. The rst one of each pair of structures is a
frame structure, whereas the second one is a dual (framewall) structural system. Behaviour
factors used in the design of the frame structures and the framewall structures were q =3.75
and 3.00, respectively. The mass is taken as M =(G + 0.3Q)}q and the design base shear
force was taken as J =(0.3q}q)M. Reduced values of member moments of inertia (I
ef
) were
considered in the design; I
ef
=0.5I
g
, I
ef
= 0.9I
g
and I
ef
= 2I
g
}3 for beams, columns and walls,
respectively. In most of the cases the code minimum requirements proved to be critical for
the dimensioning of the columns.
TLFeBOOK
4.3. Three-storey structure
One 3-storey structure is also designed according to the codes EC2 and EC8, meeting the
DCM design criteria. The structure is a dual (framewall) structural system. The behaviour
factor is taken as equal to 3.0. The mass and the design characteristics were the same as
in the cases of the 4-storey framewall structure. The code minimum dimensions proved to
be critical for the dimensioning of the columns in most of the cases. The height of the rst
storey is 4.80 m and the height of the other stories is 3.20 m.
5. POUNDING TYPE A. POUNDING AT THE FLOOR LEVELS
5.1. Examined cases
Seventy-two pounding cases between structures with equal inter-storey heights designed
according to EC2 and EC8 are examined and discussed. These are the interaction cases be-
tween the 8-storey frames (DCM and DCH frames) and the shorter and stier structures (two
4-storey and two 2-storey structures) mentioned in the previous section. The pounding cases
of the 8-storey structures to stationary barriers (very sti structures) are also included. Each
pair of structures is examined for six dierent gap distances between the two structures.
Each pounding case is subjected to two dierent natural seismic excitations; the El Cen-
tro 1940 earthquake (duration 15 secs and :
max
= 0.318 g) and the Korinth (Alkyonides)
Greece 1981 earthquake (duration 12 secs and :
max
=0.306 g). The pounding cases were ex-
amined and the ratios of the elastic structural stiness and period (p
k
and p
T
, respectively)
of the 8-storey frame to the ones of the adjacent less exible structure of each case are pre-
sented in Table I. The elastic stiness is evaluated by means of a push-over analysis of the
structures.
5.2. Results
Time history comparative results of the examined pounding cases between the DCM 8-storey
frame and the 4-storey structures (Korinth 1981 excitation) are presented in Figure 4. In
Figure 4(a) the displacement time histories of the 4th level (pounding level) of the 8-storey
frame for the cases of pounding with the 4-storey frame structure and the 4-storey stationary
barrier (very sti structure) are presented and compared with the response of the 8-storey
frame vibrating without pounding. In Figure 4(b) the displacement time histories of the 8th
level of the same structure for the same pounding cases are presented.
In Figures 4(a) and (b) it can be seen that the amplitude of the response displacements
of the 8-storey frame in the cases of pounding with the 4-storey structures is signicant. It
is noted that the peak negative displacement at the pounding level is larger in the case of
pounding with the 4-storey frame (d= 0.105m) than the one in the case where the 8-storey
frame vibrates without pounding problems (d = 0.080 m). The same type of behaviour has
been observed in the case of pounding between the 8-storey frame with the 2-storey frame-
wall structure. The responses of the 4-storey and the 2-storey structures were greatly aected
by the pounding that generally decreased the response amplitude of these structures.
In this study the inuence of the pounding on the curvature ductility requirements of
the columns of the 8-storey frame in the area of the pounding is primarily examined. The
TLFeBOOK
Table I. Pounding type A. The examined cases, the rations p
k
(and p
T
) of the pounding structures and
coecients
j
, maximum values yielded from non-linear dynamic analyses for the excitations Korinth,
Alkyonides, Greece 1981 and EI Centro 1940.
Examined casesRatios p
k
(and p
T
)
(each pair is examined for 6 dierent gap distances d
g
between the structures)
4-storey structures 2-storey structures
Frame Dual system Rigid barrier Frame Dual system Rigid barrier
8-storey p
k
=0.42 p
k
=0.21 p
k
0 p
k
=0.56 p
k
=0.13 p
k
0
DCM-frame (p
T
=1.36) (p
T
=2.93) (p
T
) (p
T
=2.42) (p
T
= 7.85) (p
T
)
8-storey p
k
=0.48 p
k
=0.21 p
k
0 p
k
=0.64 p
k
=0.15 p
k
0
DCH-frame (p
T
=1.30) (p
T
=2.93) (p
T
) (p
T
=2.30) (p
T
= 7.45) (p
T
)
Coecients for the 5th oor column of the 8-storey frame
(pounding between the 8-storey frames and the 4-storey structures (d
g
=0))
4-storey 4-storey 4-storey
Frame structure Framewall structure Stationary barrier
Top end Bottom end Top end Bottom end Top end Bottom end
cross-section cross-section cross-section cross-section cross-section cross-section
8-storey
DCM-frame 1.95 2.14 2.49 2.78 2.85 4.39
8-storey
DCH-frame 1.00 1.26 1.63 1.98 2.42 4.78
Coecients for the 3rd oor column of the 8-storey frame
(pounding between the 8-storey frames and the 2-storey structures (d
g
=0))
2-storey 2-storey 2-storey
Frame structure Framewall structure Stationary barrier
Top end Bottom end Top end Bottom end Top end Bottom end
cross-section cross-section cross-section cross-section cross-section cross-section
8-storey
DCM-frame 1.22 1.31 1.30 1.38 1.41 1.42
8-storey
DCH-frame 1.17 1.10 1.38 1.21 1.40 1.31
cross-section geometry, the strength M +N interaction diagram and the available curvature of
the upper end of the 5th level column in the pounding area of the 8-storey frame are shown
in Figures 5(a), (b) and (c), respectively. The inuence of the pounding and the gap distance
between the adjacent structures on the curvature ductility requirements of the 5th level column
is investigated. The results are presented in Figure 5(d) for the top end cross-sections of the
column, in terms of the overstress coecient
j
. The coecient
j
represents the increased
requirements of the column in curvature ductility due to the pounding eect and is dened
as
j
=j
[, d
g
}j
[
, where j
is the curvature ductility requirement as deduced from seismic

analysis of the 8-storey frame without pounding problems and j
, d
g
is the curvature ductility
requirement of the same cross-section as deduced from seismic analysis of the 8-storey frame
TLFeBOOK
0 2 4 6 8 10 12 14
Time (sec)
-0.12
-0.09
-0.06
-0.03
0.00
0.03
0.06
D
i
s
p
l
a
c
e
m
e
n
t

(
m
)
without pounding
pounding case with 4-story frame structure
pounding case with 4-story rigid structure (very stiff one)
8
7
6
5
4
3
2
1
0 2 4 6 8 10 12 14
Time (sec)
-0.24
-0.18
-0.12
-0.06
0.00
0.06
D
i
s
p
l
a
c
e
m
e
n
t

(
m
)
without pounding
pounding case with 4-story frame structure
pounding case with 4-story rigid structure (very stiff one)
(a)
(b)
Figure 4. Time history comparative results of pounding cases between the DCM 8-storey frame and
the 4-storey structures. Pounding at the oor levels (pounding type A) (for the seismic excitation
of Korinth, Alkyonides, Greece 1981): (a) time history of the 4th level of the 8-storey structure;
and (b) time history of the 8th level of the 8-storey structure.
taking into account the inuence of the adjacent structure and considering various values for
the gap distance (d
g
) between the two structures.
Values of the coecient
j
for the cases of pounding of the 8-storey frames (DCM and
DCH) with (a) the 4-storey DCM frame, (b) the 4-storey DCM framewall structure, and
(c) the 4-storey stationary barrier for various gap distances (d
g
) have been evaluated by means
of non-linear dynamic step-by-step analysis. These values for the DCM frame are presented
in Figure 5(d) for the top end cross-sections of the 5th level column of the 8-storey structure
in the pounding area.
The ductility requirements of the 5th storey column as deduced from analyses without
pounding were 1.22 and 1.65 for the upper and the lower connection cross-sections of the
5th level column, respectively. From the diagrams of Figure 5(d) indicative values for the
inuence of the pounding on the ductility for similar pounding cases can be obtained. Thus,
TLFeBOOK
Example: Ductility requirement
100 200 300
M (kNm)
-1000
1000
3000
5000
N

(
k
N
)
24 6 8 10
ductility
-1000
1000
3000
5000
N

(
k
N
)
0 2 4 6 8 10
Gap distance (cm) - d
g
1
2
3
4
3
2
1
1. pounding with DCM frame
2. pounding with DCM frame-wall structure
3. pounding with rigid (very stiff) structure
- without pounding

,req
= 1.22
- pounding with frame-wall
structure d
g
=0.5 cm

,req,dg
=
,req
=
= 2.25
.
1.22 = 2.75
(a) (b) (c)
(d)
Figure 5. Pounding of 8-storey frames with 4-storey structures (Type A). Values of coecients
j
in
relation to the gap distance (d
g
) of the adjacent structures for the column of the 8-storey frame above
the pounding area as evaluated from the non-linear seismic analyses. Max. coecient values obtained
for two seismic motions (Korinth, Alkyonides, Greece 1981 and El Centro 1940): (a) cross-section;
(b) interaction diagram M+N; (c) available ductility j
,av
= 3.87 (curvature ductility); and (d) values
of cocients
j
for the top cross-section of the 5th level column of the 8-storey frames.
in the case of pounding of the DCM 8-storey frame with the 4-storey frame-wall structure
and gap distance d
g
=0.5cm the ductility requirements can be evaluated based on Figure 5(d)
and similar diagrams as
j
, d
g
=0.5
=
j
j
=2.25 1.22 =2.75 and j

, d
g
=0.5
=
j
j
=2.00 1.65 =3.30

for the top and bottom end cross-sections of the column, respectively. Where the value
j
=2.25 for the top end cross-section is taken from Figure 5(d) and the value
j
=2.00
from similar diagrams for the bottom cross-section [18].
Values for the coecients
j
have been evaluated for all the examined cases for the columns
in the area of the pounding. Values of the coecients
j
for the column of the 5th oor of
the 8-storey structure for the pounding cases of the DCM and DCH 8-storey frames with the
4-storey structures without gap (structures in contact) are presented in Table I. Likewise, the
values of the coecients
j
for the column of the 3rd oor of the 8-storey structure for the
pounding cases of the DCM and DCH 8-storey frames with the 2-storey structures without
gap (structures in contact) are also presented in Table I.
TLFeBOOK
6. POUNDING TYPE B. INTERACTION CASES BETWEEN STRUCTURES WITH
UNEQUAL INTER-STOREY HEIGHTSPOUNDING AT THE COLUMNS
6.1. Examined cases
Thirty-six interaction cases between the 8-storey frame and the 3-storey structures with un-
equal inter-storey heights are examined and discussed. In these cases it is considered that the
pounding takes place at points of the deformable height of the columns of the more exible
8-storey frame structure.
It is expected that the important problem in the case of inter-oor pounding of reinforced
concrete structures (pounding type B) is the development of critical shear state, since in these
cases the demands of exural ductility can more safely be satised. Furthermore, the fact
that the failure of reinforced concrete members due to shear is brittle led the investigation of
inter-oor pounding to the examination of the developing shear forces and their comparisons
to the corresponding shear strength. Pounding at the mid-height of the column produces the
maximum exural moment for the element but not the maximum shear force as well; in fact
the two parts of the column (the upper and the down part) suer the same shear force which
is equal to half of the impact force. Thus pounding cases at points of the column dierent to
the mid-height have been chosen for this investigation.
Series 1. Three interaction cases between the 8-storey DCM frame and the 3-storey structure
are investigated. Each of these cases is examined for three dierent gap distances between
the two structures and is analysed using two seismic excitations. In these interaction cases
the total height of the 3-storey structure is greater than the total height of the 3rd oor and
less than the total height of the 4th oor of the 8-storey frame and thus the contact point of
the two structures lies between the levels of the 3rd and the 4th oor of the 8-storey frame.
Each of these cases is examined for three positions of the contact point.
(a) The highest contact point of the two structures lies between the levels of the 3rd and
the 4th oor of the 8-storey frame at 2}3 of the height of the column of the 4th oor.
(b) The highest contact point is located at 1}3 of the inter-storey height of the column of
the 4th oor of the 8-storey frame.
(c) The total height of the 3-storey structure is equal to the height of the 4th oor of the
8-storey frame. This is pounding type A but is included in this series for comparison
reasons.
Series 2. Similarly to the previously mentioned cases of Series 1, pounding cases between the
8-storey frame and the 3-storey height stationary barrier (very sti structure) are examined in
this series.
(a) The highest contact point of the 8-storey frame with the 3-storey stationary barrier is
located at 1}3 of the inter-storey height of the 4th oor column of the 8-storey frame.
(b) The highest contact point of the 8-storey frame with the 3-storey stationary barrier is
located at 2}3 of the inter-storey height of the 4th oor column of the 8-storey frame.
(c) The contact point is located at the 4th oor level of the 8-storey frame. This is pounding
type A but is included in this series for comparison purposes.
TLFeBOOK
All pounding cases are examined: (a) for structures in contact from the beginning
(gap distance d
g
=0), (b) for initial gap distance between the two structures of d
g
= 2 cm,
and (c) for initial gap distance of d
g
= 5 cm. In this case (d
g
= 5 cm) the structures vibrate
without pounding eect.
Each pounding case is subjected to two dierent natural seismic excitations; the El Centro
1940 earthquake (duration 15 secs and :
max
=0.318q) and the Korinth (Alkyonides) Greece
1981 earthquake (duration 12secs and :
max
=0.306q) (pounding cases type B). The maximum
acceleration (:
max
) of these excitations is very close to the design acceleration of the examined
structures (A=0.3q).
The ratio (p
k
) of the elastic structural stiness of the 8-storey frame to the one of the
3-storey structure is p
k
=0.22. The ratio (p
T
) of the elastic period of the 8-storey frame to
the one of the 3-storey structure is p
T
= 2.51.
6.2. Results
The results and the conclusions deduced from the analyses are sorted and presented in two
parts. The rst part includes the observed overall response of the 8-storey frame and the
ductility requirements of its columns. In the second part of the results attention is focused on
the response of the 4th storey column of the 8-storey frame where the pounding takes place.
Ductility requirements. The curvature ductility requirements for the columns of the 8-storey
frame are presented in Figures 6 and 7 for the pounding cases (a) 8-storey frame and 3-storey
structure and (b) 8-storey frame and 3-storey stationary barrier, respectively. In these cases
the seismic excitation of El Centro is used. Each pounding case includes the study of the
ductility requirements of the external columns at the pounding side of the 8-storey frame for:
(a) contact point at 1}3 of the inter-storey height of the 4th oor column of the 8-storey
frame (Figures 6(a) and 7(a)); (b) contact point at 2}3 of the inter-storey height of the 4th
oor column of the 8-storey frame (Figures 6(b) and 7(b)); (c) contact point at the 4th oor
level of the 8-storey frame.
It is observed that the ductility requirements of the columns of the 8-storey frame, and
especially the ductility requirements of the internal ones, are substantially increased for the
oors above the oor of the contact (4th oor). This is probably attributed to a whiplash
type of behaviour of the taller structure; the whiplash type of behaviour can also be deduced
indirectly from the ductility distributions with height. This kind of response has been also
observed in the interaction cases of pounding type A. The whiplash type of behaviour becomes
especially intense in the pounding cases between the 8-storey frame and the 3-storey stationary
barrier. In this extreme case (p
k
0) the curvature ductility requirements of the upper oor
columns of the 8-storey frame exceed the available curvature ductilities.
Column that suers the pounding. The most important issue in the pounding type B cases is
obviously the local eect on the external column of the tall building that suers impact with
the upper oor slab of the adjacent shorter and stier structure. This impact usually takes
place at a point in the deformable height of the column. The consequences of the impact can
be very severe for the capacity of the column. In this work special attention has been given
for the study of the local response of this structural member. For this purpose two special
TLFeBOOK
8th
7th
6th
5th
4th
3rd
2nd
1st
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0
F
l
o
o
r

l
e
v
e
l
Available Required-Structures in contact (dg = 0)
Required without pounding Required for dg = 2cm
Available Required-Structures in contact (dg = 0)
h/3
2nd
3rd
4th
h
1st
3rd
2nd
h
h
h
8th
3rd
4th
2nd
1st
5th
6th
7th
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0
Curvature ductility,
F
l
o
o
r

l
e
v
e
l
h
h
3rd
2nd
1st
2nd
3rd
4th
2h/3
h
h
(a)
(b)
Figure 6. Pounding type B. Interaction of 8-storey DCM frame with the 3-storey framewall struc-
ture. Ductility requirements of the external columns at the pounding side of the 8-storey frame
(seismic excitation El Centro 1940): (a) pounding at the point h}3 of the 4th oor column; and
(b) pounding at the point 2h}3 of the 4th oor column.
purpose elements of distributed plasticity type are used for the simulation of the behaviour
of this column (see also Figure 2).
Results concerning the exural and the shear demands of this column are presented and
compared with the corresponding available values for all the examined pounding cases. For
the pounding case of the 8-storey frame with the 3-storey structure the ductility requirements
of the external columns (pounding side) of the 8-storey frame for the cases of pounding at the
points 1}3 and 2}3 of the column height are presented in Figures 6(a) and (b), respectively.
In these gures results are presented for the structures in contact from the beginning (d
g
=0)
and for d
g
= 2 cm. From these gures it can be observed that the ductility demands for the
column that suers the pounding impact (4th storey column) are increased when compared
with the ones without the pounding eect (Figure 6) and in the cases where the two buildings
are in contact these demands appear to be higher than the available ductility values. In the
cases where there is a small gap distance (d
g
= 2 cm) between the interacting buildings the
TLFeBOOK
=10.82
(dg=0cm)
=20.29
(dg=0cm)
8th
7th
6th
5th
4th
3rd
2nd
1st
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0
F
l
o
o
r

l
e
v
e
l
Available Required-Stuctures in contact (dg = 0)
h
3rd
2nd
h
4th
h/3
8th
7th
6th
5th
4th
3rd
2nd
1st
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0
F
l
o
o
r

l
e
v
e
l
2h/3
2nd
3rd
4th
h
h
Available Required-Stuctures in contact (dg = 0)
(a)
(b)
Figure 7. Pounding type B. Pounding of 8-storey DCM frame with the 3-storey stationary barrier.
Ductility requirements of the external columns at the pounding side of the 8-storey frame (seismic
excitation El Centro 1940): (a) pounding at the h}3 of the 4th oor column; and (b) pounding
at the 2h}3 of the 4th oor column.
ductility demands of the column are also higher than the ones of the same column with-
out the pounding eect (Figure 6) but they appear to be lower than the available ductility
values.
The developing shear forces of the critical part of the column that suers the impact
for the pounding case of the 8-storey frame with the 3-storey structure are presented in
Figure 8 for the case of pounding at the point 1}3 of the column height. In these gures
results are presented for the case that the two structures are in contact from the beginning
(d
g
=0) and for the case where there is a gap distance equal to 2 cm between the two
structures. In these gures each point represents the developing shear force, J, and the axial
force, N, at a step of the seismic analysis, whereas the lateral solid lines show the available
capacity of the reinforced concrete element for the combination of shear versus axial force
(EC2 and EC8 [7, 8]). This way a direct comparison of the developing shear force at the
TLFeBOOK
-900 900
-2000
N (KN)
V (KN)
Vrd3
Column part A
0.0
El Centro
dg = 0.0cm
Pounding at the point 1/3h of the 4th floor column
-900 900
-2000
N (KN)
V (KN)
Vrd3
Column part A
0.0
El Centro
dg = 2.0cm
A
B
A
B
Available
Shear Strength
Available
Shear Strength
Each point represents the pair of the developing Shear force and Axial force at a step of the seismic analysis
N : Axial force, V : Shear force
Figure 8. Pounding type B. Pounding of 8-storey DCM frame with the 3-storey framewall structure
at the point h}3 of the 4th oor column. Shear forces developed in the critical column (lower part A)
of the 4th storey of the 8-storey frame and available strength.
-900 900
-2000
N (KN)
V (KN)
Vrd3
Column part A
0.0
El Centro
dg = 0.0cm
-900 900
-2000
N (KN)
V (KN)
Vrd3
Column part A
0.0
El Centro
dg = 2.0cm
Available
Shear Strength
Available
Shear Strength
B
A
B
A
Each point represents the pair of the developing Shear force and Axial force at a step of the seismic analysis
N: Axial force , V: Shear force
Figure 9. Pounding type B. Pounding of 8-storey DCM frame with the 3-storey stationary barrier at
the point h}3 of the 4th oor column. Shear forces developed in the critical column (lower part A)
of the 4th storey of the 8-storey frame and available strength.
steps of the analysis with the available shear strength can be presented. It is noted that in
all the examined cases the developing shear forces exceed the shear strength of the column
many times during the excitation.
Analyses results for the pounding cases between the 8-storey frame and a stationary barrier
are presented in Figures 7 and 9. Ductility requirements of the external columns (pounding
TLFeBOOK
side) for the cases of pounding of the 8-storey frame with the stationary barrier at the points
1}3 and 2}3 of the height of the 4-storey column are presented in Figures 7(a) and (b),
respectively. From these gures it can be observed that the ductility demands for the column
that suers the pounding (4th storey column) are increased when compared with the ones
without the pounding eect (Figure 7). In the cases where the two buildings were initially
in contact these demands appear to be higher than the available ductility values. The devel-
oping shear forces of the critical part of the column that suers the impact are presented in
Figure 9 for the cases of the pounding at the point 1}3 of the height of the 4th storey column.
In these gures results are presented for the structures in contact with the barrier from the
beginning of the excitation (d
g
=0) and for the case where there is a gap distance equal to
2 cm between the structure and the barrier. It can be observed that in all the examined cases
the developing shear forces exceeded the shear strength of the column many times during the
seismic excitation.
7. CONCLUSIONS
An investigation of the inuence of the structural pounding on the ductility requirements and
the seismic behaviour of multistorey reinforced concrete structures with non-equal heights is
presented. Special purpose elements of distributed plasticity are employed for the study of the
columns and especially for the columns in the area of the pounding.
Two distinct types of the structural pounding problem are identied and examined. Type A:
collisions may occur between the oor slabs and consequently between the storey masses. Type
B: the heights of the storey levels of the adjacent structures are not equal and consequently
the slabs of one structure hit the columns of the other one.
Seventy-two pounding cases of Type A between structures with unequal total heights are ex-
amined. The inuence of the pounding on the curvature ductility requirements of the columns
in the area of the contact of the structures is studied and presented in the form of coecients
j
. The coecient
j
represents the ratio of the increased ductility requirements (due to the
pounding) to the requirements without the pounding eect. Based on the values of the coe-
cient that are yielded from non-linear dynamic step-by-step analyses and are presented in the
form of diagrams the following remarks can be deduced. (i) Pounding between frame struc-
tures designed according to EC2 and EC8 increased signicantly the ductility requirements
of the columns in the pounding area. Nevertheless, these requirements do not appear to be
critical for all the examined cases of the DCM and the DCH 8-storey frames. (ii) Pounding
of the 8-storey frames with the 4-storey framewall structures yielded critical
j
values only
for the cases were the structures were in contact (d
g
=0). (iii) Values of the
j
coecient
for the pounding of the 8-storey frames with 4-storey stationary barriers proved to be critical
in all the examined cases.
Furthermore, based on the examined cases and the presented diagrams it could be suggested
that for problem cases with similar characteristics the use of the indicative
j
coecients might
be useful in the design process of a structure potentially under pounding.
Thirty-six pounding cases of Type B between structures with unequal total heights and
unequal heights of the storey levels are examined. From the results yielded from the non-
linear dynamic step-by-step analyses the following remarks can be deduced for the examined
cases. (i) Ductility requirements for the columns of the taller structure and especially for the
TLFeBOOK
internal ones are substantially increased for the oors above the highest contact storey level.
This is probably attributed to a whiplash type of behaviour of the taller structure. This kind
of behaviour has also been observed in the examined interaction cases of pounding type A.
(ii) The most important issue in the pounding cases of Type B is the local response of the
external column of the tall structure that suers the impact of the upper oor slab of the
adjacent shorter and stier structure. (iii) It has been observed that the ductility demands
for the column that suers the pounding hit are substantially increased compared with the
ones without the pounding eect. In the cases where the two buildings are in contact these
demands appear to be critical since they are higher than the available ductility values. In the
cases where there is a small gap distance (d
g
= 2 cm) between the interacting buildings the
ductility demands of this column are also higher than the ones of the same column without
the pounding eect but they appear to be lower than the available ductility values. (iv) It has
to be stressed that in all the examined cases the observed shear forces of the critical part of
the column that suers the impact exceed the shear strength of the column. Thus, it can be
concluded that in pounding type B the column that suers the impact is always in a critical
condition due to shear action and, furthermore, in the cases where the two structures are in
contact from the beginning this column appears to be critical due to high ductility demands
as well. This means that special measures have to be taken in the design process rst for the
critically increased shear demands and secondly for the high ductility demands.
Thus, it can be summarized that in situations where pounding potentially may occur, ne-
glecting its possible eects leads to non-conservative building design or evaluation. This obser-
vation becomes critical in cases of pounding type B. The response behaviour trends presented
in this work and especially the indicative coecients
j
may be generalized to other building
analyses provided the problem physical characteristics are similar to those used here. Such
extrapolation may be used to assist in the building design process. Attention must be given
to the key assumptions inherent to this study regarding their applicability to other situations.
REFERENCES
1. Arnold C, Reitherman R. Building Conguration and Seismic Design. Wiley: New York, 1982.
2. Rosenblueth E, Meli R. The 1985 earthquake: Causes and eects in mexico city. Concrete International (ACI)
1986; 8(5):2324.
3. Anagnostopoulos SA. Earthquake induced pounding: State of the art. Proceedings of the 10th European
Conference on Earthquake Engineering 1995; vol. 2. pp. 897905.
4. Bertero VV. Observations on structural pounding. Proceedings of the International Conference on Mexico
Earthquakes, ASCE, 1986; 264287.
5. Anagnostopoulos SA. Building pounding re-examined: How serious a problem is it? Proceedings of the 11th
World Conference on Earthquake Engineering, Acapulco, Mexico, 1996.
6. ACI Committee 318. Building code requirements for structural concrete (ACI 318-95) and Commentary (ACI
318R-95), American Concrete Institute, Detroit, 1995.
7. Eurocode 2. Design of Concrete StructuresPart 1. General Rules and Rules for Building. CEN, Technical
Committee 250}SG2, ENV 1992-1-1, 1991.
8. Eurocode 8. Structures in Seismic Regions, DesignPart 1. General and Building. CEC, Report EUR 12266
EN, May 1994.
9. Athanasiadou CJ, Penelis GG, Kappos AJ. Seismic response of adjacent buildings with similar or dierent
dynamic characteristics. Earthquake Spectra 1994; 10.
10. Anagnostopoulos SA. Pounding of buildings in series during earthquakes. Earthquake Engineering and
Structural Dynamics 1988; 16:443456.
11. Westermo BD. The dynamics of interstructural connection to prevent pounding. Earthquake Engineering and
12. Anagnostopoulos SA, Spiliopoulos KV. An investigation of earthquake induced pounding between adjacent
buildings. Earthquake Engineering and Structural Dynamics 1992; 21:289302.
TLFeBOOK
13. Liolios AA. A numerical approach to seismic interaction between adjacent buildings under hardening and
softening unilateral contact. Proceedings of the 9th European Conference on Earthquake Engineering 1990;
7A:2025.
14. 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.
15. Stavroulakis G, Abdalla K. Contact between adjacent structures. Journal of Structural Engineering (ASCE)
1991; 117(10):28382850.
16. Maison BF, Kasai K. Analysis for type of structural pounding. Journal of Structural Engineering (ASCE)
1990; 116(4):957977.
17. Maison BF, Kasai K. Dynamics of pounding when two buildings collide. Earthquake Engineering and Structural
Dynamics 1992; 21:771786.
18. Karayannis CG, Fotopoulou MG. Pounding of multistorey RC structures designed to EC8 and EC2. 11th
European Conference on Earthquake Engineering (Proceedings in CD form), 1998; Balkema, ISBN 90-5410-
982-3.
19. Rahman AM, Carr AJ, Moss PJ. Seismic pounding of a case of adjacent multiple-storey buildings of diering
total heights considering soil exibility eects. Bulletin of the New Zealand Society for Earthquake Engineering
2001; 34(1):4059.
20. Prakash V, Powell GH, Gampbell S. DRAIN-2DX base program description and users guide. UCB}SEMM
Report No. 17}93, 1993; University of California.
TLFeBOOK
Identication of structural and soil properties from vibration
tests of the Hualien containment model
J. Enrique Luco
1, ,
and Francisco C. P. de Barros
2
1
Department of Structural Engineering, University of California, San Diego, La Jolla,
California 92093-0085, U.S.A.
2
Depto. Ci encias Fundamentais, Radiac oes e Meio Ambiente, Instituto Militar de Engenharia}CNEN,
Praca General Tib urcio 80, CEP 22290-270, Rio de Janeiro, Brazil
SUMMARY
Measurements of the response of the
1
4
-scale reinforced concrete Hualien (Taiwan) containment model
obtained during forced vibration tests are used to identify some of the characteristics of the super-
structure and the soil. In particular, attempts are made to determine the xed-base modal frequencies,
modal damping ratios, modal masses and participation factors associated with translation and rocking
of the base. The shell superstructure appears to be softer than could have been predicted on the basis
of the given geometry and of test data for the properties of concrete. Estimates of the shear-wave
velocity and damping ratio in the top layer of soil are obtained by matching the observed and theo-
retical system frequency and peak amplitude of the response at the top of the structure. The resulting
models for the superstructure and the soil lead to theoretical results for the displacement and rotations
at the base and top of the structure which closely match the observed response. Copyright ? 2004
John Wiley & Sons, Ltd.
KEY WORDS: identication; structures; soils; dynamics; interaction; vibration tests
INTRODUCTION
The extensive data recorded during forced vibration tests of the
1
4
-scale Hualien, Taiwan
containment model [1, 2] are used to test the possibility of using identication techniques
to determine some of the modal properties of the superstructure and some of the charac-
teristics of the top layers of soil. The study focuses on: (i) assessment of the adequacy of
the experimental forced vibration test program to provide data from which structural and
Correspondence to: J. Enrique Luco, Department of Structural Engineering, University of California, San Diego,
La Jolla, California 92093-0085, U.S.A.
Contract}grant sponsor: National Science Foundation, U.S.A.; contract}grant number: BCS-9315680
Contract}grant sponsor: U.S. Nuclear Regulatory Commission
Contract}grant sponsor: CNEN (Brazilian Nuclear Regulatory Commission)
Received 10 February 1997
Revised 3 April 2004
Copyright
?
2004 John Wiley & Sons, Ltd. Accepted 3 April 2004
TLFeBOOK
22 J. E. LUCO AND F. C. P. de BARROS
soil information can be obtained by use of identication techniques, (ii) discussion of vari-
ous structural identication techniques in the presence of signicant soilstructure interaction
eects, (iii) discussion of a direct approach to obtain some of the soil properties of the top
layers of soil, (iv) assessment of the quality of the basic structural and soil data by compar-
ison with the identied properties, and (v) evaluation of a mathematical model to calculate
the dynamic response of structures including soilstructure interaction eects.
The study starts with a brief description of the Hualien containment model including rele-
vant characteristics of the superstructure, foundation and soil. The experimental procedure is
then summarized and followed by a discussion of the characteristics of the observed response
of the structure. The observed response at the top of the structure together with the observed
translation and rotation of the base are used to isolate and determine some of the xed-base
modal characteristics of the superstructure including modal frequencies, modal damping
ratios, modal masses and participation factors associated with the translation and rocking
of the base. Next, the characteristics of the top layer of soil are determined by matching the
observed and calculated responses at the top of the structure. Finally, the resulting models
for the superstructure and the soil are used to calculate a set of theoretical results which are
compared with the observations.
DESCRIPTION OF THE HUALIEN CONTAINMENT MODEL
Characteristics of the superstructure and foundation
The
1
4
-scale Hualien containment model is illustrated in Figure 1. The structure has a total
height of 16.13 m, a basal diameter of 10.82 m and is founded at a depth of 5.15 m below
grade level. The base slab of diameter 10.82 m has a thickness of 3.00 m and rests on a
layer of lean concrete of 0.15 m thickness. The total mass of the foundation (including the
lean concrete) is estimated at 695 10
3
kg. The reinforced concrete containment shell has an
external diameter of 10.52 m, a height of 11.63 m and a uniform thickness of 0.30 m. The
cylindrical roof slab has a diameter of 13.28 m and a thickness of 1.50 m and is supported on
four beams with cross-sections of 0.60 m0.30 m. The top slab has a square 2.20 m2.20 m
access hole at its center. The mass of the shell and top slab are estimated at 264 10
3
and
505 10
3
kg, respectively (after the mass of the beams is partitioned between the slab and
the shell). A variety of tests to determine Youngs modulus for the concrete in the shell
have been conducted at the National Taiwan University. These tests have led to estimates of
E
c
=2.61 10
10
N}m
2
, 2.82 10
10
N}m
2
and 4.05 10
10
N}m
2
based, respectively, on average
28-day cylinder strength, standard compression tests and resonant frequency tests. The values
E
c
=2.82 10
10
N}m
2
(2.88 10
5
kgf }cm
2
), j=2400 kg}m
3
and v =0.16 for concrete were
recommended for the blind prediction exercises.
The wide scatter of the estimates for the Youngs modulus E
c
and the lack of information,
at the time, on the strain levels during the tests, has led the authors to use the average of
the compression and resonant frequency test data (E
c
=3.44 10
10
N}m
2
) as the initial value
for E
c
. In the initial model, the containment shell was modeled as a Timoshenko beam for
horizontal-rocking vibrations and as a hollow shaft for vertical and torsional vibrations. Some
of the xed-base modal characteristics of the superstructure calculated by use of the initial
model are listed in Table I. These results were later conrmed by a nite element model of
TLFeBOOK
VIBRATION TESTS OF THE HUALIEN CONTAINMENT MODEL 23
0.30 0.30
3
.
0
0
Dimensions in meters
G.L.
2.00 2.50 1.50
2
.
5
0
2
.
6
5
0
.
1
5
0.15 0.15
0.60 0.60
3.26 3.26 2.20
1
6
.
2
8
1
.
5
0
0
.
3
0
1.38 1.38
13.28
10.82 2.65
Figure 1. Schematic representation of the Hualien containment model.
Table I. Initial calculated values for xed-base modal characteristics of the
superstructure (based on E
c
=3.44 10
10
N}m
2
).
Type of Modal Fraction of
motion frequency (Hz) total mass
M
1
M
b
1
I
b
1
Horizontal}rocking 12.03 0.822 0.714 1.073 1.012
Torsion 24.21 0.891 0.762 1.082
Vertical 34.86 0.899 0.780 1.073
Modes normalized by response at the top.

the shell. The calculated fundamental xed-base mode involving horizontal}rocking vibrations
has a characteristic frequency of 12.03 Hz (for E
c
=3.44 10
10
N}m
2
) and accounts for 82
per cent of the total mass. The second horizontal}rocking mode has a frequency of 45.59 Hz
and accounts for 9.2 per cent of the total mass. The fundamental xed-base torsional and
vertical modes have frequencies of 24.2 and 34.9 Hz, respectively and account for about 90
per cent of the mass of the superstructure.
TLFeBOOK
0 200 400 600
-20
-18
-16
-14
-12
-10
-8
-6
-4
-2
0
Shear-Wave Velocity (m/sec)
D
e
p
t
h

(
m
)
0 200 400 600
-20
-18
-16
-14
-12
-10
-8
-6
-4
-2
0
Shear-Wave Velocity (m/sec)
A C B
(a)
5 - 12 m
= 0.47
= 2420 kg/m
3
= 0.47
= 0.02
12 - 46 m
= 476 m/sec
= 0.47
= 2420 kg/m
3
= 0.02
12 - 46 m
(b)
= 476 m/sec
= 2420 kg/m
3
= 0.02
5 - 12 m
= 0.47
= 333 m/sec
= 2420 kg/m
3
= 0.02
0 - 2 m,
= 231 m/sec
= 133 m/sec, = 0.38

= 1690 kg/m
3
, = 0.02
2 - 5 m,
= 1930 kg/m
3
= 0.02
= 0.48
Figure 2. (a) Shear-wave velocities below the foundation for Models A, B and
C; and (b) prescribed free-eld velocities.
Characteristics of the soil
The soil at the Hualien site consists of sands and gravels which have been studied extensively
[3] by CRIEPI (Central Research Institute of Electric Power Industry, Japan) and IES (Institute
for Earth Sciences, Taiwan). The characteristics of the free-eld soil over the top 20m of soil
are shown in Figure 2(b). These velocities were obtained by CRIEPI by cross-hole (10 m)
and down-hole (10:20 m) measurements. There is some uncertainty about the abrupt
change in velocity from 333 m}sec to 476 m}sec at the depth of 12 m. This abrupt change
in properties is also shown in the IES down-hole data but is not apparent in the blow count
of the penetration tests conducted at the site. Estimates of the shear-wave velocities obtained
by Tajimi Engineering Services [4] on the basis of correlation of shear-wave velocity with
blow count and eective overburden suggest an average velocity of about 375 m}sec in the
depth range from 12 to 18 m instead of 458 to 474 m}sec as obtained by IES and CRIEPI.
Overall, signicant dierences exist between the CRIEPI, IES and Tajimi ES estimates of
the free-eld velocities [5]. The maximum deviations with respect to the mean of the three
estimates amount to 22, 34, 5 and 14 per cent for the 02 m, 25 m, 512 m and 1218 m
layers, respectively.
After excavation to a depth of 5 m and construction of the containment model and prior
to backll of the soil surrounding the foundation, extensive measurements of the shear-wave
velocity immediately below the foundation (in the depth range from 510 m below grade
level) were undertaken by CRIEPI. The resulting estimates of the shear-wave velocity are
widely scattered over the range from 200 m}sec to 475 m}sec and have a weighted average
value of 317 m}sec.
The participants in the Hualien project [1, 2] were asked to submit blind predictions for the
response of the containment model to forced vibration tests for two prescribed soil models
TLFeBOOK
(Models A and B). The soil properties in both models are identical to the CRIEPI free-eld
properties except for the soil in the rst 7m immediately beneath the foundation (depth range
from 5 to 12 m). In Model A, the soil beneath the foundation in the depth range from 5 to
12m is characterized by the average of the measured shear-wave velocities in that depth range
([ =317 m}sec). In Model B, the shear-wave velocities in this area are based on calculations
reecting the changes in conning stress as a result of excavation and subsequent construction
of the model. In our blind predictions we considered a third model (Model C) based on our
own estimates of the eects of changes of conning stress on shear-wave velocity. It was also
suggested that the eects of strain on the soil shear modulus and damping ratio should not
be considered. Finally, soil damping ratios of 2% were recommended. The characteristics of
the upper layers of Models A, B and C for the soil column below the foundation are shown
in Figure 2(a).
FORCED VIBRATION TESTS AND RESULTS
Description of the forced vibration tests
Harmonic forced vibration tests of the Hualien containment model were conducted by the
Tokyo Electric Power Corporation [6, 7] in November 1992 before the backll soil was put
into place. The experiment included four tests with a shaker exerting a harmonic horizontal
force in two orthogonal directions at the top and base of the model and one test with the
shaker exerting a vertical force at the base of the model. The tests covered frequency ranges
from 2 to 20 Hz and 2 to 25 Hz for horizontal and vertical excitations, respectively.
The measured displacement response components for each excitation frequency were pre-
sented in the form D=Aexp(i[) where [ represents the phase of the response with respect
to the forcing function and A represents the amplitude of the response. Although the shaker
generates a force which varies with the square of the forcing frequency, the amplitude A
provided to the participants was linearly scaled to a force of amplitude 1 tonf (9806 N) for
all frequencies.
For horizontal excitation, the recorded response appears to show a high degree of cross-axis
coupling whereby a force in the NS direction excites not only vibrations in the NS direction
but also vibrations in the orthogonal EW direction and vice-versa. The perpendicular response
component can be as large as 60 per cent of the parallel response component. In addition,
the frequency response curves for both the NS and EW components instead of presenting one
peak show two peaks at frequencies of 4.14.2 and 4.6 Hz. The cross-axis coupling and the
bimodal characteristics of the response have been ascribed to azimuthal variations of the soil
properties in the vicinity of the foundation. It has been found [6] that the coupling between
the response in two directions could be minimized if the axes are rotated counterclockwise
by : =34
(it should be noted that the nominal North direction is 31.4
to the west of the

true North).
It must be noted here that the process of scaling the data to a nominal force of 9806 N
and of combining the results of NS and EW main and sub-tests to obtain the response for
excitation in the D
1
(N34
W) and D
2
(N124
W) directions presumed that the structure, soil

and data acquisition system are responding linearly.
TLFeBOOK
0
50
100
150
200
250
300
A
m
p
l
i
t
u
d
e

(
m
/
t
o
n

f
)
2 3 4 5 6 7 8
-60
0
60
120
180
240
300
Frequency (Hz)
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0
25
50
75
100
2 3 4 5 6 7 8
-60
0
60
120
180
240
300
Frequency (Hz)
0
1
2
2 6 10 14 18 22 26
-60
0
60
120
180
240
300
Frequency (Hz)
Horizontal Force at Top Horizontal Force at Base Vertical Force at Base
(a)
D
1
D
2
(b)
D
1
D
2
(c)
D
1
D
2
(d)
D
1
D
2
(e)
(f)
U
T
H
T
H
b
U
b
U
T
H
T
H
b
U
b
U
T
U
b
Figure 3. Amplitude and phase of the experimental response for horizontal excitation at the top (a, b);
horizontal excitation at the base (c, d); and vertical excitation at the base (e, f).
Characteristics of the experimental response
The amplitude and phase characteristics of the rotated response for the conditions prior to
backll are illustrated in Figure 3 which shows the amplitude (a c) and phase (b d) of the total
horizontal displacement

U
T
, the total rotations H

T
and H

b
at the top and base normalized
by the height H =13.13 m, and the total horizontal displacement

U
b
at the top of the base
slab. The results labeled D
1
correspond to the response in the direction D
1
=N34
W for a
horizontal force of amplitude 1 tonf (9806 N) acting at the top (a b) or base (c d) of the
structure in direction D
1
. The results labeled D
2
correspond to the response in the direction
D
2
=N124
W for a force at the top or base of the structure acting in the direction D
2
. Finally,
Figures 3(e) and (f) show the amplitude and phase of the total vertical displacements at the
top

U
T
and base

U
b
of the structure for a vertical force of amplitude 1tonf (9806 N) applied
at the base of the structure.
All of the phase angle data shown in Figure 3 show signicant uctuations for frequen-
cies below 4 Hz. The amplitude data for horizontal excitation at the base (Figure 3(c))
TLFeBOOK
and for vertical excitation (Figure 3(e)) also show strong uctuations for frequencies
below 46 Hz.
Prior to backll, the peak response of the model in direction D
1
occurs at frequencies of
4.1 and 4.2 Hz for horizontal excitation at the top and base of the structure, respectively. The
peak response for horizontal excitation in the D
2
direction occurs at a frequency of 4.6 Hz
independently of the location of the shaker. The amplitude of the response in the D
1
direction
is 20 to 40 per cent larger than the response in the D
2
direction, depending on the response
component and on the location. For vertical excitation, the peak response at the top and base
of the structure occur at frequencies of 11.0 and 9.5 Hz, respectively.
The eects of soilstructure interaction on the response are quite large. Rigid-body rocking
with respect to the bottom of the base slab accounts for 61 to 69 per cent of the total
response at the top of the structure at the fundamental system frequency. Rigid-body swaying
of the foundation accounts for 11 to 13 per cent of the total response at the top while the
deformation of the structure only accounts for 21 to 26 per cent of the total response. For
vertical vibrations, the motion of the base accounts for 83 per cent of the total motion at
the top.
IDENTIFICATION OF STRUCTURAL PROPERTIES
Model of the superstructure
To identify structural properties out of forced vibration test results that include a signicant
amount of soilstructure interaction it is necessary to consider that the deformation of the
superstructure depends on the force applied by the shaker and also on the translation and
rotation of the base. To derive the necessary equations we consider the lumped mass model
shown in Figure 4. In the case shown, the superstructure is excited by the force F
T
e
ict
that
the harmonic shaker exerts at the top of the structure. The total translation at the top of the
rigid foundation is represented by

U
b
e
ict
and the total harmonic rotation of the base about
the horizontal axis is represented by

b
e
ict
.
The total harmonic displacement

U
)
e
ict
() =1, N) and the total rotation

)
e
ict
() =1, N)
at the )-th level can be written in the form
U
)
=

U
b
+ h
)

b
+ U
)
(1)
)
=

b
+ [
)
(2)
where U
)
and [
)
represent the relative displacement and relative rotation at the )-th level
with respect to a system of coordinates attached to the rigid foundation. In Equation (1), h
)
denotes the height of the )-th level with respect to the top of the foundation. Equations (1)
and (2) can be summarized in the form
{

U} ={1}

U
b
+ {h}

b
+ {U} (3)
where {

U} =(

U
1
,

1
,

U
2
,

2
, . . . ,

U
N
,

N
)
T
and {U} =(U
1
, [
1
, U
2
, [
2
, . . . , U
N
, [
N
)
T
represent
the total and relative displacement vectors, respectively, {1} =(1, 0, 1, 0, . . . , 1, 0)
T
and {h} =
(h
1
, 1, h
2
, 1, . . . , h
N
, 1)
T
.
TLFeBOOK
Figure 4. Lumped mass model for derivation of identication approach.

The equation of motion of the superstructure in terms of the generalized relative displace-
ment is given by
(c
2
[M] + ic[C] + [K]){U} ={F} + c
2
[M]({1}

U
b
+ {h}

b
) (4)
where [M], [C] and [K] are the xed-base mass, damping and stiness matrices, and {F}
=(0, 0, . . . , 1, 0)
T
F
T
represents the force vector acting on the superstructure.
Now, for vibrations in the vicinity of the fundamental xed-base natural frequency of the
superstructure we can approximate the deformation of the superstructure by
{U} ={[
(1)
}U
T
(5)
where U
T
=U
N
is the relative translational displacement at the top of the superstructure and
{[
(1)
} is the fundamental xed-base mode normalized so that the translation at the top is
equal to one. Substitution from Equation (5) into Equation (4), pre multiplication by {[
(1)
}
T
and use of Equation (1) for ) =N lead to
U
T
=

U
b
+ H

b
+
F
T
z
1
c
2
M
1
+ [
1

U
b
+
1
H

b
(c
1
}c)
2
1 + 2i
1
(c
1
}c)
(6)
where

U
T
=

U
N
=U
T
+

U
b
+ H

b
is the total displacement at the top of the structure and
H =13.13m is the height from the top of the base mat to the top of the roof slab. The modal
mass M
1
and the participation factors [
1
and
1
appearing in Equation (6) are given by
M
1
={[
(1)
}
T
[M]{[
(1)
} (7a)
TLFeBOOK
[
1
=
1
M
1
{[
(1)
}
T
[M]{1} (7b)
1
=
1
HM
1
{[
(1)
}
T
[M]{h} (7c)
The xed-base natural frequency c
1
and the xed-base modal damping ratio
1
are given by
c
2
1
=
1
M
1
{[
(1)
}
T
[K]{[
(1)
} (8a)
1
=
1
2M
1
c
1
{[
(1)
}
T
[C]{[
(1)
} (8b)
The term z
1
appearing in Equation (6) represents a correction that needs to be incorporated
when the shaker acts at a location dierent from that of the observation point at the top of the
structure. The term z
1
is dened as the ratio of the amplitude of the translational component
of the mode shape {[
(1)
} at the location of the shaker divided by the corresponding amplitude
at the location of the observation point at the top of the structure. In the present case, when
the shaker is acting at the top of the structure z
1
=1.0345.
During forced vibration tests the total displacements

U
T
and

U
b
and the total rotation

b
are measured at dierent frequencies c for a known force F
T
. All or some of the structural
properties M
1
, [
1
,
1
, c
1
and
1
can then be determined by tting Equation (6) to the observed
data. In here we note that if the shaker is acting on the oor slab (base excitation) then the
term (F
T
}c
2
M
1
) appearing in Equation (6) should be deleted (z
1
=0). In addition, for vertical
excitation the term involving

b
should be deleted from Equation (6).
Approximate identication approach (Method 1)
To start the process of identication of structural properties on the basis of the observed
response during forced vibration tests we consider Equation (6) at the system frequency c
1
.
From Equation (6) we obtain
(c
1
} c
1
)
2
1 =Re
_
F
T
z
1
c
2
1
M
1
U
T
+ [
1
U
b
U
T
+
1
H

b
U
T
_
c
1
(9)
where U
T
=

U
T

U
b
H

b
is the relative displacement at the top of the structure.
Since

U
T
,

U
b
and H

b
are approximately in phase at the system frequency c
1
and are
90
out of phase with respect to F

T
(Figure 3), we nd that
c
1
c
1
U
T
+ [
1

U
b
+
1
H

b
U
T
c
1
(10)
which reduces to
c
1
c
1
U
T
U
T
c
1
(11)
TLFeBOOK
Table II. Inferred modal properties by dierent identication methods.
Top force Base force
Modal Average Base force
property Method D
1
D
2
D
1
D
2
horizontal vertical
[
1
(Hz) 1 9.06 9.72 8.32 9.72 9.21 26.76
2 9.25 9.25 9.70 9.70 9.48 27.60
3 9.05 9.67 9.39 9.74 9.46 27.35
4 9.23 8.81 9.29 9.72 9.26
100
1
3 6.83 4.42 0.64 0.57 3.12 3.11
4 4.24 11.08 1.07 0.16 4.14
:
1
3 1.359 1.561 1.460
4 1.311 1.195 1.253
[
1
3 1.169 1.216 1.203 1.071 1.165 1.009
4 1.147 0.940 1.444 1.011 1.136
1
3 0.975 1.188 1.063 1.027 1.063
4 0.950 0.803 1.242 0.955 0.988
after the approximations [
1
1 and
1
1 are introduced. An alternative derivation of Equa-
tion (11) has been presented by Luco et al. [8, 9].
Applying Equation (11) to the data for the vibration tests prior to backll (FVT-1) leads to
the results for [
1
=c
1
}2 listed in Table II. The obtained values of [
1
for horizontal}rocking
vibrations range from 8.32 to 9.72Hz, while the theoretical value based on the initial model of
the superstructure was 12.03 Hz. For vertical vibrations, the identied value of [
1
is 26.76 Hz
which must be compared with the theoretical value of 34.86 Hz based on the initial model of
the superstructure.
The large dierences between the estimates of the xed-base natural frequencies [
1
based on
the FVT-1 data and the initial theoretical values suggest that the assumed value for concretes
Youngs modulus E
c
was largely overestimated or that the structure as built diers from the
initial plans. To match the vertical frequency of 26.76 Hz it would be necessary to reduce E
c
from 3.44 10
10
N}m
2
to 2.02 10
10
N}m
2
which would be 28% lower than the recommended
value of 2.82 10
10
N}m
2
.
Identication frequency-by-frequency (Method 2)
A second identication approach [8, 9] assumes that the modal properties M
1
, [
1
and
1
are
known. In this approach Equation (6) is written in the form
(c
1
}c)
2
1 + 2i
1
(c
1
}c) =A(c) (12)
where
A(c) =
:
1
F
T
z
1
c
2
M
b
+ [
1

U
b
+
1
H

U
T

U
b
H

b
(13)
TLFeBOOK
and :
1
=M
b
}M
1
in which M
b
=769 10
3
kg is the total mass of the superstructure. The term
A(c) involves known or measured quantities. Taking real and imaginary parts of Equation (12)
results in
c
1
=c
_
1 + Re A(c) (14)
and
1
=
ImA(c)
2
_
1 + Re A(c)
(15)
which lead to independent estimates of c
1
and
1
for each forcing frequency. Again, if
the force is applied at the base then the term F
T
}c
2
M
b
should be deleted from Equa-
tion (13). In the case of vertical vibrations the terms involving

b
should also be deleted from
Equation (13).
The results of applying Equations (14) and (15) to the FTV-1 data prior to backll are
shown in Figure 5 as a function of the excitation frequency [. The calculations for horizontal
excitation are based on :
1
=1.400, [
1
=1.073, and
1
=1.012. The parameter z
1
takes values
z
1
=1.0345 for excitation at the top and z
1
=0 for excitation at the base. For vertical excitation
at the base the only required parameter is [
1
=1.073. The estimate of the fundamental xed-
base frequency for vertical vibration is about 27.6 Hz and this estimate is quite stable for
excitation frequencies above 8 Hz. For horizontal vibrations induced by a horizontal force
at the base of the structure, the estimates of [
1
in the directions D
1
and D
2
converge to a
value of [
1
=9.70 Hz for excitation frequencies above 8 Hz. The estimates of [
1
from tests
involving a horizontal force at the top of the structure lead to values that oscillate between
8.0 and 10.5 Hz with a best estimate of [
1
=9.25 Hz.
The fact that the results for [
1
obtained from tests with excitation at the base which do
not involve the term F
T
}c
2
M
1
U
T
in Equation (13) are more stable at high frequencies than
those involving a force at the top suggest that there may be some error in the experimental
values of the phase of the response with respect to the forcing function. This phase dierence
aects the estimates of [
1
and
1
for forces at the top of the structure but not for forces at
the base.
The results obtained for the xed-base damping ratio
1
for horizontal}rocking vibrations
are quite erratic. There is some indication that the damping ratio for horizontal excitation at
the base is less than 2 per cent, while larger values would apply to the case of excitation
at the top. The result for vertical excitation would correspond to a damping ratio inversely
proportional to the cube of the forcing frequency. In general, it seems that data for the
relative phase between response components and the external force are not accurate enough
for a precise identication of the damping ratio within the superstructure.
Identication by use of the amplitude of the frequency response (Method 3)
In a third approach we attempt to determine all of the modal properties M
1
}M
b
, [
1
,
1
, c
1
and
1
by imposing Equation (6) at all of the observation frequencies. In particular, we
TLFeBOOK
2 6 10 14 18 22 26
0
6
12
18
24
30
36
V
e
r
t
i
c
a
l

F
o
r
c
e

a
t

B
a
s
e
2 6 10 14 18 22 26
.00
.02
.04
.06
.08
.10
2 4 6 8 10 12 14
0
4
8
12
16
H
o
r
i
z
o
n
t
a
l

F
o
r
c
e

a
t

B
a
s
e
2 4 6 8 10 12 14
.00
.02
.04
.06
.08
.10
2 4 6 8 10 12 14
0
4
8
12
16
Forcing Frequency (Hz)
H
o
r
i
z
o
n
t
a
l

F
o
r
c
e

a
t

T
o
p
2 4 6 8 10 12 14
.00
.02
.04
.06
.08
.10
Forcing Frequency (Hz)
f
1
f
1
f
1

1

1

1
Fixed-Base Frequency f
1
(Hz) Fixed-Base Damping
1
Direction D
1
Direction D
2
Figure 5. Estimates of the xed-base frequency [
1
(Hz) and of the xed-base modal damping ratio
1
as a function of the forcing frequency for vertical excitation at the base, horizontal excitation at the
base (solid D
1
, segmented D
2
) and horizontal excitation at the top (solid D
1
, segmented D
2
).
require that the observed amplitude of the relative displacement at the top |U
T
| be matched
by the corresponding theoretical value, i.e.
|U
T
| =
:
1
_
F
T
z
1
c
2
M
b
_
+ [
1

U
b
+
1
H

|(c
1
}c)
2
1 + 2i
1
(c
1
}c)|
(16)
TLFeBOOK
where z
1
=1.0345 and :
1
=M
b
}M
1
in which M
b
is the total mass of the superstructure. For
this purpose we introduce the function
q(c, :
1
, [
1
,
1
, c
1
,
1
) =
_
_
|U
T
|
:
1
_
F
T
z
1
c
2
M
b
_
+ [
1

U
b
+
1
H

_
[(c
1
}c)
2
1]
2
+ [2
1
(c
1
}c)]
2
_
_
1
[(c)
(17)
where
[(c) =:
1o
F
T
z
1
c
2
M
b
+ [
1o
|

U
b
| +
1o
|H

b
| (18)
is a normalizing factor in which :
1o
=1.400, [
1o
=1.073 and
1o
=1.012 are reference values
for :
1
, [
1
and
1
, respectively. To nd the modal properties :
1
=M
b
}M
1
, [
1
,
1
, c
1
and
1
that minimize the error q we write
:
1
= :
1
+ :
1
, [
1
=

[
1
+ [
1
,
1
=
1
+ o
1
(19a)
c
1
= c
1
+ c
1
,
1
=

1
+
1
(19b)
where :
1
,

[
1
,
1
, c
1
and

1
are initial estimates of the modal properties. The next step is to
linearize q in terms of :
1
, [
1
, . . . ,
1
. The resulting equations can be written in the form
[a]
_
_
:
1
[
1
1
c
1
1
_
_
={b} (20)
where
[a] =
_
_
cq
c:
1
[
1
cq
c[
1
[
1
cq
c
1
[
1
cq
cc
1
[
1
cq
c
1
[
1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
cq
c:
1
[
M
cq
c[
1
[
M
cq
c
1
[
M
cq
cc
1
[
M
cq
c
1
[
M
_
_
(21)
and
{b} =(q|
[
1
, . . . , q|
[
M
)
T
(22)
The elements of [a] and {b} are calculated at a set of observation frequencies [
1
, [
2
, . . . , [
M
and at the assumed initial values for the modal properties :
1
,

[
1
,
1
, c
1
and

1
. Expressions
for the derivatives appearing in Equation (21) can be obtained analytically [10].
TLFeBOOK
Equation (20) can be solved in the least squares sense and new estimates :
1
+:
1
,

[
1
+[
1
,

1
+
1
, c
1
+ c
1
and

1
+
1
for the modal properties can be calculated. The process
can be repeated until stable estimates of :
1
, [
1
,
1
, c
1
and
1
are obtained.
The values of the modal characteristics c
1
,
1
, :
1
, [
1
and
1
identied by use of Equa-
tion (20) (Method 3) for horizontal excitation at the top and base of the structure are listed in
Table II. All of the data in the frequency ranges of 2 to 20Hz (horizontal vibrations) and 2 to
25Hz (vertical vibrations) were used in the identications. The average inferred values for the
mass ratio :
1
, for the horizontal participation factor [
1
and the rocking participation factor
1
dier by 4.3%, 8.1% and 5.0% from the corresponding theoretical values of 1.400, 1.073 and
1.012, respectively. For vertical vibrations, the identied value for the vertical participation
factor [
1
is 1.009 which diers by 6.0% from the theoretical value 1.073.
The identied values for the horizontal xed-base natural frequency [
1
depend on the
direction of excitation and location of the shaker. The obtained results for [
1
have an average
value of 9.46 Hz. The values of [
1
for the larger vibrations induced by excitation at the top
are somewhat lower than those for excitation at the base (9.36 Hz versus 9.56 Hz). Also,
the identied values of [
1
for the larger vibrations in direction D
1
are lower than those in
the D
2
direction (9.22 Hz versus 9.70 Hz). However, it must be kept in mind that additional
identications of [
1
and
1
for prescribed values of :
1
, [
1
and
1
listed elsewhere [10] indicate
that the dierences between the estimates of [
1
in the D
1
and D
2
directions tend to disappear
when the same values of :
1
, [
1
and
1
are used in both directions. In particular, for the
theoretical values of :
1
=1.40, [
1
=1.073 and
1
=1.012, the dierence between the values
for [
1
in the D
1
and D
2
directions amount to 0.4 and 1.0% for horizontal excitation at the top
and base, respectively. For vertical vibrations, the estimate of [
1
is 27.35 Hz which is close
to the estimate obtained by the second method.
The inferred values for the xed-base modal damping
1
are not very stable. Small variations
of the parameters :
1
, [
1
and
1
lead to signicant changes in the values obtained for
1
.
The inferred values of
1
for base excitation appear to be signicantly lower than those for
excitation at the top (0.60% versus 5.63% as averages over D
1
and D
2
directions). While
the larger damping values for horizontal excitation at the top may reect the eect of larger
deformations of the superstructure, it is more likely that this result may reect the error in
the measurement of the phase of the response with respect to the forcing function. This error
aects the estimates of
1
for excitation at the top which depend on F
T
but not the estimates
for excitation at the base which are independent of F
T
.
The quality of the t between theory and experiment can be judged by comparison of the
experimental and theoretical values for the amplitudes and phases of the relative displacement
given by
(U
T
)
exp.
=

U
T
H

b

U
b
(23)
and
(U
T
)
theor.
=
:
1
_
F
T
z
1
c
2
M
b
_
+ [
1

U
b
+
1
H

b
_
c
1
c
_
2
1 + 2i
1
_
c
1
c
_
(24)
It should be noted that the theoretical relative displacement given by Equation (24) is really a
hybrid term that depends on both observed quantities (F
T
,

U
b
,

b
) and on estimated properties
TLFeBOOK
2 4 6 8 10 12
0
20
40
60
80
H
o
r
i
z
o
n
t
a
l

F
o
r
c
e

a
t

T
o
p
2 4 6 8 10 12
0
20
40
60
80
2 4 6 8 10 12
0
5
10
15
20
25
Frequency (Hz)
H
o
r
i
z
o
n
t
a
l

F
o
r
c
e

a
t

B
a
s
e
2 4 6 8 10 12
0
5
10
15
20
25
Frequency (Hz)
Amplitude (m/ton f) Amplitude (m/ton f)
(a) (b)
(c) (d)
D
1
D
2
D
1
D
2
D
1
D
2
D
1
D
2
Experimental
Theoretical
Experimental
Theoretical
Figure 6. Comparison of experimental (solid lines) and theoretical (segmented lines)
relative displacements (m}tonf ) for horizontal excitation at the top (a, b); and base
(c, d) for Methods 3 (a, c) and 4 (b, d).
(:
1
, [
1
,
1
, c
1
,
1
). For horizontal excitation at the top, the agreement between the experimental
and theoretical relative displacements is very good in direction D
1
and reasonable in direction
D
2
(Figure 6(a)). The comparisons in Figure 6(c) for horizontal excitation at the base indicate
good agreement for the relative displacement in direction D
2
and poor agreement in direction
D
1
. Finally, for vertical excitation at the base good agreement between the theoretical and
experimental results for frequencies above 10 Hz is obtained [10].
The results obtained indicate that an identication approach based on matching the amplitude
of the observed and calculated relative displacements at the top of the structure can be used
to determine :
1
, [
1
,
1
and c
1
with some condence but the xed-base modal damping ratio
1
is not well determined. The problem may be intrinsic to the approach and may result from
the large dierence between the system frequency at which the peak response occurs and the
xed-base natural frequency. On the other hand, the possible experimental error in the values
of the phase of the response with respect to the exciting force may be the source of most
of the diculties in trying to determine the xed-base modal damping ratio. The phase data
below 4Hz for horizontal vibrations and below 6Hz for vertical vibrations are clearly in error.
Identication by use of complex frequency response (Method 4)
Finally, we attempt to use both the amplitude and phase information available to determine
the modal properties M
1
}M
b
, [
1
,
1
, c
1
and
1
. In this case, we impose Equation (6) in its
TLFeBOOK
complex form at all of the observation frequencies. For this purpose we introduce the new
complex function
q(c, :
1
, [
1
,
1
, c
1
,
1
) =
_
_
U
T

:
1
_
F
T
z
1
c
2
M
b
_
+ [
1

U
b
+
1
H

b
(c
1
}c)
2
1 + 2i
1
(c
1
}c)
_
_
(25)
where z
1
=1.0345 and :
1
=M
b
}M
1
. The function q represents the (complex) error of the
theoretical relative displacement with respect to the experimental relative displacement U
T
=
U
T

U
b
H

b
at the top of the structure.
To nd the modal properties :
1
=M
b
}M
1
, [
1
,
1
, c
1
and
1
that minimize the error q
we write again :
1
= :
1
+ :
1
, [
1
=

[
1
+ [
1
,
1
=
1
+ o
1
, c
1
= c
1
+ c
1
,
1
=

1
+
1
where :
1
,

[
1
,
1
, c
1
and

1
are initial estimates of the modal properties and linearize q in
terms of :
1
, [
1
, . . . ,
1
. The resulting equations, after separating the real and imaginary
parts of q, can be written in the form
_
Re[a]
Im[a]
_
_
_
:
1
[
1
1
c
1
1
_
_
=
_
Re{b}
Im{b}
_
(26)
where the matrix [a] and the vector {b} have the same form as in Equations (21) and
(22), respectively. The elements of [a] and {b} are calculated at the observation frequen-
cies [
1
, [
2
, . . . , [
M
and at the assumed initial values for the modal properties :
1
,

[
1
,
1
, c
1
and

1
. Again, the derivatives appearing in Equation (21) are obtained analytically from
Equation (25). Equation (26) can be solved in the least squares sense and new estimates
of :
1
+ :
1
,

[
1
+ [
1
,
1
+
1
, c
1
+ c
1
and

1
+
1
for the modal properties can
be calculated. The process can be repeated until stable estimates of :
1
, [
1
,
1
, c
1
and
1
are obtained. In the case of forces applied to the foundation the term :
1
(F
T
}c
2
M
b
) appear-
ing in Equation (25) should be deleted and the parameter :
1
should be excluded from the
identication.
The modal characteristics c
1
,
1
, :
1
, [
1
and
1
identied by solving Equation (26) in the
least square sense for horizontal excitation at the top and base of the structure are listed in
Table II. The identications considered only the data in the frequency range from 3 to 14Hz.
The average of the inferred values for :
1
, [
1
and
1
dier by 11%, 6% and 2%, respectively,
from the initial theoretical values.
The inferred value for the xed-base natural frequency in horizontal}rocking vibrations [
1
appears to depend on the direction of excitation and location of the shaker. The results for
[
1
have an average value of 9.26Hz. The identied value [
1
for excitation at the top appears
to be somewhat lower than for excitation at the base (9.02 Hz vs. 9.51 Hz).
The inferred values for the xed-base modal damping ratio
1
appear to be much larger
for excitation at the top of the structure than for base excitation.
TLFeBOOK
Comparisons of the experimental and theoretical relative displacements calculated on the
basis of the parameters listed in Table II are shown in Figure 6(b) and (d). In general, the t
to the relative displacement is very good for excitation at the top and base of the structure. In
general, it appears that the identication approach which considers both the real and imaginary
parts of the relative displacement (Method 4) leads to slightly better ts than the approach
based only on the amplitude of the relative displacement (Method 3).
Summary of structural identication
Averaging over the direction of excitation, the location of the shaker and over two identica-
tion methods (3 and 4) we nd the representative values :
1
=1.357, [
1
=1.151 and
1
=1.026
for horizontal}rocking vibrations which can be compared with the theoretical values :
1
=1.400,
[
1
=1.073 and
1
=1.012. The average (over directions D
1
and D
2
and Methods 3 and 4) of
the identied values for [
1
for horizontal excitation at the top and base of the structure are
9.19 and 9.54 Hz, respectively. The corresponding values for the xed-base modal damping
ratio
1
are 6.64% and 0.61%, respectively.
The inferred values of [
1
averaged over source location and identication methods (3 and
4) correspond to 9.24 and 9.49Hz, respectively, for directions D
1
and D
2
. This 2.7% dierence
is not considered signicant and, in particular, does not explain the 12% dierence between
the observed system frequencies

[
1
in directions D
1
and D
2
. In fact, since

[
1
=(

[
1
}[
1
)
3
[
1
,
a 2.7% change in the xed-base frequency [
1
would imply only a 0.5% change in the system
frequency

[
1
if everything else remains constant.
In future calculations we will ignore the dierences between the values of [
1
and
1
obtained
for excitation at the top and base of the structure and in directions D
1
and D
2
, and use the
average values of [
1
=9.37Hz and
1
=3.6%. It is quite likely that the estimate of
1
=3.6%
is not accurate as already indicated. For vertical vibrations we have obtained [
1
=27.35 Hz,
1
=3.1% and [
1
=1.009.
It should be noted that Kobayashi et al. [11] have also used modal identication to de-
termine estimates of the fundamental xed-base frequency and modal damping ratio for the
simpler case of horizontal excitation at the base in which the term (:
1
F
T
z
1
}c
2
M
b
) is absent.
By using prescribed values for the modal participation factors [
1
and
1
, Kobayashi et al.
[11] obtained xed-base frequencies of 9.42 and 9.56 Hz in the direction D
1
and D
2
, respec-
tively. These values can be compared with the present results of 9.34 and 9.73Hz obtained by
averaging the results of Methods 3 and 4. Kobayashi et al. [11] also obtained modal damping
ratios of 0.9% and 0.4%, respectively for directions D
1
and D
2
which can be compared with
the present results of 0.86% and 0.37%.
Revised structural model
The results of the identication of the xed-base natural frequencies [
1
for horizontal}rocking
vibrations and for vertical vibrations indicate that the superstructure is considerably softer than
initially assumed. Based on the identication studies a revised model for the superstructure
was derived (Table III). This model is the same model used for the blind prediction except
that the Youngs modulus for concrete is reduced to E
c
=2.084 10
10
N}m
2
. This model leads
to calculated xed-base natural frequencies for horizontal}rocking and vertical vibrations of
9.37 Hz and 27.15 Hz, respectively, which coincide with or are close to the average identied
values of 9.37 Hz and 27.35 Hz. The relatively low value for E
c
may reect an actual low
TLFeBOOK
Table III. Fixed-base modal characteristics of the superstructure. Blind prediction model,
mean inferred values and revised model.
Blind prediction Mean Revised model
(E
c
= 3.44 10
10
N}m
2
) inferred values (E
c
= 2.084 10
10
N}m
2
)
[
1

1
[
1

1
[
1

1
(Hz) (%) (Hz) (%) (Hz) (%)
Horiz}rocking 12.03 0.5 9.37 3.6 9.37 3.6
Horiz}rocking 45.59 0.5 35.50 3.6
Torsion 24.21 0.5 18.85 3.6
Vertical 34.86 0.5 27.35 3.1 27.15 3.1
value of the Youngs modulus for the concrete used in the structure or it may reect other
weaknesses of the structure as built. This low value for E
c
is in excellent agreement with
the result of Yun et al. [12] who obtained E
c
=2.11 10
10
N}m
2
as a mean value over three
segments of the shell, and in relative agreement with the result of Kobayashi et al. [11] who
obtained E
c
=2.38 10
10
N}m
2
.
In the revised model, the xed-base modal damping ratios for horizontal}rocking and vertical
vibrations are also increased to 3.6 and 3.1 per cent, respectively, to match the average
identied values. Although there is some indication that the structural damping during the
smaller vibrations induced by the shaker placed at the base is smaller ( 0.61%) than the
damping during the stronger vibrations induced by the shaker placed at the top ( 6.5%)
we have enough doubts about these results that we prefer to use the overall average of 3.6%
for horizontal}rocking vibrations. This average value is probably somewhat large but does not
aect drastically the response in the vicinity of the system frequency.
The revised model of the structure and various soil models are used in the next two
sections to calculate the response of the Hualien containment model during forced vibration
tests prior to backll (FVT-1). The calculations are done by use of the CLASSI approach
in which the soil underneath the foundation is modeled as a layered half-space consisting of
several parallel viscoelastic layers overlying a uniform viscoelastic half-space. The foundation
is represented as a rigid cylinder of diameter 10.82 m, height 3.15 m and mass 695 10
3
kg.
The containment shell is modeled as a Timoshenko beam with an external diameter of 10.52m,
eective height of 11.555 m, and thickness of 0.30 m. The concrete shell is characterized by
E
c
=2.084 10
10
N}m
2
, j=2400 kg}m
3
and v =0.16. The top of the containment model is
represented as a rigid block with an external diameter of 13.28 m, an equivalent height of
1.575 m and with an internal cavity of equivalent diameter 2.42 m. The mass of the top slab
was estimated at 505 10
3
kg while the total mass of the superstructure (shell and top slab)
was estimated at 769 10
3
kg.
The xed-base model characteristic of the superstructure was calculated by representing the
shell by 40 elements distributed over the height. A total of 6 xed-base modes were used in
the analysis (two horizontal}rocking modes in each of the D
1
and D
2
directions, one vertical
mode and one torsional mode). Fixed-base modal damping ratios of 3.6 per cent were used for
horizontal, rocking and torsional modes and 3.1 per cent for the vertical mode. The response
was calculated for frequencies in the range from 0 to 20 Hz with a step of 0.02 Hz.
TLFeBOOK
IDENTIFICATION OF SOIL PROPERTIES
Some of the characteristics of the soil are determined by a relatively simple and direct ap-
proach based on matching the observed values for the system frequencies and peak amplitudes
of the response at the top of the structure with those calculated by use of the CLASSI method-
ology. This approach allows us to consider separately the cases of excitation at the top and
base of the structure which lead to dierent response amplitudes and may require the use of
dierent soil models as a result of the strain dependence of the soil properties. A dierent
identication of soil properties based on comparison of experimentally-based and theoretical
impedance functions has been presented elsewhere [5]. The later approach combines data from
the tests with excitation at the base and top of the structure and assumes that an unchanged
soil model applies to both cases.
In the context of the isotropic soil models being employed here it is necessary to develop
two soil models: one for direction D
1
and another for direction D
2
. To simplify the problem
we consider as variables the shear-wave velocity

[
S
and the material damping ratio

S
in the
7 metre thick layer immediately below the foundation. The Poissons ratio v =0.47, density
j=2420 kg}m
3
, and ratio of P to S damping ratios
P
}
S
=0.10 for this layer are kept xed.
In a rst set of revised models (labeled A
1
and A
2
) the properties of the soil at depths below
7 metres from the foundation (12 metres from the free surface) are kept the same as in the
predictive Models A, B and C.
To determine the shear-wave velocity in the top layer we consider rst the variation of
the system frequency

[
1
= c
1
}2 with the average shear-wave velocity over the top 7 metres.
Figure 7(a) shows the system frequencies of 4.82, 5.10 and 5.72Hz calculated for blind Models
A, C and B for the initial model of the structure ([
1
=12.03 Hz) plotted versus the average
250 300 350 400
3
4
5
6
Average Shear-Wave Velocity
Top 7 m (m/s)
S
y
s
t
e
m

F
r
e
q
u
e
n
c
y

(
H
z
)
2 3 4 5
100
200
300
400
500
Average Soil Damping Ratio
Top 7 m (%)
P
e
a
k

A
m
p
l
i
t
u
d
e

a
t

T
o
p

(

m

/

t
o
n

f

)
2 3 4 5
0
50
100
150
Top 7 m (%)
P
e
a
k

A
m
p
l
i
t
u
d
e

a
t

T
o
p

(

m

/

t
o
n

f

)
A
1
E
1
A
2
E
2
D
1
D
2
(obs.)
(obs.)
FORCE AT BASE
f
1
= 9.37 Hz

1
= 3.6% A
1
E
1
A
2
E
2
D
1
D
2
(obs.)
(obs.)
FORCE AT TOP
f
1
= 9.37 Hz

1
= 3.6%
A
C
B
A
1
A
2
C
E
1
E
2
D
2
D
1
(obs.)
(obs.)
f
1
=12.05Hz
f
1
=9.37Hz
f
1
=9.37Hz
(a) (b) (c)
Figure 7. (a) Variation of calculated horizontal}rocking system frequency

[
1
with average
shear-wave velocity in the top 7 metres of soil; and (b,c) variation of calculated peak hori-
zontal motion at the top of the structure versus soil damping ratio over the top 7metres of soil
for horizontal excitation at the top (b) and base (c) of the model.
TLFeBOOK
Table IV. Characteristics of soil Models A, E and M.
Shear-wave velocity
S1

SB

S1
=
SB
(m}sec) % % %
Layer Thickness
# (m) A
1
E
1
M
1
A
2
E
2
M
2
A}E A}E M
1 7 275 282 278 317 328 326 4.2}3.8 3.3}3.0 3.5
2 6 476 375 359 476 375 424 2.0 2.0 2.0
3 9 476 425 359 476 425 424 2.0 2.0 2.0
4 19 476 476 476 476 476 476 2.0 2.0 2.0
5 70 520 520 520 520 520 520 2.0 2.0 2.0
6 50 626 626 626 626 626 626 2.0 2.0 2.0
7 888 888 888 888 888 888 2.0 2.0 2.0
(v = 0.47, j = 2420 kg}m
3
,
P
= 0.1
S
for all layers).
shear-wave velocities over the 7 metres of soil beneath the foundation for each of the models
(317, 337 and 382 m}sec, respectively). It is apparent that there is an approximately linear
relation between the system frequency and average shear-wave velocity. When the revised
structural model with an identied xed-base fundamental horizontal}rocking frequency of
9.37Hz is used, soil Model A (

[
S
=317m}sec) leads exactly to the observed system frequency
in direction D
2
. By extrapolation it is found that to match the average of the two observed
frequencies in direction D
1
(4.14 and 4.20 Hz) it is necessary to reduce the average shear-
wave velocity

[
S
in the top 7metres below the foundation from 317m}sec to about 275m}sec.
The resulting proles of shear-wave velocities that match the observed system frequencies
in directions D
1
and D
2
are designated Models A
1
and A
2
, respectively, and are listed in
Table IV.
To determine the material damping ratio

S
in the top 7 metres of soil below the founda-
tion we turn our attention to the peak amplitudes of the horizontal response at the top of
the structure. Figures 7(b) (force at top) and (c) (force at base) show the calculated peak
amplitude of the horizontal response at the top of the containment building as a function
of the material damping ratio

S
for S-waves over the top 7 metres of soil. These results
were calculated on the basis of the revised structural model with a fundamental xed-base
frequency of [
1
=9.37 Hz and xed-base damping ratio of
1
=3.6%.
The results in Figure 7(b) indicate that soil damping ratios of 4.3 and 4.1 per cent would
lead to the observed peak amplitudes in directions D
1
and D
2
for excitation at the top. For
excitation at the base (Figure 7(c)) a soil damping ratio of 3.3 per cent leads to the ob-
served peak amplitudes in both directions D
1
and D
2
. Since the actual observed amplitudes
were clipped (Figure 3) we consider it suciently accurate to use soil damping ratios of 4.2
and 3.3 per cent for excitation at the base and top of the structure, respectively, indepen-
dently of the direction of excitation. The resulting revised proles of damping ratios are listed
in Table IV.
Next, we turn our attention to the response to vertical excitation. Figure 8(a) shows the
calculated frequency at which the peak response at the top of the structure occurs versus the
average shear-wave velocity

[
S
over the top 7 metres of soil. The solid line in Figure 8(a)
corresponds to the blind predictions for a structure characterized by [
1
=34.86 Hz and for
soils which below 7 metres have the same properties as in Models A, B and C. The line
TLFeBOOK
250 300 350 400
9
10
11
12
13
Average Shear-Wave Velocity
Top 7 m (m/s)
V
e
r
t
i
c
a
l

S
y
s
t
e
m

F
r
e
q
u
e
n
c
y

(
H
z
)
2 3 4 5
1.0
1.5
2.0
2.5
3.0
Top 7 m (%)
P
e
a
k

A
m
p
l
i
t
u
d
e

a
t

T
o
p

(
m

/

t
o
n

f

)
A
C
B
A
1
A
2
E
1
E
2
(obs.)
f
1
=27.15Hz
f
1
=34.86Hz
A
1
E
2
(obs.)
(a) (b)
Figure 8. (a) Variation of calculated frequency at peak vertical displacement versus average
shear-wave velocity in the top 7metres of soil; and (b) variation of calculated peak vertical
motion at the top of the structure versus soil damping ratio over the top 7 metres of soil.
Vertical excitation at the base of the model.
A
1
A
2
represents the corresponding results when the revised models for the structure with
[
1
=27.15 Hz are used. Since the structure is very sti in the vertical direction, the system
frequency is only slightly aected by the changes in the superstructure. The observations in
Figure 3(e) indicate that for vertical vibrations the peak frequency is not clearly dened but
lies in the range from 9.5 Hz (base) to 11 Hz (top). To match this system frequency it would
be necessary to consider a soil model with a top-layer shear-wave velocity

[
S
somewhat
smaller than 275 m}sec. This result would be in contradiction with the requirements for the
horizontal response which lead to velocities of 275m}sec (D
1
) and 317m}sec (D
2
). A second
diculty has to do with the calculated amplitude at the top of the containment as shown
in Figure 8(b). In this gure, the calculated peak vertical amplitude for Model A
1
is shown
versus the damping ratio

S
in the top layer of the soil model. It is apparent from Figure 8(b)
that the calculated response from Model A
1
is signicantly larger (2.5 vs. 1.42 m}tonf ) than
the observed response even if a large soil damping is used.
Clearly, soil models characterized by a marked discontinuity at a depth of 7 m below
the foundation with velocities increasing from 275317 m}sec to 476 m}sec at that depth
do not provide sucient radiation damping to properly represent the vertical response. We
consider next a second family of models (E-models) based on the shear-wave velocities
estimated from penetration tests in which the second 34 metre thick layer of Model A with a
velocity of 476m}sec is replaced by three sub-layers with velocities of 375, 425 and 476m}sec
and thicknesses of 6, 9 and 19m, respectively. For this family of models we repeat the process
of determining the shear-wave velocity

[
S
and the damping

S
in the top layer by matching the
observed system frequency and the peak response at the top (Figures 7(a)(c)). The resulting
models in directions D
1
and D
2
for horizontal excitation at the top and base of the structure
TLFeBOOK
are designated by E
1
and E
2
and are also listed in Table IV. One of the advantages of the
E-models is that the calculated vertical response for Model E
2
also matches the observed peak
amplitude and the range of frequencies at which the peak vertical response occurs.
It should be noted that the best soil model found in this study (Model E) is similar to the
models inferred by de Barros and Luco [5] by inversion of the impedance data. In particular,
model M of de Barros and Luco [5] which resulted from identications using as unknowns
the shear-wave velocities in the top two layers of 7 and 15m thicknesses resulted in velocities
in the top layer similar to those in Model E (Table IV). Model M also shows that there is
a transition from the velocities in the top layer to the velocity of 476 m}sec at a depth of
22metres. The present study conrms the 1518% dierence between the shear-wave velocities
in directions D
1
and D
2
in the top layer of soil below the foundation. Since similar results
have been found [1316] from analyses of earthquake ground motions recorded at free-eld
downhole arrays located away from the foundation it must be concluded that soil anisotropy
must be involved.
COMPARISON OF THEORETICAL RESULTS FOR REVISED
MODELS WITH OBSERVATIONS
Some typical comparisons of the observed response with the results calculated by the CLASSI
approach are illustrated in Figures 9 to 11. Comparisons for the peak amplitudes are presented
in Table V. For horizontal excitation (Figures 9 and 10) there is very good agreement
between the calculations for the revised A and E models, and the observations. The error in
the theoretical response at the top of the structure ranges from 1.2% to 1.8% for Model A and
2.7% to 3.4% for Model E depending on location of the shaker and direction of excitation.
The largest dierences can be seen in the horizontal response at the base of the structure
where the observed peak response is somewhat larger than the calculated amplitudes. The
error in this case ranges from 14.5% to 7.0% for Model A and 13.9% to 4.5% for
Model E.
For vertical excitation, the theoretical peak amplitude based on Model A
1
is 71% larger
than the observed value. However, the revised Model E
2
with

[
S
=328 m}sec and

S
=3.0%
gives a much better agreement with the observations as shown in Figure 11 and Table V. In
this case, the calculated peak response is only 4.2% lower than the observed response. The
vertical response then serves as a discriminant between the two sets of models.
The comparisons in Figures 9 to 11 and in Table V as well as additional comparisons for
the response in direction D
2
[10] indicate that the mathematical model of the complete soil
structure interaction problem used in this study accurately accounts for most of the observed
response.
CONCLUSIONS
The conclusions of the study with respect to the identication of structural and soil properties
at the Hualien site are listed below.
TLFeBOOK
2 4 6 8 10
0
50
100
150
200
250
300
350
U
T
2 4 6 8 10
-60
0
60
120
180
240
300
2 4 6 8 10
0
50
100
150
200
H
b
2 4 6 8 10
-60
0
60
120
180
240
300
2 4 6 8 10
0
20
40
60
80
Frequency (Hz)
U
b
2 4 6 8 10
-60
0
60
120
180
240
300
Frequency (Hz)
Amplitude (m/ton f) Phase (degrees)
(a) (b)
(c) (d)
(e)
(f)
Model A
Model E
Figure 9. Comparison of observations (open circles) and calculations of the response for horizontal
excitation at the top in direction D
1
. Revised Models A
1
and E
1
.
Structural identication
(i) The rst result of the structural identication studies is that the superstructure is signif-
icantly softer than rst thought. The eective Youngs modulus required to match the
identied average horizontal (9.37 Hz) and vertical (27.35 Hz) xed-base frequencies
for the superstructure is of the order of 2.084 10
10
N}m
2
which is 26% lower than
the suggested value of 2.82 10
10
N}m
2
. This low eective value for E
c
may reect
the actual properties of the concrete in the shell or it may reect other weaknesses of
the structure as built.
(ii) There is some indication that the containment shell may be slightly softer in direction
D
1
([
1
9.24 Hz) than in direction D
2
([
1
9.49 Hz) but this result is not very strong
TLFeBOOK
Table V. Comparison of observations with calculations for revised models.
Direction D
2
Direction D
1
Theoretical Theoretical
Observed A
2
E
2
Observed A
1
E
1
(a) Horizontal excitation at the top
Frequency (Hz) at:
- Peak amplitude 4.60 4.60 4.60 4.10 4.16 4.15
- Phase [=90
4.52 4.61 4.61 4.14 4.16 4.16

Peak amplitude
(m}tonf )
U
T
235.8 232.0 231.6 283.0 288.1 292.7
H

T
167.3 170.9 170.2 201.5 211.6 214.3
H

b
126.2 126.2 125.5 156.5 166.2 168.4
U
b
56.7 48.5 48.8 68.5 63.7 65.4
(b) Horizontal excitation at the base
Frequency (Hz) at:
- Peak amplitude 4.60 4.60 4.60 4.20 4.16 4.15
- Phase [=90
4.61 4.60 4.60 4.20 4.16 4.15

Peak amplitude
(m}tonf )
U
T
61.3 60.58 59.62 81.57 80.95 81.23
H

T
43.46 44.42 43.56 55.97 59.22 59.27
H

b
32.96 32.85 32.17 40.31 46.57 46.57
U
b
14.61 12.85 12.79 20.46 18.09 18.31
(c) Vertical excitation at the base
Theoretical
Observation A
1
E
2
Frequency (Hz) at:
- Peak amplitude 11.00 10.78 10.01
- Phase [ = 90
10.99 11.08 11.44

Peak amplitude
(m}tonf )
U
T
1.42 2.43 1.36
U
b
1.18 2.03 1.18
and we prefer to use the average value of [
1
=9.37 Hz for the horizontal xed-base
natural frequency of the shell in both directions. The small dierence between the xed-
base frequencies [
1
inferred in directions D
1
and D
2
is not sucient to explain the 12%
dierence in observed values for the system frequencies

[
1
in these directions.
(iii) There is some indication that the containment shell may be slightly softer during the
stronger vibrations generated by the shaker at the top of the structure ([
1
9.19Hz) than
for the weaker vibrations induced by the shaker at the base ([
1
9.54 Hz). Again this
TLFeBOOK
2 4 6 8 10
0
15
30
45
60
75
90
U
T
2 4 6 8 10
-60
0
60
120
180
240
300
2 4 6 8 10
0
10
20
30
40
50
H
b
2 4 6 8 10
-60
0
60
120
180
240
300
2 4 6 8 10
0
5
10
15
20
25
Frequency (Hz)
U
b
2 4 6 8 10
-60
0
60
120
180
240
300
Frequency (Hz)
(a) (b)
(c) (d)
(e) (f)
Model A
Model E
Figure 10. Comparison of observations (open circles) and calculations of the response for horizontal
excitation at the base in direction D
1
. Revised Models A
1
and E
1
.
result may not be fully reliable and we prefer to use the average value of ([
1
9.37Hz)
for excitation at either location.
(iv) The identied values for the xed-base structural damping ratio
1
are not very reliable
due to lack of accuracy in the phase of the observed response with respect to the
forcing function. This is particularly true for estimates of
1
for excitation at the top
of the structure and less so for estimates of
1
for excitation at the base which depend
only on the relative phase between dierent response components. The estimates of
1
for horizontal excitation at the base and top of the structure correspond to 0.61% and
6.48% with an average value of 3.6%. It is unlikely that an increase in internal damping
by a factor of 10 could be associated with an increase in response by a factor of only
3.5. The estimated xed-base damping for vertical vibrations is 3.1%.
TLFeBOOK
2 4 6 8 10 12 14 16 18 20
0
1
2
3
U
T
2 4 6 8 10 12 14 16 18 20
-60
0
60
120
180
240
300
2 4 6 8 10 12 14 16 18 20
0
1
2
3
Frequency (Hz)
U
b
2 4 6 8 10 12 14 16 18 20
-60
0
60
120
180
240
300
Frequency (Hz)
(a) (b)
(c) (d)
Model A
Model E
Figure 11. Comparison of observations (open circles) and calculations of the vertical response for
vertical excitation at the base. Revised Models A
1
and E
2
.
(v) The structural identication techniques used in this study permitted us to determine with
reasonable accuracy some additional xed-base modal parameters such as modal mass
M
1
and participation factors [
1
and
1
even in the presence of strong soilstructure
interaction eects. The identied values diered from the calculated values based on
the initial model of the structure by 3.2%, 7.3% and 1.4%, respectively.
Identication of soil properties
(i) Identication of the properties of the layer of soil (7 m thick) immediately below the
foundation resulted in two families of soil models (A and E) with variances in directions
D
1
and D
2
and for excitation at the top and base of the structure. The soil models in
class A are slightly softer in the top layer [317m}sec (D
2
) and 275 m}sec (D
1
) vs. 328
and 282 m}sec for E models], stier in the second layer (476 m}sec vs. 375 m}sec) and
involve larger soil damping ratios in the top layer (4.2 and 3.3% vs. 3.8 and 3.0%). Both
sets of models lead to reasonably accurate results for the horizontal}rocking response
of the structure but only the second set of models (E) which involve a larger amount
of radiation damping give reasonable results for the vertical response.
(ii) The results of the soil identication on the basis of the structural response seem to
indicate that the material damping in the soil would be larger (3.8 to 4.2%) during the
stronger vibrations induced by a force at the top than during vibrations by a force at
the base of the structure (3.0 to 3.3%).
TLFeBOOK
(iii) The results obtained conrm previous studies suggesting soil anisotropy. Dierences of
1518% are found in the shear-wave velocities of the layer below the foundation for
excitations in directions D
1
and D
2
.
(iv) A simple method of soil identication based on matching the observed system frequency
and the observed peak response at the top of the structure has led to results for the
upper layer of soil fully consistent with more complex identication methods.
Comparisons of observations with the results for revised models
(i) As expected, the theoretical results based on the revised models of the superstructure
and the soil compare favourably with the observed response. This suggests that the
current methods to calculate the eects of soilstructure interaction can account for
most of the observed phenomena. All of the changes in the properties of the structure
and the soil required to arrive at the revised models are within the accuracy of the
basic data.
(ii) It has been shown that slightly dierent models for the soil (A and E) may lead to
similar results for the horizontal}rocking response but that the use of the response to
vertical excitation may permit discrimination between the two soil models.
ACKNOWLEDGEMENTS
The work described here was supported by Grant BCS-9315680 from the National Science Foundation
with funding from the U.S. Nuclear Regulatory Commission. The support of CNEN (Brazilian Nuclear
Regulatory Commission) is also acknowledged. Some of the results of this work were rst presented
in report form in 1994.
REFERENCES
1. Tang H-T, Stepp JC, Cheng YH, Yeh YS, Nishi K, Iwatate T, Kokusho T, Morishita H, Shirasaka Y, Gatenbein
F, Touret JP, Sollogonb P, Graves H, Costello J. The Hualien large-scale seismic test for soilstructure interaction
research. Transactions of the 11th International Conference on Structural Mechanics in Reactor Technology,
vol. K, 1991; 6974.
2. Graves HL, Tang H-T, Liao YC. Large-scale seismic test program at Hualien, Taiwan. Nuclear Engineering
and Design 1993; 163:323332.
3. Kokusho T, Nishi K, Okamoto T, Kataoka T, Tanaka Y, Kudo K, Tang HT, Cheng YH. Geotechnical
investigation in the Hualien large scale seismic test project. 12-SMIRT, K1, Stuttgart, Germany, August 1993;
8590.
4. Tajimi Engineering Services. Post-test correlation analysis of the forced vibration test before backll. Hualien
LSST Project, 24 Nov. 1993, 68 pp.
5. de Barros FCP, Luco JE. Identication of impedance functions and soil properties from vibration tests of Hualien
containment model. Soil Dynamics and Earthquake Engineering 1995; 14(4): 229248.
6. Tokyo Electric Power Company. Status report of the forced vibration tests (before backll and after backll).
Hualien LSST Project, Nov. 1993, 62 pp.
7. Morishita H, Tanaka H, Nakamura N, Kobayashi T, Kan S, Yamaya H, Tang HT. Forced vibration test of
Hualien large scale SSI model. 12-SMIRT, K02}1, Stuttgart, Germany, August 1993; 1520.
8. Luco JE, Wong HL, Trifunac MD. Soilstructure interaction eects on forced vibration tests. Report 86-05,
Department of Civil Engineering, University of Southern California, Los Angeles, California, 1986.
9. Luco JE, Trifunac MD, Wong HL. Isolation of soilstructure interaction eects by full-scale forced vibration
tests. Earthquake Engineering and Structural Dynamics 1988; 16:121.
10. Luco JE, de Barros FCP. Identication of structural and soil properties from forced vibration tests of the Hualien
containment model prior to backll. Report, Department of Applied Mechanics and Engineering Sciences,
University of California, San Diego, California, June 1994, 114 pp.
11. Kobayashi T, Kan S, Yamaya H, Kitamura E. Seismic identication of the Hualien LSST model structure.
Earthquake Engineering and Structural Dynamics 1997; 26:11571167.
TLFeBOOK
12. Yun C-B, Choi J-S, Kim J-M. Identication of the Hualien soilstructure system. Soil Dynamics and Earthquake
Engineering 1999; 18:395408.
13. Ueshima T, Okano H. Further investigation on seismic response of soil and embedded structure in Hualien
LSST program. 11th World Conference on Earthquake Engineering, Acapulco, Mexico, 1996, Paper No. 1930,
Elsevier Science.
14. Gunturi VR, Elgamal A-W, Tang HT. Hualien seismic downhole data analysis. Engineering Geology 1998; 50:
929.
15. Chen C-H, Chiu H-C. Anisotropic seismic ground responses identied from the Hualien vertical array. Soil
Dynamics and Earthquake Engineering 1998; 17(6):371395.
16. Choi J-S, Lee JS, Yun C-B. Input and system identication of the Hualien soilstructure interaction system
using earthquake response data. Earthquake Engineering and Structural Dynamics 2003; 32:19551975.
TLFeBOOK
Simplied non-linear seismic analysis of inlled
reinforced concrete frames
Matja z Dol sek
;
and Peter Fajfar
Faculty of Civil and Geodetic Engineering; University of Ljubljana; Jamova 2, SI-1000 Ljubljana; Slovenia
SUMMARY
The N2 method for simplied non-linear seismic analysis has been extended in order to make it appli-
cable to inlled reinforced concrete frames. Compared to the simple basic variant of the N2 method,
two important dierences apply. A multi-linear idealization of the pushover curve, which takes into ac-
count the strength degradation which occurs after the inll fails, has to be made, and specic reduction
factors, developed in a companion paper, have to be used for the determination of inelastic spectra.
It is shown that the N2 method can also be used for the determination of approximate summarized
IDA curves. The proposed method was applied to two test buildings. The results were compared with
the results obtained by non-linear dynamic analyses for three sets of ground motions, and a reasonable
accuracy was demonstrated. A similar extension of the N2 method can be made to any structural sys-
tem, provided that an appropriate specic RT relation is available. Copyright ? 2004 John Wiley
& Sons, Ltd.
KEY WORDS: simplied non-linear analysis; inelastic spectrum; inlled RC frame; N2 method; incre-
mental dynamic analysis
1. INTRODUCTION
In recent years a breakthrough of simplied inelastic procedures for seismic design and assess-
ment has been observed. These procedures combine the non-linear static (pushover) analysis
of a relatively simple mathematical model and the response spectrum approach. One such
procedure is the N2 method [1], which has been implemented in the nal draft of the
Eurocode 8 standard [2].
In this paper the application of the N2 method has been extended to inlled reinforced
concrete (RC) frames. Inlled frames are typically characterized by a substantial strength
Correspondence to: Matja z Dol sek, Faculty of Civil and Geodetic Engineering, University of Ljubljana, Jamova 2,
SI-1000 Ljubljana, Slovenia.
Contract=grant sponsor: Ministry for Education, Science and Sport of the Republic of Slovenia
Contract=grant sponsor: European Commission, SPEAR Project; contract=grant number: G6RD-CT-2001-00525
Received 23 June 2003
Revised 2 July 2004
Copyright
?
2004 John Wiley & Sons, Ltd. Accepted 2 July 2004
TLFeBOOK
50 M. DOL
SEK AND P. FAJFAR

degradation after the inll fails. A specic RT relation, which takes into account this
characteristic, was developed by Dol sek [3] and has been slightly modied in the companion
paper [4]. By using this relation, which requires a multi-linear idealization of the pushover
curve, the N2 method can be applied to inlled frames. A similar approach can be used for
any structural system, provided that an appropriate RT relation is available.
A lot of information on seismic response can be obtained from the IDA (incremental
dynamic analysis) curve introduced by Vamvatsikos and Cornell [5]. However, the compu-
tational procedure for the determination of the IDA curve is time-consuming. The procedure
can be greatly simplied if the N2 method is used for the determination of an approximate
summarized IDA curve. The simplied procedure will be called the IN2 method.
The extended N2 method was applied to two test buildings. The results were compared
with the results obtained by non-linear dynamic analyses for three sets of ground motions.
2. SUMMARY OF THE ANALYSIS METHODS
2.1. The N2 method
The N2 method (N stands for non-linear analysis, and 2 for two mathematical models) is a
relatively simple non-linear method for the seismic analysis of structures. It combines push-
over analysis of a multi degree-of-freedom (MDOF) model with the response spectrum anal-
ysis of an equivalent single-degree-of-freedom model (SDOF). The N2 method has been
described in detail in several publications. Details about the basic version of the N2 method,
limited to planar structural models, can be found in References [1, 6]. Here the planar version
of the method will be briey summarized.
Determination of capacity. A non-linear static (pushover) analysis is performed by subjecting
a MDOF structural model to a monotonically increasing pattern of lateral forces. As a result,
a non-linear relationship between base shear and roof (top) displacement (the pushover curve)
is determined. The lateral load distribution is related to the assumed displacement shape
by
=M (1)
where M is the mass matrix. There are no xed rules for the choice of the displacement
shape. In regulatory documents some guidelines may be given.
The forcedisplacement relationship for the equivalent SDOF system has to be idealized
and the yield point has to be determined. In the majority of cases a bilinear (elasto-plastic)
idealization is appropriate. In some cases, for example in the case of inlled frames, a multi-
linear idealization is necessary. For idealization, engineering judgement has to be used. Again,
some guidelines may be given in regulatory documents.
Next, an equivalent SDOF model has to be determined. The forces and displacements
corresponding to the equivalent SDOF model are obtained by dividing the base shear and
top displacement of the MDOF system with a transformation factor , which is a function
of the displacement shape (
n
=1, n denotes the roof level) and the mass distribution,
TLFeBOOK
ANALYSIS OF INFILLED RC FRAMES 51
dened as
=

T
M1
T
M
=
m
(2)
The elastic period T of the idealized system can be determined from the relation
T =2
D
y
F
y
(3)
where F
y
and D
y
are the yield strength and displacement, respectively.
The capacity diagram in AD (accelerationdisplacement) format is obtained by dividing the
forces in the forcedeformation diagram by the equivalent mass m
.
Determination of seismic demand. Seismic demand is dened by the spectrum formulated
in the AD format. In principle, any spectrum can be used. However, the most convenient is
a spectrum of the NewmarkHall type, i.e. a spectrum with constant acceleration, constant
velocity and constant displacement regions.
Inelastic spectra for constant ductility can be determined if R, the reduction factor due
to ductility, i.e. due to the hysteretic energy dissipation of ductile structures, is known as a
function of the ductility and period T. The inelastic spectral acceleration S
a
and displacement
S
d
are dened as
S
a
=
S
ae
R
(4)
S
d
=

R
S
de
=

R
T
2
4
2
S
ae
=
T
2
4
2
S
a
(5)
where S
ae
and S
de
are the values in the elastic acceleration and displacement spectrum, respec-
tively, corresponding to the period T and a xed viscous damping ratio, and is the ductility
factor dened as the ratio between the maximum displacement and the yield displacement.
In principle, any RT relation can be taken into account. However, the procedure
becomes very simple and transparent if a simple but fairly accurate relation is employed, i.e.
R =( 1)
T
T
C
+ 1 TT
C
(6)
R = TT
C
(7)
where T
C
is the characteristic period of the ground motion. It is typically dened as the transi-
tion period where the constant acceleration segment of the response spectrum (the short-period
range) passes to the constant velocity segment of the spectrum (the medium-period range).
Equations (5) and (7) suggest that, in the medium- and long-period ranges, the equal displace-
ment rule applies, i.e. the displacement of the inelastic system is equal to the displacement
of the corresponding elastic system with the same period.
For inlled frames, Equations (6) and (7) are not appropriate. A specic RT relation
has been proposed in the companion paper [4].
TLFeBOOK
52 M. DOL
SEK AND P. FAJFAR

Starting from the elastic design spectrum, and using Equations (4) to (7), the demand
spectra for the constant ductility factors in AD format can be obtained. They represent the
inelastic demand spectra.
The seismic demand for the equivalent SDOF system with a period T can be determined
as follows. Elastic demand in terms of the acceleration S
ae
and the displacement S
de
is deter-
mined from the elastic spectrum. The inelastic acceleration demand S
a
is equal to the yield
acceleration S
ay
, which represents the acceleration capacity of the inelastic system. The reduc-
tion factor R can be determined as the ratio between the accelerations corresponding to the
elastic and inelastic systems (Equation (4)). Mean ductility demand is determined from the
RT relation and the inelastic displacement demand S
d
is computed from Equation (5).
The maximum top displacement D
t
of the MDOF system (the target displacement) is ob-
tained by multiplying the displacement demand for the SDOF model S
d
with the transforma-
tion factor . Pushover analysis is employed once more and the structure is pushed to D
t
.
Seismic demands for all relevant local quantities are obtained by assuming that the distribu-
tion of deformations throughout the structure in the static (pushover) analysis approximately
corresponds to that which would be obtained in the dynamic analyses.
Performance evaluation (Damage Analysis). Expected performance can be assessed by com-
paring the seismic demands with the capacities for the relevant performance level. The deter-
mination of seismic capacity is not discussed here.
2.2. Incremental dynamic analysis
The fundamentals of incremental dynamic analysis (IDA) were dened by Vamvatsikos and
Cornell [5]. This is a parametric analysis method, in which a structural model is subjected
to multiple levels of seismic intensity using one or more ground motion records. The goal
of an IDA study is the understanding of structural behaviour at dierent levels of seismic
intensity. IDA is also a part of a probabilistic framework for seismic performance assessment
developed by Cornell and co-workers at Stanford [7, 8].
In spite of many problems related to the input data on ground motions and mathematical
modeling, there is no doubt that IDA analysis provides a very thorough image of the seismic
behaviour of a structure. However, it is very time-consuming, and the question arises as to
whether it is possible to create summarized IDA curves with less input data, with less eort,
but still with acceptable accuracy. One possible approach is to determine seismic demand
for multiple levels of seismic intensity using the N2 method. Such an analysis, called the
Incremental N2 (IN2) method, has been employed in our study. In the paper a comparison
between IDA analysis and the IN2 method has been performed for RC inlled frames.
3. THE EXTENDED N2 METHOD
In order to apply the N2 method to inlled RC frames, two modications of the basic (simple)
version of the method described in Reference [1] need to be made. First, the pushover curve
has to be idealized as a multi-linear forcedisplacement relation rather than a simple bilinear
elasto-plastic one. Secondly, inelastic spectra have to be determined by using specic reduction
factors (i.e. the RT relation), appropriate for inlled frames.
TLFeBOOK
Top displacement
Pushover curve
Idealised
F
e
F
min
E
h, Fmax
D
e
B
a
s
e

s
h
e
a
r
Fmax=Fy
D
Fmax
D
Fmin
D
y
D
s
E
h,Fmin
Figure 1. Idealization of forcedisplacement relationship.
Idealization of the pushover curve. The basic characteristic of the pushover curve of inlled
frames is a substantial decrease in strength after the inll has begun to degrade. This fea-
ture has to be taken into account in analyses. Consequently, a multi-linear idealization of the
pushover curve is required (Figure 1), as explained in the companion paper [4]. The ideal-
ization procedure presented in this sub-section and applied in test examples is one possibility
which worked quite well in our analyses. Note, however, that any other idealization is pos-
sible, provided that the three characteristic points (yielding, the initiation of degradation, and
complete failure of the inll) are properly dened. Idealization of the pushover curve can be
performed as follows.
First, two points are dened on the pushover curve (Figure 1). They correspond to the
maximum (D
F max
; F
max
) and minimum strength (in the degrading branch) of the structure
(D
F min
; F
min
). The latter point is usually related to the complete failure of the inll
in one or more stories. After that, only the frame resists the horizontal loading (see
Figure 1).
Secondly, the characteristic points of the idealized pushover curve are determined. The
yield force F
y
of the idealized system is assumed to be equal to the maximum strength
of the structure F
max
. Displacements at yield D
y; E
and at the start of the degradation D
s
are determined by the equal energy rule, which is applied to the idealized system and
pushover curve separately for intervals from 0 to D
F max
and D
F max
to D
F min
D
y; E
= 2
D
F max

E
h; F max
F
max
(8)
D
s
=
2
F
max
F
min
(E
h; F max
E
h; F min
+F
max
D
F max
0:5D
F min
(F
max
+F
min
)) (9)
where E
h; F max
and E
h; F min
represent the area under the pushover curve in the intervals
from 0 to D
F max
and from 0 to D
F min
, respectively. The third characteristic point is
(D
F min
; F
min
).
TLFeBOOK
54 M. DOL
SEK AND P. FAJFAR

It has been observed [3] that the above idealization may underestimate displacements of
systems with the fundamental period in the short-period range. In order to improve the
accuracy, Dol sek [3] has proposed a procedure for the reduction of the eective stiness
of the idealized system. This approach is promising, but needs further investigation and has
not been applied in this paper.
RT relation for inlled frames. The relation has been proposed by Dol sek and Fajfar [4].
It should be used in the form which allows the mean ductility demand to be computed
from the known reduction factor R (Equation (2) in Reference [4]). Note that in the same
way as shown in this paper for inlled frames, any other RT relation corresponding to
any structural system can be used in the N2 method.
Improvement of accuracy in the case of low seismic demand. The N2 method is not intended
to be used for structures which remain in the (equivalent) elastic region. However, if the N2
method is employed for the determination of an approximate summarized IDA curve, one may
wish to compute realistic displacement demand also for acceleration demand S
ae
(T), which is
lower than the yield acceleration S
ay
of the idealized pushover curve, i.e. for R1 . This can
be done by approximating the rst part of the pushover curve by a bilinear curve rather than a
linear one. The two parts of the bilinear curve are separated by the point (D
e
; F
e
) (Figure 1),
which represents the boundary of the initial ideal elastic behaviour, and is arbitrarily dened
as the crossing point of the radial line with a slope equal to 95% of the initial stiness of
the structure with the computed pushover curve. The relation between ductility (normalized
displacement) and reduction factor (normalized force) R for R1 is dened as
=
e
R
r
e
Rr
e
R r
e
1 r
e
(1
e
) +
e
r
e
R1
(10)
where
e
=D
e
=D
y
and r
e
=F
e
=F
y
.
The improvement of accuracy, presented above, is applicable to any structural system which
is characterized by a pushover curve that substantially deviates from a line in the equivalent
elastic region. Note that the procedure inuences results only in the case when R1.
4. TEST STRUCTURES AND GROUND MOTIONS
With the intention of demonstrating and evaluating the proposed simplied seismic analy-
sis of RC inlled frames, the seismic responses of two test structures have been studied
(Figure 2).
The rst example is a four-story existing building, for which the frame had been designed
to reproduce the design practice in European and Mediterranean countries about forty to fty
years ago [9]. However, it may also be typical of buildings built more recently, but with-
out the application of capacity design principles (especially the strong columnweak beam
concept), and without up-to-date detailing. In such buildings a soft rst story eect may ap-
pear even though the structure is uniformly inlled in its elevation [10]. The design base shear
TLFeBOOK
4
.
0
m
4.0m 6.0m 2.5 5.0m 5.0m
2
.
7
2
.
7
2
.
7
2
.
7
3
.
0
m
3
.
0
m
3
.
5
m
3
.
0
m
L
o
a
d
i
n
g
5
.
0
m
5
.
0
m
Four-story
contemporary building
Four-story
existing building
Figure 2. Test structures.
coecient amounted to 0.08. In the design, concrete of quality C16/20 (according to
Eurocode 8) and smooth steel bars of class Fe B22k (according to Italian standards) were
adopted.
The four-story contemporary building was designed according to Eurocode 8, as a high-
ductility class structure [11]. The design peak ground acceleration amounted to 0:3 g, which
results in a base shear coecient equal to 0.15. The specied steel quality was B500, which
does not fulll the Eurocode 8 requirements, and the concrete quality was C25/30. For the
inll, UNIBRICK 11:2 cm thick clay bricks, with vertical perforations, were used.
For both test structures full-scale pseudo-dynamic tests have been performed at the ELSA
Laboratory, Ispra [9, 11]. The technique proposed by the authors [12], which employs the
results of pseudo-dynamic test, has been used to construct the mathematical model of the
inlled frame of the contemporary structure. The inll in the existing structure, tested in
Ispra, contained many openings, which were not taken into account in our mathematical
model. Only the bare frame was modeled by the technique proposed in Reference [12], while
the model for inll was similar to the model used in the contemporary building with reduced
stiness and strength. All the beams and columns were modeled by perfectly elastic, massless
beam elements with two non-linear rotational springs at each of the two ends. The moment
rotation relationship for each spring was dened by a trilinear envelope and Takedas hysteretic
rules. Asymmetric backbone curves were used for the beams. In addition to these elements,
simple rotational connection elements were placed between the beams and joints to model
the pinching behaviour of the beams. The inll panels were modeled by equivalent diagonal
TLFeBOOK
56 M. DOL
SEK AND P. FAJFAR

struts, which carry loads only in compression. The characteristics of diagonal struts of the
contemporary building were determined with the technique proposed by the authors [12].
The same values of the initial stiness and the maximum strength can be obtained from the
formulas used in Reference [12] if the elastic modulus of 8:0GPa and the cracking strength of
0:6MPa is assumed. The displacement at the maximum strength amounts from 0.2% to 0.25%
of the length of the diagonals and the cracking strength amounts to about 40% of the maximum
strength. The diagonal struts for the existing building were determined by arbitrarily reducing
the elastic modulus and the cracking strength of inlls for 15% and 60% respectively. The
displacement at the maximum strength of inlls was also reduced to 0.1% of the length of the
diagonal. The shearslip hysteretic model has been used for modeling the cyclic behaviour
of the inll panels. Strength deterioration was modeled only for the elements representing
inlls, whereas for the elements representing RC beams and columns unlimited ductility was
assumed. All non-linear analyses were performed using a modied version of the computer
program DRAIN-2DX [13]. More details on the mathematical models of the test structures
can be found elsewhere [3, 12].
In all analyses 5% damping was assumed. In the case of the N2 method, damping was
reected in the elastic spectrum. In non-linear dynamic analyses, an appropriate mass-prop-
ortional damping matrix was formed, which produced 5% damping if related to the equivalent
elastic period T. Since the initial periods of the test structures are smaller than the equiv-
alent elastic periods, and since mass-proportional damping increases with an increasing period,
a smaller damping ratio (3.0% and 2.7% for the existing and contemporary building,
respectively) was used for the determination of the damping matrix, in order to approximately
obtain the eect of 5% damping.
Ground motion was represented by three sets of ground motions, which were also used in
the companion paper, where more details can be found. Sets 1 and 2 each consist of 14 semi-
articial ground motions. They were based on recorded ground motions. A special procedure
was used for the modication of records aimed at matching the target spectrum in the range
beyond the fundamental period of the structure. The target spectrum was a NewmarkHall
type spectrum with constant acceleration, constant velocity and constant displacement regions.
Set 3 consists of 20 accelerograms recorded in California and Montenegro.
In the N2 analyses, the target NewmarkHall type spectra represented seismic demand
corresponding to sets 1 and 2. For set 3, the mean spectrum was idealized in the Newmark
Hall form. The characteristics of the spectra are summarized in Table I, where T
B
, T
C
and T
D
are the corner periods at the beginning of constant acceleration regions, between the constant
acceleration and constant velocity regions, and between the constant velocity and the constant
Table I. Characteristics of the NewmarkHall type spectra
applied in the N2 analyses.
Ground
motion set T
B
(s) T
C
(s) T
D
(s)
1 0.09 0.30 1.89 2.27
2 0.14 0.59 1.76 2.61
3 0.22 0.55 1.76 2.39
TLFeBOOK
displacement region, respectively, whereas is the ratio between the spectral value in the
constant acceleration region and peak ground acceleration.
5. RESULTS OF ANALYSES
For the test buildings global seismic demand in terms of top displacement and story drifts
was calculated using the N2 method and compared with results of dynamic analysis. In this
section, the application of the N2 method is presented in detail for the four-story existing
building and the third set of ground motions. Selected results for the other two sets of ground
motions as well as for the other test building are also shown. The results are compared with
results of non-linear dynamic analyses.
5.1. The four-story existing building
A basic description of the building is provided in Section 4 and in Figure 2. The story
masses amount to 46 tons for the rst three stories, and to 40 tons for the roof, and the total
mass amounts to 178tons. An inverted triangular displacement shape was arbitrarily assumed,
which was used for the determination of the lateral force pattern used in the pushover analysis
(Equation (1)), and for the MDOF to SDOF transformation (Equation (2)). The normalized
lateral force pattern is dened as [0.288 0.575 0.862 1]. The base sheartop displacement
relationship obtained by pushover analysis is presented in Figure 3. Its characteristic points,
(D
F max
; F
max
=F
y
) and (D
F min
; F
min
), are dened by (1:84; 706) and (4:89; 331). The units are
cm and kN.
In the next step, idealization of the pushover curve is performed according to the procedure
described in Section 3. The yield displacement D
y; E
(Equation (8)) and the displacement at
the beginning of the strength degradation D
s
(Equation (9)) amount to 1:39 cm and 2:69 cm,
respectively. Consequently, the ductility at the beginning of the strength degradation
s
0 0.5 1 1.5 2 2.5
0
10
20
30
40
50
60
70
80
Top displacement / Height (%)
B
a
s
e

S
h
e
a
r

/
W
e
i
g
h
t

(
%
)
Contemporary
Existing
Plastic mechanism
Figure 3. Base sheartop displacement relationships for the test buildings.
TLFeBOOK
58 M. DOL
SEK AND P. FAJFAR

0 2 4 6 8 10
0
0.3
0.6
0.9
1.2
S
d
(cm)
S
a

(
g
)
Elastic spectrum
T
=
0
.
2
9
s
Capacity diagram
6.59 2.25
0.49
1.07
Inelastic spectrum = 6.5
Demand point
Figure 4. Elastic and inelastic demand spectra versus capacity diagram for the four-story existing build-
ing, for the third set of ground motions, and for a PGA of 0:45 g.
amounts to 1.94. The parameter r
u
, which represents the ratio between the minimum force
F
min
and the yield force is equal to 0.468.
The MDOF quantities are transformed to SDOF quantities by using the transformation factor
=1:36 (Equation (2)). The equivalent mass m
of the SDOF system (see Equation (2)) is

equal to 109 tons, and the period of the equivalent SDOF system T (Equation (3)) is equal
to 0:29 s.
The capacity diagram in AD format (Figure 4) is obtained by dividing the forces of the
equivalent SDOF system by the equivalent mass m
. The yield acceleration S

ay
amounts to
0:486 g.
Seismic demand is dened by the idealized elastic spectrum corresponding to the third set
of ground motions, normalized to a peak ground acceleration of 0:45 g. Elastic demand in
terms of acceleration and displacement (S
de
; S
ae
), calculated for the known elastic period of
the equivalent SDOF system T, amounts to (2:25 cm, 1:07 g) (Figure 4). Thus the reduction
factor R
(Equation (4)) is equal to 2.21.

The mean ductility demand is calculated by using the specic RT relation, developed
in Reference [4] for inlled frames. Equation (7) of Reference [4] yields R(
s
) equal to 1.35.
The reduction factor R
=2:21 is then larger than R(

s
) =1:35, so the parameters
0
and R
0
are equal to
s
and R(
s
) (Equations (5) and (6) in Reference [4]). Because the period T
is less than T
C
(Table I), and R
is larger than R(
s
), parameter c depends on T, T
C
and r
u
(see Equation (3) in Reference [4]) and amounts to 0.19. Finally, the mean ductility demand
is obtained from Equation (2) of Reference [4]. It amounts to 6.46.
The mean inelastic displacement of the SDOF system can be obtained by multiplying the
yield displacement D
y
by the mean ductility . It is equal to 6:59 cm (Figure 4).
TLFeBOOK
A graphical presentation of the basic quantities of the equivalent SDOF system is provided
in AD format in Figure 4. In the case of a structure modeled with an ideal elasto-plastic
capacity diagram (as in the basic N2 method), the demand point is dened by the intersec-
tion of the capacity diagram and inelastic demand spectrum for the relevant ductility. For the
inlled frame, which is characterized by the strength-degrading capacity diagram, extension
of the horizontal yield plateau should be used instead of the capacity diagram for the deter-
mination of the demand point. Namely, the reduction factor in the RT relation, used for
determining the inelastic spectrum, is dened as the ratio between the elastic and inelastic
acceleration demand (Equation (4)), where the inelastic acceleration demand is equal to the
yield acceleration.
The top displacement of the MDOF model is obtained by multiplying the displacement of
the SDOF model with the constant . It amounts to 8:95 cm, whereas the top displacement
obtained by non-linear dynamic analysis amounts to 9:5 cm.
Story drifts and all local quantities of interest are obtained by performing pushover analysis
with an assumed force pattern, up to the calculated top displacement. Maximum drift occurs
in the rst story and amounts to 7:7cm (2.85% of the story height) (Figure 5). The rst story
plastic mechanism is formed already at a story drift of 1.05%, after collapse of the inll in
the rst story. The rst story drift obtained with the N2 method slightly overestimates the
mean value obtained with the non-linear dynamic analysis. On the other hand, story drifts
in all stories above the rst story are underestimated by the N2 method. These observations
are typical for almost all examples investigated in the study. After a story plastic mechanism
is formed, pushover analysis tends to increase the deformations mostly in this story. On the
other hand, further mechanisms can form in addition to the rst one during the non-linear
time-history response. Moreover, in some rare cases, a dierent plastic mechanism (e.g. in the
second or third story) can form in the dynamic analysis for a specic accelerogram. Thus, in
the case of IDA analysis, maximum story drift sometimes occurs in the second or third story,
whereas in the case of the IN2 analysis the maximum story drift always occurs in the rst
story. Note the large dispersion of results obtained by non-linear dynamic analysis.
The seismic demand of the SDOF system for the rst and second sets of ground motion
parameters is shown in Figure 6. The top displacement of the MDOF system (11:6cm) and the
corresponding ductility ( =8:4) obtained with the second set of ground motion parameters
are a little higher then those obtained for the third set of ground motion parameters. The
small increase occurs because of higher T
C
and (Table I). A much larger dierence can
be seen in the case of the rst set of ground motions, due to the much lower T
C
. The mean
ductility demand amounts to only 3.0. Story drifts are presented in Figure 5 and compared
with results of non-linear dynamic analyses. Some characteristics of the mathematical model
and some results obtained with the N2 method and non-linear dynamic analysis are presented
in Table II.
5.2. The four-story contemporary building
A basic description of the building is provided in Section 4 and in Figure 2. The story
masses amount to 96 tons, 95 tons, and 88 tons for the rst, the second and the third, and the
fourth stories, respectively. An inverted triangular displacement distribution was employed for
the determination of the lateral force pattern for the pushover analysis. The base sheartop
displacement relationship obtained from the pushover analysis, and the idealized pushover
TLFeBOOK
60 M. DOL
SEK AND P. FAJFAR

0 0.3 0.6 0.9 1.2 1.5
0
2.7
5.4
8.1
10.8
H
e
i
g
h
t

(
m
)
Story drift (%)
0 1.2 2.4 3.6 4.8 6
0
2.7
5.4
8.1
10.8
H
e
i
g
h
t

(
m
)
Story drift (%)
0 0.6 1.2 1.8 2.4 3
0
3.5
6.5
9.5
12.5
Story drift (%)
0 2 4 6 8 10
0
2.7
5.4
8.1
10.8
Third set of ground motions
H
e
i
g
h
t

(
m
)
Story drift (%)
0 0.8 1.6 2.4 3.2 4
0
3.5
6.5
9.5
12.5
Story drift (%)
Third set of ground motions
Second set of ground motions Second set of
ground motions
First set of ground motions
Single acc. (dynamic)
Mean (dynamic)
N2
Single acc. (dynamic)
Mean (dynamic)
N2
0 0.1 0.2 0.3 0.4 0.5
0
3.5
6.5
9.5
12.5
Story drift (%)
First set of
ground motions
(a) (b)
Figure 5. Mean values of story drifts obtained using the N2 method and non-linear dynamic
analysis for the: (a) existing building; and (b) contemporary building, for a PGA equal to
0:45 g and for all three sets of ground motions. Results from non-linear dynamic analysis for
single accelerograms are also presented.
curve used in the N2 method, are shown in Figure 3. The maximum strength of the building
is equal to 2471 kN, which is equal to 67% of the total weight of the building. It is reached
concurrently with the maximum strength of inll in the rst story. The rst story plastic
mechanism is formed when the top displacement and the corresponding rst story drift amount
TLFeBOOK
0
0.3
0.6
0.9
1.2
0 2 4 6 8 10
0
0.3
0.6
0.9
1.2
I
N
2
c
u
r
v
e
Second set of ground motions Second set of ground motions
S
d
(cm)
S
a

(
g
)
S
a

(
g
)
0 2 4 6 8 10
S
d
(cm)
0 2 4 6 8 10
S
d
(cm)
0 2 4 6 8 10
S
d
(cm)
C
ap
ac
ity
diagram
Capacity diagram
C
apacity
d
iagram
E
l
a
s
t
ic
s
p
e
ctru
m
E
l
a
s
t
ic
s
p
e
c
tru
m
I
n
e
l
a
s
t
i
c
s
p
e
c
tru
m
3
=
.0
I
n
e
l
a
s
t
i
c
s
p
e
c
tr
u
m
=
1
.1 Capacity diagram
I
n
e
l
a
s
t
ic
s
p
e
c
tru
m
=
8
.4
IN2 curve
In
elastic
s.
=2.
3
I
N
2
c
u
r
v
e
I
N
2
c
u
r
v
e
Elastic spectrum
T
=
0
.
2
9
s
T
=
0
.
2
9
s
Elastic
spectrum
T
=
0
.
3
7
s
T=0.37 s
(a) (b)
Figure 6. Elastic and inelastic demand spectra versus capacity diagrams for the four-story:
(a) existing building; and (b) contemporary building, for the rst and second sets of ground
motions, and for a PGA of 0:45 g. IN2 curves are also shown.
Table II. Summary of results obtained with the N2 method for the existing building
and a comparison with the results of non-linear dynamic analysis.
MDOF, PGA=0:45 g
SDOF Ground motion set 1 2 3
F
y
=W(%) 40 D
y
(cm) 1.0 N2 4.1 11.6 9.0
D
t
(cm)
m
(ton) 109 T
(s) 0.29 Dyn. 3.0 11.4 9.5

S
ay
(g) 0.49 r
u
0.47 Max. story N2 1.0 3.8 2.8
1.36
s
1.94 drift (%) Dyn. 0.5 2.9 2.5
TLFeBOOK
62 M. DOL
SEK AND P. FAJFAR

to 8:1cm (0.65%) and 4:3cm (1.6%), respectively. At this deformation, the strength of the inll
has substantially degraded (Figure 3). A local minimum of strength is obtained when the inll
in the rst story completely fails. After that, strength increases again. A local maximum of
strength is reached concurrently with the maximum strength of inll in the second story. With
increasing deformation, the strength of the inll in the second story degrades until the inll
fails and the minimum strength is reached. (Note that an unlimited ductility of RC elements is
assumed in the mathematical model.) In the idealized pushover curve, the degrading branch
is directed toward minimum strength. Because the slope of the degrading branch does not
inuence the reduction factors used in the analysis, and because the strength at the rst local
minimum of the pushover curve is only slightly higher than the second minimum, idealization
of the degrading branch with respect to the rst local minimum would yield practically the
same results.
Seismic demand is dened by the idealized elastic spectrum corresponding to the three sets
of ground motions, and normalized to a peak ground acceleration of 0:45g. In Figure 6 elastic
and inelastic demand spectra, as well as the capacity diagram, are presented in AD format for
the rst and second sets of ground motions. In the case of the rst set of ground motions, the
structures remain practically in the elastic region. For the second set of ground motions, the
mean ductility demand amounts to 2.3, which corresponds to 8:2 cm (0.66%) in terms of
the top displacement. From Figure 3 it can be seen that the calculated top displacement is in
the degrading branch of the pushover curve, near the start of the plastic mechanism, which
is formed in the rst story at a drift of 1.7%.
In Figure 5 the story drifts obtained using the N2 method are compared with the story
drifts obtained by non-linear dynamic analyses for three sets of ground motions. Owing to
the much larger strength of the contemporary building, the displacements are smaller than in
the case of the existing building, especially the rst story drift. Otherwise, similar conclusions
as in the case of the existing building apply. Some additional data on the equivalent SDOF
system, which were used in order to obtain the seismic demand, are presented in Table III.
In the same table a comparison of the global seismic demand obtained by the N2 method and
by non-linear dynamic analysis is also presented. Good agreement can be observed.
5.3. Dispersion of results
The results presented in Sections 5.2 and 5.3 represent mean values. In the case of non-linear
dynamic analysis, the maximum coecients of variation for story drift (Figure 5) are, for the
Table III. Summary of results obtained with the N2 method for the contemporary
building and a comparison with the results of non-linear dynamic analysis.
MDOF, PGA=0:45 g
SDOF Ground motion set 1 2 3
F
y
=W(%) 67 D
y
(cm) 2.7 N2 3.9 8.2 6.4
D
t
(cm)
m
(ton) 236 T
(s) 0.38 Dyn. 3.3 8.0 7.2

S
ay
(g) 0.79 r
u
0.61 Max. story N2 0.4 1.3 0.8
1.34
s
1.68 drift (%) Dyn. 0.4 1.1 0.9
TLFeBOOK
existing and contemporary buildings, equal to 0.32 and 0.30, 0.74 and 0.58, and 1.07 and 0.78
for the rst, second and third sets of ground motions, respectively. Much lower coecients of
variation were obtained for the top displacement. They amount for the existing and contempo-
rary buildings to 0.17 and 0.12, 0.41 and 0.29, and 0.78 and 0.35 for the three sets of ground
motions, respectively. All the above results are determined for the ground motions scaled
to elastic spectral acceleration 0:45 g (for see Table I) at the period of the equivalent
SDOF system T. Most realistic are the values for the third set of ground motions, because
accelerograms for rst two sets were corrected [4]. Therefore the coecients of variation for
the rst two sets of ground motions are underestimated.
The coecient of variation for top displacement of the existing (0.78) and the contemporary
buildings (0.35), calculated for the third set of ground motions, are in good agreement with
the coecients of variation for the SDOF model, which are presented in Figure 12 (Chart for
T =0:3s) of the companion paper [4]. Note that the mean ductility, for which the coecient of
variation was calculated, amounts to 6.5 and 1.8 for the existing and contemporary buildings,
respectively. A low value of the coecient of variation for the contemporary building is the
consequence of a low ductility demand.
In the companion paper [4] we suggested the value 0.7 as a rough estimate for the coecient
of variation for short-period buildings. That value is in good agreement with the coecient
of variation for the top displacement of the existing building. The coecient of variation
for the top displacement of the contemporary building is much smaller than the suggested
estimate due to low ductility demand. The coecient of variation for the top displacement of
the contemporary building would also increase to about 0.7 if the building was subjected to
a higher ground motion intensity.
More details on the dispersion measure used in the N2 method and comparisons with results
of non-linear dynamic analyses are presented in Reference [14].
6. INCREMENTAL N2 ANALYSES
The incremental N2 (IN2) method was employed for both test examples. The elastic spectral
acceleration at a period T (Equation (3)) was chosen as the intensity measure, whereas top
displacement and maximum story drift were chosen as damage measures.
The results of IDA analysis in terms of the top displacement for dierent levels of spectral
acceleration and dierent accelerograms are presented in Figure 7 for the existing building
and for two sets of ground motions. Additionally, the summarized (mean) IDA curves have
been plotted.
In Figure 8(a) the results of the incremental N2 method are compared with the mean IDA
curves for the two test buildings and all the ground motion sets used in the analysis. Pushover
curves for MDOF models, normalized by the eective masses m
e
= m
, i.e. the capacity

diagram for the MDOF model, are also shown. As expected, the top displacement demand for
the rst set of ground motions is much lower than the demand obtained with the other two
sets of ground motions. The demand for the second and third sets of ground motions is very
similar due to the similar mean elastic spectra, although both sets consist of quite dierent
accelerograms. Very good agreement can be observed between the IN2 and IDA results.
In Figure 8(b) the maximum story drifts obtained using the IN2 method and the IDA
analysis are shown. Maximum story drift was employed as the damage parameter, following
TLFeBOOK
64 M. DOL
SEK AND P. FAJFAR

0 5 10 15 20 0 5 10 15 20
0
0.5
1
1.5
2
S
a

(
g
)
Top displacement (cm) Top displacement (cm)
IDA
Single acc.
Capacity diagram
First set of
ground motions
Second set of
ground motions
Figure 7. Elastic spectral acceleration at period T versus top displacement: results obtained
for single accelerograms and the mean IDA curve for the existing building and for the rst
and second sets of ground motions. The capacity diagram of the MDOF model, obtained
by pushover analysis, is also presented.
the approach proposed by Vamvatsikos and Cornell [5]. Note, however, that in the case of the
IN2 analysis, the maximum story drift occurs always in the rst story, whereas in the case of
IDA analysis it sometimes occurs in the second or third story. The relations between the base
shear divided by the eective mass and the rst story drift obtained by pushover analyses are
also shown in Figure 8(b). Again, good correlation of the results can be observed for ground
motion sets 2 and 3. In the case of ground motion set 1, the IN2 results are conservative
for high intensities of ground motion. Owing to the gentle slope of the curves at higher
intensities, a better approximation is obtained for spectral acceleration given displacement,
than for displacement demand given ground motion intensity.
IN2 curves for top displacement were constructed with the suggested improvement of ac-
curacy for low seismic intensity (Equation (10)). The inuence of that improvement can be
observed in Figure 6 where IN2 curves are presented for SDOF models. In the region below
the yield point the IN2 curve is bilinear, whereas the capacity diagram, determined without
the consideration of Equation (10), is linear.
7. CONCLUSIONS
The N2 method for simplied non-linear seismic analysis has been extended to inlled frames.
Compared to the simple basic variant, two important dierences apply. A multi-linear ideal-
ization of the pushover curve, which takes into account strength degradation after the inll
fails, has to be made, and specic reduction factors, developed in the companion paper [4],
have to be used for the determination of inelastic spectra. The proposed approach can also
be employed for the determination of an approximate summarized IDA curve with much less
computational eort than in the case of incremental dynamic analysis. Reasonably accurate
results can be obtained, as demonstrated by the test examples. Typically, the results obtained
TLFeBOOK
0 1 2 3 4
0 1 2 3 4
Maximum story drift (%)
Maximum story drift (%)
0 5 10 15 20
S
a

(
g
)
S
a

(
g
)
0 10 20 30 40
0
0.5
1
1.5
2
0
0.75
1.5
3
2.25
Top displacement (cm)
Top displacement (cm)
first
second
third
set of ground motion
IDA
IN2
Figure 8. Comparison between the results of the IN2 and IDA analysis for mean values of top dis-
placement and maximum story drift of the existing (above) and contemporary building (below).
using the N2 analysis are conservative in the most critical story, and non-conservative in the
other stories.
The extended N2 method is, like any approximate method, subject to several limitations.
In addition to the basic limitations, described in References [1, 15], limitations related to
the range of parameters dening the idealized pushover curve apply [4]. However, in many
cases the extended N2 method may provide a suitable tool for a rational evaluation procedure
for inlled frame building structures for multiple performance objectives. Formulation of the
method in the accelerationdisplacement format enables a visual interpretation of the procedure
and of the relations between the basic quantities controlling the seismic response. This feature
is attractive to designers.
A similar extension of the N2 method can be made to any structural system, provided that
an appropriate specic RT relation is available.
Note an essential dierence between the determination of the demand point in the original
capacity spectrum method, which is based on equivalent elastic spectra, and in the N2 method,
TLFeBOOK
66 M. DOL
SEK AND P. FAJFAR

which is based on inelastic spectra. In the rst case the demand point is dened as the
intersection of the demand spectrum and the capacity curve. In the second case the demand
point is represented by the intersection of the demand spectrum and the horizontal line
representing the yield plateau. In the special case of an ideal elasto-plastic capacity curve,
there is no dierence between the two procedures.
ACKNOWLEDGEMENTS
The research was supported by the Ministry for Education, Science and Sport of the Republic of
Slovenia and by the European Commission. This support and valuable comments of two anonymous
reviewers are gratefully acknowledged.
REFERENCES
1. Fajfar P. A nonlinear analysis method for performance-based seismic design. Earthquake Spectra 2000; 6:
573592.
2. CEN. Eurocode 8Design of structures for earthquake resistance, Part 1, European standard prEN 1998-1, Draft
No. 6, European Committee for Standardization, Brussels, 2003.
3. Dol sek M. Seismic response of inlled reinforced concrete frames. Ph.D. Thesis (in Slovenian), University of
Ljubljana, Faculty of Civil and Geodetic Engineering, Ljubljana, Slovenia, 2002.
4. Dol sek M, Fajfar P. Inelastic spectra for inlled reinforced concrete frames. Earthquake Engineering and
5. Vamvatsikos D, Cornell CA. Incremental dynamic analysis. Earthquake Engineering and Structural Dynamics
2002; 31:491514.
6. Fajfar P, Ga sper si c P. The N2 method for the seismic damage analysis of RC buildings. Earthquake Engineering
and Structural Dynamics 1996; 25:3146.
7. Cornell CA, Krawinkler H. Progress and challenges in seismic performance assessment. PEER Center News
2000; 1(4):13.
8. Cornell CA, Jalayar F, Hamburger RO, Foutch DA. Probabilistic basis for 2000 SAC Federal Emergency
Management Agency Steel Moment Frame Guidelines. Journal of Structural Engineering (ASCE) 2002;
128:526533.
9. Carvalho et al. Preparation of the Full-scale Tests on Reinforcement Concrete FramesDenition of the
Specimens, Loads and Testing ConditionsICONS Topic 2, LNEC Report 1999.
10. Dol sek M, Fajfar P. Soft storey eects in uniformly inlled reinforced concrete frames. Journal of Earthquake
Engineering 2001; 5(1):112.
11. Fardis MN (ed.). Experimental and numerical investigations on the seismic response of RC inlled frames and
recommendations for code provisions. ECOEST/PREC 8, Report No. 6, LNEC, Lisbon, 1996.
12. Dol sek M, Fajfar P. Mathematical modelling of an inlled RC frame structure based on the results of pseudo-
dynamic tests. Earthquake Engineering and Structural Dynamics 2002; 31:12151230.
13. Prakash V, Powell GH, Campbell S. DRAIN-2DX Base program description and user guide, Version 1.10,
University of California, Berkeley, 1993.
14. Dol sek M, Fajfar P. IN2A simple alternative for IDA. Proceedings of the 13th World Conference on
Earthquake Engineering, Paper No. 3353, Vancouver, Canada, August 2004.
15. Fajfar P. Structural analysis in earthquake engineeringa breakthrough of simplied non-linear methods,
Keynote lecture. Proceedings of the 12th European Conference on Earthquake Engineering, London, 2002.
TLFeBOOK
Published online 17 November 2004 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/eqe.417
Optimum design of structures against earthquake by wavelet
neural network and lter banks
Eysa Salajegheh
,
and Ali Heidari
Department of Civil Engineering, University of Kerman, Kerman, Iran
SUMMARY
Optimum design of structures for earthquake is achieved by simulated annealing. To reduce the com-
putational work, a fast wavelet transform is used by means of which the number of points in the
earthquake record is decreased. The record is decomposed into two parts. One part contains the low
frequency of the record, and the other contains the high frequency of the record. The low-frequency
content is the eective part, since most of the energy of the record is contained in this part of the
record. Thus, the low-frequency part of the record is used for dynamic analysis. Then, using a wavelet
neural network, the dynamic responses of the structures are approximated. By such approximation, the
dynamic analysis of the structure becomes unnecessary in the process of optimization. The wavelet
neural networks have been employed as a general approximation tool for the time history dynamic
analysis. A number of structures are designed for optimal weight and the results are compared to those
corresponding to the exact dynamic analysis. Copyright ? 2004 John Wiley & Sons, Ltd.
KEY WORDS: simulated annealing; fast wavelet transform; wavelet neural network; dynamic analysis;
lter banks
INTRODUCTION
The optimum design of structures is usually performed to select the design variables such
that the weight or cost of the structure is minimized, while all the design constraints are
satised. The external loads on the structures can be static [13] or dynamic [4]. In the
present study, the design variables are considered as the member cross-sectional areas, which
are chosen as discrete variables. The design constraints are bounds on member stresses
and joint displacements. The optimum design problem is formulated as a mathematical
non-linear programming problem and the solution is obtained by simulated annealing (SA)
by Reference [5].
Correspondence to: Eysa Salajegheh, Department of Civil Engineering, University of Kerman, Kerman, Iran.
Contract}grant sponsor: Ministry of Science, Research and Technology, Iran
Received 17 February 2004
Revised 31 May 2004
Copyright
?
2004 John Wiley & Sons, Ltd. Accepted 21 July 2004
TLFeBOOK
68 E. SALAJEGHEH AND A. HEIDARI
For problems with a large number of degrees of freedom, the structural analysis is time
consuming. This makes the optimal design process inecient, especially when a time history
analysis is considered. In order to overcome this diculty, discrete wavelet transforms (DWT)
and fast wavelet transforms (FWT) are used. Using these transformations, the main earthquake
record is modelled as a record with a very small number of points. Thus the time history
dynamic analysis is carried out with fewer points. In References [68], DWT and FWT are
used for dynamic analysis of structures. These transformations are powerful means for dynamic
analysis and the time required is far less than the classical methods. Therefore, the DWT
and FWT are used for optimization of structures with earthquake loading [9, 10]. Despite
a substantial reduction in the dynamic analysis, the optimization process requires a great
number of time history dynamic analyses; thus the overall time of the optimization process
for earthquakes can still be very long. In this work, in order to overcome this diculty, the
dynamic responses of the structures are approximated using wavelet neural networks (WNN)
[11, 12]. By such an approximation, the dynamic analysis of the structure is not necessary
during the optimization process. This network is inspired by both feedforward neural networks
and wavelet decompositions. An algorithm of backpropagation type is proposed for training
the network. In this network, the input is the damping ratio and the angular natural frequency
of the structure, and the output is the dynamic responses of a single degree of freedom
structure against these reduced points. After training the network, using the inverse wavelet
transform (IWT), the results of the dynamic analysis are obtained for the original earthquake
accelerograph record from the output of the network. The numerical results of optimization
show that this approximation is a powerful technique and the required computational eort
can be substantially reduced.
In the following, rst a brief discussion of SA is presented. Then a brief discussion of the
FWT and WNN are outlined. The details of the optimization approach with approximation
concepts are discussed and some numerical examples for the optimum design of structures
are presented. The computational time is compared to that of the exact optimization method.
SIMULATED ANNEALING FOR OPTIMIZATION
Simulated annealing (SA) is an eective optimization method among various available tech-
niques [13]. The common approach shared by the SA techniques is the avoidance of a gradient-
based search and thus reduction of the possibility of getting stuck in local optima. The method
does not require the sensitivities of the functions and it has the capability of nding the optimal
solution with discrete design variables. The basis of the SA is a stochastic search procedure
for nding the minimum of the function by emulating the natural process found in metals
during a temperature drop. On the other hand, a system initially at a high-energy state is
cooled down to reach the lowest energy state. The SA method makes random design changes
with a probabilistic acceptance criterion during the search. This unique characteristic enables
the search process to escape from a local minimum. In the SA, the value of an objective
function that we want to minimize is analogous to the energy in a thermodynamic system.
The temperature starts at a high value and is reduced by a decay rate during the search.
This implies that the algorithm accepts the almost random moves of the search at the high
temperature, and the probability of acceptance of any move drops at the low temperature. At
a given temperature, t, the algorithm perturbs the position of an atom randomly and evaluates
TLFeBOOK
OPTIMUM DESIGN OF STRUCTURES AGAINST EARTHQUAKE 69
the resulting change in the energy of the system. If the new energy state is lower than the
initial state, then the new conguration of the atoms is accepted. Otherwise, the new state
that increases the energy may be accepted or rejected based on an acceptance function, P
r
.
There are several acceptance functions and the most frequently used function is the Boltzmann
probability distribution:
P
r
=
1
1 + e
(O
A
O
C
)}0
(1)
where O
A
is the value of the appropriate objective function for a candidate design at an
adjacent point randomly selected, O
C
is the value of the objective function for current design,
and 0 is an adjustable parameter described as temperature. The criterion to accept or reject
the new states when O
A
O
C
, is that a number in the interval (0, 1) is randomly selected and
compared with P
r
. If the number is less than P
r
, then the perturbed state is accepted, otherwise
it is rejected. The starting temperature and the method of reducing temperature are important
parts of the method. In an optimization problem, 0 is just a control parameter that has no
physical concepts and it only regulates the convergence of the process.
The cooling schedule requires the choices of starting temperature, nal temperature and
cooling factor, which are introduced to adjust the reduction of the temperature in the conse-
quent cooling cycles [13, 14].
The starting temperature will be high for higher values of the starting acceptance probability.
Therefore, it is generally chosen in the range of (0.5, 0.9), usually 0.5 is used [1315]. The
nal acceptance probability is equated to small values, 10
7
or 10
8
. The cooling factor serves
to reduce the temperature gradually between successive cooling cycles. For this purpose, rst
the number of cooling cycles (NC) is assigned, and then the temperature of the next cooling
cycle with reference to that of the previous cycle is calculated. A small value for the cooling
factor might cause a rapid cooling schedule, which may end up with the stagnation of the
algorithm in a local optimum [16]. On the other hand, a large value of NC will eliminate
this situation by careful annealing, however, it will cause a prohibitively heavy computational
burden. It has been experienced that for NC=100 a premature local optimum is found, and
so NC=200 and 300 are appropriate values [1315].
In this work, SA is used to minimize the weight of the structures under consideration while
the design constraints are limits on member stresses and joint displacements. In the process
of the SA, the numerical values of the constraints are evaluated at all the time intervals of the
earthquake and the non-critical values are deleted to form the appropriate objective function
required by the SA.
DECOMPOSITION OF EARTHQUAKE RECORD BY FWT
Wavelets are mathematical functions that cut up data or function into dierent frequency
components, and then study each component with a resolution matched to its scale. They
have advantages over traditional Fourier methods in analysing physical states where the signal
contains discontinuities and sharp spikes. Unlike the Fourier transform, the wavelet transform
has dual localization, both in frequency and in time [17, 18]. These characteristics make
wavelets an active subject with many exciting applications. There are various types of wavelet
transforms: continuous wavelet transforms [17], discrete wavelet transforms [18] and fast
TLFeBOOK
wavelet transforms [19]. In this paper, both FWT and DWT are used. The FWT is used to
reduce the number of points of the earthquake record. The DWT is used to construct the
wavelet neural network.
It turns out that the wavelet transform can be simply achieved by a tree of digital l-
ter banks. The main idea behind the lter banks is to divide a signal into two parts; the
rst is the low-frequency part and the other is the high-frequency part. This idea can be
achieved by a set of lters. A lter bank consists of a low-pass lter and a high-pass l-
ter, which separate a signal into dierent frequency bands. A lter may be applied to a
signal to remove or enhance certain frequency bands of the signal [20]. By applying a low-
pass lter to a signal s[t] of length N (number of points), the high-frequency bands of
the signal are removed and an approximate version of the original signal is obtained. A
high-pass lter removes the low-frequency components of the original signal, and the re-
sult is a signal containing the details of the main signal. By combining these two lters
into a lter bank, the original signal is divided into an approximate signal and a detail
signal.
A multi-level decomposition of the original signal is performed by repeating the decom-
position process. In the next stage, the low-pass ltered output signal is used as input to
the lter bank [21]. If the computation of a wavelet transform is reduced to a FWT, then
the resulting implementation is very ecient. Several FWT algorithms have been devised
for computation of wavelet transform coecients. In this paper the Mallat algorithm [19] is
used for dynamic analysis of structures against earthquake [6, 7]. In this method the original
record is decomposed into two records: the detail record D
)
and the approximate record A
)
for ) =1, . . . , J and k =1, . . . , K, which are obtained as follows:
D
)
=
t
s[t]h
)
[t 2
)
k] (2)
A
)
=
t
s[t]q
)
[t 2
)
k] (3)
where ) and k are integer numbers. The symbol denotes complex conjugate, the value of 2
)
is the dilation factor; the value of k is the shifting factor, and [.] is used for discrete notation:
h
)
[t 2
)
k] is discrete wavelet analysis and q
)
[t 2
)
k] is scaling function analysis, dened
as [19]:
h
)
[t 2
)
k] =2
)}2
(2
)
(t 2
)
k)) (4)
q
)
[t 2
)
k] =2
)}2
(2
)
(t 2
)
k)) (5)
Dierent functions can be chosen for and . In the present study, the following functions
are selected. In the wavelet literature, these functions are referred to as the Harr functions
[18] that are easy to implement and the numerical results indicate that the Harr functions are
appropriate.
(t) =
1 06t60.5
1 0.5t61
(t) =
1 06t61
0 otherwise
(6)
TLFeBOOK
Wavelets and scaling functions must be deduced from one stage to the next as follows:
q
1
[t] =q[t], h
1
[t] =h[t] (7a)
q
)+1
[t] =
k
q
)
[k]q[t 2k] (7b)
h
)+1
[t] =
k
h
)
[k]q[t 2k] (7c)
The number of points of the earthquake record is reduced by the FWT. The decomposition
starts from the original record of the earthquake (s[t]), and produces two sets of records,
D
)
, and A
)
. These vectors are obtained by convolving s[t] with the low-pass lter for A
1
by
Equation (3), and with the high-pass lter for D
1
by Equation (2). Then, from record A
1
,
the two records A
2
and D
2
are evaluated and the process is continued until A
J
and D
J
are
evaluated for the J-th stage. For the earthquake record, the approximation record (A
)
) with
low-frequency components is the eective part. The detail record (D
)
) with high-frequency
components is not eective. Therefore A
)
is used for dynamic analysis. For all the earthquake
records, the low-frequency content is the eective part, because most of the energy of the
record is in the low-frequency part of the record. On the other hand, for a record, the shape and
the eects of the entire low-frequency component are similar to those of the main record. For
an earthquake record if we remove the high-frequency components, the record is dierent, but
we can still distinguish the pattern of the record. However, if we remove enough of the low-
frequency components, we dont distinguish the record. The numerical results of the dynamic
analysis show that this approximation is a powerful technique and the required computational
work can be reduced greatly. The errors in the proposed methods are small. In Figure 1, the
El Centro earthquake record (S-E 1940) for the A
)
signal is shown.
In this paper, this process is repeated in three stages, and the number of points of the
original record is reduced to 0.125 of the primary points. The error is negligible, in par-
ticular in the rst three stages of decomposition [6, 7]. To back up the applicability of the
approach, the results of the decomposition up to the 5th stage are presented in one of the
examples.
In this method, the process can be inversed and the original record can be computed. This
process is named as inverse wavelet transform (IWT). The original signal s[t], can be achieved
through the IWT process, by using the D
)
and A
)
as:
s[t] =
J
)=1
k
D
)
h
)
[t 2
)
k] +
J
)=1
k
A
)
q
)
[t 2
)
k] (8)
where

h
)
is called the synthesis wavelets and q
)
is called synthesis scaling functions [6, 7]. In
the present study, the A
)
record is only used in the process of the IWT, because the eects
of the D
)
are negligible, as explained.
WNN FOR DYNAMIC RESPONSE APPROXIMATION
The wavelet neural networks (WNN) [22, 23] used to approximate the dynamic responses of
the structure consist of three layers: input layer, hidden layer, and output layer. Each layer
TLFeBOOK
Figure 1. (a) The El Centro record, and decomposition of the El Centro
records; (b) A
1
; (c) A
2
; (d) A
3
; (e) A
4
; and (f) A
5
.
has one or more nodes. Figure 2 depicts the schematic diagram of the three layers of the
WNN. As illustrated in Figure 2, the input data vector X is connected to the input nodes
of the networks. The connections between input units and hidden units, and between hidden
units and output units are called weights U and W, respectively. The modied training steps
of the WNN are as follows:
(a) The number of nodes in the three layers is dened. These numbers in input, hidden
and output layers are Q, R and M, respectively.
(b) The activation function of the hidden layer is chosen. Dierent functions can be
employed. In this paper the mother wavelet is used as follows [24]:
(x) =xe
0.5x
2
(9)
Then, the function
a
i
,b
i
can be calculated from the mother wavelet, with dilation a
i
and translation b
i
, as:
a
i
, b
i
(x) =2
0.5a
i
(2
a
i
x b
i
) (10)
where a
i
and b
i
are integer numbers. Now, the activation function of the wavelet nodes
by substitution of Equation (9) into (10), has the following form:
a
i
, b
i
(x) =2
0.5a
i
(2
a
i
x b
i
)e
0.5(2
a
i
xb
i
)
2
(11)
TLFeBOOK
Figure 2. Wavelet neural network.
(c) The weights, u
qr
and w
rm
(components of U and W), are endowed with random values.
(d) Input learning samples X
n
(q) and the corresponding target output values J
n
(m)
1
are
chosen, where n is the number of learning samples, 1 stands for target value, and q
and m are the appropriate components of the vectors X
n
and J
n
, respectively.
(e) The output value of the sample J
n
is calculated as:
J
n
(m) =
JK
r=1
w
rm
2
0.5)
2
)
Q
q=1
u
qr
X
n
(q) k
(12)
where JK =(J + 1)(K + 1) and the values of ) and k are computed as:
) =r}(2K + 1) J (13)
k =rem(r, 2K + 1) K (14)
in which rem(r, 2K + 1) is the remainder of r}(2K + 1).
(f) The instantaneous gradient vectors are computed as follows:
ow
rm
=
cE
cw
rm
=
N
n=1
[J
n
(m)
1
J
n
(m)]2
0.5)
2
)
Q
q=1
u
qr
X
n
(q) k
(15)
ou
qr
=
cE
cu
qr
=
M
m=1
n=1
[J
n
(m)
1
J
n
(m)]w
rm
2
0.5)
c
cq
X
n
(q)
(16)
TLFeBOOK
where q
=

Q
q=1
JK
r=1
u
qr
X
n
(q). If
n
=2
)
q
k, then the derivative of with respect

to q
is:
c
cq
=2
)
n
2
e
0.5
n
2
2
)
e
0.5
n
2
(17)
The error function is the mean square error function, that is
E =0.5
N
n=1
M
m=1
[J
n
(m)
1
J
n
(m)]
2
(18)
In this paper, the steepest descent method is used to minimize E.
(g) The modied weight in backpropagation (BP) is calculated by:
w
new
rm
=w
old
rm
+ w
new
rm
(19)
u
new
qr
=u
old
qr
+ u
new
qr
(20)
in which the values of w
new
rm
and u
new
qr
are computed as:
w
new
rm
=p
cE
cw
old
rm
+ :w
old
rm
(21)
u
new
qr
=p
cE
cu
old
qr
+ :u
old
qr
(22)
where p is the learning rate factor, and : is a momentum factor.
(h) If the output error falls below a setting value, the learning procedure of the WNN is
stopped, otherwise it will return to Step (d).
In fact, the method of evaluation of the errors in the WNN is similar to BP but the
node activation function is the mother wavelet with varying translation and scale values.
This modication results in better output than standard BP as the activation function in the
standard BP is a simple function and this is not adequate for approximating the functions such
as the earthquake records. The results of the training process are the approximate dynamic
responses (displacements) of a structure with single degree of freedom at all the time intervals
as explained further in the subsequent sections.
SIMULATION OF DYNAMIC RESPONSES BY WNN
One of the methods used for the evaluation of the dynamic responses of a linear analysis
of structures is the modal superposition. In the modal superposition method it is shown that
a multi-degree of freedom (MDOF) system can be converted into multi-systems of single
degree of freedom (SDOF). The solution of each SDOF system may be found by Duhamels
integral. After calculating the response of each SDOF system, the response of the MDOF
system can be found by the superposition method. The Duhamels integral is shown as [25]:
,(t) =
1
=
D
t
i+1
t
i
,(t)e
=(tt)
sin =
D
(t t) dt (23)
TLFeBOOK
where is the damping ratio, = is the angular natural frequency, , is the acceleration record,
and =
D
is equal to =
1
2
. In Duhamels integral for a specic earthquake record, ,,
t
i
and t
i+1
are constants. Therefore, two independent variables exist and the input vari-
ables of WNN are as and =. The target vector is the response of the SDOF system.
In this method, the responses of the SDOF are computed at all time intervals of the earth-
quake record. The number of nodes of input layer, hidden layer and output layer are cho-
sen as 2, 12 and 336, respectively. If we use a =12, the value of 2
a
is equal to 4096.
This number is greater than the number of points of the earthquake record. According to
wavelet theory choosing at least 12 values can show the variant of the record suitability
[12]. In fact, the values of J and K in Equations (2) and (3) should be chosen such that
(J + 1)(K + 1) =a. In this paper J =3 and K =2 are used, as dierent values of these
parameters would not aect the nal results. The total time intervals of the El Centro earth-
quake are 2688 and the number of reduced intervals by FWT (3rd stage of decomposition) is
evaluated as 336.
In fact, the analysis of the SDOF structure by the standard Duhamels integral is not
required. By changing the values of and =, the trained network provides the complete time
history dynamic analysis of any SDOF against the A
3
record with less computational eort.
The eciency of the dynamic analysis of the SDOF structure is due to two factors. The major
factor is that the number of the points of the record is reduced, and the other factor is that
the computational time of the dynamic analysis of the structure against the reduced points
by the WNN is less than that of Duhamels integral [26]. Then, by modal superposition, the
MDOF system can be converted into a number of SDOF systems, each of which has dierent
and =. Thus, the dynamic time history analysis of the MDOF system under investigation
can be obtained by using the results of the WNN.
MAIN STEPS OF OPTIMIZATION WITH FWT AND WNN
The main steps in the optimization process employing FWT and WNN for earthquake loading
are as follows:
(a) The functions and are dened. In this study, Equation (6) is employed.
(b) The number of stages for decomposition of the record is chosen. Here three stages are
used [6, 7].
(c) The FWT of the earthquake record in three stages is computed.
(d) The approximate version of the earthquake record in the 3rd stage (A
3
) is used for
dynamic analysis.
(e) The dynamic responses of a SDOF structure against A
3
are calculated. This process is
repeated for a number of the SDOF structures with dierent and = for training the
network.
(f) The number of nodes in the input layer, hidden layer, and output layer are chosen. In
this paper 2, 12 and 336 nodes are used, respectively.
(g) The activation function in the wavelet layer is chosen by using Equation (11). In this
paper, 12 activation functions are employed.
(h) The weights of the wavelet neural network are optimized. This process is repeated until
the network is converged.
TLFeBOOK
(i) For each and =, the trained network is used to calculate the dynamic responses of
any SDOF structure against the A
3
record.
(j) The responses of the SDOF structure against the original earthquake record are evaluated
by Equation (8). Now, the dynamic responses of the structure under investigation are
determined by the modal superposition method.
(k) SA is used for optimization. The optimization convergence is checked, if convergence
is satised, the process is stopped, otherwise the member cross-sections are updated
and the process is repeated from Step (i).
It can be observed that in the process of optimization, the direct dynamic analysis of the
structure is not required. In fact, the necessary responses are found by the trained WNN. In
addition, the time history analysis of the structure required for the training is achieved for a
record with a smaller number of time intervals. For the El Centro earthquake, the best ratio
of the reduced intervals with respect to the original record is chosen as 0.125 [6, 7, 26]. It
should be mentioned that the training process should be carried out once for the specied
earthquake record. Thus, for optimum design of any structure against this earthquake, the time
history dynamic analysis is not required. The numerical results show that the evaluation of the
structural responses by trained network is faster than employing the Duhamels integral [26].
In summary, the FWT and WNN are used in two dierent ways to enhance the eciency
of the optimization process. The rst application is to use the FWT to reduce the cost of the
time history dynamic analysis. By the FWT, the original record is ltered and the number of
points in the resulting record (A
3
) is about 0.125 of the original record. The second aspect
is to reduce the computational cost of the overall optimization process by the WNN. This
is achieved through training a special network; the result of which is that the analysis of
the structures is not necessary during the optimization process. In addition, the WNN was
applied to the A
3
record. The WNN was not used for the original record because the number
of output nodes would be increased and training the system would be impossible. In fact, the
main application of the FWT is to reduce the number of points of the original earthquake
accelerogram for the process of the WNN, and the application of the WNN is to omit the
time history dynamic analysis of the SDOF systems during the optimization process which is
faster than the Duhamels integral [26].
It should be noted that the procedure outlined in the present study is for structural op-
timization under earthquake loads when the dynamic analysis is carried out by the modal
superposition method. However, investigation is in progress for other methods of dynamic
analysis as the training process is dierent.
NUMERICAL EXAMPLES
Two examples are optimized for minimum weight against the El Centro earthquake record
(S-E 1940). The computer times for analysis are obtained in clock time by a personal Pentium
2 PC. The number of points for the El Centro is 2688. The time interval for the record was
0.02 seconds. The number of points in A
3
was 336. The optimization is carried out by the
following methods:
(a) Simulated annealing with the original earthquake record (SAE).
(b) Simulated annealing using the FWT and WNN (SFW).
TLFeBOOK
Table I. Available member areas (cm
2
).
No. Area No. Area No. Area No. Area No. Area
1 0.8272 9 3.789 17 10.57 25 25.11 33 68.35
2 1.127 10 4.303 18 12.99 26 27.54 34 70.7
3 1.427 11 4.479 19 13.66 27 29.69 35 71.4
4 1.727 12 5.693 20 15.11 28 33.93 36 73.0
5 2.267 13 6.563 21 17.13 29 40.14 37 80.7
6 2.777 14 7.413 22 18.74 30 43.02 38 85.2
7 3.267 15 8.229 23 19.15 31 51.03 39 87.4
8 3.493 16 9.029 24 21.15 32 63.6 40 90.5
Table II. Results of optimization for ten bar truss.
Areas (cm
2
)
Member no. SAE SFW
1 25.11 29.69
2 13.66 17.13
3 25.11 29.69
4 13.66 17.13
5 0.827 0.827
6 0.827 0.827
7 25.11 27.54
8 25.11 27.54
9 51.03 51.03
10 51.03 51.03
Weight (kg) 731.2 782.9
Time (min) 142 16
In all the examples, the allowable stress is taken as 1100 kg}cm
2
, Youngs modulus is
2.1 10
6
kg}cm
2
, weight density is 0.0078 kg}cm
3
, and the damping ratio for all modes
is 0.05. The members are pipes, with radius to thickness less than 50. The problems are
designed with stress, the Eulers buckling and horizontal displacement constraints. The set of
available discrete values considered for the cross-sectional areas of the members is given in
Table I. The training time for the El Centro earthquake for a SDOF system is 47 minutes.
Example 1
The ten bar truss shown in Figure 3 is designed with height and span of 3 m. The truss is
simply supported at the joints 1 and 4. The mass of 5000 kg is lumped at each free node.
The horizontal displacement at joint 6 is considered to be less than 10 cm.
The results of optimization are given in Table II. In the cases of SAE and SFW the nal
weights are 731.2 and 782.9 kg, respectively. The time of computation for SAE and SFW are
142 and 16 minutes, respectively.
TLFeBOOK
Figure 3. Ten bar truss.
Table III. Member grouping for double layer grid.
No. Member no. No. Member no. No. Member no.
1 14; 3740 6 7; 16; 25; 34 11 47; 50; 58; 61; 69;
72; 80; 83; 91; 94
2 1013; 2831 7 4145; 96100 12 48; 49; 59; 60; 70;
71; 81; 82; 92; 93
3 1922 8 5256; 85-89 13 All diagonal
members
4 5; 9; 14; 18; 23; 9 6367; 7478
27; 32; 36
5 6; 8; 15; 17; 24; 10 46; 51; 57; 62; 68;
26; 33; 35 73; 79; 84; 90; 95
Example 2
A double layer grid of the type shown in Figure 4 is chosen with dimensions of 10 10 m
for the top layer and 8 8 m for the bottom layer. The height of the structure is 0.5 m and
is simply supported at the corner joints 1, 5, 21 and 25 of the bottom layer. The mass of
300 kg is lumped at each free node. The vertical displacement of joint 13 at the centre of the
bottom layer must be less than 10 cm. The members are grouped into 13 dierent types as
shown in Table III.
TLFeBOOK
Figure 4. (a) Double layer grid; (b) top layer; and (c) bottom layer.
For this example, the results of maximum displacements are compared for the original
earthquake record and those of the FWT. Results of the analysis of maximum displace-
ments of joints 2, 6, 13, 16, 23, 33, 40, 49, 51 and 57 in the x, , and : directions, for
the original earthquake record (OER), A
1
, A
2
, A
3
, A
4
and A
5
for the El Centro record for
the optimal results of the SFW method are given in Tables IV to VI, respectively. In this
study, the allowable error for the average maximum displacement is considered to be less
than 10%.
The results show that the maximum displacement of joints in A
4
and A
5
are not suitable,
therefore A
3
is used for dynamic analysis and optimization of the structures.
TLFeBOOK
Table IV. Results of maximum displacement of Example 2 in the x direction.
Maximum dynamic displacement 10
2
|OER A
)
|}OER 100
Joints
no. OER A
1
A
2
A
3
A
4
A
5
A
1
A
2
A
3
A
4
A
5
2 117 117 108 105 97 90 0.3 7.8 10.3 17.1 23.1
6 117 117 108 107 98 91 0.4 8.2 8.5 16.2 22.2
13 114 115 105 103 93 87 0.2 7.9 9.6 18.4 23.7
16 116 117 106 105 95 86 0.8 8.4 9.5 18.1 25.9
23 104 105 96 94 85 75 1.6 7.7 9.6 18.3 27.9
33 125 129 119 117 103 90 3.4 4.9 6.4 17.6 28
40 91 89 85 83 75 65 3.0 6.7 8.8 17.6 28.6
49 82 80 77 73 67 56 2.4 6.1 10.9 18.3 31.7
51 97 95 89 88 80 72 1.8 8.2 9.3 17.5 25.8
57 89 87 82 81 73 63 1.8 8.0 8.9 18 29.2
Table V. Results of maximum displacement of Example 2 in the , direction.
2
|OER A
)
|}OER 100
Joints
no. OER A
1
A
2
A
3
A
4
A
5
A
1
A
2
A
3
A
4
A
5
2 112 116 122 123 135 147 3.6 8.9 9.8 20.5 31.3
6 104 107 108 99 91 80 3.3 3.6 4.5 12.5 23.1
13 101 103 104 108 120 132 2.2 2.6 6.9 19.3 30.7
16 68 65 61 67 81 91 4.2 10.3 2.5 19.3 33.8
23 46 44 42 44 55 63 4.3 8.7 4.1 19.6 36.9
33 72 69 66 65 60 51 5.2 8.3 9.7 16.7 29.2
40 66 65 60 59 54 46 2.5 9.1 10.6 18.5 30.3
49 58 55 53 52 46 39 5.2 8.6 10.3 20.7 32.8
51 74 70 68 67 59 48 6.2 8.1 9.5 20.2 35.1
57 43 41 39 38 34 29 4.7 9.3 11.6 20.9 32.6
Table VI. Results of maximum displacement of Example 2 in the : direction.
3
|OER A
)
|}OER 100
Joints
no. OER A
1
A
2
A
3
A
4
A
5
A
1
A
2
A
3
A
4
A
5
2 3490 3689 3861 3891 4147 4329 5.7 10.6 11.5 18.8 24.1
6 3398 3594 3735 3803 3322 4138 5.7 9.9 11.9 2.3 21.8
13 3191 3353 3482 3532 3895 3989 5.1 9.1 10.7 22.1 25.1
16 2811 2942 2995 3065 3316 3596 4.6 6.5 9.1 18 27.9
23 2608 2716 2824 2905 3068 3364 4.1 8.3 11.3 17.6 28.9
33 2893 3001 3106 3151 3410 3609 3.7 7.4 8.9 17.9 24.7
40 3210 3376 3518 3587 3890 4106 5.2 9.6 11.7 21.2 27.9
49 2485 2590 2691 2734 3003 3312 4.2 8.3 10.1 20.8 33.3
51 2949 3078 3221 3278 3577 3716 4.4 9.2 11.1 21.3 26.1
57 2725 2822 3016 3051 3228 3401 3.5 10.7 11.9 18.5 24.8
TLFeBOOK
Table VII. Results of optimization for double layer grid.
Areas (cm
2
)
Group no. SAE SFW
1 68.35 68.35
2 12.99 12.99
3 3.493 4.479
4 10.57 12.99
5 10.57 12.99
6 17.13 19.15
7 10.57 13.66
8 10.57 12.99
9 10.57 13.66
10 12.99 12.99
11 5.693 9.029
12 18.74 21.15
13 21.15 21.15
Weight (kg) 5354.2 5716.3
Time (min) 432 47
The results of optimization are given in Table VII. In the cases of SAE and SFW the nal
weights are 5354.2 and 5716.3 kg, respectively. The time of computation for SAE and SFW
are 432 and 47 minutes, respectively.
CONCLUSIONS
An ecient method is presented for discrete optimum design of structures for earthquake load.
The main goal is to reduce the computational cost of the optimum design procedure. Numerical
optimization techniques such as simulated annealing require the evaluation of the objective
function and the design constraints at a great number of design points. In each design point
the structure should be analysed to evaluate the necessary information. It can be seen that the
number of structural analyses is excessive and for large-scale structures with many degrees of
freedom, the optimization is dicult. In particular, when a time history dynamic analysis is
required. In the present study, to reduce the overall cost of optimization, some attempts have
been made to reduce the cost of dynamic analysis as well as the cost of the optimization
process. As far as the dynamic analysis is concerned, the idea of signal processing together
with wavelets and lter banks is employed. By this method, the acceleration record is ltered
and the number of points of the record is reduced. The structure is analysed dynamically with
a smaller number of time intervals. To reduce the cost of optimization, a neural network type
is employed. A wavelet neural network is presented to approximate the dynamic analysis of
the structure during the optimum design process. Thus, during optimum procedure the dynamic
analysis of the structure is not required.
The FWT is used in two dierent aspects to enhance the eciency of the optimization
process; the rst application is to reduce the cost of time history dynamic analysis. The
second aspect is to reduce the computational cost of the overall optimization process. The
numerical results show that in the proposed method, the time of optimization is reduced to
TLFeBOOK
about 10% of the time required for exact optimization. But the error is increased by a factor
of about 7%.
ACKNOWLEDGEMENTS
The nancial support of the project provided by the Ministry of Science, Research and Technology and
the Shahid Bahonar University of Kerman is gratefully acknowledged.
REFERENCES
1. Salajegheh E. Approximate discrete variable optimization of frame structures with dual methods. International
Journal for Numerical Methods in Engineering 1996; 39:16071617.
2. Salajegheh E. Discrete variable optimization of plate structures using dual methods. Computers and Structures
1996; 58:11311138.
3. Salajegheh E, Salajegheh J. Optimum design of structures with discrete variables using higher order
approximation. Computer Methods in Applied Mechanics and Engineering 2002; 191:13951419.
4. Papadrakakis M, Lagaros ND. Advances in computational methods for large-scale structural optimization.
Computational Mechanics for the Twenty-rst Century, Topping BHV (ed.). Saxe-Coburg Publications:
Edinburgh, U.K., 2000; 431449.
5. Kirkpatrik S, Gelatt CD Jr., Vecchi MP. Optimization by simulated annealing. Science 1983; 220:671680.
6. Salajegheh E, Heidari A. Time history dynamic analysis of structures using lter banks and wavelet transforms.
Computers and Structures, in press.
7. Salajegheh E, Heidari A, Saryazdi S. Approximate dynamic analysis of structures against earthquake by fast
wavelet transform. Iranian Journal of Science and Technology, submitted.
8. Salajegheh E, Heidari A. Dynamic analysis of structures against earthquake by combined wavelet transform and
fast Fourier transform. Asian Journal of Civil Engineering 2002; 3:7587.
9. Salajegheh E, Heidari A. Optimum design of structures against earthquake by an adaptive genetic algorithm using
wavelet transform. In Proceedings of the 9 th AIAA}USAF}NASA}ISSMO Symposium on Multidisciplinary
Analysis and Optimization, AIAA, Atlanta, Georgia, U.S.A., 2002.
10. Salajegheh E, Heidari A, Saryazdi S. Optimum design of structures against earthquake by a modied genetic
algorithm using discrete wavelet transform. International Journal for Numerical Methods in Engineering, in
press.
11. Zhang Q, Beveniste A. Wavelet networks. IEEE Transactions on Neural Networks 1992; 3:889898.
12. Thuillard M. Wavelets in Soft Computing. World Scientic: New York, 2001.
13. Balling RJ. Optimum steel design by simulated annealing. Structural Engineering and Mechanics 1991;
117:17801795.
14. Hasancebi O, Erbatur F. Layout optimization of trusses using simulated annealing. Advances in Engineering
Software 2002; 33:681696.
15. Bennage WA, Dhingra AK. Single and multiobjective structural optimization in discretecontinuous variables
using simulated annealing. International Journal for Numerical Methods in Engineering 1995; 38:27532773.
16. Chen GS, Bruno RJ, Salama M. Optimal placement of active}passive members in truss structures using simulated
annealing. AIAA Journal 1991; 29:13271334.
17. Grossmann A, Morlet J. Decomposition of Hardy function into square integrable wavelets of constant shape.
SIAM Journal on Mathematical Analysis 1984; 15:723736.
18. Daubechies I. Ten lectures on wavelets. CBMS-NSF Conference Series in Applied Mathematics, Montpelier,
Vermont, 1992.
19. Mallat S. A theory for multiresolution signal decomposition: the wavelet representation. IEEE Transactions on
Pattern Analysis and Machine Intelligence 1989; 11:674693.
20. Oppenheim AV, Schafer RW, Buck JR. Discrete-Time Signal Processing. Prentice-Hall: New Jersey, 1999.
21. Strang G, Nguyen T. Wavelets and Filter Banks. WellesleyCambridge Press: New York, 1996.
22. Zhang X, Qi J, Zhang R, Liu M, Hu Z, Xue H, Fan B. Prediction of programmed-temperature retention values
of naphthas by wavelet neural networks. Computers and Chemistry 2001; 25:125133.
23. Xu J, Ho DWC. A basis selection algorithm for wavelet neural networks. Neurocomputing 2002; 48:681689.
24. Farge M. Wavelet transforms and their application to turbulence. Annual Review of Fluid Mechanics 1992;
24:395457.
25. Paz M. Structural Dynamics: Theory and Computation. McGraw-Hill: New York, 1997.
26. Heidari A. Optimum Design of Structures against Earthquake by Advanced Optimization Methods. Ph.D.
Thesis, Department of Civil Engineering, University of Kerman, Kerman, Iran, 2004.
TLFeBOOK
Published online 17 November 2004 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/eqe.422
Incorporating low-cycle fatigue model into duration-dependent
inelastic design spectra
Y. H. Chai
;
Department of Civil and Environmental Engineering; University of California; Davis; Engineering Unit III; One
Shields Avenue; Davis; CA 95616; U.S.A.
SUMMARY
Seismic performance of structures is related to the damage inicted on the structure by the earthquake,
which means that formulation of performance-based design is inherently coupled with damage assess-
ment of the structure. Although the potential for cumulative damage during a long-duration earthquake
is generally recognized, most design codes do not explicitly take into account the damage potential of
such events. In this paper, the classical low-cycle fatigue model commonly used for seismic damage
assessment is cast in a framework suitable for incorporating cumulative damage into seismic design.
The model, in conjunction with a seismic input energy spectrum, may be used to establish an energy-
based seismic design. In order to ensure satisfactory performance in a structure, the cyclic plastic strain
energy capacity of the structure is designed to be larger than or equal to the portion of seismic in-
put energy contributing to cumulative damage. The resulting design spectrum, which depends on the
duration of the ground motion, indicates that the lateral strength of the structure must be increased
in order to compensate for the increased damage due to an increased number of inelastic cycles that
occur in a long-duration ground motion. Examples of duration-dependent inelastic design spectra are
developed using parameters currently available for the low-cycle fatigue model. The resulting spec-
tra are also compared with spectra developed using a dierent cumulative damage model. Copyright
? 2004 John Wiley & Sons, Ltd.
KEY WORDS: cumulative damage; ground motion duration; seismic design spectra; input energy
1. INTRODUCTION
An issue that has gained considerable attention in recent years concerns the eects of cumu-
lative damage that occurs during a strong earthquake. Although the potential for cumulative
damage is generally recognized, most seismic codes do not explicitly take into account such
damage. The use of a displacement ductility factor, as assumed in current design practice,
fails to account for cumulative damage since it is implicitly assumed that structural damage
Correspondence to: Y. H. Chai, Department of Civil and Environmental Engineering, University of California,
Davis, Engineering Unit III, One Shields Avenue, Davis, CA 95616, U.S.A.
Received 21 October 2003
Revised 29 July 2004 and 1 September 2004
Copyright
?
2004 John Wiley & Sons, Ltd. Accepted 1 September 2004
TLFeBOOK
84 Y. H. CHAI
occurs only due to the maximum response deformation and is independent of the number of
non-peak inelastic cycles or strain energy dissipation. For damage assessment, however, all
inelastic cycles must be considered contributory to damage, the accumulation of which may
become important depending on the characteristics of the ground motion. For a long-duration
ground motion, yielding structures undergo an increased number of cycles into the inelastic
range, in which case, the accumulation of damage may signicantly aect the overall per-
formance of the structure. Cumulative damage may also arise from multiple seismic events,
in which case, a series of pre- or after-shocks in combination with the main shock may be
treated as a single event with an extended duration.
Currently most of the seismic design codes do not explicitly take into account the ef-
fects of ground motion duration. A notable exception is the 1991 NEHRP Commentary [1]
where a suggestion was made to increase the lateral strength of the structure in order to
compensate for the cumulative damage that may occur during a long-duration earthquake.
Following that suggestion, Chai et al. [2] developed the so-called duration-dependent in-
elastic design spectra in order to account for the eects of cumulative damage. A key el-
ement in the formulation is the requirement that the structure must have sucient plastic
strain energy capacity so that the portion of seismic input energy contributing to cumula-
tive damage can be dissipated by the structure without a serious degradation of strength.
In estimating the plastic strain energy capacity, however, Chai et al. [2] used the damage
model proposed by Park and Ang [3], modied slightly to account for the plastic energy
dissipated under monotonic loading. In this paper, duration-dependent inelastic design spec-
tra are re-examined using the classical low-cycle fatigue model instead of the modied Park
and Ang model. An objective of this paper is to demonstrate that the previously proposed
procedure for duration-dependent spectra can be easily adapted for the classical low-cycle
fatigue model. Also as a part of this study, duration-dependent spectra derived using the
low-cycle fatigue model are compared with that derived using the modied Park and Ang
model [2].
2. SEISMIC ENERGY DEMAND
Since the excitation from an earthquake ground motion may be viewed as a process of impart-
ing seismic energy to a structure, various studies [412] have suggested the use of earthquake
input energy as a criterion for seismic hazard assessment or design. In the context of energy-
based seismic design, it may be postulated that, for satisfactory performance of the structure,
the cyclic plastic strain energy capacity of the structure must be equal to or greater than the
portion of seismic energy that is contributing to damage [2]. Mathematically, the energy-based
design criterion can be written as:
E
h
m
E
I
m
(1)
where the term E
h
=m on the left-hand side of Equation (1) corresponds to the plastic strain
energy capacity per unit mass and the term E
I
=m on the right-hand side of Equation (1)
corresponds to the maximum input energy per unit mass, and the parameter represents the
ratio of hysteretic energy to input energy, which will be discussed later. The seismic input
TLFeBOOK
LOW-CYCLE FATIGUE MODEL AND DURATION-DEPENDENT SPECTRA 85
Figure 1. Amplication factor for equivalent input energy velocity spectrum.
energy is commonly expressed as:
E
I
1
2
mv
2
e
(2)
where v
e
is the equivalent input energy velocity, which may further be expressed as a product
of an amplication factor
v
and the peak ground velocity ( x
g
)
max
, i.e.
v
e
=
v
( x
g
)
max
(3)
The seismic input energy imparted to a structure, as characterized by the amplication factor
v
, depends on the ground motion and the period of the structure. To facilitate the formulation
of duration-dependent inelastic design spectra, the bilinear equivalent velocity spectrum, as
proposed earlier by Akiyama [5] and Kuwamura and Galambos [13], is followed in this
paper. In this case, the amplication factor for spectral input energy is assumed to increase
linearly with the period in the short-period range but remain constant in the medium- and
long-period ranges, as shown in Figure 1. In equation form, the bilinear equivalent velocity
amplication factor
v
may be specied as:
v
=
1:2
v
T=T
c
for T6T
c
=1:2
v
TT
c
=1:2
(4)
where T is the period of the structure, and T
c
is the characteristic period of the ground motion,
which is dened as [14]:
T
c
=2
c
v
c
a
( x
g
)
max
( x
g
)
max
(5)
TLFeBOOK
86 Y. H. CHAI
and
v
is the peak amplication factor of the input energy spectrum to be specied later.
The terms ( x
g
)
max
and ( x
g
)
max
are the peak ground acceleration and peak ground velocity,
respectively, and the coecient c
a
corresponds to the ratio of spectral-acceleration-to-peak-
ground-acceleration in the acceleration-controlled region and the coecient c
v
corresponds to
the ratio of spectral-velocity-to-peak-ground-velocity in the velocity-controlled region. Con-
stant values of c
a
=2:5 and c
v
=2:0 are used in this paper, and these values are based on a
study of inelastic design spectra by Vidic et al. [14]. It is worth noting that the constant am-
plication factor
v
in the medium- and long-period range is intended to include the seismic
energy contained in the higher modes of exible structures since the spectral input energy in
multi-degree-of-freedom systems is given by a direct summation of the input energy for all
the modes in the structure [15].
A key parameter for specifying the seismic input energy spectrum is the peak amplication
factor
v
, which characterizes the spectral input energy near the predominant period of the
ground motion. By using the Parseval theorem to evaluate the total power of the ground
acceleration, the peak amplication factor of the equivalent input energy spectrum has been
shown to be [16]:
v
=0:343
( x
g
)
max
t
d
( x
g
)
max
(6)
where t
d
is the duration of the strong motion phase of the ground motion, and is dened
according to Trifunac and Brady [17]:
t
d
=t
0:95
t
0:05
(7)
where t
0:05
and t
0:95
are the times at 5% and 95% of the Arias intensity I
A
, respectively, and
the Arias intensity of the ground motion is dened as:
I
A

2g
t
o
0
x
2
g
dt (8)
where x
g
is the ground acceleration history and t
o
is the time to the end of the ground accel-
eration. It should be noted that, although Equation (6) has been shown to predict reasonably
well the seismic input energy to structures for a wide range of epicentral distances and site
conditions, it has not been calibrated against near-fault ground motions, and consequently, it
may not be applicable to such ground motions.
For structures yielding under an earthquake ground motion, only a portion of the spectral
input energy will be dissipated as the plastic strain energy. The portion of spectral input energy
to be dissipated as the plastic strain energy depends on the amount of damping as well as the
level of inelastic deformation experienced by the structure. For structures designed with a low
lateral strength, a large ductility demand may be imposed on the structure, in which case, the
plastic strain energy to be dissipated by yielding of its members would be correspondingly
large. On the other hand, for structures with a large damping or protected by supplementary
energy dissipators, the plastic strain energy dissipated by the structure would be relatively
small. The portion of spectral input energy to be dissipated as the plastic strain energy is
characterized by the parameter in Equation (1). Various empirical expressions have been
proposed for the ratio of plastic strain energy to spectral input energy [5, 13, 18]. In this study,
TLFeBOOK
the expression proposed by Fajfar and Vidic [18] for structures with 5% damping ratio and
force-deformation characterized by stiness-degrading hysteretic model is followed:
=1:13
(
c
1)
0:82
c
(9)
The above equation has been shown to calibrate well for a fairly large number of earthquake
ground motions and up to a displacement ductility factor of 10 [2, 19].
3. SEISMIC ENERGY CAPACITY
To ensure satisfactory performance of a structure during severe earthquakes, the structure
must be detailed for good energy dissipation in addition to the provision of adequate lateral
strength and stiness. In an earlier paper [2], the plastic strain energy capacity of structures
was calculated from a modied form of the cumulative damage model proposed by Park and
Ang for reinforced concrete structures [3]. The plastic strain energy capacity may also be
estimated from the classical low-cycle fatigue model [20], which was originally developed for
metals [21] but has recently been calibrated against reinforced concrete columns [11, 22, 23].
The adaptation of the low-cycle fatigue model for predicting the plastic strain energy capacity
of a structure is described next.
For seismic damage assessment, it is commonly assumed that the damage sustained by
structures during a strong ground motion is similar to metal fatigue under a variable amplitude
cyclic loading [2426]. Seismic damage, however, is normally assessed in terms of only
the plastic strain component since a relatively small number of load cycles is imposed by
the earthquake ground motion. Assuming a correspondence between material and structural
damage [2628], the low-cycle fatigue model under constant displacement amplitude cycles
may be written as:
y
=(
um
y
)(2N
f
)
c
(10)
where N
f
is the number of load cycles to cause failure, c is an exponent characterizing the
low-cycle fatigue characteristic of the structure,
um
is the ultimate displacement under a
monotonic loading,
m
is the peak displacement under constant amplitude cycles, and
y
is the yield displacement. Testing of metals indicated that the exponent c generally varies
between 0:5 and 0:8 [21, 29, 30], with an average value of c = 0:6 commonly assumed
for steel [26]. Experimental testing of reinforced concrete columns, however, indicated that
the fatigue ductility coecient is smaller than that of steel and a value of c = 0:285 has
been suggested for concrete [22]. By dening a monotonic displacement ductility factor as:
um
y
(11)
and a cyclic displacement ductility factor as:
y
(12)
TLFeBOOK
88 Y. H. CHAI
Equation (10) can be written as:
c
=(
m
1)(2N
f
)
c
+ 1 (13)
The above equation signies a relation between the displacement ductility capacity and the
number of constant displacement amplitude cycles to cause failure.
The number of load cycles N
f
in Equation (13) may be estimated from the plastic strain
energy dissipated by the structure during an earthquake ground motion [26]. The energy
dissipated in one full cycle with equal displacement in both directions may be written as:
(E
h
)
1cycle
=4
1
V
y
(
m
y
) (14)
where
1
is a parameter characterizing the shape of the lateral forcedisplacement hysteresis
loop. Average values for the loop shape factor have been proposed [19]:
1
=
0:75 for bilinear; non-degrading systems

0:50 for non-pinched; degrading systems
0:25 for severely-pinched; degrading systems
(15)
The total plastic strain energy dissipated by the structure for N
f
number of cycles is given
by:
E
ht
=N
f
(E
h
)
1cycle
(16)
Upon substituting Equations (10) and (14) into Equation (16), the total plastic strain energy
capacity may be written as:
E
ht
=2
1
V
y
um
(
c
1)
1+1=c
m
(
m
1)
1=c
(17)
For seismic design, however, only a portion of the total plastic strain energy should be used.
This is to ensure that the structure performs satisfactorily since repeated cycles may result
in excessive degradation of lateral strength [20]. In this case, the design plastic strain energy
should be taken as a fraction of the total plastic strain energy i.e.
E
h
=E
ht
=2
1
V
y
um
(
c
1)
1+1=c
m
(
m
1)
1=c
(18)
where 61. Small-scale testing of exural members indicated that a value of =0:7 is
appropriate for seismic design [20]. The predictive form of Equation (18) may be used in
the basic energy-based design criterion, i.e. Equation (1), to develop the duration-dependent
design spectrum.
Figure 2 shows a plot of the normalized design plastic strain energy, E
h
=V
y
um
, versus the
cyclic displacement ductility factor,
c
, for a range of fatigue ductility exponent from c =0:3
to 0:8. A monotonic displacement ductility factor of
m
=8, a design-to-total-plastic-strain-
energy ratio of =0:7 and a loop shape factor of
1
=0:5 are assumed. It can be seen
from the gure that the normalized design plastic strain energy is sensitive to the value of the
exponent c in the small displacement ductility range. For a given exponent c, the design plastic
strain energy decreases rapidly with increased cyclic displacement ductility factor. In the large
TLFeBOOK
Figure 2. Design plastic strain energy capacity.
displacement range, however, all curves converge to 2
1
(1 1=
m
) =0:6125 irrespective of
the fatigue ductility coecient.
4. SUPPLEMENTARY RELATIONS
In developing the duration-dependent inelastic seismic design spectrum, relations in addition
to seismic input energy and plastic strain energy capacity are needed. More specically,
elastic response acceleration spectra and force-reduction factor, expressed as a function of
cyclic displacement ductility factor and period, are needed. Relations adopted in this paper are
described next.
4.1. Elastic design spectra
Demands for seismic design of new structures or evaluation of existing structures have tra-
ditionally been prescribed in terms of spectral response acceleration using smooth curves or
connected straight lines with one curve for each damping ratio [31]. In this paper, the slightly
modied form of the 5% damped elastic design spectra proposed by Vidic et al. [14] is
followed. In this case, the spectral acceleration is determined by scaling the ground motion
parameters in the acceleration-controlled and velocity-controlled regions using an amplication
factor
a
, which is specied piece-wise in three period ranges:
a
=
1:0 + 2:5(c
a
1)T=T
c
for 0T60:4T
c
c
a
0:4T
c
T6T
c
2c
v
( x
g
)
max
=(( x
g
)
max
T) T
c
T
(19)
TLFeBOOK
90 Y. H. CHAI
The three period ranges are dened using the characteristic period T
c
, which has been given
previously in Equation (5). The same values of c
a
=2:5 and c
v
=2:0, suggested earlier for
the characteristic period, are used to generate the elastic design spectrum.
4.2. R
c
T relation
Typical lateral strength of structures, established on the basis of current design codes, is
considerably lower than the force level required for an elastic response. The appropriate level
of lateral strength depends on the ductility capacity of the structure, and may be established on
the basis of constant-ductility inelastic design spectra. Extensive studies have been conducted
in the past to determine the appropriate level of force reduction for a given ductility capacity,
and various expressions have been proposed in the literature [14, 3234]. In this paper, the
relation proposed by Vidic et al. [14] and adopted by Fajfar [33] for a stiness-degrading
hysteretic model is used:
R
(
c
1)T=T
o
+ 1 for T6T
o
c
TT
o
(20)
where
T
o
=0:65T
c
0:3
c
(21)
and T
o
corresponds to the transition period between the two regions of R
. In this case, the

force reduction factor is characterized by a linear increase with period T in the short-period
range (T6T
o
) but assumes a constant value that is equal to the cyclic displacement ductility
factor in the long-period range (TT
o
). Note that the transition period T
o
increases with
increased cyclic displacement ductility factor.
5. DURATION-DEPENDENT INELASTIC DESIGN SPECTRA
The duration-dependent inelastic design spectrum is determined by solving for the force re-
duction factor R
in the set of non-linear algebraic equations for seismic input energy, plastic
strain energy capacity, elastic design spectrum, R
c
T relation, and plastic-strain-energy-to-
input-energy ratio. The elastic seismic design spectrum is then divided by the force reduction
factor R
to arrive at the inelastic design spectrum. Examples of duration-dependent inelas-

tic design spectra are developed for concrete and steel structures using parameters currently
available for the low-cycle fatigue model.
5.1. Duration-dependent spectra for steel and concrete
The current database for structural members tested under cyclic loading indicates that
parameters characterizing the cumulative damage properties of steel and concrete compo-
nents are dierent. It is therefore instructive to investigate the duration-dependent spectrum
for these two types of structures. To this end, the fatigue ductility exponent of steel is taken
as c = 0:6, as reported by McCabe and Hall [26], while the fatigue ductility exponent of
concrete is taken as c = 0:3, which is rounded o from a value of 0:285, as reported by
El-Bahy et al. [22]. The monotonic displacement ductility factor of steel is taken as
m
=12
TLFeBOOK
(b)
(a)
Figure 3. Duration-dependent design spectra for concrete and steel structures:
(a) low a=v ratio = 0:6 g=ms
1
; and (b) high a=v ratio = 1:4 g=ms
1
:
or one plus the monotonic plastic ductility factor of 11, as reported by McCabe and Hall
[26], while the monotonic displacement ductility factor of concrete is taken as
m
=9:25 [19].
The loop shape factor is taken as
1
=0:75 for steel and
1
=0:5 for concrete, while the
design-to-total-plastic-strain-energy ratio is taken as =0:7 for both cases. It is important to
note that, even though these values have been selected from reported test results, their actual
values may vary signicantly especially when dierent details are used. For example, testing
of reinforcing bars under large amplitude cyclic loading indicated that the low-cycle fatigue
behavior of reinforcing bars may be characterized by an exponent of c = 0:5 [35], which
implies that the exponent for concrete members may be larger than the value of c = 0:3
adopted in this paper. Thus the duration-dependent spectra generated in this paper are intended
for illustration purposes only.
Figures 3(a) and (b) show the duration-dependent design spectra of steel and concrete
for low and high a=v ratios, respectively. The duration is varied from t
d
=10 s to 60 s.
TLFeBOOK
92 Y. H. CHAI
A peak ground acceleration of ( x
g
)
max
=0:4g is assumed, resulting in a peak ground ve-
locity of ( x
g
)
max
=0:67 m=s and 0:29 m=s for low and high a=v ratios, respectively. It can be
seen from the gures that the base shear coecient C
y
, which is plotted on the y-axis, in-
creases with increased duration of the ground motion, but the increase is dierent for steel
and concrete. For the selected parameters, the increase in C
y
of concrete is smaller than of
steel, and the smaller increase for concrete is observed for both low and high a=v ratios. For
example, for the low a=v ratio of 0:6g=ms
1
in Figure 3(a), the base shear coecient of steel
at T =0:31 s increases from C
y
=0:269 for t
d
=10 s to C
y
=0:455 for t
d
=60 s, representing
a 69% increase. On the other hand, the base shear coecient of concrete increases from
C
y
=0:293 to 0:381 for a duration increase from t
d
=10 s to 60 s, representing only a 30%
increase. Similarly, for the high a=v ratio of 1:4g=ms
1
in Figure 3(b), the base shear coe-
cient of steel increases from C
y
=0:284 to 0.468 for a duration increase from t
d
=10 s to 60 s,
corresponding to a 65% increase. For concrete, on the other hand, the base shear coecient
increases from C
y
=0:229 to 0:305, or a 33% increase, for a duration increase from t
d
=10 s
to 60 s. The lesser increase in the base shear coecient of concrete is possibly due to the
small value of the fatigue ductility exponent adopted for concrete. As evident in Figure 2, a
value of c = 0:3 for concrete implies a plastic strain energy capacity that is signicantly
larger than that of steel with a value of c = 0:6. The apparently larger plastic strain energy
capacity of concrete translates into a greater ability of concrete to dissipate seismic energy
without the need to increase the lateral strength in order to compensate for cumulative dam-
age. Since the accuracy of the proposed procedure depends on the accuracy of the selected
parameters, it will be prudent to conduct further tests with dierent details for both concrete
and steel members in order to verify their low-cycle fatigue characteristics.
5.2. Comparison between low-cycle fatigue model and modied Park and Ang model
A common approach for cumulative damage assessment, particularly for concrete structures,
involves a direct use of the plastic strain energy as a characterizing parameter. Most notably,
the Park and Ang model [3, 36] assumes that damage inicted on a structure is caused by a
linear combination of damage due to the peak response displacement and damage due to the
plastic strain energy dissipation. By setting the damage index of the modied Park and Ang
model to unity to correspond to the ultimate limit state, the design plastic strain energy may
be written as [2]:
E
h
m
=
a
( x
g
)
max
T
2R
+ 1

c
1
(22)
where
is a weighting factor for the damage due to plastic strain energy dissipation and
is called the strength deterioration parameter. It should be noted that Equation (22) is based
on a modied form of the original Park and Ang model where the plastic strain energy
dissipation under monotonic loading has been incorporated into the model. The predictive
form of the plastic strain energy capacity in Equation (22) may similarly be used to develop
the duration-dependent inelastic design spectra.
Figures 4(a)(d) show a comparison of the duration-dependent inelastic design spectra for
the modied Park and Ang model and the low-cycle fatigue model for a range of a=v ratios
from 0:6g=ms
1
to 2:0g=ms
1
. The duration-dependent inelastic design spectra are developed
for a fatigue ductility coecient of c =0:3 and a strength deterioration parameter of
=0:15
TLFeBOOK
(a) (b)
(c) (d)
Figure 4. Comparison of duration-dependent spectra between the low-cycle fatigue model and the
modied Park and Ang model: (a) low a=v ratio = 0:6g=ms
1
; (b) normal a=v ratio = 1:0g=ms
1
;
(c) high a=v ratio = 1:4g=ms
1
; and (d) ultra-high a=v ratio = 2g=ms
1
.
for concrete. The monotonic displacement ductility factor is taken as
m
=9:25, while the
loop shape factor and the design to total plastic strain energy ratio are taken as
1
=0:5 and
=0:7, respectively. Except for the elastic design spectrum in Figures 4(a)(d), the solid line
corresponds to the modied Park and Ang model while the dashed line corresponds to the
low-cycle fatigue model. It can be seen from the gures that both models predict an increase
in the base shear coecient for an increased duration. The increase in base shear coecient,
however, is dierent for the two models, with generally a larger value of C
y
predicted for the
modied Park and Ang model. The following numerical values are used to compare the two
models. For the low a=v ratio of 0:6g=ms
1
in Figure 4(a) and for a short duration of t
d
=10 s,
the base shear coecient at a period of T =0:31 s is C
y
=0:293 for the modied Park and
Ang model, which is only slightly larger than the base shear coecient of C
y
=0:279 for the
low-cycle fatigue model. At a longer duration of t
d
=60 s, however, the base shear coecient
TLFeBOOK
94 Y. H. CHAI
for the modied Park and Ang model increases to C
y
=0:355, but is smaller than the base
shear coecient of C
y
=0:381 for the low-cycle fatigue model. For the ultra-high a=v ratio
of 2:0g=ms
1
in Figure 4(d) and for a short duration of t
d
=10 s, the base shear coecient
at a period of T =0:31 s is C
y
=0:185 for the modied Park and Ang model, which is only
very slightly higher than the base shear coecient of C
y
=0:182 for the low-cycle fatigue
model. For a duration of t
d
=60 s, however, the base shear coecient increased to C
y
=0:362
for the modied Park and Ang model, compared to the relatively small base shear coecient
of C
y
=0:256 for the low-cycle fatigue model. The generally larger base shear coecient for
the modied Park and Ang model is due to the smaller plastic strain energy predicted by the
modied Park and Ang model compared to the low-cycle fatigue model, and the duration-
dependent spectra based on the modied Park and Ang model appear to be more sensitive to
the a=v ratio than the low-cycle fatigue model.
6. CONCLUSIONS
In this paper, a previously developed procedure for generating duration-dependent inelastic
design spectra is extended to the commonly used low-cycle fatigue model. The approach is
based on the premise that, in order to ensure satisfactory performance of a structure, the cyclic
plastic strain energy capacity of the structure must be larger than or equal to the portion of
seismic input energy contributing to cumulative damage. The seismic input energy, prescribed
in terms of a bilinear equivalent velocity spectrum, is used as the seismic demand, while
the plastic strain energy capacity is estimated from a low-cycle fatigue model. The resulting
design spectrum, which depends on the duration of the ground motion, indicates that the lateral
strength of a structure must be increased in order to compensate for the increased damage due
to an increased number of inelastic cycles expected during a long-duration ground motion.
To illustrate the procedure as well as demonstrate the sensitivity of the spectrum to various
parameters, duration-dependent inelastic design spectra are generated using currently available
parameters for steel and concrete for ground motions characterized by low and high a=v
ratios. Results indicate that, although a higher lateral strength is required for an increased
duration, the increase is smaller for concrete than for steel, and the lower lateral strength of
concrete is consistently observed for both low and high a=v ratios. The smaller increase in
the lateral strength is likely due to the small value of the fatigue ductility exponent assumed
for concrete, which is taken as c = 0:3 in this paper. The small value of the exponent
implies that the plastic strain energy capacity of concrete is larger than that of steel. The
larger plastic strain energy capacity translates into the greater ability of concrete to dissipate
seismic energy without the need to increase the lateral strength in order to compensate for
the cumulative damage. Testing of steel reinforcement under large amplitude cyclic loading,
however, suggested that the low-cycle fatigue behavior of reinforcing bars may be better
characterized by an exponent of c = 0:5. This implies that for reinforced concrete columns
governed by low-cycle fatigue failure of its reinforcement, a fatigue ductility exponent of
c = 0:5 may be more appropriate. Since the overall accuracy of the procedure depends on a
proper selection of these parameters, it is important that further tests be conducted on concrete
and steel members in order to verify their low-cycle fatigue characteristics.
The duration-dependent inelastic design spectra developed using the low-cycle fatigue model
are also compared with a dierent cumulative damage model; in this case, the modied Park
TLFeBOOK
and Ang model. Although both models provide a similar lateral strength demand in the short-
duration range, say a duration of 10 s, the dierence between the two models increases with a
longer duration. The lateral strength is generally larger for the modied Park and Ang model,
and is likely due to the smaller plastic strain energy predicted by the modied Park and
Ang model. The spectra based on the modied Park and Ang model also appear to be more
sensitive to the a=v ratio than the low-cycle fatigue model.
REFERENCES
1. FEMA 223. NEHRP Recommended Provisions for the Development of Seismic Regulations for New
BuildingsPart 2. Commentary (1991 Edition). Building Seismic Safety Council, Washington D.C.
2. Chai YH, Fajfar P, Romstad KM. Formulation of duration-dependent inelastic seismic design spectrum. Journal
of Structural Engineering (ASCE) 1998; 124(8):913921.
3. Park YJ, Ang AHS. Mechanistic seismic damage model for reinforced concrete. Journal of the Structural
Division (ASCE) 1985; 111(4):722739.
4. Zahrah TF, Hall WJ. Earthquake energy absorption in SDOF structures. Journal of Structural Engineering
(ASCE) 1984; 110(8):17571772.
5. Akiyama H. Earthquake-Resistant Limit State Design for Buildings. University of Tokyo Press, 1985.
6. Housner GW. Limit design of structures to resist earthquakes. In Selected Earthquake Engineering Papers of
George W. Housner, ASCE, 1990, pp. 186198.
7. Fajfar P, Fischinger M. Earthquake design spectra considering duration of ground motion. In Proceedings of
the Fourth U.S. National Conference on Earthquake Engineering, Vol. 2, Palm Springs, California, 2024
May, 1990, pp. 1524.
8. Leelataviwat S, Goel SC, Stojadinovic B. Toward performance-based seismic design of structures. Earthquake
Spectra 1999; 15(3):435462.
9. Chapman MC. On the use of elastic input energy for seismic hazard analysis. Earthquake Spectra 1999;
15(4):607635.
10. Manfredi G, Polese M, Cosenza E. Cumulative demand of the earthquake ground motions in the near source.
Earthquake Engineering and Structural Dynamics 2003; 32:18531865. DOI: 10.1002=eqe.305.
11. Dutta A, Mander JB. Energy based methodology for ductile design of concrete columns. Journal of Structural
Engineering (ASCE) 2001; 127(12):13741381.
12. Uang C-M, Bertero VV. Use of Energy as a Design Criterion in Earthquake-Resistant Design. UC Berkeley
EERC Report 88=18, 1988.
13. Kuwamura H, Galambos TV. Earthquake load for structural reliability. Journal of the Structural Division
(ASCE) 1989; 115(6):14461462.
14. Vidic T, Fajfar P, Fischinger M. Consistent inelastic design spectra: Strength and displacement. Earthquake
Engineering and Structural Dynamics 1994; 23:507521.
15. Kuwamura H, Kirino Y, Akiyama H. Prediction of earthquake energy input from smoothed Fourier amplitude
spectrum. Earthquake Engineering and Structural Dynamics 1994; 23:11251137.
16. Chai YH, Fajfar P. A procedure for estimating input energy spectra for seismic design. Journal of Earthquake
Engineering 2000; 4(4):539561.
17. Trifunac MD, Brady AG. A study on the duration of strong earthquake ground motion. Bulletin of the
Seismological Society of America 1975; 65(3):581626.
18. Fajfar P, Vidic T. Consistent inelastic design spectra: Hysteretic and input energy. Earthquake Engineering and
19. Kunnath SK, Chai YH. Cumulative damage-based inelastic cyclic demand spectrum. Earthquake Engineering
and Structural Dynamics 2004; 33:499520. DOI: 10.1002=eqe.363.
20. Chai YH, Romstad KM. Correlation between strain-based low-cycle fatigue and energy-based linear damage
models. Earthquake Spectra 1997; 13(2):191209.
21. Collins JA. Failure of Materials in Mechanical Design: Analysis, Prediction, Prevention (2nd Edn). Wiley:
New York, 1992.
22. El-Bahy A, Kunnath SK, Stone WC, Taylor AW. Cumulative seismic damage of circular bridge columns:
Benchmark and low-cycle fatigue tests. ACI Structural Journal 1999; 96(4):633641.
23. El-Bahy A, Kunnath SK, Stone WC, Taylor AW. Cumulative seismic damage of circular bridge columns:
Variable amplitude tests. ACI Structural Journal 1999; 96(5):711719.
24. Kasiraj I, Yao JTP. Fatigue damage in seismic structures. Journal of the Structural Division (ASCE) 1969;
95(8):16731692.
25. Suidan MT, Eubanks RA. Cumulative fatigue damage in seismic structures. Journal of the Structural Division
(ASCE) 1973; 99(5):927943.
TLFeBOOK
96 Y. H. CHAI
26. McCabe SL, Hall WJ. Assessment of seismic structural damage. Journal of Structural Engineering (ASCE)
1989; 115(9):21662183.
27. Krawinkler H. Performance assessment of steel components. Earthquake Spectra 1987; 3(1):2741.
28. Krawinkler H, Nasser A. Damage potential of earthquake ground motions. In Proceedings of the Fourth
U.S. National Conference on Earthquake Engineering, Vol. 2, Palm Springs, California, 2024 May, 1990,
pp. 945964.
29. B aumel A Jr., Seeger T. Materials Data for Cyclic LoadingSupplement 1. Elsevier Science, 1990.
30. Suresh S. Fatigue of Materials. Cambridge University Press, 1991.
31. Chopra AK. Dynamics of Structures: Theory and Applications to Earthquake Engineering (2nd Edn). Prentice-
Hall, 2000.
32. Veletsos AK, Newmark NM. Eect of inelastic behavior on the response of simple systems to earthquake
motions. In Proceedings of the Second World Conference on Earthquake Engineering, Vol. 2, Japan, 1960,
pp. 895512.
33. Fajfar P. Capacity spectrum method based on inelastic demand spectra. Earthquake Engineering and Structural
Dynamics 1999; 28:979993.
34. Paulay T, Priestley MJN. Seismic Design of Reinforced Concrete and Masonry Buildings. Wiley Interscience,
1992.
35. Mander JB, Panthaki FD, Kasalanati A. Low cycle fatigue behavior of reinforcing steel. Journal of Materials
in Civil Engineering (ASCE) 1994; 6(4):453468.
36. Park YJ, Ang AHS, Wen YK. Seismic damage analysis of reinforced concrete buildings. Journal of the
Structural Division (ASCE) 1985; 111(4):740757.
TLFeBOOK
Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/eqe.427
DISCUSSION OF PAPER
An improved capacity spectrum method for ATC-40 by
Y.-Y. Lin and K.-C. Chang, Earthquake Engineering and
Structural Dynamics 2003; 32(13):20132026
Anil K. Chopra
;
Department of Civil and Environmental Engineering; University of California; Berkeley; CA 94720-1716; U.S.A.
The above-referenced paper recommends using the true-acceleration spectrum instead of the
pseudo-acceleration spectrum to dene the demand diagram in the capacity spectrum method
(CSM). Although this change may improve the approximate resultsas demonstrated in the
paperin my opinion it lacks a theoretical basis.
To provide a basis for this assertion, consider a linearly elastic SDF system of mass m and
stiness k. As is well known, the peak value of the equivalent-static lateral force f
So
(and
base shear V
bo
) is
f
So
=kD=mA
where D and A are the ordinates of the deformation and pseudo-acceleration response (or
design) spectra, respectively.
The equivalent static lateral force (and base shear) used to compute design member forces
and deformations in the structure is not equal to m times the true acceleration u
t
o
(except for
undamped systems) because, in general, this includes a velocity-dependent damping force. It
is inappropriate to include the damping force in computing element stresses because they are
to be compared with allowable stresses that are specied based on static tests of materials.
Returning to the CSM in the paper, the demand diagram (spectrum) is based on true
acceleration (thus including the damping force), but the capacity diagram (spectrum) is based
on static properties of the structure. These dierent choices are theoretically inconsistent. If
the choice of the true-acceleration spectrum is justied in order to compensate for the errors
in the parameters of the equivalent linear system, conceptually, it is inappropriate to introduce
a second error to compensate for the rst.
Correspondence to: Anil K. Chopra, Department of Civil and Environmental Engineering, University of California,
Berkeley, CA 94720-1716, U.S.A.
Copyright
?
2005 John Wiley & Sons, Ltd.
TLFeBOOK
Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/eqe.428
AUTHORS REPLY
Authors reply to discussion by Anil K. Chopra of
An improved capacity spectrum method for ATC-40
Earthquake Engineering and Structural Dynamics
2003; 32(13):20132026
Yu-Yuan Lin
;
Department of Civil and Water Resources Engineering; National Chiayi University; Taiwan
Yes, the demand diagram and the capacity diagram in the paper are not consistent. In order to
reduce the errors resulting from the parameters of the equivalent linear system, we introduce
the damping force into the demand diagram.
The reason for doing so is that we found that the period of the equivalent linear system
corresponding to the true acceleration response spectrum (Sa) is always longer than that
corresponding to the pseudo-acceleration response spectrum (PSa). Figure 1 shows the mean
Sa and PSa derived from 1053 earthquakes for a viscous damping ratio of 30%. For a given
level of acceleration, it is observed that the Sa always corresponds to an equivalent linear
system with longer periods than that for the PSa. For example, the periods corresponding
to the acceleration of 0:4g are 1:54 sec for Sa and 1:21 sec for PSa. Because longer period
systems generally result in larger displacement response (Figure 2), we suggest using Sa as
a modication factor to improve the accuracy of the capacity spectrum method in ATC-40.
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
3.0 2.5 2.0 1.5 1.0 0.5 0.0
Period (sec)
S
a

o
r

P
S
a

(
g
)

Sa
PSa
30% Damping
1.21s 1.54s
0.4g
0
0.2
0.4
0.6
0.8
1
1.2
1.4
0 5 10 15 20 25 30 35 40
Sd (cm)
S
a

o
r

P
S
a

/

W

Sa vs. Sd
PSa vs. Sd
18.1cm
30% Damping
T
eq
=1.54s
T
eq
=1.21s
T
n
= 0.78s
14.4cm
Figure 1. Mean acceleration response spec-
trum derived from 1053 earthquakes.
Figure 2. Capacity spectrum method.
Correspondence to: Yu-Yuan Lin, Department of Civil and Water Resources Engineering, National Chiayi
University, Taiwan.
Copyright
?
2005 John Wiley & Sons, Ltd.
TLFeBOOK

Earthquake Engineering & Structural Dynamics

Uploaded by

Copyright:

Available Formats

Earthquake Engineering & Structural Dynamics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Earthquake Engineering & Structural Dynamics

Uploaded by

Copyright:

Available Formats

EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2005; 34:120

is the curvature ductility requirement as deduced from seismic

=2.25 1.22 =2.75 and j

=2.00 1.65 =3.30

Modes normalized by response at the top.

= 133 m/sec, = 0.38

(it should be noted that the nominal North direction is 31.4

to the west of the

W) directions presumed that the structure, soil

Figure 4. Lumped mass model for derivation of identication approach.

out of phase with respect to F

4.52 4.61 4.61 4.14 4.16 4.16

4.61 4.60 4.60 4.20 4.16 4.15

10.99 11.08 11.44

SEK AND P. FAJFAR

SEK AND P. FAJFAR

SEK AND P. FAJFAR

SEK AND P. FAJFAR

SEK AND P. FAJFAR

of the SDOF system (see Equation (2)) is

. The yield acceleration S

(Equation (4)) is equal to 2.21.

=2:21 is then larger than R(

SEK AND P. FAJFAR

(s) 0.29 Dyn. 3.0 11.4 9.5

SEK AND P. FAJFAR

(s) 0.38 Dyn. 3.3 8.0 7.2

, i.e. the capacity

SEK AND P. FAJFAR

SEK AND P. FAJFAR

k, then the derivative of with respect

0:75 for bilinear; non-degrading systems

. In this case, the

to arrive at the inelastic design spectrum. Examples of duration-dependent inelas-

You might also like