JEE 2002 Vol 5 Issue 1

December 6, 2001 14:43 WSPC/124-JEE
00050
Journal of Earthquake Engineering, Vol. 6, No. 1 (2002) 115

c Imperial College Press

DAMAGE DIRECTIVITY IN BURIED PIPELINES OF

KOBE CITY DURING THE 1995 EARTHQUAKE
SHIRO TAKADA and NEMAT HASSANI

Civil Engineering Department, Kobe University, Japan
KATSUMI FUKUDA
Department of Science & Technology,
Kobe University, Japan
Received 7 July 2000
Revised 12 January 2001
Accepted 27 March 2001
In this paper, we analysed the damage ratio in water distribution, gas and telecommunication pipelines buried normal and parallel to the fault direction using GIS database.
We attempt to simulate strong ground motions in these two directions at near-field considering the S-wave radiation pattern. We have also analysed and compared the buried
pipes behaviour located normal or parallel to the fault direction in the ground with
horizontal or inclined bedrock, using a three-dimensional finite element approach. Both
statistical and analytical results show that response and damage ratio of the pipes buried
normal to the fault direction are larger than those in parallel direction. It is also shown
that the effect of seismic directivity at near field is large, while at far field, the effect
due to inclination of the bedrock on the ratio of pipe damage in these two directions is
predominant.
Keywords: Damage directivity; pipeline damage; wave radiation pattern; seismic
directivity; bedrock inclination.
1. Introduction
Lifelines suffered extensive damage in the 1995 Hyogoken Nambu Earthquake and
the role of urban lifelines were much more closed up. Damage to the water and
gas distribution pipes in Kobe City, which is located near to the source fault,
happened in thousands of locations. The analyses of the recorded accelerations
show that the ground displacement and velocity were larger in normal direction
to the fault comparing to those in parallel [CIHAE, 1998; Mori, 1998]. There is
also a special characteristic in this city and that is the inclination of bedrock
from the mountainside to the seaside [CIHAE, 1998; Mori, 1998] (north to south).
Professor
Assistant
Graduate
Professor
Student
1
00050
S. Takada, N. Hassani & K. Fukuda
Therefore, these two characteristics are supposed to have effective role on damage directivity in buried pipelines during the earthquake in Kobe City. The GIS
based statistical analysis of data collected from the performance of buried pipes
in water, gas and telecommunication networks in this city showed that damage
in those pipes which were buried normal to the faults direction was considerably more than those in the parallel direction. Also for studying the effect of
bedrock inclination on damage directivity, the generated earthquake waves, considering the radiation pattern of S-wave are used in the FEM analyses of pipeground system with and without bedrock inclination using the ABAQUS computer
code.
2. Relation Between Damage and Burial Direction of Pipelines
The damage ratios of water, gas and telecommunication pipelines in two directions of parallel and normal to the fault were statistically analyzed using
the GIS medium [Takada et al., 1999a; 1999b]. The four faults used for the
present analyses were not the main ones having a rapture origin. However surface
fault displacements along these faults were recognized by SAR analysis [Fujiwara
et al., 2000]. Here, the method that we have used for determining angles between pipes and fault is explained. Assuming an axis for the fault and using an
angle-step of 30 , the pipe directions have been divided into six parts as shown in
Fig. 1.
Then considering the end points of each pipe in five wards of the city, their
angles with the fault direction are calculated. In more general division the pipes in
directions 3 and 4 are assumed normal to the fault while the ones in directions
1 and 6 have been considered as parallel to the fault. Using such a classification, the pipe length, damage number and ratios have been determined for each
divided part, which we named them directions 1 to 6 to avoid complicated description in the paper. We have showed the directivity characteristics by concentrating

Fig. 1.
Burial directions of the pipes.
00050
Damage Directivity in Buried Pipelines of Kobe City
on directions 3 (as normal to the fault) and 6 (as parallel to the fault) [Fukuda,
2000].
2.1. Water distribution pipelines
The water distribution pipes in the five wards of Nada, Higashinada, Chuo, Hyogo
and Suma which are located close to the four faults of Gosukebashi, Suwayama,
Egeyama and Suma in Kobe City are selected for the analysis. The materials of
pipelines are ductile iron with different joint type in more than 90% of total length in
Kobe City. Those pipes in the back fill lands which suffered more from liquefaction
than being near to the fault are excluded in this analysis. Comparisons of directions
3 and 6 in different wards are given in Table 1 and Fig. 2.
The results show that the damage ratio in direction 3, which is normal to the
fault direction, is more than that in the area 6, which is parallel to the fault.
In those wards such as Nada, Higashinada and Chuo where the damage ratio is
large, the difference between the ratios in two directions is also large, while in the
other areas, there is not so big differences between the ratios in two directions.
Table 1. Comparison among damage condition of water distribution pipes in different wards of
Kobe City.
The wards in
Kobe City
Length of buried pipes (km)
Damage ratio
(location/km)
Damage locations
Direction 3
Direction 6
Direction 3
Direction 6
Direction 3
Direction 6
152.05
228.83
150
166
0.99
0.73
Chuo
74.92
107.85
61
44
0.81
0.41
Hyogo
51.13
60.93
28
23
0.55
0.38
Suma
46.24
59.42
22
26
0.48
0.44
Damage ratio in water distribution

pipes(location/km)
Nada and
Higashinada
Fig. 2.
1.20
Direction 3
Direction 6
1.00
0.80
0.60
0.40
0.20
0.00
Nada and
Chuo Ward
Higashinada Wards Suwayama Fault
Gosukebashi Fault
Hyogo Ward
Egeyama Fault
Suma Ward
Suma Fault
Comparison of damage ratios in water distribution pipes for different burial direction.
00050
Therefore, it is concluded that in those areas with big differences in damage ratios
for two directions, the earthquake motion in the normal direction to the fault was
predominant.
2.2. Gas pipes
Considering the reliability degree of damage data of gas pipes in three wards of
Chuo, Nada and Higashinada the total damage location is decreased somewhat.
The same method of analysis in water distribution pipes, is used for finding the
relation between damage ratio and pipe directions against the Suwayama fault.
The length, damage location and damage ratio of gas pipes, are given for each
direction in Table 2. The materials of gas pipelines are mostly ductile iron and
steel pipe with screw joint.
The comparison between the damage ratios in directivity areas 3 and 6 are given
in Fig. 3. The damage directivity analysis was also done for the different types of
pipes. The results show that in the gas pipes damage ratio for the pipes in direction
Table 2. Comparison of damage condition of gas pipes in
different directions.
Direction
Buried length
(km)
Damage
location
Damage ratio
(location/km)
201.89
33
0.16
228.67
55
0.24
813.05
292
0.36
143.82
34
0.24
231.12
35
0.15
1147.95
266
0.23
Fig. 3.
Comparison of damage ratio for different directions.
00050
3, is more than that in direction 6 (considering total whole lines). This difference is
much greater (about 1.7 times) for the pipes with screw joints (considering damage
ratio in pipelines with screw joints).
2.3. Telecommunication pipelines
In a damage investigation, NTT performed the Street Mandrel Test for two vacant
conduits among several ones in each span between manholes. The judgement on
damage was done whether the Mandrel passes smoothly or not. The results showed
that the interruption rate has increased to 36% which is bigger than usual rate by
6% [CIHAE, 1998]. For telecommunication analysis, the damage data of pipelines
in Nada and Higashinada (the Nada, Mikage and Higashinada blocks) were used
against the Gosukebashi fault. Assuming 3 number interruption of Mandrel as the
damage to the pipeline, the interruption rate is calculated for whole length of the
pipeline. Figure 4 shows the comparison between the interruption rate in direction
3 and 6. It is shown that the damage ratio in normal direction to the fault is
somewhat greater than in the parallel direction without Mikage Block.
However, comparing to the water and gas pipes, there is different tendency for
damage difference in block with less damage. Due to this reason, it is supposed that
there are some other causes for damage except earthquake that is necessary to be
included in the analysis of data. One of the main reason for interruption of Mandrel
is the influence of rust in the pipe accumulated during the past time.
One of the reasons for more damage in pipes buried in normal direction to the
fault is the effect of the predominant ground motion in this direction. Generally
speaking, the behavior of buried pipelines follows the ground deformation and the
longitudinal behavior is much predominant compared with perpendicular one due to
soil stiffness in each direction. Since the motion in the normal direction to the fault
is predominant, the longitudinal strain in those pipes which are buried in this
direction is also predominant. Due to the pipe longitudinal severe compression and
tension, there is a lot of damage to the joints (such as pulling-out) and to the pipebody (such as buckling) both for water and gas buried pipes. However, since the
Fig. 4.
Comparison of interruption percentage in different areas.
00050
pipelines in Kobe City are buried also parallel to the bedrock inclination from the
mountains in the north to the sea in the south, there is possibility of topography
influence on the pipe damage in this direction. Therefore, the effect of bedrock
inclination is also investigated in this paper.
3. Introduction of Function of S-Wave Radiation Pattern

For analysis of damage directivity in pipelines, the earthquake waves are generated
through a statistical synthesis method. In order to make waves for small earthquakes (due to incremental fracture of the fault), the random phase is applied
to the spectrum model of the source fault using the Boor Statistical Simulation
Method [Mori, 1998; Boore, 1983]. The large earthquake wave is generated through
the summation of these small waves. It is not necessary to estimate the path of wave
propagation or ground characteristics in this method. Although the advantage of
empirical Green Function is not used very well, the best use of the propagation of
fault surface rupture and geometrical relation among fault and any arbitrary point
is the advantages of the method. By introducing the Radiation Pattern Function
in the relation between the location of epicenter and observation point, the same
earthquake is obtained in any direction at observation point. In this way, after
making ground motion for each short path in the fault, the radiation pattern of
displacement amplitude directivity is used and then the main earthquakes in normal and parallel direction to the fault are generated by summation of the small
quakes.
Hirasawa and Stauder [1965] have considered the characteristics of rupture propagation in a single direction of fault with vertical-horizontal displacements and
calculated the radiation pattern of the S-waves. Here, by assuming the unilateral
epicenter line of a vertical-horizontal fault as the fault rupture direction, the radiation pattern of P - and S-waves are calculated in any arbitrary point. If the angle
between the rupture propagation direction and the line to the observation point is
Fig. 5.
Radiation pattern of S-wave.
00050
, the radiation pattern (Fig. 5) for S-wave is obtained through Eq. (1).
R =
cos 2
Vp /Vs
.
Vr
1
cos
Vs
(1)
Here, R is the Radiation Pattern Function of the S-wave, Vr ; rupture propagation

velocity, Vp and Vs ; velocities of P - and S-waves, respectively and R is the distance
between the epicenter and observation point. As the analytical parameters it is
assumed that Vs = 3.0 km/s, Vr = 2.7 km/s, Vr /Vs = 0.9, Vp /Vs = 1.5 and
Vp = 4.5 km/s [Hirasawa and Stauder, 1965].
4. Simulated Earthquake Motions using Radiation Pattern

The resulted waves by the statistical synthesis method in an observation point
1.0 km far from the epicenter for any direction without considering the radiation
pattern is shown in Fig. 6.
Acceleration(gal)
1500
The same wave in any direction
1000
500
0
-500
-1000
0
5
6
time(sec)
Example of statically synthesised wave at ground surface (distance from the fault 1.0 km).
Acceleration(gal)
Fig. 6.
Parallel to the fault

Normal to the fault
1500
1000
500
0
-500
-1000
-1500
0
Time(sec)
Fig. 7.
Wave at the ground surface using radiation pattern (distance from the fault is 1.0 km).
00050
By introducing the Radiation Pattern Function, it became possible to estimate

the wave components in normal and parallel directions to the fault direction separately as shown in Fig. 7. Here, the distance from the fault as shown in Fig. 8,
is measured from the end point of fault rupture to the observation point in the
normal direction to the fault. Figure 7 shows that for 1.0 km far from the fault,
the maximum acceleration in normal direction is 1.5 times larger than the one in
parallel direction to the fault. The relation between the maximum acceleration at
ground surface and the distance from the fault is depicted in Fig. 9 for ground
with horizontal bedrock. This figure clearly shows that the maximum acceleration
at ground surface in normal direction is much greater than the parallel one and the
former one is 1.55.2 times larger than the latter one.
Comparison among the results with those from the attenuation formula of
Fukushima and Tanaka [1990] shows that the maximum acceleration in normal
direction at a distance of 10.0 km to the fault is more than that obtained by the
formula. While in the parallel direction, by increasing the distance from the fault, it
decreases very fast, and for a distance of 2.0 km, it becomes less than that calculated
by the attenuation formula. The displacement hysteric paths are shown in Fig. 10
for different distances from the fault. The displacement amplitudes in the normal direction are also larger than the ones in parallel direction to the fault. Figure 10 also
shows that the displacement amplitudes for normal to parallel direction increases
by increasing the distance from the fault. The next is determination of ground motion at underground bedrock. Figure 11 shows the velocity response spectra for the
M=7.0
L=40km
W=20km
Source fault
Distance from the fault

Observation point
Fig. 8.
Relation between presumption point and its distance from the fault.
Maximum acceleration at ground

surface(gal)
2000
Normal to the fault

Fukushima-Tanaka formula
1000
500
100
1
10
distance from the fault(km)
Fig. 9.
Relation between maximum acceleration at ground surface and distance from the fault.
00050
(a) Distance from the fault: 1.0 km
(b) Distance from the fault: 2.0 km
(c) Distance from the fault: 3.0 km
(d) Distance from the fault: 4.0 km
Fig. 10.
Path of a ground particle at different distances from the fault.
200
2km
300
4km
Sv (cm/sec)
300
Sv (cm/sec)
6km
100
8km
50
10km
Water
Water
guideline 90% guideline 70%
Non-excess
Non-excess
probability probability
10
1
0.1
0.5
200
100
2km
4km
6km
Water guideline90%
Non-excessprobability
8km
50
10km
10
1
0.1
probability 70%
Non-excess probability
0.5
Period (sec)
(a) Normal to the fault

Fig. 11.
Period (sec)
(b) Parallel to the fault
Velocity spectra obtained from numerically generated wave.
10
00050

300
300
200
Water guideline 90%

200
Sv (cm/sec)
Sv (cm/sec)
Water guideline 90%

Water guideline 70%

100
100
50
Water guideline 70%

10
Higashi-Ohashi.y(GL-33m)
Port Island.y(GL-83m)
50
10
Higashi-Ohashi.y(GL-33m)
Port Island.y(GL-83m)
Kansai Electric Co.

Research Center.y(GL-100m)
1
0.1
0.5
Kansai Electric Co.

Research Center.y(GL-100m)
1
0.1
0.5
(a) Normal to the fault

Fig. 12.
Period (sec)
Period (sec)
(b) Parallel to the fault
Velocity spectra from the waves recorded during Kobe Earthquake.
generated earthquakes in normal and parallel directions to the fault. The two bold
lines in the figure are the design velocity spectra, which are proposed in the Recommended Anti-Seismic Construction of Water Facilities (revised 1997) [JWWA,
1997] for design of underground structures against level 2 earthquakes in Japan.
These lines are corresponded to 90% and 70% non-excess probability. This figure
shows that the spectrum characteristics are clearly different for normal and parallel
directions to the fault and the one in the normal direction is much larger than that
in the parallel direction. At a distance of 6.0 km in the normal direction to the fault
and for periods more than 1.0 second, the velocity spectrum exceeds the 100 kine
(given for 90% non-excess probability), which shows the importance of considering
directivity in velocity spectra for seismic design of buried structures. This is also
clear from the Fig. 12 which depicts velocity spectra for two directions obtained
from recorded waves during Kobe Earthquake.
5. Performance Characteristics of Buried Pipes Network in
Different Distances from the Fault
5.1. Outlines and conditions of analyses
In this part, the generated waves by the Statistical Synthesis Method using Radiation Pattern Function are used for increment distances from the fault as input
waves for response analysis of a 3D model of buried pipes in two directions using the
ABAQUS code. The generated waves in two directions are applied in two orthogonal directions, normal and parallel to the faults, for comparison and investigation
of pipe seismic responses. The rupture length and width of the fault corresponded
to a magnitude 7.0 are obtained as 40.0 km and 20.0 km empirically. The waves at
the bedrock are generated for every 1.0 up to 10.0 km from the fault and applied
in direction 1 (parallel to the fault), direction 2 (normal to the fault) and both
directions simultaneously. The input earthquakes for distances 1.0 km and 2.0 km
are given in Fig. 13 as examples.
00050

Normal to the fault
600
Acceleration(gal)
Acceleration(gal)

Normal to the fault
600
Maximum acceleration
405.0gal
400
200
0
-200
139gal
400
200
0
-200
-400
-400
604.4gal
-600
0
(a) Distance from the fault: 1.0 km

Fig. 13.
Maxmum acceleration
574gal
-600
11
5
Time(sec)
(b) Distance from the fault: 2.0 km
Example of input ground motions.
100m
Normal to the
fault buried pipe
100m
100m
100m
Normal to the
fault buried pipe
Parallel to the
fault buried pipe
25m
Burial
depth
Direction 3
Burial
1.5mdepth
1.5m
Direction 1
iNormal to the fault j
5m
Parallel to the
fault buried pipe
Burial
depth
20m
4m
1.5m
1.5m
Direction 3
Direction 1
(Normal to the fault)
20m
20m
20m
20m
20m
Direction 2
iParallel to the fault j
Direction 2
Parallel to the fault j
(a) Case 1
(b) Cases 2 and 3

Fig. 14.
Model for analysis.
Table 3.
Cases of analyses.
Case
Ground
Input ground motion
Case 1
Uniform
Direction 1: Normal to the fault , Direction 2: Parallel to the fault
Case 2
Inclined bedrock
Normal one was applied in directions 1 and 2
Case 3
Inclined bedrock
Direction 1: Normal to the fault, Direction 2: Parallel to the fault
There are three cases of analyses, which are given in Table 3. The models of
the cases are also shown in Fig. 14. The length, width and depth of the models are
100.0, 100.0 and 25.0 m, respectively. The N -value is 10 and there is 45 bedrock
inclination in the model for Cases 2 and 3. The MohrCoulomb criterion is used
for soil failure. The bottom of the models is fixed while the other boundaries are
viscous. The pipe is a 150 mm welded steel with bilinear characteristics buried
in a depth of 1.5 m and simulated as a beam element in both directions.
12
00050
Maximum strain in pipe axial direction
0.3
Normal to the fault

0.25
0.2
0.15
0.1
0.05
0
3
4
5
6
7
8
Distance from the fault (km)
10
Fig. 15. Relation between the maximum axial strain in the pipes and distance from the fault
(Case 1).
Strain in normal direction /

Strain in parallel direction
1.5
1.4
1.3
1.2
1.1
1
0
10
Distance from the fault(km)
Fig. 16.
Distance from the fault and ratio of axial strains of pipe (Case 1).
5.2. Results of the analyses and discussion

The relation between the pipes maximum strain and the distance from the fault
for the Case 1 is shown in Fig. 15. According to this figure, up to 7.0 km from
the fault, the strains in normal direction, are larger than those in parallel direction. The relation between the maximum strain ratio of these two directions
and the distance from the fault is given in Fig 16. The ratio of strains in normal
to parallel direction in this figure for 1.0 to 7.0 km distances from the fault is
about 1.2 to 1.3 times, while for the distances more than 8.0 km, it decreases to
1.1 times.
The similar results are given in Fig. 17 for Cases 2 and 3. This figure also
shows that the maximum strain of pipe buried in normal direction are greater
than those buried in parallel. The ratio of maximum strains of normal to parallel
direction is shown in Fig. 18 for Cases 2 and 3. Figure 19 shows the influence ratio
00050
13
Case2 Normal to the fault
Maximum axial strain of pipes (%)
Case2 Parallel to the fault
0.5
Case3 Normal to the fault
0.45
Case3 Parallel to the fault
0.4
0.35
0.3
fault
0.25
0.2
0.15
0.1
0.05
0
0
10
Strain in normal to the fault/

Strain parallel to the fault
Fig. 17. Relation between the maximum axial strains in the pipes and distance from the fault
(Cases 2 and 3).
3.8
3.6
3.4
3.2
3
2.8
2.6
2.4
2.2
2
1.8
P [
Case2
XP Q[
Case3
XR
0
Fig. 18.
and 3).
2 3 4 5 6 7 8
Distanse from the fault(km)
1
0
Distance from the fault and the ratio of maximum axial strains in the pipes (Cases 2
of bedrock inclination (Case 2) in the total ratio of maximum strain of normal to

parallel direction, which includes the effects of both seismic directivity and bedrock
inclinations (Case 3 in Fig. 18. The strain ratio is different for any distance from
the fault and is larger in normal direction for any distance in Case 3, in which the
directivity is also included.
The results in Figs. 18 and 19 show that at near field of the fault, the directivity
of ground motion (seismic directivity) is considerable. However for the distances
Effect of bedrock inclination on pipe strain ratio

in 2 directions
14
00050
0.9
0.8
0.7
0.6
0.5
0
10
Fig. 19. Distance from the fault and the influence ratio of bedrock inclination on ratio of strain
in pipe axial direction (Case 3).
more than 7.0 km from the fault (far field), there is almost no effect of seismic
directivity and the effect of bedrock inclination becomes predominant.
6. Conclusions
Damage directivity in buried pipelines in Kobe City during the 1995 Hyogoken
Nambu Earthquake are investigated. For investigating the effect of seismic directivity and bedrock inclination effects at near and far fields, two orthogonal pipes
are analyzed in a 3D model of ground. The input waves for normal and parallel
direction are generated using a Statistical Synthesis Method considering Radiation
Pattern Function. The main findings are as follows:
(1) The statistical analyses of the GIS based data in Kobe City show that the
water, gas and telecommunication pipelines, which were buried normal to the
faults suffered much damage comparing to those in the parallel direction.
(2) Comparison among the results with those from the attenuation formula of
FukushimaTanaka shows that the maximum acceleration in normal direction
at a distance of 10.0 km to the fault is more than that obtained by the formula. While in the parallel direction, by increasing the distance from the fault,
it decreases very fast, and for a distance of 2.0 km, it becomes less than that
obtained by the attenuation formula.
(3) The spectrum characteristics are clearly different for normal and parallel directions to the fault and the one in the normal direction is quite larger than that
in the parallel direction, which shows the importance of considering directivity
in velocity spectra for seismic design of buried structures.
00050
15
(4) In the case of horizontal bedrock the ratio of pipe strain in normal to parallel
direction in the near field is not increasing so severely (about 1.2 to 1.3 times)
and for the distances more than 8.0 km decrease to 1.1 times.
(5) At near field of the fault, the directivity of ground motion is very effective.
However for the distances more than 7.0 km from the fault, there is almost no
effect of ground motion directivity and the effect of bedrock inclination is the
predominant one.
References
Boore, D. M. [1983] Stochastic simulation of high-frequency ground motions based
on seismological models of the radiated spectra, Bull. Seism. Soc. Am. 73(6),
18651894.
CIHAE [1998] Committee of Investigation on Hanshin-Awaji Earthquake, Report on
Hanshin-Awaji disaster, JSCE, pp. 5868 (in Japanese).
Fujiwara, S., Ogawa, S., Murakami, M. and Tobita, M. [2000] Estimation of fault positions
of the 1995 Hyogo-ken Nanbu Earthquake using displacement gradient detected by
SAR interferometry, JISIN 53(2), pp. 127136.
Fukuda, K. [2000] Study on dynamic behavior of buried pipelines and aseismic design in the vicinity of active faults, Ph.D. dissertation, Kobe University, pp. 7394
(in Japanese).
Fukushima, Y. and Tanaka, T. [1990] A new attenuation for peak horizontal acceleration
of strong earthquake ground motion in Japan, Bull. Seism. Soc. Am. 80(4), 757783.
Hirasawa, T. and Stauder, W. [1965] On the seismic body waves from a finite moving
source, Bull. Seism. Soc. Am. 55(2), 237262.
JWWA [1997] Japan Water Work Association, Recommendation for anti-seismic construction of water facilities, JWWA, pp. 1718 (in Japanese).
Mori, K. [1998] A study about the strong motion estimation considering ground motion
observation near source fault and using asperity model, Master thesis, Civil Eng.
Dept. Kobe University, pp. 3256 (in Japanese).
Takada, S., Fukuda, K. and Kitada, T. [1999] Damage directivity in buried pipelines,
The 4th Symposium on Urban Inland Earthquakes, pp. 511514 (in Japanese).
Takada, S., Hassani, N., Fukuda, K., and Kitada, T. [1999a] Damage directivity in buried
pipelines and ground motion characteristics at near field, The 25th Conference on
Earthquake Engineering, pp. 10531056 (in Japanese).
00049


ON THE OPTIMAL LOCATION OF SENSORS IN

MULTI-STOREYED BUILDINGS
ALOKE K. DATTA, MANISH SHRIKHANDE and DILIP K. PAUL

Department of Earthquake Engineering,
University of Roorkee, Roorkee247 667, India
mshrifeq@rurkiu.ernet.in
dpaulfeq@rurkiu.ernet.in
Received 23 August 2000
Accepted 30 May 2001
A new approach has been developed for finding optimal location of sensors in 3D multistoreyed buildings. This approach is based on the compact probabilistic representation
of acceleration response in terms of its covariance matrix. For a specified number of
sensors, the optimal location has been taken to be the one for which the computed
covariance matrix is closest to the exact covariance matrix of the random field constituted
by the acceleration response process. It has been found that the determined sensor
locations match favourably with those predicted by earlier studies for the special case of
shear buildings. Further, the optimality of the determined sensor locations has also been
verified by identifying the system parameters from the time series data and comparing
them with those of the Finite Element Model.
Keywords: Optimal sensor location; system identification; structural health monitoring.
1. Introduction
The response of large, complex structural systems to external loads is usually
predicted by using a mathematical model developed with the help of a suitable discretization procedure, e.g. finite element method. The reliability of such
mathematical models can only be assessed by comparing the predicted response
with the response of the prototype structure recorded in field conditions. The
mathematical model, if required, can be refined with the help of structural parameters estimated from field measurements by using system identification techniques [Friswell and Mottershead, 1995]. In addition to calibrating the analytical
model, the identified structural parameters are also very useful in damage assessment or health monitoring, re-evaluating the original design, and for controlling the
vibrations. For this purpose, response has to be recorded at several locations in the
Research
Scholar, on leave from Regional Engineering College, Durgapur, India.

Professor, Corresponding author.
Professor.
Assistant
17
18
00049
A. K. Datta, M. Shrikhande & D. K. Paul
building/structure and it is desirable to install as many sensors as possible. The high

costs of data acquisition systems (accelerometers, recorders, etc.) and accessibility
limitation, however, often put a constraint on the number of sensors. Therefore, it is
important to determine the optimal location of sensor(s) in the building such that
selected locations maximize the information content with respect to the parameter
identification.
Shah and Udwadia first formulated the problem of optimal location of sensors as
a statistical inference problem, wherein a norm of the covariance matrix of the estimated parameters (from the time-domain records) is minimised [Shah and Udwadia,
1978]. The covariance matrix was computed by using a linear relationship between
small perturbations in a finite dimensional representation of system parameters and
finite sample of observations of the system time response. Another approach to solve
the optimal sensor location problem was presented by Udwadia [1994], in which the
optimal sensor locations were defined as those for which a norm of the Fishers
information matrix is maximised. A method for locating sensors in reinforced concrete structures to gain maximum information about the damage distribution has
also been presented on the basis of numerical simulation studies [Skjrbk et al.,
1996]. The method proposed by Udwadia was extended by Heredia-Zavoni and
Esteva [1998] to account for the uncertainty about the structural parameters and
the seismic ground motion excitation within the framework of Bayesian Decision
Theory [Goodwin and Payne, 1977]. The optimal locations were so selected that
the expected Bayesian loss function, expressed in terms of the Fishers information
matrix of the response records, is minimised. Recently, this approach has been further extended to the case of optimal instrumentation of structures on flexible base
[Heredia-Zavoni et al., 1999].
All the above mentioned studies have focussed on the issue of locating sensors
in a structure so as to facilitate the best possible identification of a set of structural
parameters selected a priori. In spite of a rigorous formulation of the optimal sensor location problems, the two approaches followed by Udwadia and Heredia-Zavoni
and his co-workers respectively, provide contradictory results in so far as the effect
of noise on optimal sensor locations is concerned. Moreover, these formulations may
not appeal to a majority of civil engineering community not trained in information
theory. On the other hand, the method proposed by Skjrbk et al. [1996], using
sub-structuring technique, does not guarantee convergence of the iterative procedure. Therefore, we look for a simple and robust technique for solving the optimal
sensor location problem in this study. The excitation and response processes have
been modelled as stationary Gaussian with zero mean. The problem of optimal
sensor locations has been formulated in the form of compact probabilistic representation of random processes similar to the one considered by Masri and Miller [1982].
Two example cases (for different building configurations) have been considered for
illustration of the proposed approach.
00049
On the Optimal Location of Sensors in Multi-Storeyed Buildings
19
2. Formulation
The equation of motion for a multi-degree-of-freedom (MDOF) system subjected
to earthquake excitation can be written as
M
x + Cx + Kx = Mr
xg
(1)
where, M, C and K are the inertia, damping and stiffness matrices of size N N
representing the structural system, x is the N 1 vector of nodal displacements,
r denotes the N 3 matrix of the rigid body influence coefficients, x
g is the 3 1
vector of translational ground acceleration time histories, and a dot () over a
variable indicates the time derivative.
A 3-dimensional finite element model of multi-storeyed buildings is considered
in this study. It is assumed that each node of the the model represents a possible sensor location and the nodal accelerations (translational) correspond to the
sensor records. The ground acceleration process is assumed to be stationary Gaussian with zero mean. Assuming that the structural system responds linearly, the
response process will also be stationary Gaussian with zero mean (neglecting the
effect of start-up transients). Since a Gaussian process is completely described by
the first two spectral moments, it follows that a complete statistical representation
of the zero mean, Gaussian random process representing the response of a linear
system requires an accurate representation of the covariance kernel. Therefore, to
preserve the information content of the response process, its covariance needs to be
reproduced as faithfully as possible. The problem of determining optimal locations
of the sensors can be now recast in the form of determining the degrees of freedom
(DOFs) in the structural system for installation of a given number of sensors, say
m N , which provide the best estimate of the covariance kernel of the response
process.
As a first approximation, let us assume the response process to be a zero mean,
ergodic Gaussian process. The sample realisation being characterised by concatenation of the response of all nodes for increasing the length of the time history
in order to get more stable estimates of the temporal statistics. For example, the
sample realization in the longitudinal direction is considered as the time history
obtained by appending the response in longitudinal direction of a node to the tail
of the response of previous node resulting into one long stretch of response time history in longitudinal direction. Similarly, realisations of the response time history in
other two orthogonal directions can be characterised. The 33 temporal covariance
matrix of these three time histories represents the exact covariance matrix, C of the
response process. Thus the ijth element of the temporal covariance matrix, Cij , is
given by
1 X
(ai (k) i )(aj (k) j ) ;
N 1
N
Cij =
i, j = 1, 2, 3
(2)
k=1
where, ai and aj denote the two orthogonal components of acceleration, N is the

PN
total number of data points in the time history, and i (= N1 k=1 ai (k)) is the
20
00049
temporal mean of the ith component of acceleration. Assuming that m number

of tri-axial sensors
are available, these can be distributed among the N possible
locations in N
m different ways. For each such combination, sample realisations
can be characterised similarly by the concatenation of response of each of the m
selected nodes. The temporal covariance matrix of these realisations characterises
the approximate covariance matrix, C of the response process. The optimal locations
for these m number of sensors will be those for which the difference between some
suitable norm of C and C is minimised. In this study, the goodness of approximation
of C by C is measured by using (i) the determinant norm, (ii) the trace norm, and
(iii) the eigenvalues of these matrices.
The response records from the sensors thus located may be used for the system
identification studies. It will now be shown with the help of illustrative examples
that the response records corresponding to the derived optimal sensor locations
indeed lead to robust estimation of the system parameters.
3. Illustrative Examples
3.0m
3.0m
3.0m
3.0m
3.0m
The optimal sensor locations have been derived for two multi-storeyed reinforced
concrete buildings shown in Figs. 1 and 2. One of these buildings is symmetric
in plan, while the other building, with asymmetry in both plan and elevation, is
typical of constructions on hill slopes. The column cross-section (300 mm 300 mm)
Fig. 1.
3.0m
2.0m
ELEVATION
SIDE VIEW
Schematic diagram of 5-storeyed symmetric building.
00049
21
1.5m
3.0m
1.5m
3.0m
1.5m
3.0m
3.0m
3.0m
5.0m
5.0m
ELEVATION
Fig. 2.
5.0m
4.0m
SIDE VIEW
Schematic diagram of 6-storeyed asymmetric building.
is assumed to be uniform throughout the height for both the buildings, whereas the
beam cross-sections are taken as 300 mm 300 mm and 300 mm 230 mm for the
symmetric and asymmetric buildings, respectively. The slab thickness is taken to
be 150 mm at all levels. The first six modes were used in the mode superposition
for computation of the response of buildings to the earthquake excitations. The
damping has been assumed to be of Rayleigh type with the viscous damping ratio
of 5% in the first and sixth mode. The resulting modal damping for intermediate
modes is approximately 5%, whereas the higher modes with modal damping proportional to the square of the natural frequency are damped out of the response.
An in-house general purpose dynamic finite element analysis code [Kumar 1996]
has been used to compute the acceleration response at each node of the linear finite
element models of these buildings for two different types of earthquake motions,
viz., (i) Uttarkashi Earthquake of 21 Oct., 1991 in Western Himalayas recorded at
Bhatwari, 20 km from the epicentre, shown in Fig. 3, and (ii) Michoacan Earthquake
of 19 Sept., 1985 in Mexico recorded at CDAO station, 400 km from the epicentre,
shown in Fig. 4. While the Uttarkashi Earthquake records are representative of
near-field motion due to a moderate size earthquake, the Michoacan Earthquake
time histories represent the far-field motions due to a large (magnitude 8+) earthquake. No appreciable effect of the flexibility of the floor diaphragm was noticed in
the analysis. Hence the floor diaphragm is assumed to be rigid for all calculations
and the response of any one node on a floor is considered to be representative of
the floor response.
Acceleration (g)
0.30
0.15
0.00
-0.15
-0.30
Acceleration (g)
0.30
0.15
0.00
-0.15
-0.30
N05W
10
15
20
25
Time (s)
30
35
40
Vert
10
15
20
25
30
35
40
Acceleration (g)
Time (s)
0.30
0.15
0.00
-0.15
-0.30
N85E
10
15
20
25
30
35
40
Time (s)
Acceleration (g)
0.08
0.04
0.00
-0.04
-0.08
Acceleration (g)
Fig. 3.
0.08
0.04
0.00
-0.04
-0.08
Acceleration (g)
22
00049
0.08
0.04
0.00
-0.04
-0.08
Time histories of Uttarkashi Earthquake.
N90E
20
40
60
80 100
Time (s)
120
140
160
180
Vert
20
40
60
80 100
Time (s)
120
140
160
180
N00E
20
Fig. 4.
40
60
80 100
Time (s)
120
140
Time histories of Michoacan Earthquake.
160
180
00049
23
For each component of motion, the sample response time history is obtained by
concatenating the response of all the floors together. The covariance matrix computed from these time histories are considered to be the exact representation of
the the zero mean, Gaussian random process characterising the building response.
Further, let us consider for the purpose of illustration, the optimal location of
two tri-axial sensors in these two buildings. The approximate temporal covariance
Table 1. Covariance matrix representation with two sensors (for
Uttarkashi Earthquake, Symmetric Building).
Combination
| C |
Tr[C]
(104 g 6 )
(101 g 2 )
Eigenvalues (101 g 2 )
1
Exact
0.97
1.61
0.94
0.44
0.23
First + Roof
1.28
1.68
0.89
0.51
0.29
Second + Roof
1.99
1.97
1.07
0.57
0.33
Third + Roof
2.57
2.20
1.24
0.63
0.33
Fourth + Roof
3.52
2.48
1.41
0.72
0.35
First + Second
0.12
0.78
0.45
0.21
0.12
First + Third
0.23
1.02
0.62
0.26
0.14
First + Fourth
0.44
1.29
0.79
0.34
0.17
Second + Third
0.47
1.31
0.81
0.32
0.18
Second + Fourth
0.79
1.58
0.98
0.40
0.20
Third + Fourth
1.06
1.82
1.14
0.48
0.19

Michoacan Earthquake, Symmetric Building).
| C |
Tr[C]
(1017 g 6 )
(105 g 2 )
Exact
1.18
2.79
17.22
10.61
0.06
First + Roof
1.25
2.95
18.25
11.19
0.06
Second + Roof
2.01
3.42
21.06
13.04
0.07
Third + Roof
3.47
4.11
25.48
15.53
0.09
Combination
Fourth + Roof
5.56
4.89
30.47
18.34
0.10
First + Second
0.04
0.77
4.52
3.16
0.03
First + Third
0.21
1.46
8.93
5.66
0.04
First + Fourth
0.64
2.25
13.91
8.49
0.05
Second + Third
0.48
1.93
11.74
7.49
0.05
Second + Fourth
1.14
2.71
16.72
10.32
0.07
Third + Fourth
2.18
3.40
21.14
12.81
0.08
24
00049
computed from the concatenated records of these two sensors placed on

matrix, C,
trace (Tr[C]),
and eigenvaldifferent floors. The three norms, viz., determinant (|C|),
ues (i ), of C for different combinations of sensor locations are compared with those
of C in Tables 14 corresponding to the four cases of two buildings excited by two
different earthquake motions. It may be inferred from Tables 1 and 2 that for the
case of symmetric buildings, the records from two sensors located at either the first
floor and roof or at the second floor and fourth floor provide the best approximation for the covariance matrix for both types of excitations. Further, the sensors
located either at the first and second floor, or at the fourth floor and roof present
the two worst sensor placement strategies for this building. It may be noted that in
the case of symmetric building, response records from only two optimally located
sensors are sufficient for a good estimation of the covariance matrix of the response
process. The maximum error in all the three norms of the covariance matrix is less
than 6% for the case of best location of sensors in the symmetric building excited
by Michoacan Earthquake motions. However, the maximum error is in the range of
5% to 20% in the case of Uttarkashi Earthquake. This wide variation in the errors of
approximation could be due to the short duration of Uttarkashi earthquake motion,
and presence of noise in Uttarkashi Earthquake records, which were derived from
analog traces [Chandrasekaran and Das, 1991]. These results for the identification
of optimal locations of sensors are in good agreement with the earlier published

Uttarkashi Earthquake, Asymmetric Building).
| C |
Tr[C]
(106 g 6 )
(101 g 2 )
Exact
0.84
0.59
3.53
2.24
0.10
First + Roof
4.16
0.97
6.00
3.48
0.19
Second + Roof
2.12
0.99
6.45
3.42
0.09
Third + Roof
1.86
1.00
6.28
3.67
0.08
Fourth + Roof
3.59
1.21
7.24
4.77
0.10
Fifth + Roof
4.37
1.21
7.64
4.37
0.12
First + Second
0.06
0.15
0.96
0.39
0.15
First + Third
0.06
0.16
0.85
0.56
0.13
First + Fourth
0.48
0.36
1.77
1.72
0.16
First + Fifth
0.53
0.37
2.14
1.34
0.19
Combination
Second + Third
0.02
0.19
1.23
0.58
0.03
Second + Fourth
0.20
0.39
2.20
1.69
0.05
Second + Fifth
0.27
0.39
2.60
1.28
0.08
Third + Fourth
0.15
0.40
2.02
1.94
0.04
Third + Fifth
0.24
0.40
2.41
1.54
0.06
00049
25

Michoacan Earthquake, Asymmetric Building).
Combination
Exact
First + Roof
Second + Roof
Third + Roof
Fourth + Roof
Fifth + Roof
First + Second
First + Third
First + Fourth
First + Fifth
Second + Third
Second + Fourth
Second + Fifth
Third + Fourth
Third + Fifth
| C |
Tr[C]
(1016 g 6 )
(104 g 2 )
12.52
60.81
26.51
22.97
43.75
91.15
0.02
0.07
3.46
21.56
0.02
1.05
8.45
0.83
7.15
2.68
4.24
4.27
4.38
5.27
6.78
0.07
0.17
1.07
2.58
0.20
1.10
2.61
1.20
2.71
225.20
357.84
360.67
370.75
443.60
571.86
4.83
14.52
87.39
215.67
17.47
90.31
218.53
100.31
228.57
42.79
65.98
66.46
66.69
83.53
106.25
1.91
2.50
19.24
42.03
2.84
19.64
42.47
19.92
42.72
0.13
0.26
0.11
0.09
0.12
0.15
0.20
0.18
0.21
0.24
0.03
0.06
0.09
0.04
0.07
results for shear buildings [Heredia-Zavoni and Esteva, 1998; Udwadia, 1994]. This
is not surprising since the optimality criterion used in earlier studies, e.g. maximisation of a norm of Fishers information matrix, or minimisation of the Bayesian
loss function can be related to the accurate representation of the covariance matrix
of the Gaussian random process [Goodwin and Payne, 1977]. It may, however, be
noted that the derived optimal sensor locations for the symmetric building are not
applicable for asymmetric building. The first and fifth floors of the six storeyed
asymmetric building correspond to the best possible locations for two sensors as
can be inferred from Tables 3 and 4. Further, it should be noted that the response
records from only two sensors are not sufficient for accurately representing the covariance matrix of the response process and more sensors should be used in the
case of asymmetric buildings.
The floor response data corresponding to the best location of sensors will now
be used for estimation of modal parameters of the building by using system identification techniques. It will be shown that data from the best sensor locations provide
a robust estimate of the model parameters vis-a-vis any arbitrarily selected pair of
response data.
3.1. Parameter estimation
The ground acceleration records and the building response computed at different
floors are assumed to be the input-output pairs for this identification study. In the
26
00049
system identification terminology such systems are known as multi-input multioutput (MIMO) systems and the system parameters may be identified by using
either parametric or non-parametric methods. However, parametric methods exhibit better convergence properties in the case of aperiodic data sequences as compared to the non-parametric methods [Ljung, 1987]. Therefore, the seismic response
of buildings as considered in the numerical examples is modelled as auto-regressive
sequence with external inputs (ARX model). The predictor of such a model is
characterised by a linear regression of model parameters on a known data set comprising of the past input and output sequences as:
y(t) + a1 y(t 1) + + ana y(t na )
= b1 u(t 1) + + bnb u(t nb ) + e(t) .
(3)
Equation (3) is a linear difference equation relating the system response y(t) to
the past values of y(t), input u(t) and a random noise e(t). The model parameters
are then determined by minimizing the quadratic norm of the error between the
recorded and predicted response vectors [Ghanem and Shinozuka, 1995; Ljung,
1987; Shinozuka and Ghanem, 1995]. These parameters are then used to determine
the transfer function of the model, in z-transform, as
nb
X
H(z) =
k=1
na
X
bk z k
;
al z
a0 = 1
(4)
l=0
where, H(z) is the transfer function for the ARX model defined in Eq. (3). The conventional transfer function in the frequency domain can be obtained by evaluating
H(z) on the unit circle. The poles, with positive imaginary part, of this transfer
function are used to determine the natural frequency and damping ratio [Ghanem
and Shinozuka, 1995]. In particular, the pole corresponding to the jth mode of
vibration is given as
q
h
i
zj = exp j j + ij 1 j2 t
(5)
where, j and j respectively represent the damping ratio and the natural frequency
of vibration in jth mode, and t is the sampling interval of time histories. Alternatively, the natural frequencies can be identified from the resonant peaks in the
transfer function plots and the damping ratio can be estimated from the bandwidth
of transfer function in the neighbourhood of the natural frequency by the half-power
bandwidth method. Further, the mode shapes can be obtained by using quadrature
curve fitting technique [Richardson, 1997]. In this technique, the peak values of the
imaginary part of the acceleration/force frequency response function, or transfer
function are taken as components of mode shapes at the sensor locations. In case
of velocity/force frequency response functions, the peak values of the real part are
taken as the mode shape components.
00049
27
Frequency response
10
Amplitude
8
6
4
2
0
10
5
6
Frequency (Hz)
10
Phase (deg)
200
400
600
800
1000
Fig. 5. Sample transfer function of the ARX model for symmetric building. Solid curves: ARX
model estimate, Dotted curves: Periodogram estimate.
The input-output pairs corresponding to the Uttarkashi Earthquake excitations

for both symmetric and asymmetric buildings are considered for the identification purpose. The identification of model parameters in this case is relatively more
difficult because of the short duration of records and presence of noise. Temporal
evolution of the model parameters have been monitored by using the data windowing approach, wherein only a part of the data is used at a time. For each data
window of 5 s length the highest order stable ARX model is determined for the
input-output pairs by using MATLAB. The transfer functions were computed from
the identified parameters of the ARX model for the six data windows for different
combination of sensor locations. A sample of a transfer function for the case of
optimal location of sensors, i.e. first floor and roof is shown in Fig. 5. The transfer
function estimate obtained from the non-parametric (periodogram) method is also
shown in the figure as dotted curve for comparison. It can be seen that the parametric modelling permits a finer resolution of resonant peaks which may be helpful
in estimation of damping by the half-power bandwidth method. However, the parametric transfer function tends to smoothen out some closely spaced peaks in the
transfer function as seen in the figure in the neighbourhood of 9.2 Hz. The phase
spectrum of the transfer function shown in the figure correspond to the unwrapped
phases [Childers et al., 1977; Tribolet, 1977] as computed by MATLAB routines. The
resonant frequencies of the system can be identified as 1.8, 5.3, and 9.2 Hz, which
28
00049
compare favourably with the frequencies of the finite element model. The first six
natural frequencies of the finite element model of the symmetric building were computed as 1.92, 1.96, 2.80, 5.96, 6.20, and 8.46 Hz, respectively. It may be noted that
the translational modes of the symmetric building in longitudinal and transverse
directions are very closely spaced and it is difficult to distinguish such closely spaced
modes from the analysis of response data. Thus the first identified frequency from
the model may be considered to be applicable for the first and second mode of the finite element model, while the second identified frequency corresponds to the fourth,
and fifth mode of vibration of the building. The third mode of vibration, a vertical
mode, could not be identified from the ARX model. Similar estimates of the natural frequencies were obtained from the model parameters identified from different
windows when the response records were those corresponding to the optimal sensor
locations i.e. either the first floor and roof or the second floor and fourth floor.
However, the response records from other locations/floors did not yield consistent
frequency estimates for all windows, and in some cases the lowest mode could not
be identified at all. Thus the choice of the response records plays a very important
role in the identification process. The records from derived optimal sensor locations
are found to lead to robust estimates of the system parameters.
Similarly, the first six natural frequencies of the finite element model of the
asymmetric building were determined as 1.97, 2.79, 2.84, 5.00, 5.37, and 6.39 Hz,
Frequency response
10
Amplitude
8
6
4
2
0
10
5
6
Frequency (Hz)
10
Phase (deg)
500
500
1000
Fig. 6. Sample transfer functions of the ARX model for asymmetric building. Solid curves: ARX
model estimate, Dotted curves: Periodogram estimate.
00049
29
respectively. The frequencies estimated from the identified ARX model of the system
using records from optimal sensor locations are 1.4, 3.4, 4.5, and 6.2 Hz. A sample
transfer function of the ARX model identified from the data is shown in Fig. 6 along
with the transfer function estimate by non-parametric method (shown as dotted
curve). These frequency estimates appear to be reasonable since it is not possible
to capture adequate information about the dynamic behaviour of the asymmetric
building by using only two sensors. It may be noted that though these frequencies
were identified from the models for most of the data windows, the peaks of the
transfer functions at lower modes were not very prominent. Further, for a viscous,
lightly damped system, the phase should change by approximately 180 in the
neighbourhood of a natural model. However, the change in the phase spectrum at
4.5 Hz is much less than 180 indicating that this mode is very heavily damped in
the developed model. This is also apparent from the relatively flat nature of the
transfer function amplification at 4.5 Hz. This indicates that the data set used for
developing the model does not adequately capture the dynamic behaviour of the
system under consideration. Inclusion of response from one or two more locations
may improve the quality of the parameter estimates. Further, the non-parametric
transfer function contains several peaks which are not present in the smooth transfer
function of the parametric model. Authenticity of these additional peaks seen in the
non-parametric transfer function estimate can be ascertained only by considering
more response data from different locations for system identification, which will also
help in better modelling of phases. Moreover, as in the case of symmetric building,
the response records from other locations provide very inconsistent estimates of the
system frequencies identified from different data windows. Similar trends were also
observed in the case of parameter identification with response records for Michoacan
earthquake excitation.
4. Conclusions
A simple methodology for optimal installation of a specified number of sensors in
multi-storeyed buildings for system identification studies has been proposed. The
method is based on the concept of compact probabilistic representation of random
processes and is very easy to apply. The proposed methodology can also be used
to assess the adequacy of a specified number of sensors in capturing sufficient information for system identification purposes. The methodology has been validated
by using it to derive the optimal locations of two sensors in two different types
of buildings. The derived optimal locations of two sensors in a five-storeyed symmetric building compare favourably with the locations predicted by other sensor
placement methodologies. It has been found that the optimal sensor locations for the
symmetric building are different from those for the asymmetric building. Further,
the placement of sensors in the building are found to have an important bearing
on the reliability of the model parameters identified by using system identification
techniques.
30
00049
References
Chandrasekaran, A. R. and Das, J. D. [1991] Analysis of strong motion accelerograms of
Uttarkashi earthquake of October 20, 1991, Technical Report EQ:91-10, Department
of Earthquake Engineering, University of Roorkee, Roorkee, India, December 1991.
Childers, D. G., Skinner, D. P. and Kemerait, R. C. [1977] The cepstrum: A guide to
processing. Proc. IEEE 65, 14281443.
Friswell, M. I. and Mottershead, J. E. [1995] Finite Element Model Updating in Structural
Dynamics (Kluwer Academic Publishers, Dordrecht, The Netherlands).
Ghanem, R. and Shinozuka, M. [1995] Structural system identification. I: Theory, J.
Engrg. Mech. ASCE 121, 255264.
Goodwin, G. C. and Payne, R. L. [1977] Dynamic System Identification: Experiment
Design and Data Analysis (Academic Press, New York).
Heredia-Zavoni, E. and Esteva, L. [1998] Optimal instrumentation of uncertain structural
systems subject to earthquake ground motions. Earthq. Engrg. Struct. Dyn. 27,
343362.
Heredia-Zavoni, E., Montes-Iturrizaga, R. and Esteva, L. [1999] Optimal instrumentaion
of structures on flexible base for system identification. Earthq. Engrg. Struct. Dyn.
28, 14711482.
Kumar, S. [1996] Seismic Analysis of Stepback and Setback building, Ph.D. thesis
Department of Earthquake Engineering, University of Roorkee, Roorkee, India.
Ljung, L. [1987] System Identification: Theory for the User (Prentice-Hall, Inc., Englewood
Cliffs, New Jersey).
Masri, S. F. and Miller, R. K. [1982] Compact probabilistic representation of random
processes, J. Appl. Mech. ASME 49, 871876.
Richardson, M. H. [1997] Is it a mode shape, or an operating deflection shape? Sound
Vib. Mag. 30th Anniversary Issue 1997.
Shah, P. C. and Udwadia, F. E. [1978] A methodology for optimal sensor location for
identification of dynamic systems, J. Appl. Mech. ASME 45, 188196.
Shinozuka, M. and Ghanem, R. Structural system identification. II: Experimental verification, J. Engrg. Mech. ASCE 121, 265273.
Skjrbk, P. S., Nielsen, S. R. K. and C
akmak, A. S
. [1996] Identification of damage in
reinforced concrete structures from earthquake records optimal location of sensors,
Soil Dyn. Earthq. Engrg. 15, 347358.
Tribolet J. M. [1977] A new phase unwrapping algorithm, IEEE Trans. Acoustics, Speech
Signal Processing ASSP-25(2), 170178.
Udwadia, F. E. [1994] Methodology for optimum sensor locations for parameter identification in dynamic systems, J. Engrg. Mech. ASCE 120(2), 368390.
00048


SIMULATION OF NONLINEAR SOIL-STRUCTURE

INTERACTION ON A SHAKING TABLE
KAZUO KONAGAI and RAQUIB AHSAN

Institute of Industrial Science, University of Tokyo, 4-6-1 Komaba,
Meguro-ku, Tokyo 153-8505, Japan
Received 23 May 2000
Accepted 28 February 2001
A new method to simulate soil-structure interaction effects in shaking table tests has
recently been presented by the authors. In this method, analog circuits or digital signal
processors are used to produce soil-foundation interaction motions in real time. Their
expressions of soil-structure interaction motions are based on published rigorous formulations of impulse response functions or flexibility functions of foundations resting on
or embedded in homogeneous or layered soils of semi-infinite extent. In this paper the
method is further extended to take the far field soil non-linearity into account. An
example of non-linear soil-structure interaction simulation using the present method is
also described.
Keywords: Nonlinear soil-structure interaction; shaking table test; dynamic pile-grouphead stiffness; model experiment.
1. Introduction
When a structure is subjected to a ground excitation, it interacts with its substructure, i.e. foundation and soil. In other words, the motion of the ground is
altered because of the vibration of the structure. This dynamic soil-structure interaction is associated with the influx and efflux of energy, which is generated by
the earthquake excitation and transmitted through the soil-structure interface. The
difference between the influx and efflux corresponds to the energy stored up within
the structure, and thus, is closely related to the extent of damage to it. If this interaction effect can be rationally simulated in shaking table tests, one may obtain the
information necessary to interpret the failure processes of prototype structures in
terms of energy. For shaking table tests conducted without taking the interaction
into account, the input energy is totally consumed by the structure, producing quite
conservative results. On the other hand, tests conducted using a physical ground
model to incorporate associated non-linearity have twofold disadvantages. Firstly,
the finite dimension of the ground model fails to provide radiation damping, and
thus, yields conservative results. Secondly the weight of the ground model causes
an extra burden on the performance of the shaking table, restricting superstructure
31
32
00048
K. Konagai & R. Ahsan
models to a smaller size. Hence, if observation of the behavior of a superstructure

is the main concern, neither of the above methods seems to be satisfactory.
In order to incorporate the interaction between soil and structure without using
a physical ground model, Konagai and Nogami [1997, 1998] introduced a new
method for shaking table tests. In their method, appropriate soil-structure interaction motions are added to free-field ground motions to simulate soil-structure
interaction effects. The expressions of soil-structure interaction motions are based
on published rigorous formulations of flexibility functions of foundations resting on
or embedded in homogeneous or layered soils of semi-infinite extents. The method
considers radiation damping which, in general, causes the total damping of a soilstructure system to be greater than that of the structure itself. Thus the incorporation of soil-structure interaction effects in a shaking table test leads to a reduction
in the demands on the capacity of the shaking table.
The method was initially developed with the assumption that soil behaves linearly. In the present paper, the method is extended to take far field soil nonlinearity into account through an equivalent linear approach. The non-linearity
produced in the vicinity of foundations, which is usually associated with large strain
and separation between soil and foundation, is not considered in this study. Here,
the dynamics of the interaction between a superstructure and a pile-group, along
with the semi-infinite soil, are modeled with a simple flexibility function. A computer program, based on the Thin Layered Element Method [Tajimi et al., 1976], is
used to derive the flexibility function for the lateral sway of the pile-group. In this
method the dynamic soil parameters are varied in real time by means of a digital
signal processor. The method, on the one hand captures the non-linear soil behavior
of softening and rehardening during the course of an earthquake, and on the other
hand may allow testing of larger superstructure models by obviating the need for a
heavy physical ground model. A comparison of the results of a test incorporating
soil-structure interaction and those of a conventional test is also presented in this
paper.
2. Outline of the Present Method
In this study, a soil-structure system is divided into two substructures, the superstructure and the unbounded soil extending to infinity; the latter includes an
embedded foundation as illustrated in Fig. 1, and is represented by a shaking table.
The multi-step method is used to describe two primary causes of soil-structure interaction: the inability of the foundation to match the free-field deformation, and
the effect of the dynamic response of the superstructure on the movement of its
supporting soil-foundation system. In the lower substructure of soil, an earthquake
will cause soil displacements, {uf }. The foundation embedded in this soil deposit,
however, will not follow the free-field deformation pattern. This deviation of the
displacements from the free-field soil displacements, {uf }, is denoted by {us }. The
mass of the superstructure then causes it to respond dynamically, and the forces,
00048
Simulation of Nonlinear Soil-Structure
Fig. 1.
33
Two primary causes of soil-structure interaction.
{p}, transmitted to the lower substructure of soil and foundation will produce further deformation of soil, {uR } (inertia interaction), that would not occur in a fixed
base structure. Thus, the displacements of soil, {u}, are eventually expressed by
the following equation:
{u} = {uf } + {us } + {uR } .
(1)
Consider the case that a foundation has two degrees of freedom in sway and rocking
(x, ) at the base of its super-structure, as illustrated in Fig. 1. The interaction
forces, {p}(= {px p }T ), from the superstructure cause the inertia interaction motions, {uR }, in the frequency domain to be:
(
) "
#( )
uR
Hxx (s) Hx (s)
px
x
=
(2)
R
u
Hx (s) H (s)
p
where
"
Hxx (s)
Hx (s)
Hx (s)
H (s)
#
= [H]
(3a)
is the flexibility (compliance) at the top of the foundation, and

s=i
(3b)
in which i = 1 and is the excitation circular frequency.

In the present method, the motion of a shaking table is controlled directly
following the actual process of soil-structure interaction. Figure 2 shows a schematic
view of the set-up of a shaking table test, in which a superstructure model is placed
directly on the table without a physical ground model. Soil-structure interaction
effects are simulated by adding appropriate soil-structure interaction motions to
free-field ground motions on the shaking table. In the simulation, first, the transducers at the base of the foundation pick up the signals of the base forces, px and
p in sway and rocking motions, respectively. These two amplified signals are then
applied to the circuits Hxx , Hx , Hx and H to produce outputs corresponding
R
to the soil-structure interaction motions, uR
x and u . The output signals are then
34
00048
Fig. 2.
Simulation of soil-structure interaction on a shaking table.
added to the signals of the base input motions, ufx + usx and uf + us , to produce
f
s
R
signals of foundation motions, ufx + usx + uR
x and u + u + u . The method is,
f
f
s
s
thus, based on the premise that ux + ux and u + u are known beforehand as the
base input motions. Signals of the foundation motions are finally translated into
the shaking table motions by the shaking table controller.
The above method requires a device that generates signals corresponding to the
soil-structure interaction motions, and a digital signal processor (DSP) which is capable of producing a variety of transfer functions is used as this device. The transfer
functions to be realized on the DSP are designed from the analytical expressions for
stiffness or flexibility functions of the foundations. Konagai et al. [2000] have shown
that np piles closely grouped together beneath a superstructure can be viewed as
a single equivalent upright beam whose stiffness matrix is described with two stiffness parameters, EIp and EIG . The parameter, EIp is identical to np Ep Ip , where
Ep Ip is the bending stiffness of an individual pile and np is the number of piles in
a pile-group. EIG is evaluated following the same procedure as that used for the
evaluation of the bending stiffness of a reinforced concrete beam; in this analogy,
piles and the soil mass caught among them are respectively compared with reinforcing bars and concrete. In other words, EIG is assumed to be equal to the sum
of the Youngs-modulus-weighted products of all the elementary areas times their
distances squared from the centroid of the cross-section AG (Fig. 3). Careful examination of deflections of pile groups reveals that most piles are indeed flexible in
practice in the sense that they do not deform over their entire lengths. Instead, pile
deflections become negligible below their active lengths, La ; the active pile length
is given as a function of EIp and the shear modulus of the surrounding soil, , as:
s
EIp
La = L0 =
.
(4)
00048
35
centroid
Ag
R0
w1
wj
j
hj
nL
remain on one plane
np piles
Fig. 3.
Assumptions for evaluation of equivalent single beam.
The parameter in the above equation differs in different soil profiles. Konagai,
Ahsan and Maruyama [2000] carefully examined the solutions of pile cap stiffness,
Sxx , in sway motion by using the upright single beam analogy, and showed that
they are closely approximated by the following expression:
Sxx = ks + ia cs a2 ms
(5a)
where
a=
R0
s
(5b)
with = circular frequency and s = shear wave velocity; R0 = the radius of

the equivalent upright beam, which is assumed to be identical to the radius of a
circle with the same area as the cross-section AG that includes all the grouped piles
enclosed by the broken line in Fig. 3. For a homogeneous soil,

ks = 2R0 + L0 ,
(5c)
2
cs = 2L0
(5d)
and
ms =
L0
.
4
(5e)
Equation (5a) clearly shows that the flexibility function, which is the inverse of the
dynamic-stiffness of a soil-pile group system is approximated by that of a simpledamped oscillator with a spring K(= ks ), a damper C(= R0 cs /s ) and a mass
M (= R02 ms /s2 ) (Fig. 4). For an inhomogeneous soil, soil stiffness in Eqs. (5c)
(5e) will be a mode-weighted average of shear modulus over the active pile length,
and thus, will vary with time as the nonlinear feature of soil develops.
36
00048
M
K
C
Fig. 4.
Modeling of a pile-group as a simple-damped oscillator.
3. Control of Shaking-Table
The system illustrated in Fig. 2 is realised on condition that a shaking table loses no
time in faithfully producing its input motion. The motion produced by the shaking
table, however, is not exactly identical to the intended time history because the
ratio of the output to input amplitude of the shaking table system does not remain
the same over the desired frequency range. The performance of the systems transfer
function is also affected by the presence of models on the shaking table; this fact
may cause the motion of the table to further deviate from the intended time history.
A controller with the transfer function T normally performs like a low pass filter,
and experiments on the table are often conducted below its cut-off frequency. Below
this frequency there yet remains a time delay t between the produced motion and
the input signal. The effect of the time delay, described in the frequency domain
as T
= eit , can be canceled by multiplying the flexibility function H by T 1 .
Assuming that the performance of a soil-foundation system is approximated by that
of a simple-damped oscillator with a spring K, a dashpot C and a mass M (Fig. 4),
the flexibility function Hxx is expressed as:
Hxx =
1
K
2M
+ iC
(6)
Thus, the cancellation of the time-delay effects is achieved by

Hxx T 1
=
eit
.
K 2 M + iC
(7a)
For smaller values of t, Eq. (7a) is rewritten as:

Hxx T 1
=
1
K 2 (M M ) + i(C C)
(7b)
where
M = C t
(8a)
C = K t .
(8b)
Equation (7b) shows that the equivalent mass and the viscous damping coefficient
are reduced by C t and K t, respectively. The reduced mass M M and
00048
37
the damping coefficient C C must be positive, calling for:

M
tc t
= 4 2 2 < 1
M
t0
(9a)
C
t
=
<1
C
tc
(9b)
with
tc = C/K (T ime constant)
p
t0 = 2 M/K .
(10a)
(10b)
The above conditions [Eqs. (9a) and (9b)] are usually satisfied in reality for many
cases of soil-structure interaction, because radiation of waves from a foundation
leads the motion of the structure to be noticeably damped.
It is, however, necessary for the time delay to be minimised when Eqs. (9a) and
(9b) are not satisfied. One possible measure for reducing the time delay is to increase the feedback gain of the servo-amplifier of a shaking table. The increase of the
servo-amplifier gain, however, leads to a decrease in the margin for unstable clattering of the table that is caused by the noise echoing through the closed circuit of
the servo-amplifier [Konagai et al., 1999a]. A form of adaptive control known as the
Minimal Control Synthesis (MCS) algorithm has recently been used on a shaking
table at the University of Bristol, UK [Gomez and Stoten, 2000]. The MCS algorithm responds in real-time to dynamic changes experienced by a structure model
and tunes the control parameters to optimum values in an automated manner. The
main advantage of this method is that it does not require any prior information on
the plant dynamic parameters. Implementation of another new control method in
a shaking table system is currently being developed in the Institute of Industrial
Science, University of Tokyo [Yamauchi et al., 2000]. In this method, an H-infinity
robust control scheme has been adopted in conjunction with an adaptive filter to
give the controller the ability to track changes in the controlled system characteristics. Thus, one of the major impediments to implementing the present method,
namely that preventing reliable realisation of the input motion in shaking tables,
is being overcome.
4. Experiments
To provide a proper perspective on the usefulness of the present method, its
application to a particular prototype is described herein. The prototype is an expressway viaduct supported by pile foundations embedded in an alluvial soil deposit [Fig. 5(a)]. This soil-viaduct system is divided into two substructures: the
superstructure of the steel-pipe pier carrying the 666 tf weight of the viaduct
deck [Fig. 5(b)] and the lower substructure of soil, which includes the pile foundation [Fig. 5(c) and Table 1]. A model of the upper substructure is mounted on a
38
00048
shaking table whose motion expresses virtually the unbounded soil substructure.
Since the shaking table used in the experiment has only one sway degree of freedom,
discussions on rocking motions were excluded.
Table 1.
Parameters for steel piles.
Ep (tf/m2 )
(t/m3 )
r0 (m)
Thickness (m)
Length (m)
2.1 107
0.406
0.012
50
Fig. 5.
Prototype viaduct.
00048
39
Lateral force (MN)
6
4
2
0
-2
-4
-6
-0.4
-0.2
0.0
0.2
0.4
Relative displacement (m)

Fig. 6.
Force displacement relationship of pier (PC-15).
4.1. Superstructure model

Since the objective of the experiment was mainly focused on studying the change in
dynamic behavior of the structure due to the incorporation of soil-structure interaction effects, an exact physical model of the prototype structure was not necessary.
Therefore, instead of making such an exact model, an attempt was made to simplify the dynamic features of the prototype. A typical pier of the viaduct sustaining
666 tf weight has a fundamental resonance frequency of about 1 Hz. The dynamic
force displacement relationship of the pier was calculated by using SFRAME [Li
and Goto, 1998] as shown in Fig. 6 [JSSC (Japan Society of Structure Construction) Report, 2000]. The force-displacement curve in Fig. 6 was approximated as a
bilinear one. One simple way to realise this bilinear force-displacement relationship
is to place a mass on top of a pedestal supported by a spring so that the mass slips
when the acceleration exceeds the level that the friction between the mass and the
pedestal can resist. Below this level the mass moves with the pedestal. If the spring
carrying the pedestal possesses a linear force-displacement relationship, stick and
slip motions of the mass yield a bilinear force-displacement curve.
In the present experiment, a mass of 6.75 kg was put on the crossbeam (pedestal)
of a steel frame. A teflon sheet was put on the crossbeam to realise a smooth
transition of the blocks motion from stick to slip and vice versa. A photograph of
the experimental set-up is shown in Fig. 7. The frequency of the steel frame with
the block mounted was 2 Hz. Since the frequency of the frame was two times that
of the prototype, the duration of the input motion as well as the time-constant
[Eq. (10a)] of the flexibility function was halved. From Fig. 6, it is found that the
pier carrying a mass of 666 tf shows plastic deformation beyond an acceleration
of about 600 cm/s2 . But the average frictional coefficient between the mass and
40
00048
Fig. 7.
Experimental setup.
the teflon sheet was found to be around 0.2. Therefore, the mass slipped when
acceleration exceeded about 200 cm/s2 . Again the ratio of model displacement, um
and prototype displacement, up can be expressed in terms of the ratio between
model frequency, m and prototype frequency, p and the ratio between model
acceleration, am and prototype acceleration, ap as,
p2 am
um
= 2
.
up
m ap
(11)
For the present case, therefore, the input motion was scaled down by a factor of
1/12.
4.2. Foundation input motion

The signal of the foundation input motion, ufx + usx , which is not affected by the
presence of the superstructure, must be obtained prior to the experiment, and for
this, the free-field ground motion ufx was first obtained.
The ground response analysis was carried out considering only vertically propagating horizontal SH waves, and the presence of the flexible group-piles was ignored
in this analysis. Hence the soil profile (Table 2) was modeled as a one-dimensional
horizontally layered soil column having non-linear soil properties. For the analysis,
the Finite Element Method was adopted in the spatial domain assuming piecewise
linear variations of displacement for the soil slices and a Central Difference Time
Marching Scheme in the time domain. A schematic illustration of the soil column
is shown in Fig. 8.
To express the non-linear stress-strain relationship of each layer of the soil column, the Hardin-Drnevitch model was adopted in association with the Modified
00048

Table 2.
41
Ground profile at the site of prototype.
Thickness
h (m)
Unit Weight
(tf/m3 )
Shear Wave
Velocity
Vs (m/s)
Shear Modulus
G0 (tf/m2 )
3.8
1.9
144.0
4020
2.6
1.9
129.3
3241
12.2
1.4
159.8
3648
0.9
1.9
173.2
5816
11.5
1.4
159.5
3634
5.0
1.9
161.0
5026
6.0
1.9
188.2
6867
Fig. 8.
One-dimensional soil column.
Masing rule. The backbone curve of the model is expressed in differential form as,

n
d
= G0 1
(12)
d
f
where, and are respectively the shear stress and shear strain of soil; moreover
G0 is the initial tangent modulus and f is the shear strength of soil. The value of
the arbitrary constant n depends on the actual stress-strain relationship of the soil.
In this analysis the value of n was assumed to be 2. In order to take into account
the effect of the bedrock underlying the soil column, a dashpot was attached at the
bottom of the soil column. For shear wave propagation through a one-dimensional
semi-infinite medium, the damping coefficient of this dashpot is given by,
C = Gr /Vsr
(13)
42
00048
Acceleration (cm/s )
1000
500
-500
-1000
0
Time (s)
Displacement (cm)
Fig. 9.
Sample base excitation.
5
0
-5
-10
-15
0
Fig. 10.
10
15
20
Time (s)
Displacement history on the ground surface.
where, Gr and Vsr are the shear modulus and shear wave velocity of the underlying
half-space of soil respectively.
The soil column was subjected to a sample base excitation for the non-linear
ground response analysis. The north-south component of the acceleration record of
the September 3, 1998 Middle-Iwate Earthquake (Fig. 9, Shizukuishi, JMA) was
used as the sample base excitation. The analysis yielded the displacement history
at the surface layer (Fig. 10) and stress-strain histories for all the layers. Since the
basic concept of the present experiment is the superposition of the foundation input
motion ufx + usx and the inertia interaction motion uR
x in real time, an equivalent
linear approach, which allows for the Multi-Step Method, was required to reflect the
non-linear features of soil. Hence, the equivalent linear material parameters, namely
secant shear moduli and damping ratios, were obtained for all soil slices following
the sequence of loading cycles. At each reversal point of the stress-strain history,
the corresponding secant shear modulus was determined as the slope of the line
joining the present and previous reversal points, and the damping ratio from the half
breadth of the latest hysteresis loop enclosed by both the line determining the
secant modulus and the newest arc of the loop (Fig. 11). Figs. 12(a) and 12(b),
respectively, show the variations of the secant shear moduli and the corresponding
damping ratios of different layers of the soil profile with time.
00048
43
Half breadth
of loop
Gsec
Fig. 11.
Determination of secant shear moduli.
>
>
&
'
"
;<
;<
7

9
"

78
345
&
'
(a) Variations of secant shear moduli

Fig. 12.
(b) Variations of damping ratio
Variations of equivalent linear parameters with time.
The presence of the pile group in the soil causes the soils motion to be altered
from {uf } to {uf } + {us }, and consequently, the pile caps motion in sway is
described as ufx + usx . The pile group is replaced with an equivalent upright beam
[Konagai et al., 1999b, 2000], the concept of which has already been described in
Sec. 2. In the evaluation of ufx +usx, a soil deposit was treated as an infinite stratified
medium with the inclusion of a cylindrical hollow [Thin Layered Method: Tajimi
and Shimomura, 1974]; into that hollow, the pile-group-equivalent upright beam
was fitted. Based on these assumptions, a numerical program, TLEM (Ver. 1.2)
was developed for the evaluation of the kinematic interaction [Konagai et al., 1999b,
2000] whose effect is portrayed in the form of two kinematic displacement factors
in sway and rocking motions
Te,sway =
Te,rocking =
ufx + usx
ufx
ufz + usz usz

= f
ufx
u1
(14a)
(14b)
plotted as functions of frequency. In the above equations, rocking motions are expressed in terms of the anti-symmetric vertical motions of uz at the outermost edge
of the equivalent upright beam (r = R0 ) with respect to the beams centroid. It is
real (sway)
imag (sway)
real (rocking)
imag (rocking)
1.0
(u +u ) / u x
0.5
Kinematic displacement factors
44
00048
0.0
10
Frequency (Hz)
Fig. 13.
Kinematic displacement factors.
noted in Eq. (14b) that the presence of the pile foundation produces the vertical
component of the ground motion usz even though there is no vertical component
ufz included in the free-field ground motion. Figure 13 shows the kinematic displacement factors for the pile foundation. It is recognized from this figure that the
kinematic interaction effect, which is described as the deviation of Te,sway from 1.0
and the increase in Te,rocking , is noticeably small below 4 Hz, implying that the pile
foundation is indeed flexible enough to follow closely the surrounding soils motion.
For this reason, as well as the restriction that the shaking table used in the experiments does not have the freedom for rocking motion, the kinematic interaction
effect was ignored in the experiment.
4.3. Pile cap stiffness and substructuring by means of DSP
In the present method, the inertial interaction uR
x is simulated in real time on a
shaking table. For this simulation, pile cap stiffness should be rationally evaluated.
Equations (5c)(5e) certainly provide the necessary parameters K, C and M describing the approximate stiffness. The examined cases in deriving Eqs. (5c)(5e),
however, were oversimplifications of some typical soil profiles. For this reason, the
aforementioned TLEM (Ver. 1.2) was repeatedly performed at different times to
cover the total duration of the excitation, with the variations of the secant shear
moduli and the damping ratios of the surrounding soils provided in Figs. 12(a) and
12(b). Then the obtained value of the pile cap stiffness at each time was expressed
in terms of the frequency invariant K, C and M parameters that best fit the stiffness in the least square sense. The obtained variations of K, C and time-constant,
tc are shown in Figs. 14(a)14(c) with linear interpolation between the values obtained from the analysis. The value of M was found to remain nearly constant at
20 tf-s2 /m.
00048

(a) Variation of the stiffness of the substructure
"

&
'
(b) Variation of the damping of the substructure

"
(c) Variation of the time-constant of the substructure

Fig. 14.
Variations of equivalent linear parameters with time.
45
46
00048
The variations of the parameters should be realized in the real time simulation
of the interaction motion through a Digital Signal Processor (DSP). For this purpose, a SIMULINK program (Fig. 15) was implemented in the DSP. The motion
realized by means of the DSP on the shaking table was adjusted by the FFI gain
to correspond to 1/12th of the prototype motion (Fig. 10). The signals picked up
by the force transducers attached at the bottom of the columns of the steel frame
were transmitted to the SIMULINK program through an AD converter. The signal
was then passed through the flexibility function to produce soil-structure interaction motion. The K and C parameters of the flexibility function were changed,
according to their variations shown in Figs. 14(a) and 14(b), in real time with an
MLIB routine. The SSI Gain was used for consistent modeling of the substructure
dynamics. The initial stiffness of the prototype substructure was found to be about
50 times the stiffness of the pier. The value of the SSI Gain was adjusted so that,
in the absence of any input motion, a static displacement of the frame produced a
displacement of the shaking table that was 1/50th of that of the frame.
A gap sensor was used to measure the relative displacement between the mass
and the frame. The displacements of the frame and the shaking table were measured
with laser sensors. Accelerometers were used to measure acceleration of the mass
and of the frame.
Displacement of
shaking table (cm)
Fig. 15.
SIMULINK model.
0.5
0.0
input motion
realized motion
-0.5
-1.0
0
Time (s)
Fig. 16.
Reliability of the realised motion.
10
00048
47
4.4. Results
One of the major concerns in the present experimental method is the reliability
of the realized input motion. For the purpose of checking the reproduceability of
motions on the shaking table, the signal of the input motion and the motion realized
on the shaking table were both measured simultaneously. During this measurement,
the SSI Gain was set at zero i.e. there was no interaction effect involved. Both the
measured signals are shown in Fig. 16, which shows that the input motion was
satisfactorily realised.
In Fig. 17(a), the displacement of the shaking table is shown for both the cases
when the interaction effect was considered and when there was no interaction. Although the difference in the motion of the shaking table may not be noticeable, the
difference in the acceleration histories is evident. The peak acceleration was reduced
when the soil-structure interaction was considered [Fig. 17(b)]. The differences due
to the interaction effect in the displacements of the frame and the mass were thus
,-
"

+
&

'(
&
(a) Displacements
<
>
'
&
"#
(b) Acceleration
Fig. 17.
Displacement and acceleration of shaking table.
48
00048
Displacement of frame (cm)
0.5
0.0
-0.5
without soil-structure interaction

with soil-structure interaction
-1.0
-1.5
0
10
15
20
Time (s)
(a)
Slip of the block (mm)
Without soil-structure interaction

With soil-structure interaction
-1
0
10
Time (s)
(b)
Fig. 18.
Displacement of (a) frame and (b) mass with respect to frame.
clear, as shown in Figs. 18(a) and 18(b). Marked difference can be noticed in the
displacement of the frame between the 2nd and 5th second when the stiffness of the
flexibility function is small and the time-constant is high due to the non-linearity
of soil. During the latter part of the displacement history of the frame when input
motion is about zero, the effect of the increased damping due to the soil-structure
interaction is evident. It should be mentioned here that the slippage of the block
was not always the same even for the same input motion and even without interaction. After repeating the same experiment a number of times, a representative
result was produced [Fig. 18(b)].
In the absence of any soil-structure interaction, the total input energy is consumed through the friction of the mass and the damping of the frame. Whereas
the interaction between soil and structure causes a large amount of energy to dissipate as outwardly propagating waves into the virtually spreading soil medium,
00048
49
this dissipation of energy causes less energy to be consumed by friction and thereby
reduces the slippage of the mass. The present method allows both influx of energy,
Einflux and efflux of energy Eefflux through the foundation to be measured in real
time. These two quantities are respectively:
Z
Einflux =
px u x dt
(15a)
px u R
x dt .
(15b)
and
Z
Eefflux =
0
The energy, Econsumed used up within the model on the shaking table is then
obtained as:
Econsumed = Einflux Eefflux .
(15c)
The energy is consumed within the model in two mechanisms: one due to the
damping of the frame and the other through friction of the mass and the teflon
sheet. The energy consumed due to the damping of the structure is given by,
Z
cu 2frame dt
Edamp =
(16a)
where c and u frame are respectively the damping coefficient and velocity of the
frame. The energy used up due to friction can be calculated as,
Z
m
ublocku slip dt
Efric =
(16b)
where m and u
block are respectively the mass and acceleration of the block, and
uslip is the distance the block slips with respect to the frame. Thus Econsumed can
also be expressed as
Econsumed = Edamp + Efric .
(16c)
Figures 19(a) and 19(b) show the distribution of energy in two different cases:
one without any consideration of the interaction and the other considering the
interaction effects. The difference between the energy influx and the total consumed
energy in these figures corresponds to the kinetic and potential energy components
of the mass and the frame when they are in motion. In relation to the prototype, it
can be said that the soil-structure interaction causes dissipation of energy through
soil which results in less plastic deformation of the pier. Any prediction based on
experimental results not considering the interaction may overestimate the amount
of plastic deformation accumulated in the pier after an earthquake.
50
00048
"
*+
(

"

&
'
&
'
(a) without interaction

Fig. 19.
1
,
(b) with interaction
Influx, efflux and consumption of energy.
5. Conclusions
The objective of this study was to incorporate soil non-linearity into an existing
shaking table test where, instead of using a physical ground model, the soil-structure
interaction effects are realized through the introduction of appropriate flexibility
functions in the control of the shaking table. In this paper, a method has been
described to achieve the non-linear behavior of soil in shaking table tests. The
method is basically an equivalent linear method where non-linear parameters obtained from a non-linear ground response analysis are idealized by equivalent linear
parameters. These parameters are then employed to determine appropriate variation of the flexibility function parameters. This variation is then realised in real
time in a shaking table test by means of digital signal processors. An application
of the method has been elaborated in this paper. The results obtained from the
experiment show some important features of the soil-structure interaction effects.
Unless soil-structure interaction effects are considered in shaking table tests, the
predictions of such tests may provide too conservative information.
Acknowledgements
Partial financial supports for this study have been provided by the Japanese Society
of Steel Construction and the Ministry of Education, Science and Culture (Grant
in Aid for Scientific Research, No. 10450174). The authors would like to express
their sincere gratitude towards Mr. Toshihiko Katagiri, IIS, University of Tokyo
and Dr. Atsushi Mikami, Tokushima University for their efforts that have made the
experiments possible.
References
Gomez, E. G. and Stoten, D., [2000] A comparative study of the adaptive MCS control algorithm on European shaking-tables, 12th World Conference on Earthquake
Engineering.
00048
51
Konagai, K. and Nogami, T. [1997] Simulation of soil-structure interaction on a shaking

table, Geotech. Special Tech. Publ. ASCE 64, 91106.
Konagai, K. and Nogami, T. [1998] Analog circuit to simulate dynamic soil-structure
interaction in shaking table test, Int. J. Soil Dyn. Earthq. Engrg. 17(5), 279287.
Konagai, K., T. Nogami, T., Katsukawa, T., Suzuki and A., Mikami [1999a] Real time
control of shaking table for soil-structure interaction simulation, J. Struct. Mech.
Earthq. Engrg. JSCE 16(1), 91s95s.
Konagai, K. [1999b] Real-time control of shaking table for the simulation of structurebase interaction during earthquakes, Report of Research Project, Japan Ministry of
Education, Science, Sports and Culture, 09875109.
Konagai, K., Ahsan, R. and Maruyama, D. [2000] Simple expression of the dynamic stiffness of grouped piles in sway motion, Int. J. Earthq. Engrg. accepted for publication.
Tajimi, H. and Shimomura, Y. [1976] Dynamic analysis of soil-structure interaction by
the thin layered element method, Trans. Architect. Inst. Japan 243, 4151.
Japan Society of Steel Construction [2000] Seismic design of entire soil-foundation-bridge
systems and its future prospects.
Li, X. S. and Goto, Y. [1998] A three-dimensional nonlinear seismic analysis of frames
considering panel zone deformations, J. Struct. Mech. Earthq. Engrg. JSCE 15(3),
113.
Yamauchi, Y., Uchiyama, Y., Ueno, K. and Fujita, M., [2000] Vibration control of multiDOF excitation systems using modern control theory, Fatigue.
00054


WHEN SHOULD SEISMIC RETROFITTING OF EXISTING

STRUCTURES BE IMPLEMENTED IN ORDER TO
MINIMIZE EXPECTED LOSSES
IVO VANZI
Department of Structural Design, Retrofitting and Control (PRICOS),
University G. DAnnunzio of Chieti, Italy
Received 24 January 2001
Revised 25 June 2001
Accepted 28 June 2001
The usual outcomes of the seismic safety analysis for an existing civil engineering structure are the probability of exceedance of specified limit states and the increase in safety
due to retrofitting interventions. This information can be used in several ways. For a
single structure, one can compare it with desired target reliability values; for structures
belonging to a network, e.g. highway bridges [Donferri et al., 1998], electric networks
[Vanzi 1996, 2000] or strategic buildings [Nuti and Vanzi, 1998] they can also be used
to assess the priority of interventions.
In this study, an alternative use of the reliability values for existing structures is
proposed, which answers the following question: when, i.e. in which year from the date of
construction, should seismic retrofitting be implemented so as to minimize the expected
total cost? In the expected total cost, here, both the costs of retrofitting and possible
disruption, due to delayed retrofitting, are accounted for.
The method proposed computes the expected costs by analysing the branches of
the event tree for the problem built after strong but reasonable and highly simplifying
assumptions on the problem. Although these assumptions limit the general applicability
of the solutions obtained, they allow the building up of an extremely agile and effective
solution scheme.
The results obtained from the study, i.e. the year in which it is economically best to
implement retrofitting and what the expected annual equivalent cost is, are presented in
diagrams and in analytical form, as a function of the most important variables. Finally,
an example application on a real structure is presented, which shows all the steps to
undertake with the proposed method.
Keywords: Seismic retrofitting; optimization; fragility; safety; cost minimisation;
structural reliability.
1. Introduction
Seismic retrofitting of existing structures is a key issue in any earthquake-prone
region. Moderate to strong earthquakes, in fact, claim many lives and cause extensive damage to structures and infrastructures [Schiff, 1988; O Rourke and Palmer,
1994; O Rourke, 1996; Lund, 1996] at a regional scale, halting normal life and
the economy.
In developed countries, earthquakes may cause severe economic damage also
at a national scale; estimates of the monetary damage [Tiedemann, 1992; The
53
54
00054
I. Vanzi
Table 1. Damages expressed in monetary values for some past earthquakes (adjusted
by inflation).
Country
Earthquake
Year
Costs (billion US$)
Costs/GDP(1) (%)
Japan
Great Hanshin
1995
110
2.3
California, USA
Northridge
1994
20
2.4
California, USA
Loma Prieta
1989
7.1
0.9
(1): For Northridge and Loma Prieta, the GDP considered is the one of the State of
California.
Economist, 1995] put the figure for direct costs only at some percentage points
of the gross domestic product (GDP) as shown in Table 1. Estimates of total costs,
which include also indirect costs, indicate [The Economist, 1995] values almost
double.
To cope with the problem, a variety of technical solution has been implemented
for the seismic retrofitting of structures, ranging from strengthening of parts of the
structure to seismic isolation and active control. However a systematic application
of retrofitting techniques on a national, or even regional, scale has not been undertaken, because of the high costs involved. As a result, in case of interventions
on a territorial scale, massive retrofitting of structures that have been identified as
vulnerable has never taken place.
This is true both for developed and developing countries in which seismic risk is
a concern. In Italy, where most of the territory is at risk, no special public policy is
adopted for earthquake prevention of existing structures, apart from temporary and
partial eligibility of retrofitting costs for fiscal deduction. Earthquake insurance is
moreover used by a negligible minority and so the costs of repairing the earthquake
damages have been up to now footed by the state interventions (only the direct part
of the costs). This situation is similar to Japan, another country where earthquake
insurance is not used (only 3% of home owners in the Kobe area were insured) and
different from the US, where about half properties are covered by insurance. The
scheme in the US looked like a working one because insurance costs for property
owners were a function of the structural risk of failure (considering both hazard,
i.e. earthquake intensity at the site, and fragility, i.e. the capability of the structure
to withstand the action) and so owners were economically motivated to retrofit
their houses or facilities. The drawback of the scheme showed up right after the
Northridge event: US$11.4 billion worth claims for property damage, about US$4
billion collected as insurance fees between 1970 and 1994 [The Economist, 1995].
Insurers decided then to drop the residential earthquake insurance market.
The drawback is, in the authors opinion, that earthquake damage is highly
random both in place and in time and in magnitude and so, in order to have the
economics of investment under risk work, costs and risk must be spread on a large
community for a long period of time. Financial resources ought to be used to make
00054
When Should Seismic Retrofitting of Existing Structures
55
preventive interventions (isnt it common knowledge that prevention is better than

cure?) under a compulsory scheme (retrofit or pay the damage by yourself) fitted
to the local conditions and customs.
A second important issue concerns cost-evaluation: in Europe and especially in
Italy, objective cost-evaluation is more difficult than elsewhere because most civil
engineering works are old or ancient [Istat, 1999], with artistic or historical values,
and have not been designed for seismic action, and hence the assessment of their
safety is subject to a high degree of uncertainty.
These two issues (Is retrofitting worth the cost? How can costs be computed?)
have been dealt with in the state-of-the-art literature. Cost-benefit models have
been proposed for the evaluation of the profitability of public or private investment
in seismic retrofitting [Fema, 1992; Chang and Shinozouka, 1996; Kanda and Shah,
1997]. These models permit comparison among alternatives by assigning monetary
values to costs and benefits happening in the future and discounting them at the
present time accounting for inflation and interest rate but their application to the
real world is somewhat problematic because of their complexity and of the subjectivity in the phase of assignment of monetary values to costs and benefits of
disomogeneous nature and of difficult evaluation.
The model presented in this study tries to address both issues. It does not
solve the problem of cost-evaluation but, by using a very agile representation of the
economic problem of retrofitting, identifies the few basic variables which govern the
problem, to perform parametric studies using them and to produce results readily
exploitable. Among the results, the annual equivalent cost for earthquake damage
is obtained as a function of the year of retrofitting.
In the model, structural behavior is represented with a (fixed in time) fragility
curve, relative to a selected limit state, which may be of the serviceability or ultimate type. The seismic action is represented with the classical Cornells model
[Cornell, 1968] and the convolution of the structural fragility curve with the probability density function (pdf) of the earthquake intensity yields the mean rate in time
of structural failures. The different possibilities which may occur when retrofitting
is delayed, are next examined; for each of these, identified by a branch of an event
tree, the annual equivalent cost is computed. In this way, the expected cost at
present time due to structural failures is computed and is minimised with respect
to the year at which the seismic retrofitting intervention is made.
2. Statement of the Problem
In the proposed model, the structure is described by its fragility curve Pfl,z (z),
the function expressing the probability of exceedance of the limit state l versus an
earthquake intensity parameter z, usually the Mercalli intensity or the peak ground
acceleration. Methods to establish structural fragility curves are well established
[Der Kiureghian, 1999; Giannini et al., 1981] and will not be dealt with in this
paper. An application example is however given in Sec. 5.
56
00054
I. Vanzi
In the present version of this study, ageing effects [Ciampoli, 1998; Mori and
Ellingwood, 1993] are disregarded because they would cause sensible increase in
complexity of the mathematical passages which follow. Ageing effects would however
only cause the structural fragility function to be time-dependent and they might
be included in an improvement of this method.
The seismic action at the site, following Cornells method [Cornell, 1968], is
modeled as a Poisson process with mean rate equal to Earthquake intensity Z,
given an event, is distributed according to the GutenbergRichter law:
Fz (x) =
ex ezmin
ezmin
ezmax
(1)
between the minimum and maximum values zmin and zmax . is the so-called severity
parameter for the site. Physically, it is the slope of the minimum error interpolation
line in a diagram having the recorded intensities on the abscissa axis and the natural
logarithm of the frequency on the ordinate axis. In Eq. (1), and in the following,
fX (y) and FX (y) respectively indicate the probability and cumulative distribution
functions of the random variable X evaluated at y.
(l,e)
of exceedance of the limit state l, henceforth indicated as
The probability Pf
failure, given an event is equal to:
Z zmax
(l,e)
(l,z)
Pf
=
Pf (x) fz (x) dx .
(2)
zmin
In Eq. (2), the familiar resistance and action terms may be recognised: the
(l,z)
first term, PF (x), is inversely proportional to the structural resistance under the
action x, while the second term fz (x) expresses the likelihood of action x. In short,
Eq. (2) says the probability of failure is obtained summing up the contributions at
each value of the action x. The distribution of failures in time is then a homogeneous
Poisson process with mean rate equal to:
(l,e)
l = Pf
(3)
The time elapsed between the beginning of exposure of the structure to earthquakes (at time t = 0) and the first failure(at time t = ) is a random variable
exponentially distributed:
F (t) = P ( t) = 1 el t .
(4)
Now, let T the time at which the retrofitting intervention is planned. Two
possibilities may occur: no failure before T (T < ) or failure before T (T ).
The present study allows to determine which T minimises the expected annual
equivalent cost AE considering both outcomes. Synthetically, the problem can be
expressed as:
Find T : {AE(T ) = min} .
(5)
00054
57
3. Solution Procedure and Assumptions

Once the value of T , the time at which the retrofitting intervention is made, is set,
two possibilities may occur: failure before T (T ) or no failure before T (T < ).
The event tree for the problem is thus composed of two branches only, A and B,
and is shown in Fig. 1.
Probabilities PA and PB are computed from Eq. (4):
PA = P [ T ] = F (t) = 1 el T
(6)
PB = 1 PA = el T .
The cash-flows for the two branches of the event tree are evaluated with the
following assumptions, which are discussed at the end of this section:
in branch A:
at time the structure fails. It is rebuilt immediately afterwards so that it can
be assumed that the service offered by the structure is never discontinued
after rebuilding structural fragility becomes negligible
in branch B:
during retrofitting, the service offered by the structure is never discontinued
after retrofitting structural fragility becomes negligible
in both branches
after rebuilding or retrofitting, the structure has a service life equal to Lpi ,
after which it has no economic value.
From the above assumptions, it follows that the benefits in time deriving from
utilisation of the structure are equal in both branches. Since the cash flows in the
two branches will be expressed in terms of their Annual Equivalent [Thuesen and
Fabrycky, 2000], benefits in the two branches can be canceled out.
The cash-flows, hence, simplify to a lump expenditure equal to the cost of failure
Ccollapse at time for branch A and a lump expenditure equal to the costs due to
retrofitting Cretrofitting at time T for branch B.
The cash-flow in branch A is shown in Fig. 2.

0
Fig. 1.
Event-tree for the problem.
&
'
*
'
'
"
58
00054
I. Vanzi
+Lpi
collapse
Economic life of the structure

Fig. 2.
Cash-flow in branch A of the event tree.
Ccollapse contains all the items necessary to restore full functionality for utilisation of the structure plus any other cost deriving from failure. It is composed
of fixed yard costs, demolition and rebuilding of the structure, service-disruption,
damage to third-parties and damage to the reputation of the owner of the structure.
After rebuilding, the structure has a remaining service life equal to Lpi ; on the
whole, the structural service life has been equal to the period + Lpi . In order
to compute the expected value of the Annual Equivalent AE of the cost of failure
Ccollapse , one first needs to find the statistics of . In branch A, the random variable
is conditioned to the event T and is distributed according to:
F (t)
F / T (t) = P [ t/ T ] = F (T )
tT
.
t>T
(7)
After some algebra, its first and second moments are computed as:
=
s
=
l T (el T + 1) 1
1 (el T 1)
el T + l 2 T 2 (el T 1) 3 + 3 el T e2l T
.
(el T 1)(el T 1)2 l 2
The Annual Equivalent AE to the cash-flow in Fig. 2 can be computed as follows

[Thuesen and Fabrycky, 2000]:
(i) compute the present value Ccollapse (0) of Ccollapse ( ).
The computation is made using in , the rate of interest discounted for the value
of the inflation f .
00054
1
Ccollapse (0) = Ccollapse ( )
(1 + in )
59
in (i, f ) = interest rate discounted for inflation =
if
1+f
(8)
i = money interest rate for the owner of the structure

f = inflation rate .
Notice that, for small values of f , in (i, f ) i f . In Eq. (8), the value for
Ccollapse ( ) can be directly estimated using the costs at the present time, without
having to consider inflation which is already accounted for in in .
It has further to be noticed that the value of i in Eq. (8), and in the other
equations, is not the money market interest rate but the so called internal rate
of return [Fabrycky et al., 1998], i.e. the money interest rate relative to the owner
of the structure.
(ii) compute, on the period + Lpi , the annual equivalent AE (A) to Ccollapse (0)

i (1 + i) +Lpi
(A)
= {Ccollapse (0)}
AE
(1 + i) +Lpi 1

(1 + i)
= {Ccollapse ( ) i (1 + i)Lpi }
. (9)
(1 + in ) ((1 + in ) +Lpi 1)
On the right hand side of the lowest part of Eq. (9), the terms contained in
the first braces are deterministic, whereas the remaining terms are function of the
random variable . Let:
K = Ccollapse ( ) i (1 + i)Lpi
g( ) =
(1 + 1)
(1 + in ) [(1 + i) +Lpi 1]
(10)
and substitute in Eq. (9). The annual equivalent is then equal to:
AE (A) = K g( ) .
(11)
The function g( ) is a random variable with unknown distribution. It is however

sufficiently accurate to compute its moments with a first order approximation for
the mean and a second order approximation for the variance [Ang and Tang, 1975].
(
)
1
2 g( )
g( ) g( ) +
2
2
2
(12)
(
)2
g(
)
2
g(
2 .
)

The moments of AE (A) are equal to:
AE (A) = K g( )
AE (A) = K g( ) .
(13)
60
00054
I. Vanzi
+Lpi
retrofitting
Economic life of the structure

Fig. 3.
Cash-flow in branch B of the event tree.
The cash-flow in branch B of the event tree is analogous to the one in branch
A and is shown in Fig. 3.
The structure is seismically retrofitted at time T ; the cost associated with
retrofitting, Cretrofitting , contains all the costs necessary to seismically upgrade the
structure, e.g. fixed yard costs and retrofitting operations. After retrofitting, the
structure has a remaining service life equal to Lpi ; on the whole, the structure has
carried out its service for a period equal to T + Lpi .
The value of the annual equivalent is computed with Eqs. (10) and (11), substituting Cretrofitting (T ) to Ccollapse ( ) and T to . The problem, in this case, is fully
deterministic.
Now, since the Annual Equivalents on the two branches of the tree are homogeneous values and are independent on the length of the period of study [Fabrycky
et al., 1998], the mean value of the annual equivalent in node O of Fig. 1 can be
computed as:
AE (O) = PA AE (A) + PB AE (B) .
(14)
Substituting in Eq. (14) the expected value of Eq. (13), and the analogous one
for branch B, the (long and not further simplifiable) expression for AE (O) written
in Appendix A is found. AE (O) depends on:
Costs, evaluated at time t = 0: Cretrofitting , Ccollapse .
Rates, assumed constant over time: i = internal rate of return, f = inflation rate.
Temporal distribution of failures with reference to limit state l: l = mean rate
of structural failures.
Time at which the retrofitting intervention is planned: T .
Service life of the structure after retrofitting or reconstruction: Lpi .
AE (O) (Cretrofitting , Ccollapse , i, f, l , T, Lpi ) .
(15)
00054
61
Now, if one assumes f constant and equal to fc , and sets Ccollapse = r Cretrofitting ,
Eq. (15) may be minimized as a function of T , yielding the value of:
T (Cretrofitting , r, fc , l , T, Lpi )
= T : {AE (O) (Cretrofitting Ccollapse = r Cretrofitting , i, fc , l , T, Lpi) = min .
(16)
Notice further that the value of T (Cretrofitting , r, i, l , Lpi ) in Eq. (16) is independent on Cretrofitting (see also Appendix A), since expression (15) is a linear
function of Cretrofitting once Ccollapse has been set equal to r Cretrofitting .
T (.) also proves very weakly dependent on the value of Lpi even for large
variations of the variable: with Lpi in the range 50100 years, the difference between
the maximum and minimum values of T (.) is below 5% of the minimum value. Once
this is realised, the simplification:
T (Cretrofitting , r, i, l , Lpi ) = T (l , r, i)
(17)
can be adopted.
In the following section, a parametric study on T (l , r, i) will be performed
and the results for T obtained numerically will be shown.
Three important points need to be discussed before passing to the next section:
the assumptions made in the two branches of Fig. 1, which are reasonable values
for r and how to deal with retrofitting decisions conditioned by several limit states.
As for the first point, the assumptions made allow to model the problem in a
very agile way. As a drawback, when applying the results of the study, attention
must be paid to checking whether the real problem considered is compatible with
the assumptions.
The assumption of service never discontinued could be easily removed from the
procedure and would just imply a time shift in the algebra to discount the capital
back and forth in time [e.g. exponents in Eq. (9)]. However, provided that the
time during which, in reality, service is discontinued is small compared to Lpi , the
economic life of the structure, this assumption is realistic.
The assumption of negligibility of the structural fragility after retrofitting/
rebuilding must be checked more closely. Its applicability depends both on the
characteristics of the rebuilding/retrofitting intervention, on the limit state considered and on the hazard at the site. As a first indication it can be said that if the
(1)
mean rate of failures [Eq. (3)] computed after the intervention, l (consider the
one relative to retrofitting, which normally is the highest one between rebuilding
(1)
and retrofitting) is such that the computed value of T (l , r, i) is high, say 3 Lpi
or more, the procedure can still be applied.
As for the second point, the values for r are extremely variable, depending on
the structure and on the limit state considered. As a general remark, high values, higher than 50, are typical of simple retrofitting interventions on costly (or
precious in case of an earthquake) equipments. Examples are equipment located
62
00054
I. Vanzi
in strategic infrastructures (components of the stations of electric networks [Vanzi,

1996], medical equipment located in hospitals, telecommunication equipment), deck
unseating in bridges, structural failures in buildings having/containing artistic values. As an example, consider a voltage transformer located in a high-tension electric
station. Overturning is a frequent collapse mode and can be prevented anchoring
the transformer to the ground with steel bolts. Cretrofitting can be estimated in
US$5 thousands; since overturning destroys this piece of equipment, which costs
US$500 thousands, and may cause electric power black out to a wide area after an
earthquake, r is in excess of 500 000/5000 = 100. On the other hand, low values,
equal to some units, are typical of complex retrofitting interventions on lower values
structures. An ordinary masonry building, provided that life protection is ensured,
shows values of r of about 5.
Finally, the procedure proposed can give useful information when several limit
states are considered. If the different limit states are such that their T (.) are close,
and there is synergy among the different interventions, then it is convenient to
schedule all of them at the same time. Different outcomes must instead be examined
on a case by case basis, taking into account the interaction between interventions
relative to different limit states, but the plan of interventions can be in most cases
more rationally organised considering also the information on T (.).
4. Parametric Study
The procedure which yields the values of T (l , r, i) in Eq. (17) has been programmed and numerically solved with the following values:
fc = 0.02
l [104 ; 101 ]
r [1; 100]
i [0.1; 0.3].
For the sake of conciseness, the results for T are shown only for i equal to 0.1
and 0.3 in Figs. 4 and 5. These values for i should contain most real cases.
In each figure, the contour lines for constant values of T (l , r, i), between 0 and
95 years with 5 years step, are shown. On the abscissas and ordinates, the values of
l , in logarithmic scale, and of r = Ccollapse /Cretrofitting , in natural scale, evaluated
at time = 0, are respectively shown.
In the figures, three regions, A, B and C, can be identified. In region A,
T (l , r, i) = 0 while in region B T (l , r, i) > 95 years. In region C, T (l , r, i)

depends on the values of l and r.
Notice that, for r and i fixed, lower values of T (l , r, i) correspond to higher
values of the mean rate of failure; this is equivalent to saying that, in areas of high
seismicity and/or for highly fragile structures, retrofitting must be done before than
what should be done for low seismicity and/or fragility.
00054
63
Fig. 4.
Contour lines at constant T (l , r, i) for i = 0.1.
Comparing now Figs. 4 and 5, it can be seen that an increase in the value
of the internal rate of return i, for r and l fixed, generally causes an increase
of T (l , r, i) (compare the amplitudes of regions B). This happens because when
capital is more efficiently allocated, as measured by higher values of the interest
rate i, postponing any other investment is mathematically worthwhile, even if the
risk of failure augments with time.
At the border between regions A and B, T (l , r, i) is very steep and highly
sensitive to the values of r. This happens for high values of l , higher than about
0.01. Such a high mean rate of annual failure should be avoided altogether unless
the estimated values of r are well away from the border.
For real cases, in which the values of (l , r, i) may differ from those shown in
Figs. 4 and 5, the results for T (l , r, i) must be calculated with the minimization
procedure for AE (O) (.) of Eq. (15) and Appendix A, as a function of T .
Next, the values of AE (O) (Cretrofitting , Ccollapse = r Cretrofitting , i, f, l , T, Lpi )
of Eq. (15) are shown, for the following values for the variables:
Cretrofitting = 1
r [1; 100]
i = 0.1
f = 0.02
64
00054
I. Vanzi
Fig. 5.
Contour lines at constant T (l , r, i) for i = 0.3.
100
90
80
70
60
50
40
30
0.1
8
0. 0
6
0. 0
04
0.
0.
02
10
-3
0
2
0. 0 4
0. 0 . 08
60
0. 0
20
-2
10
10
Fig. 6.
Contour lines at constant AE (O) (.) for Lpi = 50 years.
-1
10
00054
65
l [104 ; 101 ]
T = T (l , r, i)
Lpi = 50 years.
The minima of the expected values of the Annual Equivalent are equal to
some percentage points of Cretrofitting and, obviously, the longer Lpi , the lower
AE (O) (.).
Finally, to have a feeling of how much the optimization procedure can save,
the value of AE (O) (Cretrofitting , Ccollapse = r Cretrofitting , i, f, l , T, Lpi) with the
following values for the variables:
Cretrofitting = 1
r = 23
i = 0.1
f = 0.02
l = 1.7 103
T [0.01; 2] T (l , r, i)
Lpi = 50 years.
The value of T (l , r, i) for this last case is equal to 32 years.

Fig. 7. AE (O) with Cretrofitting = 1, r = 23, i = 0.1, f = 0.02, l = 1.7 103 , T [0.01; 2]
T (l , r, i), Lpi = 50 years.
66
00054
I. Vanzi
Notice that, within reasonable values for T , optimization can save up to 0.5
the cost.
5. An Example: The Hospital of Fivizzano in Tuscany, Italy

A simple application, which contains all the steps of the method, is presented in this
section. It deals with an existing structure, the hospital of Fivizzano in Tuscany,
Italy. The hospital is composed of nine buildings, some in masonry and some in
reinforced concrete, built between 1926 and 1974, with height of three or four floors.
A schematic plan of the hospital, drawn from Ferrini et al. [2000], is shown in Fig. 8.
The immediate occupancy limit state, as defined by the Applied Technology
Council [1996], is considered since the hospital is a strategic infrastructure after the
earthquake. For this limit state, the functional scheme of the hospital considering
both the structural and non structural components (e.g. partitions and medical
equipment) was modelled.
To give an example of the procedure, Fig. 9 shows the functional scheme for the
equipment of the hospital. The reader can notice that the scheme is organised as a
chain, each component (e.g. emergency generator) being a ring of the chain.
Using the traditional methods of structural reliability, for which the reader is
referred to Ang and Tang [1975], for a general introduction on the topic or to Ferrini
K
#

"
'
'
#
$
Fig. 8.
Plan of the hospital of Fivizzano.
<
00054
67
(48,30(17
(/(&75,&
7UDQVIRUPHU
EXLOGLQJ < JURXQG IORRU
(PHUJHQF\ JHQHUDWRU
(OHFWULF FRQQHFWLRQV
EXLOGLQJ 5
:$7(5
7DQN
([WHUQDO ZDWHU OLQH
&RQQHFWLRQ ZLWK LQWHUQDO OLQH

:DWHU OLQH WR EXLOGLQJ %
:DWHU OLQH WR EXLOGLQJ :
7DQN 0
0(',&$/ *$6(6
/LQH IURP WDQN 0 WR FRPSUHVVRU 1

&RPSUHVVRU 1
/LQH IURP FRPSUHVVRU 1 WR EXLOGLQJ %
/LQH IURP EXLOGLQJ % WR EXLOGLQJ 5
Fig. 9.
Logical scheme for the equipment in the Fivizzano hospital.
et al. [2000] and Nuti et al., [2001] for the reliability of hospitals in particular, the
fragility curve of the system was determined. The curve, shown in Fig. 10 with the
thicker line, gives the value of the probability of failure of the system (this event
being for the case at hand the impossibility to use the hospital after an earthquake)
as a function of the seismic intensity (the peak ground acceleration was chosen).
Figure 10 shows also a different fragility curve, indicated with the letter C. This
is relative to the system after retrofitting and will be used later in this example.
68
00054
I. Vanzi
-/
-.
*+
Fig. 10.
"
&
&
Fragility curve of the hospital of Fivizzano for the immediate occupancy limit state.
'
+,
(
%&'
"
#$
!
Fig. 11.
Hazard curve in Fivizzano.
Seismic intensity for the site is defined via the hazard curve, shown in Fig. 11.
The curve shows the values of the expected peak ground acceleration as a function
of the return period.
The hospital is situated in an area of low seismicity but its fragility is high, since
the immediate occupancy limit state is considered and both the structure and its
content are not designed for horizontal forces.
00054
69
The values of zmin , zmax have been assumed respectively equal to 0.3 m/s2 and
2.5 m/s2 . Below 0.3 m/s2 , in fact, the structural fragility is disregardable; while
2.5 m/s2 is the asymptotic value for the hazard curve of Fig. 9. [Eq. (1)] is
estimated equal to 2.81, from the data shown in Fig. 11. The mean annual rate of
earthquakes (see Sec. 2) is equal to 0.017. The value of Pfl,e [Eq. (2)] is 0.78 and
the mean annual rate of failures l [Eq. (3)] is 0.013.
The values of T (l , r, i) are shown in Fig. 12 as a function of r, assuming the
value of 0.1 for the internal rate of return i.
For values of r greater than seven, which is the likely outcome for a hospital
with the immediate occupancy limit state, the best option would have been to build
since the beginning a less fragile structure.
In Ferrini et al. [2000], three different retrofitting strategies, A, B and C, had
been identified for this hospital, with increasing costs and benefit. Strategy C looked
the most promising, requiring some interventions on the equipment and structural
upgrade in only two buildings, and delivering a significant decrease of the structural
fragility curve. Assume that upgrade C has been independently adopted by the
owner of the structure. Will it suffice or more incisive interventions are still needed?
This can be checked via a further application of this procedure. Convolution of the
fragility curve after adoption of upgrade C (Fig. 10, curve C) with the distribution
(l,e)
and of 0.0068
of the earthquake intensity [Eq. (2)] gives the value of 0.403 for Pf
for the mean annual rate of failures l [Eq. (3)]. From Fig. 4, one can read that
for r 17, T (.) equals 0. Considering the strategic importance of hospitals right
after an earthquake, especially when they are located in a rural (no other hospital
Fig. 12.
T (, r, i) versus r.
70
00054
I. Vanzi
in the neighbourhood) and hilly (long transportation times) area, like this one, this
upgrading level would still be insufficient.
6. Conclusions
The method presented in this paper answers in a simple format a central question
in seismic retrofitting: is it worth doing it? Would it not be better to choose the
easy option of doing nothing to seismically retrofit structures and simply wait and
see what happens?
The answer is a rocksolid it depends. This is intuitively understandable, since
the convenience of structural retrofit depends on a number of things, firstly the
considered limit states, then the structural risk, the costs of the consequences of
a structural failure, and the cost of borrowing (or diverting from other uses) the
capital employed in retrofitting. Here, the problem is inserted in a coherent frame
which says when retrofitting is economically best and also quantifies the expected
costs deriving from different choices of intervention time.
The model presented is set-up departing from strong simplifying assumptions,
stated in Sec. 3, that, though questionable, allow to derive results in a straightforward manner. Only three independent variables, structural risk, interest rate and
ratio of failure cost to retrofitting cost, are retained to evaluate the final results,
the year in which the retrofitting intervention is most economically implemented.
The main results are that the convenience of retrofitting strongly depends on
these three parameters; and that the year in which it is best to upgrade the structure can be computed (respecting the assumptions made) minimising AE (O) (.) in
Appendix A. Finally, depending on the combination of the three main variables,
the right choice of the time of intervention may significantly reduce overall costs
(due to retrofitting and to the risk of failure), by half or more.
Appendix A
The translation of AE (O) (Cretrofitting , Ccollapse , i, f, l , T, Lpi) of Eq. (15) in fortran
code is as follows:
t1 = lambda*T
t2 = exp(-t1)
t3 = 1-t2
t4 = 1+i
t10 = -1/t3
t11 = (t2*lambda*T+t2-1)/lambda*t10
t12 = t4**t11
t15 = t4/(1+f)
t16 = t15**t11
t17 = 1/t16
00054
71
t18 = t12*t17
t20 = t4**(t11+Lpi)
t21 = t20-1
t22 = 1/t21
t24 = lambda**2
t25 = T**2
t26 = t24*t25
t27 = exp(t1)
t31 = exp(2*t1)
t38 = alog(t4)
t39 = t38**2
t40 = t12*t39
t41 = t17*t22
t44 = alog(t15)
t48 = t21**2
t49 = 1/t48
t54 = t44**2
t64 = t20**2
t75 = t4**Lpi
t78 = t4**T
t80 = t15**T
t84 = t4**(T+Lpi)
t91 = (t3*(t18*t22+(t2+t26*t27-3+3*t27-t26-t31)*t10/t24/(t27-1)*(t
#40*t41-2*t12*t38*t41*t44-3*t40*t17*t49*t20+t18*t22*t54+2*t18*t49*t
#44*t20*t38+2*t18/t48/t21*t64*t39)/2)*Ccollapse*t75+t2*t78/t80/(t84
#-1)*Cretrofitting*t75)*i
The last expression, t91, equals AE (O) (Cretrofitting , Ccollapse , i, f, l , T, Lpi ).
Appendix B: Meaning of the Symbols Used in the Paper

AE
Ccollapse ( )
Cretrofitting (T )
f
fX (y)
FX (y)
i
in
l
Lpi
annual equivalent cost

cost due to structural failure at time
cost due to retrofitting at time T
inflation rate
probability distribution function of the random variable X,
computed at the value y
cumulative probability distribution function of the random
variable X, computed at the value y
interest rate
interest rate discounted for inflation
limit state
economic life of the structure after retrofitting or rebuilding
72
00054
I. Vanzi
probability
(l,z)
Pf (z)
probability of structural failure with respect to limit state l given

an earthquake
(l,z)
Pf
(z)
r
t
T
T
z
zmin
zmax
l
X
probability of structural failure with respect to limit state l

given intensity z i.e. structural fragility curve
ratio between cost of failure and cost of collapse
time
time at which the structure is retrofitted
T : AE(t) is minimum
earthquake intensity
minimum value considered for the seismic intensity at the site
maximum value considered for the seismic intensity at the site
severity parameter in the GutenbergRichter law
mean rate of Poisson process of failures
mean value of the random variable X
mean frequency of earthquakes in time
standard deviation of the random variable X
time at which the first failure occurs.
References
Ang. A. H. S. and Tang, W. H. [1975] Probability concepts in engineering planning and
design, Decision, Risk and Reliability, Vol. 2 (John Wiley and Sons, New York).
Applied Technology Council, [1996] Seismic evaluation and retrofit of concrete buildings,
report SSC 96-01, USA.
Chang, S. E. and Shinozouka, M. [1996] Life-cycle cost analysis with natural hazard risk,
J. Infrastructure Syst. ASCE 2(3).
Ciampoli, M. [1998] Time dependent reliability of structural systems subject to
deterioration, Comput. Struct. 67, 2935.
Cornell, C. A. [1968] Engineering seismic risk analysis, Bull. Seism. Soc. Am. 58,
15831606.
Der Kiureghian, A. [1999] A Bayesan framework for fragility assessment, Proc. Icasp
8th Conference, Sydney, Australia.
Donferri, M., Giannini, R., Nuti, C. and Pinto, P. E. [1998] Analysis of seismic risk on
the bridges of Autostrade network, Autostrade 2, 715.
Fabrycky, W. J., Thuesen, G. J. and Verma, D. [1998] Economic Decision Analysis
(Prentice Hall, New Jersey).
Fema-227, [1992] A benefit-cost model for the seismic rehabilitation of buildings, Vols. 1
and 2 (Federal Emergency Management Report, Sacramento).
Ferrini, M., Nuti, C. and Vanzi, I. [2000] Seismic risk assessment of two existing hospitals
in Italy, XII World Conference on Earthquake Engineering, Auckland, New Zealand.
Giannini, R., Nuti, C., Ortolani, F. and Pinto, P. E. [1981] Seismic risk analysis of a
highway system, European Simulation Meeting, Modeling and Simulation of LargeScale Structural Systems, Capri, Italy.
ISTAT [1999] Annuario statistico italiano, Istituto Nazionale di Statistica, Roma.
00054
73
Kanda, J. and Shah, H. [1997] Engineering role in faiulre cost evaluation for buildings,
Struct. Safety 19(1), 7990.
Lund, P. E. [1996] Utility lifeline lessons learned, January 17, 1994, Northridge
earthquake, 11th W.C.E.E., Acapulco, Mexico.
Mori, Y. and Ellingwood, B. [1993] Methodology for reliability based condition assessment. Application to concrete structures in nuclear power plant, NUREG/CR-6052ORNL/Sub/93/-SD684.
Nuti, C. and Vanzi, I. [1998] Assessment of post-earthquake availability of hospital system
and upgrading strategies, Earthq. Eng. Struct. Dyn. 27, 14031423 (software on the
topic can be downloaded from www.vanzi.it).
Nuti, C., Santini, S. and Vanzi, I. [2001] Seismic risk of the Italian hospitals, Eur.
Earthq. Eng. n.1/2001, Patron publ.comp., Bologna, Italy.
O Rourke, M. and Palmer, C. [1994] The Northridge California earthquake of January
17, 1994; performance of gas transmission pipelines, N.C.E.E.R. Tech. Rep. 940011.
O Rourke, T. D. [1996] Lessons learned for lifeline engineering from major urban
earthquakes, 11th W.C.E.E., Acapulco, Mexico.
Schiff, A. J. [1988] The Whittier Narrows, California earthquake of October 1, 1987
response of lifelines and their effects on emergency response, Earthq. Spectra 4(2).
The Economist, [1995] Fear of trembling, a survey of earthquake engineering, print
edition of 4.22.1995, The Economist, London.
Thuesen, G. J. and Fabrycky, W. J. [2000] Engineering Economy (Prentice Hall,
New Jersey).
Tiedemann, H. [1992] Earthquakes and Volcanic Eruptions, a Handbook on Risk
Assessment, Swiss Re, Zurich.
Vanzi, I. [1996] Seismic reliability of electric power networks: methodology and
application, Struct. Safety 18(4), 311327.
Vanzi, I. [2000] Structural upgrading strategy for electric power networks under seismic
action, Earthq. Eng. Struct. Dyn. 29(7), 10531073 (software on the topic can be
downloaded from www.vanzi.it).
00060


SEISMIC HAZARD IN GREECE BASED ON DIFFERENT

STRONG GROUND MOTION PARAMETERS
S. I. KOUTRAKIS, G. F. KARAKAISIS and P. M. HATZIDIMITRIOU

Geophysical Laboratory, Aristotle University of Thessaloniki,
GR 540 06, Thessaloniki, Greece
P. K. KOLIOPOULOS
Technological Education Institute of Serres,
Eng. Mechanics Div., 62124 Serres, Greece
V. N. MARGARIS
Institute of Engineering Seismology and Earthquake Engineering (ITSAK),
P.O. Box 53, GR 551 02 Finikas, Thessaloniki, Greece
Received 19 February 2001
Revised 4 July 2001
Accepted 30 July 2001
The available Greek strong ground motion records to date are used in order to study
the duration of strong-motion in Greece, covering magnitudes between 4.5 and 6.9 and
distances from 1 km to 128 km. An attenuation relation of strong-motion duration is
calculated and compared to earlier existing similar relations proposed for Greece and
Japan. Furthermore, the seismic hazard for the area of Greece is assessed, using the
strong-motion parameters of duration and peak ground acceleration. The results are
presented in the form of a map according to which Greece is classified in four different
categories of equal seismic hazard.
Keywords: Strong ground motion duration; attenuation relations; probabilistic hazard
assessment.
1. Introduction
The assessment of seismic hazard in Greece has become the subject of a growing number of studies in the last 25 years. These studies are mainly concerned
with the evaluation of seismic hazard of macroseismic intensities [e.g. Papazachos
et al., 1985; Papaioannou, 1986; Papoulia and Slejko, 1997; Papaioannou and
Papazachos, 2000], as well as with the estimation of peak ground accelerations, velocities and strong-motion durations [for example, Algermissen et al.,
1976; Drakopoulos and Makropoulos, 1983; Papazachos et al., 1990, 1992, 1993;
Theodoulidis and Papazachos, 1992]. Based on the results of these publications, a
map was compiled according to which Greece is divided into four zones of equal
seismic hazard [Papazachos et al., 1993]. This map was proposed jointly, by four
75
76
00060
S. I. Koutrakis et al.
Fig. 1.
Code.
The division of Greece into zones of equal seismic hazard according to the Greek Seismic
different seismological research institutions and constitutes part of the Seismic Code
of Greece since 1995 (Fig. 1).
Several studies have also been made concerning duration and energy characteristics of strong ground motion in Greece, due to the interest of this issue to both
seismologists and engineers. Theofanopoulos and Drakopoulos [1986] accomplished
the first of these studies, and a new measure of strong-motion duration was introduced and was subsequently used to study a number of major earthquakes that
occurred in Greece. Margaris et al. [1990] attempted to utilise the strong-motion
database available at that time to study the duration of strong ground-motion in
Greece. More specifically, the bracketed and significant duration of strong-motion
were computed and their relation to rupture duration was studied. Furthermore, relations were defined to associate bracketed duration with macroseismic intensity and
peak ground acceleration. Finally, an attempt was made to calculate attenuation
00060
Seismic Hazard in Greece
77
relations for these two different measures of strong-motion duration. Papazachos

et al. [1992] studied the bracketed duration if strong-motion and proposed a corresponding attenuation relation. Koliopoulos et al. [1998] studied various measures
of strong-motion duration along with energy characteristics of Greek strong-motion
records. Koutrakis et al. [1999] and Koutrakis [2000] proposed a new attenuation
relation for the bracketed duration in Greece and made an effort to use this duration
measure for the assessment of seismic hazard.
The fact that duration is an important parameter that affects both inelastic deformation and energy dissipation demand, is well established [Bommer and
Martinez-Perreira, 1999]. However, there is no generally accepted procedure for
duration to be taken into account in design codes. One attempt to account duration, is via the construction of 3-D response spectra [Safak, 1998; Koliopoulos
and Margaris, 2001]. Yet, another attempt is to compute energy demand spectra,
which incorporate the effect of duration [Chapman, 1999]. Nevertheless, the implementation of the resulting (3-D or energy) spectra into the design process is yet
to be established. A third line of approach, adopted herein, is to firstly perform a
duration-based seismic hazard assessment that will modify the PGA seismic hazard
results. This methodology will in fact, introduce penalty factors in the PGA values
of seismic zones with long duration seismic events. The resulting seismic hazard
assessment has the advantage to be expressed in terms of modified PGA values
and conventional design spectra, and therefore does not introduce any significant
deviation from standard design process.
Firstly, in order to assess seismic hazard in Greece based on strong-motion duration, the evaluation of an accurate attenuation relation for strong-motion duration
using all strong-motion data available for Greece was performed. Secondly, the assessment of seismic hazard has also been made for the parameter of peak ground
acceleration, using earlier published attenuation relations. Finally, the composite
result of these calculations is presented in a form of a map where Greece is divided
into four equal seismic hazard categories.
2. Strong-motion Data Used
The dataset used in the present study, summarised in Table 1, consists of strongmotion records provided by the Institute of Engineering Seismology and Earthquake
Engineering (ITSAK), the Geophysical Laboratory of the Aristotle University of
Thessaloniki and the Geodynamic Institute of the National Observatory of Athens.
These records span a time range of 20 years, from 1978 to 1997. They consist of 141
sets of two horizontal components of strong-motion acceleration, representing 93 different seismic events. The majority of these recordings correspond to surface events,
with the exception of one earthquake (two records in total), which is of intermediate
depth. Due to the fact that focal depths of Greek seismic events are only qualitatively known (http://geohazards.cr.usgs.gov/iaspei/europe/greece/the), this information is not included in Table 1. In the seismic hazard analysis, shallow and
78
00060
Table 1.
Parameters of the earthquakes used in the present study.
EQ.
DATE
OR. TIME
LAT.
LONG.
Mw
REC. CODE
NC
REF.
780620
20:03:21
40.80
23.20
6.4
THE78-1
1, 2
2
3
4
5
6
800716
800811
800926
810224
830117
00:23:30
09:15:59
04:19:18
20:53:38
12:41:29
39,25
39,31
39,26
38.22
38.09
22,75
22,86
22,75
22.93
20.19
5.3
5.6
5.2
6.6
6.9
7
8
9
830117
830119
830131
16:53:30
00:02:14
15:27:00
38.11
38.17
38.11
20.37
20.23
20.30
5.1
5.8
5.4
10
11
12
13
830220
830316
830323
830323
14:42:29
21:19:39
19:04:00
23:51:06
37.76
38.80
38.78
38.33
21.11
20.88
20.83
20.22
5.2
4.9
5.3
6.2
14
15
830324
830806
04:17:32
15:43:53
38.10
40.18
20.49
24.73
5.4
6.6
16
830826
12:52:10
40.51
23.92
5.1
17
18
840219
840709
03:47:22
18:57:10
40.61
40.69
23.40
21.82
4.9
5.2
19
20
841004
841025
10:15:12
09:49:16
37.64
36.83
20.85
21.71
5.2
5.3
21
22
23
850322
850322
850831
20:37:39
20:38:54
06:03:46
38.98
38.91
38.99
21.11
21.06
20.59
4.8
4.5
5.5
24
851109
23:30:42
41.24
23.93
5.2
25
26
27
860218
860913
860915
14:34:03
17:24:34
11:41:30
40.77
37.03
37.04
22.11
22.20
22.13
5.2
5.9
5.5
29
32
34
870610
880303
880424
14:50:11
11:38:58
10:10:33
37.17
38.87
38.83
21.39
21.23
20.54
5.5
5.0
4.9
ALM80-2U
ALM80-4U
ALM80-5U
KOR81-1
AGR83-1
LEF83-1
ARG83-1
ARG83-3
ARG83-4
ARG83-6
ZAK83-1
ZAK83-2
LEF83-2
LEF83-3
ZAK83-3
LEF83-4
ARG83-7
ARG83-8
POL83-1
IER83-2
POL83-2
IER83-3
POL84-1
EDE84-1
VER84-1
ZAK84-1
KYP84-2
PEL84-1
AMF85-3
AMF85-4
LEF85-1
PRE85-1
DRA85-1
KAV85-1
EDE86-1
KAL86-1
KAL86-8
KAL286-2
MES186-1
KYP87-1
AMF88-2
LEF88-2
3, 4
5, 6
7, 8
9, 10
11, 12
13, 14
15, 16
17, 18
19, 20
21, 22
23, 24
25, 26
27, 28
29, 30
31, 32
33, 34
35, 36
37, 38
39, 40
41, 42
43, 44
45, 46
47, 48
49, 50
51, 52
53, 54
55, **
56, 57
58, 59
60, 61
62, 63
64, 65
66, 67
68, 69
70, 71
72, 73
74, 75
76, 77
78, 79
80, 81
82, 83
84, 85
2
2
2
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
00060

Table 1.
(Continued)
EQ.
DATE
OR. TIME
LAT.
LONG.
Mw
REC. CODE
35
880518
05:17:42
38.36
20.42
5.7
ARG88-2
36
37
880705
881016
20:34:51
12:34:06
38.11
37.95
22.85
20.90
4.6
5.8
40
41
890831
901221
21:29:30
06:57:44
38.08
40.98
21.81
22.34
4.7
6.1
42
43
44
46
47
49
50
51
52
53
54
55
57
910319
910319
920123
920530
921110
930326
930326
930326
930326
930326
930429
930604
930714
12:09:26
21:29:26
04:24:00
18:55:38
22:14:59
11:45:16
11:56:13
11:58:15
12:26:32
12:49:17
07:54:29
03:24:27
12:31:49
34.67
34.74
38.40
37.93
38.78
37.66
37.69
37.49
37.55
37.77
37.40
38.70
38.24
26.35
26.37
20.57
21.44
20.66
21.39
21.43
21.49
21.27
21.33
21.58
20.45
21.78
5.7
5.0
5.4
5.0
4.8
5.3
5.1
5.4
5.2
4.9
5.8
4.8
5.6
58
940225
02:30:49
38.76
20.55
5.5
60
940523
06:46:12
35.16
24.59
5.7
61
62
941126
950504
14:30:30
00:34:11
38.87
40.54
20.48
23.63
5.1
5.4
63
950513
08:47:15
40.16
21.67
6.5
64
65
950514
950515
14:46:57
04:13:57
40.13
40.07
21.66
21.67
4.9
5.2
KOR88-2
ZAK88-4
AML88-6
PAT189-3
EDE90-1
KIL90-1
SIT91-1
SIT91-2
ARG92-1
PAT92-1
LEF92-1
PYR93-6
PYR93-7
PYR93-8
PYR93-10
PYR93-11
PYR93-17
VAS93-2
PAT193-2
PAT393-2
LEF94-1
VAS94-1
HAN94-1
HRA94-1
LEF94-6
POL95-6
IER95-5
PRO16-14
SAR95-1
STC16-18
STE12-18
TST16-12
KOZ195-1
KAR195-1
KRP195-1
KAS195-1
VER195-1
FLO195-1
CHR195-5
CHR195-13
KOZ295-4
NC
86, 87
88,
90,
92,
94,
96,
98,
100,
102,
104,
106,
108,
110,
112,
114,
116,
118,
120,
122,
124,
126,
128,
130,
132,
134,
136,
138,
140,
142,
144,
146,
148,
150,
152,
154,
156,
158,
160,
162,
164,
166,
168,
89
91
93
95
97
99
101
103
105
107
109
111
113
115
117
119
121
123
125
127
129
131
133
135
137
139
141
143
145
147
149
151
153
155
157
159
161
163
165
167
169
REF.
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
79
80
00060
Table 1.
EQ.
DATE
OR. TIME
LAT.
(Continued)
LONG.
Mw
66
67
68
950515
950516
950516
08:17:00
23:00:42
23:57:28
40.11
40.02
40.09
21.50
21.56
21.62
4.6
4.7
5.1
69
950517
04:14:26
40.07
21.61
5.2
70
950517
09:45:07
40.01
21.56
4.8
71
72
950518
950519
06:22:55
06:48:50
40.03
40.03
21.56
21.62
4.5
5.0
74
950520
21:06:25
40.00
21.58
4.5
75
950606
04:36:00
40.14
21.61
4.7
78
79
80
950717
950718
960804
23:18:15
07:42:55
08:03:21
40.10
40.12
40.04
21.58
21.61
20.70
5.0
4.5
4.7
81
960804
10:08:47
40.04
20.70
4.8
82
960805
22:46:42
40.06
20.66
6.3
83
960805
23:58:46
40.03
20.72
4.5
84
85
960806
960811
05:13:51
07:57:15
40.03
40.08
20.69
20.73
4.5
5.4
86
960811
08:30:27
40.04
20.74
4.5
87
88
960820
960821
01:26:46
07:10:34
40.04
40.02
20.70
20.78
5.4
4.7
REC. CODE
NC
REF.
KOZ195-7
170, 171
GRE195-1
GRE195-3
CHR195-29
CHR195-30
KOZ295-6
KOZ295-7
KOZ195-8
GRE195-9
CHR195-32
GRE195-11
CHR195-34
CHR195-35
KOZ295-13
KPE195-1
GRE295-1
GRE195-12
KPE195-3
KND195-1
KPE195-7
KND195-4
KEN195-55
CHR195-49
KOZ295-22
GRE195-20
GRE195-22
KONL-028
KONU-105
KONL-029
KONU-106
KONL-032
KONU-107
KONL-039
KONU-108
KONU-109
AETP-005
KONL-070
KONU-123
KONL-074
KONU-127
AETP-010
KONL-101
172, 173
174, 175
176, 177
178, 179
180, 181
182, 183
184, 185
186, 187
188, 189
190, 191
192, 193
194, 195
196, 197
198, 199
200, 201
202, 203
204, 205
206, 207
208, 209
210, 211
212, 213
214, 215
216, 217
218, 219
220, 221
222, 223
224, **
225, 226
227, **
228, 229
230, 231
232, 233
234, **
235, **
236, 237
238, 239
240, **
241, 242
243, **
244, 245
246, 247
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
00060

Table 1.
81
(Continued)
EQ.
DATE
OR. TIME
LAT.
LONG.
Mw
REC. CODE
89
960901
07:41:45
40.07
20.72
4.7
90
960901
21:15:00
40.01
20.74
5.0
91
960926
12:31:49
40.05
20.75
5.4
92
971013
13:39:46
36.10
22.04
6.4
93
971118
13:07:53
37.33
20.84
6.6
NC
REF.
KONL-112
248, 249
KONU-132
AETP-013
KONL-117
KONU-134
AETP-016
KONL-132
KONU-137
KRN197-2
KYP197-2
KAL197-1
GTH197-1
ANS197-1
ZAK197-3
KYP197-3
KRN197-3
PYR197-1
KAL197-2
250, **
251, 252
253, 254
255, 256
257, 258
259, 260
261, 262
263, 264
265, 266
267, 268
269, 270
271, 272
273, 274
275, 276
277, 278
279, 280
281, 282
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1. Geodynamic Institute, Athens National Observatory.

2. Geophysical Laboratory, Aristotle University of Thessaloniki.
3. Institute of Engineering Seismology and Earthquake Engineering.
The recording instrument was not functioning.
intermediate depth earthquakes are considered through the seismotectonic model

[Papaioannou and Papazachos, 2000] assuming shallow and intermediate depth seismic sources. For the shallow seismic sources a mean depth of approximately 10 km
and for intermediate depth seismic sources mean depths of 80 and 140 km are taken
into consideration. The magnitudes used are taken from Papazachos et al. [1997] and
range between 4.5 and 6.5. In the above study, an effort was made to produce a statistically homogeneous catalogue of historical and recent earthquakes of the Aegean
area expressing the size of these earthquakes in the moment-magnitude scale. It was
shown that the proposed magnitude M is equivalent to moment magnitude, Mw ,
for a wide range of earthquake sizes (5 M 8) and to the surface-wave magnitude, Ms , for large earthquake sizes (6 M 8). Concerning seismic events with
magnitude lower than 5, moment-magnitude relations [Margaris and Papazachos,
1999] based on strong-motion records are used.
The distance considered herein, is the epicentral distance. Although other relevant studies use the closest distance to the fault rupture or to the surface projection of the rupture, these distances could not be calculated in the majority of the
cases considered in the present study since the fault rupture was not adequately
determined. In addition, due to the uncertainty of the focal depth determination,
82
00060
the application of the hypocentral distance was avoided. Thus, the epicentral distance was adopted, varying from 1 to 128 km. The distribution of the records in
terms of epicentral distance and earthquake magnitude are shown in Fig. 2. In
order to facilitate the comparison of the data sets used in study by Papazachos
et al. [1992] and the present study, two sets of plots are shown in Fig. 2. In particular, the former study includes 37 seismic events and 54 accelerograms while the
latter study compromises 93 seismic events and 141 accelerograms. It is obvious
from the first graph of Fig. 2 that there is a lack of strong-motion records from
large earthquakes in the near field. This lack is partly due to the limited number of
strong-motion installations developed throughout Greece, along with the fact that
most large earthquakes occur offshore.
The strong-motion data have been recorded from ITSAK (Institute of Engineering Seismology and Earthquake Engineering) and NOA-GI (National Observatory of Athens-Geodynamic Institute) accelerographic networks (www.itsak.gr,
www.gein.noa.gr). The recording instruments are mainly consisted of SMA-1 type
and few digital installations constituting aftershock arrays. The available records
have been processed and corrected by the standard procedure applied in ITSAK.
Processing included interpolation to 600 samples/s, low pass filter at 2550 Hz depending on instrument type, high-pass filter based on the signal to noise ratio per
each record, and decimation to 200 samples/s.
The available dataset, stems from national network located mainly at basements
of low-rise governmental buildings (up to 3 storey). Numerous studies have shown
[Kavazanjian et al., 1985] that the effect of soil-structure interaction on duration
is negligible for low-rise buildings. This was further supported from the fact that
duration measures of Greek records are rather short, in accordance with the Greek
seismotectonic environment.
Rupture directivity [Boore and Joyner, 1978; Boatwright and Boore, 1982], can
cause significant variations in ground motion amplitudes and duration around the
faults. Deviations from the mean duration from earthquake to earthquake are observed and this is often due to the multiplicity of the earthquake source. Those
deviations become significant at a period of 0.6 s and generally grow in size with
increasing period [Somerville et al., 1997]. Analytical studies taking into account the
wave propagation and directivity effects have been made after strong seismic events
recorded by a significant number of accelerographs adequately covering the epicentral area [Boatwright and Boore, 1982; Somerville et al., 1997]. Unfortunately, a lack
of such strong-motion data exist in Greece and therefore, dependence of duration in
terms of rupture directivity and wave propagation it is not possible to accomplish.
Nevertheless, application of an anisotropic radiation of seismic waves for estimation of the strong-motion attenuation in Greece [Papazachos, 1992] is adopted in
seismic hazard assessment through the input model [Margaris and Papazachos,
1994].
00060
83

"
&
'
&
'
"
&
'
&
'

'
I
I
'
EFG
&

&
B
>
'
'
Fig. 2. (a) Distribution of data in terms of epicentral distance. (b) Distribution of data in terms
of earthquake magnitude. (c) Distribution of epicentral distance vs. earthquake magnitude. The
left column includes data from previous work [Papazachos et al., 1992], in the right column are
the data from the present study.
84
00060
3. Attenuation of the Bracketed Duration of Strong Motion

No universally accepted measure of the duration of strong ground-motion exists.
In a recent publication concerning the study of strong-motion duration [Bommer
and Martnez-Pereira, 1999], approximately 30 different definitions of strong-motion
duration are referred. Thus, the first problem encountered was the selection of an
appropriate definition of strong-motion duration. After some consideration, the definition adopted is the one proposed by Bolt [1973], known as bracketed duration.
For a given strong-motion record, it is defined as the time interval between the first
and last excursion of the peak ground acceleration beyond a certain predefined level
of it, usually 5%g. The rationale behind the choice of bracketed duration, rather
than any other duration definition [Bommer and Martinez-Perreira, 1999], lies on
the fact that bracketed duration correlates well with rupture duration [Margaris
et al., 1990; Papazachos et al., 1992], significant duration (for specific level of bracketed duration) and macroseismic intensity IMM [Koliopoulos et al., 1998]. In fact
Kawashima and Aizawa [1989] also used the same definition of duration and applied
multiple regression analyses using the bulk of the available Japanese strong-motion
data.
The available records were inspected in order to remove any possible secondary
events following the main shock, which may lead to an overestimation of the calculated strong-motion duration. The bracketed durations where calculated for a series
of different acceleration levels equally spanning the interval between 1 and 20%g. It
was decided to keep only the bracketed durations corresponding to levels between
2 and 10%g, specifically 2, 2.5, 3, 3.5, 4, 5, 6.5, 8 and 10%g. Lower acceleration
levels were not considered, to avoid an overestimation of corresponding durations,
in case the signal/noise ratio of the record is low. The reason for the removal of
the measured durations for levels above 10%g, lies in the fact that the resulting
number of data is very small and the calculated durations are extremely short and
exhibit a large statistical scatter. This often led to a behaviour entirely different
than expected (for example, the calculated duration decreases with the increase of
earthquake magnitude). The finally accepted values of strong-motion duration are
given in Table 2, while the distribution of these values for the level of 5%g is shown in
Fig. 3. It can be seen that scatter of the bracketed duration is considerable, however
as a general trend bracketed duration increases as earthquake magnitude increases
and epicentral distance decreases [Kawashima and Aizawa, 1989]. Notably the dependence of duration on the epicentral distance is not very strong. Earlier studies
[Papazachos et al., 1992] revealed that taking into account the epicentral distance
in the same functional form decreases the standard deviation by 40%. Although it
has been reported [Bommer and Martinez-Pereira, 1999] that bracketed duration
can be rather unstable for low threshold of accelerations (0.02, 0.03g) in our case
no significant instabilities have been observed. Furthermore previous studies using
Greek data [Koliopoulos et al., 1998], shows that 0.03g bracketed correlates well
with significant duration, which is generally considered as stable.
00060

Table 2.
85
Calculated values of bracketed duration for the earthquakes of Table 1.
NC
Mw
SC
BD02
BD025
BD03
BD035
BD04
BD05
BD065
BD08
BD10
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
6.4
6.4
5.3
5.3 6
5.6
5.6
5.2
5.2
6.6
6.6
6.9
6.9
6.9
6.9
6.9
6.9
5.1
5.1
5.8
5.8
5.4
5.4
5.4
5.4
5.2
5.2
4.9
4.9
5.3
5.3
6.2
6.2
6.2
6.2
6.2
6.2
5.4
5.4
6.6
6.6
6.6
6.6
5.1
30
30
6
1
15
15
7
7
31
31
128
128
105
105
33
33
13
13
15
15
6
6
56
56
19
19
16
16
12
12
68
68
75
75
13
13
10
10
119
119
85
85
45
0
0
1
4.47
1
1
1
1
0
0
0
0
0
0
1
1
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
2
2
0
0
2
8.82
13.19
4.31
2.93
5.74
5.85
4.08
2.25
28.27
22.41
5.13
4.90
15.90
13.82
17.94
17.22
3.20
1.48
6.04
7.12
3.60
3.43
8.23
8.99
3.47
2.92
4.71
4.86
2.58
2.23
18.99
20.28
3.77
4.61
9.28
12.94
17.61
15.21
1.23
0.43
6.03
3.88
2.30
1.74
8.21
7.70
2.79
1.92
4.70
3.49
2.12
1.73
18.46
17.61
0.01
2.79
9.24
11.30
16.24
13.64
0.28
8.20
7.68
2.77
1.43
4.68
3.48
2.11
1.49
18.43
17.47
8.19
7.46
2.03
1.18
0.82
2.66
2.10
0.46
17.18
14.68
5.94
5.73
2.02
1.05
0.81
1.40
0.32
0.38
17.14
13.14
5.89
5.56
1.18
0.16
1.34
5.05
0.05
0.02
1.06
0.91
0.04
8.43
7.65
5.88
6.18
1.25
9.17
10.33
11.86
11.35
0.22
0.52
6.58
8.11
11.85
8.69
0.21
0.02
4.90
2.66
10.50
8.28
0.08
4.50
3.94
3.29
1.66
3.75
3.11
1.80
1.57
3.63
3.10
1.36
1.51
3.43
2.97
0.94
1.47
3.07
2.52
0.49
0.87
7.29
2.80
0.15
6.65
1.25
0.02
6.63
7.98
2.99
0.02
2.43
5.06
4.01
16.23
14.82
7.03
4.77
0.01
0.01
3.67
1.99
2.17
2.64
2.26
0.50
0.02
0.48
0.34
0.18
1.28
13.08
14.79
5.69
4.37
13.07
10.30
3.59
4.32
10.65
10.29
1.61
3.78
10.65
8.83
1.57
1.88
10.63
7.98
1.50
1.53
6.11
7.77
1.47
1.23
6.09
5.54
1.23
1.21
2.27
5.26
0.55
0.53
3.27
1.97
1.95
1.85
1.45
1.39
0.25
0.15
0.07
0.01
8.47
11.18
0.01
6.38
8.26
0.02
0.51
0.25
0.02
0.01
86
00060
Table 2.
(Continued)
NC
Mw
SC
BD02
BD025
BD03
BD035
BD04
BD05
BD065
BD08
BD10
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
5.1
5.1
5.1
4.9
4.9
5.2
5.2
5.2
5.2
5.2
5.2
5.3
5.3
5.3
4.8
4.8
4.5
4.5
5.5
5.5
5.5
5.5
5.2
5.2
5.2
5.2
5.2
5.2
5.9
5.9
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.0
5.0
4.9
4.9
5.7
45
10
10
25
25
21
21
28
28
18
18
22
11
11
14
14
12
12
22
22
15
15
17
17
49
49
13
13
9
9
1
1
1
1
10
10
20
20
6
6
14
14
22
2
0
0
2
2
1
1
2
2
0
0
2
2
2
2
2
2
2
0
0
1
1
1
1
2
2
1
1
1
1
1
1
1
1
0
0
2
2
2
2
0
0
1
1.21
4.13
5.27
1.13
3.37
3.06
1.05
3.37
3.06
0.62
2.91
2.86
0.61
2.15
2.86
0.01
1.45
2.85
1.26
1.33
0.20
0.77
0.19
0.51
6.44
4.29
4.56
3.19
0.23
2.56
1.51
1.29
4.27
1.28
0.02
8.82
10.45
1.00
2.25
0.17
2.14
7.76
8.32
5.87
3.10
3.50
4.41
6.49
5.43
6.17
6.18
11.84
17.19
6.14
4.78
3.73
4.56
10.19
5.04
2.45
2.82
0.91
0.94
4.37
0.05
2.99
2.77
0.83
2.09
0.67
0.65
0.35
0.20
0.18
8.80
7.68
0.63
2.14
0.10
1.62
5.31
6.20
3.27
3.07
2.98
1.03
5.94
1.96
5.20
5.22
11.51
11.33
4.37
4.37
3.10
3.93
1.33
2.84
2.37
2.80
0.81
0.85
0.58
5.55
7.60
0.52
1.99
0.08
0.86
5.29
6.16
2.91
2.45
2.36
0.93
0.60
0.75
4.30
4.96
6.42
10.73
4.35
1.80
2.80
3.91
0.01
0.68
1.93
2.53
0.72
0.83
0.57
5.19
6.45
0.40
0.28
0.07
0.50
2.66
6.02
2.90
2.44
5.18
4.98
0.34
0.07
4.28
3.94
0.02
0.05
3.22
3.02
2.66
2.01
2.07
1.98
0.33
0.20
0.03
0.74
0.20
0.17
0.04
0.91
0.33
0.88
0.04
0.37
0.01
0.33
0.02
2.98
3.46
6.40
10.69
4.18
1.79
2.78
2.84
2.96
0.86
5.71
8.14
3.42
1.78
2.75
2.28
2.05
0.09
0.04
5.69
8.02
3.19
1.13
1.11
1.45
5.62
3.88
1.12
0.88
1.00
1.44
4.08
3.51
0.99
0.65
0.45
1.15
2.61
2.46
0.98
0.25
0.23
1.01
0.03
1.91
2.52
0.50
0.01
0.54
0.02
1.29
2.51
0.42
0.40
1.52
0.32
0.38
0.10
0.11
0.10
0.04
0.09
0.37
0.29
0.08
2.90
2.21
1.46
1.46
0.02
0.01
0.63
00060

Table 2.
87
(Continued)
NC
Mw
SC
BD02
BD025
BD03
BD035
BD04
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
5.7
4.6
4.6
5.8
5.8
5.8
5.8
4.7
4.7
6.1
6.1
6.1
6.1
5.7
5.7
5.0
5.0
5.4
5.4
5.0
5.0
4.8
4.8
5.3
5.3
5.1
5.1
5.4
5.4
5.2
5.2
4.9
4.9
5.8
5.8
4.8
4.8
5.6
5.6
5.6
5.6
5.5
5.5
22
21
21
16
16
34
34
20
20
31
31
44
44
39
39
65
65
16
16
44
44
11
11
6
6
2
2
12
12
21
21
14
14
34
34
16
16
5
5
5
5
15
15
1
0
0
0
0
0
0
0
0
1
1
1
1
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
3.51
1.46
2.85
18.34
20.00
11.34
13.92
4.03
4.21
11.61
12.04
5.01
3.96
1.80
1.93
0.68
0.87
6.06
6.91
1.29
1.56
1.32
0.62
2.47
5.63
5.65
5.03
7.31
8.10
0.03
0.10
0.03
0.51
0.41
1.30
4.03
5.10
12.30
13.14
12.67
11.92
14.69
16.52
3.50
0.55
2.23
16.84
17.33
9.82
7.69
4.02
4.17
11.27
6.36
4.98
2.62
1.32
1.02
0.67
0.02
5.23
5.28
0.18
1.42
0.29
0.61
2.14
5.62
1.84
4.49
7.23
7.35
1.58
0.54
0.94
16.62
17.02
9.52
7.66
0.19
3.91
7.31
6.34
3.43
0.35
1.16
0.70
0.66
1.57
0.52
0.93
16.37
16.07
9.51
7.63
0.19
3.75
6.69
6.31
1.46
0.34
0.95
0.06
0.64
5.22
4.89
0.12
0.78
0.27
0.59
1.65
5.61
1.81
2.52
4.55
5.12
3.76
4.17
0.11
0.76
0.16
0.49
1.47
3.46
1.24
2.51
4.53
4.93
3.75
3.93
0.11
0.76
0.14
0.47
1.46
1.77
1.23
2.50
4.52
4.91
0.09
0.07
0.05
0.02
0.31
0.38
0.94
2.76
4.24
9.39
7.81
9.85
10.03
10.70
9.10
0.06
0.41
0.91
0.77
9.38
7.81
9.24
10.01
8.63
9.07
0.39
0.90
0.66
9.37
6.35
7.22
6.91
8.61
4.19
0.03
0.21
0.65
6.59
5.10
7.22
6.76
8.58
3.89
BD05
BD08
BD10
6.24
7.14
0.56
1.09
2.63
6.31
0.01
1.07
1.41
6.27
0.06
3.46
0.41
0.04
1.55
0.34
0.01
3.74
3.92
0.09
0.01
3.72
3.58
0.06
3.49
1.06
0.25
0.61
0.41
1.08
1.76
0.64
1.39
4.50
4.88
0.40
1.03
1.21
0.56
0.28
2.48
2.11
0.38
1.01
0.95
0.33
0.19
0.65
1.61
0.34
0.03
6.58
4.61
6.31
6.75
7.56
3.31
6.44
4.23
4.59
5.89
3.59
3.28
2.71
4.22
2.53
5.88
0.60
3.22
2.24
2.45
1.66
1.60
0.55
0.37
1.27
0.48
0.38
0.66
0.01
11.40
6.81
11.28 10.96
7.80
4.11
7.35
1.76
0.18
0.04
0.09
0.07
5.85
5.61
6.06
4.08
0.109
0.01
0.53
0.03
0.04
0.27
BD065
0.02
0.34
0.94
0.02
0.41
1.57
88
00060
Table 2.
(Continued)
NC
Mw
SC
BD02
BD025
BD03
BD035
BD04
BD05
BD065
BD08
BD10
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
5.5
5.5
5.7
5.7
5.7
5.7
5.1
5.1
5.4
5.4
5.4
5.4
5.4
5.4
5.4
5.4
5.4
5.4
5.4
5.4
5.4
5.4
6.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
4.9
4.9
5.2
5.2
5.2
5.2
5.2
5.2
5.2
12
12
65
65
50
50
28
28
27
27
27
27
35
35
59
59
31
31
30
30
32
32
20
20
90
90
138
138
53
53
61
61
72
72
8
8
9
9
27
27
28
28
22
0
0
0
0
0
0
0
0
2
2
0
0
1
1
1
1
1
1
2
2
0
0
2
2
0
0
2
2
2
2
2
2
2
2
1
1
1
1
1
1
2
2
1
8.18
8.16
3.10
3.93
22.22
19.84
6.69
10.04
1.81
2.00
0.70
1.65
4.92
6.26
1.56
2.10
17.49
17.51
5.93
7.52
1.73
1.72
0.02
0.49
4.38
6.24
1.53
1.68
13.05
17.28
5.91
4.67
1.67
1.65
4.37
3.85
1.51
0.36
8.57
12.37
5.78
4.64
1.48
1.63
4.34
3.84
1.49
0.07
2.46
12.33
4.02
1.56
1.47
1.21
1.88
3.81
0.03
0.80
1.39
1.20
0.05
0.94
0.70
0.46
0.03
0.17
0.79
0.08
0.01
0.63
0.43
0.80
1.75
2.20
0.02
14.10
14.22
1.49
6.37
0.28
0.01
0.61
1.67
0.03
0.03
0.48
1.14
0.02
0.50
0.48
8.62
11.62
7.35
11.61
7.04
9.92
5.67
9.91
4.96
6.83
4.45
3.89
4.44
3.41
2.83
2.94
0.23
0.42
3.88
4.39
0.01
0.39
2.24
2.10
0.11
2.23
1.83
0.03
2.04
0.46
1.13
0.08
2.44
0.01
5.90
0.10
2.34
0.06
2.75
2.78
5.49
5.43
1.23
1.48
1.10
1.75
0.95
3.94
4.58
0.01
0.16
1.69
0.94
3.89
4.48
0.58
0.93
3.88
4.40
0.09
0.02
2.91
0.81
0.30
0.03
0.03
00060

Table 2.
89
(Continued)
NC
Mw
SC
BD02
BD025
BD03
BD035
BD04
BD05
BD065
BD08
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
5.2
4.6
4.6
4.7
4.7
5.1
5.1
5.1
5.1
5.2
5.2
5.2
5.2
5.2
5.2
5.2
5.2
4.8
4.8
4.8
4.8
4.5
4.5
5.0
5.0
5.0
5.0
5.0
5.0
5.0
5.0
4.5
4.5
4.5
4.5
4.7
4.7
4.7
4.7
4.7
4.7
4.7
4.7
22
9
9
20
20
12
12
27
27
29
29
30
30
22
22
14
14
15
15
21
21
19
19
33
33
8
8
19
19
19
19
6
6
12
12
22
22
5
5
13
13
12
12
1
1
1
1
1
1
1
1
1
1
1
2
2
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
2
2
1
1
2.14
0.36
0.81
0.19
0.02
0.58
0.35
0.34
0.18
3.16
3.27
2.87
2.23
2.85
2.22
2.55
2.22
1.30
0.92
0.18
0.73
7.21
6.36
0.46
3.18
0.72
4.26
4.57
0.29
4.24
4.55
0.01
11.17
11.18
1.19
0.18
1.15
1.20
3.10
2.54
0.31
10.68
10.02
9.68
9.25
1.13
0.01
1.34
2.52
3.34
4.37
0.01
1.55
2.61
1.90
BD10
2.53
2.20
1.93
1.07
1.58
0.09
0.16
3.19
4.47
2.87
2.16
1.60
2.14
1.49
1.58
1.17
0.92
0.31
0.04
9.66
8.46
6.93
6.67
5.65
5.96
3.61
4.74
2.50
4.37
2.37
4.35
0.69
1.88
0.04
1.87
0.03
0.71
0.01
3.27
4.04
2.37
3.65
2.09
2.56
1.21
1.21
0.88
0.89
0.56
0.01
1.57
1.56
1.47
1.54
1.46
0.68
0.25
0.14
0.12
90
00060
Table 2.
NC
Mw
SC
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
4.7
4.7
5.0
5.0
4.5
4.5
4.7
4.7
4.7
4.8
4.8
4.8
6.3
6.3
6.3
6.3
4.5
4.5
4.5
4.5
5.4
5.4
5.4
5.4
5.4
4.5
4.5
4.5
5.4
5.4
4.7
4.7
4.7
4.7
4.7
5.0
5.0
5.0
5.0
5.0
5.0
5.4
5.4
23
23
14
14
17
17
5
5
5
4
4
5
9
9
9
9
3
3
4
6
8
8
4
4
4
1
1
2
5
5
4
4
3
3
4
10
10
4
4
4
4
11
11
1
1
1
1
1
1
0
0
2
0
0
2
0
0
2
2
0
0
2
2
0
0
0
0
2
0
0
2
0
0
0
0
0
0
2
0
0
0
0
2
2
0
0
(Continued)
BD02
BD025
BD03
BD035
BD04
BD05
BD065
BD08
BD10
1.70
1.02
0.01
1.00
0.38
0.36
0.01
0.46
0.29
0.13
0.22
0.03
0.06
0.02
0.71
0.11
0.01
0.21
5.09
6.26
4.64
4.27
0.05
5.08
6.18
4.48
2.44
5.06
4.37
4.30
2.42
4.92
4.36
4.09
2.29
3.00
4.19
2.42
2.26
2.99
3.72
2.34
1.95
2.89
3.13
2.22
0.86
1.35
2.90
2.11
0.82
1.14
2.74
1.98
0.74
0.04
0.03
0.02
3.68
3.44
3.25
0.01
0.54
3.39
3.02
3.23
3.24
3.00
2.72
2.79
2.60
2.58
2.79
2.59
1.68
2.77
2.20
1.38
2.15
2.13
1.52
1.97
1.51
1.95
0.04
0.03
0.03
0.01
3.48
2.32
0.38
0.90
0.88
1.19
2.70
1.91
2.68
1.90
1.77
1.15
1.22
1.14
1.20
1.13
0.14
0.02
0.13
0.03
0.26
0.53
0.64
0.19
0.35
0.51
0.18
1.49
1.64
0.37
0.54
1.47
1.49
0.23
0.19
0.99
1.44
0.12
0.18
0.80
1.21
0.11
0.01
0.45
0.85
0.11
0.43
0.09
0.03
0.01
0.03
0.41
0.02
00060

Table 2.
91
(Continued)
NC
Mw
SC
BD02
BD025
BD03
BD035
BD04
BD05
BD065
BD08
BD10
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
5.4
5.4
5.4
5.4
6.4
6.4 7
6.4
6.4
6.4
6.4
6.4
6.4
6.4
6.4
6.6
6.6
6.6
6.6
6.6
6.6
6.6
6.6
6.6
6.6
1
1
1
1
78
8
132
132
104
104
86
86
105
105
51
51
116
116
82
82
65
65
117
117
0
0
2
2
2
2
2
2
1
1
2
2
2
2
0
0
2
2
2
2
0
0
1
1
2.66
2.68
2.85
2.95
21.07
26.82
2.04
2.59
2.12
1.81
19.78
19.83
1.64
2.43
1.34
1.48
16.28
13.97
1.28
2.42
1.33
1.47
11.39
10.09
1.12
2.08
1.32
1.38
10.12
8.73
1.11
1.60
0.22
0.92
8.90
5.14
0.58
0.79
0.20
0.03
7.69
3.46
0.48
0.78
0.05
0.46
0.38
0.02
7.62
0.54
6.64
0.36
7.47
6.38
0.01
3.33
1.09
3.30
25.17
30.84
6.49
7.92
14.61
11.92
11.86
9.31
5.90
4.53
20.01
28.50
6.41
7.41
14.58
9.71
7.23
7.65
4.15
0.02
17.17
21.83
5.01
5.99
9.92
9.59
6.44
7.64
2.79
17.14
20.54
4.30
4.34
6.66
5.60
3.23
3.70
16.35
20.51
4.04
3.37
5.25
4.94
1.16
0.72
4.80
0.05
0.02
1.89
2.48
0.60
0.94
0.50
0.89
0.04
0.06
30.45
30.89
9.34
11.79
15.30
12.62
12.78
13.30
10.47
7.01
1.40
The independent variables, which were selected to describe the attenuation relation of the dependent variable of bracketed duration, are the earthquake magnitude,
the epicentral distance, a variable representing the soil conditions at the recording
site and a variable representing the level of ground acceleration for which the bracketed duration was estimated. Typically, studies aiming to develop empirical ground
motion models select a single, preferred mathematical relationship to describe the
attenuation of strong-motion parameters. Models are selected on the basis of fundamental physical arguments that lead to the basic functional dependence of ground
motion on the independent variables, usually magnitude and distance. The dependence of the strong-motion duration, in terms of magnitude distance and soil
conditions, has been examined by many authors [Esteva and Rosenblueth, 1964;
Trifunac and Brady, 1975; McGuire and Barnhard, 1979]. The same mathematical models of the empirical prediction of the duration as a function of magnitude,
distance and soil conditions have been proposed for Greece and other regions [Xie
and Zhang, 1988; Kawashima and Aizawa, 1989; Margaris et al., 1990; Papazachos
92
00060

'
&
"

>
"
&
&
"

"
&
'
Fig. 3. (a) Distribution of calculated values of bracketed duration 5%g, (b) distribution of the
same values versus epicentral distance and (c) distribution against earthquake magnitude.
et al., 1992; Novikova and Trifunac, 1994; Trifunac and Novikova, 1995]. The general
form of the attenuation relation is:
ln D = c1 + c2 M + c3 ln(R + R0 ) + c4 L + c5 S + ln D P
(1)
where D represents the bracketed duration (in seconds), M is the earthquake magnitude, R is the epicentral distance (in kilometers), R0 is a coefficient used to take
00060
93
into account the effect of the saturation of the values of D with distance, L is the
level of acceleration for which the bracketed duration is calculated (in g) and finally
S is the variable used to describe the soil conditions at the recording site. S, takes
three different values: zero, one and two for soft alluvium deposits, intermediate
soil conditions and tertiary or older rock respectively. P is 0 for 50 percentile and
1 for 84-percentile level of non-exceedence, assuming normal distribution of the
residuals and ln D is the residuals root mean square resulting from Eq. (1). The
inclusion of soil conditions as an independent parameter was decided in order to
clarify an apparent contradiction between the findings of early works [e.g. McGuire
and Barnhard, 1979] and more recent works [e.g. Kawashima and Aizawa, 1989;
Papazachos et al., 1992] regarding the importance of site characterization on duration. In particular, the former study showed the significance of incorporating
soil conditions on duration empirical predictions while the later works have found
that duration is practically independent of soil characterization. Recording sites
were classified into three categories soft alluvium, intermediate and rock soil
conditions. The grouping was carried out based on the stiffness of the material
at the accelerographic site, geological description and the available geotechnical
data (Vs30 , NSPT etc.). This soil classification followed a methodology proposed
by Trifunac and Brady [1976] and applied by Theodulidis and Papazachos [1992]
for Greek strong-motion data. In addition, recent geotechnical information for a
few accelerographic sites was taken into account in this classification [Klimis et al.,
1999].
The values ci represent the coefficients of the attenuation relation under determination and were calculated with the application of multiple regression analysis.
The method used to calculate the coefficients ci , is the multi-linear regression analysis [Papoulis, 1990; Everitt, 1991; Bevington and Robinson 1992; Press et al., 1992].
This method solves the problem of defining a function of the form:
y = c1 w1 + c2 w2 + + cm wm
(2)
which fits the data:

w1i , . . . , wmi ,
yi i = 1, . . . , n
where, n is the number of available strong-motion records. Each record corresponds

to a single set of values for the variables D, M , R, S and L.
The solution of the above problem, is the function which minimizes the sum:
X
Q=
vi2
(3)
where vi = yi (c1 w1i + c2 w2i + + cm wmi ).
Following this mathematical procedure the problem of determining coefficients,
ci , eventually reduces to the solution of a system of algebraic equations. The inclusion of the coefficient R0 results in a system of non-linear equations, which is
difficult if not impossible to solve. The procedure followed in this study is to assume
distinct values for R0 that lie inside a range of expected values. In this way, the
94
00060
Table 3.
Coefficients of the proposed attenuation relation [Eq. (1)].

C1
C2
C3
C4
C5
C6
ln Y
Including the soil

conditions variable, S
1.88
2.05
2.05
30
0.05
27.74
1.50
Without the inclusion of

variable S
1.88
2.05
2.05
30
27.75
1.49
system of algebraic equations becomes linear and is easily solved. The coefficient R0
has been defined by investigating the standard deviation values of Eq. (1) through
successive regression analyses with R0 = 0, 5, 10, . . . , 40. The value, which is finally chosen to represent R0 , is the one that minimises the standard deviation of
the equation. By following this procedure, the value of R0 was found equal to 30.
This value R0 = 30 is in a good agreement with values which have been derived
from similar analyses regarding effective acceleration [Mortgat, 1979; R0 = 25] or
duration of strong-motion [Kawashima and Aizawa, 1989; Yayong and Minxian,
1990; R0 = 30]. Table 3 shows the values of the coefficients of the attenuation
relation. As it is clearly seen, the value of the coefficient c5 , for the variable S
is very low. This means that bracketed duration is practically independent of the
soil conditions at the recording site, provided that a simple soil description index
like the one adopted herein is used. For further confirmation, a binary variable S
(zero for alluvium and one for rock site conditions) was also used. The resulting
coefficient c5 was found equal to 0.09 while all remaining coefficients were almost
identical to the values given in Table 3. The coefficients of the empirical relation
were recalculated without taking into account the soil conditions parameter and
they were found identical to the previously calculated values, as shown in Table 3.
For this reason, the final relation does not include the soil condition parameter at
the recording site.
In order to evaluate the reliability of the empirical relation, a residual analysis
was performed. The residuals were plotted against the independent variables of
the attenuation relation shown in Fig. 4. These plots are used in order to identify
any trends that would indicate an inadequacy of the selected model leading to a
modification of the functional form. Although a mild negative bias can be observed
regarding low threshold values, this trend becomes negligible for higher threshold
values, which are of practical interest. Consequently, the relation was considered
acceptable. It is worth noting here, that this form of the empirical prediction of
duration has been successfully used by McGuire and Barnhard [1979], Kawashima
and Aizawa [1989], Margaris et al. [1990] and Papazachos et al. [1992].
Finally, the derived relation was compared with similar relations reported in
two of the above-mentioned works. Specifically, a comparison was made with the
relation proposed by Papazachos et al. [1992] in Greece and the one proposed by
Kawashima and Aizawa [1989] in Japan. The latter work in fact, proposed two
00060
95

>

"
'
Fig. 4. Distribution of residuals of the calculated attenuation relation vs. earthquake magnitude
(a), epicentral distance (b) and bracketed duration level (c).
different relations: one for the level of 50 cm/s2 and one for the level of 100 cm/s2
(roughly 5 and 10%g). Papazachos et al. [1992] proposed a relation that contains
a parameter concerning the level of ground acceleration for which the bracketed
duration is being calculated. It should be noted that all definitions of magnitude
used (Mw , Ms and MJMA ) in the above works are comparable for magnitude range
96
00060
5 to 7 and shallow earthquakes, covering the bulk of the available data [Katsumata,
1996; Papazachos et al., 1997]. In Figs. 5 and 6, the values of bracketed duration
predicted by these three relations, are plotted against the epicentral distance for
a constant magnitude of 6.5 and for two different levels of acceleration (5 and
10%g). From these two figures it can be seen that the relation proposed in the
present study, is different from the one proposed by Papazachos et al. [1992]. In
particular, the proposed relations predict higher durations for epicentral distances
up to 30 km and lower durations for greater distances. These deviations are probably
due to the substantially richer data set available herein compared to Papazachos
et al. data set. The aforementioned conclusion is further supported from the fact
that similar deviations have been reported in a comparison between Greek and
Japanese empirical predictions of strong-motion duration [Papazachos et al., 1992;
Fig. 5). It is noteworthy that the Japanese empirical predictions [Kawashima and
Aizawa, 1989] were based on a very rich data set comprising 197 accelerograms.
(5 < M < 8 and 5 km < R < 500 km). A part of these data were also used by
Theodulidis and Papazachos [1992] to enrich the Greek dataset in order to derive
empirical predictions for pga, pgv, pgd and 5%-psv spectra. This fact necessitates
the comparison between Greek and Japanese empirical predictions.
Despite certain quantitative differences, the general trend of the relation proposed in the present study, show a satisfactory qualitative agreement with the trend
of the relation proposed for Japan. In contrast, the differences between the Papazachos et al. empirical prediction and the Japanese predictions are both quantitative
and qualitative.
CD
>
<=
"
'
"
Fig. 5. Comparison between the attenuation relation of the present study and two existing relations for a constant magnitude of 6.5 and for a level of bracketed duration of 5%g.
00060
97
EF
>?
<
'
"
"
'
"
'
Fig. 6. Comparison between the attenuation relation of the present study and two existing relations for a constant magnitude of 6.5 and for a level of bracketed duration of 10%g.
<
9:
23
"
"
"
'
>
"
"
Fig. 7. Observed bracketed duration values for a level of 3%g from the 7/9/99 earthquake in
Athens (solid circles). The solid line corresponds to the attenuation relation proposed in the
present study. The dashed lines define the standard deviation interval.
Considering that, (in addition to the rich data set), the relation proposed by
Kawashima and Aizawa [1989] was calculated using both shallow and intermediate
depth earthquakes covering a more extended range of magnitudes and epicentral
distances, one can conclude that the relation of the present study clearly improves
the existing one for Greece.
98
00060
<
:;
34
'
>
Fig. 8. Observed bracketed duration values for a level of 5%g from the 7/9/99 earthquake in
Athens (solid circles). The solid line corresponds to the attenuation relation proposed in the
present study. The dashed lines define the standard deviation interval.
The relation has been further tested by comparing its predictions with the corresponding results from the strong-motion records of the Athens earthquake of
7-9-1999 [Koliopoulos and Margaris, 2001]. These records were not included in the
database used for the evaluation of the empirical relation and thus the comparison of the computed values of strong-motion duration of these recordings with the
theoretical values predicted by the attenuation relation serves as a good test of
the accuracy of the proposed relation. Figures 7 and 8 show the measured values
of strong-motion duration for two different levels 3 and 5% of g along with the
empirical theoretical attenuation curves one standard deviation. It is obvious
that, given the statistical uncertainty, the proposed attenuation relation predicts
the calculated values with acceptable accuracy.
4. Seismic Hazard Assessment for the Area of Greece

Using the new attenuation relation for strong-motion duration, a probabilistic
seismic hazard analysis was performed in Greece [McGuire, 1976; Margaris and
Papazachos, 1994]. The most recent model regarding geographical distribution of
seismotectonic zones in Greece was used, as has been proposed by Papaioannou
and Papazachos [2000]. In particular, 67 shallow and 7 intermediate seismic sources
are used, which include seismicity information, maximum observed magnitudes and
radiation coefficients. Regarding the distinction between shallow and intermediate
depth seismic sources, the model implements the seismic hazard analysis adopting
two characteristic depths namely 10 km and 80 km for shallow and intermediate
00060
99
seismic sources respectively. Moreover the above mentioned intermediate depth seismic sources are characterised by high seismic activity with magnitude between 6.2
and 7.8 [Papaioannou and Papazachos, 2000].
At first, calculations were made for the 136 different sites throughout Greece,
which are included in the Greek Seismic Code, in order to examine the potential use
of strong-motion duration in such an analysis. According to the Greek Seismic Code,
the territory is divided in four different categories of seismic hazard (Fig. 1). The
maximum expected values of bracketed duration were calculated for two different
levels of acceleration, L = 5%g and L = 10%g and for eight different values of
return period, 10, 25, 50, 80, 100, 200, 476 and 952 years. The results obtained
for the aforementioned 136 sites were subsequently categorized according to the
seismic hazard category in which they belong, according to the Seismic Code of
Greece. Finally, the average values of the maximum expected bracketed durations
were estimated for each category. These average values are plotted in Fig. 9 against
the return periods. It can be seen that the four different curves corresponding to
the four different categories of equal seismic hazard, are parallel to each other and
follow the same shape starting from the first category, exhibiting the lowest seismic
hazard category, up to the fourth category with the highest seismic hazard. The
division into four different categories of seismic hazard of Greece as proposed by the
Greek Seismic Code, has been made using the parameters of macroseismic intensity
and peak ground acceleration. From the previous observation stems the conclusion

"
'
Fig. 9. Maximum expected values of bracketed duration for levels of 5%g (continuous curves)
and 10%g (dashed curves), regarding the four categories of equal seismic hazard of the Greek
Seismic Code.
100
00060
that the same division also applies for the case of bracketed duration, which exhibits
similar geographic distribution.
The next step was to estimate the seismic hazard of strong ground motion duration for a grid of points covering the entire Greek region. In this way, a new attempt
to classify Greek territory into categories of equal seismic hazard was made, using
the bracketed duration as the strong-motion parameter. The grid spacing was 0.2
degrees while the bracketed duration for the level of 10%g was used. The maximum
expected values of bracketed duration were estimated for the eight values of return
period, mentioned above. The level of 10% of g has been chosen because it was
considered more appropriate for seismic design, since only strong earthquakes that
exhibit high damage potential will contribute non-zero values for this particular
level of measurement. In order to be able to check the results of the seismic hazard analysis of strong-motion duration, a similar analysis was performed for the
peak ground acceleration using the attenuation relations proposed by Thedoulidis
and Papazachos [1992; Eq. (8)], one for shallow and one for intermediate depth
earthquakes. Both relations include a parameter describing the soil conditions at
the recording site, unlike the attenuation relation of bracketed duration. For this
reason, the estimation of seismic hazard of peak ground acceleration was performed
assuming intermediate soil conditions at the recording sites. The results of these two
analyses were used to divide Greece into four different categories of equal seismic
hazard. The methodology used was the one proposed by Papazachos et al. [1985].
Specifically, a certain value of the strong-motion parameter is given as a criterion
and the total number of maximum expected values of the strong-motion parameter
for each point of the grid and for all values of return period is used in order to divide the area under consideration into categories of equal seismic hazard. The four
categories derived in each case are shown in Figs. 10 and 11. A sample of seismic
hazard results for different mean return periods including the most significant cities
in Greece is presented in Fig. 12.
The conclusion from the inspection of the two maps in Figs. 10 and 11 is that
the geographical distributions of the seismic hazard based on bracketed duration
and peak ground acceleration follows roughly the same pattern. The main difference
between the two appears at the area extending along the southern Aegean (Hellenic
trench) where, in the case of bracketed duration, the second category covers a much
larger area compared to the case of peak ground acceleration. The reason for this is
that this area is governed by intermediate depth earthquakes that show an increased
duration compared to the shallow ones. It should be noted that, although the proposed empirical predictions [Eq.(1)] are based (with only one exception) on shallow
seismic events, nevertheless the focal depth is taken into account as an input to
seismic hazard assessment [McGuire, 1976]. Another difference appears in the area
of the northern Aegean trench (which consists the westward continuation of the
northern Anatolian strike-slip fault). Again in the case of bracketed duration, the
third category extends at a slightly larger area compared to the case of peak ground
acceleration. This area is dominated by strong earthquakes (M 7.0), and as a
00060
101
Fig. 10. The four categories of equal seismic hazard for Greece based on bracketed duration for
a level of 10%g.
result, the expected values of strong-motion duration are significantly larger. These
differences reveal that strong earthquakes contribute more to a duration-based seismic hazard assessment, compared to an acceleration-based hazard assessment. For
example, an increase of magnitude from 5.5 to 6.5 for a design distance R = 20 km,
causes 200% increase in terms of PGA (from 65 cm/s2 to 200 cm/s2 ), while in
terms of 5%g Bracketed Duration causes 670% (from 1 s to 7.7 s). It should be
mentioned that a similar conclusion is derived from a comparison between elastic
input energy-based and elastic response spectra-based seismic hazard assessments,
[Chapman, 1999].
In the next step, an attempt was made to combine the two previous maps into a
single one. For this reason, an empirical relation was derived to associate the bracketed duration for the level of 10%g with the peak ground acceleration. Implementing this relation, the calculated values of bracketed duration can be transformed
102
00060
Fig. 11.
The four categories of equal seismic hazard for Greece based on peak ground acceleration.
to values of peak ground acceleration. The combined map will then result from the
average values of peak ground acceleration. The resulting relation reads:
ln D10 = 7.45 + 1.41 ln am
(4)
where D10 , stands for the bracketed duration for the level of 10%g and am is
the peak ground acceleration. The corresponding standard deviation is equal to
= 1.35. Because the total number of observed values of bracketed duration for
the level of 10%g was not sufficiently large, values of bracketed duration for lower
levels of acceleration, scaled to the level of 10% of g, were also used. The scaling
of bracketed duration of lower threshold levels to the duration of level of 10% of
g was done by using the theoretical attenuation relation. The values of bracketed
duration, D1 and D2 , for two different levels of bracketed duration, L1 and L2 ,
for the same record (i.e. keeping the same values of magnitude, M and epicentral
00060
103
"

<
>
>
>
<
&
>
>
>
&
<
<
"
\_`

YZ
>
>
>
>
&
<
>
>
>
>
>
&
Fig. 12. Seismic hazard results for different mean return periods including the most significant
cities in Greece.
distance, R) according to the attenuation relations are:

ln D1 = 1.88 + 2.05M 2.05 ln(R + 30) 27.75L1
ln D2 = 1.88 + 2.05M 2.05 ln(R + 30) 27.75L2 .
Performing the necessary algebra, the following relation is derived:
D2 = D1 e27.75(L1 L2 ) .
(5)
Using the above equation for the records that exhibited zero values of bracketed
duration for the level of 10% of g, all non-zero values of duration corresponding
to lower threshold levels were scaled to the level of 10% of g and the average of
these scaled values was used for the calculation of the Eq. (4). The aforementioned
104
00060
Fig. 13.
Relation between bracketed duration for a level of 10%g and peak ground acceleration.
Fig. 14. Composite map of the categories of equal seismic hazard for Greece based on peak
ground acceleration and bracketed duration for a level of 10%g.
00060
105
relation is plotted with the observed values in Fig. 13. The data show a large scatter,
reflected in the computed high value of standard deviation in Eq. (4).
By implementation of the above relation, the calculated average maximum expected values of strong-motion duration were transformed to peak ground acceleration values. The average values between the actual and the scaled were finally used
in order to divide Greece into categories of equal seismic hazard using the same
methodology mentioned at the previous stage. The results shown in the map of
Fig. 14 consist the final proposal for classifying Greece into areas of equal seismic
hazard. This proposal may prove useful in future attempts of improving the current
Greek Seismic Code. In Figs. 15 and 16, the mean maximum expected values (
standard deviation) of bracketed duration and peak ground acceleration are plotted
versus the respective return periods. The mean values were computed from 2052
examined sites in Greece after their classification into the four seismic categories.
Comparing the map of Fig. 14 with the map that constitutes part of the current
Greek Seismic Code (Fig. 1), it can be seen that the distribution of the two lowest
seismic hazard categories is roughly the same. The major differences appear in the
geographical distribution of the two higher seismic hazard categories. The fourth
category, which exhibits the highest seismic hazard, in Fig. 14 covers a more extended area compared to the map of the current Seismic Code and also appears in
two other parts of Greece (southeastern Crete and the area near Patras). Also the
third category covers a larger area in Fig. 14 since it covers the largest part of the
Hellenic arc (it appears in north Greece at the area of the North Aegean trough,
/
+

&
Fig. 15. Maximum expected values of bracketed duration for a level of 10%g, vs. return period,
for the categories of Fig. 14.
106
00060
0
*
1000
0
100

10
100

"
1000
#
&
Fig. 16. Maximum expected values of peak ground acceleration vs. return period, for the categories of Fig. 14.
in Southern Aegean at the area of intermediate depth earthquakes and in most of

central Greece).
5. Conclusions
In the preceding sections the duration of strong-motion in Greece was studied along
with its potential use as a parameter for seismic hazard analysis in Greece. From
the results of these analyses, the following conclusions can be deduced:
The correlation of the measured duration values with both magnitude and epicentral distance is not very strong. In general, duration shows a significant scatter,
a fact that was also mentioned in previous studies [McGuire and Barnhard, 1979;
Kawashima and Aizawa, 1989] and seems to be a characteristic of the selected measure of duration. However, as it can be seen from Fig. 3, bracketed duration in
general increases with the increase of earthquake magnitude and the decrease of
epicentral distance. Based on available strong-motion data in Greece, an empirical prediction for bracketed duration is proposed with a validity range in terms of
magnitude between 4.5 and 6.9 and in terms of distances between 1 and 128 km.
No significant relation between bracketed duration and soil conditions was found
and thus no parameter describing soil conditions was included in the proposed
attenuation relation. This does not necessarily mean that the measured values of
bracketed duration are independent of soil conditions, but that a simple definition
like the one used herein may not adequately describe them. A more comprehensive
definition of soil conditions should be adopted in future works, which will take
00060
107
into account the depth of the overlying layer of sediments and/or the shear wave
velocity.
The results from seismic hazard assessment based on strong-motion duration,
are consistent with the results derived from similar analysis based on peak ground
acceleration. However, the seismic hazard maps that resulted in both cases show
small, yet significant, differences from the corresponding that constitutes part of the
Greek Seismic Code. Notably, these differences concern the areas in which the higher
seismic hazard categories extend. These categories appear now to cover larger areas
than assumed before and these differences are a consequence of two factors. The
differences between the ground acceleration hazard maps are mainly due to the new
improved seismotectonic model adopted here. Yet, the differences between durationbased seismic hazard and acceleration-based hazard assessment, both using the
same seismotectonic model, are due to the fact that strong earthquakes contribute
more to a duration-based hazard assessment.
It is one of the aims of an ongoing work between Greece and Italy to confirm
our findings regarding the role of a duration-based hazard assessment. It is hoped
that the implementation of the proposed methodology will lead to the improvement
of current Seismic Codes.
Acknowledgements
The authors would like to express their gratitude to the Technological Educational
Institute of Serres for the partial funding of this work. They would also like to thank
the Institute of Engineering Seismology and Earthquake Engineering (ITSAK),
for providing the strong-motion data, used in this work. This work was partially
supported by the General Secretariat of Research and Technology of Greece (project
2965/1999) and is a contribution of the Geophysical Laboratory of the Aristotle
University of Thessaloniki (no. 586/2001). We finally, express our gratitude to the
two anonymous reviewers and Pr. E. Faccioli for the constructive comments, which
improved the final version of this work.
References
Algermissen, S. T., Perkins, D. M., Isherwood, W., Gordon, D., Reagor, G. and Howard
C. [1976] Seismic risk evaluation of the Balkan region, Proc. Sem. Seismic Zoning
Maps, Vol. 2, UNESCO, Skopje, pp. 172240.
Bevington, P. R. and Robinson, D. K. [1992] Data Reduction and Error Analysis for the
Physical Sciences (McGraw-Hill), 328 pp.
Boatwright, J. and Boore, D. M. [1982] Analysis of the ground accelerations radiated by the 1980 Livermore Valley earthquake for directivity and dynamic source
characteristics, Bull. Seism. Soc. Am. 72, 18431865.
Bolt, B. [1973] Duration of strong ground motion, Proc. 5th WCEE, Rome, Italy,
pp. 13041313.
Bommer, J. J. and Martnez-Pereira, A. [1999] The effective duration of earthquake
strong motion, J. Earthq. Engrg. 3(2), 127172.
108
00060
Boore, D. M. and Joyner, W. B. [1978] The influence of rupture incoherence on seismic

directivity, Bull. Seism. Soc. Am. 68, 283300.
Chapman, M. C. [1999] On the use of elastic input energy for seismic hazard analysis,
Earthq. Spectra 15, 607635.
Drakopoulos, J. and Makropoulos, K. [1983] Seismicity and seismic hazard analysis studies in the area of Greece, (Publ. Univ. of Athens, Seism. Lab.).
Esteva, L. and Rosenblueth, E. [1964] Espectros de templores a distancias moderadas y
grandes, Bol. Soc. Mex. Ing. Sism. 2(1), 118.
Everitt, B. S. [1991] Applied Multivariate Data Analysis (Halsted Press, UK), 306 pp.
Katsumata, A. [1996] Comparison of magnitudes estimated by the Japan Meteorological Agency with moment magnitudes for intermediate and deep earthquakes, Bull.
Seism. Soc. Am. 86(3), 832842.
Kavazanjian, E., Echezuria, H. and Cann, M. W. [1985] RMS acceleration hazard for San
Francisco, J. Soil Dyn. Earthq. Engrg. 4, 106123.
Kawashima, K. and Aizawa, K. [1989] Bracketed and normalized durations of earthquake
ground acceleration, Earthq. Engrg. Struct. Dyn. 18, 10411051.
Klimis, N. S., Margaris, B. N. and Koliopoulos, P. K. [1999] Site-dependent amplification
functions and response spectra in Greece, J. Earthq. Engrg. 2, 237270.
Koliopoulos, P. K. and Margaris, V. N. [2001] The 1999 Athens (Greece) earthquake:
Energy and duration-related response spectral characteristics of different site conditions, Proc. 4th Intl. Conf. on Earthq. Engrg. Soil Dyn., 2431 March 2001, San
Diego, California.
Koliopoulos, P. K., Margaris, B. N. and Klimis, N. S. [1998] Duration and energy characteristics of Greek strong-motionrecords, J. Earthq. Engrg. 2(3), 391417.
Koutrakis, S. I. [2000] A study of the duration of strong-motionin Greece, M.Sc. thesis,
University of Thessaloniki, Greece.
Koutrakis, S. I., Margaris, V. N., Koliopoulos, P. K. and Karakaisis, G. F. [1999] New
trends in seismic hazard evaluation in Greece, Abstracts book, IUGG XXII General
Assembly, Vol. B, Birmingham, pp. 181.
Margaris, V. N. and Papazachos, B. C. [1994] Implication of the azimuthal variation of
seismic intensities in seismic hazard assessment, Proc. XXIV Gen. Ass. Eur. Seism.
Com., Athens, 3, pp. 17181727.
Margaris, B. N. and Papazachos, C. B. [1999] Moment-magnitude relations based on
strong-motion records in Greece, Bull. Seism. Soc. Am. 29, 442455.
Margaris, V. N., Theodulidis, N. P., Papaioannou, Ch. A. and Papazachos, B. C. [1990]
Strong-motion duration of earthquakes in Greece, Proc. XXII Gen. Ass. E.S.C.,
Vol. 2, Barcelona, Spain, pp. 865870.
McGuire, R. K. [1976] FORTRAN computer program for seismic risk analysis, Open-File
Rept 76-67, pp. 190.
McGuire, R. K. and Barnhard, T. P. [1979] The usefulness of ground motion duration
predicting the severity of seismic shaking, Proc. of 2nd U.S. National Conf. on Earthquake Engrg. August 1979, Stanford, 713722.
Mortgat, C. P. [1979] A probabilistic definition of effective acceleration, Proc. of 2nd
U.S. National Conf. on Earthquake Engrg., August 1979, Stanford, pp. 743752.
Novikova, E. I. and Trifunac, M. D. [1994] Durations of strong ground motion in terms
of earthquake magnitude, epicentral distance, site conditions and site geometry, J.
Earthq. Engrg. Struct. Dyn. 23, 10231043.
Papaioannou, Ch. A. [1986] Seismic hazard assessment and long term earthquake prediction in Southern Balkan region, Proc. 2nd Int. Sem. Earthq. Prognostics, A. Vogel
and K. Brandes (Editors), Berlin, pp. 223241.
00060
109
Papaioannou, Ch. A. and Papazachos, B. C. [2000] Time-independent and timedependent seismic hazard in Greece based on seismogenic sources, Bull. Seism. Soc.
Am. 90, 2233.
Papazachos, B. C., Kiratzi, A. A. and Karakostas, B. G. [1997] Toward a homogeneous
moment-magnitude determination for earthquakes in Greece and the surrounding
area, Bull. Seis. Soc. Am. 87, 474483.
Papazachos, B. C., Kiratzi, A. A., Hatzidimitriou, P. M., Papaioannou, Ch. A. and
Theodoulidis, N. P. [1985] Regionalization of seismic hazard in Greece, Proc. of
the 12th Reg. Sem. on Earthq. Engrg. EAEE-EPPO, Halkidiki-Greece, 12 pp.
Papazachos, B. C., Papaioannou, Ch. A., Papastamatiou, D. J., Margaris, V. N. and
Theodulidis, N. P. [1990] On the reliability of different methods of the seismic hazard
assessment in Greece, Natural Hazards 3, 141151.
Papazachos, B. C., Margaris, V. N., Theodulidis, N. P. and Papaioannou, Ch. A. [1992]
Seismic hazard assessment in Greece based on strong-motion duration, Proc. 10th
W.C.E.E. 1, pp. 425430.
Papazachos, B. C., Papaioannou, Ch., Margaris, V. and Theodulidis, N. [1993] Regionalization of seismic hazard in Greece based on seismic sources, Natural Hazards 8,
118.
Papazachos, C. B. [1992] Anisotropic radiation modeling of macroseismic intensities for
estimation of attenuation structure of the upper crust in Greece, Pure Appl. Geophys.
138, 445469.
Papoulia, J. E. and Slejko, D. [1997] Seismic hazard assessment in the Ionian Islands
based on observed macroseismic intensities, Natural Hazards 3, 4958.
Papoulis, A. [1990] Probability and Statistics (Prentice-Hall Inc.), 454 pp.
Press, W. H., Teukolsky, S. A., Vetterling, W. T. and Flannery, B. P. [1992] Numerical Recipes in FORTRAN, The Art of Scientific Computing 2nd Ed. (Cambridge
University Press), 968 pp.
Safak, E. [1998] 3D response spectra: A method to include duration in response spectra,
Proc. of 11th E.C.E.E., Paris, September 611, 1998, pp. 19.
Somerville, P. G., Smith, N. F., Graves, R. W. and Abrahamson, N. A. [1997] Modification
of empirical strong ground motion attenuation relations to include the amplitude and
duration effects of rupture directivity, Seism. Res. Lett. 68, 199222.
Theodulidis, N. P. and Papazachos, B. C. [1992] Dependence of strong ground motion on
magnitude-distance, site geology and macroseismic intensity for shallow earthquakes
in Greece: I, Peak horizontal acceleration, velocity and displacement, J. Soil Dyn.
Earthq. Engrg. 11, 387402.
Theofanopoulos, N. A. and Drakopoulos, J. [1986] A study of strong ground motion
duration of earthquakes in Greece, Proc. 8th E.C.E.E., Vol. 1, Lisbon, Portugal,
pp. 3.2/653.2/67.
Trifunac, M. D. and Brady, A. G. [1975] A study on the duration of strong earthquake
ground motion, Bull. Seism. Soc. Am. 65, 581626.
Trifunac, M. D. and Brady, A. G. [1976] Correlations of peak acceleration, velocity and
displacement with earthquake magnitude, distance and site conditions, J. Earthq.
Engrg. Struct. Dyn. 4, 455471.
Trifunac, M. D. and Novikova, E. I. [1995] Duration of earthquake fault motion in
California, J. Earthq. Engrg. Struct. Dyn. 24, 781799.
Xie, L.-L. and Zhang, X. [1988] Engineering duration of strong-motionand its effects on
seismic damage, Proc. of 9th W.C.E.E., August 29, 1988, Kyoto, Japan, II.
Yayong, W. and Minxian, C. [1990] Dependence of structural damage on the parameters
of earthquake strong motion, J. Eur. Earthq. Engrg. 1, 1323.
00059


MOMENT RESISTING WELDED CONNECTIONS: AN

EXTENSIVE REVIEW OF DESIGN PRACTICE AND
EXPERIMENTAL RESEARCH IN USA, JAPAN AND EUROPE
ELENA MELE
Dip. Analisi e Progettazione Strutturale,
Universita di Napoli Federico II P. le Tecchio 80-80125 Napoli, Italy
elenmele@unina.it
Received 21 November 2000
Revised 5 June 2001
Accepted 10 June 2001
Welded connections, widely used in seismic moment resisting frames (MRFs) in USA,
Japan and Europe, have been extensively investigated during the 1990s in order to
improve their performance under severe earthquakes. In order to correctly evaluate the
results of experimental research on welded connections, the differences among the American, Japanese and European current practices in designing the frame structural systems
and in detailing the beam-to-column connections should be firstly appraised. In this paper, the major aspects characterising the USA, Japan and Europe design practice in
moment resisting welded connections are reported, the differences are underlined and
the main issues recently addressed in experimental research are reviewed. Among the
several parameters which affect the connection performance, attention is focused on the
effect of the beam cross-section size. Thus conclusions are drawn in terms of plastic
rotation capacity as dependant on the beam size.
Keywords: Welded connections; cyclic behaviour; plastic rotation capacity; failure mode;
repaired connections; retrofitted connections.
1. Introduction
The earthquakes of Northridge (1994) and Hyogoken-nanbu (1995) can be considered as milestones in the study of the seismic behaviour of steel structures: starting
from the observations of unexpected damage and brittle failure modes, occurred
during these earthquakes at moment resisting connections of framed structures, the
need for further investigations has been established. Thus research on steel structures in the second half of the 1990s has been strongly characterised by significant
efforts toward understanding the causes of the observed damages, assessing the
major parameters which affect the cyclic behaviour of welded beam-to-column connections and suggesting improvements to the connection configurations. With this
aim, valuable experimental programs on beam-to-column connections have been
developed in the United States [Mahin et al., 1996; Engelhardt and Sabol, 1997;
Malley, 1998] and in Japan [JCSS, 1997; Tanaka et al., 1997; Nakashima et al.,
111
112
00059
E. Mele
1998]. The renewed interest in the seismic behaviour of connections has also affected the recent research trends in Europe, where several experimental projects
have been undertaken [Mele et al., 1997; Plumier et al., 1998; Taucer et al., 1998;
Calado et al., 1999; Calado and Mele 2000a, 2000b; Castiglioni et al., 2000].
It has to be underlined that significant differences arise among the layouts of
welded connections utilised in USA, Japan and Europe, due to both different design
choices in the structural system and different constructional arrangements and detailing of the connection. Correspondingly, also the experimental researches carried
out in USA, Japan and Europe deal with different specimens and investigate on different aspects. However, as a consequence of these research efforts, the experimental
database on welded connection has been remarkably enriched in the last years and
a deeper understanding of the connection cyclic behaviour has been gathered.
This paper provides an overview on the most recent experimental research undertaken in USA, Japan and Europe, and is aimed to unitarily assess the test
outcomes for deriving design implications. In particular, firstly the aspects characterising the USA, Japan and Europe design practice in frame structures and
beam-to-column connections are reported and the main differences are underlined.
Secondly, the experimental research programs recently carried out are reviewed and
the major results are discussed. Finally, conclusions are derived in terms of plastic
rotation capacity as dependant on the beam size, which is directly related to the
redundancy degree of the structural system. Since redundancy in MRF systems is
a major difference among the design practices in USA, Japan and Europe, different
performances are expected in terms of ductility of the connections.
2. Design Practice
2.1. USA
Steel MRF is the most popular structural system among the ones adopted in the
USA for low to medium rise buildings in seismic zones. In this system, which could
be appointed as full MRF, all beam-to-column connections are designed for transferring bending moment. During the last decades in the USA economic considerations have led to establish a design trend of reducing the number of moment connections, thus obtaining the so-called partial MRF. As reported by [Bruneau et al.,
1998] in the extreme, prior the Northridge earthquake, some engineers routinely
designed buildings having only four single-bay ductile moment frames (two in each
principal direction, with each in a different plan to provide torsional resistance).
The adoption of such partial perimeter MRFs lead to a significant increase in the
depth of the beams as compared to more redundant structural systems consisting of
full MRFs. Typical values of beam depth range between 600 and 1000 mm, depending on the bay span and the number of stories; even low-rise buildings commonly
have beam depth of 1000 mm [Nakashima et al., 2000]. Column cross sections are
instead very limited in height, due to the need of saving space, but have very thick
00059
Moment Resisting Welded Connections
Fig. 1.
113
USA (pre-Northridge) welded connection.
flanges for providing high moment capacity and satisfying the strong-column-weakbeam (SCWB) requirement [UBC, 1991; 1994].
2.1.1. Detailing of the connection
The typical welded flanges and bolted web (WFBW) moment connection currently
adopted before the Northridge earthquake (pre-Northridge connection) is depicted
in Fig. 1.
In this connection type, the beam flanges are field welded to the column, while
the beam web is field bolted to a single plate shear tab, which is welded to the column in the shop. The beam flange welds are single bevel complete joint penetration
(CJP) groove welds with backup bars. Continuity plates are applied to the column
flanges and web, according to the seismic code provisions [UBC, 1991; 1994], by
means of either fillet or groove welds. Supplemental welding between the beam web
and the shear tab is required when the flexural strength of the beam flanges is less
than 70% of the flexural strength of the entire section. Doubler plates are required
when the shear strength of the panel zone is less than 80% of the shear strength
corresponding to the beam plastic moment. Yielding of the panel zone is accepted
as a means for energy dissipation.
114
00059
E. Mele
Self-shielded flux-cored metal arc welding (FCAW) process is used for the welds
made in field. The electrodes most frequently used were those designated as E70T-4
and E70T-7 by the AWS. The 3.05 mm diameter electrode was the most popular
one since it provided very high metal deposition rates (of the order of 18 kg/hour)
[Kurobane et al., 1997; Nakashima et al., 2000]. The steel backup and weld tabs
were normally left in place after completion of the weld.
After Northridge significant efforts toward understanding the reasons of the poor
performance of this pre-Northridge welded connection have been spent and several
experimental programs have been carried out and are currently in progress. As
a consequence of these efforts, new configurations of beam-to-column connections
(post-Northridge) have been proposed and tested under cyclic loading histories. It is
therefore likely that in the near future the USA design practice in MRF connections
will have remarkable modifications on the basis of these recent experimental results.
In Sec. 3, the cyclic performance of both pre-Northridge and post-Northridge
connections is examined through a detailed review of the available experimental
results.
2.2. Japan
The adoption of partial perimeter MRFs, which was common design practice in
the USA before the Northridge earthquake, is not followed in Japan. Steel framed
buildings in Japan are usually highly redundant structures, where all columns are
designed to contribute to the lateral load resistance [Kurobane et al., 1997], thus
sizes of both members and welds are significantly smaller than in the US framed
structures; as reported by [Nakashima et al., 2000], beam depths of 400600 mm are
very common. In more than 90% of steel multistory frame buildings the columns are
cold-formed box sections (Rectangular or Square Hollow Section, RHS or SHS) due
to the excellent cross sectional properties to resist biaxial bending loads [Kurobane
et al., 1997], while the beam sections usually are either hot-rolled or built-up wideflange sections.
Two typical welded beam-to-column sub-assemblages reflect the common Japanese
construction practice in steel frame structures: shop welding type (mostly adopted
for low to medium rise buildings) and the site welding type (commonly used for
high rise buildings). The first one consists of the so-called column tree or Christmas tree [Nakashima et al., 1998] configuration, in which the column is fully
welded in the shop to a short segment of the beam; then the beam splice is realised through high strength bolts (Fig. 2). This type of connection is referred
to as through-diaphragm connection where continuity plates, also appointed as
diaphragm plates, are located at the position of the beam flanges within the
column. In particular, the tube section column is cut into three pieces, one used
for the column at the lower story, one for the connections panel zone, and one
00059
Fig. 2.
115
Japan beam-to-column connection: shop welding type.
for the column of the upper storey. The two diaphragm plates are inserted between
the three separate pieces and shop welded all around by complete joint penetration
(CJP) groove welds. A short segment of wide flange beam is then welded to the
column in the shop and the entire subassembly (called column tree or Christmas
tree) is then transported to the site and connected with the midportion of the beam
by means of bolted splice. This shop welded and field bolted type of construction is
very common in Japan, particularly for low and medium rise buildings [Nakashima
et al., 1998]. Beam flanges are connected to the diaphragm plates through CJP
single bevel groove welds made with gas shielded metal arc welding (GMAW)
[Nakashima et al., 2000], and fillet welds connect the beam web to the column.
Since the whole connection is welded in the shop, welding for the beam top and
bottom flanges are both performed in the flat position after turning over the connection [Kurobane et al., 1997], and the root of the CJP weld is located on the
interior side of both the top and the bottom flange [Nakashima et al., 1998].
The second connection type, the site welding type, is similar to the previous
one, except that the beam flanges are field welded to the column and the beam web
is field bolted with high-tension bolts to a shear tab, which is shop welded to the
column. The beam bottom flange is beveled so that welding can be performed from
the upper side at the erection site [Kurobane et al., 1997].
As alternative to the through-diaphragm plates, inside-diaphragm or
exterior-diaphragm plates [Tadao et al., 1997; Masuda and Tanaka, 2000] can be
introduced for guarantying the continuity of the force path from the beam flanges
to the column, even though these last diaphragm configurations are less common
than the first one.
116
00059
E. Mele
Also in Japan, several experimental investigations have been carried out after
the Kobe earthquake. On the basis of this recent experimental evidence, it can
be stated that presently the Japanese practice is moving towards adopting slight
modifications in the typical pre-Kobe connection, which do not alter the original configurations but mainly consist in improvements in details and in welding
procedures.
In Sec. 4, a detailed review of the experimental tests recently carried out in
Japan is reported.
2.3. Europe
The Europe connections in full MRFs are either welded or bolted-welded (extendedend-plate). Concerning welded connections, several differences with both the American and the Japanese ones can be evidenced.
The US design trend of adopting partial perimeter MRFs is not established in
Europe, where beams are therefore less deep (beam depth db = 300450 mm) than
the ones adopted in the current US design, also due to smaller bay span in the
building configuration.
The wide flange section beam is fully welded to the wide flange section column, with
complete joint penetration (CPJ) welds applied at the beam flanges and fillet welds
applied at both sides of the beam web (Fig. 3). Continuity plates are commonly
utilised for stiffening the column panel zone, while doubler plates are generally
Fig. 3.
Europe type beam-to-column welded connection.
00059
117
not present, thus leading to a substantial contribution of panel zone to the overall
deformation of the connection.
Also the welding technique currently adopted in Europe presents some differences with the USA procedure: in fact, after welding the beam flanges from the top,
the root opening is cleaned of impurities from the bottom, and the opening is filled
by means of fillet welds [Taucer et al., 1998]. This procedure minimises the porosities that may result at the bottom of the flange, due to the fusion of back up plates
during the welding process, and thus reduces the potentials of crack formation.
The common welding technique used is the FCAW (Flux Cored Arc Welding) with
ATAL gas protection, currently called MIG semi automatic; welding material is
basic or extra-basic (FLUXOFIL 31, AWF51).
Some experimental investigations on both welded and several other types of
beam-to-column connections have been recently carried out in Europe. As above
mentioned, beams adopted throughout these experimental programs have depth
equal to or less than 450 mm. The major findings of these recent tests are widely
illustrated in Sec. 5 of the paper.
3. Review of Cyclic Tests on USA Type Welded Connections
3.1. Premise
A great deal of experimental activity on the cyclic behaviour of moment beam-tocolumn connections has been developed in the USA since the late 1960s. As reported
in [Popov, 1988], the first cyclic investigations were mainly carried out on small
size specimens [Popov and Pinkney, 1969] and on one-third scale subassemblages
[Krawinkler et al., 1971]. It was clearly stated that small scale specimens could
not properly simulate the effects related to the actual connection details, i.e. large
welds and bolts. Therefore later experimental investigations, during the 1970s and
the 1980s, were mainly planned and performed on full scale specimens, with beam
height comprised between 400 and 600 mm. The tests showed that the WFBW connection type, described in the paragraph 2.1, had a sufficiently ductile behaviour
under cyclic loads thus it was defined as a pre-qualified detail. Only very recently,
at the beginning of the 1990s, the experimental researches carried out on the large
scale beam-column subassemblies, with member cross sections close to the ones actually adopted in partial perimeter frames [Engelhardt and Husain, 1992, 1993; Tsai
et al., 1995], suggested that, as an effect of the large size of member cross sections
and of the welds, the expected ductility of the connection could be undermined.
The results presented hereafter are divided in two main groups: before and after
Northridge. In the presentation of the test results, a brief description of the specimen
geometry, loading history, cyclic behaviour and failure modes is provided; however,
due to space limitations no graphical representation of the above information is
provided. The performance of all analysed specimens is mainly evaluated in terms
of plastic rotation capacity, thus bar charts reporting these values are provided for
each set of experimental results.
118
00059
E. Mele
Fig. 4.
USA test results (70s90s): maximum plastic rotations.
3.2. Experimental results (before Northridge)

In the bar chart of Fig. 4, five sets of test results, expressed in terms of maximum
plastic rotation experienced by the connection under increasing amplitude loading
histories, are reported. The results have been obtained in experimental investigations carried out in the USA, starting from the early 1970s up to the first 1990s, by:
(1) Popov and Bertero [1973]; (2) Popov et al. [1986]; (3) Tsai and Popov [1988];
(4) Engelhardt and Husain [1992, 1993]; (5) Tsai et al. [1995].
The major part of the tested specimens are welded flange bolted web (WFBW)
connections; among the results provided by [Popov and Bertero, 1973] two fully
welded (WW) specimens and one welded flange (WF) with no web connection
specimen are also reported. Also among the specimens tested by [Tsai and Popov,
1988] and by [Engelhardt and Husain, 1992, 1993], respectively two specimens and
one specimen with fully welded (WW) connections are also included. All tests,
except the ones reported by [Popov et al., 1986] which refer to half-scale cruciform
specimens, have been performed on beam-column subassembly representative of
exterior frame joints, with depth of the beams varying between 450 and 600 mm.
In the following some details on each single experimental research are given.
The data provided by [Popov and Bertero, 1973] refer to a wide experimental
research carried out in the early 1970s at University of California at Berkeley
on full-scale specimens of large beam-to-column connections to be used in highrise buildings. The specimen configuration was representative of an exterior frame
joint, with the beam, either W1850 (db = 457 mm) or W2476 (db = 608 mm),
attached to a short column stub in order to simulate a very rigid panel zone. A
total of 8 specimens (2 fully welded (WW) connections, 5 welded flange bolted
web (WFBW) connections, 1 welded flange (WF) with no web connection) were
tested under cyclic increasing amplitude loading history. The two WW and two
WFBW specimens showed excellent loop stability and very high ductility. The
00059
119
other three WFBW specimens failed in an abrupt manner, with negligible buckling of the beam flanges and fracture through the weld. The behaviour of the
specimen with no-web connection (WF) was remarkably good and showed that
an amount of shear can be transmitted by the flanges. Despite of some brittle
failure modes were observed, all eight specimens developed quite large plastic
rotations (1.935.49%).
During the 1980s additional experimental studies have been conducted by Popov
at the University of Berkeley. In a first series of tests [Popov et al., 1986], 8
half-scale cruciform specimens of WFBW connections, designed for a 47-storey
perimeter moment resisting frame, were subjected to cyclic loads in order to assess
the need of joint stiffeners and doubler plates. An increasing amplitude loading
history was applied to the beam tip, while a constant axial load, corresponding
to the effect of gravity loads plus seismic overturning, was applied to the column.
The experimental results showed that only two specimens experienced ductile
failures outside HAZ, with plastic rotations greater than 5% radians (5.2 and
5.8%) and significant yielding occurred both in the beam and in the panel zone.
One specimen showed failure through HAZ but still a quite large plastic rotation
(5.3%), mainly arising from panel zone distortion; two specimens reached plastic
rotations around 2.5% (2.5 and 2.7%), while three connections presented abrupt
failures through weld, with rather limited plastic rotations (0.2, 1.8 and 2.2%).
Additional analyses and experimental studies on full-scale beam-to-column subassemblies have been carried out during the 1980s still by Popov [Tsai and Popov,
1988]. In particular, 18 specimens of an exterior frame joint, representative of different connection solutions (9 WFBW strong-axis column; 5 WFBW weak-axis
column, 2 fully welded (WW), 2 extended end plate) were fabricated as cantilever
beam, with different cross section (W18 40, W18 35, W21 44, W18 46,
respectively db = 455, 450, 525, 460 mm) attached to column stub. The specimens were tested under increasing cyclic loading history. With reference to the 16
welded connections, two specimens showed plastic rotation less than 0.5%, four
showed rotation between 0.5 and 1%, four between 1 and 1.5%, one around 2%,
and only five displayed rotations greater than 2% (2.39, 3.21, 4.49, 4.82, 5.24%).
At the beginning of the 1990s experimental cyclic tests were also performed by
[Engelhardt and Husain, 1992, 1993] at the University of Texas. The tests were
performed on large scale cantilever type specimens, with different beam cross
sections (W24 55 db = 600 mm, W18 60 db = 463 mm, W21 57 db =
535 mm) and different web connection details. The beam-column subassemblage
was designed to provide a strong panel zone, thus forcing inelastic deformations
mainly in the beam. Seven WFBW and one WW moment connections were tested
and all of them fractured at the bottom flange groove weld. Three specimens failed
unexpectedly early in the loading program and showed a plastic rotation less
than 0.5%. One specimen achieved plastic rotation of 0.9% gradually fracturing
through bottom beam flange, outside of weld, while the other four specimens
developed plastic rotations between 1.2 and 1.5%.
120
00059
E. Mele
Other tests on ten beam-to-column connections using WFBW details have been
more recently conducted at National Taiwan University by Tsai, Wu and Popov
[Tsai et al., 1995], with the aim of collecting data on the performance of this type
of connection with supplemental web welds or bolts, and of assessing the effect of
column panel zone strength on the inelastic deformation of the connection. The
beam-column subassemblages were fabricated as cantilever attached to column
stubs, adopting different beam cross sections (W21 101, W21 83, W21 62,
W21 50, db ranging from 545 to 530 mm) with the same column cross section
(W14 159, dc = 380 mm). The results obtained on the ten specimens indicate
that the cyclic plastic rotation capacity never reaches 2%, ranging between about
0.9 and 1.8%: seven specimen showed rotation between 1 and 1.5% radians (1.14,
1.24, 1.12, 1.38, .92, 1.42, 1.33) and only three specimen reached rotation values
greater than 1.5% (1.68, 1.56. 1.82%).
According to some recent reviews of experimental results obtained on welded
connection specimens [De Luca and Mele, 1996, 1997; Roeder and Foutch, 1996;
AISC, 1997], a strong adverse effect of the beam depth on the inelastic rotation
capacity has been suggested: as beams become deeper and heavier, their ability
to develop inelastic rotation may be somewhat diminished. Although such scale effects are not completely understood, at least two possible detrimental scale effects
have been identified. First, as a beam gets deeper, larger inelastic strains are generally required to develop the same level of inelastic rotation. Second, the inherent
restraint associated with joining thicker materials can affect joint and connection
performance [AISC, 1997]. In order to tentatively appraise this size effect, the
plastic rotation values, already provided in Fig. 4, are reported in the chart of
Fig. 5 as a function of the beam depth. The data show the expected decreasing
trend, even though consistent scatters among the results are evident, mainly due
to the lack of full homogeneity of the data. In particular the following reasons for
&
'
<
>
Fig. 5.
USA test results (70s90s): maximum plastic rotations as a function of the beam depth.
00059
121
the observed scatters can be identified: (1) no distinction has been made between
WFBW, WW and WF connections results; (2) the results refer to different column
sizes and panel zone strengths; (3) differences exist in the cyclic loading histories
applied to the beam tip as well as in the axial load applied to the column.
3.3. Experimental results (after Northridge)
In the context of the experimental research efforts developed following the
Northridge earthquake, some results obtained in the experimental programs sponsored by the SAC (SEAOC, ATC, CUREE) Joint Venture, by the AISC (American
Institute of Steel Construction) and by the NSF (National Science Foundation) are
reported in the following.
Four classes of connection specimens are distinguished with reference to these
experimental tests:
(1) the so-called pre-Northridge specimens, close to the pre-qualified WFBW
connections currently realised up to 1994;
(2) the post-Northridge specimens, which are connections designed and detailed
according to different strategies, in order to obtain an improved cyclic behaviour
and to develop the target plastic rotation of 0.03 rad, as specified in the
Interim Guidelines developed by the SAC Joint Venture (FEMA 267, 1995);
(3) the retrofitted specimens, which are fabricated as pre-Northridge connections
and then retrofitted according to post-Northridge design strategies;
(4) the repaired specimens, which are pre-Northridge connections either tested
up to failure or taken off from actual steel MRF buildings damaged by the
Northridge earthquake, and then repaired, according to post-Northridge design strategies, and re-tested.
For all four classes of specimens, the tests have been performed on large
beam-column sub-assemblages, with depth of the beams varying between 500 and
1000 mm.
3.3.1. Pre-Northridge specimens
Concerning the pre-Northridge connection types, some results of tests performed
in the context of the SAC and other research programs are shown in Fig. 6, where
five sets of experimental data are given in bar chart format.
The bar values correspond to the maximum plastic rotation experienced by the
specimens, and the five sets of bars respectively refer to tests performed: (1) at the
University of California at Berkeley, Davis Hall, by Popov and Blondet [1996], on
three specimens, (in the chart appointed as UCB-DH 96); (2) at the University
of California at Berkeley, Earthquake Engineering Research Center by [Whittaker
and Bertero, 1995], on three specimens, (in the chart appointed as UCB-EERC
95); (3) at University of Texas at Austin, by Engelhardt et al. [1995] and Sabol
122
00059
E. Mele
Fig. 6.
USA test results (19952000): maximum plastic rotations.
et al. [1996] on four specimens (in the chart appointed as UTA 9596); (4) and
at the University of California at San Diego by Uang and Bondad [1996] on five
specimens (in the chart appointed as UCSD 96); (5) at the University of Southern
California by Anderson and Duan [1998], on three specimens (in the chart appointed
as USC 98). Some of these tests have been carried out for obtaining data to be
utilised as a benchmark for evaluating the efficacy either of improved configurations
for new beam-to-column connections [Engelhardt et al., 1995; Sabol et al., 1996] or
of different repair and retrofit techniques for connections in existing frame structures
[Popov and Blondet, 1996; Uang and Bondad, 1996; Anderson and Duan, 1998].
In Fig. 6, for the UCB-EERC and UCSD test results, both the beam and the
beam plus panel zone plastic rotations are reported. In the latter experimental
program two dynamic tests have also been performed in order to assess the effect
of strain rate on crack initiation and propagation.
The chart shows very low values of the attained plastic rotation (in the beam),
being always lower than 1% radians, and in many cases lower than 0.5% radians.
When the rotation capacity embodies the contribution of the inelastic deformation
in the panel zone, an increase in the above values is obtained, with values reaching
12% radians. For all specimens, failure modes similar to the ones observed after
the Northridge earthquake, in the connections of actual frame structures, have been
reproduced in the laboratory tests.
The comparison between the results on statically and dynamically loaded specimens (UCSD results) allows to derive the effect of strain rate on the cyclic behaviour of the connections: lower plastic rotation and energy dissipation capacities
can be observed for the dynamically tested connections, and a different failure mode,
namely propagation of fracture in the column flange, was also observed.
As for the experimental results of tests carried out before Northridge, also the
above data have been reported as a function of the beam depth in the chart of Fig. 7,
00059
123
'
&
Fig. 7.
&
%
&
USA test results (19952000): maximum plastic rotation as a function of the beam depth.
confirming the decreasing trend in the plastic rotation capacity at increasing depth
of the beam.
3.3.2. Post-Northridge specimens
The major part of experimental tests after Northridge has concerned improved
configurations of connections to be adopted in new structures (post-Northridge
connections) as well as for retrofitting and repairing existing frame joints. The
post-Northridge as well as the retrofitted and the repaired specimens are designed
according to two basic strategies:
(a) increasing the connection strength capacity, thus making the connection
stronger than the beam, by reinforcing and stiffening the portion of beam immediately close to the welds (beam flange cover plates, widened beam flanges,
upstanding ribs and haunches, vertical plates, etc.) [Figs. 8(a)8(c)];
(b) decreasing the demand at the connection, by making the beam weaker than the
connection, through a localised reduction, away from welds and HAZ, of the
beam strength (either reduced beam section, RBS, obtained through tapering
the beam flanges, or perforated beam section, PBS, obtained through drilling
holes in the beam flanges) [Fig. 8(d)].
In both the design strategies, the occurrence of yielding and plastic deformation
is forced in the beam away from the most vulnerable welded region at the column
face.
With reference to post-Northridge connections, in Fig. 9 some experimental results are reported. Also in this chart, the bars report the values of maximum plastic rotation obtained for the five sets of specimens respectively tested by:
(1) Popov et al. [1996] (in the chart appointed as UCB-DH 96); (2) Engelhardt
et al. [1995], Sabol et al. [1996] and Engelhardt and Sabol [1998] (UTA
124
00059
E. Mele
Fig. 8.
Post-Northridge connections [taken from: Engelhardt and Sabol, 1997].
9598); (3) Chen et al. [1996] (NTIT 96), and Tsai and Chen [1996] (NTU 96);
(4) Stojadinovic et al. [2000a, 2000b] (Univ. Michigan 2000); (5) Lin et al. [2000]
(NTU Taiwan 2000).
The bars in the chart are distinguished according to the strategy adopted for
improving the connection performance.
In the following some details on the various test programs are given.
At the University of California at Berkeley, Davis Hall (UBC-DH 96), a test program on both reinforced-connection specimens and weakened-beam specimens
has been carried out by Popov [1996]. Firstly, three identical specimens, appointed as APD1, APD2 and APD3, with bottom and top beam flange cover
plates were fabricated and tested up to collapse, showing quite high values of
maximum plastic rotation, respectively equal to 4.3%, 4.4% and 1.9%, but a sudden failure, with abrupt fracture of the column flange. For the APD1 and APD2
specimens, a non-symmetrical behaviour was observed at the last cycles, due to
some initial offset of the beam end. On the basis of the plastic rotation range
experienced at the last cycles (respectively +4.3%2.5% for APD1, +4.4%2.2%
00059
Fig. 9.
125
USA test results on post-Northridge connections.
for APD2), the average value of the maximum plastic rotation, has been assumed
in the bar chart of Fig. 9, as a more representative parameter of the plastic rotation capacity. The connection was then re-designed in order to achieve a better
performance, and two identical specimens (APD3 and APD4) were fabricated
with circular cuts on the beam flanges (reduced beam section, RBS) to reduce
flexural capacity thus inducing plastic deformation in the beam away from the
welded region. Improvement of the cyclic performance was observed in the tests
and maximum plastic rotations of 3.8% and 4.1% were achieved.
A large experimental program, sponsored by the AISC, on 18 full-scale cantilever
specimens (beam W36 150, db = 911 mm; column W14 455 and W14 426)
representing both pre-Northridge and post-Northridge connections has been
carried out at the University of Texas at Austin (UTA 9598) by [Engelhardt
et al., 1995; Sabol et al., 1996]. Different post-Northridge modified configurations
have been considered for the specimens, consisting of addition of cover plates (8
specimens), upstanding ribs (2 specimens) or side plates (2 specimens), in order
to reinforce the connection and to shift the location of beam plastic hinge away
from the face of the column. Two replicates of each specimen have been fabricated
and tested under cyclic increasing amplitude loading history. The conventional,
pre-Northridge specimens, despite of some improvements in the welding procedures, details and workmanship, showed a very poor behaviour with very limited
ductility (maximum plastic rotation between 0.25% and 0.9%) and sudden failure
mode due to brittle fracture in the beam bottom flange welds. On the contrary,
for all but two post-Northridge specimens, high plastic rotation capacities and
ductile deformation and failure modes were observed; in particular among eight
cover plates specimens, six exhibited plastic rotation greater than 2.5% (up to
5%), while two specimens collapsed due to brittle fracture in the weld connecting
126
00059
E. Mele
the beam to the cover plate. The performance of the specimens with upstanding
ribs and with side plates was very good as well.
The data on the performance on cover plates reinforced connections have been
enlarged by the outcomes of additional test carried out at the UTA (UTA 95
98) by [Engelhardt and Sabol, 1998]. In addition to the eight specimens already
provided in [Engelhardt et al., 1995; Sabol et al., 1996], other four large scale
specimens, characterised by different panel zone strength (beam W36 150 and
W30 148; column W14 426, W14 257, W14 455) have been tested. The
experimental results confirm the dramatic improvement in the cyclic performance
and plastic rotation capacity (always greater than 3.3%) that can be obtained by
using this joint configuration.
Even though not performed in the USA, but at the National Taiwan Institute of
Technology and at the National Taiwan University, the experimental tests presented by [Chen et al., 1996] (NTIT 96) and by [Tsai and Chen, 1996] (NTU 96)
are here reported, since they are carried out on specimens which can be embedded in the post-Northridge category, i.e. connections detailed with the specific
aim of improving the cyclic behaviour and the maximum plastic rotation. In the
same way as [Popov et al., 1996], the basic idea in [Chen et al., 1996] is tapering
the beam flange in order to localise the plastic hinge region; in five specimens the
geometrical parameters of the cut in the beam flanges are varied. The results of
the tests reveal high plastic rotations, going from 2.35% to 4.88% rad. In [Tsai
and Chen, 1996] the cyclic tests performed on 12 specimens, representative of
differently modified WFBW connections are presented. The tentative improvement of the cyclic behaviour is obtained by different schemes, which rely on one
of the following principles: (a) increasing the connection strength capacity, (fully
welding the beam web to the shear tab, applying flange cover plates, flange wing
plates and widened beam flanges) or (b) decreasing the strength capacity of the
beam away from the connection (employing drilled holes in the beam flanges).
The maximum plastic rotations obtained in the cyclic tests are satisfactory, varying between 1.4% and 3.8% rad.
More recently some experimental tests on improved connection specimens have
been carried out always at the National Taiwan University by [Lin et al., 2000]
(NTU Taiwan 2000). The tests have been planned on four double-side beamcolumn subassemblage with RC slab, with the aim of assessing the effect of panel
zone stiffness and strength (presence and thickness of doubler plates) on the
experimental behaviour. The welded connection configuration was very similar
to the pre-Northridge practice, since only improvements in the welding details
and procedures have been adopted. The test results prove the efficacy of the
improvements on the performance of the connections, all showing plastic rotations
greater than 3% (between 3.1 and 5.6%), with increasing values for decreasing
panel zone yielding strengths. The failure mode of the three stiffened panel zone
specimens involves ductile fracture in the beam bottom flange, while the weakest
panel zone specimen (no doubler plates) collapsed due to ductile fracture arising
00059
127
from severe local links in the beam and column flanges at the four corners of the
panel, after that large plastic rotations were experienced by the specimen.
Tests on modified connection details have been recently carried out also at the
University of Michigan by [Stojadinovic et al., 2000a, 2000b] (Univ. Michigan
2000). In particular some weld fracture mitigation measures, including use of
notch-tough weld metal and improvements in weld detailing practice, have been
applied to ten full scale specimens, representing beam-to-column connections
characterised by global configuration close to the pre-Northridge ones. Different
couplings of beam and column cross sections (beams: W36 150, W30 99,
W24 68; columns: W14 257, W14 176, W14 145, W14 120) have been
utilised for addressing the effect of beam size and of panel zone strength on
the cyclic behaviour of the specimens. As reported by the authors, the plastic
rotation was of 1.5% on average (with values ranging between 0.6 and 2.6%), and
all specimens failed in the same way, due to fracture in the beam flanges between
the weld and the web access hole. The poor cyclic behaviour demonstrated that
fracture mitigation measures alone cannot give rise to significant improvements,
which can be obtained by modifying also the connection global configuration and
the force transfer path.
3.3.3. Retrofitted specimens
Several experimental studies have also been devoted to evaluate the possibility of
utilising the same Post-Northridge details proposed for new connections (based either on the addition of reinforcing elements or on the reduction of beam strength)
for retrofitting of existing pre-Northridge connections. Some major experimental
results are reported in Fig. 10, again in terms of maximum plastic rotation experienced by the retrofitted specimens. In particular the tree sets of bar values refer
to results respectively provided by: (1) Uang et al. [2000a, 2000b] by Yu and Uang
[2000] and by Civjan et al. [2000], concerning RBS specimens (in the chart appointed as UCSD and UTA [2000]; (2) Uang et al. [2000a, 2000b] by Yu and Uang
[2000] and by Civjan et al. [2000], concerning welded haunches specimens [UCSD
and UTA, 2000]; (3) Anderson and Duan [1998] (USC 98).
The experimental results are discussed in the following.
An experimental research project, funded by NIST and AISC, has been carried
at the University of California at San Diego and at the University of Texas at
Austin [UCSD and UTA 2000], respectively by [Uang et al., 2000a, 2000b; Yu
and Uang, 2000] and by [Civjan et al., 2000]. Six (at UCSD) plus six (at UTA)
full size interior joint specimens have been fabricated and assembled according
to the pre-Northridge practice, both with and without RC slab, and then have
been retrofitted. Two alternative solutions have been chosen for retrofitting the
specimens: addition of welded haunch at the beam bottom flange and reduction of
the beam section (RBS) through dog-bone cut of the bottom flange. The above
solutions have been selected as well suited for retrofit applications since they
128
00059
E. Mele
Fig. 10.
USA test results on retrofitted connections.
require little or no modifications to the beam top flange, thus minimising the need
to remove or alter the floor slab [Civjan et al., 2000]. The six UCSD specimens
consisted of a W36150 beam and a W14426 column, while the UTA specimens
consisted of a W30 99 beam and a W12 279 column. The results showed that
the RBS solution does not prevent brittle fracture in the top flange groove weld,
while the addition of welded haunches gives rise to significant improvements in
the cyclic performance. Some limited benefits have been obtained for the RBS
retrofitted specimens by replacing existing groove welds with higher toughness
welds.
Different alternative retrofit solutions for pre-Northridge connections, consisting
in addition of cover plates, addition of vertical plates, weld overlays, reduction
of beam section through drilling holes in the flanges have been tested at the
University of Southern California (USC 98) by [Anderson and Duan, 1998]. A
total of ten retrofitted specimens have been tested, and the authors report that
both the specimens retrofitted with weld overlay and the specimens with cover
plates met or exceeded the 3% plastic rotation target value. The specimens with
vertical plates showed plastic rotation ranging between 2.5 and 3%, thus still
satisfactory even though not excellent. Also the specimen with drilling holes
in the beam flange (Perforated Beam Section, PBS) developed 2.5% of plastic
rotation, showing a good, but not outstanding, performance.
3.3.4. Repaired specimens
In Fig. 11, three sets of experimental results on repaired connections, i.e. preNorthridge connection specimens, firstly tested up to failure and then repaired
according to different procedures, are reported.
The data in the bar chart provide a comparison between the maximum plastic
rotations experienced by the original and the repaired specimens, tested at: (1)
the University of California at Berkeley by Popov and Blondet, 1996 (in the chart
00059
Fig. 11.
129
USA test results on repaired connections.
appointed as UCB-DH 96); (2) at the University of California at San Diego by

Uang et al. [1998] (UCSD 98); (3) at the University of Southern California by
Anderson and Duan [1998] (USC 98).
In the following these experimental results are discussed.
The three couples of full-scale specimens tested by [Popov and Blondet, 1996]
(UCB-DH 96) consist of nominally identical exterior beam-column subassemblage, with W36 150 beam and W14 257 column. The behaviour of the three
original, pre-Northridge connections was particularly poor, with maximum plastic rotation between 0.34 and 1.07%. The specimens collapsed suddenly with
brittle fractures in the column or in the beam bottom flange weld. Current repair
procedures, which consisted of replacing the cracked portions of the elements and
the existing groove welds, have been utilised for all damaged specimens before
re-testing. The repaired connections showed a slightly better performance, even
though the maximum plastic rotations (1.55, 1.6, 1.77%) were still quite low and
the failure modes were a non-ductile sudden fracture.
Four identical full-scale pre-Northridge specimens, simulating an exterior beamcolumn subassembly, have been tested up to failure statically and dynamically
at UCSD (UCSD 98) by [Uang et al., 1998], showing poor performance and low
plastic rotations (0.152%). After repair, by adding a haunch beneath the beam
bottom flange, the specimens have been re-tested and have showed a significant
increase of the cyclic performance both in the static and dynamic range: the
adoption of the haunch effectively shift the critical plastic hinge section in the
beam away from the column face, allowing for the attainment of high level of
inelastic deformations (up to 5.2%). The not very high values of rotation reported
in the chart of Fig. 11 for the repaired specimens tested under dynamic loads is
due to the fact that a maximum displacement was imposed to the specimens
during testing. Thus the differences between the dynamic performance of the
specimens before and after repair can be better appreciated in terms of energy
130
00059
E. Mele
dissipation capacities, which for all repaired specimens were much higher than
those of original specimens.
The tests carried out by [Anderson and Duan, 1998] (USC 98) on pre-Northridge
and repaired specimens were aimed to compare the effectiveness of different repair/retrofitting procedures. The data reported in Fig. 11 concern the results on
specimens tested up to failure and then repaired by weld replacement or weld
overlay. The specimens repaired through weld replacement class B (gouging out
existing weld and re-welding using SMAW with an E7018 electrode) showed an
increase of ductility capacity, but the improvement was not always enough to
reach the 3% target plastic rotation (2% and 2.4%), while a plastic rotation
equal to 3.5% was experienced by the specimen repaired using a class A weld
overlay with SMAW and E7018 electrodes.
4. Review of Cyclic Tests on Japan Type Welded Connections
In the Japanese scientific community the 1995 Hyogoken-Nanbu earthquake has
played the same role as the Northridge earthquake for the USA, since the performance of steel structures was not as good as expected and brittle failure modes were
evidenced at several beam to column connections. Therefore also in Japan, in the
second half of the 1990s a great deal of experimental activity has been developed to
clear out the reasons of the observed damage. In particular large experimental programs have been carried out by the Japanese Society of Steel Construction [JSSC,
1997], by the Steel Committee of the Kinki Branch of the Architectural Institute of
Japan [Nakashima et al., 1998], and also by all major Universities and construction
companies of the country [Tanaka et al., 1997].
Despite of the comparable research efforts spent in USA and in Japan, significant differences exist between the approaches adopted for improving the connection behaviour: while in the research developed in USA after Northridge,
modified configurations of the beam-to-column connection (the so-called postNorthridge connections) have mainly proposed and tested in experimental programs, the investigations carried out in Japan after Kobe started from the idea
that modifications only of small details (mainly presence, shape and size of weld
access hole) combined with improved welding procedures (weld sequence, placement of backing bars and of weld tabs), without modifying the global joint configuration, could be considered a solution to the connection fracture problem. Weld
tabs and backing bars, panel strength and diaphragm thickness, loading rate and
temperature, were additional major test parameters investigated in the Japanese
laboratories. Among the several experimental researches carried out in Japan, some
test results obtained under increasing applied displacement histories are reported
in Fig. 12, in terms of maximum plastic rotation. The test data are grouped in
four sets: (1) Matsui and Sakai [1992]; (2) JSSC [1997]; (3) Kurobane et al. [1997];
(4) Nakashima et al.[1998].
In the following the major experimental findings are discussed in detail.
00059
Fig. 12.
131
Japan test results: maximum plastic rotations.
The first data shown in the chart refer to the experimental program carried out
before the 1995 Hyogoken-Nanbu earthquake, in the 1992, by [Matsui and Sakai,
1992] on welded beam-to-column connections, with the specific aim of assessing
the effect of panel zone design and of the connection details (presence of scallops
in the beam web) on the failure modes of the connections. Both the beam and
the column sections were built-up wide flange, with depth equal to 200 mm.
The experimental results, particularly interesting under different points of view,
show maximum plastic rotations ranging from 2.5 to 6.5% radians. The best
performance has been observed for the specimens designed in order to have plastic
deformations both in the beam and in the panel zone. The test results also show
a significant effect of the scallop on the ductility and energy dissipation capacity
of the connections: in the specimens with scallop, a crack generally occurs at the
weld of beam flange and propagated in the HAZ, driving to a rapid deterioration
of the connection strength; on the contrary, the specimens without scallops show
a gradual deterioration of the strength due to buckling of the beam flange at the
plastic hinge location.
The data indicated as JSSC 97 are some experimental results obtained in the
context of a wide research promoted by a Special Task Committee of the Japanese
Society of Steel Construction [JSSC, 1997], which was formed for assessing the
impact of the Hyogoken-Nanbu earthquake on steel building frames. In particular
several experimental investigations on specimens representing actual connections
were carried out in this context for understanding the causes of damage and suggesting improvements to the connection details and welding procedures. Many
of these test programs have been planned as multi-specimen testing programs,
where several identical specimens are subjected to constant amplitude loading
histories, thus in Fig. 12 only the available results under cyclic increasing amplitude loading histories, obtained by [Teraoka et al., 1997], are provided in terms
of maximum plastic rotations. These results refer to eight full scale specimens
132
00059
E. Mele
representing interior frame joints. A H-type beam cross section (H500 200
12 19, db = 500 mm) and a square tubular column cross section (375 375 32,
dc = 375 mm) were adopted for all specimens. CJP welds were applied to the
beam flanges, while high strength friction type bolts were adopted for the beam
web connection. Several parameters were varied among the specimens: shape and
location of the weld access hole, welding method and details, number of web bolts.
The test outcomes provide plastic rotations between 2.03% and 2.88% and stable
hysteresis loops.
The experimental tests reported by [Kurobane et al., 1997] have been carried
out on four full scale one-sided specimens, representing an exterior beam column
subassemblage. The beam cross section in all specimens was WF3503571919
(db = 350 mm), while the columns were wide flange sections WF500 500
25 259 (dc = 500 mm) for two specimens and RHS BOX-500 500 22 (db =
500 mm). In all specimens a cover plate was welded on the beam top flange and an
artificial weld defect was introduced in the beam bottom flange groove weld. All
tested specimens developed a plastic hinge at the beam end close to the column
face and the results show quite high plastic rotation capacities, even though the
global performance and the failure mode proved to be strongly affected by the
geometry and locations of weld defects.
Finally in Fig. 12 are reported some results obtained by [Nakashima et al., 1998]
in the context of a comprehensive experimental project organised by the Steel
Committee of the Kinki Branch of the Architectural Institute of Japan. In this
project a total of 86 full scale beam-to-column specimens have been tested under
cyclic loads. The results reported in the figure refer to 14 single sided specimens,
tested up to failure under both static and dynamic cyclic loading histories. The
cross sections adopted for the subassemblage members, a wide flange section 600
2501925 (db = 600 mm) for the beam and a square tube section 45045019
(dc = 450 mm) for the column, are typical in low-to-medium rise steel buildings
in Japan. The column cross section give rise to panel zone strength much larger
than the beam flexural strength, thus the effect of panel zone deformations on
the specimen behaviour has been excluded. The beam-to-column connection was
shop welded with two through-diaphragm plates. Testing variables concerned:
type of steel and of connection, details of the weld access hole, diaphragm plates
and backing bars, weld tabs and welding parameters. The results of tests showed
plastic rotation capacities quite high, going from 3.5% to 5.1%, both for statically
and dynamically loaded specimens, thus evidencing that dynamic loading has
not a detrimental effect on the performance of the connection. Another major
outcome concerns the detailing of the weld access hole, which significantly affects
the failure mode of the specimens.
In Fig. 13, all above test results have been plotted as a function of the beam
depth: also in this case the beam size seems a key factor in affecting the inelastic
capacity of the beam-to-column connection.
00059
133
'
<
>
"
"
#
Fig. 13.
"
$
#
Japan test results: maximum plastic rotation as a function of the beam depth.
5. Review of Cyclic Tests on Europe Type Welded Connections

In last years the European design approach of beam-to-column connections has
been strongly influenced by the recently drafted Eurocode 3 on Design of Steel
Structures, particularly the Annex J Joints in Building Frames, which is specifically devoted to beam-to-column joints. In this annex, a component approach for
modelling, design and interpretation of the joint behaviour is utilised.
In this perspective, several experimental researches have been planned and carried out both on single joint components and on entire beam-column subassemblies.
The first type of testing is generally devoted to the assessment of the single components behaviour, while the second type of tests is focused to the global joint
response. Quite often these latter tests have been planned and carried out on alternative connection solutions designed for the same beam-to-column joint. This is
the case of the experimental investigations carried out: in Italy, at the University of
Milano in the context of different research programs [Ballio et al., 1987; Ballio and
Chen, 1993; Plumier et al., 1998; Castiglioni et al., 2000]; in Portugal at the IST
of Lisbon during the 1990s [Calado et al., 1999; Calado and Mele, 2000a, 2000b];
at the European Joint Research Center in Ispra [Kakaliagos and Bouwkamp, 1993;
Taucer et al., 1998].
As already pointed out, in the current European practice of MRF design, beam
depth ranging between 300 and 450 mm is usually adopted, therefore tests are
carried out on full scale beam-column subassemblies characterised by these member
dimensions.
With specific reference to tests on welded connections, some results obtained
under increasing amplitude loading histories are reported in Fig. 14 in terms of
maximum plastic rotation values.
In the following each series of experimental results is briefly discussed.
One of the first cyclic test programs carried out in Europe on beam-to-column
connections is the one developed in the late 1980s at the University of Milano
134
00059
E. Mele
Fig. 14.
Europe test results: maximum plastic rotations.
and reported in [Ballio et al., 1987]. A total of 14 specimens representative of

exterior frame joint, belonging to 4 different connection typologies (top and seat
angles, cover plates, end plate, fully welded) have been tested by applying the
ECCS stepwise increasing amplitude loading history [ECCS, 1986]. An axial load,
approximately equal to 1/3 of the nominal squash load, was applied to the column
and left constant during the tests. Both the beam and the column cross section
were hot-rolled IPE300 (db = dc = 300 mm). Major aims of the research were the
assessment of the comparative behaviour of the connection typologies and of the
effect, for each connection type, of different local stiffening elements on strength,
ductility, energy dissipation capacity and collapse mode. Two tests have been
performed on CPJ fully welded connections, with and without doubler plate
(respectively appointed as D1 and D2 specimens). Quite high values of plastic
rotation have been observed for both the specimens (respectively 8% and 7%),
while the inelastic deformation and collapse modes of the specimens have been
remarkably different: mainly governed by the panel zone behaviour in the first
case, while in the second case the supplementary web plates, giving rise to an
increase of shear strength of the panel zone, shift the occurrence of major inelastic
deformations in the beam. The comparison of the hysteresis loops reflects the
above differences: while for the D1 specimen the loops are wide, stable and quite
regular up to failure, for the D2 specimen an increase of stiffness and strength,
but a reduction of maximum rotation and of energy dissipation capacity can be
derived, due to the strength degradation at increasing number of cycles related
to the local buckling of the beam flanges.
Another wide experimental research has been carried out again at the University
of Milano by Ballio and Chen [Ballio and Chen, 1993] on beam-to-column subassemblies representative of both exterior and interior joints of frame structures
adopting different connection solutions, i.e. end plate, welded flange cover plates
and bolted web, fully welded. In all tests both beam and column sections were
H-shaped, respectively HE260A (db = 250 mm) and HE300B (dc = 300 mm).
00059
135
The tests have been performed both on bare steel and steel-concrete composite
joints differently detailed, i.e. with and without concrete slab, with and without
stiffeners, with and without doubler plates. The cyclic loading history applied to
the specimens throughout the experimental program is the one suggested by the
ECCS Recommendations [ECCS, 1986]. With reference to the bare steel specimens with fully welded connections (appointed as E1 and F1), the experimental
results show very high plastic rotation capacities, being the ultimate plastic rotation (conventionally defined as the maximum plastic rotation corresponding to
three stable loops) always grater than 7% radians. Interior welded flanges bolted
web specimen (I1) shows even higher plastic rotation (> 10%). The adoption of a
reduced beam section (RBS) in proximity of the connection leads to an increase
of ultimate plastic rotation from 7.66% to 8.68% radians (specimens E1 and F1),
a decrease of maximum bending moment, and a more significant plastic engagement of the beam than the panel zone. As a consequence, strength degradation
is evident in the hysteresis loops of the RBS specimen F1, while the specimen E1
shows a very stable cyclic behaviour.
In [Kakaliagos and Bouwkamp, 1993] a collection of 7 cyclic tests on a typical first
storey exterior joint of a 11-storey-3-bay frame is reported. The test specimens
have been designed to eliminate the shear distortion of the column web, thus
consisting of a cantilever beam and a short column stub rigidly connected to
the reaction frame. The beam cross section is IPE600 (db = 600 mm), while the
column cross section is HE400M (dc = 432 mm). The connection between the
beam and the column is either fully welded or extended end plate. The tests have
been aimed to evaluate the effect of continuity plates and of filled-in concrete on
the behaviour of the joints under cyclic stepwise increasing displacements. The
results reported for the bare steel fully welded specimens show a stable behaviour,
with slight strength degradation due to plastic buckling of beam flanges, and
maximum plastic rotations of 2.7% radians, with a very low contribution of panel
zone (0.18%).
Among the several experimental investigations carried out during the 1990s at the
Instituto Superior Tecnico of Lisbon [Calado et al., 1999; Calado and Mele, 2000a,
2000b], in the chart of Fig. 14 only the test results concerning fully welded connections under increasing amplitude history are reported [Calado et al., 1999]. The
specimens are T-shaped beam-column subassemblages, consisting of a 1000 mm
long beam and a 1800 mm long column. In the three types of specimens, respectively appointed as BCC5, BCC6 and BCC8, the beam cross section is the same
(IPE300, db = 300 mm), while the column cross section is varied, in order to
assess the effect of the column size and of the PZ design on the connection behaviour. In fact, the beam plastic modulus is respectively larger, approximately
equal and smaller than the column plastic modulus for the BCC5, BCC6 and
BCC8 specimens. The tests have evidenced the effect of beam-to-column strength
ratio and of the panel zone design on the cyclic behaviour and failure modes of
the connections, as well as the dependency of cyclic performance parameters
136
00059
E. Mele
and of the failure modes upon the applied loading history. In Fig. 14, both the
values of the total (beam + panel zone) plastic rotation and the values of the
beam plastic rotation are reported for the three specimens. It can be observed
that, with the panel zone contribution, the connections experience large inelastic
deformations, respectively equal to 5.5% for the BCC5 specimen, 4.5% for the
BCC6 specimen and 3.9% for the BCC8 specimen. Disregarding the contribution
of panel zone distortion, quite low values of beam plastic rotations are obtained,
respectively equal to 0.57%, 1.75% and 2.42% for the three specimens, thus confirming that large rotations can be experienced thanks to column web panel
deformations.
A significant contribution to a deeper understanding of the cyclic behaviour of
welded beam-to-column connections is provided by the experimental researches
developed in the context of the EC Steelquake Project [Calado et al., 1998;
Plumier et al., 1998; Taucer et al., 1998; Castiglioni et al., 2000]. (The results of
this experimental research are not reported in the chart of Fig. 14, since all tests
have been carried out under constant amplitude loading histories.) A total of 32
cyclic tests on both fully welded and end plate connections have been carried out
at the Universities of Milano and Liege. The beam-column subassembly, consisting of a IPE450 beam (db = 450 mm) and a HEB300 (dc = 300 mm) column
represents an exterior frame joint. The 21 tests on the CPJ welded connections
have been planned with the major aim of assessing the effect of doubler plates as
well as of welding procedure and details (double or single bevel groove weld, V
or K shape of groove weld, shape and material of backing bar, presence of web
access hole). A multi-specimen testing program, in which identical specimens
are subjected to constant amplitude loading histories, characterised by different
amplitude of the applied displacement, has been adopted. A major outcome of
experimental results on the welded specimens concerns the effect of loading history on the failure mode of the specimen: two failure modes, brittle and ductile,
and an intermediate (mixed) one, have been observed depending on the displacement ductility required to the specimen through the applied displacement history
(ratio of the applied displacement v to the yielding displacement vy ). In particular for small ductility ranges the specimens show brittle failure mode, with
sudden collapse occurring without previous degradation of the connection performance parameters, due to crack at weld toes or in the base material. On the
contrary, for large ductility ranges, a ductile failure has been observed: a plastic hinge usually develops in the beam, local buckling of the flanges occurs and
consequently gradual deterioration of strength and energy dissipation properties
can be observed from the hysteresis loops.
The above test data are provided in the chart of Fig. 15 as a function of the
beam depth. As already stated with reference to other experimental results, the
detrimental effect of the beam depth on the inelastic rotation capacity of the connection can be derived.
00059
137
&
'
<
<
Fig. 15.
Europe test results: maximum plastic rotation as a function of the beam depth.
6. Evaluation of Experimental Results and Observations

6.1. USA connections
From the analysis reported in the previous sections and neglecting all other phenomena which are presently being investigated and considered to be possibly coresponsible of the connection earthquake damages, (i.e. large vertical component
in the seismic input, low redundancy of the structural systems, poor detailing and
workmanship, etc.), it can be stated that the USA pre-Northridge welded connections very often exhibit abrupt and premature failure. The inelastic behaviour is
characterised by a large variability, the rotational capacity ranging from less than
1% to more than 5% radians. However in many cases, especially when deep beams
are adopted, the connections are characterised by very poor rotation capacities,
approximately of the order of 0.01 radians.
Significant improvements in the cyclic behaviour of the US welded connections
can be obtained by modifying the connection global configuration and the force
transfer path (Post-Northridge connections) either through reinforcing the connection or through weakening the beam. Cyclic tests carried out on the different postNorthridge specimens as well as on pre-Northridge specimens repaired or retrofitted
according to post-Northridge strategies, have shown a dramatic improvement of the
cyclic global performance and maximum plastic rotations frequently higher than
3% have been achieved.
In addition to modified design approaches, important outcomes of the experimental investigations concern the assessment of the effects of a number of fabrication related factors (welding procedure and details, removal of backing bar and
weld tabs), material related factors (overstrength, notch toughness), and applied
loading related factors (static vs. dynamic loading, strain rate effects, pulse type
loading) on the seismic performance of the beam-to-column joints.
138
00059
E. Mele
6.2. Japan connections

The investigations carried out in Japan after Kobe started from the idea that there
is no need for modifying the joint configuration, but only minor modifications of
details (mainly presence, shape and size of weld access hole) combined with improved welding procedures (weld sequence, placement of backing bars and of weld
tabs) could be considered a solution to the connection fracture problem. Weld
tabs and backing bars, panel strength and diaphragm thickness, loading rate and
temperature, were additional major test parameters investigated in the Japanese
laboratories. Some major experimental findings summarised in the following:
The loading rate, though affecting the connection yield load and maximum
strength, has a little influence on the energy dissipation and collapse mode
[Terada et al., 1997]. Furthermore, Nakashima et al. [1998] showed that specimens subject to dynamic loading generally exhibited higher rotation capacity,
and increase of strength by about 510%.
Temperature affects both energy absorption and fracture modes [Terada et al.,
1997].
Absence of the so-called scallop (weld access hole) leads to improvements both
of the strength and deformation capacity of the specimens [Tadao et al., 1997]
as well as appropriate modifications of the weld access hole can lead to decrease
of the stress and strain concentration and to higher cumulative plastic rotation
capacity [Nakashima et al., 1998].
Inside-diaphragm specimens show better cyclic behaviour than throughdiaphragm specimens [Tadao et al., 1997].
6.3. Europe connections
Experimental researches carried out in Europe have evidenced the following aspects:
High global and plastic rotation capacities of the Europe type fully-welded
connections;
Dependence of the failure modes on the applied cyclic rotation history;
Role of the panel zone deformation in reducing the beam rotation demands.
The quite high values of the plastic rotations experienced by the European connections, as compared to US-type welded connections can be related to the beam
size, much smaller than the ones adopted in the US practice, and to a significant
contribution of panel zone deformation. In addition, fully welded connections, have
already shown in past experimental tests higher rotation capacity than the weldedflange-bolted-web connections, as widely reported in the inherent literature [Popov
and Bertero, 1973; Popov and Tsai, 1989; Usami et al., 1997], mainly due to the full
engagement of both the flanges and the web in resisting the applied cyclic loads.
It has to be noted that when the beam has depth more similar to the ones
adopted in the USA and when the design of panel zone drives to a little contribution
00059

139
3
3
"
<
Fig. 16.
depth.
7
=
>
>
USA, Japan, Europe test results: maximum plastic rotation as a function of the beam
of the panel distortion to the global deformation, the plastic rotation of the beamcolumn assembly significantly decreases: the results presented by [Kakaliagos and
Bouwkamp, 1993], where db = 600 mm, confirm this trend.
6.4. Effect of the beam depth
The review of the large number of experimental results has suggested to tentatively
appraise the adverse effect of the beam depth on the inelastic rotation capacity of
the connection, as already done in the Figs. 5, 7, 13, 15, separately for the USA,
Japan and Europe experimental data.
In order to analyse this effect on a wide range of beam depth, in Fig. 16 all
previous experimental values of plastic rotation are plotted against the beam depth.
Among the US test results reported in the chart, only the ones obtained on preNorthridge connection specimens have been considered, since most of the postNorthridge connections have a global configuration which strongly modifies the
force transfer path.
The values of plastic rotation in Fig. 16 vary from less than 1% to 8.5% and are
differentiated as results coming from Europe, USA or Japan researches. The wide
scatters among the experimental data which is evident in the chart are related to
the following reasons:
(1) The information on the tests have been mainly derived from scientific papers,
and the values of plastic rotation capacity are the ones directly provided by
the authors, thus they refer either to a conventionally defined collapse state
(attainment of a prescribed value of strength, stiffness or energy loss; occurrence of significant buckling; manifestation of cracks; physical rupture of the
specimen; etc.) or to the attainment of the test set-up capacity;
(2) Some results have been obtained from tests where premature weld fractures
occurred, due to weld defects;
140
00059
E. Mele
(3) No distinction has been made among fully welded specimens, welded flange
bolted web specimens, welded flange with no web connection specimens;
(4) The specimens have been assembled with different welding techniques and details (type of electrode, treatment of steel backing, geometry of weld access
hole, etc.);
(5) The test results refer to different column sizes and panel zone strengths;
(6) The tests have been carried out with different values of the axial load applied
to the column, which can significantly affect the panel zone rotation capacity;
(7) Differences exist in the increasing cyclic loading histories applied to the beam
tip, i.e. differences in the increase of applied displacement in each step and
differences in the number of cycles per step;
(8) Finally, some scatters are expected also for nominally identical specimens tested
in identical conditions, due to the randomness of the actual mechanical properties of materials.
It must be underlined that a deeper evaluation of the experimental data collected in
the literature is still needed, and that a selection and re-adjustment of these values
could reduce the scatters.
However, despite of the existing scatters, a significant trend for reduced plastic
rotation with greater beam depth can be clearly observed in Fig. 16, which confirm
the strong scale effect on the flexural ductility of connections, already found by
[Roeder and Foutch, 1996] on the basis of only US test results, and by [De Luca and
Mele, 1996, 1997] on the basis of a data base more limited than the one presented
in this paper.
A linear regression analysis of the data provides the following relationship:
pl = 6.163 0.006 db
(1)
with pl in % and db in mm.

On the basis of this relationship, the ranges of plastic rotation capacity, as
dependant on the ranges of beam depth commonly utilised in the design practice,
can be predicted for the USA, Japan and Europe type connections. In Table 1,
these ranges of values, computed by means of Eq. (1), are reported.
Table 1. Rotation capacity of connections

as a function of the beam depth.
db [mm]
pl [%]
USA
6001000
2.56 0.16
Japan
400600
3.76 2.56
Europe
300450
4.36 3.46
00059
141
The values provided in Table 1 can be considered as a rough estimate of the

plastic rotation capacity, since Eq. (1) is derived from data characterised by large
scatters and only accounts for a single, though important, parameter, among the
several ones which affect the cyclic performance and rotation capacity of connections. Nevertheless, the ranges of values of Table 1, quite closely match the ranges
of actual rotation values observed in the laboratory specimens and reported in
this paper.
7. Summary and Conclusive Remarks
In this paper, the aspects characterising the USA, Japan and Europe design practice in beam-to-column connections have been reported, the differences have been
underlined and the main issues recently addressed in experimental researches have
been briefly reviewed.
The review of the Europe, USA and Japan test results has led to the following
major observations:
Recent US tests on pre-Northridge specimens confirm that this WFBW connection type very often exhibits poor inelastic behaviour, abrupt failure, and low
plastic rotations, approximately equal to or less than 0.01 radians. On the contrary,
cyclic tests carried out on the different post-Northridge specimens (reinforced
connections and RBS connections) as well as on pre-Northridge specimens repaired
or retrofitted according to post-Northridge philosophies, have shown a dramatic
improvement of the cyclic global performance and maximum plastic rotations frequently higher than target value of 3% have been achieved.
Results of experimental researches carried out in Japan have shown plastic rotation capacities quite high both for statically and dynamically loaded specimens,
thus evidencing that dynamic loading has not a detrimental effect on the performance of the connection. Major findings concern the temperature effect both on
energy absorption and fracture modes, as well as the importance of size and shape
of the weld access hole on the failure mode of the specimens.
European-type welded connections exhibit high global and plastic rotation capacities and the role of the panel zone in reducing the beam rotation demands is
confirmed in several experimental tests.
The values of the maximum plastic rotations experienced by American, Japanese
and European connection specimens have been compared; from a preliminary
screening of this wide number of experimental data a significant dependence of
plastic rotation capacity on the beam depth has been found. A linear relationship,
resulting from the regression analysis of the experimental data, has been provided
for approximately predict the plastic rotation capacity as a function of the beam
depth. However, since all other factors which affect the plastic rotation capacity
of welded connections (such as contribution of panel zone deformation, effect of
loading history, value of the axial load in the column, etc.) have been neglected,
the prediction derived from the linear relationship should be regarded as indicative
of the order of magnitude.
142
00059
E. Mele
References
Anderson, J. C. and Duan, X. [1998] Repair/Upgrade procedures for welded beam to column connections, Rep. No. PEER-98/03. Pacific Earthquake Engineering Research
center, University of California, Berkeley, California.
AISC [1997] Seismic provisions for structural steel buildings.
Ballio, G. and Chen, Y. [1993] An experimental research on beam to column joints:
exterior connections, Proc. XIV National Conf. Steel Structures CTA, Viareggio,
Italy, pp. 110120.
Ballio, G., Calado, L., De Martino, A., Faella, C. and Mazzolani, F. M. [1987] Cyclic behaviour of steel beam-to-column joints experimental research, Costruzioni Metalliche
2, 6990.
Bruneau, M., Uang, C. M. and Whittaker, A. [1998] Ductile Design of Steel Structures
(McGraw-Hill, New York).
Calado, L., Castiglioni, C. A., Barbaglia, P. and Bernuzzi, C. [1998] Seismic damage
criteria based on cumulative damage concept, Proc. 11th ECEE, Euro. Conf. Earthq.
Engrg., Paris, France, September, A.A. Balkema/Rotterdam.
Calado, L., Mele, E. and De Luca, A. [1999] Experimental investigation on the cyclic
behaviour of welded beam-to-column connections, Proc. Eurosteel, 2nd European
Conf. Steel Structures, Praha, Czech Republic, May, Paper No. 215.
Calado, L. and Mele, E. [2000a] Experimental behaviour of steel beam-to-column joints:
fully welded vs bolted connections, Proc. 12th WCEE, World Conf. Earthq. Engrg.,
Auckland, New Zealand, January, Paper No. 2570/6/A.
Calado, L. and Mele, E. [2000b] Cyclic tests on bolted and welded beam-to-column
connections, ISET J. Earthq. Tech. 37(4), 6588, eds. Gupta, V. K. and Krawinkler,
H., Special Issue on Experimental Techniques.
Castiglioni, C. A., Bernuzzi, C. and Calado, L. [2000] Cyclic behaviour of steel beam-tocolumn joints with concrete slab, Proc. STESSA 2000, 3rd Int. Conf. Steel Structures
in Seismic Areas, Montreal, Canada, August, eds. Mazzolani, F. M. and Tremblay,
R. (A.A. Balkema), pp. 147154.
Chen, S., Yeh, C. H. and Chu, J. M. [1996] Ductile steel beam-to-column connections for
seismic resistance, J. Struct. Engrg. ASCE 122(11), 12921299.
Civjan, S. A., Engelhardt, M. D. and Gross, J. L. [2000] Experimental program
and proposed design method for the retrofit of steel moment connections, Proc.
12th WCEE, World Conf. Earthq. Engrg., Auckland, New Zealand, January,
Paper No. 257.
De Luca, A. and Mele, E. [1996] The recent Kobe earthquake: general data and lessons
learned on steel structures, in Earthquake Performance of Civil Engineering Structures (COST-C1-Seismic Working group Report), ed. Pinto, A. V., JRC-ELSA Special Publication No. I.96.56, July, pp. 120.
De Luca, A. and Mele, E. [1997] Behaviour of steel structures in seismic zones: design
aspects and research developments, Proc. of 1st ENCMM Portugal National Meeting
on Steel and Composite Construction, Porto, Portugal, November, pp. 589618.
ECCS [1986] Seismic design. Recommended testing procedure for assessing the behaviour
of structural steel elements under cyclic loads, Tech. Comm. 1 Structural Safety
and Loadings, TWG1.3 Rep. No. 45.
Engelhardt, M. D. and Husain, A. S. [1992] Cyclic tests on large scale steel moment
connections, Proc. 10th WCEE, World Conf. Earthq. Engrg., Madrid, Spain, July,
pp. 28852890.
Engelhardt, M. D. and Husain, A. S. [1993] Cyclic-loading performance of welded flangebolted web connections, J. Struct. Engrg. ASCE 119(12), 35373550.
00059
143
Engelhardt, M. D., Sabol, T. A., Aboutaha, R. S. and Frank, K. H. [1995] Development

of interim recommendations for improved welded moment connections in response to
the Northridge earthquake, Proc. 3rd International Workshop Connections in Steel
Structures, eds. Bjorhovde, R., Colson, A. and Zandonini, R., Trento, Italy, May,
pp. 381390.
Engelhardt, M. D. and Sabol, T. A. [1997] Seismic-resistant steel moment connections:
development since the 1994 Northridge earthquake, Prog. Struct. Engrg. Mat. 1,
6877.
Engelhardt, M. D. and Sabol, T. A. [1998] Reinforcing of steel moment connections with
cover plates: benefits and limitations, Eng. Struct. 20(46), 510520.
FEMA 267 [1995] Interim Guidelines, Rep. No. SAC-95-02, SAC Joint Venture, Sacramento, California.
JCSS [1997] Kobe earthquake damage to steel moment connections and suggested
improvement, Japanese Society of Steel Construction. JSSC Tech. Rep. No. 39,
July.
Kakaliagos, A. K. and Bouwkamp, J. G. [1993] Tests on steel and composite beam-tocolumn connections strength and stiffness aspects, Earthq. Spectra 9(4), 755780.
Krawinkler, H., Bertero, E. P. and Bertero, V. V. [1971] Inelastic behavior of steel
beam-to-column subassemblages, Earthquake Engineering Research Center, Rep.
No. EERC 71-7, University of California, Berkeley, California.
Kurobane, Y., Azuma, K. and Ogawa, K. [1997] Brittle fracture in steel building frames
Comparative study of Northridge and Kobe earthquake damage, International Institute of Welding, Annual Assembly, San Francisco, California, July.
Lin, K. C., Tsai, K. C., Kong, S. L. and Hsieh, S.-H. [2000] Effects of panel zone deformations on cyclic performance of welded moment connections, Proc. 12th WCEE,
World Conf. Earthq. Engrg., Auckland, New Zealand, January, Paper No. 1252.
Mahin, S. A., Hamburger, R. O. and Malley, J. O. [1996] An integrated program
to improve the performance of welded steel frame buildings, Proc. 11th WCEE,
World Conf. Earthq. Engrg., Acapulco, Mexico, August, Elsevier Science Ltd.,
Paper No. 1114.
Malley, J. O. [1998] SAC Steel Project: summary of Phase-1 testing investigation results,
Engrg. Struct. 20(46), 300309.
Masuda, H. and Tanaka, A. [2000] Statical characteristics of the improved rigid beamto-column connections of steel structures, Proc. 12th WCEE, World Conf. Earthq.
Engrg., Auckland, New Zealand, January, Paper No. 213/6/A.
Matsui, C. and Sakai, J. [1992] Effect of collapse modes on ductility of steel frames, Proc.
10th WCEE, World Conf. Earthq. Eng., Madrid, Spain, A.A. Balkema/Rotterdam,
pp. 29492954.
Mele, E., Calado, L. and Pucinotti, R. [1997] Indagine sperimentale sul comportamento ciclico di alcuni collegamenti in acciaio, (in Italian). Proc. 8th National Conf.
Earthq. Engrg. ANIDIS Lingegneria Sismica in Italia, Taormina, Italy, September,
pp. 10311040.
Nakashima, M., Suita, K., Morisako, K. and Maruoka, Y. [1998] Tests on welded beamcolumn subasemblies. I: global behaviour. II detailed behaviour, J. Struct. Engrg.
ASCE 124(11), 12361252.
Nakashima, M., Roeder, C. W. and Maruoka, Y. [2000] Steel moment frames for earthquakes in United States and Japan, J. Struct. Engrg. ASCE 126(8), 861868.
Plumier, A., Agatino, M. R., Castellani, A., Castiglioni, C. A. and Chesi, C. [1998] Resistance of steel connections to low-cycle fatigue, Proc. 11th ECEE, European Conf.
Earthq. Engrg., Paris, France, September, A.A. Balkema/Rotterdam.
144
00059
E. Mele
Popov, E. P. and Pinkney, R. B. [1969] Cyclic yield reversal in steel building joints, J.
of Struct. Div. ASCE, 95(ST3), 327353.
Popov, E. P. and Bertero, V. V. [1973] Cycling loading of steel beams and connections,
J. Struct. Div. ASCE 99(ST6).
Popov, E. P., Amin, N. R., Louie, J. J. C. and Stephen, R. M. [1986] Cyclic behaviour
of large beam-to-column assemblies, Engrg. J./AISC, 1st quarter, 23(1), 923.
Popov, E. P. [1988] Seismic moment connections for MRFs, J. Construct. Steel Res. 10,
163198.
Popov, E. P. and Tsai, K. C. [1989] Performance of large seismic steel moment connections
under cyclic loads, Engrg. J./AISC, 2nd quarter, 923.
Popov, E. P. and Blondet, M. [1996] Behaviour of large steel beam-column connections,
Proc. 11th WCEE, World Conf. Earthq. Engrg., Acapulco, Mexico, August, Elsevier
Science Ltd., Paper No. 166.
Popov, E. P., Blondet, M. and Stepanov, L. [1996] Application of dog bones for
improvement of seismic behaviour of steel connections, Earthquake Engineering
Research Center, Rep. No. UCB/EERC-96/05, University of California, Berkeley,
California.
Roeder, C. W. and Foutch, D. A. [1996] Experimental results for seismic resistant steel
moment frame connections, J. Struct. Engrg. ASCE 122(6), 581588.
Sabol, T. A., Engelhardt, M. D., Aboutaha, R. S. and Frank, K. H. [1996] Overview
of the AISC Northridge moment connection test program, Proc. 11th WCEE,
World Conf. Earthq. Engrg., Acapulco, Mexico, August, Elsevier Science Ltd.,
Paper No. 857.
Stojadinovic, B., Goel, S. C. and Lee, K. H. [2000a] Development of post-Northridge steel
moment connections, Proc. 12th WCEE, World Conf. Earthq. Engrg., Auckland,
New Zealand, January, Paper No. 1269.
Stojadinovic, B., Goel, S. C., Lee, K. H., Margarian, A. G. and Choi, J. H. [2000b] Parametric tests on unreinforced steel moment connections, J. Struct. Engrg. ASCE
126(1), 4049.
Tadao, N., Tetsuya, F. and Keisuke, M. [1997] At welding connections on shop assemling
recommendations for beam-end details in steel structures2, Proc. STESSA 97, 2nd
Int. Conf. Steel Struct. Seismic Areas, Kyoto, Japan, August, eds. Mazzolani, F. M.
and Akiyama, H., 632639.
Tanaka, A., Kato, B., Kaneko, H., Sakamoto, S., Takahashi, Y. and Teraoka, M. [1997]
Seismic damage of steel beam-to-column connections evaluation from statical
aspects, Proc. STESSA 1997, 2nd Int. Conf. Steel Struct. Seismic Areas, Kyoto,
Japan, August, eds. Mazzolani, F. M. and Akiyama, H., pp. 856865.
Taucer, F., Negro, P. and Colombo, A. [1998] Cyclic testing of the steel frame, JRC
ELSA Special Publication No. I.98.160, December.
Teraoka, M., Matsuo, Y., Morita, K., Sasaki, S. and Fujiwara, T. [1997] Effects of details
on structural behaviors for square tubular steel column to H-shaped steel beam with
mixed connection, in Kobe earthquake damage to steel moment connections and
suggested improvement, Japanese Society of Steel Construction. JSSC Technical
Report No. 39, July.
Tsai, K. C. and Popov, E. P. [1988] Steel beam-column joints in seismic moment resisting frames, Earthquake Engineering Research Center, Rep. UCB/EERC-88/19,
University of California, Berkeley, California.
Tsai, K. C. and Chen, C. [1996] Performance of ductile steel beam-column moment connections, Proc. 11th WCEE, World Conf. Earthq. Eng., Acapulco, Mexico, August,
Elsevier Science Ltd., Paper No. 405.
00059
145
Tsai, K. C., Wu, S. and Popov, E. P. [1995] Experimental Performance of seismic steel
beam-to-column moment joints, J. Struct. Engrg. ASCE 121(6), 925931.
Tsai, K. C. and Popov, E. P. [1995] Seismic steel beam-column moment connections, in
Metallurgy, Fracture Mechanics, Welding, Moment Connections and Frame Systems
Behavior, Rep. No. SAC/BD-95/09, SAC Joint Venture, Sacramento, California,
pp. 4-14-39.
Uang, C. M. and Bondad, D. M. [1996] Dynamic testing of full-scale steel moment connections, Proc. 11th WCEE, World Conf. Earthq. Engrg., Acapulco, Mexico, August,
Elsevier Science Ltd., Paper No. 407.
Uang, C. M., Bondad, D. and Lee, C. H. [1998] Cyclic performance of haunch repaired
steel moment connections: experimental testing and analytical modeling, Engrg.
Struct. 20(46), 552561.
Uang, C. M., Yu, Q. S. and Gilton, C. S. [2000a] Effects of loading history on cyclic
performance of steel RBS moment connections, Proc. 12th WCEE, World Conf.
Earthq. Engrg., Auckland, New Zealand, January, Paper No. 1294.
Uang, C. M., Yu, Q. S., Noel, S. and Gross, J. [2000b] Cyclic testing of steel connections
rehabilitated with RBS or welded haunch, J. Struct. Engrg. ASCE 126(1), 5768.
UBC [1991] Uniform Building Code, International Conference of Building Officials,
Whittier, California.
UBC [1994] Uniform Building Code, International Conference of Building Officials,
Whittier, California.
Usami, T., Kaneko, H., Kimura, M., Une, H. and Kushibe, A. [1997] Real scale model
tests on flange fracture behaviour of beam adjacent to beam-to-column joint and the
seismic resistance after repairing and strengthening, Proc. STESSA 1997, 2nd Int.
Conf. Steel Struct. Seismic Areas, Kyoto, Japan, August, eds. Mazzolani, F. M. and
Akiyama, H., pp. 955962.
Whittaker, A. and Bertero, V. V. [1995] Testing of steel moment-resisting frame connections, News EERC, UCB, Vol. 16, No. 3, July.
Yu, Q. S. and Uang, C. M. [2000] Cyclic performance and retrofit design of pre-Northridge
steel moment connections with welded haunch, Proc. 12th WCEE, World Conf.
Earthq. Engrg., Auckland, New Zealand, January, Paper No. 660.

JEE 2002 Vol 5 Issue 1

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

JEE 2002 Vol 5 Issue 1

Uploaded by

Copyright:

Available Formats

December 6, 2001 14:43 WSPC/124-JEE

Journal of Earthquake Engineering, Vol. 6, No. 1 (2002) 115

DAMAGE DIRECTIVITY IN BURIED PIPELINES OF

SHIRO TAKADA and NEMAT HASSANI

December 6, 2001 14:43 WSPC/124-JEE

S. Takada, N. Hassani & K. Fukuda

Burial directions of the pipes.

December 6, 2001 14:43 WSPC/124-JEE

Damage Directivity in Buried Pipelines of Kobe City

Length of buried pipes (km)

Damage ratio in water distribution

December 6, 2001 14:43 WSPC/124-JEE

S. Takada, N. Hassani & K. Fukuda

Comparison of damage ratio for different directions.

December 6, 2001 14:43 WSPC/124-JEE

Damage Directivity in Buried Pipelines of Kobe City

Comparison of interruption percentage in different areas.

December 6, 2001 14:43 WSPC/124-JEE

S. Takada, N. Hassani & K. Fukuda

3. Introduction of Function of S-Wave Radiation Pattern

Radiation pattern of S-wave.

December 6, 2001 14:43 WSPC/124-JEE

Damage Directivity in Buried Pipelines of Kobe City

Here, R is the Radiation Pattern Function of the S-wave, Vr ; rupture propagation

4. Simulated Earthquake Motions using Radiation Pattern

Parallel to the fault

December 6, 2001 14:43 WSPC/124-JEE

S. Takada, N. Hassani & K. Fukuda

By introducing the Radiation Pattern Function, it became possible to estimate

Distance from the fault

Maximum acceleration at ground

Normal to the fault

distance from the fault(km)

December 6, 2001 14:43 WSPC/124-JEE

Damage Directivity in Buried Pipelines of Kobe City

(a) Distance from the fault: 1.0 km

(b) Distance from the fault: 2.0 km

(c) Distance from the fault: 3.0 km

(d) Distance from the fault: 4.0 km

Path of a ground particle at different distances from the fault.

Distance from the fault

Distance from the fault

(a) Normal to the fault

(b) Parallel to the fault

Velocity spectra obtained from numerically generated wave.

December 6, 2001 14:43 WSPC/124-JEE

S. Takada, N. Hassani & K. Fukuda

Water guideline 90%

Water guideline 90%

Water guideline 70%

Water guideline 70%

Kansai Electric Co.

Kansai Electric Co.

(a) Normal to the fault

(b) Parallel to the fault

Velocity spectra from the waves recorded during Kobe Earthquake.

December 6, 2001 14:43 WSPC/124-JEE

Damage Directivity in Buried Pipelines of Kobe City

Parallel to the fault

(a) Distance from the fault: 1.0 km

(b) Distance from the fault: 2.0 km

Example of input ground motions.

(b) Cases 2 and 3