To appear in Proceedings of 8th International Conference on Structural Safety and Reliability

Newport Beach, CA, June 2001

Probabilistic Seismic Demand Analysis of Controlled Steel Structures

Luciana R. Barroso

Texas A&M University, College Station, Texas, USA

Scott E. Breneman

Real-Time Innovations Inc, Sunnyvale, California, USA

H. Allison Smith

Stanford University, Stanford, California, USA (retired)

ABSTRACT: The structural engineering community has been developing performance-based earthquake engineering methodologies. Structural control can provide an additional method to meet desired performance
objectives. One approach is the development of probabilistic seismic demand curves, so that the probability
that any damage measure exceeds a pre-determined allowable limit can be determined. Cornell developed a
procedure that couples conventional Probabilistic Seismic Hazard Analysis with nonlinear dynamic structural
analyses in order to determine the annual probability of exceedance of a given parameter, resulting in the
Probabilistic Seismic Demand Analysis. These concepts are then extended to the analysis of controlled structural response, referred to as Probabilistic Seismic Control Analysis. This research then evaluates the performance of various control systems under seismic excitations. It focuses on the SAC Phase II structures for
the Los Angeles region, and three types of possible controllers: (1) base isolation system; (2) linear viscous
brace dampers; (3) and active tendon braces.
1 INRODUCTION
The structural engineering community has been
making great strides in recent years to develop performance-based earthquake engineering methodologies for both new and existing construction. Both
SEAOC's Vision 2000 projects (SEAOC 1995) and
BSSC's NEHRP Guidelines for Seismic Rehabilitation of Buildings (BSSC 1997 ) present the first
guidelines for multi-level performance objectives.
Structural control can provide an additional method
to meet desired performance objectives. Design of a
structure/controller system should involve a thorough understanding of how various types of controllers enhance structural performance, such that the
most effective type of controller is selected for the
given structure and seismic hazard. Current performance-based codes provide guidelines for calculation of seismic demand for a scenario event; the
resulting seismic demand is then compared with the
allowable limit given. This procedure generally accounts for the uncertainty present in ground motion
excitation; however, the uncertainty and variability
in the demand estimation generally is not considered. One approach to addressing this issue is the
development of probabilistic seismic demand
curves, so that the probability that any damage
measure exceeds a pre-determined allowable limit
can be determined.

Procedures for developing probabilistic seismic

demand curves can be found in literature. These
methodologies generally relate a record-specific
quantity, typically the linear elastic spectral acceleration at a characteristic building frequency, to a
structural response parameter, such as interstory
drift, so as to estimate the probability of exceeding
the selected structural response parameter
This research utilizes the Probabilistic Seismic
Control Analysis to evaluate the performance of
various control systems under seismic excitations.
This study focuses on steel moment resisting frames,
specifically the SAC Phase II structures for the Los
Angeles region, and three types of possible controllers: (1) base isolation system (passive); (2) linear
viscous brace dampers (passive); (3) and active tendon braces. Simulations of these systems, both controlled and uncontrolled, are prepared using the three
suites of earthquake records, also from the SAC
Phase II project, representing three different return
periods for the area. Several controllers are developed for each structure, and their performance is
judged based on the drift, dissipated hysteretic energy, and floor acceleration demands.
The use of a probabilistic format allows for a
consideration of structural response over a range of
seismic hazards. Stable relationships can be developed between the spectral acceleration and controlled structural demands. Similar relationships are
also possible for the demands on the control system,

such as the peak bearing displacement for the isolation system. As a result, fewer control analyses may
be required to estimate the expected structural behavior. The resulting annual hazard curves can be
used to evaluate the effect of different control parameters as well as provide a basis for comparison
between different control strategies.
2 BACKGROUND
The final objective of the methodology devised by
Cornell (1996) is the estimation of the annual probability of exceedance of a given level of inelastic response in a specific MDOF structure. The structure
is located at a specific site with an associated potential seismic hazard. The response of interest can be
any structural response parameter, representing either local or global structural demands. The procedure couples conventional Probabilistic Seismic
Hazard Analysis (PSHA) with nonlinear dynamic
structural analyses in order to determine the annual
probability of exceedance of a given response parameter. This procedure is referred to as the Probabilistic Seismic Demand Analysis (PSDA). These
concepts are then extended to the analysis of controlled response, referred to as Probabilistic Seismic
Control Analysis (PSCA).
2.1 Probabilistic Seismic Hazard Analysis (PSHA)
In view of the rareness of seismic events and the
variability of their characteristics, it has become
common to describe the seismic threat in probabilistic terms. The earliest work in this area has defined
this threat in terms of some measure of the ground
motion intensity, such as peak ground acceleration
(PGA). More recent work has commonly sought to
characterize the ground motion by its response spectral acceleration Sa -- i.e., the peak acceleration the
earthquake will induce in a 1DOF system with a
specified period T and damping ratio . Commonly,
T and are chosen to describe the properties of the
first mode of the building of interest. In any case, the
aim is to describe the probability that the ground
motion characteristic of interest (PGA, Sa, spectral
velocity Sv, or spectral displacement Sd) is exceeded
over a reference period of interest.
For example, the probability that Sa exceeds a
critical value scr in an arbitrary earthquake is formally written as

P [ S a > scr ] =
m.r

P [ Sa > scr | M = m, R = r ] f M , R ( m, r ) dmdr

(1)

in which fM,R(m,r) is the joint probability density

function that describes the magnitude M and site-tosource distance R of an arbitrary event. Equation (1)

is commonly evaluated numerically, and has become

known as PSHA' or Probabilistic Seismic Hazard
Analysis (e.g., Cornell, 1968). Final results are generally rescaled by the rate of all events being modelled, leading to a mean hazard rate H ( s ) associated with those seismic events that produce Sa > scr .
As a simple parametric representation, it has been
suggested (Cornell 1996) that the net result is often
well-approximated by a relationship of the form:

H ( S a ) = k0 S a

k1

(2)

The coefficients k0 and k1 thus serve to characterize the seismic threat at a given site of interest.
The approximation in Equation (2) has the advantage of being linear in log-log space and has been
shown to be satisfactory over a range of spectral accelerations (Shome 1999)
2.2 Probabilistic Seismic Demand Analysis (PSDA)
Unfortunately, Sa does not provide sufficient information to determine the precise response of actual
buildings, which generally show both nonlinear and
multi-degree-of-freedom (MDOF) behavior. Indeed,
there is a growing trend toward performancebased seismic design, in which a range of increasingly rare hazards are specified, together with a correspondingly increasing level of permissible damage (i.e., nonlinear behavior). This implies the
need for explicit recognition, and statistical quantification, of the degree of nonlinear behavior-- e.g., the
drift demand -- as a function of the desired exceedance rate (return period).
Extension of conventional PSHA, to directly describe the seismic demand of a complex, nonlinear structure, has become known as probabilistic
seismic demand analysis (PSDA) (Cornell 1996,
Luco et. al. 1998).
The power of this method hinges on the following
observations. First, given knowledge of the ground
motion's intensity, as measured by Sa at the building's first mode, the nonlinear behavior is often
found to be not substantially influenced by additional ground motion parameters (e.g., M, R, duration). While somewhat counterintuitive, this result
has been demonstrated by detailed comparisons
(Shome 1999, Shome et. al. 1996), at least in the
context of non-degrading systems. This result permits Sa to be used as a powerful, scalar descriptor
used to summarize the ground motion threat.
Secondly, in practice, considerable variability exists in Sa, which reflects the elastic demand (of a
simplified, 1DOF building). A positive consequence
of this fact is that one need not have a very precise
description of the true nonlinear (MDOF) demand
as a function of the elastic 1DOF demand Sa. In particular, it generally suffices to have a reasonably accurate estimate of the median demand for a given

To appear in Proceedings of 8th International Conference on Structural Safety and Reliability

Newport Beach, CA, June 2001
is based should change as a function of the effective
level of Sa. Formally, we may convey this idea by
perio, Tc, and damping, c, of the controlled system.
the expression:
Computationally, the only practical difference
= ( Sa )
(3)
that arises is the need to perform sufficient analyses
of the controlled structure to predict the median dein which the random variable has unit median, and
mand as a modified power-law function as folvariability that is small compared with that of . In
lows:
view of Equation (3), one may construct a result
b
analogous to Equation (1).
= ac Sa c
(7)
The median relationship between spectral accelThe subscripts c here reflect that the parameters
eration and drift is established be performing nonare to be fit to the controlled, rather than the origilinear dynamic analyses of the model structure for
nal, structural system. One expects that the benefit
numerous ground motions at different levels of inof control is reflected, statistically, through reduced
tensity, as measured by spectral accelerations. The
values of ac that describe the original system.
spectral acceleration at the fundamental period of
the structure for each ground motion is simply obtained from its elastic response spectrum. The re3 PROBLEM DEFINITION
sponse of the model structure subjected to each
earthquake record provides the corresponding drift.
This investigation has the specific objective of
The functional relationship between drift and specevaluating the effect of the various controller architral acceleration is taken to be:
tectures on seismic demands as described through
b
= a Sa
(4)
performance-based design criteria. We believe it is
critical, in the practical evaluation of the performwhere the exponent b is included to capture potential
ance of buildings---whether controlled or not--softening of the nonlinear relationship between
under seismic threats to reflect (1) the realistic pospectral acceleration and median drift. Utilizing this
tential for nonlinear behavior and (2) the realistic
relationship, an expression for the drift hazard rate,
characteristics of ground motion excitations (i.e., by
H, can be expressed as (Cornell 1996, Barroso
imposing a suite of recorded ground motion records,
1999):
as opposed to an idealized probabilistic, random vi k1
bration description).
cr ( b )
H ( cr ) = k0
(5)
Cf
a
3.1 Structure
The last term in Equation (5), Cf, represents the
The structures analyzed are a three- and nine-story
first moment of a lognormally distributed random
steel moment-resisting frame buildings (SMRF) devariable, given by:
signed as part of the SAC steel project for the Los
2
Angeles area. These buildings conform to local code
1
ln ( )|S a
2
requirements. The structure is an office building de(6)
C f = exp k1

2 b
signed for gravity, wind, and seismic loads, with a

basic live load of 2.4 kPa (50 psf). The structural

system for all buildings consists of steel perimeter
Note that a value of 1 for Cf would ignore the effects
moment frames and interior gravity frames with
of dispersion of the structural response in the drift
shear connections. All columns in the perimeter
demand hazard. As a result, this coefficient is reframe that are part of the lateral force-resisting sysferred to as a dispersion coefficient.
tem bend about the strong axis. The North-South
frame of the 3-story structure has three fully moment
2.3 Probabilistic Seismic Control Analysis (PSCA)
resisting bays and one simply-connected bay. The
columns are fixed at the base and run the full height
This research proposes an extension of PSDA,
of the structure. The North-South frame of the 9which statistically describes the drift demand of a
story structure has four fully moment resisting bays
complex nonlinear structure, to analyze the modified
and one partially moment-resisting bay. The simple
behavior of the same structure in the presence of
shear connection in the partial moment resisting bay
supplemental control (PSCA). The presence of supoccurs as a result to avoid bi-axial bending in the
plemental control devices, whether active or passive,
corner columns. The columns are pinned at the base
often changes the effective first-mode period and/or
of the structure and spliced at every other story, with
damping of the system. As a result, the definition of
the splice location six feet above the floor. Detailed
Sa, on which the response regression in Equation (4)
descriptions of these structures, including dimen-

sions and member sizes are provided in several

works involving these structures (Barroso 1999; Gupta
1999; Krawinkler and Gupta 1998)

The structures are modeled as twodimensional frames that represent half of the structure in the north-south direction. The frame is given
half of the seismic mass of the structure at each floor
level. A basic centerline model of the bare moment
resisting frame is developed for both structures. The
strength, stiffness, and shear distortions of panel
zones is neglected. A finite element model of the
structure was developed where an assembly of interconnected elements describes the hysteretic behavior
of structural members. The inelastic behavior of the
members is taken to be concentrated at the end of
girders and beams. Thus each structural member is
constructed using a lumped plasticity model with
nonlinear rotational springs at each end joined by a
linear beam-column element. The resulting modal
properties for both the 3- and 9-story structures are
given in Table 1.
Table 1. Modal Properties of the SAC3 and SAC9 Elastic
Models
SAC3 Frame
SAC9 Frame
Damping
Period
Damping
Period
1st Mode
1.02 sec
2.0%
2.27 sec
2.0%
2nd Mode
0.33 sec
1.5%
0.85 sec
1.1%
3rd Mode
0.17 sec
2.2%
0.49 sec
1.1%

3.2 Earthquakes
Suites of ten time histories were generated by Paul
Somerville (1997) to represent ground motions having probabilities of exceedance of 50% in 50 years,
10% in 50 years, and 10% in 250 years in the Los
Angeles region. These sets of ground motions are referred to as the 50 in 50 Set, 10 in 50 Set, and 2 in
50 Set respectively. A single scaling factor was
found for each time history that minimized the
squared error between the target spectrum and the
average response spectrum of the two horizontal
components of the time history assuming lognormal
distribution of amplitudes. The weights used were
0.1, 0.3, 0.3, and 0.3 for periods of 0.3, 1, 2, and 4
seconds respectively. The scale factor was then applied to all components of the time history. An important note regarding the earthquake sets is that
they should be used only as a set, and not individually or as small sub-sets as representative of the
probability levels specified. At any particular period
the median spectral acceleration of the set may
match the target value reasonably well; however,
any individual record may have a value quite different than the expected target spectral acceleration.

3.3 Control systems

Three basic types of control systems are used: (1)
isolation systems, (2) passive damper systems, and
(3) active control systems. A representative control
system was chosen from each type listed above for
implementation. These systems are: (1) the friction
pendulum isolation system, (2) the linear fluid viscous damper system, and (3) the active tendon brace
system. The basic design and implementation of
these systems is described in the following sections,
with more detailed information found in Barroso
(1999) and Breneman (2000).
The isolation system studied is the Friction Pendulum System (FPS). The design of this system requires the selection of the isolation period, the sliding surface of the bearings, and the number and
location of the bearings. The isolation systems designed are located at the ground level, have a sliding
surface of unfilled teflon, and the direction of sliding
is parallel to the lay of the teflon. In the 9-Story
structure, so as to place the bearings at ground level,
the basement columns are cut and the bearings are
placed at their top. During the design of the isolation
systems, various isolation periods and number of
bearings are investigated to determine the impact of
these parameters. The isolation system selected has a
3 second isolation period and a teflon-steel sliding
interface with frictional properties: maximum friction coefficient = 11.93%, minimum friction coefficient = 2.66%.
Viscous dampers (VS), such as fluid cylinders,
can be designed to provide a purely viscous force to
the surrounding structure. The VS dampers studied
are assumed to be linear in nature. The brace support
for the damper is assumed to be rigid compared to
the damper, so that all deformation in the damper
system occurs through damper deformation. The
dampers are located in or adjacent to the center of
the moment resisting frame and are arranged so that
one damper runs diagonally across each story.
Damping systems were designed through the selection of the damping constant for each damper. The
system selected for comparison has equal size
dampers on all three stories of the structure and provides 30% effective damping in the first mode of the
structure.
The active tendon system was chosen because its
force application is similar to that of a passive viscous brace system, allowing for a more direct comparison between the two systems. The actuators are
located in the center of the moment-resisting frame
and are arranged so that one actuator runs diagonally
across each story in the 3-Story structure and on
first, second, and eighth stories for the 9-Story structure. Accelerometers are utilized as sensors and are
placed to measure horizontal floor accelerations at
all floors above ground level. A robust H controller
was developed for the above architecture using in-

To appear in Proceedings of 8th International Conference on Structural Safety and Reliability

Newport Beach, CA, June 2001
Annual Hazard Curve for Spectral Acceleration
terstory drifts as the regulator response quantity. The
10
design model represents the real system to the controller design optimization procedures, thus should
be a realistic as needed to characterize the behavior
of the physical system that impact the effectiveness
of the controller.
10
1

4 RESULTS
The Probabilistic Seismic Control Analysis described in Section 2 is applied to the response of
both structures. The details of the application of the
procedure are discussed for the 3-Story Structure
and the final hazard curves are shown for both systems.

Annual Hazard

10

10

0.5

1.5

S (T = 1.01 sec, = 2%) [g]

a

2.5

Figure 1. Annual Hazard Curve for Sa, 3-Story Structure

4.1 Spectral acceleration Hazard

The spectral acceleration hazard curve is obtained
by fitting a curve of the form given in Equation (2)
to the points defined by the annual probabilities of
exceedance and corresponding median spectral accelerations for the three sets of ground motions. Median values were utilized as the ground motions
were scaled so that, on average, the entire set
would correspond to the specified hazard level. Recall that, due to the addition of supplemental control
devices, different spectral acceleration hazard curve
needs to be developed for each system. Figures 1
shows the hazard curve for the uncontrolled 3-story
structure as well as the three Sa points, representing
the three ground motion sets, that were used to obtain the curve. Similar curves were developed for the
9-Story structure and the various control strategies
and the values of the parameters k0 and k1 for all systems are given in Table 2.
Table 2. Parameters for Spectral Acceleration Hazard Curve
Control
T1

Structure
k0
k1
System
(sec)
(%)
Uncontrolled 1.01
2
1.42e-3
3.25
FPS
3.00
2
1.10e-4
2.14
3-story
VS
1.01
30
2.84e-4
2.33
ATB
1.00
36.7
2.59e-4
2.42
Uncontrolled 2.27
2
2.62e-4
2.08
FPS
4.00
2
3.90e-5
1.89
9-story
VS
2.27
30
6.67e-5
2.22
ATB
2.27
35.6
2.62e-4
2.08

4.2 Relationship between ground motion and peak

story drift
A functional relationship of the form given in Equation (7) can be fit to the data by: 1) using the median
values for Sa and max for each hazard set, or 2) utilizing the full data set of Sa and max points. A linear
regression of the log of the data selected was performed to establish the parameters ac and bc.
Regardless of whether the full data set or only the
median values were used in the regression, the
dispersion of max given Sa must be calculated as the
square root of the mean squared deviation of the full
data set from the fitted curve. Plots of Sa vs. max
over the height of the structure from nonlinear structural analyses are presented in Figures 2 through 5.
The fit of the regression curve for the 3-story uncontrolled structure, shown in Figure 2 displays a
significant amount of scatter to the fitted curve.
Most of the contribution to this scatter comes from
the results of the 2 in 50 set of ground motions. The
resulting parameters for the regression fit to all the
records are listed in Table 3. When fitting the curve
through the median values, the resulting exponential
coefficient, b, is only slightly larger than 1, indicating a slight softening of the system. However, when
the curve is performed using the entire data set, the
exponent b is significantly smaller. This trend
proves to be consistent for all records analyzed. One
possible explanation lies in the scaling procedure
used to the ground motions. As discussed previously, the earthquake records were scaled to match
specified spectral acceleration values at four specific
period values. As a result, a record with a higher
than average spectral acceleration value at the period
of interest to the structure is associated, by design,
with spectral acceleration values that are systematically lower than average at some if the other periods.
As nonlinearities occur, the period of the structure

Relationship Between Spectral Acceleration and Maximum Story Drift Angle

LA 3Story Structure: VS 30 D1
0.06

0.04

0.03

0.02

0.01

0
0

0.4

0.6

0.8

Sa (T1 = 1.01 sec, = 30%) [g]

1.2

1.4

1.6

Relationship Between Spectral Acceleration and Maximum Story Drift Angle

LA 3Story Structure: ATB S1k
0.14

LA 3Story Structure: Viscous Dampers, = 30%

0.12

0.12
50 in 50
10 in 50
2 in 50

50 in 50
10 in 50
2 in 50

Drift Angle

0.1

0.08

Drift Angle

0.2

Figure 4. Plot of Sa and max for 3-Story Structure with VS

Relationship Between Spectral Acceleration and Maximum Story Drift Angle

0.1

50 in 50
10 in 50
2 in 50

0.05

Drift Angle

moves away from the fundamental period and into

the range where spectral acceleration is lower than
average. So this record will produce lower drifts
than the typical records. A reverse effect occurs
with records with spectral accelerations lower than
average at the period of interest. The net result is a
flattening of the fitted curve and lower b values
than might be expected from a random record.
Another concern with this structure is the results
due to the inclusion of simulated ground motions.
Previous researchers (Luco 1998) have found that
for structures of 1 second the results can be misleading when correlating spectral acceleration and drift.
This effect has been attributed to the methodology
used in generating the synthetic ground motions. Results excluding synthetic ground motions are also
presented in Table 3. Note that though dispersion is
lower when excluding synthetic records, the resulting curve fit is close to the original.

0.08

0.06

0.06
0.04
0.04
0.02
0.02
0
0

0.5

0
0

0.5

1.5

S (T = 1.01 sec, = 2%) [g]

a

2.5

1.5

S (T = 1.01 sec, = 2%) [g]

2.5

Figure 5. Plot of Sa and max for 3-Story Structure with ATB

Figure 2. Plot of Sa and max for 3-Story Uncontrolled Structure

Relationship Between Spectral Acceleration and Maximum Story Drift Angle

LA 3Story Structure: FPS Isolation, T3 f1
0.06

0.05

50 in 50
10 in 50
2 in 50

Table 3. Parameters for Fit of Relationship between Sa and max

for 3-Story Uncontrolled Structure
Fitted
Records
a
b
ln()|Sa
Data
Median
0.0291
1.099
0.383
All
All
0.0288
0.842
0.346
No
Median
0.0283
0.984
0.251
Simulated
All
0.0267
0.789
0.219

Drift Angle

0.04

0.03

0.02

0.01

0
0

0.5

1.5

S (T = 1.01 sec, = 2%) [g]

a

2.5

Figure 3. Plot of Sa and max for 3-Story Structure with FPS

The relationship between ground motion and

peak story drift is shown in Figure 3. As this system
alternates between a fixed base response, with a period of 1 second, and a sliding system, with a period
of 3 seconds, analyses were performed utilizing both
conditions. The resulting expression utilizing median drift values for each data set is poor, especially
at high Sa values. Attempts at improving the fit utilizing all data points or correlating to another ground
motion parameter had only minimal effects. The
analyses were then performed neglecting the synthetic ground motions, resulting in significantly reduced scatter in the response. Also, these results indicate that the structural demand is more closely

To appear in Proceedings of 8th International Conference on Structural Safety and Reliability

Newport Beach, CA, June 2001
associated with the spectral response at the fixed
most effective at all target drift limits. The drift hazbase period of 1 second.
ard for the VS damper system and ATB system are
very similar over the drift range considered. The VS
Table 4. Parameters for Fit of Relationship between Sa and max
system has a smaller drift hazard over this range;
for 3-Story Structure with FPS
however, the two curves converge at the high end.
Fitted
The FPS isolation drift hazard curve crosses both
Records
Period
a
b
ln()|Sa
Data
other controlled drift hazard curves, exhibiting
Median 1.3e-2
0.754
0.377
1.0 sec
higher drift hazard at low return periods and lower
All
1.2e-2
0.525
0.348
All
drift hazard at high return periods. This condition reMedian 2.2e-2
0.484
0.377
3.0 sec
All
2.0e-2
0.380
0.291
flects the fact that frictional isolation systems only
Median 1.2e-2
0.567
0.216
impact the structural response if the forces through
1.0 sec
All
1.1e-2
0.432
0.199
No
the bearing are large enough to overcome the fricSimulated
Median 1.8e-2e
0.378
0.324
3.0 sec
tional slip force for the bearing. The other factor afAll
1.4e-2
0.205
0.296
fecting the performance of the FPS system with respect to the two other control systems is the scatter
In contrast with the previous systems, the relain the structural response for a given spectral acceltionship between Sa and max for both the VS and
eration.
ATB systems is very strong, as seen in Figures 4 and
Similar curves were developed for the control
5 respectively. The resulting curves are nearly linstrategies
designed for the 9-Story structure, shown
ear, as indicated by a value of b being nearly unity.
in
Figure
7. Again, though all three controlOne of the advantages of this result is that only a
strategies reduce the drift hazard, no one strategy is
small number of nonlinear dynamic analyses should
the most effective at all return periods or target drift
be required to determine the functional relationship
levels. The ATB system is consistently less effective
between Sa and max for these systems. Also, these
than the two passive control system. However, recall
systems are insensitive to the inclusion of the simuthat this system only utilizes 3 braces located at the
lated time histories in the data analysis. The specific
first, second, and eighth stories, and as such has a
parameters for both systems are given in Table 5.
much smaller capacity than the VS system plotted
here. The drift hazard curves for the FPS isolation
Table 5. Parameters for Fit of Relationship between Sa and max
VS systems cross at a drift limit of about 3%. The
for 3-Story Structure with VS and ATB
Fitted
VS system has a lower drift hazard at the low return
Control
Record
a
b
ln()|Sa
Data
periods, while the FPS isolation has a lower drift
Median 2.97e-2
1.10
0.132
hazard at higher return periods.
All
No
Simulated
All
ATB

No
Simulated

All
Median
All
Median
All
Median
All

2.84e-2
2.74e-2
2.63e-2
3.31e-2
3.30e-2
2.99e-2
2.98e-2

1.04
1.04
0.996
1.02
1.01
0.944
0.943

0.128
0.0978
0.0935
0.287
0.287
0.280
0.280

4.3 Drift Demand Hazard Curves

Once the median relationship between spectral acceleration and drift (i.e., the median drift given spectral acceleration), and the dispersion of drift given
spectral acceleration are known, the spectral acceleration hazard curve can be used to create a drift
demand hazard curve using Equation (5). Median
values are used as the ground motion sets were developed to be used as a unit, so that the seismic demand for a structure for a given hazard is best approximated by the median response. The resulting
drift demand curves are then used to provide a basis
for comparison between systems.
The curves showing the maximum interstory drift
hazard for each of the control strategies for the 3Story structure are plotted in Figure 6. All three control strategies result in significant reductions in drift
hazard. However, no single strategy proves to be

10

Annual Hazard Curves for Maximum Peak Story Drift Angle

Uncontrolled

Annual Probability of Exceedance [ P( > cr) ]

VS

10

10

10

10

10

FPS T3, f1
Viscous 30%, D1
ATB S1k

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

0.05

Maximum Story Drift Ratio

Figure 6. Comparison of Drift Demand Hazard Curves for

Maximum Peak Drift for 3-Story Structure

the lowest drift demands. However, at higher return

periods, the FPS isolation system becomes the most
effective strategy.

Annual Hazard Curves for Maximum Peak Story Drift Angle

LA 9Story Structure

10

Uncontrolled
FPS T4, f1

Annual Probability[ P( > ) ]

Viscous 30%, D1
ATB S1k

cr

10

REFERENCES

10

10

10

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

0.05

Maximum Peak Story Drift Ratio

Figure 7: Comparison of Drift Demand Hazard Curves for

Maximum Peak Drift for 9-Story Structure

5 CONCLUSION
Current performance guidelines provide for the determination of seismic demands for a given scenario
event, which is then compared with an allowable
limit. These procedures do not account for the variability in seismic demand estimation. One approach
to address this issue is the development of probabilistic seismic demand curves. This study uses the
Probabilistic Seismic Demand Analysis methodology developed by Cornell (1996) that combines information regarding the seismic hazard and nonlinear structural response. The procedure relates a
record specific quantity, such as the elastic spectral
acceleration, to a structural response parameter, such
as drift, to estimate the probability of exceeding the
selected structural parameter.
The elastic spectral acceleration at the fundamental period of the structure is a measure of ground
motion intensity that is structure specific. However,
the addition of supplemental control devices changes
the fundamental properties of the structural system.
Therefore, the determination of spectral acceleration
hazard curves and relationships between record
quantities and structure response need to be determined with respect to the new system. Relationships
between spectral acceleration and interstory drift are
developed for all systems. For the uncontrolled
structure, a significant amount of scatter is present,
especially due to responses from the 2 in 50 set of
ground motions. The addition of control reduces the
scatter of the response in all 3 cases.
Once this information is combined with the spectral hazard, the resulting demand hazard curves can
be used to compare the effects of different parameters for a single control type as well as the performance between the different control strategies. Results
from these curves indicate that no single control
strategy is the most effective at all hazard levels. For
example, at low return periods the VS system has

Barroso, L. 1999. Performance Evaluation of Vibration Controlled Steel Structures Under Seismic Loading. Ph.D. Thesis. Department of Civil & Environmental Engineering.
Stanford University.
Barroso, L.R., S.E. Breneman and H.A. Smith. 1998. Evaluating the Effectiveness of Structural Control within the Context of Performance-Based Engineering. Proc. of the 6th
U.S. National Conference on Earthquake Engineering, Seattle, Washington
Breneman, S.E. 2000. Practical Design Issues in the Design of
Active Control for Civil Engineering Structures. Ph.D. Thesis, Dept. Civil & Environmental Engineering. Stanford
University.
BSSC. 1997. NEHRP Guidelines for the Seismic Regulation of
Exhisting Buildings and Other Structures. Technical Report
FEMA 273. Building Seismic Safety Council, U.S.A.
Cornell, C. A. (1968). Engineering Seismic Risk Analysis.
Bulletin of Seismological Society of America 58(5), 15831606.
Cornell, C. A. 1996. Calculating Building Seismic Performance Reliability; A Basis for Multi-level Design Norms."
Proc. of 11th World Conference on Earthquake Engineering, Acapulco, Mexico.
Gupta, A. (1999). Seismic Demands for Performance Evaluation of Steel Moment Resisting Frame Structures. Ph. D. for
Civil and Environmental Engineering, Stanford University.
Krawinkler, H.K. and A. Gupta. 1998. Story Drift Demands for
Steel Moment Frame Structures in Different Seismic Regions. Proc. of the 6th U.S. National Conference on Earthquake Engineering, Seattle, Washington.
Luco, N. and C. A. Cornell. 1998. Effects of Random Connection Fractures on the Demands and Reliability for a 3-Story
Pre-Northridge SMRF Structure. Proc. 6th U.S. National
Conference in Earthquake Engineering, Seattle, Washington.
SEAOC. 1995. Vision 2000 - A Framework for Performance
Based Design, Volumes I, II, III. Technical report, Structural Engineers Association of California. Vision 2000
Committee, Sacramento, California.
Shome, N. and Cornell, C.A. 1999. Probabilistic Seismic Demand Analysis of Nonlinear Structures. Report No. RMS-35 (Ph.D. Thesis), RMS Program, Stanford University.
Shome, N, Cornell, C.A., Bazzurro, P., and Carballo, J.E.
1997. Earthquakes, Records and Nonlinear MDOF Records. Report No. RMS--29, RMS Program, Stanford University.
Somerville, P., D. Anderson, J. Sun, S. Punyamurthula, and N.
Smith. 1998. Generation of Ground Motion Time Histories
for Performance-Based Seismic Engineering. Proc. 6th U.S.
National Conference in Earthquake Engineering, Seattle,
Washington.