Professional Documents
Culture Documents
Eerc 05 01
Eerc 05 01
Eerc 05 01
By
Keri L. Ryan
Anil K. Chopra
COLLEGE OF ENGINEERING
By
Keri L. Ryan
Department of Civil and Environmental Engineering
Utah State University
Logan, UT
and
Anil K. Chopra
Earthquake Engineering Research Center
University of California, Berkeley
Although base-isolated buildings are designed using dynamic analysis, current code provi-
sions require that design values of the earthquake-induced lateral deformations and forces do not
fall below minimum values determined by a simpler procedure. Isolation systems are strongly
nonlinear, but this procedure estimates the isolator deformation by approximate methods based
on equivalent-linear systems.
isolation system, based on rigorous nonlinear analysis, is developed in Chapter 3. The procedure
offers an alternative to the iterative equivalent-linear methods used by current U.S. building
codes. The governing equation is reduced to a form such that the median normalized deforma-
tion of the system due to an ensemble of ground motions with given corner period Td is found
to depend on only two parameters: the natural isolation period, defined from the post-yield
stiffness, and the normalized strength, or strength normalized by peak ground velocity. The
implying that the median normalized deformation is a meaningful estimate of response. The
simple trends shown by the median normalized deformation led to the development of suitable
design equations for isolator deformation. These design equations reflect a 13% increase when
the excitation is biaxial (two lateral components of ground motion) compared to single com-
are unconservative by up to 50% compared to those found from the more accurate nonlinear
spectrum, and building codes include at most a 4.4% increase for a second component.
Subsequent chapters extend the procedure to more complex models of the isolated build-
ing. Chapter 4 extends the procedure to asymmetric-plan systems, deriving equations to es-
timate the peak deformation among all isolators in an asymmetric building. A variety of
ity is introduced by varying the stiffnesses and strengths of individual isolators. The idealized
system is shown to approximate the peak deformation in asymmetric-plan systems with less
than 1% error.
i
In addition to the isolation period and normalized strength, the median normalized de-
the normalized stiffness eccentricity. However, the influence of each of these parameters, ex-
cept for eccentricity, on the deformation ratio – the ratio of peak deformations in asymmetric
developed to estimate the largest deformation ratio among all isolators depends only on the
stiffness eccentricity and the distance from the center of mass to the outlying isolator. This
equation, multiplied by the design equation for symmetric systems in Chapter 3, gives the peak
deformation of asymmetric systems. This design equation conservatively estimates the peak
deformation among all isolators, but is generally within 10% of the ‘exact’ value.
In Chapter 5, the procedure is applied to buildings isolated with the friction pendulum
system. Friction pendulum (FP) isolators can also be modeled by a nonlinear force-displacement
relation, but with yield displacement on the order of 0.05 cm compared to 1 cm for rubber bear-
ings. Design equations for the peak slip, or displacement, in FP isolators reflect the significant
– 20 to 38% – increase when the excitation is biaxial. Equivalent-linear methods are shown to
history analysis for one component of ground motion, and building codes include at most a
Discussed in Chapter 6, existing models for isolation bearings neglect certain aspects
of their response behavior. For instance, rubber bearings have been observed to decrease in
stiffness with increasing axial load, as well as soften in the vertical direction at large lateral
deformations. The yield strength of lead-rubber bearings has also been observed to vary with
axial load, such that a lightly loaded bearing may not achieve its theoretical strength.
A series of bearing models are developed to include these observed behaviors, referred to
as “axial-load effects”. The models are considered to be most accurate for lead-rubber bear-
ings. Extending an existing two-spring model (shear spring plus rotational springs) developed
from linear stability theory of multi-layer bearings, the constant-strength model is achieved
by incorporating a nonlinear constitutive model for the shear spring. Numerically, this model
ii
is implemented by solving its equilibrium and kinematic equations (a system of five nonlinear
equations) by Newton’s method for the bearing forces, and taking differentials of these equations
to derive the instantaneous bearing stiffness matrix. An empirical equation is developed that
can be calibrated to match the experimentally observed varying yield strength in lead-rubber
bearings; this effect is included in the variable-strength model. The response behavior of these
Not considered in previous chapters, the peak axial forces in individual isolators need
also be estimated for design due to code testing requirements. For this purpose, the isolated
structure model is modified in Chapter 7 to include rocking about one axis and incorporate the
improved bearing models of Chapter 6. However, rocking of the structure and bearing axial-
load effects are found to have little influence on the peak lateral bearing deformation, whereas
even if rocking is neglected entirely, median response spectra are within, perhaps, 10% of those
when rocking and axial-load effects are included. Furthermore, bearing axial-load effects can
usually be neglected in determining the maximum and minimum bearing axial forces. Based
on analyses that indicate more than 10% error in neglecting such effects, the variable strength
model is recommended only when the normalized strength exceeds 0.75 and the rocking-to-
The design equation for peak lateral deformation is updated to include the slight influence
of rocking, and a design equation for peak axial force is developed for the first time. It is not
surprising that the peak lateral deformation follows trends from earlier chapters and does not
depend on any additional parameters. The design equation for axial force depends on the
isolation period, normalized strength, and rocking-to-vertical frequency ratio. It is shown how
to use this design equation to predict and subsequently eliminate bearing tension early in
the design process. Decreasing the normalized strength of the system is the simplest way to
eliminate tension, but at the expense of larger lateral deformation in the bearings. Since the
peak axial forces do not vary with the vertical-to-lateral frequency ratio, the use of high shape
factor bearings does not provide the expected benefit of avoiding instability.
iii
analysis of the response of the isolated block. The concept of accidental torsion due to axial-
load effects is introduced, where variation of the axial forces and hence stiffnesses and strengths
the isolation system from this source is found to be insignificant. Design equations are updated
to estimate the peak bearing deformation and axial forces in both nominally symmetric and
asymmetric-plan isolation systems due to biaxial excitation. Determined from these design
equations, the peak deformation in a symmetric system is again found to increase by 13%
when the excitation is biaxial, and the further increase in deformation for asymmetric-plan
systems again depends on the eccentricity and the distance from the center of mass to an
outlying corner. Additional parameters that relate to the distance from the center of mass to
the outermost edge in each rocking direction are applied to the axial force design equation for
symmetric-plan systems. These parameters provide only marginal improvement to the axial
force estimates in symmetric systems, but significant improvement to the force estimates in
Although not obviously applicable when rocking of the system is included, the earlier
is justified since the design equations are as accurate for general asymmetric-plan systems as
for the idealized systems for which they were developed. However, these design equations are
overall less effective than in previous chapters, perhaps indicating the limit of application of the
simplified procedure. While equations to estimate deformations are simple and accurate, those
estimating axial forces are complicated and can err on the order of -25 to 25% for asymmetric-
plan systems.
iv
ACKNOWLEDGEMENTS
This research is funded by National Science Foundation Grants CMS-9812414 and CMS-
9812531 and a Graduate Student Fellowship to the first author, as well as a California state
legislative grant. We gratefully acknowledge this support. Except for editorial changes, this
report is the same as the doctoral dissertation of Keri Lynn Ryan at the University of California,
Berkeley. We especially thank Professor Jim Kelly for his guidance throughout the project. We
also wish to thank Professor Nicos Makris for his guidance in defining the scope of the project
v
TABLE OF CONTENTS
1 INTRODUCTION 1
SIS 8
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
vi
3.9 Comparison to Code Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.10 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3A.1 R2 statistic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
SYSTEM 46
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
vii
4.5.2 Equations to Estimate Corner Deformation . . . . . . . . . . . . . . . . . 67
4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
LUM ISOLATOR 86
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
viii
6.2.1 Approximate Force-Deformation Relation Based on Linear Two-Spring
Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
ix
7.2.3 Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
ING 187
x
8.3.2 Uncoupled Nonlinear Bearing Model . . . . . . . . . . . . . . . . . . . . . 196
References 225
xi
1: INTRODUCTION
Although base-isolated buildings are designed using dynamic analysis, the current Inter-
national Building Code (IBC) provisions require that design values of the earthquake-induced
lateral deformations and forces do not fall below minimum values determined by a simpler
procedure [1, 2]. This procedure assumes that the response is dominated by the first mode,
and estimates the isolator deformation from a design spectrum by approximate methods based
on equivalent-linear systems. Isolation systems are strongly nonlinear, however, due to added
energy dissipation devices, such as high-damping fillers, lead cores, and friction devices [2, 3].
Therefore, use of the simple procedure requires defining an equivalent-linear system with period
and damping ratio based on the nonlinear force-deformation relation of the isolator. Because
the period and damping ratio of the equivalent-linear system depend on the absolute maximum,
The two horizontal components of ground motion acting simultaneously should be con-
sidered to determine the peak deformation of the isolation system that can occur in any lateral
direction. The IBC is ambiguous, making no mention of bidirectional excitation in its static
lateral response procedure. The usual 100 + 30 rule – the response to 100% of the ground
motion applied in the most critical direction and 30% applied in the transverse direction –
is used for response spectrum analysis, resulting in at most a 4.4% deformation increase for
bidirectional excitation.
buildings due to torsion. To account for this additional deformation, the IBC increases the
system. Again, the justification for this increase is questionable since isolation systems are
strongly nonlinear.
the IBC, leave room for improvement. A rational approach to estimate isolator deformations
should be developed that accounts for the nonlinear force-deformation relation of the isolators
1
Initially, rocking or overturning of the structure was not a concern for designing isolation
systems, which were traditionally implemented in short, squat structures. However, recently
isolation has been extended to taller buildings, such as the 32-story LA City Hall, the 18-story
A number of approaches have been taken to accommodate the time-varying axial forces
in an isolation system. Due to uncertainty whether isolation bearings and sliding isolation
systems could sustain tension, unrestrained uplift of the structure was allowed in early designs.
Concern about this uplift led researchers to experiment with various restraint systems, which
both limited uplift of the structure and provided tensile resistance through an alternate load
path in the isolation system [6, 7, 8]. It was also proposed that uplift in sliding isolation systems
be accommodated at the foundation level rather than the isolation level [9].
Even with eventual evidence that elastomeric bearings can sustain moderate tensile forces
[10, 11], designers still wish to minimize or avoid tension altogether. Sometimes extreme mea-
sures are taken to meet this goal, such as the “loose-bolt” connection used in the designs of LA
City Hall (retrofit) [12], San Bernadino Medical Center and LAC/USC Medical Center. This
questionable connection detail delays engagement of the bearing in tension by allowing a small
amount of unrestrained uplift, but forces the bolt to transfer shear while acting as a cantilever
element [13].
Underlying the desire to eliminate tension in isolation bearings is the following: First,
the tensile force that can be withstood by elastomeric bearings without excessive damage is
unknown. Second, for lead-rubber (LRB) bearings, the absence of confining pressure when the
bearing is in tension appears to compromise the lateral resistance and diminish the energy dis-
sipation that is relied on to control lateral deformation. Third, satisfying the IBC requirement
that the bearings be tested laterally under the maximum and minimum axial loads predicted
by dynamic analysis is difficult since most testing machines cannot apply simultaneous tension
and shear [2]. Thus, accurate determination of overturning forces on the isolators is essential
To further complicate this matter, the behavior of isolation bearings appears to change
2
considerably when the variation of axial forces on the bearings is taken into account [13, 14,
15, 16, 17, 18]. Existing nonlinear bearing models do not account for the influence of axial
forces on the response of the bearing, nor allow for its consideration in dynamic analysis of the
system. If the behavior of the isolation bearings depends on axial force, among other things,
the time-variation of the strengths and stiffnesses of individual bearings can induce torsion in
The overall objective of this dissertation is to develop a procedure suitable for estimating
the peak lateral deformation and axial forces – needed for preliminary design of the isolation
system – for a given design spectrum. Although the structure is treated as rigid, an advanced
model of the isolation system is used to achieve accuracy. This includes characterization of
each individual isolator by a new nonlinear model that incorporates the effects of axial force
on the bearing response. Design equations to estimate the responses of interest are developed
motions representative of the spectrum. The equations of motion are normalized such that the
deformation response of the system is insensitive to ground motion intensity, thereby minimizing
its statistical variation over the ground motion ensemble. This makes the design equations more
Chapter 2 presents the ground motion ensemble and statistical analysis equations that will
LRB bearings, the normalization procedure is first developed in Chapter 3 for an isolated block
The equivalent-linear procedure used by the IBC is shown, for a given design spectrum, to be
generally unconservative in estimating the peak lateral deformation. Chapter 4 extends the
lateral deformation due to plan-asymmetry can be accounted for by a simple factor. This factor
that minimize the dependence of the peak response on the plan layout and individual bearing
3
systems by changing the yield deformation in the force-deformation relationship.
A suitable model to account for the interaction between lateral response and axial force
for dynamic analysis. Using the new bearing model, the isolated structure is modified to include
rocking about one axis in Chapter 7. Design equations for the lateral deformation are updated
for this planar rocking analysis, and design equations to estimate the maximum and minimum
axial forces are developed for the first time. In Chapter 8, the lateral-rocking analysis of
(3D) analysis of the response of the isolated block. Design equations are extended to estimate
the peak bearing deformation and axial forces in both nominally symmetric – leading to the
concept of accidental torsion due to axial-load effects – and asymmetric-plan isolation systems.
Estimating the axial forces in asymmetric-plan buildings proves to be difficult, perhaps signaling
4
2: GROUND MOTIONS AND STATISTICAL ANALYSIS
An ensemble of 20 ground motions has been selected for this study, recorded from four California
earthquakes: (1) 1971 San Fernando (Mw = 6.6), (2) 1987 Superstition Hills (Mw = 6.7),
(3) 1989 Loma Prieta (Mw = 6.9), and (4) 1994 Northridge (Mw = 6.7). These motions,
obtained from PEER Strong Motion Database [19], are listed in Table 2.1. The data includes
the recording site, the earthquake, closest distance of the site to fault rupture, and peak ground
acceleration (ügo ), velocity (u̇go ), and displacement (ugo ) for the two horizontal components
of motion. For each motion, the component having the larger peak ground velocity (PGV) is
defined as the “strong” component, and the other as the “weak” component. Designated the
Large Magnitude Small Distance (LMSR) ensemble, these motions are representative of ground
shaking from a large magnitude earthquake at a site near the fault rupture. Recorded on firm
soil (USGS site class C) in the far-field of these earthquakes, they are broad frequency band
Figure 2.1 shows the median linear response spectra for both the strong and weak-
component ensembles as four-way log plots. In constructing these spectra, each stronger com-
ponent of motion was scaled to a PGV of 35 cm/s, and the same scale factor was applied to the
weaker component; which amplifies the median velocities of 29.4 and 20.9 cm/s before scaling to
35 and 24.9 cm/s for the strong and weak-component ensembles, respectively. The spectra have
region Tc < Tn < Td , and displacement-sensitive region Tn > Td [21, Section 6.8]; Fig. 2.1 also
shows the idealized spectra upon which these spectral regions are based.
By the procedures of the following section, the dynamic response of an isolation system to
each motion in the ensemble was determined and from this the median value x̂ and dispersion
measure δ were computed. For n observed values xi , the median, defined as the geometric
5
Table 2.1: Characteristics of excitations in the LMSR ground motion ensemble.
mean, is
n
i=1 ln xi
x̂ = exp (2.1)
n
and the dispersion measure is
n 1/2
− ln x̂)2
i=1 (ln xi
δ= (2.2)
n−1
For small values, δ is close to the coefficient of variation of x. The median plus one standard
by
These definitions assume the data is sampled from a lognormal distribution, which has been
6
(a) Strong-Component Ensemble (b) Weak-Component Ensemble
Acc Sensitive Vel Sens Disp Sensitive Acc Sensitive Vel Sens Disp Sensitive
2 2
)
)
10 10
m
m
10
10
(c
(c
A
A
D
D
(g
(g
10
10
)
)
0
0
1
1
V (cm/s)
V (cm/s)
1 1
10
10
10 10
1
1
0.
0.
01
01
1
1
0.
0.
0 0
0.
0.
10 10
Td =2.06 s
Td =2.35 s
1
1
Tc =.45 s
Tc =.39 s
−1 −1
10 10
−2 −1 0 1 2 −2 −1 0 1 2
10 10 10 10 10 10 10 10 10 10
Tn (sec) Tn (sec)
Figure 2.1: Median linear response spectra for 5% damping (a) strong-component ensemble,
median ügo = 0.324 g, u̇go = 35 cm/s and ugo = 10.2 cm, and (b) weak-component ensemble,
median ügo = 0.274 g, u̇go = 24.9 cm/s and ugo = 7.6 cm. Also shown are idealized spectra
with spectral regions.
7
3: ESTIMATING ISOLATOR DEMANDS BASED ON NONLINEAR
ANALYSIS
3.1 Introduction
Preliminary analysis is an important step in the design of a base-isolated building. How, then,
given the nonlinear behavior typical of most isolation systems, to estimate the peak lateral
deformation of the isolators based on a specific design ground motion or design spectrum? This
deformation estimate could be based on nonlinear analysis due to a single ground motion but
would be sensitive to the particular motion selected, unlike deformation determined from a
design spectrum. A better estimate could be obtained from nonlinear analyses to an ensemble
of ground motions that represent the design spectrum, but a preliminary design procedure that
requires multiple nonlinear analyses is impractical. An additional difficulty is that the response
of a nonlinear isolation system is much more sensitive to ground motion intensity than a linear
system; thus the ground motions in an ensemble must be of similar intensity for the analyses
to be meaningful.
Unlike building frames that resist lateral forces primarily in one direction, the peak de-
excitation – controls the design of the isolation system. Furthermore, isolator nonlinearity
causes bidirectional interaction consistent with the assumption of a circular interaction surface,
as shown by extensive testing of both high damping and lead-rubber bridge bearings [22]. An-
alytical models that included this interaction usually estimated the measured response of an
isolated block shaken by various ground motions more accurately than those that neglected it
The objective of this chapter is to develop a procedure suitable for estimating isolator
deformations for a given design spectrum. The procedure developed is based on rigorous non-
effective because the governing equation for the system is rewritten such that its normalized
8
fb
(a) (b)
Q kb
m ub (t) ki
ub
kb , Q
üg (t)
Figure 3.1: (a) Single isolator supporting a rigid mass, and (b) force-deformation relation for
a bilinear isolator.
deformation is insensitive to ground motion intensity, and statistical variation of the normalized
oped for a mass-isolator system subjected to one component of excitation, is then extended for
and forces is evaluated against the nonlinear procedure developed in this chapter.
The system analyzed is idealized as a rigid mass mounted on a single isolator (Fig. 3.1). The
mass m represents the total mass above the isolator, which includes the structure mass and the
base mass. The isolator has bilinear force-deformation, parameterized by the postyield stiffness
Q
uy = (3.1)
kI − kb
In a lead-rubber bearing, which will be the focus here, kb is the stiffness of rubber and Q is the
strength of the lead core. The force in the isolator is determined from the force-deformation
where ub is the isolator deformation and z represents the fraction of the yield strength applied.
This function z, which depends on the initial stiffness, deformation, and velocity; equals ±1 on
9
300 20
(a) (b)
150 10
üg (cm/s2 )
ub (cm)
0 0
−150 −10
−300 −20
0 10 20 30 40 0 10 20 30 40
t (sec) t (sec)
2
(c) 0.2 (d)
1
0.1
fb /w
0 0
z
−1 −0.1
−0.2
−2
0 10 20 30 40 −20 −10 0 10 20
t (sec) ub (cm)
Figure 3.2: (a) Strong component of ground motion for LMSR Record No. 11, (b) isola-
tor deformation, (c) yield function z(t), and (d) isolator force-deformation relation. System
parameters are Tb = 2.5 seconds and µ = 0.07.
the upper and lower bounding surfaces – represented by dashed lines in Fig. 3.1b – and varies
The deformation of the isolator that supports mass m (or weight w), subjected to ground
where ωb = kb /m and µ = Q/w. Equation 3.3 is solved for a single ground acceleration (see
Fig. 3.2a) to obtain the deformation history ub (t) (Fig. 3.2b). Also shown is the yielding history
through z(t) (Fig. 3.2c) and the variation of the isolator force coefficient fb /w with deformation
(Fig. 3.2d).
The parameters ωb and µ in Eq. 3.3 characterize the nonlinear system. Because the
system vibrates mostly at its postyield stiffness (see Fig. 3.2d and the time durations when
to characterize the isolation system; ωb and Tb are generally known as the isolation frequency
10
and period, respectively. Known as the characteristic strength ratio, µ quantifies the strength
of the system relative to the structure weight w, and target ranges of µ provide a basis for
designing the yield strength of the isolation system. Equation 3.3 indicates that besides the
isolation period and the strength ratio, the deformation depends on the initial stiffness kI and
is sensitive to ground motion intensity. We will show how to reduce the number of parameters
Unlike the postyield stiffness, which is associated with a physical property of the isolator, the
initial stiffness is open to interpretation, because the transition from initial to yielded state
in actual bearing tests is gradual. Naeim and Kelly [2, pg106] suggest that initial stiffness be
modeled as a fixed multiple of the postyield stiffness. To demonstrate the effect of fixing kI /kb
to a constant, peak deformation is plotted against the isolation period Tb for various strength
coefficients µ in Fig. 3.3a. This figure shows that an increase in yield strength leads to an
increase in deformation at long periods, which is contrary to the expectation that the lead core
Using a different model for initial stiffness, Makris and Chang [23] concluded that the
response is insensitive to initial stiffness. With this model, the yield deformation – rather than
the ratio of the initial to postyield stiffness – is fixed; and the initial stiffness depends on the
yield strength instead of the postyield stiffness. For yield deformation fixed at 1 cm with all
other parameters remaining the same, the peak deformation tends to decrease as yield strength
increases (Fig. 3.3b), consistent with expectations. Although the response may not be sensitive
to small changes in initial stiffness, Fig. 3.3 shows that it is sensitive to the large discrepancies
in initial stiffness resulting from these two disparate models. Thus, selecting a model that best
If the lead core is press fit into a hole that is slightly too small, causing it to extrude into
the rubber and lock with the steel plates, it deforms primarily in shear throughout its volume
11
30
(a) (b)
25
20
ubo (cm)
15
10 µ=.04
.06
.08
5 .10
.12
0
0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec)
Figure 3.3: Peak deformation ubo vs Tb for different values of µ for (a) constant kI /kb = 10
and (b) constant uy = 1 cm; excitation is the strong component of LMSR Record No. 8.
when subjected to load [3]. Consider the following model for this behavior: The initial stiffness
rubber. These stiffnesses depend on their respective shear moduli Gl and G, areas Al and A,
and total thickness tr of rubber in the bearing, according to the equations kl = Gl Al /tr and
kb = GA/tr . The yield strength is Q = Al σyl , where σyl is the yield stress of the lead core.
kI Gl Al
=1+ (3.4)
kb GA
which leave the yield deformation unchanged (Eq. 3.5), or by modifying tr , which leaves the
ratio kI /kb unchanged (Eq. 3.4). On the other hand, the yield strength Q at a fixed period
Tb may be altered only by modifying Al , leaving the yield deformation unchanged. Thus,
although the yield deformation may vary with period, such variations should be independent
of yield strength; therefore, fixing kI /kb to a constant is inadequate because it causes the yield
12
Because practical concerns dictate that adjustment to the isolation period Tb over a wide
range requires modifications to not only A and G, but also tr , Eq. 3.5 suggests that a model
where yield deformation is dependent to a certain degree on the isolation period is appropriate.
Unfortunately, because these properties do not vary with period in a predictable way, devising
such a model is difficult. Given this obstacle, fixing the yield deformation to 1 cm seems
reasonable and shall be adopted throughout the study, as consideration of multiple values of
strength of the system should be defined relative to the intensity of individual ground motions.
Consider a∗y = Q/m = µg, equal to the acceleration at yield of a rigid system with strength Q,
and
which is related to the true yield deformation uy of the system by u∗y = (kI /kb − 1)uy . Dividing
the equation of motion (Eq. 3.3) by u∗y leads to the following equation:
üg
¨b + ωb2 ūb + ωb2 z(ub , u̇b ) = −ωb2
ū (3.7)
a∗y
The normalized strength, which characterizes the system strength relative to the PGV
u̇go , is defined as
a∗y µg
η= = (3.8)
ωd u̇go ωd u̇go
The frequency ωd , included to make η a unitless parameter, corresponds to the period Td mark-
ing the transition from the velocity-sensitive to the displacement-sensitive region of the median
spectrum (Fig. 2.1). Peak ground velocity is the preferred measure of ground motion intensity
for base-isolated structures with isolation period Tb typically in the velocity or displacement-
13
10 2
(a) (b)
5 1
¨g (1/s)
ūb
0 0
ū
−5 −1
−10 −2
0 10 20 30 40 0 10 20 30 40
t (sec) t (sec)
Figure 3.4: (a) Strong component of normalized ground acceleration for LMSR Record No. 11
and (b) normalized isolator deformation ūb (t) for a system with Tb = 2.5 seconds and η = 0.5.
ωb2
¨b + ωb2 ūb + ωb2 z(ub , u̇b ) = −
ū ¨g
ū (3.9)
ηωd
¨g = üg /u̇go has been normalized such that its corresponding velocity
where the acceleration ū
¨g (t)
varies from -1 to 1. Figure 3.4 demonstrates the normalized ground acceleration history ū
of the strong component of Record No. 11 and the resulting normalized deformation ūb (t), in
Equation 3.9 implies that if Tb and η are fixed, the intensity of the ground motion has no
effect on the peak normalized deformation ūbo . This important property permits meaningful
statistical analysis of the responses to an ensemble of motions with common frequency content
but variable intensity. Dispersion in ūbo is expected to be small, allowing the peak response to
be estimated with higher confidence than if the governing equation had not been normalized.
While not dependent on intensity, ūbo appears to depend on the system parameters Tb and η,
¨g (t).
the ground motion ensemble parameter ωd , and the frequency content of the excitation ū
¨g (t) have been normalized to a common intensity, their
However, because the excitations ū
variability is limited to that inherent in a random process. Thus, for a given ensemble, the
median of the normalized deformation ūbo of a system depends on only two parameters: the
Alternative Normalizations
To demonstrate that normalizing the strength by the PGV effectively reduces the dispersion in
14
1. No normalization: the system is characterized by its isolation period Tb and strength
coefficient µ.
µg
η = (3.10)
ügo
The dispersion (Eq. 2.2) of peak normalized deformation ūbo (where applicable) and peak
deformation ubo due to the strong-component ground motion ensemble is presented in Fig. 3.5
for each alternative: no normalization (Fig. 3.5a), PGA normalization (Figs. 3.5b and c), and
PGV normalization (Figs. 3.5d and e). For consistency, µ and η are the medians of their
corresponding values computed to match η for each ground motion; µ was determined from
Eq. 3.8 and η from Eq. 3.10. Thus, for five selected values of η: {0.25, 0.5, 0.75, 1.0, 1.5}, the
matching values for µ and η that have been rounded off are: µ = {.023, .045, .068, .09, .137}
The dispersion of normalized deformation ūbo is usually smallest using the PGV normal-
ization, demonstrating its superiority over the alternatives. Compare the dispersion of ūbo for
the PGV normalization, ranging from 0.3 to 0.6 (Fig. 3.5d), to that of the PGA normalization,
ranging from 0.4 to 1.0 (Fig. 3.5b), and finally to the dispersion of ubo for no normalization
(since ūbo is undefined), ranging from 0.5 to 1.0 (Fig. 3.5a). If the normalization is effective,
the dispersion of normalized deformation ūbo , which is independent of ground motion intensity,
should be smaller than the dispersion of deformation ubo , which is influenced by individual
ground motion intensity. Note that the PGA normalization provides little benefit because the
(Fig. 3.5c). For the PGV normalization, however, the dispersion of normalized deformation
(Fig. 3.5d) is lower than the dispersion of deformation (Fig. 3.5e), demonstrating the effective-
15
1.2
(a)
1
Dispersion of ubo
0.8
0.6
0.4 µ=.023
.045
0.2 .068
.090
.137
0
0 1 2 3 4 5
Tb (sec)
1.2 1.2
(b) (c)
1 1
Dispersion of ūbo
Dispersion of ubo
0.8 0.8
0.6 0.6
1.2 1.2
(d) η=.25 (e)
1 0.5 1
Dispersion of ūbo
Dispersion of ubo
0.75
1.0
0.8 1.5 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec)
Figure 3.5: Dispersion δ of various responses: (a) ubo for no normalization, (b) ūbo and (c)
ubo for PGA normalization, (d) ūbo and (e) ubo for PGV normalization.
16
3.3.3 Isolator Deformation and Force
The peak isolator deformation ubo and isolator force coefficient fbo /w can be readily obtained
where u∗y (Eq. 3.6) is related to the normalized strength η by Eq. 3.8, indicating that the
deformation demand depends, as it should, on PGV in addition to the isolation period and the
normalized strength.
Assuming the system yields, the isolator force is fbo = Q + kb ubo , leading to a correspon-
fbo
= 1 + ūbo (3.12)
Q
fbo
= µ(1 + ūbo ) (3.13)
w
where µ is related to η by Eq. 3.8, indicating that the isolator force coefficient also depends on
In Fig. 3.6, the deformation ubo and force coefficient fbo /w due to the strong component
of Record No. 8 are plotted as functions of the isolation period for several values of normalized
strength. Figure 3.6a presents data similar to Fig. 3.3b, but for specified values of η = {0.25,
0.5, 0.75, 1.0, 1.5} instead of µ. Equation 3.8 gives the corresponding values of µ equal to
{.035, .069, .104, .139, .208}, using u̇go = 44.7 cm/s for this excitation and ωd = 3.05 rad/s
(corresponds to Td = 2.06 sec) for the median spectrum of the strong-component ensemble.
While increasing the isolator yield strength reduces the deformation demand over a wide range
of isolation periods (Fig. 3.6a), it may increase the isolator force at longer periods (Fig. 3.6b).
This can be understood by recognizing that at long isolation periods Tb the system approaches
an elastic-perfectly plastic system, in which case the deformation has no limiting value, and the
17
30 0.5
η=.25
0.5
25 0.75
0.4
1.0
1.5
20
0.3
ubo (cm)
fbo /w
15
0.2
10
0.1
5
0 0
0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec)
Figure 3.6: (a) Isolator deformation ubo and (b) force coefficient fbo /w vs Tb for several values
of η due to strong component of LMSR Record No. 8.
reminiscent of the standard approach for analyzing nonlinear structural response. While similar
ζ, yield strength fy , and yield deformation uy . The yield strength fy of the structure corresponds
to the yield deformation uy whereas the yield strength Q of the isolation system is the y-intercept
of the fb -ub relation (Fig. 3.1b). If fo is the minimum strength required for the structure to
remain elastic, uo = fo /k is the corresponding deformation, and f¯y = fy /fo is the normalized
strength of the system. In contrast to the isolator strength that was normalized relative to
the intensity of the ground motion (Eq. 3.9), the structural strength is normalized relative to
the elastic demand on the structure. Although the isolator strength η (Eq. 3.8) has no upper
Table 3.1 compares the governing equations for the structural system and the isolation
system. In step 1, each equation is written in terms of a relevant system frequency: the
initial, small-oscillation frequency ωn for the structure, and the isolation frequency ωb for the
18
Table 3.1: Governing equations for nonlinear structural systems vs. isolation systems.
fy Q
ü + 2ζωn u̇ + z(u, u̇) = −üg (t) üb + ωb2 ub + z(ub , u̇b ) = −üg (t)
m m
(3) Note that ay = f¯y An , giving: (3) Define η so that a∗y = ηωd u̇go , giving:
differential equation for the ductility factor µs = u/uy , where ay = fy /m = ωn2 uy . In contrast,
the isolation equation is normalized by u∗y , which is more appropriate for the isolation system
because it depends on the isolation frequency rather than the initial frequency. Aside from the
normalizing parameter, the approaches for the two systems are parallel.
Thereafter, however, the approaches for the two systems diverge to reflect differing defi-
nitions of normalized strength. In step 3, the structural equation of step 2 has been rewritten
in a form similar to the isolation equation by relating ay to f¯y and the pseudo-spectral acceler-
ation An of the reference linear system: ay = ωn2 uy = ωn2 f¯y uo = f¯y A; thus the ground motion
in the structural equation has been normalized by spectral acceleration An . Unable to find a
reference linear system for the isolation system that consistently reflected an appropriate level
of damping, we instead normalized its strength by a ground motion intensity parameter. While
such an approach has also been suggested for structures [24, 25, 26], it has not been widely
adopted. An elastic-plastic structure worked best in the preceding comparison; however, the
19
3.5 Earthquake Response Spectra
As demonstrated earlier, the PGV normalization reduces dispersion of the normalized deforma-
tion ūbo , allowing the response to an ensemble of ground motions to be estimated with a high
degree of confidence. For the normalization to be useful, nonlinear spectra based on statistical
analysis of response data are needed. Such spectra are developed next for the strong-component
ensemble.
The median response spectrum for normalized deformation ūbo is constructed by performing
1. Select ranges for the isolation period and normalized strength. We chose Tb from 1 to
5 seconds and five values of η: {0.25, 0.5, 0.75, 1.0, 1.5}. Using the median properties
of the strong-component ensemble – ωd = 3.05 rad/s and median u̇go = 29.4 cm/s for
the unscaled motions – and a generous range for µ from 0.04 to 0.12, Eq. 3.8 led to an
estimated range of η from about 0.44 to 1.31. The five values of η chosen cover a slightly
larger range to account for some variation in intensity among ground motions.
2. Determine the normalized deformation ūbo by nonlinear response history analysis (RHA)
of Eq. 3.9, repeating over the desired range of Tb and η and for each ground motion of
the ensemble. Figure 3.4 presented such results for one time history.
3. Compute the median normalized deformation (Eq. 2.1) over all ground motions for each
value of Tb and η.
Implementing steps 2 and 3 for the strong-component ensemble led to the median normalized
Figure 3.7b shows that the median normalized deformation ūbo varies linearly with Tb on a log
scale plot. This suggested fitting a regression equation to ln (ūbo ) that is linear in the parameters
ln (Tb ) and ln (η). Regression analysis on a data set more comprehensive than the one shown
20
2
20 10
(a) exact median (b)
design spectrum
1
15 10
η=0.25
0
ūbo
ūbo
10 10
η=0.25
0.5
−1 0.75
5 0.5 10 1.0
0.75 1.5
1.0
1.5 −2
0 10
0 1 2 3 4 5 0 1
10 10
Tb (sec) Tb (sec)
Figure 3.7: Median normalized deformation ūbo for strong-component ensemble along with
the design spectrum (Eq. 3.14b), plotted in two formats: (a) linear scales and (b) log scales.
or
with coefficients estimated by the method of least squares. The design spectrum given by this
equation – shown as dashed lines in Fig. 3.7 – is very close to the exact (median) spectrum and,
Because the dispersion δ of ūbo is essentially independent of Tb and η (except for small
values of η, Fig. 3.5d), regression analysis can be used to estimate a single value for δ applicable
to all Tb and η. This method of computing dispersion, alternative to applying Eq. 2.2 for each
value of Tb and η, allows estimation of the median+σ, or 84th percentile response, without
further regression analysis. By such methods, the dispersion of ūbo was estimated to be 0.38;
further details are provided in Appendix 3A. Thus, the 84th percentile (Eq. 2.3) of ūbo is
approximated by
This equation could be used in place of Eq. 3.14b whenever a more conservative estimate
21
(a) (b)
15
10 η=0.25 η=0.25
ubo (cm)
0.5 0.5
0.75 0.75
5 1.0 1.0
1.5 1.5
exact median exact median
design equation simplified equation
0
0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec)
−10
−20
0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec)
Figure 3.8: Comparison of design equations for ubo (a) Eq. 3.16, (b) Eq. 3.18 (simplified equa-
tion) to the exact median by nonlinear RHA with design PGV of 35 cm/s; percent discrepancy
in (c) Eq. 3.16 and (d) Eq. 3.18 relative to the exact median.
A design equation for the median deformation is obtained by substituting Eq. 3.14b for
ūbo and Eqs. 3.6 and 3.8 (with ωd = 3.05) for u∗y into Eq. 3.11:
This equation provides the deformation for a given median PGV u̇go that reflects the intensity
of the design ground motions. The deformation given by Eq. 3.16 is shown to be close to its
exact median determined by nonlinear RHA of the system for the strong-component ensemble
(Fig. 3.8a), and the percent discrepancy between this design equation and the exact median is
22
that neglects quadratic variation with η, and hence may be desirable for code applications is
This equation for normalized deformation leads to a simplified equation (compared to Eq. 3.16)
Plotted against the exact median in Fig. 3.8b, the deformation estimated by Eq. 3.18 results in
only a slight loss of accuracy compared to Eq. 3.16 (Fig. 3.8a). Although the simplified equation
(Eq. 3.18) leads to a slightly larger discrepancy relative to the exact median on average, the
error bounds are about the same as for Eq. 3.16 (Fig. 3.8d vs. 3.8c).
(Eq. 3.14b or Eq. 3.17) into Eq. 3.13, or directly from the deformation (Eq. 3.16 or Eq. 3.18)
Consider the system of Fig. 3.1a with the isolator properties – postyield stiffness kb , yield
strength Q, and yield deformation uy – identical in both the x and y-directions, subjected to
bidirectional excitation. The x and y-components of bearing force are given by a generalization
of Eq. 3.2: ⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎨fbx ⎪
⎪ ⎬ ⎪
⎨ubx ⎪
⎬ ⎨zx (kI , ub, u̇b)⎪
⎪ ⎬
= kb +Q (3.19)
⎩fby ⎪
⎪ ⎭ ⎪
⎩uby ⎪
⎭ ⎩zy (kI , ub , u̇b)⎪
⎪ ⎭
where ub = ubx , uby T , u̇b = u̇bx , u̇by T and z = zx , zy T are the isolator deformation, velocity
Generalizing Eq. 3.3 gives the deformation of the isolator in the x and y-directions due
23
with ωb and µ, the isolation frequency and characteristic strength ratio, as defined before.
Equation 3.20 is solved for a system with Tb = 2.5 seconds and µ = 0.07, with the weak and
strong components of Record No. 11 applied in the x and y-directions, respectively, resulting
in the evolution of deformation traced in Fig. 3.9a. Identified in this figure are ubyo , the peak
deformation in the y-direction, and the peak deformation in any direction, subsequently referred
to as peak deformation:
ubo = max (ubx (t)2 + uby (t)2 ) (3.21)
t
Implied by Eq. 3.19, the components of the yield function interact, governed by the con-
straint that |z| ≤ 1, which results in a circular yield surface. Yielding of the system (and
vibration at the postyield stiffness kb ) is defined by |z| = 1; the system is elastic otherwise.
This interaction is also evident in the force versus deformation plots for the x and y-directions
(Fig. 3.9c and d), which appear erratic compared to comparable plots for unidirectional excita-
tion (Fig. 3.2d). Examples comparing a system with no yield-function interaction (rectangular
yield surface) to a system with yield-function interaction as assumed here are available in Fenves
To extend the normalization procedure of Sec. 3.3.2 to systems excited by two components of
This has been achieved by first dividing Eq. 3.20 by u∗y , such that ūbx = ubx /u∗y and ūby = uby /u∗y
are normalized deformations in the x and y-directions, and then substituting the normalized
strength into the right-hand side. The definition of normalized strength in Eq. 3.8 is still valid
where ωd (Fig. 2.1) and u̇go refer to the stronger component of ground motion, which is applied
in the y-direction; thus u̇go is replaced by u̇gyo and Eq. 3.8 becomes
a∗y µg
η= = (3.23)
ωd u̇gyo ωd u̇gyo
24
20 2
(a) (b)
ubo =18.7 cm
uby (cm) 10 1
zy
0 ubyo = 0
15.7 cm
−10 −1
−20 −2
−20 −10 0 10 20 −2 −1 0 1 2
ubx (cm) zx
0.2 0.2
(c) (d)
0.1 0.1
fbx /w (g)
fby /w (g)
0 0
−0.1 −0.1
−0.2 −0.2
−20 −10 0 10 20 −20 −10 0 10 20
ubx (cm) uby (cm)
Figure 3.9: Bidirectional excitation of a system with Tb = 2.5 sec and µ = 0.07 to LMSR
Record No. 11 (strong component applied in y-direction) results in: (a) deformations ubx vs
uby , (b) yield functions zx vs zy , (c) force-deformation in x and (d) y-directions
Therefore, the components of ground acceleration in Eq. 3.22 are normalized by the PGV in
¨gx = ügx /u̇gyo and ū
the y-direction, i.e., ū ¨gy = ügy /u̇gyo , such that the corresponding velocity
in the y-direction varies from -1 to +1, while the magnitude of velocity in the x-direction is
Similar to the procedure outlined in Sec. 3.5.1, a median spectrum for the deformation ubo
repeating the following steps over a suitable selection of Tb and η: (1) determine the peak
normalized deformation ūbo by nonlinear RHA of Eq. 3.22 for every motion in the ensemble, (2)
compute the median of ūbo by Eq. 2.1, and (3) convert ūbo to the deformation ubo by multiplying
by u∗y (Eq. 3.6), a function of η, PGV and ωd for the strong-component ensemble.
25
15 15
(a) (b)
10 10
ubyo (cm)
ubo (cm)
η=0.25
η=0.25
0.5 0.5
0.75 0.75
5 1.0 5 1.0
1.5 1.5
unidirectional excitation unidirectional excitation
bidirectional excitation bidirectional excitation
0 0
0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec)
Figure 3.10: Median deformation ubo for unidirectional excitation compared to the median
deformations (a) ubyo in y-direction and (b) ubo – peak in any direction – for bidirectional
excitation, both due to the LMSR ensemble with median PGV of 35 cm/s.
Figure 3.10 compares the median deformation spectra for bidirectional and unidirectional
excitations; the latter case was already shown in Fig. 3.8. The similarity between the comparable
deformations ubyo due to bidirectional excitation and ubo due to unidirectional excitation implies
that adding a transverse component of excitation has little influence on the median deformation
in the direction considered (Fig. 3.10a). However, the peak deformation ubo of the isolator
(Eq. 3.21) is significantly larger when the excitation is bidirectional (Fig. 3.10b).
Because the y-direction deformation due to bidirectional excitation is very similar to the
deformation for unidirectional excitation, both sets of design equations (Eqs. 3.14b and 3.16,
or Eqs. 3.17 and 3.18) are expected to be valid even for bidirectional excitation. The first set
is shown to be a good fit in Fig. 3.11a, where Eqs. 3.14b and 3.16 (shown as dashed lines) are
compared against the exact median values of ūbyo and ubyo due to bidirectional excitation.
Regression analysis of the response data for bidirectional excitation assuming an equation
of the form of Eq. 3.14b or of the simplified Eq. 3.17 led to coefficients for Tb and η that are
suitably close to those for unidirectional excitation. Therefore, they have been constrained to
be identical, and ūbo is estimated to be a constant 13% greater for bidirectional excitation than
26
(a)
2
10 15
exact median
1 design equation
10
10
ubyo (cm)
η=0.25
ūbyo
0
10 η=0.25 0.5
0.5 0.75
−1 0.75 5 1.0
10 1.0 1.5
1.5
−2
10 0
0 1 0 1 2 3 4 5
10 10
Tb sec Tb sec
2
(b)
10 15
exact median
1 design equation
10
10 η=0.25
ubo (cm)
0 0.5
ūbo
10 η=0.25 0.75
0.5 1.0
0.75 5
−1
10 1.5
1.0
1.5
−2
10 0
0 1 0 1 2 3 4 5
10 10
Tb sec Tb sec
Figure 3.11: Design equations for (a) y-direction deformations ūbyo (Eq. 3.14b) and ubyo
(Eq. 3.16) and (b) peak deformations ūbo (Eq. 3.24) and ubo (Eq. 3.25) compared to their exact
medians determined by nonlinear RHA due to the LMSR ensemble with PGV of 35 cm/s.
which applies for both versions of ūbo(unidirectional) (Eq. 3.14b and Eq. 3.17). (Because of this
constraint, the estimated dispersion δ is the same as for unidirectional excitation, and the same
factor of 1.46 [see Eq. 3.15] can be applied to Eq. 3.24 to achieve an 84th percentile estimate
of response. Further details of the regression methods used and statistical significance of these
Similarly, design equations for the peak isolator deformation reflect a 13% increase over
those for unidirectional excitation, and are written out here in their entirety for future reference:
27
Design Eqs. 3.24 (with Eq. 3.16 for ūbo(unidirectional) ) and 3.25 are shown to be excellent fits for
Because of the interaction of the yield functions zx and zy in the x and y-directions, the
peak force in any direction cannot be directly related to the peak deformation for bidirectional
excitation, and Eq. 3.13 is not strictly valid. However, it is shown that Eq. 3.13 is an upper
bound for the actual peak force, and is exact if the yield function z is in the same direction as
the deformation ub when its peak occurs (Appendix 3B). Because z and ub are likely to be
close to the same direction, Eq. 3.13 is a good upper bound; even when the directions differ by
If the isolation period Tb and strength coefficient µ of a base-isolation system are known, estimat-
ing the isolator deformation from a design equation, such as Eq. 3.25 or 3.26, is straightforward.
First, however, the normalized strength η, which depends on design PGV, must be determined.
Example 1
Determine the seismic deformation of an isolated building with Tb = 2.5 seconds and µ = 0.06
due to the LMSR ground motion ensemble and a design PGV u̇gyo = 35 cm/s in the direction
of strong excitation.
For comparison, the deformation computed from the median of ūbo that was obtained by non-
The design equation for deformation could also be used to find suitable choices for the isolation
period and strength coefficient, such that the deformation and force demands do not exceed
28
allowable values. For example, suppose the design problem is posed as follows: Given an
isolation period Tb and a design PGV u̇gyo , select the strength coefficient µ of the system to
the shear strain, while the height controls the diameter of the bearing needed for stability. By
limiting the deformation, the overall size of the bearing can be reduced, ultimately making
it easier to achieve the desired flexibility. The second constraint may be imposed to limit the
forces transferred to the structure; for instance, it may be necessary to keep the structure elastic
A procedure to find strength coefficients such that the constraints are satisfied is sum-
marized.
1. Find a lower bound ηL such that the deformation demand does not exceed the allowable
deformation. For the LMSR ensemble, substitute Eq. 3.25 into Eq. 3.27 to give the
constraint:
6.59 0.19 (−0.55−0.08 ln η)
T η u̇gyo ≤ uallowable (3.29)
4π 2 b
Solve this nonlinear equation numerically or by trial and error to obtain the lower bound.
2. Find an upper bound ηU such that the isolator force demand does not exceed the allowable
force coefficient. With Eq. 3.14b for ūbo and Eq. 3.23 for µ, Eq. 3.13 is substituted into
ηωd u̇gyo f
−1.81 (−1.55−0.08 ln η)
1 + 1.91 Tb η ≤ (3.30)
g w allowable
Solving this nonlinear equation numerically or by trial and error gives a conservative
estimate of ηU , because the left side is an upper bound to the true force.
3. If ηL < ηU , then both constraints (Eqs. 3.27 and 3.28) can be met by any η within the
bounds. Convert the bounds for η to bounds for µ using Eq. 3.23, giving µL < µ < µU .
29
Select a strength coefficient µ, and compute the deformation demand as demonstrated in
4. If ηL > ηU , then allowable deformation and force cannot both be satisfied, thus requiring
Example 2
For an isolation system with Tb = 2.5 seconds, select the strength coefficient to sustain bidirec-
tional shaking represented by the LMSR ensemble and a design ground velocity of 35 cm/s in
the direction of strong excitation, so that the deformation does not exceed 10 cm and the force
3. An acceptable range for η is 0.48 to 0.83, equivalent to µ from 0.052 to 0.09 (Eq. 3.23).
calculated from Eq. 3.25 and Eq. 3.13, of ubo = 8.72 cm and fbo /w = 0.126. These
demands are below the allowable limits and the design is acceptable.
The International Building Code 2000 [1] and previous editions of the Uniform Building Code
whose properties are generally determined by iteration. This section first evaluates the accuracy
of this method to estimate the isolator deformation for unidirectional excitation, defined by
the strong-component spectrum in Fig. 2.1a, and then evaluates a code rule to increase the
deformation for bidirectional excitation. Recall that the this spectrum was developed by scaling
the strong component of each motion to a PGV of 35 cm/s (Sec. 2.1), representing the same
30
3.9.1 Estimating Isolator Deformation
Suppose the deformation of a system with known isolation period Tb and strength coefficient µ
due to unidirectional excitation is to be estimated from a linear response (or design) spectrum.
An iterative procedure, which determines the equivalent-linear system for a nonlinear isolation
1. Using initial guesses for the effective period Tef f and effective damping ζef f of the
2. Evaluate the isolator force from Eq. 3.2, taking z to be 1: fbo = Q + kb ubo .
3. Update estimates of effective period: Tef f = 2π w/(kef f g), and effective damping:
4. Repeat steps 1-3 with updated values of Tef f and ζef f , until successive estimates of ubo
converge.
Example 3
Estimate the deformation for an isolated building with Tb = 2.5 seconds and µ = 0.06 due to
unidirectional excitation defined by the median spectrum of Fig. 2.1a for 5% damping. Spectral
displacements for other damping ratios, needed in the iterative procedure, are determined by
additional response history analyses. Select the weight of the structure as w = 1600 kN, giving
1. For initial guesses of Tef f = 2 seconds and ζef f = 0.15, the median ubo = 9.27 cm.
2. The isolator force, computed from Eq. 3.2, is fbo = 96 + 10.29 ∗ 9.27 = 191.39 kN.
3. The new effective stiffness is kef f = 191.39/9.27 = 20.65 kN/cm, leading to updated
effective period Tef f = 2π 1600/(20.65 ∗ g) = 1.765 sec, and damping ζef f = (2/π) ∗ 96 ∗
31
4. The deformation demand for the updated equivalent-linear system is ubo = 6.75 cm.
Steps 1-3 are repeated until convergence is attained, with the iterations summarized in
Table 3.2.
This equivalent-linear procedure converges after six iterations, giving deformation demand
ubo = 5.88 cm. This deformation is unconservative compared to the exact median deformation
determined by nonlinear RHA of Eq. 3.9 for all 20 motions: ubo = 8.38 cm.
The accuracy of using a linear spectrum to estimate the deformation of a system subjected to
unidirectional excitation is evaluated over a broader range of isolation period Tb and normalized
strength η. Figure 3.12a plots the median deformation for a system with η = 0.5 due to the
The nonlinear deformation is based on the exact median of ūbo determined by nonlinear RHA
instead of the design equation (Eq. 3.16). These plots indicate that the equivalent-linear pro-
cedure fails to recognize the trend of increasing deformation with period, thereby increasingly
positive when the equivalent-linear estimate exceeds the nonlinear deformation. These plots
confirm that this estimate is unconservative for a wide range of isolation period and normalized
strength. The isolator deformation is underestimated by 20 to 50% for most practical systems,
Table 3.2: Computation of iterations for the equivalent-linear analysis procedure of Example 3.
Teff (s) ζeff ubo (cm) fbo (kN) fbo/w keff (kN/cm)
2.000 0.150 9.27 191.39 0.120 20.65
1.765 0.285 6.75 165.47 0.103 24.51
1.620 0.315 6.12 158.99 0.099 25.98
1.574 0.322 5.94 157.13 0.098 26.45
1.559 0.323 5.89 156.62 0.098 26.59
1.555 0.324 5.88 156.52 0.098 26.62
1.555 0.324 5.88
32
10 50
(a) (b) η=0.25
0.5
8 0.75
% Discrepancy in ubo
25 1.0
1.5
6
ubo (cm)
0
4
Nonlinear −25
2 Equiv-Linear
0 −50
0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec)
becoming increasingly unconservative for longer isolation periods and higher strengths.
systems has been documented by several researchers [28, 29, 30]. For example, the ATC-40
inelastic design spectra [28]. Thus, it is not surprising that the isolator deformations determined
Although the equivalent-linear procedure of the IBC does not address the issue of bidirectional
excitation[1], the 100% + 30% rule is recommended for response spectrum analysis. Specifi-
cally, the maximum deformation of the system is calculated as the vector sum of orthogonal
deformations resulting from 100% of the ground motion applied in the most critical direction
and 30% applied orthogonally. Adopting this rule for the IBC equivalent-linear procedure, the
ratio of peak deformations due to bidirectional and unidirectional excitations is given by:
ubo
= 12 + 0.32 = 1.044 (3.31)
ubo(unidirectional)
33
This estimate is based on the conservative assumption that both components of ground motion
have the same intensity. Compare this to a 13% increase in the median deformation due to
bidirectional excitation given in Eq. 3.24. The estimated increase in isolator deformation by the
100% + 30% rule is only about one-third of the exact increase determined by nonlinear RHA
to bidirectional excitation; hence, another possibility for improvement to the code procedure.
3.10 Conclusions
This investigation to develop a procedure to estimate the isolator deformation and force of a
base-isolated building, based on nonlinear response history analysis, has led to the following
conclusions:
1. For a bilinear isolation system, the common modeling strategy of fixing the ratio of the
initial to the postyield stiffness is inadequate when the response is to be compared over a
wide range of parameters of the system. This strategy, which causes the yield deformation
to grow in proportion to the yield strength, does not correctly represent the behavior of
2. The median – over an ensemble of ground motions – of the normalized deformation ūb
depends only on the isolation period Tb and normalized strength η. This was achieved by
defining the normalized strength as the system strength ÷ ωd u̇go , where u̇go is the peak
ground velocity and ωd is the corner frequency separating the velocity and displacement-
This implies that the peak response can be estimated with a high degree of confidence
by the median response of the system to the ground motion ensemble. For the strong-
34
4. The approach for isolation systems presented here differs from the preferred approach for
structural systems as follows: the controlling frequency, which determines the normalized
deformation, is the isolation frequency rather than the initial frequency, because it has
proved unfeasible to define a linear system that accurately reflects the energy dissipation
of the isolators; the strength is normalized by a ground motion intensity parameter rather
5. The median normalized deformation of the isolation system showed simple trends across
sion analysis, which led to a design equation for the isolator deformation. Comparable
equations for the peak deformation due to unidirectional excitation and the peak defor-
a 13% increase due to the second component of ground motion. Simplified versions of
these equations that may be preferable for code applications were also included. Exam-
ples demonstrated the use of these equations to determine the deformation of a system
with given parameters, or to select the yield strength of a system such that the isolator
6. For a given spectrum, equivalent-linear procedures in building codes were shown to under-
estimate the isolator deformation by 20 to 50% compared to the exact median deformation
30% rule allows for at most a 4.4% increase in deformation due to bidirectional excitation,
compared to the exact increase of 13% determined by nonlinear response history analysis.
This investigation has provided a framework for an alternative procedure that could be
integrated into the base-isolation code to replace the current equivalent-linear procedure.
35
Appendix 3A: Regression Statistics
The R Project, an open-source environment for statistical computing available through GNU,
was used for regression analysis that led to the design equations. Excerpts of the R Project
output for Eq. 3.14b (Table 3A.1a) and Eq. 3.17 (Table 3A.1b) are listed, which also apply
to Eq. 3.24. Before this output is discussed, a terminology for linear least-squares regression
is developed. Any textbook on statistical methods [31, 32, 33] could serve as a reference for
the following concepts. For the general problem, the observed data Y is fit to an equation
containing regressors X1 , X2 , etc. The vector Y contains individual observations yi , and the
matrix X contain the values xi1 , xi2 , etc., corresponding to each observation yi . With linear
regression, the model or predictor for Y , Y̌ , is simply a linear function of the regressors, i.e.,
yi = y̌i + ei = xik βk + ei (3A.1)
k
where ei is the residual error between the observed value yi and the predicted value y̌i . By the
least-squares method, the coefficients β are selected to minimize the norm of the residual error,
X (Y − Xβ) = 0 (3A.2)
Applying the above notation to the model of Eq. 3.14a, which is linear by logarithmic
transformation of the variables, the observed data is Y = ln (ūbo ), and the regressors are X0 = 1
(a vector of ones), X1 = ln (Tb ), X2 = ln (η) and X3 = (ln η)2 . Table 3A.1 shows an additional
regressor Ind , an indicator variable that distinguishes observations of ūbo due to bidirectional
excitation (Ind = 1) from the single-direction responses ūbo(unidirectional) and ūbyo (Ind = 0).
This indicator allows the three different responses to be fit to the same model, whereby ubo
due to bidirectional excitation is permitted to increase by a constant relative to the other two.
Compared to the model for the design equation (Eq. 3.14a, Table 3A.1a), the model for the
36
Table 3A.1: Excerpts of R Project output for analysis of response to both unidirectional and
bidirectional excitation
3A.1 R2 statistic
Listed in Table 3A.1, the ‘Multiple R-Squared’ statistic is a comprehensive measure of the
where Ȳ is the average of Y over the number of observations, n. Interpreting this equation, R2
is the fraction – between 0 and 1 – of total variability in Y accounted for when Y̌ is used to
37
predict Y . Exactly equal to 1 when all data points lie on the regression line, R2 should ideally
be close to 1; however, there is no line between acceptable and unacceptable values [31].
With resulting values of R2 equal to 0.91 and 0.9095 (Table 3A.1), both models (for
Eqs. 3.14b and 3.17) led to good fits of our data. This is especially encouraging since three
different responses (ūbo(unidirectional) , ūbyo and ūbo ) were merged into a common model. Fur-
thermore, the negligible difference in R2 for the models leading to the design equation and the
simplified equation suggests that the additional regressor (ln η)2 accomplishes little. However,
The residual errors are assumed to be independent and normally distributed with mean 0 and
ei ∼ N (0, σ 2 ) (3A.4)
When these assumptions are satisfied, the user can go beyond simple data fitting to calculating
distribution parameters, constructing confidence intervals, etc. A zero mean for the residual
errors is an intrinsic property of linear least-squares regression. For the data considered here,
typical normality tests [31, 32, 33] verified that the observed data ūbo is lognormally distributed,
or Y = ln ūbo is normally distributed, and hence the residual errors are also normally distributed.
Furthermore, the residual errors were nearly constant when plotted against individual regressors
When this assumption (Eq. 3A.4) is satisfied, the variance σ 2 is estimated from the
where d degrees of freedom is equal to n minus the number of regressors. In this derivation, the
residual error mean ē = 0, and X (Y − Xβ) = 0 (Eq. 3A.2). Listed as the ‘Residual standard
error’ (Table 3A.1), the square root of the variance or σ was estimated to be about 0.38 for
both regression models. In Sec. 3.5.2, this value was referred to as the dispersion of ūbo , which
38
is the standard deviation of Y (= ln ūbo ) (Eq. 2.2). The dispersion of ūbo equals σ because the
predictor Y̌ has zero variance (Y̌ = Xβ), which means that the variance of Y is identical to
√
σβk = σ ckk (3A.6)
where ckk is the kth diagonal element of (X X)−1 . The output of Table 3A.1 lists four statistics
for each regressor: the ‘estimate’ of βk , its ‘std. error’ (σβk ), the ‘t-value’, equal to βk ÷σβk , and
a probability ‘Pr(>|t|)’. The probability tests the hypothesis that βk is zero. Specifically, the
t-value results from substituting zero with mean βk and standard deviation σβk into a student’s
are used. Pr(>|t|) lists the probability of the current t-value occurring if βk = 0 (or βk is
within an interval that contains zero). Thus, when the t-value for βk is sufficiently large or its
corresponding probability sufficiently small, the hypothesis that βk is zero is rejected, meaning
Applying these concepts to our output, all regressors for the model of Table 3A.1a have
large t-values and small probabilities of their coefficients being zero. In this sense, each coef-
being nonzero). These high t-values justify keeping each regressor in the equation, including
the questionable regressor (ln η)2 . However, because the simplest reasonable model may be
preferred for practical application, and the R2 statistic (Eq. 3A.3) indicated little improvement
when (ln η)2 was included (Table 3A.1), Eq. 3.17 was offered as an alternative.
The merging of responses ūbo(unidirectional) , ūbyo and ūbo into one regression equation was
justified by analysis of their regression coefficients determined independently. Table 3A.2 shows
that the corresponding coefficients for the three responses are close in value. To improve on
this, a 95% confidence interval for each coefficient was computed from its estimated standard
39
Table 3A.2: R Project output for independent analysis of various responses
40
Table 3A.3: 95% confidence intervals for coefficients
error; these intervals are listed in Table 3A.3 and compared to the coefficients from analysis
of the combined responses (Table 3A.1a). The combined model should not be used if it gives
coefficients outside the confidence intervals for any of the individual models. However, the
estimated coefficients from Table 3A.1a are within all confidence intervals, justifying acceptance
41
Appendix 3B: Peak Force due to Bidirectional Excitation
fb = Qz + kb ub (3B.1)
where fb = fbx , fby T , ub = ubx , uby T , and z = zx , zy T are vectors containing the x and
|fb|2 = fb · fb
= Q2 z · z + 2Qkb z · ub + kb2 ub · ub
Presumably, the peak force occurs when the system is fully yielded (|z| = 1) and the deformation
is near its peak value (|ub| ≈ ubo ). Thus, with cos θ taking on any value from -1 to +1, the
These upper and lower bounds correspond to the cases when the vectors z and ub are collinear
and point in the opposite directions, respectively. In general, the hysteretic contribution to the
force (Qz) is in the direction of instantaneous velocity [34], which is unlikely to be collinear
with the deformation. However, as the isolator deformation increases toward a maximum, the
yield surface translates outward such that the direction of instantaneous velocity is close to the
direction of deformation. The angle between the vectors is certainly less than 90 degrees, which
represents a translation along the yield surface that results in no deformation increase. If the
directions of z and ub vary by as much as 60 degrees, the upper bound estimate of fbo is about
15% conservative when Q and kb ubo are equal, and less conservative when they are unequal.
fbo ≈ Q + kb ubo
42
Appendix 3C: Notation
bearing
of lead core
of linear spectrum
43
Tn initial period of nonlinear fixed-base structure
tion
u∗y Q/kb
44
η strength of isolator or isolation system normalized by peak ground
yield deformation
45
4: ESTIMATING THE PEAK DEFORMATION IN AN
ASYMMETRIC-PLAN SYSTEM
4.1 Introduction
The earthquake response of asymmetric, base-isolated structures has been studied using analyt-
ical, numerical and experimental techniques. Analysts working with a simplified model of the
system – a rigid structure on linear isolators – derived closed-form equations for the absolute
maximum or peak response to an acceleration impulse [35, 36]. When the system’s torsional
and lateral frequencies were equal, the corner deformation was predicted to be uninfluenced
by torsion because the peak rotation, although large, occurred with a significant time lag after
the peak lateral deformation [35, 36]. On the other hand, when the torsional frequency of the
system was much greater than its lateral frequency, the corner deformation was estimated by
summing the peak lateral and (now smaller) rotational deformations, which were likely to occur
near the same time [36]. Compared to this estimate, a code factor to account for the increase
lateral frequency ratios less than 1.485, which, in our view, includes all isolated systems. A later
investigation of the same simplified system [37], which compared results from response history
analysis and response spectrum analysis to this static code factor, concluded that when the
torsional frequency is greater than the lateral frequency, response spectrum analysis is accurate
Researchers employing numerical techniques [38, 39, 40, 41] incorporated bilinear force-
deformation of the isolators and superstructure models of varying detail. Some included yield-
surface interaction of the isolators and varied system parameters over wide ranges [40, 41].
Despite the sophistication, earlier studies [38, 39] sought only to show that the superstructure
forces and deformations are reduced compared to a fixed-base structure, as had already been
accepted for symmetric systems. Clearly, torsion leads to significantly increased isolator defor-
mations only if large rotations and large lateral deformations at the center of mass (CM) occur
at or very near the same time during the earthquake, but some of these studies considered the
46
peak rotations and lateral deformations independently [38, 41]. One investigation [40] related
the increase in deformation at the corner relative to the CM to system parameters: superstruc-
ture or isolation eccentricity, superstructure period, and isolation or structure frequency ratios
[40], but the conclusions were based on only two ground motions.
[18] resulted in large rotations of the structure, which contributed significantly to the corner
of a small number of lead-rubber and natural-rubber bearings, may have been unrealistically
Although the analytical studies provide insight into the behavior of the system, they are
limited by oversimplification, and previous numerical studies lacked sufficient data to determine
reliably the influence of system parameters on corner deformation. Armed with faster computers
and a growing database of recorded ground motions, response trends can be determined based
on statistical analysis of response due to many ground motions, paving the way for better
plan systems, thereby developing equations to estimate the peak lateral deformation among all
isolators. These equations are based on global properties of the system, which have been limited
allow one equation, determined from analysis of an idealized system subjected to bidirectional
excitation, to be used to estimate the deformation of many systems with varying planwise layout
of isolators, even though these systems are highly nonlinear. The results demonstrate that the
of a symmetric system, by a factor that depends only on eccentricity between the CM and
center of rigidity (CR) and distance from the CM to the outermost corner. The code static
procedure for determining the design displacement is evaluated in light of the results presented.
47
CM axis
Isolator i
Position (xi , yi )
Properties kbi , Qi
Deformations ubxi , ubyi
Consider a rigid slab with mass m uniformly distributed in plan, and radius of gyration r,
supported on a system of isolators (Figure 4.1). Each isolator is distinguished by its position
(xi , yi ) from the CM, postyield stiffness kbi and yield strength Qi . The properties kbi and Qi
are independent of the direction of loading. The x and y-components of force in the ith isolator,
where ubxi and ubyi are the x and y-components of isolator deformation, and zxi and zyi are the x
and y-components of the yield function. These components interact when loaded bidirectionally
(Sec. 3.6.1), such that the magnitude of the yield function equals 1 when the isolator is yielding
and is less than 1 otherwise. Yielding depends on the initial stiffness, determined for every
48
4.2.2 Equations of Motion
Equations governing ubx and uby , the x and y-components of deformation at the CM, respec-
Equation 4.1 is substituted for fbxi and fbyi , with isolator deformations expressed in terms of
where kb is the lateral postyield stiffness of the isolation system, kbθz is the torsional postyield
stiffness about the CM, and ebx and eby are the x and y-components of the stiffness eccentricity,
lateral direction. The stiffness eccentricity defines the location of the CR relative to the CM,
where the CR is the point about which the first moment of isolator stiffnesses is zero, and a
Strength properties analogous to the stiffness properties of Eq. 4.4 include the lateral
strength Q (defined identically in Sec. 3.2.1), the torsional strength about the CM, Qθz , and
the strength eccentricity, with components epx and epy in the x and y-directions, all defined
below:
Q= Qi Qθz = Qi (x2i + yi2 ) (4.5a)
i i
49
1 1
epx = Qi xi epy = Qi y i (4.5b)
Q Q
i i
Together, these parameters describe the distribution of strength in the system, but unlike the
stiffness parameters, cannot be separately identified in the equations of motion (Eq. 4.3) due
to the system nonlinearity. Previous studies apparently ignored strength eccentricity [38, 41],
considered strength eccentricity instead of stiffness eccentricity [40], or considered both [39].
Later assumptions will clarify how the strength parameters should be defined.
where
ωbθz
Ωθz = (4.7)
ωb
and
kb kbθz
ωb = ωbθz = (4.8)
m mr 2
are the lateral and torsional isolation frequencies. As Ωθz , the ratio of the torsional to the lateral
frequency, increases, the system becomes stiffer in torsion relative to its lateral stiffness. To
obtain Eq. 4.6, the nonlinear terms in Eq. 4.3 were rewritten by introducing γi = Qi /Q, allowing
Q to be factored out; µg was substituted for Q/m. The isolation frequency ωb and strength µ
are identical to corresponding properties that govern the response of the single isolator system
(Sec. 3.2.2); for a given asymmetric system the single isolator system with the same values of
Equation 4.6 indicates that the deformations ubx , rθbz and uby of the asymmetric system
depend on ωb , µ, Ωθz , ebx /r and eby /r, which shall be referred to as the global properties of
the system, and on the planwise layout of isolators. However, the response of a linear system
(µ = 0) depends only on the global properties and is independent of the isolator layout.
50
4.2.3 Characteristics of Response Histories of Asymmetric System
Response histories are presented for the above defined inelastic, asymmetric system for the
special case of unidirectional (y-direction) excitation ügy (t) and one-way asymmetry (eby =0),
depend only on the system’s global properties and planwise layout of isolators, and thus not
explicitly on the distance r (Eq. 4.6). When eccentricity ebx /r is positive, u−ry refers to the
‘flexible’ side, which tends to deform more than the ‘stiff’ side.
ratio of 2 that is asymmetric due to an imbalance of isolator stiffnesses; this idealized system
is developed in detail in Sec. 4.3. Figure 4.2 presents the response histories for a linear system
(µ = 0) with Tb = 2.5 seconds and ebx /r = 0.1 for two different frequency ratios, Ωθz = 1 and
1.25, subjected to a selected ground motion. Stiffness proportional damping that would reduce
to ζ = 0.05 if the system were symmetric is included. The deformation ub of the corresponding
symmetric system is shown dotted for reference. The response of this single isolator system
was studied in detail in Chapter 3. Also shown are the lateral deformation uby at the CM, the
The response histories of this linear system are influenced significantly by the frequency
ratio. When the torsional and lateral frequencies are equal (Ωθz = 1, Fig. 4.2a), the peak
rotational deformation rθbz is greater than 50% of the peak lateral deformation ubyo . Concurrent
to a slow build-up of the rotational component, the lateral component dies out, and thus differs
significantly from the deformation ub of the symmetric system. However, when the torsional
frequency exceeds the lateral frequency (Ωθz = 1.25, Fig. 4.2b), the peak rotational deformation
rθbzo is only 20% of the peak lateral deformation ubyo , and absent a rotation build-up, the lateral
for systems with equal lateral and torsional frequencies (Ωθz = 1), the time lag of the peak
51
(a) (b)
25
−25
ub (t) ub (t)
0
25
25
Deformation (cm)
Figure 4.2: Deformation ub of a linear symmetric system and deformations uby , rθbz , u+ry
and u−ry of a linear, asymmetric system with Tb = 2.5 sec, ebx /r = 0.1, and (a) Ωθz = 1 and
(b) Ωθz = 1.25, subjected to the strong component of LMSR Record No. 12
rotational deformation, observed in Fig. 4.2a, prevented it from contributing to the corner de-
formation. For systems with torsional frequency much greater than lateral frequency (Ωθz 1),
although the rotational deformation was much smaller, the peak lateral and rotational deforma-
tions were assumed to occur at the same time to allow an estimate of the corner deformation.
These features are discernible in Fig. 4.2, although the conclusions of [36], based on impulse
excitation, apply only partially for realistic ground motions, which is not surprising.
lustrated in Fig. 4.3 for a system with µ = 0.06. The rotational deformation does not build
up over time for Ωθz = 1, and though its maximum is still larger for Ωθz = 1 than 1.25, the
difference is less significant. In contrast to linear systems, both the lateral (uby ) and rotational
52
(a) (b)
25 25
ubo =15.45 cm ubo =15.45 cm
ub (t) ub (t)
0 0
−25 −25
25 25
ubyo =15.52 cm ubyo =15.11 cm
uby (t) uby (t)
0 0
Deformation (cm)
−25 −25
25 25
rθbz (t) rθbzo =1.20 cm rθbz (t)
0 0
rθbzo =2.02 cm
−25 −25
25 25
u+ryo =13.94 cm u+ryo =14.49 cm
u+ry (t) u+ry (t)
0 0
−25 −25
25 25
u−ryo =17.14 cm u−ryo =15.80 cm
u−ry (t) u−ry (t)
0 0
−25 −25
0 10 20 30 40 0 10 20 30 40
Time (s) Time (s)
Figure 4.3: Deformation ub of a nonlinear symmetric system and deformations uby , rθbz , u+ry
and u−ry of a nonlinear asymmetric system with Tb = 2.5 sec, µ = 0.06, ebx /r = 0.1, and (a)
Ωθz = 1 and (b) Ωθz = 1.25, subjected to the strong component of LMSR Record No. 12
(rθbz ) deformations die out quickly for the nonlinear system, and the ratio of peak rotational to
lateral deformations is smaller; compare rθbzo /ubyo = 0.13 for Ωθz = 1 and 0.08 for Ωθz = 1.25
(Fig. 4.3) to 0.55 and 0.20 for the linear system (Fig. 4.2). Thus, the lateral deformation history
uby of the asymmetric system closely resembles ub of the corresponding symmetric system.
The deformations u±ry , applicable for one component of excitation, can be generalized to a
Let uro be the peak deformation over all rx , ry located at distance r from the CM, thereby
satisfying:
53
The x and y-components of deformation at location rx , ry are, respectively:
urx (t) = ubx (t) − ry θbz (t) ury (t) = uby (t) + rx θbz (t) (4.11)
By the method of Lagrange multipliers, ur (t) is maximized subject to the constraint of Eq. 4.10,
the details of which are provided in Appendix 4A. The simple result is
where ub (t) = [u2bx (t) + u2by (t)]1/2 is the magnitude of the deformation at the CM. Implicit in
Eq. 4.13, the deformation is first maximized over the circle of radius r at each time instant and
then maximized over time. Like u±ry , uro depends on the global properties and planwise layout
Next, the equations of motion for the asymmetric system are normalized comparable to the
equation for the single isolator system (Sec. 3.3.2). Let a∗y and u∗y for an asymmetric system be
Q a∗y
u∗y = = 2 (4.14)
kb ωb
where x̄ notation denotes response x normalized by u∗y . Identical to the definition for bidirec-
tional analysis of a single isolator system (Eq. 3.23), η is the strength of the isolation system
54
with u̇gyo and ωd referring to the stronger component of ground motion, always applied in the
Similar to the single isolator system (see Secs. 3.3.2 and 3.6.2), Eq. 4.17 indicates that any
median normalized deformation of interest (such as ūbx , r θ̄bz and ūby ) over an ensemble of ground
motions depends primarily on the system’s global properties that were defined previously, with
is weakly affected by ground motion intensity, and the dispersion of normalized deformation over
the ensemble is expected to be small, as proven for the single isolator system in Sec. 3.3.2. These
properties permit meaningful statistical analysis of the relevant deformations of the asymmetric
system.
For given global properties, the median deformation over an ensemble of ground motions, with
weak and strong-component ensembles applied in the x and y-directions, respectively, is deter-
mined by the following steps: (1) Compute the peak normalized deformation by nonlinear RHA
of Eq. 4.17 to every motion in the ensemble, (2) compute the median value (over the ensemble)
of the normalized deformation by Eq. 2.1, and (3) convert the normalized deformation to actual
deformation by multiplying by u∗y (Eq. 4.14), expressed as a function of η (Eq. 4.16), design
PGV u̇gyo and the corner frequency ωd for the strong-component ensemble.
55
4.3 Modeling Asymmetric-Plan Systems
To focus this study, the ranges of global properties most applicable to base-isolated buildings are
determined first. Based on Chapter 3, the isolation period Tb is varied from 1 to 4 seconds and
the normalized strength η from 0.25 to 1.5. A linear system (η = 0) with stiffness proportional
Previous studies considered the frequency ratio Ωθz varying from 0.8 to 1.7 [40], and normalized
eccentricities ebx /r and eby /r upwards of 0.2 [18, 36, 40, 41]; however, these are unrealistically
Consider a system with uniformly distributed mass and identical isolators located regu-
larly over the building plan. Since isolators are mass-produced by manufacturers, limiting to
one prototype is economical, justifying the assumption of identical isolators. Exceptions occur,
for example, when the isolators in the interior of the plan are larger and stiffer than those on
the exterior to support larger tributary mass, or elastomeric bearings are used in the interior
and lead-rubber bearings on the exterior to increase the torsional resistance. With identical
isolators, the planwise distributions of strength and stiffness are identical, implying that the
ratio of torsional to lateral strengths Qθz /Q (Eq. 4.5a) equals the ratio of the torsional to lat-
eral stiffness kbθz /kb (Eq. 4.4a), and the strength eccentricities epx and epy (Eq. 4.5b) equal the
Given these assumptions, as the number of isolators approaches infinity, Eq. 4.4 becomes
kbθz ⇒ k̄b x2 + y 2 dA (4.18a)
A
1 1
ebx ⇒ k̄b xi dA eby ⇒ k̄b yi dA (4.18b)
kb A kb A
where k̄b is the stiffness per unit area. When the stiffness and mass are both uniformly
distributed, Eq. 4.18a results in kbθz = kb r 2 with r equal to the mass radius of gyration, Ωθz
(Eq. 4.7) reduces to 1, and the eccentricities (Eq. 4.18b) become ebx = eby = 0. Thus, for a
system containing many isolators, the frequency ratio is close to 1 and the eccentricity is close
56
(a) Rectangular Plan (b) L-plan
1.5
b/d=1 b/t=2
1.4 2 3
3 5
1.3 5 10
10
Ωθz
1.2
1.1
1
0 50 100 150 200 0 50 100 150 200
Number of isolators Number of isolators
(c) L-plan
0.06
b/t=2
0.05 3
5
ebx /r = eby /r
0.04 10
0.03
0.02
0.01
0
0 50 100 150 200
Number of isolators
Figure 4.4: Torsional to lateral frequency ratio Ωθz for (a) rectangular-plan system with
aspect ratios b/d = {1, 2, 3, 5, 10} and (b) asymmetric L-plan system with length-to-thickness
ratios b/t = {2, 3, 5, 10}; and (c) eccentricity for L-plan system, all presented as a function of
plan-dimension ratio and number of isolators
system and an asymmetric L-plan system that satisfy the above assumptions. Figure 4.4a and
b show the frequency ratios Ωθz for a rectangular plan with various aspect (length-to-width)
ratios b/d, and an L-plan (Plan A1 in Fig.4.5a) with various length-to-thickness ratios b/t. This
figure confirms a lower bound frequency ratio of 1, and suggests an upper bound of 1.3 for any
system with more than 10 isolators, thus Ωθz is selected as {1 - 1.3}. Figure 4.4c shows that the
maximum normalized eccentricity in either direction (ebx /r, eby /r) for the L-plan is less than
0.06. The frequency ratios for two other examples, Plans A2 and A3 in Fig. 4.5a, fall within
the range suggested above. The normalized eccentricity is largest (ebx /r = 0.108) for Plan
A2, which has the smallest number of isolators and irregular spacing. Because it was hard to
57
(a)
Plan A1: Plan A2: Plan A3:
ebx /r = eby /r = 0.049 ebx /r = 0.108, eby /r = 0 ebx /r = 0, eby /r = 0.057
Ωθz = 1.180 Ωθz = 1.280 Ωθz = 1.172
t 3
3
3 4
d 4
b d
2 1
2
t
1
2 1
b
d (=b) b
Figure 4.5: (a) Configuration of 3 asymmetric-plan systems, and (b) configuration of idealized
system
contrive even these examples, normalized eccentricity will typically be smaller than 0.1. These
bounds derived for frequency ratio and eccentricity are not intended to be firm, but simply
As discussed earlier, for a linear isolation system, uro is independent of the planwise layout of
isolators. Although individual elements typically factor into the response of a nonlinear system,
the planwise layout of isolators has little influence on the deformations of a nonlinear isolation
aspect ratio b/d = 2, where the stiffnesses of individual isolators are varied to match the global
properties Tb , η, Ωθz , ebx /r and eby /r of Plans A1-A3, and the strength of each isolator is varied
58
(a) Plan A1 (b) Plan A2 (c) Plan A3
30
Plan A1 Plan A2 Plan A3
Idealized Idealized Idealized
25
0
0 1 2 3 4 0 1 2 3 4 0 1 2 3 4
Tb (sec) Tb (sec) Tb (sec)
Figure 4.6: Comparison of median deformations uro of systems with Plans A1-A3 and the
idealized, rectangular-plan (b/d = 2) system with identical global properties
Computed for the LMSR ground motion ensemble with median PGV of 35 cm/s (Sec. 4.2.6),
Fig. 4.6 compares median uro for each of Plans A1-A3 and their corresponding idealized sys-
tems. The deformations of a linear system with η=0, known to be identical for the actual
and idealized systems, are shown for reference. The deformations of the two systems appear
identical, with discrepancies less than 1.5% over all values of Tb and η for all three plans.
These examples suggest that the idealized system is representative of various geometries with
different isolator layouts. The discrepancies for different plans are minimized because isolators
have identical resisting properties in any direction and tend to be arranged at regular intervals
To verify that the idealized, rectangular system with b/d = 2 can represent a sufficiently
wide range of asymmetric plans, this comparative analysis is extended to the extremes, a square
plan with b/d = 1 and a slender plan with b/d = ∞. Figure 4.7 compares median uro , computed
as above, for these extreme plans and the idealized system with identical global properties. The
deformations of a linear system, identical for the actual and idealized systems, are shown again
for reference. Even though these geometries result in greater differences in isolator layout than
before, the deformations compared in Fig. 4.7 still appear identical, with discrepancies less than
59
30
(a) b/d=1 (b) b/d=∞
Idealized Idealized
25
20 η=0 η=0
uro (cm)
0.25 0.25
15
0.5 0.5
10
1.0 1.0
5 1.5 1.5
0
0 1 2 3 4 0 1 2 3 4
Tb (sec) Tb (sec)
Figure 4.7: Comparison of median deformations uro of systems with extreme aspect ratios [(a)
b/d = 1 and (b) b/d = ∞] and the idealized, rectangular-plan system (b/d = 2) with identical
global properties: ebx /r = 0.1 and Ωθz = 1.25
(a) (b) y
y
x
CR ügx
x
ügx CM CM CR
ügy ügy
Figure 4.8: (a) Two-way asymmetric system with ground motions applied in the x and y -
directions and (b) idealized system with same ground motions applied in the x and y-directions;
both with identical global properties Tb , η, Ωθz , and magnitude of eccentricity eb /r (not to scale)
Given the outcome of these tests, we conclude that the median deformation, such as uro ,
based only on the global properties of the system is sufficiently accurate, differences in geometry
and isolator layout are unimportant, and the idealized, rectangular-plan system with b/d = 2
where the x -axis is directed toward the CR. The eccentricities in the x and y -directions are
ebx = e2bx + e2by and eby = 0. Thus, a two-way asymmetric system can be viewed after axis
60
(a) Ωθz = 1 (b) Ωθz = 1.25
30
2-way asymmetric
Idealized
25
20 η=0 η=0
uro (cm)
15 0.25 0.25
0.5 0.5
10
1.0 1.0
5 1.5 1.5
0
0 1 2 3 4 0 1 2 3 4
Tb (sec) Tb (sec)
Figure 4.9: Comparison of median deformations uro of a 2-way asymmetric system with
rotated axes and the idealized, rectangular-plan system with identical global properties; ebx /r =
0.1 and (a) Ωθz = 1 and (b) Ωθz = 1.25
rotation as a one-way asymmetric system. The other global properties of the system Tb , η, and
Ωθz are unaffected by axis rotation. Because isolators have identical resisting properties in any
direction, Tb and η remain the same; Ωθz (Eq. 4.7) is unchanged because kbθz (Eq. 4.4) and r
Consider the idealized system of Fig. 4.5b with global properties identical to those of the
two-way asymmetric system with rotated axes; the eccentricity ebx /r of the idealized system
equals the eccentricity ebx /r of the two-way asymmetric system. If the system is linear (η = 0),
the response of this idealized system to ground motions applied in the x and y-directions is
identical to the response of the two-way asymmetric system to the same motions now applied in
the x and y -directions (Fig. 4.8). For nonlinear isolation systems, Sec. 4.3.2 suggests that this
one-way asymmetric idealized system should give a highly accurate estimate to the response.
Figure 4.9 compares median uro , computed as before, for a two-way asymmetric system (rectan-
gular plan as in Fig. 4.8a) with eccentricities ebx /r = 0.086 and eby /r = 0.05 and the idealized
system (Fig. 4.8b) with identical global properties and ebx /r = (0.086)2 + (0.05)2 = 0.1. The
deformations of the two systems appear to be identical, with discrepancies limited to 1.1% over
61
These results demonstrate that the response of a two-way asymmetric system is predicted
accurately by a one-way asymmetric model with the same magnitude of eccentricity. Therefore,
the results and conclusions presented in subsequent sections, while based on the idealized, one-
way asymmetric system, are applicable to any two-way asymmetric system with eccentricity
The median deformations uro (Eq. 4.13) of the asymmetric system and ubo of its corresponding
symmetric system – both systems subjected to the same bidirectional ground excitation – are
uro
ûro = (4.19)
ubo
a measure of the increase of deformation due to torsion. Results are presented over the ranges
of global properties (Tb , η, Ωθz and eb /r) defined in Sec. 4.3.1, for the idealized, rectangular-
plan system of Fig. 4.5b with a one-way, x-direction eccentricity eb /r. The subscript x in ebx /r
unimportant. The median deformation is computed for the LMSR ensemble with a median
PGV of 35 cm/s for the strong component of excitation, by the procedure of Sec. 4.2.6.
Figure 4.10 presents the median deformations uro and ubo and median deformation ratio
as a function of Tb for several different normalized strengths η and fixed Ωθz = 1.25 and
eb /r = 0.1. Included are the deformations of a linear system with η = 0. The deformation uro
in Fig. 4.10a of nonlinear systems displays smoother variation with Tb compared to the linear
system (η = 0). In Fig. 4.10b, the mostly unsystematic variations of ûro with Tb and η fall
between 1.05 and 1.15, implying that for an eccentricity eb /r of 0.1, plan asymmetry leads to
a 5 to 15% increase in the deformation uro at a distance r from the CM, over the deformation
When the results for uro , ubo and ûro are presented as a function of normalized strength
η for different values of Tb , the deformations of the asymmetric and corresponding symmetric
system are seen to be similar, decreasing monotonically with increase in strength (Fig. 4.11a),
62
30 1.5
(a) Symmetric (b) η=0
Asymmetric 0.25
25 0.5
1.4
1.0
20 η=0 1.5
ubo , uro (cm)
1.3
0.25
ûro
15
0.5 1.2
10
1.0
1.1
5 1.5
0 1
0 1 2 3 4 0 1 2 3 4
Tb (sec) Tb (sec)
Figure 4.10: (a) Median asymmetric system deformations uro and deformations ubo of a
corresponding symmetric system, and (b) median deformation ratio ûro , both plotted against
Tb ; Ωθz = 1.25 and eb /r = 0.1
20 1.5
(a) Symmetric (b) Tb =1.5 s
Asymmetric 2.5 s
1.4 3.5 s
15
ubo , uro (cm)
1.3
ûro
10
1.2
5
1.1
0 1
0 0.5 1 1.5 0 0.5 1 1.5
η η
Figure 4.11: (a) Median asymmetric system deformations uro and deformations ubo of corre-
sponding symmetric system for Tb = 2.5 sec, and (b) median deformation ratio ûro for varying
Tb , both plotted against strength η; Ωθz = 1.25 and eb /r = 0.1
and the deformation ratio ûro varies unsystematically, with weak dependence on strength
(Fig. 4.11b).
Figure 4.12 shows the median uro against frequency ratio for several values of η, eb /r =
0.1, and Tb = 2 and 4 and seconds; wherein the most relevant range of 1.0-1.3 for Ωθz is indicated.
The deformation ubo of the corresponding symmetric system, unaffected by frequency ratio, is
included for reference. The median deformations of the linear asymmetric system, included
63
(a) Tb = 2.0 sec (b) Tb = 4.0 sec
30
Symmetric
Asymmetric η=0
25
20
ubo , uro (cm)
η=0
15 0.25
0.25 0.5
10
0.5 1.0
5 1.0
1.5
1.5
0
0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5
Ωθz Ωθz
Figure 4.12: Median deformations uro and ubo of the asymmetric and corresponding symmetric
system, respectively, plotted against frequency ratio Ωθz ; eb /r = 0.1, (a) Tb = 2 sec, and (b)
Tb = 4 sec
for comparison, display a slight dip in the neighborhood of Ωθz = 1. However, in a nonlinear
system these deformations are essentially independent of the frequency ratio, even outside the
range most relevant to base-isolated buildings (Fig. 4.12), verifying our observations of the
response histories (Sec. 4.2.3). This suggests that intentionally increasing the relative torsional
stiffness of the system, as is sometimes done in design, is unlikely to have much effect on the
response. This contradicts the conclusion of [40] that a decrease in frequency ratio (with all
other properties fixed) leads to increased corner deformation. Figures 4.10-4.12 together leave
the impression that the increase in deformation in an asymmetric system, remarkably, depends
only weakly on each of Tb , η, and Ωθz . Developing a design factor to estimate the deformation
Since all results so far were for fixed eccentricity, Fig. 4.13 presents the variation of median
uro and ubo , and ûro against normalized eccentricity eb /r for a system with Tb = 2.5 seconds,
Ωθz = 1.25, and several values of η. Eccentricities eb /r up to 0.2, beyond the range identified
earlier, are considered, to admit special design situations that we may have missed. Figure 4.13a
shows that uro increases with increasing eccentricity, and suggests that the increase is roughly
linear, which Fig. 4.13b confirms, except perhaps for large values of η. This figure also shows
that the increase of ûro with eccentricity varies only slightly with normalized strength.
64
20 1.5
(a) Symmetric (b) η=0.25
Asymmetric 0.5
1.4 0.75
15 1.0
η=0.25
1.5
ubo , uro (cm)
1.3
0.5
ûro
10
0.75
1.0 1.2
5 1.5
1.1
0 1
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2
Eccentricity ebx /r Eccentricity ebx /r
Figure 4.13: (a) Median asymmetric system deformations uro and deformations ubo of corre-
sponding symmetric system and (b) median deformation ratio ûro , both plotted against eb /r;
Tb = 2.5 sec and Ωθz = 1.25
An equation for the deformation ratio ûro (Eq. 4.19) is now developed by regression analysis
of the data presented in Sec. 4.4; this equation multiplied by the deformation ubo (Sec. 3.7) of
a corresponding symmetric system will provide an estimate of the resultant deformation uro of
The regression equation for ûro is required to be linear upon logarithmic transformation,
consistent with Eq. 3.14a for ubo (Appendix 4B). The chosen regression equation should also
reduce to ûro = 1 for symmetric systems (eb /r = 0) and approach this limit continuously. To
satisfy this theoretical requirement, the equation for ln (ûro ) cannot vary linearly with ln (Tb ),
ln (η) or ln (Ωθz ), which is permissible since the median of ûro depends only weakly on each
of these properties (Sec. 4.4), and supplementary regression analyses showed the coefficients
The variation of ûro with normalized eccentricity eb /r (Fig. 4.13) suggests ln (1 + eb /r)
be included in the regression equation; a factor of 1 is added so that this term reduces to zero
when the eccentricity is zero. Because the other global properties are not included as variables
65
(a) Tb = 2.0 sec (b) Tb = 4.0 sec
20
exact median
design equation
15 η=0.25
η=0.25
0.5
uro (cm)
10 0.5 0.75
0.75 1.0
1.0
1.5
1.5
5
0
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2
Eccentricity eb /r Eccentricity eb /r
Figure 4.14: Design equation for uro (Eq. 4.22) compared to exact median computed by
nonlinear RHA to LMSR ground motion ensemble with median PGV = 35 cm/s, for systems
with Ωθz = 1.15; (a) Tb = 2 sec, (b) Tb = 4 sec
and, by omitting the constant coefficient, the theoretical requirement that ûro = 1 when eb /r =
0 is satisfied. This model is shown to be suitable in Appendix 4B, and the value of β1 resulting
A more general model to estimate corner deformation, which includes the effect of distance
from the CM, will be introduced in the next section. For consistency with this general model,
β1 is rounded to 0.88, which for small eccentricities (eb /r < 0.1) increases ûro by less than 1.2
66
20
exact median
design spectrum
15
uro (cm)
η=0.25
10
0.5
1.0
5 1.5
0
0 1 2 3 4
Tb (sec)
Figure 4.15: Design equation for uro (Eq. 4.22) compared to exact median computed by
nonlinear RHA to LMSR ground motion ensemble with median PGV = 35 cm/s, for systems
with Ωθz = 1.25 and eb /r = 0.1
A comparison of Eq. 4.22 with the exact median – over the LMSR ensemble – demon-
strates that the design equation generally follows the increase in deformation with increasing
eccentricity (Fig. 4.14). In this and later figures, the median deformation, computed by the
procedure of Sec. 4.2.6, assumes u̇go of 35 cm/s, as does the deformation given by the compa-
rable design equation. While overall an excellent fit, some discrepancy between Eq. 4.22 and
the median occurs because the design equation for ubo does not match its median exactly. This
can be seen by comparing the fit of uro for fixed eccentricity, presented in Fig. 4.15, to the
comparable fit for ubo in Fig. 3.11b. This comparison suggests that the design equation for uro
(Eq. 4.22) fits the median of uro as well as the design equation for ubo (Eq. 3.25) fits its median.
Regression models that were rejected even though they provided better fits to the median of
All that remains is to find a simple way to apply the preceding concepts to estimate the largest
deformation among all isolators, assumed to occur at the outermost corner of the building.
Unlike uro , which depends little on plan geometry, the deformation of the outermost corner
depends on its location, specified as a distance c and direction from the CM. However, a practical
method that is independent of plan geometry should be feasible if a reasonable estimate for the
67
3
b/d=1
1
2
3
y/r
0 CM ∞
−1
−2
−3
−3 −2 −1 0 1 2 3
x/r
Figure 4.16: Rectangular plans of various aspect ratios with corners on a circle of radius c/r
deformation of the outermost corner can be attained from the deformation maximized over a
c
uco = max ( ub (t) + r|θbz (t)| ) (4.23)
t r
This equation will give essentially the same deformation for all rectangular plans with
√
uniformly distributed mass, for two reasons. First, c/r = 3 regardless of plan aspect ratio
b/d (r = (b2 + d2 )/12 and c = (b2 + d2 )/2). Second, b/d has little influence on median
uro (Sec. 4.3.2), implying little influence on the CM deformations ub and rθbz . Equation 4.23
determines the largest deformation over a circle of radius c instead of the largest deformation
The accuracy of Eq. 4.23 is evaluated for rectangular plans with different aspect ratios in
Fig. 4.17, which compares the median deformation computed by Eq. 4.23 to the exact median
of corner deformation, largest among the four corners. Because aspect ratio has little influence
on the CM deformations, the approximate (Eq. 4.23) and exact values of corner deformation
are both computed from the CM deformation ub and rotation rθbz of the idealized system with
68
20 11.5
(a) uro (b)
uco
b/d=1
15 η=0.25 b/d=2
11 b/d=3
b/d=∞
uco (cm)
uco (cm)
0.5
10 0.75
1.0
1.5 10.5
5
0 10
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15
Eccentricity eb /r Eccentricity eb /r
Figure 4.17: Median corner deformations uco for rectangular plan with various aspect ratios
compared to conservative estimate given by Eq. 4.23; uro included for reference, Tb = 2.5 sec
and Ωθz = 1.15: (a) several values of η, (b) η = 0.5
b/d = 2 (Fig. 4.5b) rather than a different model for each aspect ratio. The locations where the
exact corner deformations are sampled are meaningless for the idealized system but appropriate
for the systems they represent. A cluster of curves for each of several values of η appears in
Fig. 4.17a; in each cluster the exact corner deformations for different aspect ratios lie between uco
determined from Eq. 4.23 and uro . A single value of strength, η = 0.5, is the focus of Fig. 4.17b,
where median corner deformation is seen to be largest for plans with b/d = 2 (though nearly
equal for b/d = 1, 2 and 3) and smallest for plans with b/d = ∞. Equation 4.23 overestimates
the deformation by at most 2% when b/d = 2 and 5% when b/d = ∞. Therefore, deformation
maximized over the circle of radius c (Eq. 4.23) should estimate the corner deformation with
sufficient accuracy.
In an asymmetric plan, the relative distance c/r from the CM to the outermost corner
√
varies according to its geometry; the smallest possible c/r is 2 for a circular plan, and no
theoretical maximum exists. However, c/r generally falls within a narrow range; for example,
for Plans A1, A2, and A3 (Fig. 4.5a), c/r = 1.81, 2.05, and 1.83, respectively. To examine how
deformation varies with relative distance from the CM, uco was computed for different values of
c/r by Eq. 4.23 from ub (t) and rθbz (t) of the idealized system. The median – over all excitations
in the ensemble – of uco is plotted as a function of c/r (Fig. 4.18) for a system with Ωθz = 1.15,
69
(a) Tb = 2.0 sec (b) Tb = 4.0 sec
20
exact median
design equation
η=0.25
η=0.25
15
0.5
uco (cm)
0.5
0.75
10 0.75 1.0
1.0 1.5
1.5
5
0
1 1.5 2 2.5 1 1.5 2 2.5
Distance to Corner c/r Distance to Corner c/r
Figure 4.18: Design equation for uco (Eq. 4.25) compared to exact median computed by
nonlinear RHA to LMSR ground motion ensemble with median PGV = 35 cm/s, for systems
with Ωθz = 1.15 and eb /r = 0.15; (a) Tb = 2 sec, (b) Tb = 4 sec
eb /r = 0.15, and varying Tb and η; dot markers indicate where c/r was sampled.
With these results, the pieces are in place to develop an equation to estimate deformation
at the outermost corner. To summarize, the equation will be based on regression of uco for dif-
ferent c/r, which is the deformation maximized over a circle of radius c, shown to conservatively
estimate the deformation at the outermost corner as a function of its distance c from the CM.
To extend the present regression model (Eq. 4.21) to include corner distance, we hypothesized
c eb
that ln (ûco ) varies linearly with ln (1 + r r ). This particular model gave nearly the best fit
among regression models considered (see Appendix 4B for details). Linear regression on re-
sponse data for ln (ûco ), shown by dot markers in Fig. 4.18, resulted in a regression coefficient
Multiplying this by the deformation ubo of a corresponding symmetric system (Eq. 3.25) gives
70
Table 4.1: Corner deformations for asymmetric Plans A1-A3 computed by nonlinear RHA
and estimated by design equation (Eq. 4.25)
Plan Tb (s) η eb/r c/r (Crn) uco (Crn) Eq. 4.25 % Err
A1 2.0 0.46 0.069 1.81 (1,3) 10.15 (2) 10.80 6.4
A1 4.0 0.46 0.069 1.81 (1,3) 11.53 (2) 12.32 6.9
A1 2.0 0.92 0.069 1.81 (1,3) 7.24 (2) 7.73 6.8
A1 4.0 0.92 0.069 1.81 (1,3) 7.38 (3) 8.82 19.5
A2 2.0 0.46 0.108 2.05 (1,4) 10.91 (3) 11.61 6.4
A2 4.0 0.46 0.108 2.05 (1,4) 12.22 (3) 13.24 8.4
A2 2.0 0.92 0.108 2.05 (1,4) 7.95 (3) 8.32 4.7
A2 4.0 0.92 0.108 2.05 (1,4) 8.79 (3) 9.49 8.0
A3 2.0 0.46 0.057 1.83 (1) 10.23 (2) 10.62 3.8
A3 4.0 0.46 0.057 1.83 (1) 11.39 (2) 12.12 6.4
A3 2.0 0.92 0.057 1.83 (1) 7.09 (2) 7.61 7.3
A3 4.0 0.92 0.057 1.83 (1) 7.86 (1) 8.68 10.4
written out here in its entirety (Eq. 4.25b) for completeness. Alternatively, the simplified
equation for ubo (Eq. 3.26) could be used in Eq. 4.25a. Note that (by design) Eq. 4.25a reduces
to the equation for uro (Eq. 4.22) when c/r = 1, and to ubo when eb /r = 0. Figure 4.18
demonstrates that Eq. 4.25 (dashed line) is a good fit to the median of uco ; however, this fit is
not as good as Eq. 4.22 for uro , as discrepancies increase with distance from the CM.
Since the design equation for corner deformation (Eq. 4.25) is developed from the response of
the idealized, rectangular-plan system, the applicability of this equation to various asymmetric-
plan systems is evaluated next. For two different isolation periods (Tb = 2 and 4 seconds) and
strengths (µ = 0.05 and 0.1), Plans A1-A3 (Fig. 4.5a), each with a different eccentricity eb /r
and outermost corner distance c/r, are subjected to the LMSR ground motion ensemble. With
assumed median PGV of 35 cm/s and ωd = 3.05 for the strong-component ensemble, the
For each plan, the exact deformation at each corner (as numbered in Fig. 4.5a) is the
median over the deformations at that corner, computed from nonlinear RHA for each excitation.
Table 4.1 lists the largest exact corner deformation (among all corners), the value estimated
71
by Eq. 4.25, and the percent discrepancy (% Err) in the latter. A positive discrepancy signifies
that Eq. 4.25 gives a conservative estimate of the largest corner deformation. Also noted are
the outermost corner that enters in the estimate of Eq. 4.25 and the corner that actually sees
Table 4.1 reveals that Eq. 4.25 tends to be conservative, which is not surprising since
the design equation is based on deformation maximized over the circle of radius c while the
exact medians are determined at the corner locations. Another source of conservatism is that
the deformation is not largest at the outermost corner, and the distance in Eq. 4.25 is over-
estimated. However, Eq. 4.25 is typically within 10% of the exact deformation, which is ideal
for an estimate. Successful development of a procedure that generally predicts the median
deformation within 10% was possible specifically because the eccentricities (representative of
An equation in the International Building Code (IBC) [1] increases the design displacement due
to torsion, according to
12e
DT D = DD 1+y 2 (4.26)
b + d2
where DT D and DD are the design displacements with and without torsion, respectively, y is
the distance from the CR (not the CM) to the outermost isolator in the direction perpendicular
to the applied load, and b and d are the longest and perpendicular to longest plan dimensions.
This equation increases DD by the rotational deformation resulting from a static torque equal
to DD kef f × eb applied about the CR, where kef f is the effective (equivalent-linear) stiffness at
Equation 4.26 implies a deformation ratio equal to DT D /DD . Because (b2 + d2 )/12 is
an estimate for r 2 (exact for a rectangular plan), the deformation ratio by this equation is
approximately (1 + yr re ), which is similar to Eq. 4.24. The two equations differ by the coefficient
0.88, and according to their single direction (y and e) or vector magnitude (c and eb ) variables.
Specifically, the code equation gives the deformation ratio resulting from loading in a principal
72
direction, while Eq. 4.24 gives the deformation ratio that results from bidirectional excitation.
However, the two deformation ratios can be compared directly, because an increase in the
deformation (estimated by Eq. 4.24) implies equivalent increases in the component deformations
The deformation ratios given by the code equation (Eq. 4.26) and Eq. 4.24 are compared
for Plans A1-A3. These data imply a 7.7, 14.9 and 6.1 percent increase in isolator deformation
due to torsion by the code equation, compared to a 10.9, 19.2, and 8.7 percent increase by
Eq. 4.24, for Plans A1-A3 respectively. The torsional increase in deformation by the code
equation is about 30% smaller than that by Eq. 4.24, but this underestimation is less significant
However, the IBC gives an inadequate estimate of DD in a symmetric system (Sec. 3.9).
It was shown that, for unidirectional excitation, equivalent-linear methods used by the IBC un-
derestimate DD by up to 50% compared to nonlinear RHA; and IBC’s at most 4.4% increase in
DD to account for bidirectional excitation is smaller than the 13% increase determined by non-
linear RHA for bidirectional excitation. All factors combined, the IBC grossly underestimates
the total deformation of the asymmetric system for a given design spectrum.
4.6 Conclusions
This investigation to estimate the median value of the peak isolator deformation in an asymmetric-
1. The median – over the ground motion ensemble – of any normalized deformation of interest
depends primarily on the following global properties of the system, which are attained
by summing over individual isolators: the isolation period Tb , the normalized strength η,
the torsional to lateral frequency ratio Ωθz and the normalized stiffness eccentricity eb /r.
This was achieved by defining the normalized strength as the system strength ÷ ωd u̇gyo ,
where u̇gyo is the peak ground velocity of the strong component of motion and ωd is the
corner frequency separating the velocity and displacement-sensitive regions of the median
73
spectrum for the strong-component ensemble.
2. Because isolators tend to be distributed at regular intervals, and their stiffnesses and
• uro (the deformation maximized over a circle of radius r from the CM) is only
weakly dependent on the system geometry and the planwise layout of isolators; thus,
identical to those of the actual asymmetric plan is within 2% of the median of uro
• the frequency ratio varies over a narrow range (1-1.3) and the eccentricities tend to
be small (< 0.1), giving focused ranges of these properties for base-isolated systems,
an axis rotation that does not alter the other global properties, allowing results for
3. The median ratio ûro (of deformation uro of the asymmetric system to the deformation
strength η, and frequency ratio Ωθz , and increases with eccentricity eb /r. Thus, a simple
equation (Eq. 4.21) that depends only on eccentricity was fit to the data for ûro , which
can be combined with an equation to estimate the deformation of the symmetric system
(Eq. 3.25 or Eq. 3.26) to obtain Eq. 4.22 for the deformation uro of the asymmetric system.
4. To extend these concepts to estimate the peak deformation in systems with different plan
geometry, which occurs at one of the corners, it was shown that for rectangular plans, the
deformation maximized over a circle of radius c from the CM provides an excellent, only
from the CM. Based on this result, the equation to estimate ûro was generalized, resulting
in Eq. 4.25, which estimates the peak corner deformation as a function of its distance from
74
the CM. For three asymmetric plans, this design equation was always conservative, but
generally within 10% of the exact median value determined by nonlinear response history
5. The percent increase in deformation due to torsion determined by the IBC equation is
about 30% smaller than the increase found in this investigation by nonlinear response his-
tory analysis; this underestimation is relatively unimportant given that the contribution
system given by the IBC can be unconservative by up to 50% compared to the deforma-
tion determined by nonlinear response history analysis; and this deformation increases
by at most 4.4% to account for bidirectional excitation, smaller than the 13% increase in
for a given design spectrum, the IBC grossly underestimates the peak isolator deformation
of an asymmetric system.
75
Appendix 4A: Maximization of ur (t)
1/2
ur (t) = urx (t)2 + ury (t)2
1/2
= (ubx − ry θbz )2 + (uby + rx θbz )2
1/2
= u2bx + u2by + 2rx θbz uby − 2ry θbz ubx + r 2 θbz
2
(4A.1)
By the method of Lagrange multipliers, ur is maximized subject to the constraint of Eq. 4.10.
To simplify the math, we maximized the function u2r . In this method, the gradients of both
u2r and r 2 (Eq. 4.10) with respect to coordinates rx and ry are computed and substituted into
2
2rx θbz + 2uby θbz = λ2rx
2
2ry θbz − 2ubx θbz = λ2ry
Substituting these equations into the constraint equation (Eq. 4.10) and solving for λ results
in:
θbz 2
2
λ = θbz ± (u + u2by )1/2
r bx
and
uby ubx
rx = ± r and ry = ∓ r (4A.2)
ub ub
where ub , the magnitude of deformation at the CM has been substituted for (u2bx +u2by )1/2 . These
solutions for rx and ry imply that the location of the maximum on the radius r is determined
by the relative values of ubx and uby . One pair of solutions corresponds to a minimum while
the other corresponds to a maximum. Substituting Eq. 4A.2 into Eq. 4A.1 results in:
2 1/2
ur = u2b ± 2ub rθbz + r 2 θbz
76
Obviously the maximum results from taking ± to match the sign of θbz , and this equation
simplifies to
ur = ub + r|θbz | (4A.3)
(Eq. 4.9), for which, recall, deformations were evaluated at x = ±r, locations perpendicular to
the direction of deformation. In the more general case, the x-axis is replaced by a plane through
the CM perpendicular to the direction of instantaneous deformation ub (t), which intersects the
circle of radius r at two points, one being a maximum (ub + r|θbz |) and the other being a
minimum (ub − r|θbz |). Clearly, Eq. 4A.3 reduces to Eq. 4.9 for unidirectional excitation if the
This appendix discusses regression models that were considered in developing design equations
for ûro and ûco , building upon concepts and terminology introduced in Appendix 3A.
Because the response data is sampled from a lognormal distribution, a function that is linear
etc. – gives a best estimate of the median response (Eq. 2.1). Therefore, regression models for
ûro and ûco were restricted to be of this form, and relaxing the requirement did not lead to any
improvements.
for ûro , equal to ūro ÷ūbo , is identical to the ratio of regression equations for ūro and ūbo (assumes
regression models for ūbo , ūro and ûro each contain the same regressors). This suggests that
regression analysis of the deformation ratio ûro , which proves to be simpler and more convenient,
gives identical results to distinct regression analysis of the normalized deformations ūro and ūbo ,
for which dispersion is minimized. The corresponding argument can also be made for the peak
A final consideration, both deformation ratios ûro and ûco should reduce to 1, i.e, the
77
peak deformation reduces to that of a symmetric system as the eccentricity approaches 0.
The model for ûro (Eq. 4.20) can be simplified to one that depends only on eccentricity eb /r
by eliminating other regressors. Although the normalized equation of motion (Eq. 3.9) im-
plies that the median value of uro depends on the global parameters Tb , η, Ωθz and eb /r, the
theoretical limit will not be satisfied if the model for ln ûro contains regressors ln Tb , ln η, or
ln Ωθz . Figures 4.10-4.12 demonstrated that the median of ûro is only weakly dependent on
each parameter Tb , η and Ωθz , but it can also be shown conclusively that the corresponding
Direct regression analysis of ln ûro implied that ln Tb and ln η were significant. However, to
show that these regressors can be eliminated, we instead compared the coefficients for ln Tb and
ln η determined from separate regression of ln ūro and ln ūbo . This regression analysis utilized a
data set where Tb and η are varied (16 values of Tb by 5 values of η), Ωθz = 1.15 and eb /r = 0.1
ln ūro = β0 + β1 ln Tb + β2 ln η (4B.1)
where results are given in Table 4B.1. Recall that from a statistical perspective, the coefficients
β determined from regression analysis are just estimates, but their true values are likely to fall
within a 95% confidence interval (Appendix 3A). If the coefficients β1 and β2 (corresponding
to ln Tb and ln η) for ln ūro fall within the confidence intervals of the same coefficients for ln ūbo ,
or vice versa, their difference can be considered statistically insignificant. The 95% confidence
interval is about 2 × the “std error” on either side of the estimated coefficient. Since β1 for
ūro (Table 4B.1a) differs from β1 for ūbo (Table 4B.1b) by slightly more than the “std error”,
and β2 are nearly identical for ūro and ūbo , the corresponding coefficients are within confidence
intervals.
Note that the coefficients for ln Tb and ln η in Table 4B.1 differ from those in Appendix 3A
(Table 3A.1) because a smaller data set was used for analysis of asymmetric-plan systems to
limit computation time, and the regressor (ln η) was dropped to make the argument cleaner. The
78
Table 4B.1: Significance of Tb and η (excerpts of R Project output)
regression equation for ūbo in Chapter 3 (Eq. 3.14a) is more reliable because it was developed
with a much larger data set; hence the advantage of analyzing ûro and combining with the
The regressor ln Ωθz was eliminated by a similar exercise. Since the deformation ubo of the
symmetric system is independent of Ωθz , it is only necessary to test whether the deformation
uro of the asymmetric-plan system depends on Ωθz , via the following model:
The response data included Tb varied over 3 values, η varied over 4 values, and Ωθz varied over 8
values within its practical range (Fig. 4.12). For this analysis, the t-value for β3 corresponding
to ln Ωθz is small enough (Table 4B.2) that the probability is high that β3 is zero; thus ln Ωθz
79
Table 4B.2: Significance of Ωθz (excerpts of R Project output)
ln ūro = β0 + β1 ln Tb + β2 ln η + β3 ln Ωθz
Coefficients:
Estimate Std.Error t-value Pr(>|t|)
(Intercept) 0.92507 0.0291407 31.75 <2e-16 ***
lnTb -1.88116 0.0290337 -64.79 <2e-16 ***
lnEta -1.52175 0.0148070 -102.77 <2e-16 ***
lnOmega -0.00084 0.0885179 -0.01 0.992
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Once these three regressors had been eliminated, possible regression models for ûro were
limited. The model of Eq. 4.20 included the remaining parameter eb /r and satisfied the lim-
iting behavior (ûro → 1 as eb /r → 0). Regression analysis of this model (Eq. 4.20) with a
comprehensive data set (including variation of Tb over 5 values, η over 5 values, and eb /r over
10 values equally spaced from 0 to 0.15) led to the regression coefficient of 0.756 (Table 4B.3a).
The R2 statistic, which is about 0.6, does not seem high compared to other analyses where it
was above 0.9. However, due to the scatter in ûro as a function of eb /r (Fig. 4.13b), a model
To demonstrate that it is usually possible but not necessarily desirable to achieve incre-
mental improvement in the regression equation, one last regression model for ûro is considered.
In this model, Tb and η are included in such a way that the theoretical limits are still satisfied:
Resulting analysis with this model (Table 4B.3b) compared to the model of Eq. 4.20 (Ta-
ble 4B.3a) suggests that one of the additional coefficients (β2 ) is significant, and demonstrates
a small improvement in the R2 statistic (0.6072 versus 0.5941). Ultimately, we must make a
judgment call; are the improvements large enough to justify the increased complexity of the
equation? Since the actual coefficients β2 and β3 are small, and it has already been demon-
80
Table 4B.3: Variation of ûro with eccentricity (excerpts of R Project output)
strated that ûro does not vary systematically with Tb and η (Figs. 4.10b and 4.11b), including
Whereas the additional deformation due to rotation (rθbz ) in uro (Eq. 4.13) was shown to be
accounted for by eb /r, we anticipate that the additional deformation due to rotation (cθbz )
in uco (Eq. 4.23) is accounted for by eb /r multiplied by c/r. In this sense, Eq. 4.24 is the
most logical selection of regression model for ûco . The resulting regression analysis with this
model – labeled Model 1 – (Table 4B.4a), is compared to three other models – Models 2-4
– that were also considered (Table 4B.4b-d). Model 2 (Table 4B.4b), which also contained
only one coefficient and thus arguably no more complex, showed only slight and presumably
inconsequential improvement in its R2 statistic. Analysis suggested that Model 4 (Table 4B.4d),
81
Table 4B.4: Testing models for ûco (excerpts of R Project output)
82
similar to the second model for ûro (Table 4B.3b), might be worth incorporating based on
(functions of Tb and η). Once again, we rejected these arguably improved models, preferring
tors)
83
Q lateral characteristic strength of the isolation system (summed over
all isolators)
directions
ubo /u∗y
corner
ū ¨gy
¨gx , ū ground acceleration normalized by peak ground velocity
u+ry , u−ry peak deformation in the y-direction at ±r from CM, where + is the
stiff side and − is the flexible side for positive eccentricity ebx /r
from CM
u∗y Q/kb
zxi , zyi dimensionless plastic variables governing yielding of the ith isolator
γi Qi /Q
85
5: ESTIMATING THE PEAK DISPLACEMENT OF A FRICTION
PENDULUM ISOLATOR
5.1 Introduction
Recall the procedure developed in Chapter 3, by which isolator deformations due to ground
motions characterized by a design spectrum with its intensity defined by the median peak
ground velocity (PGV) can be estimated. Based on rigorous nonlinear analysis of the isolated
system for an ensemble of ground motions representative of the spectrum, this procedure is
effective because the governing equations for the system were rewritten such that the normalized
Initially developed for a system isolated with lead-rubber bearings, the nonlinear pro-
cedure is general enough to apply to other types of isolation systems. Specifically, friction
pendulum (FP) isolators can also be modeled by a nonlinear force-displacement relation, re-
quiring only a different value for the yield displacement [2]. The yield displacement for FP
isolators is on the order of 0.05 cm [42], compared to 1 cm for lead-rubber bearings (Sec. 3.3.1).
The significant change in the yield displacement suggests that different design equations are
The objective here is simply to extend earlier techniques to develop design equations to
procedure to estimate this displacement, upon which the current IBC is based, is evaluated
The system analyzed is a rigid mass m supported by a single FP isolator, whereby the weight w
of the system rests on an articulated slider, which slides over a spherical surface in response to
earthquake excitation (Fig. 5.1a). Neglecting the slight rise of the mass on the spherical surface
86
fb
(a) (b)
µw w/R
m
ub (t) ub
R, µ
üg (t)
Figure 5.1: (a) Single FP isolator supporting a rigid mass, and (b) rigid plastic force-
displacement of the isolator.
for bidirectional excitation, where ub (ubx , uby ) is the slip in the isolator (x and y-components
of slip for bidirectional excitation), or displacement relative to the ground, and u̇b (u̇bx , u̇by ) the
relative velocity. The resisting force in Eq. 5.1 is composed of a pendular component directed
toward the centered position and a friction component that opposes instantaneous velocity;
thus, the isolator is characterized completely by the radius of curvature R of the spherical
surface and the friction coefficient µ of the sliding interface. The friction coefficient is known to
be sensitive to the sliding velocity, the pressure of the supporting mass, and viscous heating of
the slider over extended excitation [34, 43, 44, 45]; however, these effects are subtle enough that
the time variation of the friction coefficient need not be included in this otherwise simple model,
and varying axial load is irrelevant in a single isolator system. The simple rigid-plastic force-
displacement relation shown in Fig. 5.1b will be modified for the x and y-components when the
excitation is bidirectional, due to interaction of the friction force in the x and y-directions; that
is, the friction force in each direction depends on its velocity component (Eq. 5.1b).
response. However, the general plasticity model is convenient to implement, and can treat
the small deformation which occurs at the sliding surface prior to slip as a nonzero yield
displacement. This pre-slip deformation is on the order of 0.05 cm [42]. Thus, the force in the
87
FP isolator can be described by Eq. 3.19 with the postyield stiffness kb equal to the pendular
stiffness w/R, the strength Q equal to the friction force µw, and the yield displacement – which
With isolator forces represented by the plasticity model, the equations governing the mass
is easily treated by Eq. 5.2 with one of the components identically equal to zero. Equation 5.2
appears identical to the equations for a mass on a single lead-rubber bearing (LRB system)
(Eq. 3.20); the two differ only because they are modeled by different yield displacement or
deformation. The FP system is characterized by its isolation frequency ωb = g/R and friction
coefficient µ (Eq. 5.2). Solving Eq. 5.2 for a given excitation gives the peak displacement – in
any direction:
ubo = max (ubx (t)2 + uby (t)2 ) (5.3)
t
For meaningful statistical analysis of the response to an ensemble of ground motions, the gov-
Identical to Eq. 3.22, this has been formulated by first dividing Eq. 5.2 by u∗y , introducing
ūbx = ubx /u∗y and ūby = uby /u∗y , the normalized displacements in the x and y-directions; and
then substituting the normalized strength into the right-hand side. The following previous
definitions apply: (1) a∗y (equal to the yield acceleration µg), (2) u∗y (Eq. 3.6):
Q a∗y
u∗y = = 2 (5.5)
kb ωb
88
and (3) normalized strength (Eq. 3.23):
a∗y µg
η= = (5.6)
ωd u̇gyo ωd u̇gyo
sensitive region of the median spectrum (Fig. 2.1) and u̇gyo is the PGV, both referring to
the strong component of motion, which is applied in the y-direction. Hence, the components of
ground acceleration in Eq. 5.4 have been normalized by the PGV u̇gyo , such that the magnitude
¨gx (=ügx /u̇gyo ) is strictly less than 1 and ū
of ū ¨gy (=ügy /u̇gyo ) varies from -1 to +1.
By virtue of Eq. 5.4, the essential benefits of normalization that were found in Chapter 3
¨gx and ū
for LRB systems also hold for FP systems. Specifically, the excitations ū ¨gy have been
normalized to a common intensity, such that their variability is limited to the variability inherent
in different realizations of a random process. Because of this normalization, the median of the
normalized displacement ūbo depends mainly on Tb and η, and its dispersion over many ground
To verify that the normalization is as effective for FP systems as it was for LRB systems
(Section 3.3.2), the dispersion of ūbo (Eq. 2.2) for FP systems is presented in Fig. 5.2 for
different values of Tb and η. The excitations are limited to unidirectional and are sampled
from the strong-component ensemble, a subset of the LMSR ensemble (Sec. 2.1). While the
dispersion remains small – between 0.4 and 0.6 – for the smaller values of η, which is consistent
with LRB systems (Fig. 3.5d), it escalates rapidly as η increases for reasons identified next.
Because an FP system is initially very stiff, the earthquake response of the system is
controlled by the peak ground acceleration if the friction force is large enough to prevent
slip. Thus, the PGV normalization, shown to be effective for LRB systems (Sec. 3.3.2), is
inappropriate if the friction force is so large that the isolator may or may not be activated by
different ground motions in the ensemble, depending on their peak ground acceleration; this
results in the sharp rise in dispersion at η = 1.5. However, an FP system is ineffective if the
isolator fails to slip, and such large normalized strength is irrelevant to practical application.
89
1.4 η=.25
0.5
1.2 0.75
1.0
Dispersion of ūbo
1 1.5
0.8
0.6
0.4
0.2
0
0 1 2 3 4 5
Tb (sec)
Thus, in this study, results for FP systems will be restricted to values of η no larger than 1.
When the preceding normalization is used to generate predictions of the median normalized
response and ultimately the median response for a specified intensity of ground motion, its full
benefit is realized. Median response spectra for the normalized displacement ūbo , and the actual
displacement ubo as a function of PGV, are constructed for the FP system as follows:
1. Determine the peak value of the normalized displacement ūbo of the system by nonlinear
response history analysis (RHA) of Eq. 5.4 for a selected value of Tb and η, and a selected
2. Compute the median normalized displacement (Eq. 2.1) over all ground motions for the
3. Convert the normalized displacement to the actual displacement by the following equation:
where u∗y (Eq. 5.5) is a function of η, ωd and PGV (Eq. 5.6); this gives the displacement
90
(a) Unidirectional Excitation
2
10 15
1 design equation
10
10
0 η=0.25
10
η=0.25
0.5 0.5
−1
5
10 0.75 0.75
1.0 1.0
−2
10 0
0 1 0 1 2 3 4 5
10 10
Tb (sec) Tb (sec)
(b) Bidirectional Excitation
2
10 15
1
10
10 η=0.25
0
10 η=0.25 0.5
0.5 5 0.75
−1
10 0.75 1.0
1.0
−2
10 0
0 1 0 1 2 3 4 5
10 10
Tb (sec) Tb (sec)
Figure 5.3: Design equations for normalized displacement ūbo and displacement ubo for (a)
unidirectional excitation (Eqs. 5.8b and 5.9) and (b) bidirectional excitation (Eqs. 5.10 and
5.11), compared to their exact medians by RHA due to the LMSR ensemble with PGV of 35
cm/s.
Determined for the LMSR ensemble (Sec. 2.1), the median normalized displacement ūbo
(step 2 above) and median displacement ubo for a PGV of 35 cm/s (step 3) are plotted (solid
lines) in Fig. 5.3, as a function of Tb and for several values of η. Included are plots for both unidi-
¨gx = 0 in Eq. 5.4, (Fig. 5.3a) and bidirectional excitation (Fig. 5.3b).
rectional excitation, i.e., ū
As already achieved for LRB systems, design equations representative of the median displace-
ments over the range of Tb and η can be determined. Considering unidirectional excitation first,
the response data is presumed to fit the same form of equation used earlier (Eq. 3.14a), which
91
results in the following linear regression equation for the normalized displacement ūbo :
or
where the coefficients were estimated by the method of least squares. An equation for the
displacement of the isolator is then obtained by substituting Eq. 5.8b into Eq. 5.7 where u∗y
is given by Eqs. 5.5 and 5.6 with ωd = 3.05, which corresponds to Td = 2.06 seconds for the
Plotted as dashed lines alongside the exact medians in Fig. 5.3a, these equations (Eqs. 5.8b and
5.9) are shown to be good fits to their corresponding medians, thus suitable for analysis and
design applications. The coefficient -0.31 in the quadratic-ln (η) term is much larger than the
corresponding coefficient -0.08 in Eqs. 3.14b and 3.16 for LRB systems. Therefore, simplification
By similar methods, the following equations were fit to the normalized displacement ūbo
and
4.36 0.14 (−0.99−0.20 ln η)
ubo = T η u̇gyo (5.11)
4π 2 b
These design equations for bidirectional excitation (Eqs. 5.10 and 5.11) are shown also to be
Evident from the design equations and as expected, bidirectional excitation causes larger
displacement of the isolator compared to unidirectional excitation. This increase depends signif-
icantly on η, varying from approximately 20% to 38% as η increases from 0.25 to 1.0 (Fig. 5.4).
92
1.5
ubo(bidirectional) /ubo(unidirectional)
1.4
1.3
1.2
η=0.25
1.1 0.5
0.75
1.0
1
0 1 2 3 4 5
Tb (sec)
Figure 5.4: Ratio of design equations for ubo due to bidirectional and unidirectional excitation
(Eq. 5.10 ÷ Eq. 5.8b)
equations for both unidirectional and bidirectional excitation is insightful for comparing FP
systems and LRB systems. The displacement of an FP system is generally much smaller than
the deformation of an LRB system for comparable values of Tb and η (Fig. 5.5). This trend is not
surprising because an FP system, due to its reduced yield displacement, dissipates more energy
than an LRB system at comparable displacements/deformations of the isolator [2, Fig. 5.5], thus
reducing its displacement. The displacements of the FP system shown in Fig. 5.5a range from
clearly the relative difference becomes larger as η increases. However, this discrepancy decreases
for bidirectional excitation (Fig. 5.5b), because the FP-system displacements increase more (22
to 38%) than the LRB-system deformations (constant 13%, see Sec. 3.7) from unidirectional to
bidirectional excitations.
The International Building Code [1] and its predecessor, the Uniform Building Code, estimate
the design displacement of an isolation system from a linear design spectrum, with the properties
linear response spectrum to estimate the displacement of an FP system, similar to current code
procedure, is evaluated next. Given a response spectrum for unidirectional excitation, steps to
93
15 15
(a) FP systems (b)
LRB systems
η=0.25
ubo (cm) (unidirectional)
0 0
0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec)
Figure 5.5: Design equations for isolator displacement/deformation compared for FP systems
and LRB systems: (a) unidirectional excitation (Eq. 5.9 vs. Eq. 3.16) and (b) bidirectional
excitation (Eq. 5.11 vs. Eq. 3.25)
iteratively determine the equivalent-linear properties for a given nonlinear isolation system are
as follows:
1. Using initial guesses for the effective period Tef f and effective damping ζef f of the
2. Evaluate the isolator force as fbo = µw + (w/R)ubo , a specialization of Eq. 5.1 for unidi-
rectional excitation.
3. Update estimates of effective period: Tef f = 2π w/(kef f g), and effective damping:
4. Repeat steps 1-3 with updated values of Tef f and ζef f , until successive estimates of ubo
converge.
The displacement of the FP system is estimated by implementing the above steps for
the median spectrum of the strong-component ensemble (Fig. 2.1a). Spectral displacements for
damping ratios other than 5%, needed in the iterative procedure, are determined by additional
response history analyses. The procedure is repeated for a range of Tb and η, whereby for each
η, the friction force is determined from Eq. 5.6 with the PGV taken to be 35 cm/s. The isolator
94
10 50
(a) Nonlinear (b) η=0.25
Equiv-Linear 0.5
8 0.75
25 1.0
Discrepancy in ubo
6
ubo (cm)
0
4
−25
2
0 −50
0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec)
Figure 5.6: (a) Comparison of median displacement determined by nonlinear and equivalent-
linear methods for a system with η = 0.5, and (b) discrepancy in displacement computed by
equivalent-linear method relative to nonlinear method, for several values of η.
displacement estimated by this equivalent-linear procedure is plotted in Fig. 5.6a for a system
with η = 0.5, alongside the exact median displacement determined from nonlinear RHA. In this
case, the equivalent-linear estimate is close to the exact value for short isolation periods, but
To represent a more diverse range of FP systems, the percent discrepancy in the equivalent-
linear estimate relative to the exact value is presented in Fig. 5.6b for several values of η. Based
on this figure, the equivalent-linear procedure overestimates the displacement (positive percent
discrepancy) for very short isolation periods at the low end of the practical range, but under-
estimates it for longer periods representative of practical systems, by up to 30%. While similar
for the three smaller values of normalized strength, the percent discrepancy for η = 1.0 displays
a very different trend, because the FP system tends to be affected by its initial rigidity at
large normalized strengths as discussed earlier. In spite of this, for most isolation periods and
equivalent-linear procedure.
95
5.4.1 Extension to Bidirectional Excitation
Although factors are included to account for bidirectional excitation in the code dynamic pro-
cedures, which include site specific response spectrum analysis and response history analysis,
no change to the displacements specified in the static lateral response procedure is required.
Bidirectional excitation is indirectly accounted for in the larger of the two code specified lev-
els of design. The design displacement, corresponding to the design basis earthquake (DBE),
“act(s) in the direction of each of the main horizontal axes of the structure”, while the maximum
displacement, corresponding to the maximum capable earthquake (MCE), “act(s) in the most
critical direction of response” [1]. Thus, it appears that the maximum displacement reflects a
spectrum increase for bidirectional excitation; however, since the MCE is also a stronger earth-
quake, it becomes difficult to separate the two effects. Nevertheless, the design displacement is
used for most applications, while the applications of the maximum displacement are limited to
stability check and vertical load testing of the isolation system, and allowance for components
5.5 Conclusions
A procedure to estimate the displacement of an FP system has been developed based on non-
linear response history analysis, as an extension of earlier procedures for LRB systems, leading
1. The median – over an ensemble of ground motions – of the normalized displacement ūb
depends only on the isolation period Tb and normalized strength η. This was achieved
by defining the normalized strength as the system strength ÷ ωd u̇gyo , where u̇gyo is
the peak ground velocity and ωd is the transition frequency separating the velocity and
the intention of minimizing the dispersion of the normalized displacement and ultimately
allowing a higher confidence estimate of the peak displacement. For FP systems, the
dispersion of normalized displacement was shown to be small except for large values of
96
normalized strength, for which the isolator either failed to slip consistently or its movement
was quite limited. However, such large normalized strength is not relevant for practical
application, because the isolation system is ineffective unless the isolator slips.
3. An equation, dependent on isolation period and normalized strength, was fit to the exact
equation for the actual displacement of the FP system. Comparable design equations
for unidirectional and bidirectional excitation predict that the peak displacement in any
on η, than the peak displacement due to unidirectional excitation. The design equations
also predict that for the same isolation period and normalized strength, the displacement
of an FP system is smaller than the deformation of an LRB system, because more energy
is dissipated.
4. For a given spectrum, equivalent-linear procedures in the International Building Code were
97
Appendix 5A: Notation
of linear spectrum
tation
ū ¨gy
¨gx , ū components of ground acceleration normalized by peak ground ve-
locity
98
u̇gyo peak ground velocity in y-direction for bidirectional excitation
u∗y Q/kb
ing surface
99
6: MODEL FOR AXIAL-LOAD EFFECTS IN LEAD-RUBBER
BEARINGS
6.1 Introduction
The method developed in previous chapters to estimate peak isolator deformation, which as-
sumes a shear model of the isolated structure, neglects the overturning associated with lateral
vibration that causes the vertical or axial forces in the isolators to vary with time. Due to
concerns about the stability of rubber isolation bearings under large compressive loads and
their ability to withstand tensile loads, the peak – both maximum and minimum – axial forces
Axial loads have been observed to have a significant effect on the response of isolation
bearings. First, a correlation between lateral stiffness and axial load has been observed in
several types of rubber bearings. In tests of disparate bearings – for example, natural rubber,
high-damping rubber (HDR) and lead-rubber (LRB) bearings – with both dowelled and bolted
connections [14, 15, 16], the secant stiffness decreased with increasing axial load. As an excep-
tion, some HDR bearings showed increased stiffness or hardening at large shear strains that was
magnified by large axial loads [16]. The stiffness reduction appeared to be greatest in bearings
Rubber bearings have also been shown to soften in the vertical direction at large lateral
deformations. In pure tension, rubber bearings tend to cavitate, or form small cavities in the
rubber that blow out from negative pressure and link together to form cracks in the rubber
matrix. In one test, this occurred at tensile strains of about .0003, corresponding to tensile stress
of 1.59 MPa (230 psi) [11]. However, in recent projects, bearings under large lateral deformation
were jacked up 12 to 20 mm (0.5 to 0.75 inches) with no evidence of cavitation damage [13].
In other characterization tests, the vertical stiffness of three different HDR bearings decreased
to about 0.7 times the nominal stiffness at lateral deformations near one half their bonded
diameter [46].
Although less documented, the yield strength of an LRB bearing, or strength of the lead
100
core, has been observed to vary with axial load, such that a lightly loaded bearing may not
achieve its theoretical strength. This effect was first noted as an immediate decay in the lateral
force-deformation hysteresis area in response to a sudden drop in axial pressure [17]. While
this undesirable behavior is thought by some to be eliminated with proper confinement of the
lead core, it continues to be observed in current research. In a 3-story structure isolated with
LRB bearings and subjected to triaxial ground excitations, the bearings beneath heavily-loaded
interior columns showed much greater strength and energy dissipation than identical bearings
In spite of the preceding evidence, existing nonlinear models do not account for the inter-
action between axial loads and lateral/vertical response of the isolators. The objective of this
chapter, therefore, is to develop a model for isolators, specifically LRB bearings, that includes
this interaction, and a numerical implementation of this model for dynamic analysis. The
model is a nonlinear extension of a two-spring model developed from linear stability theory of
of the strength variation of the lead core with axial loads. All behaviors of the model, which
include variation of lateral stiffness and yield strength with axial load, and variation of verti-
cal stiffness with lateral deformation, are confirmed by unpublished experimental data. From
hereafter these behaviors are referred to as axial-load effects. Following development of the
algorithm to implement this bearing model, the response of the new model to a simple seismic
pulse is explored. Unfortunately this model, even without the optional strength variability,
is less suitable for HDR bearings, because it lacks hardening at large shear strains and other
In the stability analysis of multi-layer bearings, [36, 47], the bearing is treated as a continuous
composite system in which the steel layers do not deform, allowing prediction of the buck-
ling load and the effective lateral stiffness in the presence of axial load. The stability theory
resembles the linearized theory of an elastic column, but accounts for shear deformation by
considering rotation of the cross-section, which is independent of the lateral deflection [47].
101
δbz
P
P
δbz
Figure 6.1: Response of an axially-loaded multi-layer rubber bearing under shear deformation,
including tilting of the rubber layers, in response to (a) compression loading and (b) tensile
loading. Note that middle reinforcing layers tilt the most.
ub P
fb
PE hb
2 Hinge δbz
θ
)
hb
θ
)
kbo
s
Hinge
PE hb
2
Figure 6.2: Two-spring model of an isolation bearing in the undeformed and deformed con-
figuration.
Also predicted by the stability analysis [47], the multi-layer bearing under simultaneous lateral
and axial loading undergoes an additional vertical displacement beyond that due to material
axial flexibility. Demonstrated visually in Fig. 6.1, the additional displacement δbz , either com-
pressive or tensile depending on the corresponding axial load, is due to tilting of the middle
102
6.2.1 Approximate Force-Deformation Relation Based on Linear Two-Spring Model
simplified two-spring model of the bearing [36] that results in explicit force-deformation rela-
tions. The two-spring model (Fig. 6.2) is a composition of rigid tees connected by a rotational
spring, subdivided at top and bottom, and a shear spring with frictionless rollers at midheight.
The bottom plate is fixed and the top plate is constrained against rotation. Axial flexibility of
the bearing is included by an additional vertical spring in series (not shown in Fig. 6.2).
Assuming linear material behavior, the nominal shear stiffness of a multi-layer bearing,
GA GAs
kbo = = (6.1)
tr hb
where G is the shear modulus, A the cross-sectional area, and tr the sum thickness of the rubber
layers. In some cases, it is convenient to use modified area As = A(hb /tr ), where hb is the total
height of the bearing, which accounts for the undeforming steel layers. Similarly, the nominal
vertical stiffness of the bearing, i.e., stiffness of the additional vertical spring, is
Ec A Ec As
kbzo = = (6.2)
tr hb
this context, the term nominal (denoted by subscript ’o’) means absent axial-load effects.
If the shear stiffness of the two-spring model (Fig. 6.2) were infinite, the rotational stiffness
divided by hb would equal the conventional Euler buckling load PE = (π 2 /h2b )EIs . Here EIs is
1 hb
EIs = Ec I (6.3)
3 tr
where I is the conventional moment of inertia: πD4 /64 or Arb2 , and rb = D/4 is the bending
radius of gyration in terms of the bearing diameter D. Thus, the rotational stiffness, divided
103
With these simple linear constitutive relations, the equilibrium equations relating the
lateral force fb and axial (compressive) force P to the deformation s across the shear spring
and the rotation θ through the rotational spring (Fig. 6.2) are
fb − kbo s + P θ = 0 (6.4a)
fb hb − PE hb θ + P (s + hb θ) = 0 (6.4b)
which assume small rotation θ. The axial force P and the deformation v across the vertical
P − kbzo v = 0 (6.4c)
The kinematic equations relating the total lateral deformation ub and vertical deformation
ub = s + hb θ (6.5a)
ubz = v + δbz
hb 2
= v + sθ + θ (6.5b)
2
In Eq. 6.5b, ubz – positive in compression – is the sum of v, the deformation resulting from
axial flexibility of the bearing, and δbz , the additional vertical displacement that occurs in the
Analysis of the linear two-spring model originally presented in [36] is summarized in the
following steps:
1. The system of equations in s and θ (Eqs. 6.4a and b) is solved for the critical buckling
where PS = GAs = kbo hb . Equation 6.6 approximates the buckling load determined from
stability analysis of the multi-layer bearing [36, Eq. 8.12], with a reasonable assumption
that PE PS .
104
2. The values of the shear deformation s and rotation θ, found by solving the same system
of equations, are substituted into Eq. 6.5a, which, with some approximation, simplifies to
Again, the lateral stiffness kb in Eq. 6.7 for the two-spring model is a good approximation
to the stiffness derived from stability analysis of the multi-layer bearing [36, Fig. 8-4].
3. The values of s and θ, determined as described above, and axial deformation v from
Eq. 6.4c, are substituted into Eq. 6.5b. This leads to the following flexibility equation for
P (PS + P ) u2b
ubz = + (6.8)
kbzo PE hb
reduced to its final form by substituting Eq. 6.3 for PE . Thus, the lateral and vertical
force-deformation relations of the bearing are defined by the coupled equations 6.7 and
6.9. If axial-load effects are neglected (take axial load P = 0 in Eq. 6.7 and lateral
deformation ub = 0 in Eq. 6.9), these equations reduce to the familiar uncoupled linear
force-deformation equations with nominal stiffnesses kbo (Eq. 6.1) and kbzo (Eq. 6.2).
Lateral force-deformation curves for different axial loads are shown in Fig. 6.3a, with the
curve corresponding to no axial load (P = 0) representing the nominal stiffness. Each curve
is linear because P is constant. The reduction in lateral stiffness as the axial load increases
can be seen here as well as in Fig. 6.4. The stiffness variation is inconsequential for typical
design values of P/Pcr < 0.2. However, the stiffness rapidly approaches zero as the axial load
105
(a) (b)
0.4 1.2
0.3 P =0
0.2Pcr 1
0.2 0.4Pcr
Lateral force fb /Pst
0.8
Figure 6.3: (a) Lateral force-deformation as a function of P/Pcr and (b) vertical force-
deformation as a function of lateral deformation ub . Lateral and axial forces are normalized by
the static, or gravity load Pst .
Lateral to nominal stiffness ratio: kb /kbo
0.8
0.6
0.4
0.2
Consideration of the negative solution to Pcr (Eq. 6.6) leads to a critical load in tension,
hence the unconventional concept of tension buckling [13]. Because the two solutions to Pcr are
equal and of opposite sign, the bearing buckling behavior in tension mirrors that in compression.
Thus, the stiffness reduction of Eq. 6.7 is the same for both compressive and tensile loads P .
Written in flexibility form (Eq. 6.8), the vertical force-deformation relation shows two
distinct contributions to the vertical deformation, a material effect and a second order geometric
effect. The additional vertical displacement δbz due to geometry of deformation (second term
106
of Eq. 6.8) increases according to the square of the lateral deformation. Note that even with
no axial load on the bearing, the additional displacement δbz is nonzero, albeit small because
PE PS , as shown in Fig. 6.3b at P = 0 for different lateral deformations ub . The net effect of
this additional displacement is an overall softening of the bearing in the vertical direction, which
depends on the lateral deformation relative to the bending radius of gyration (Eq. 6.10). This
softening is also evident in the vertical force-deformation curves for different lateral deformations
(the curves are linear because lateral deformation is constant) shown in Fig. 6.3b, wherein the
slope of the curve for the laterally undeformed configuration (ub = 0) represents the nominal
vertical stiffness (Eq. 6.2). Physically, the softening occurs as a result of the tilting of bearing
reinforcing layers (Fig. 6.1), meaning the axial loads are resisted in part by shear.
The stability of any bearing is closely related to its geometry, represented by the first and
second shape factors S and S2 . Defined as the ratio of the loaded area to the force free area,
the shape factor S for a circular bearing = D/4t, where t is the thickness of a single rubber
layer. The second shape factor S2 is simply the bearing aspect ratio based only on the rubber
The compression modulus Ec , needed in the nominal vertical stiffness (Eq. 6.2), is a
Ec = 6GS 2 (6.11)
The effect of axial load on the lateral stiffness (Eq. 6.7) is more significant for low shape factor
bearings because Pcr (Eq. 6.12) is smaller, and hence the ratio of design load to Pcr is larger.
The theory and equations presented thus far, assuming rubber to be incompressible,
should be modified for shape factors S > 10 to include the effect of bulk compressibility [36].
107
In terms of the bulk modulus κ, the compression modulus is modified to [48]:
1 1 8GS 2
= 1+ (6.13)
Ec Ec κ
The approximate Eqs. 6.13 and 6.14 are valid for shape factors S ≤ 25, and can be rewritten
as
Ec EIs
Ec = EIs = (6.15)
1 + Kc S 2 1 + Ks S 2
where Kc and Ks depend on the ratio of the shear modulus to bulk modulus G/κ, which ranges
from 1/3000 to 1/2000 for typical values of G and κ. Replacing Ec by Ec and EIs by EIs in
Eqs. 6.12 and 6.10 leads to modified equations for Pcr and kbz :
πGASS2
Pcr = (6.16)
2 2(1 + Ks S 2 )
−1
kbz 3u2 (1 + Ks S 2 )
= 1 + 2 b2 (6.17)
kbzo π rb (1 + Kc S 2 )
6.3 Variation of Yield Strength with Axial Load
Experimental evidence [17, 18] of the bearing failing to achieve its full strength when lightly
loaded has been attributed to lack of confinement of the lead plug. In the words of Skinner
[3], “the nominal upper limit of hysteretic force ... should be achieved if there is no vertical
slippage of the plug sides and no horizontal slippage of the plug ends.” Although side slip can
be reduced by decreasing the spacing between steel layers, or using a lead plug slightly larger
than the undeformed cavity to be occupied, the overall slippage unavoidably depends on the
The preceding experimental observation has not, to our knowledge, been verified by
mechanical analysis. However, we have developed an empirical equation for the yield strength
as a function of the compressive load P based on experimental data (see Sec. 6.4):
Q = Qo 1 − e(−P/Po ) (6.18)
108
1.2
P = Po
1
0.8
0.6
0.4
0.2
−0.2
−5 0 5 10
Ratio of P/Po
where Qo is the nominal yield strength of the bearing, achievable with an adequate confining
pressure; and Po is the axial load corresponding to about 63% of nominal strength. In Fig. 6.5,
a plot of Eq. 6.18 shows that the strength declines quite rapidly for loads P below Po . The
strength exceeds 95% of its nominal value at loads P ≥ 3Po , therefore we recommend a minimum
of 3Po for the bearing design load Pst , although different scenarios will be explored. When the
bearing is in tension (P < 0), the effective yield strength is taken to be zero because the lead
In conjunction with earthquake simulator tests of a model three-story reinforced concrete build-
ing, prototype HDR bearings (from Bridgestone Corporation and Malaysian Rubber Products
Research Association or MRPRA) and LRB bearings (from Oiles Industory Co.) were tested
to determine their mechanical characteristics. Details of the experimental test program and
some results were published [11, 49]; but much of the test data, especially for MRPRA and
Oiles bearings, was never reported. Although these tests were performed over 10 years ago, the
Sinusoidal displacement-controlled tests of the bearing at a specified shear strain sequence were
repeated at varying axial loads; the postyield stiffness kb and strength Q were measured from
109
the recorded data (Appendix 6A). In reality, the bearing properties vary with shear strain;
thus, the nominal stiffness kbo (at zero axial load) and strength Qo (at the maximum applied
axial load) were observed at each strain level, and the critical load Pcr (Eq. 6.16) – based on
the observed nominal shear modulus Go – was computed at each strain level. Our proposed
equations for postyield stiffness kb (Eq. 6.7) and strength Q (Eq. 6.18) obviously ignore strain
dependence, but so do typical models for LRB bearings because the strain dependence is minor
and properties at large strains control the peak response of the system.
Figure 6.6 compares the normalized postyield stiffness kb /kbo versus P/Pcr observed in
these tests for the three different bearings with Eq. 6.7. For both the Bridgestone (Fig. 6.6a)
and MRPRA bearings (Fig. 6.6b), the theoretical stiffness (Eq. 6.7) agrees well with the exper-
imental data, matching best at larger strains where the data is considered to be most reliable.
Data for the Oiles bearings (Fig. 6.6c) agrees less with Eq. 6.7, partly due to the difficulty of
obtaining a reliable estimate of the nominal stiffness kbo from the test data (Appendix 6A);
however, the slope of the experimental curves appears to match the slope of Eq. 6.7. Con-
sidering the many possibilities for experimental error, the theoretical model and experimental
On a side note, observe that the range of P/Pcr is different for each bearing, though
they are tested under identical axial loads P . The wide range of shape factors for the bearings
(S between 8.75 and 20 and S2 between 2.86 and 4, listed in Table 6A.1), led to significant
variations in the critical load Pcr (Eq. 6.16). Confirming an earlier assertion, the stiffness
variation due to axial loads is greater in bearings with low shape factors, or thick rubber layers
relative to their size, like MRPRA, than in bearings with high shape factors, like Bridgestone.
The proposed empirical model for strength variation with axial load in LRB bearings
(Eq. 6.18) is based solely on data for the Oiles LRB bearing. To show that this model is
justified by the data, the observed strength ratio Q/Qo as a function of P/Po is compared to
Eq. 6.18 for two different Oiles bearings (Fig. 6.7). The nominal strength Qo at each strain
level is the observed strength of the first bearing at its largest applied load (3Pst = 235 kN),
and the load Po for each bearing was selected visually to best fit the data. Po should be the
110
(a) HDR1 - Bridgestone
Bearing 1 (Pst =78 kN) Bearing 2 (Pst =49 kN)
1.25 1.25
Stiffness relative to nominal: kb /kbo
1 1
0.75 0.75
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
Axial load relative to Pcr : |P/Pcr | Axial load relative to Pcr : |P/Pcr |
Figure 6.6: Experimentally observed stiffness ratio kb /kbo vs P/Pcr for (a) Bridgestone HDR,
(b) MRPRA HDR, and (c) Oiles LRB, compared with Eq. 6.7. Data points at P = 0, Pst /2,
Pst , 2Pst , 3Pst , and −Pst /10, where Pst is the design axial load.
111
Bearing 1 (Pst =78 kN) Bearing 2 (Pst =49 kN)
1.2 1.2
Strength relative to nominal: Q/Qo
1 1
0.8 0.8
0.6 0.6
Model 1−exp(−P/Po)
0.4 Obs: strain = 25% 0.4
strain = 50%
strain = 75%
0.2 0.2
strain = 100%
0 0
0 2 4 6 8 0 2 4 6 8
Axial load relative to Po : P/Po Axial load relative to Po : P/Po
Figure 6.7: Strength ratio Q/Qo vs P/Po for Oiles LRB compared with Eq. 6.18. Data points
at P = 0, Pst /2, Pst , 2Pst , and 3Pst , where Pst is the design axial load.
same for both bearings, which are nominally identical; thus the differences in observed values
of 30 and 20 kN for Po may be due to accidental variation and testing under different axial
loads. The associated design loads: Pst = 78 kN and 49 kN, for interior and exterior bearings,
respectively, are just less than the recommended 3Po (Sec. 6.3).
The proposed empirical model (Eq. 6.18), depicted as a solid line, proves to be a reason-
able fit to the bearing data (Fig. 6.7), especially at the larger strains where the data is more
reliable. While the observed strength does not reduce to zero, it does fall exponentially as axial
load is removed. It was documented that at zero axial load the lead core began to extrude out of
both ends of the bearing during testing, seeming to support our claim that the lead core would
be ineffective in tension (Sec. 6.3). However, the Oiles bearings could not be tested in tension
because they were attached by dowelled connections, which may or may not have contributed
In the vertical characteristic tests, a cyclic force-controlled loading in the range Pst ± 0.3Pst
(Pst = 78 or 49 kN) was applied to each bearing, and its vertical deformation was recorded
(Appendix 6A). These tests are used to estimate damping in the vertical direction. The elliptical
appearance of the force-deformation relation (Fig. 6A.4) suggests viscous energy dissipation in
112
vertical motion. A damping coefficient for vertical motion is estimated from the dissipated
1 ED
ζeq = (6.19)
4π 1/2kbz u2bz
where kbz is the secant stiffness and ubz is one-half the deformation, measured as average peak-
to-peak over all cycles (Appendix 6A). From the experimental data, such as Fig. 6A.4, the
average damping coefficient was determined as ζeq = 0.064, 0.093 and 0.079 for the Bridgestone,
MRPRA, and Oiles bearings, respectively. Damping is observed to increase as the bearing shape
These estimated damping coefficients are not expected to be highly accurate because
techniques for recording vertical deformation were unreliable and the data used to estimate the
energy dissipation was noisy (Appendix 6A). Furthermore, although interpreted as exactly in-
phase, the axial forces and vertical deformations may have been slightly out-of-sync, introducing
a phase lag that falsely increases damping. Although such questions exist, the stated values
provide rough estimates of the damping coefficients, which support the choice of 5% viscous
The offset tests were simply a repeat of the vertical characteristic tests at different im-
posed shear strains, with vertical cycling over the loading ranges Pst ± 0.3Pst or Pst ± Pst .
Taking the nominal stiffness kbzo to be the observed stiffness at zero shear strain, the stiffness
ratio kbz /kbzo as a function of lateral deformation ÷ bending radius (ub /rb ) is compared with
Eq. 6.17 (Fig. 6.8). The theoretical and experimental stiffness ratios are in good agreement for
the Bridgestone bearings (Fig. 6.8a) and the Oiles LRB bearings (Fig. 6.8c). Unfortunately,
the predicted reduction in stiffness does not agree with test data for the MRPRA bearings
(Fig. 6.8b), and currently we have no explanation for this discrepancy. (The difficulties in mea-
suring vertical deformation should not significantly affect the relative stiffness ratio kbz /kbzo ).
However, evaluating the total data set, including both axial-load varied and offset tests, the
113
(a) HDR1 - Bridgestone (b) HDR2 - MRPRA
1.2 1.2
Vertical stiffness ratio: kbz /kbzo
0.8 0.8
0.6 0.6
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5
Deformation/bending radius (ub /rb )
Figure 6.8: Vertical stiffness ratio kbz /kbzo vs ub /rb for (a) Bridgestone HDR, (b) MRPRA
HDR, and (c) Oiles LRB. Experimentally observed values for two different bearings and two
different load sequences are compared with Eq. 6.17. Pst =78 kN for Bearing 1 and 49 kN for
Bearing 2.
The two-spring model of Sec. 6.2.1 adequately accounts for axial-load effects in a bearing with
linear material behavior. However, most isolation bearings are inherently nonlinear, especially
LRB bearings due to the yielding of the lead core, which will require extension of the two-
spring model to include various constitutive models for the shear spring. Three such models are
included: the coupled linear model uses a linear shear spring, identical to the model of Sec. 6.2.1.
The coupled nonlinear constant-strength model uses a bilinear force-deformation relation for
the shear spring, as does the coupled nonlinear variable-strength model, which additionally
incorporates the strength variation of Eq. 6.18. These latter models are hereafter shortened
114
to constant-strength and variable-strength models. Also defined for reference are comparable
bearing models that neglect axial-load effects, such that the lateral force-deformation relation,
either linear (uncoupled linear) or bilinear (uncoupled nonlinear), is uncoupled from the vertical
For the coupled linear model, the governing equations (Eqs. 6.4 and 6.5) are unchanged. Note
that we do not use the approximations that led to Eqs. 6.7 and 6.9. To consider a different
fb − fs (s) + P θ = 0 (6.20)
where the linear shear spring has been replaced by a general force fs (s).
The force of the bilinear shear spring in the constant-strength model may be deter-
mined numerically by classical rate-independent unidirectional plasticity [50], where the force-
deformation relation is elastic-plastic with kinematic hardening, and thus governed by the
following constitutive law, yield function, flow rule, and hardening law:
fs = kI (s − sp ) (6.21a)
q̇ = γ̇ Hsgn(fs − q) (6.21d)
respectively. The constitutive law (Eq. 6.21a) says the spring force equals the initial stiffness
kI times the elastic component of deformation s − sp , sp being the plastic deformation. The
initial stiffness kI = kbo + Qo /sy depends on the nominal stiffness and strength, as well as
the yield deformation sy . The yield function Φ (Eq. 6.21b) determines the set of admissible
forces, where the back force q stores the translation of the yield surface, and the yield force
fy = Qo + kbo sy . The spring response is elastic inside the yield surface (Φ < 0), and plastic
115
flow occurs on the yield surface (Φ = 0), determined by the flow rule (Eq. 6.21c) with constant
slip rate γ̇. Evolution of the back force is governed by the hardening law (Eq. 6.21d), with
To implement the strength variation due to axial load (variable-strength model), the yield
force fy in Eq. 6.21b is updated consistent with the strength Q (Eq. 6.18) at each time instant.
The initial stiffness kI remains constant, determined by the nominal strength Qo rather than
the current strength Q. For both the constant-strength and variable-strength models, fs (s)
can be computed to satisfy Eq. 6.21 via the return mapping algorithm [50]; details are given in
Appendix 6B.
A numerical routine is presented to implement the bearing model into a typical dynamic analysis
program. Based on the stiffness approach, the program is assumed to compute the deformations
of the system at each time step iteratively. Given the bearing deformations, a local routine for
the bearing computes the bearing forces to satisfy the governing equations (Eqs. 6.20, 6.4a,b
and 6.5) and the bearing stiffness matrix, which are returned for use in the global procedure.
Bearing Forces
Let F = fb , P T represent the vector of independent bearing forces that are to be computed
in the local bearing routine. The resultant moment M1 , found by equilibrium of the bearing in
the deformed configuration, should also be applied at the top of the bearing:
M1 = (fb hb + P ub ) (6.22)
The governing equations (Eqs. 6.20, 6.4b,c, and 6.5a,b) represent a system of five nonlinear
116
form: ⎧ ⎫ ⎧ ⎫
⎪
⎪ ⎪
g1 ⎪ ⎪
⎪ fb − fs (s) + P θ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ g ⎪
⎪ ⎪
⎪f h − P h θ + P (s + h θ)⎪
⎪
⎪ 2
⎨ ⎬ ⎨⎪ ⎪ b b E b b ⎪
⎬
g = g3 =
⎪ ⎪ ⎪ ⎪
⎪ P − kbzo v ⎪
(6.23)
⎪
⎪ ⎪ ⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ g ⎪
⎪ ⎪
⎪ u − s − h θ ⎪
⎪
⎪
⎪
4
⎪
⎪ ⎪
⎪
b b ⎪
⎪
⎪
⎪ ⎪ ⎪ ⎪
⎩g5 ⎭ ⎩ u − sθ − h θ 2 /2 − v ⎪
⎪ ⎪ ⎭
bz b
and solved by Newton’s method, i.e., find x = fb , P, s, θ, vT to satisfy g(x) = 0.
The converged solution x at the previous global iteration serves as an initial guess x(0) .
for k = 1, 2, . . . , where the Jacobian J(x) (Jij = ∂gi /∂xj ), determined from Eq. 6.23:
⎡ ⎤
∂fs
⎢1 θ − ∂s P 0 ⎥
⎢ ⎥
⎢h (s + h θ) (P − PE )hb 0 ⎥
⎢ b b P ⎥
⎢ ⎥
⎢ ⎥
J (x) = ⎢ 0 1 0 0 −k bzo ⎥ (6.25)
⎢ ⎥
⎢ ⎥
⎢0 −1 −h 0 ⎥
⎢ 0 ⎥
⎣ ⎦
0 0 −θ −(s + hb θ) −1
has a nonzero determinant, and hence is invertible, even when P , s, and θ are zero. Equa-
tion 6.24 is applied repeatedly until the incremental change in the solution is less than a desired
tolerance, that is: x(k) − x(k−1) < tol. At each iteration the shear spring force fs (s) in
Eq. 6.23 and tangent ∂fs /∂s in Eq. 6.25 are computed by the return mapping algorithm (Ap-
pendix 6B).
Next, the bearing stiffness matrix kb, relating a change in the bearing forces dF = dfb , dP T
1. Take differentials of the equilibrium equations (Eq. 6.20, Eq. 6.4b,c), resulting in:
keq dv = T dF (6.26)
117
where dv = ds, dθ, dvT , and the matrices keq and T are given by:
⎡ ⎤ ⎡ ⎤
∂fs
⎢ ∂s −P 0 ⎥ ⎢1 θ ⎥
⎢ ⎥ ⎢ ⎥
keq =⎢
⎢−P (PE − P )hb 0 ⎥
⎥ and T = ⎢hb (s + hb θ)⎥
⎢
⎥ (6.27)
⎣ ⎦ ⎣ ⎦
0 0 kbzo 0 1
dU = T T dv (6.28)
dU = T T feq T dF (6.29)
where feq = keq −1 . The resultant flexibility matrix fb = T T feq T , which relates the
Given the complexity of Eq. 6.30, the bearing stiffness matrix kb, the inverse of fb, is not
1. Suppose both the axial load P and deformations s and θ are zero, and the shear spring is
linear (∂fs /∂s = Ps /hb ); then the flexibility matrix (Eq. 6.30) is diagonal with elements:
(PE + PS )hb 1
f11 = and f22 = (6.31)
PE PS kbzo
Typically PE PS , giving f11 ≈ hb /PS = 1/kbo , such that the lateral and vertical bearing
stiffnesses reduce to their nominal values [k11 ≈ kbo (Eq. 6.1) and k22 = kbzo (Eq. 6.2)].
2. Suppose just the deformations s and θ are zero, and the shear spring is linear; then the
(PE + P + PS )hb 1
f11 = and f22 = (6.32)
PS PE − PS P − P 2 kbzo
118
Inverting f11 and again taking PE PS gives:
(PS PE − P 2 ) P 2
k11 ≈ = kbo 1 − (6.33)
PE hb Pcr
3. Suppose hb θ is small relative to the total lateral deformation ub , the shear spring is linear,
1 u2b 1
f22 ≈ + (6.34)
PE hb kbzo
whose inverse is the tangent vertical stiffness kbz derived earlier (Eq. 6.10).
Thus, the exact stiffness matrix kb has been shown with relevant assumptions to lead to the
same approximate force-deformation equations (Eq. 6.7 and 6.9) given earlier, a good check on
its accuracy. An outline of the complete bearing routine, including solving for the force vector
Lateral force-deformation behavior for the coupled linear model was already demonstrated in
Fig. 6.3a, where the lateral stiffness was shown to decrease as the applied force P approached
Pcr . The lateral force-deformation relation of the constant-strength model (Fig. 6.9a) is closely
related to the coupled linear model. The postyield stiffness in Fig. 6.9a is essentially identical
to the stiffness of the coupled linear model (Fig. 6.3a), showing the same successive decline as
the axial force is increased toward Pcr . The initial stiffness, and energy dissipated in a single
cycle (loop area), are affected by axial force only negligibly. As a special case, the curve with
The primary effect of including the strength variation of Eq. 6.18 (variable-strength
model) is that the yield strength, and hence energy dissipated in a single cycle, also depends
on the axial force P (Fig. 6.9b). Recall that when P = 0 or small, the strength, and hence
energy dissipated, is also zero or small, as observed in Fig. 6.9b. As axial force becomes large
119
(a) (b)
0.4 0.4
P =0 P =0
0.3 0.2Pcr 0.3 0.05Pcr , 0.38Po
0.4Pcr 0.4Pcr , 1.5Po
0.2 0.6Pcr 0.2 0.6Pcr , 3.75Po
Lateral force fb /Pst
0 0
−0.1 −0.1
−0.2 −0.2
−0.3 −0.3
−0.4 −0.4
−20 −10 0 10 20 −20 −10 0 10 20
Lateral deformation ub (cm) Lateral deformation ub (cm)
Figure 6.9: Lateral force-deformation as a function of P/Pcr for (a) constant-strength model
and (b) variable-strength model.
relative to Po and Pcr , the strength is regained but the postyield stiffness decreases similar
to the constant-strength model (Fig. 6.9a). A well-designed bearing should achieve a balance,
in the sense that P should be large enough to avoid significant strength degradation but well
below Pcr .
of the particular bearing model (linear, constant-strength or variable-strength), and the plots
of Fig. 6.3b for the coupled linear model also represent the constant-strength and variable-
strength models. Clearly, the vertical force-deformation will differ for the different models
For each variation of the bearing model presented in Sec. 6.5, a rigid block supported on two
bearings is subjected to the seismic pulse of Fig. 6.10, and the response of the left exterior
bearing is shown (Figs. 6.11-6.13). These figures demonstrate response histories of the lateral
and vertical deformations and forces, the lateral force-deformation relation, and the vertical
force-deformation relation. For the response histories, the lateral force and the vertical defor-
mation are scaled by factors indicated so that lateral and vertical deformations and lateral and
vertical forces can be viewed at similar amplitudes. These data are shown quantitatively in
120
(a) (b) (c)
150 150 100
100 100
50
50 50
üg (cm/s2 )
u̇g (cm/s)
ug (cm)
0 0 0
−50 −50
−50
−100 −100
Figure 6.10: (a) Acceleration ügo , (b) velocity u̇go and (c) displacement ugo history of seismic
excitation: approximately 1.5 cycles of a sinusoidal pulse with velocity u̇go = 90 cm/s and
period Tp = 4 seconds. Isolation period Tb for the system excited is 2 seconds.
the force-deformation plots, with forces normalized by the static load Pst . For simplicity the
The bearing response predicted by the coupled linear model is given (Fig. 6.11), with
the comparable response using the uncoupled linear model shown for reference. The lateral
deformation and axial force of the bearing are nearly in-phase (Fig. 6.11a). As a consequence,
the axial force P increases when the lateral deformation is large and positive, and an associated
drop in stiffness (Eq. 6.7) is observed in the lateral force-deformation relation (Fig. 6.11b).
The lateral stiffness does not appear to change at negative lateral deformations, because the
axial force decreases toward zero, and the lateral stiffness simply defaults to its nominal value
(Fig. 6.4). Including axial-load effects (coupled linear vs. uncoupled linear model) appears to
The vertical deformation is considerably greater when axial-load effects are included (cou-
pled linear model vs. uncoupled linear model, Fig. 6.11a and c). To understand this, recall that
the vertical force-deformation relation depends greatly on the lateral deformation (Eq. 6.9).
Thus, when the axial force is close to its static value, the lateral deformation is close to zero
(recalling the correlation between P and ub observed above) and the vertical force-deformation
121
(a) (b)
0.6
Coupled Linear
Lateral Deformation 0.4 Uncoupled Linear
ub+
0
−60 −40 −20 0 20 40 60
Lateral deformation ub (cm)
Lateral Force (x2)
fb+ (c)
2
0 Axial force P/Pst
1.5
fb−
Axial Force
Pst +P 1
Pst
0.5
Pst -P
0
0 Tp T 0 1 2 3 4 5 6 7
Time (s) Vertical deformation ubz (cm)
Figure 6.11: Response of the left exterior bearing – using the coupled linear model – of a rigid
block subjected to a seismic pulse: (a) lateral and vertical deformation and force histories, (b)
lateral force-deformation, and (c) vertical force-deformation. Also shown is the comparative
response with the uncoupled linear model; Pst /Pcr = 0.4.
is essentially that of the uncoupled linear model. However, as the axial force deviates from
the static force in either direction, the lateral deformation becomes large, resulting in vertical
softening and a considerable increase in vertical deformation. Thus the arc-shaped vertical
force-deformation (Fig. 6.11c), with vertical deformation increasing at axial forces larger or
smaller than Pst . Due to the inherently linear relation between axial force P and deformation
ubz (Eq. 6.9), the total deformation ubz is obviously greater when P > Pst than when P < Pst .
The closed loops exhibited by the above force-deformation relations leave the impression
that energy dissipation occurs (Fig. 6.11b and c), which would violate energy conservation in
an elastic system. However, a simple example in Appendix 6C verifies that energy is conserved.
122
(a) (b)
0.3
Constant-Strength
Uncoupled Nonlinear
Lateral Deformation 0.2
ub+
0
−20 −10 0 10 20
Lateral deformation ub (cm)
Lateral Force (x4)
fb+ (c)
2
0 Axial force P/Pst
1.5
fb−
Axial Force
Pst +P 1
Pst
0.5
Pst -P
0
0 Tp T 0 0.2 0.4 0.6 0.8 1
Time (s) Vertical deformation ubz (cm)
Figure 6.12: Response of the left exterior bearing – using the constant-strength model – of a
rigid block subjected to a seismic pulse: (a) lateral and vertical deformation and force histories,
(b) lateral force-deformation, and (c) vertical force-deformation. Also shown is the comparative
response with the uncoupled nonlinear model; Pst /Pcr = 0.4.
The bearing response predicted by the constant-strength model is plotted in Fig. 6.12
and compared with the response predicted by the uncoupled nonlinear model. Whereas for the
linear models the axial force history and lateral deformation history were in phase (Fig. 6.11a),
for these nonlinear models the lateral force history is more closely in-phase with the lateral de-
formation history (Fig. 6.12a). For the constant-strength model, this causes the lateral stiffness
(initial or postyield) to show the greatest decrease at large positive lateral forces (Fig. 6.12b),
which correspond to the maximum axial forces. The result is a slight increase in peak lateral
deformation compared to the nonlinear model. Because of the changing stiffness, the appar-
ent width of the hysteresis loop changes during the deformation cycle, whereas this width is
123
(a) (b)
0.3
Variable-Strength
Lateral Deformation 0.2 Uncoupled Nonlinear
ub+
0
−20 −10 0 10 20
Lateral deformation ub (cm)
Lateral Force (x4)
fb+ (c)
2
0 Axial force P/Pst
1.5
fb−
Axial Force
Pst +P 1
Pst
0.5
Pst -P
0
0 Tp T 0 0.2 0.4 0.6 0.8 1
Time (s) Vertical deformation ubz (cm)
Figure 6.13: Response of the left exterior bearing – using the variable-strength model – of a
rigid block subjected to a seismic pulse: (a) lateral and vertical deformation and force histories,
(b) lateral force-deformation, and (c) vertical force-deformation. Also shown is the comparative
response with the uncoupled nonlinear model; Pst /Pcr = 0.4 and Pst /Po = 3.
constant for the uncoupled nonlinear model. A high-frequency component is observed in the
axial force cycle (Fig. 6.12a), causing local variations in stiffness and roughness in the lateral
terpret (Fig. 6.12c). Although quite disordered, it appears hysteretic in nature relative to the
coupled linear model. Perhaps since the axial force variation resembles the lateral force vari-
ation the vertical hysteretic behavior is related to hysteretic behavior in the lateral direction
(Fig. 6.12b). Local variations in the vertical force-deformation relation are related to the high
frequency component of axial force. The peak vertical deformation is significantly less than that
124
of the coupled linear model (Fig. 6.11c), correlating to the reduction of peak lateral deformation.
The bearing response predicted by the variable-strength model, including response his-
tories (Fig. 6.13a) and force-deformation relations (Fig. 6.13b,c) – which are compared with
the response predicted by the uncoupled nonlinear model – is similar to that of the constant-
strength model (Fig. 6.12). The influence of variable strength (Eq. 6.18) is readily apparent
when the axial force is close to zero by comparing Figs. 6.12b and 6.13b. Observe that the
strength decreases and the yield surface contracts at negative lateral forces in the lateral force-
deformation relation (Fig. 6.13b). Compared to the uncoupled nonlinear model, the average
width of the force-deformation loop is smaller, and the associated decrease in energy dissipation
increases the peak lateral deformation, which in turn increases the peak vertical deformation
(Fig. 6.13c). The roughness in the lateral force-deformation increases compared to Fig. 6.12c
6.7 Conclusions
This investigation of the influence of axial load on the nonlinear response of isolation bearings
1. The following axial-load effects have been observed in various tests of both high-damping
rubber and lead-rubber bearings: decreasing lateral stiffness – measured as either the
secant or postyield stiffness – with increasing axial load, decreasing lateral yield strength
with decreasing axial load (lead-rubber bearings only), and decreasing vertical stiffness
2. Based on linear stability theory of an elastic column, a two-spring model of the bearing,
consisting of a shear spring and a rotational spring divided at top and bottom (Fig. 6.2),
accurately accounts for axial-load effects in linear models of bearings. However, bearings
with high-damping fillers or lead cores that provide energy dissipation are nonlinear and
constitutive model for the shear spring. Using unidirectional plasticity with kinematic
125
hardening for the shear spring led to the constant-strength model for the bearing. Nu-
equilibrium (Eqs. 6.20 and 6.4b,c) and kinematic (Eq. 6.5) – by Newton’s method for
the current forces in the bearing, and taking differentials of these equations to derive the
strain hardening.
4. An empirical model (Eq. 6.18) was developed to account for the varying yield strength
bearing response. This behavior is optionally included by simply updating the yield
force in the plasticity equations to reflect the time-varying yield strength, leading to the
variable-strength model.
in the response of a rigid block supported by two bearings subjected to a seismic pulse.
The effects were most pronounced in the variable-strength model, where the peak lat-
eral deformation increased by about 20% and the peak vertical deformation more than
doubled compared to the uncoupled nonlinear model (without axial-load effects). While
the bearing response predicted by the new models are hereby demonstrated for only one
case, it is desirable to investigate the response for a diverse range of system and bearing
parameters and earthquake excitations, which is the subject of the next chapter.
126
Table 6A.1: Dimensions of Prototype Bearings
This appendix describes how data recorded in the various bearing tests was interpreted. For
reference, the dimensions of the three prototype bearings are listed in Table 6A.1.
In the axial-load varied tests, two bearings from each manufacturer were subjected to the
prescribed lateral deformation, five cycles each at shear strains (lateral deformation relative to
height) of 5, 25, 50, 75 and 100%. The tests were repeated at axial loads of 0, Pst /2, Pst , 2Pst ,
3Pst and -Pst /10, one bearing each for design loads Pst of 78 kN (interior bearing) and 49 kN
(exterior bearing). The tensile tests were omitted for the Oiles LRB bearing, attached by a
dowelled rather than a bolted connection. The lateral force in the bearing in response to the
applied deformation was recorded; this force signal was smoothed using a Gaussian kernel with
full width at half maximum (FWHM) equal to 4*δt. From the deformation and force signals,
the force-deformation relation (bearing hysteresis) for each test was plotted at the various strain
levels. As an example, force-deformation for the Bridgestone bearing at the axial load P = Pst
From the bearing force-deformation, the strength and stiffness at each strain level were
determined independently. Locating the positive and negative y-axis force intercepts in each
cycle and averaging over the cycles, the strength Q was interpreted as half the distance between
the averaged intercepts, which are indicated by dots in Fig. 6A.1. Computed from data points
127
Data File: 911216.07, Bearing: Bridgestone #12, Axial Load Varied: P =78.3 kN
20 20 20
Strain = 5% Strain = 25% Strain = 50%
Bearing force (kN)
10 10 10
0 0 0
10 10 10
0 0 0
Figure 6A.1: Lateral force-deformation at different strain levels and complete bearing hys-
teresis for Bridgestone #12 with applied axial load P = 78 kN. Interpretation of Q and kb given
by dots at the y-intercept and dashed lines.
adjacent to these force intercepts, the postyield stiffness kb was interpreted as the slope at the
force intercepts, averaged over the cycles. For the Bridgestone example, the resultant stiffnesses
are the slopes of the dashed lines drawn through the force intercepts (Fig. 6A.1).
While this procedure worked well for the Bridgestone and MRPRA bearings, some ad-
justments were necessary for the Oiles bearings. A sample force-deformation for Oiles Bearing
1 at P = Pst (Fig. 6A.2) shows a slight pinching of the hysteresis near the origin, i.e., zero shear
strain. In this case, the stiffness was interpreted as the average secant stiffness over a larger
range, from about -63 to +63% of the maximum local strain. Again, the resulting stiffness is
The Oiles data at zero axial load (P = 0) presented another challenge, as the postyield
stiffness is influenced by the near complete loss of strength (Fig. 6A.3). Although not theoreti-
cally predicted, the interaction between rubber and lead appears to cause the tangent stiffness
128
Data File: 920114.09, Bearing: Oiles #12, Axial Load Varied: P =78.3 kN
20 20 20
Bearing force (kN)
0 0 0
20 20 20
Bearing force (kN)
10 10 10
0 0 0
Figure 6A.2: Lateral force-deformation at different strain levels and complete bearing hys-
teresis for Oiles #12 with applied axial load P = 78 kN. Interpretation of Q and kb given by
dots at the y-intercept and dashed lines.
to decrease at the origin compared to large shear strains. Thus, how should the stiffness,
which will represent the nominal stiffness, be interpreted to give values consistent with those
for nonzero axial loads? Believing the low stiffness near the origin to be an anomaly, we took
the stiffness as the average tangent at about 63% of the maximum local strain, determined
from adjacent data points at that strain. This stiffness, shown drawn through the points about
which it was estimated (Fig. 6A.3), appears indicative of the tangent stiffness away from the
origin. The inconsistency of the methods to determine the stiffness of the Oiles LRB bearing
could explain why it did not match theory as well as the other two (Fig. 6.6).
The vertical characteristic test was applied to 15 each of the Bridgestone and Oiles bear-
ings, and 12 MRPRA bearings, which consisted of a cyclic force-controlled loading over the
range Pst ± 0.3Pst , where Pst was 78 kN or 49 kN. Two bearings of each type were also sub-
jected to the offset tests, a repeat of the vertical characteristic test at imposed offset shear
129
Data File: 920114.07, Bearing: Oiles #12, Axial Load Varied: P =0 kN
20 20 20
Bearing force (kN)
0 0 0
20 20 20
Bearing force (kN)
10 10 10
0 0 0
Figure 6A.3: Lateral force-deformation at different strain levels and complete bearing hys-
teresis for Oiles #12 with applied axial load P = 0 kN. Interpretation of Q and kb given by
dots at the y-intercept and dashed lines.
strains of 0, 50, 100, and 150% and over the ranges Pst ± 0.3Pst and Pst ± Pst .
During the vertical characteristic tests and offset tests, the load signal was applied using
two vertical actuators. The resulting axial force in the bearing was measured by a single load cell
under the bearing, while the vertical deformation was measured by four direct current voltage
transducers (DCDTs) attached at the bearing ‘corners’, and then averaged. The observed
deformations were not very accurate for two reasons: first, the transducers measured the relative
motion between the top and bottom plates of the test machine, which was influenced by plate
bending; second, the vertical bearing deformations were small relative to the resolution of the
transducers. This problem was greatest for the Bridgestone bearings, with peak deformations
on the order of 0.08 mm (.003 in) and a resolution of only 0.025 mm (.001 in). The obvious effect
was incredibly noisy data. A Gaussian kernel was no longer a sufficient filter for smoothing the
data, and better results were obtained by applying a low pass filter that eliminated frequencies
130
Bridgestone Vertical Characteristic Tests: cycles at Pst ± 0.3Pst with Pst=78.3 kN
File: 911120.02, Bearing 1 File: 911204.02, Bearing 2 File: 911205.02, Bearing 3
−40 −40 −40
Vertical force (kN)
Figure 6A.4: Vertical force-deformation for Bridgestone bearings, as determined from vertical
characteristic tests. Interpretation of vertical stiffness kbz given by dashed lines.
Samples of the smoothed – though still quite noisy – force-deformation relations from
the vertical characteristic tests of Bridgestone bearings are shown in Fig. 6A.4. The vertical
deformation at the start of the test, which should have indicated the static deformation, was
very inconsistent, and thus the plots were centered at zero deformation. The viscous damping
coefficient, desired from the characteristic tests, was estimated by the following steps: (1) locate
the local extrema – one at each half-cycle – as the maximum of force*deformation (indicated
by dots in Fig. 6A.4), (2) draw a tangent line between each pair of adjacent extreme points, (3)
average the slope of this line over all half cycles to get the vertical stiffness (indicated by dashed
lines in Fig. 6A.4), (4) estimate the energy dissipated in each half cycle by numerically inte-
grating the area between the tangent line and the actual data, (5) average the energy dissipated
over all half cycles and multiply by 2 for energy dissipated per cycle, (6) compute the damping
131
Bridgestone Offset Tests: cycles at Pst ± 0.3Pst (top) or Pst ± Pst (bottom) with Pst=78.3 kN
File: 911218.09, 50% offset File: 911218.12, 100% offset File: 911218.13, 150% offset
−40 −40 −40
Vertical force (kN)
−0.3 −0.15 0 0.15 0.3 −0.3 −0.15 0 0.15 0.3 −0.3 −0.15 0 0.15 0.3
Vertical deformation (mm)
Figure 6A.5: Vertical force-deformation for a single Bridgestone bearing at offset shear strains
of 50, 100, 150% and load cycles of P ± 0.3P or P ± P . Interpretation of vertical stiffness kbz
given by dashed lines.
coefficient (Eq. 6.19) based on the estimates of vertical stiffness, half peak-to-peak deformation,
and energy dissipated. This damping coefficient was averaged over all characteristic tests for a
particular bearing (Bridgestone, Oiles, MRPRA) to get the values given in Sec. 6.4.2.
The estimates of energy dissipation, in particular, are likely to be influenced by the noisy
data. The numerical integration technique may overestimate the energy dissipation by picking
up noise and/or adding area on both sides of the tangent line as positive energy. For this reason,
the damping coefficients estimated by this technique may be slightly high, but are believed to
bearing is shown for shear strains of 50, 100 and 150% (Fig. 6A.5). Also shown are the dashed
lines whose slopes represent the vertical stiffness, determined in the same way as for the vertical
characteristic tests. For each load cycle (Pst ± 0.3Pst , Pst ± Pst ), a slight decrease in stiffness
132
Oiles Offset Tests: cycles at Pst ± 0.3Pst (top) or Pst ± Pst (bottom) with Pst=78.3 kN
File: 920115.03, 50% offset File: 920115.05, 100% offset File: 920115.07, 150% offset
−40 −40 −40
Vertical force (kN)
Figure 6A.6: Vertical force-deformation for a single Oiles bearing at offset shear strains of
50, 100, 150% and load cycles of P ± 0.3P and P ± P . Interpretation of vertical stiffness above
and below the design load P given by dashed lines.
A peculiar effect was observed in the offset tests for the Oiles LRB bearing (Fig. 6A.6).
The response looks normal for load cycles in the range of Pst ± 0.3Pst , but in the larger range of
Pst ± Pst , the bearing softens rapidly as the axial load approaches zero. The effect is magnified
with increase in shear strain. Because the Oiles bearings have dowelled connections, they have a
tendency to rollout at the large shear strains. The axial force couple provides a restoring moment
to resist rollout. Thus, when the axial force is removed, as in these large range offset tests,
significant rollout occurs, and the vertical deformation measurements are affected in unforeseen
ways. The proposed model for the LRB bearing does not really address this behavior, which
is hopefully limited to the increasingly uncommon dowelled connection. We dealt with this
by sampling the tangent vertical stiffness at cycles Pst ± Pst separately above and below Pst
(Fig. 6A.6), and using the stiffness at P > Pst for comparison with the theoretical stiffness
133
(Fig. 6.8).
The basic postulates of unidirectional plasticity were laid out earlier (Eq. 6.21). Recall that
plastic flow (Eq. 6.21c) occurs only when the yield function Φ equals zero (Eq. 6.21b). In
addition, consistency (γ̇ Φ̇ = 0) requires that the force persist on the yield surface for plastic
flow. The slip rate γ̇ can thus be determined by taking the time derivative of the yield function
(Eq. 6.21b), substituting in the constitutive law (Eq. 6.21a), the flow rule (Eq. 6.21c), and the
sgn(fs − q)kI ṡ
γ̇ = (6B.1)
kI + H
Substituting γ̇ into the flow rule (Eq. 6.21c), the plastic deformation sp is obtained in terms of
the total shear deformation s, which is then introduced into Eq. 6.21a, leading to the deforma-
∂fs kI H
= (6B.2)
∂s kI + H
The tangent stiffness during plastic flow equals the postyield stiffness kbo , thus, from Eq. 6B.2,
Numerical implementation of the nonlinear bearing models necessitates computing the spring
force fs and tangent stiffness ∂fs /∂s for a current estimate of the spring deformation s (Sec. 6.5.2).
This can be accomplished by enforcing the plasticity postulates (Eq. 6.21) using the return map-
ping algorithm [50]. Each of these laws are either discretized or enforced at discrete time steps.
The steps of the algorithm are outlined in pseudocode in Algorithm 1. A trial force
for the spring is computed by taking a trial elastic step (Alg. 1 line 1). If the yield function
(Eq. 6.21b) is satisfied (Φ ≤ 0) then the trial step is confirmed (line 5). If the trial force lies
outside the current yield surface (Φ > 0), the force is returned to the yield surface, through
which the incremental plastic slip ∆γ is defined (line 8). Since plastic flow occurs, the plastic
134
1: fstrial = kI (s − sp ) {Trial elastic step}
2: ξ trial = |fstrial − q|
3: Φtrial = ξ trial − fy {Test yield function for admissibility}
4: if Φtrial ≤ 0 then {Confirm elastic step}
5: (·) = (·)trial
6: ∂fs /∂s = kI
7: else {Plastic step}
8: ∆γ = Φtrial /(kI + H)
9: fs = fstrial − ∆γ kI sgn(ξ trial ) {Return force to the yield surface}
10: sp = sp + ∆γ sgn(ξ trial ) {Update plastic deformation}
11: q = q + ∆γ H sgn(ξ trial ) {Update back force}
12: ∂fs /∂s = kbo
13: end if
14: return: fs , ∂fs /∂s
deformation sp and back force q must be updated (Eqs. 6.21c and 6.21d, Alg. 1 lines 10-11).
These variables (q and sp ) are stored separately for each bearing. The tangent stiffness ∂fs /∂s
equals kI during an elastic step (line 6) and kbo during a plastic step (line 12).
To apply return mapping for the variable-strength model, the yield force fy should be updated
to reflect the variation of strength Q in time: fy = Q + kbo sy . However, if the yield deformation
sy is held constant, the initial stiffness kI varies because fy = kI sy . To avoid complicating the
constitutive law (Eq. 6.21a) with variable initial stiffness, fy is assumed to vary proportional
to Q:
fy = fyo 1 − e(−P/Po ) (6B.3)
such that the yield deformation sy also varies in time. Recall that fy = 0 when the bearing is
Substituting fy (Eq. 6B.3) into Eq. 6.21b, the yield function now reads:
Φ(fs , q) = |fs − q| − fyo 1 − e−(P/Po ) (6B.4)
135
1: fstrial = kI (s − sp ) {Trial elastic step}
2: ξ trial = |fstrial − q|
3: fy = max 0, fyo 1 − e(−P/Po ) {Update yield force}
4: Φ trial =ξ trial − fy {Test yield function for admissibility}
5: if Φ trial ≤ 0 then {Confirm elastic step}
6: (·) = (·)trial
7: ∂fs /∂s = kI
8: else {Plastic step}
9: ∆γ = Φtrial /(kI + H)
10: fs = fstrial − ∆γ kI sgn(ξ trial ) {Return force to the yield surface}
11: sp = sp + ∆γ sgn(ξ trial ) {Update plastic deformation}
12: q = q + ∆γ H sgn(ξ trial ) {Update back force}
13: ∂fs /∂s = kbo
14: end if
15: return: fs , ∂fs /∂s
Applying the consistency condition, Φ̇ = 0 during plastic flow to this new yield function leads
to:
sgn(fs − q)kI ṡ 1 e−(P/Po ) fyo Ṗ
γ̇ = − (6B.5)
kI + H Po kI + H
The flow rule (Eq. 6.21c) and hardening law (Eq. 6.21d), being functions of the slip rate γ̇,
are indirectly affected by this change in γ̇. But algorithmically, the only change needed is to
update the yield function based on the current axial force P each time the algorithm is called:
Φtrial = ξ trial − fyo 1 − e−(P/Po ) (6B.6)
Based on Sec. 6.5.2 and above, pseudocode is provided for the complete nonlinear bearing
routine. From a current estimate of the lateral and vertical deformations ub and ubz , the code
returns the force vector F that satisfies equilibrium (Eqs. 6.20 and 6.4b,c) and kinematics
136
(a) (b) (c)
(1) (2)
P
ub f3 (3)
Prescribed loads
Vertical Force
Lateral Force
f2 (2)
0
Lateral Deformation
0
Vertical Force 0
(3),v3
0
t0 t1 t2 t3 t4 0 ub 0 v1 v2
Time Lateral Deformation Vertical Deformation
Figure 6C.1: (a) Prescribed load sequence of lateral deformation and axial force, (b) lateral
force-deformation, and (c) vertical force-deformation.
and v are computed iteratively with Newton’s method (Eq. 6.24) until convergence is attained
(Alg. 3 lines 2-8). In each Newton iteration, the return mapping algorithm (line 3) updates fs
and ∂fs /∂s based on the current value of s. Upon convergence, the matrices feq , T , flexibility
To show that energy is conserved in the coupled linear model, consider the following load
sequence: (1) axial load increased linearly from 0 to P , (2) lateral deformation increased linearly
from 0 to ub , (3) axial load decreased linearly from P to 0, (4) lateral deformation decreased
linearly from ub to 0 (Fig. 6C.1a). The resulting lateral force-deformation and vertical force-
deformation are shown in Figs. 6C.1b and c, with each step in the load sequence indicated.
Since the lateral force-deformation does negative work (adds energy to the system) and the
vertical force-deformation does positive work (subtracts energy), energy is conserved if the two
In terms of the applied load P and deformation ub , the approximate lateral forces are f2 =
PS /hb (1 − P 2 /PS PE )ub and f3 = (PS /hb )ub (Eq. 6.7), and approximate vertical deformations
are v1 = P/kbzo , v2 = P/kbzo + (PS + P )u2b /(PE hb ), and v3 = PS u2b /(PE hb ) (Eq. 6.8). The
137
Require: Previous converged values fb , P , s, θ and v
1: Let x = fb , P, s, θ, vT
2: repeat
3: Solve for⎧fs (x3 ) and ∂fs /∂s with Algorithm ⎫ 1 or 2
⎪
⎪ x1 − fs (x3 ) + x2 x4 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ x h − P h x + x (x + hb 4 ⎪
x )⎪
⎨ 1 b E b 4 2 3 ⎬
4: f (x) = x2 − kbzo x5
⎪
⎪ ⎪
⎪
⎪
⎪ ub − x3 − hb x4 ⎪
⎪
⎪
⎪ ⎪
⎪
⎩ ⎭
ubz − x3 x4 − hb /2x4 − x5
2
⎡ ⎤
1 x4 −∂fs /∂s x2 0
⎢ ⎥
⎢hb (x3 + hb x4 ) x2 (x2 − PE )hb 0 ⎥
⎢ ⎥
5: J(x) = ⎢ ⎢ 0 1 0 0 −kbzo
⎥
⎥
⎢ ⎥
⎣0 0 −1 −hb 0 ⎦
0 0 −x4 −(x3 + hb x4 ) −1
6: −1
dx = −J (x) · f (x)
7: x = x + dx
8: until norm(dx) < tol
9: Update fb = x1 , P = x2 , s = x3 , θ = x4 , v = x5
10: F = fb , P T
11: det = hb (∂fs /∂s)(PE − P ) − P 2
⎡ (PE −P )hb P
⎤
det det 0
⎢ P ∂fs /∂s ⎥
12: feq = ⎣ det det 0 ⎦
1
0 0
⎡ ⎤ kbzo
1 θ
⎢ ⎥
13: T = ⎣hb (s + hb θ)⎦
0 1
14: fb = T T fbT
15: kb = fb−1
16: return: F , kb
138
negative work traced by the lateral force-deformation is thus:
1 1 P 2 u2b
W− = ub (f3 − f2 ) = (6C.1)
2 2 PE hb
1 1 1 P 2 u2b PS P u2b
W+ = P (v2 − v1 ) + P v3 = + (6C.2)
2 2 2 PE hb PE hb
By taking PS P , which is to the order of accuracy of the approximations, the second term
of Eq. 6C.2 is neglected and the negative and positive work cancel.
139
Appendix 6D: Notation
D bearing diameter
ing
ear models
models
formation
140
Kc factor by which Ec is multiplied to include bulk compressibility of
rubber
keq stiffness matrix used in local bearing routine, derived from bearing
equilibrium equations
rubber
PE Euler buckling load, neglects shear stiffness of the bearing: π 2 EIs /h2b
S2 second shape factor or bearing aspect ratio; D/tr for circular bearing
shear deformation
142
7: ESTIMATING ISOLATOR DEFORMATIONS AND FORCES
CONSIDERING LATERAL-ROCKING RESPONSE
7.1 Introduction
To estimate the peak axial forces in individual isolators, which are needed for design, it is
isolation bearings was explored [51], but the study was limited to vertical ground excitation,
where the vertical degree-of-freedom was coupled to the others through a small eccentricity.
One case study of an isolated building predicted (unrestrained) uplift off several of the bearings
[52], while another determined the uplift of 1994 UBC-designed 6 and 20-story buildings due
to various near-source ground motions [53]. The latter study concluded that resisting uplift
should be a major design consideration; and in a follow-up to that effect, the same 6-story
building was modified to meet the upgraded standards of the 1997 UBC, the bearings were per-
mitted to resist tension, and tensile demands were successfully limited by special measures in
the building frame design [54]. Although making valuable contributions, none of these studies
Recall that axial-load effects, implemented in the nonlinear constant-strength and variable-
strength models for lead-rubber bearings, were shown in the previous chapter to influence the
behavior of the isolation bearings. What is the significance of these effects on the overall system
response and pertinent design quantities, such as lateral deformation and axial forces in the
isolators? One previous study incorporated similar axial-load effects into dynamic analysis of
the isolated building [55], showing that neglecting them would increase the error in the lateral
isolator deformation as the factor of safety against buckling was reduced. However, the lateral
behavior was assumed to be linear, and vertical flexibility was neglected such that axial forces
could not be determined accurately, deficiencies that have been remedied by our nonlinear
models.
Given this background information, the objectives of this chapter are (1) to understand
143
d/2 d/2
z, uz
h/2
CM
θx y, uy
h/2
ith bearing
(yi , −h/2)
üg (t)
the response of the isolated building using the improved constant-strength and variable-strength
models for lead-rubber bearings, (2) to estimate the peak lateral deformation and maximum
and minimum axial forces in the isolators for use in preliminary design, via equations based on
response history analysis to an ensemble of ground excitations, and (3) to identify the conditions
Consider a rigid block with in-plane dimensions d (width) and h (height) supported on isolation
bearings, and subjected to lateral ground excitation üg (t) (Fig. 7.1). Planar motion of the block
is described by the lateral and vertical displacements uy and uz at the CM, and rocking angle
θx about the x-axis. With mass m uniformly distributed over the block, the rocking radius of
gyration is rx = (d2 + h2 )/12. The isolators are assumed to be evenly-spaced across the plan
in both directions. Each isolation bearing is located at (yi , −h/2) from the CM in the y- and
z-directions, respectively.
144
7.2.2 Bearing Models
The bearing models used in this chapter are derived from the equilibrium and kinematic equa-
tions developed previously for the two-spring bearing model – comprised of a shear spring and
a rotational spring (Fig. 6.2). Repeated here for the ith bearing, the equilibrium equations are
Pi − kbzi vi = 0 (7.1c)
representing the balance of lateral forces, moments, and vertical forces, respectively. The total
lateral and axial forces fbi and Pi are related to (a) the force fsi and deformation si of the shear
spring, and (b) the stiffness PE hb – the Euler buckling load times the bearing height – and
rotation θi through the rotational spring. Not included in the two-spring model (Fig. 6.2), an
additional spring with stiffness kbzi represents the axial flexibility of the bearing that permits
axial deformation vi .
ubi = si + hb θi (7.2a)
hb 2
ubzi = vi + si θi + θ (7.2b)
2 i
In the context of the two-spring model (Fig. 6.2), the total lateral deformation ubi is the sum
of the shear deformation si and the rotation θi multiplied by the bearing height (Eq. 7.2a),
while the total vertical deformation ubzi includes the axial deformation vi and an additional
vertical displacement resulting from the laterally deformed configuration (Eq. 7.2b). In contrast
to the global sign convention, the vertical deformation ubzi and force Pi are positive for bearing
compression.
1. coupled linear model: the shear spring is linear (fsi = kbi si in Eq. 7.1a)
145
2. coupled nonlinear constant-strength model: the shear spring follows a bilinear force-
deformation relation with initial stiffness kI , nominal post-yield stiffness kbi , and nominal
yield strength Qi
3. coupled nonlinear variable-strength model: like the constant-strength model, but the yield
where Po , typically about 1/3 of the bearing static load, can be determined from charac-
The latter two models will be abbreviated hereafter as the constant-strength and variable-
strength models. Each above model includes various axial-load effects, which cause local cou-
pling of the lateral and axial forces and deformations in individual bearings (Eq. 7.1). These
effects include variation of the total lateral stiffness – whether linear or nonlinear – with ax-
ial load, and variation of the vertical stiffness with lateral deformation. The variable-strength
model includes an additional variation of the yield strength with axial load, which was observed
When axial-load effects are neglected, Eq. 7.1 simplifies to the following:
where the lateral force-deformation relation (Eq. 7.4a) – defined previously (Eq. 3.2) and now
applied to individual bearings – and the linear vertical force-deformation relation (Eq. 7.4b)
• uncoupled linear model: the total lateral force-deformation relation is linear with stiffness
• uncoupled nonlinear model: the total lateral force-deformation relation is bilinear with
initial stiffness kI , post-yield stiffness kbi , and yield strength Qi , identical to that of the
146
are suitable for comparison with the three models defined above. The nominal lateral (kbi ) and
vertical (kbzi ) stiffnesses and strength Qi identified for the various bearing models are presently
Viscous damping in the vertical direction is included by an additional axial force com-
ponent for each bearing, fDi = cbzi u̇bzi where cbzi is the damping constant and u̇bzi the axial
bearing velocity. The damping coefficients cbzi are also assumed to be identical for each bearing.
When axial-load effects are included, the equations of motion for the system, subjected to
müy + fbi (ubi , ubzi ) = −müg (7.5a)
i
h
mrx2 θ̈x − yi cbzi u̇bzi + fbi (ubi , ubzi ) − yi Pi (ubi , ubzi ) = 0 (7.5b)
2
i i i
müz − cbzi u̇bzi − Pi (ubi , ubzi ) = mg (7.5c)
i i
The lateral and axial forces fbi and Pi in the bearing are implicit functions of its deformations
ubi and ubzi (Eqs. 7.1 and 7.2). (The dependence of fbi and Pi on ubi and ubzi is hereafter
implied.) In turn, the bearing deformations are related to the CM displacements (uy , uz and
θx ) by:
h
ubi = uy + θx (7.6a)
2
ubzi = −uz − yi θx (7.6b)
Substituting the time derivative of Eq. 7.6b for u̇bzi into Eq. 7.5, and normalizing Eq. 7.5b
by rx leads to
⎧ ⎫ ⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪
⎪
⎪ y ⎪⎪ ⎪
⎪ · ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎨
ü ⎪
⎬ ⎪
⎨
0 u̇ y ⎪
⎬ ⎪ ⎨ i fbi ⎪
⎬ ⎪ üg ⎪
⎨ ⎪
⎬
yi2
i cbzi rx2 · rx θ̇x ⎪
yi = −m 0
2rx fbi −
m rx θ̈x + + h (7.7)
⎪ ⎪ ⎪ ⎪ i i rx Pi ⎪ ⎪ ⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪ ⎪
⎩ ü ⎭ ⎩ ⎪ ⎪
⎭ ⎩ ⎪ ⎪
⎭ ⎩−g⎪
⎪ ⎭
z i cbzi · u̇z − i Pi
147
These equations are coupled in the three CM displacements (uy , θx and uz ), because the force-
deformation equations (Eq. 7.1) for individual bearings are coupled in the lateral and vertical
Substituting Eq. 7.4 into Eq. 7.7 gives the equations of motion:
⎧ ⎫ ⎧ ⎫ ⎡ ⎤⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪
⎪
⎪ y ⎪
⎪ ü ⎪
⎪ ⎪
⎪
⎪ 0 · u̇y ⎪
⎪
⎪ ⎢ 1 h
0 ⎥⎪
⎪
⎪ u y ⎪⎪
⎪ ⎪
⎪
⎪ i i i ⎪
γ z ⎪
⎪
⎪
⎪
⎪ üg ⎪
⎪
⎪
⎪
⎨ ⎬ ⎨ ⎬ ⎢
2rx
⎥ ⎨ ⎬ ⎨ ⎬ ⎨ ⎬
2⎢ h ⎥
rx θ̈x ⎪ + 2ω b ζθx Ωθx · rx θ̇x + ω 2
b ⎢ 2rx Ωθx 0 ⎥ ⎪rx θx ⎪ + µg h
i γi zi ⎪
= − 0⎪
⎪
⎪ ⎪ ⎪
⎪ ⎪
⎪ ⎣ ⎦⎪ ⎪ ⎪
⎪ 2rx
⎪ ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎩ üz ⎭ ⎩ ζz Ωz · u̇z ⎭ 0 0 Ω2z ⎩ uz ⎭ ⎩ 0 ⎭ ⎩−g⎪ ⎭
(7.8)
Due to uncoupling at the bearing level, the vertical displacement uz is no longer coupled with
was factored out of the nonlinear terms in Eq. 7.8 by introducing γi = Qi /Q.
Equation 7.8 has been written in terms of several parameters of a linear system (µ = 0):
(1) vertical (to lateral) and rocking (to lateral) frequency ratios:
ωbz ωbθx
Ωz = Ωθx = (7.10)
ωb ωb
cbz cbθx
ζz = ζθx = (7.12)
2mωbz 2mrx2 ωbθx
cbz = cbzi cbθx = yi2 cbzi (7.13)
i i
148
7.2.4 System Parameters
The equations of motion (Eq. 7.7) combined with those for the bearing model (Eqs. 7.1 and 7.2)
suggest that the response of the system depends on an extensive list of system and bearing prop-
erties. However, only a short list of parameters, which have been identified by trial and error,
significantly alter the response of the system when varied over their applicable ranges. These
parameters are listed below, and their ranges of variation considered are noted in parentheses:
Note that three of these parameters nearly define a linear system (Eq. 7.8 with µ = 0) and the
strength coefficient is relevant for nonlinear systems. As will be shown, this list of parameters
is both relevant and comprehensive when using bearing models that include axial-load effects
(Eq. 7.7).
The isolation period and strength coefficient were defined in Chap. 3. The vertical fre-
quency ratio Ωz (Eq. 7.10) is a function of the lateral and vertical stiffnesses kbi and kbzi
(Eq. 7.11), which are related by the ratio of the compression modulus to shear modulus Ec /G
(Eqs. 6.1 and 6.2). This ratio is a function of the bearing shape factor S (Eqs. 6.11 and 6.15),
where S = D/4t for a bearing with diameter D and thickness t of a single layer. If the shape
An applicable range of 10 to 30 for S and Kc = 0.004 (Appendix 7A) led to the range of Ωz
listed above (20 to 35). A typical shape factor of 20 leads to a vertical frequency ratio of 30.4.
Large shape factors (above 20) are commonly used today; such bearings are rather insensitive
to axial loads, but require large diameters to accommodate the design displacement, making
149
it difficult to limit the stiffness for adequate isolation. Bearings with small shape factors such
as 10 (Ωz ≈ 20) may provide some benefit of vertical isolation, but are more sensitive to axial
Substituting ωb and ωbθx (Eq. 7.11) into Eq. 7.10 gives the rocking frequency ratio as
2 12
1 h2 y kbzi
Ωθx = + 2 i i
(7.15)
4 rx2 rx i kbi
explicitly a function of many factors such as the bearing locations and stiffnesses, building
height, etc. However, it can be simplified when the bearings are identical and spaced evenly on
(in-plane).
For a fixed total lateral stiffness, the rocking frequency ratio Ωθx decreases as the number
of bearings increases (Fig. 7.2a), thus tending toward a more uniform distribution of stiffness
across the plan and less resistance to rocking. As expected, the rocking frequency ratio changes
most rapidly with number of bearings when this number is small, but seems to reach a plateau
for systems with more than 10 bearings across the plan (Fig. 7.2a). This suggests rocking
analysis of a simple system with only two bearings might not be realistic of more practical
Not surprisingly, the resistance to rocking decreases (Ωθx decreases) for more slender
buildings, as shown in Fig. 7.2a for different building slenderness – or height-to-width – ratios
h/d. The slenderness ratio affects the rocking frequency ratio through the values of h/rx and
d/rx (Eq. 7.16). The applicability of the building model used here to slenderness ratios h/d > 2
may be questionable (see Sec. 7.2.5); however, the application of base-isolation to such buildings
is considered to be rare.
Finally, the rocking frequency ratio Ωθx increases as the vertical frequency ratio Ωz in-
creases (Fig. 7.2b). In the limit as h/d → 0 and nby → ∞, Ωθx → Ωz (Eq. 7.16); thus the
150
(a) (b)
50 40
h/d=0.5
1
Rocking frequency ratio Ωθx
40 2
30 3
4
30
20
20
10
10
0 0
0 1 2 0 10 20 30 40
10 10 10
Number of bearings nby Vertical frequency ratio Ωz
Figure 7.2: Variation of rocking frequency ratio Ωθx with (a) number of bearings across the
plan in-plane (Ωz = 30.4 for S = 20) and (b) vertical frequency ratio Ωz (nby = 10).
rocking frequency ratio is close to the vertical frequency ratio for a squat building with suffi-
ciently large number of bearings, as reflected in Fig. 7.2b for h/d = 0.5. For larger slenderness
ratios, the rocking frequency takes an intermediate value between the lateral frequency and the
vertical frequency.
In summary, over the range of nby (≥ 2), building slenderness ratio h/d (0.5 to 4), and
Ωz (20 to 35), the rocking frequency ratio Ωθx varies from about 5 to 53, where the lower range
reflects slender buildings with many bearings distributed across the plan while the upper range
reflects squat buildings with fewer bearings. For details of how all the parameters in Eq. 7.7
were selected for response history analysis (RHA), consult Appendix 7A.
The validity of the rigid structure approximation comes into particular question for slender,
longer period structures. Assuming the structure to be rigid is accurate only if the fundamental
fixed-base period Ts of the structure is much shorter than the isolation period Tb [21, Chap. 20].
As a guideline, the International Building Code [1] suggests that Tb > 3Ts (required for design
based solely on static analysis), and for this range the rigid structure approximation is certainly
valid. This implies that for the range of 1.5 to 5 seconds considered for Tb , the upper limit for
151
An examination of the natural vibration periods of existing buildings measured from their
recorded response during earthquakes (Appendix 7B) indicates that most slender buildings
(h/d > 2) have periods longer than 1.66 seconds. However, these buildings were all fixed-base,
while isolated structures are required by code [1] to be stronger and stiffer. This results in
superstructure periods that are shorter than comparable fixed-base structures, and more likely
to fall within the limit of Ts = 1.66 seconds. The simplified procedure to estimate the bearing
deformation and forces presented later for preliminary design application assumes the structure
Response histories are compared for the same base-isolated structure with 5 different bearing
models: uncoupled linear (Fig. 7.3), coupled linear (Fig. 7.4), uncoupled nonlinear (Fig. 7.5),
constant-strength (Fig. 7.6), and variable-strength (Fig. 7.7). The sample structure has slen-
derness h/d = 2, supported by two bearings at the edges. Considered for each bearing model
are (a) harmonic excitation with frequency ω = 4π (chosen to be distinct from the resonant
frequency ωb = π or Tb = 2 sec) and acceleration amplitude ωvo where vo = 40 cm/s; and (b)
the strong component of LMSR Record No. 20, with peak ground velocity u̇go = 61.5 cm/s.
For this demonstration only, viscous damping is omitted from the analyses.
To summarize, the response histories shown are the lateral (uy ), rocking (rx θx ) and
vertical (uz /ust ) displacements of the CM of the block, and the axial deformation (ubz1 /ust )
and force (P1 /Pst ) in the left bearing (representative of a typical exterior bearing), where
all vertical/axial responses have been scaled relative to their static values. Also shown are
power spectra – appropriately scaled complex moduli of the discrete Fourier transforms – of the
response components uy and rx θx , and the estimated variation of the lateral stiffness kb (t)/kb
(Eq. 6.7, summed over both bearings) and vertical stiffness kbz (t)/kbz (Eq. 6.17) relative to
The most significant outcome of including axial-load effects (coupled linear model, Fig. 7.4)
relative to the uncoupled linear model (Fig. 7.3) is to greatly increase the bearing axial defor-
152
mation ubz1 , especially in compression. For harmonic excitation, this deformation increase
is a result of inciting vertical motion uz at the CM; for the recorded excitation, the rocking
bearing axial force P1 increases by only a small amount (recorded excitation only), and may
The system with the uncoupled linear bearing model – i.e., linear behavior in both the
lateral and vertical directions with no interaction between them – is well understood [51]. The
frequencies and mode shapes of the lateral and rocking degrees of freedom: ω1 = 3.13, ω2 = 65.0
and φ1 = 0.996, −0.004T , φ2 = 0.004, 0.996T , are nearly uncoupled. With tall spikes near
ω1 and barely visible or nonexistent spikes near ω2 , the power spectra (Fig. 7.3a,b) indicate
that the first mode, which is primarily lateral translation (uy ), dominates. (The second spike
For harmonic excitation, the coupled linear model (Fig. 7.4a) causes the power spectra
and the lateral (uy ) and rocking (rx θx ) displacements to change only negligibly compared to
the uncoupled linear model. Thus, despite the influence of axial load on the lateral stiffness
and the influence of lateral deformation on the vertical stiffness – both small, as reflected in
the plots for kb and kbz – this system vibrates in approximately the same modes as when axial-
load effects are neglected (uncoupled linear model). The variation in vertical stiffness, however,
causes vertical motion of the CM, which contributes to an increase in the left bearing axial
For the recorded excitation, the power spectra for the coupled linear model (Fig. 7.4b)
reflects a second spike in the rocking displacement at a much lower frequency than the second
mode frequency, an indication that the axial-load effects are significant enough to cause the
“modes” of the system to shift. While the first mode still dominates, it now contributes more
to the rocking component, and combined with participation of the second mode, the total
amplitude of rocking rx θx grows significantly. The modal shift seems to occur because this
strong ground motion simply shakes the system harder, causing large variation in the lateral
stiffness kb due to much larger axial forces, and large variation in the vertical stiffness kbz due
153
(a)
1
5 uy
uy (cm)
0.8 rx θ x
0
Power spectra
−5 0.6
rx θx (cm)
0.025
0.4
0
0.2
−0.025
0
1.5 0 20 40 60 80 100
uz /ust
Frequency ω (rad/s)
1
1
kb (t)/kb
1.3
ubz1 /ust
0.66
1
0.33
0.7
1
kbz (t)/kbz
1.3
P1 /Pst
0.66
1
0.33
0.7
0 5 10 15 20 0 5 10 15 20
Time (sec) (b) Time (sec)
1
100 uy
uy (cm)
0.8 rx θ x
0
Power spectra
−100 0.6
rx θx (cm)
6
0.4
0
0.2
−6
30 0
0 20 40 60 80 100
uz /ust
Frequency ω (rad/s)
1
1
kb (t)/kb
60
ubz1 /ust
0.66
30
0.33
0
1
kbz (t)/kbz
4
P1 /Pst
0.66
0
0.33
−4
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Time (sec) Time (sec)
Figure 7.3: Various response histories of the isolated system with rocking, and power spectra
for the lateral and rocking displacements uy and rx θx ; due to lateral excitations: (a) harmonic
with ω/ωb = 4 and (b) LMSR Record No. 20. The system uses the uncoupled linear bearing
model, and has properties Tb = 2 seconds, S = 15, S2 = 3, h/d = 2, nby = 2, and no viscous
damping.
154
(a)
1
5 uy
uy (cm)
0.8 rx θ x
0
Power spectra
−5 0.6
rx θx (cm)
0.025
0.4
0
0.2
−0.025
0
1.5 0 20 40 60 80 100
uz /ust
Frequency ω (rad/s)
1
1
kb (t)/kb
1.3
ubz1 /ust
0.66
1
0.33
0.7
1
kbz (t)/kbz
1.3
P1 /Pst
0.66
1
0.33
0.7
0 5 10 15 20 0 5 10 15 20
Time (sec) Time (sec)
(b)
1
100 uy
uy (cm)
0.8 rx θ x
0
Power spectra
−100 0.6
rx θx (cm)
6
0.4
0
0.2
−6
30 0
0 20 40 60 80 100
uz /ust
Frequency ω (rad/s)
1
1
kb (t)/kb
60
ubz1 /ust
0.66
30
0.33
0
1
kbz (t)/kbz
4
P1 /Pst
0.66
0
0.33
−4
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Time (sec) Time (sec)
Figure 7.4: Various response histories of the isolated system with rocking, and power spectra
for the lateral and rocking displacements uy and rx θx ; due to lateral excitations: (a) harmonic
with ω/ωb = 4 and (b) LMSR Record No. 20. The system uses the coupled linear bearing
model, and has properties Tb = 2 seconds, S = 15, S2 = 3, h/d = 2, nby = 2, and no viscous
damping.
155
to a much larger lateral deformation. Also related to the vertical motion at the CM, the left
bearing axial deformation in compression increases to over 60 times its static value, this time
When comparing the three nonlinear models, the consequences of including axial-load
effects – constant-strength (Fig. 7.6) or variable-strength (Fig. 7.7) models – compared to the
nonlinear uncoupled model (Fig. 7.5) are more subtle than for the linear bearing models. Like
the linear models, an increase in rocking displacement rx θx relative to the lateral displacement
uy is observed, which, when combined with vertical motion at the center of mass, increases
the left bearing axial deformation ubz1 , especially in compression. However, the deformation
increase is much smaller than when the bearings were linear, and the axial forces do not appear
to be affected.
The response of the system with the uncoupled nonlinear bearing model due to harmonic
excitation (Fig. 7.5a) is relatively straightforward, still vibrating in predominantly two modes.
In lateral motion, it reaches a steady state of vibration at the excitation frequency after the
transient excitation associated with the fundamental frequency has damped out. The second
mode influence is stronger here than for the uncoupled linear bearing model, with a higher
frequency reflected in the rocking component rx θx history as well as the power spectra. The
response of the system with the constant-strength model (Fig. 7.6a), is very similar to that with
the uncoupled nonlinear bearing model. Because the lateral deformation is small, the variation
in vertical stiffness kbz is modest, causing only a small increase in the rocking displacement rx θx
and a small vertical motion uz at the CM. Slight increases in the left bearing axial deformation
ubz1 and force P1 appear to be proportional to the increase in rocking displacement, as the
In response to recorded excitation, even the simplest nonlinear model – uncoupled non-
linear model (Fig. 7.5b) – displays much more complicated behavior. The power spectra shows
that instead of two or three frequencies, a range of frequencies are excited, i.e., the response
frequency shifts as the system is excited at different intensities. Consistent with earlier obser-
vation, the rocking component contains higher frequencies reflecting higher mode contribution.
156
The vertical softening or decrease in vertical stiffness kbz with axial-load effects (constant-
strength model, Fig. 7.6b) is more significant here, and similarities with the linear system
response to axial-load effects (Fig. 7.4b) are observed. This includes increased rocking rx θx ,
significant vertical motion uz of the CM, and a marked increase in the peak axial deforma-
tion ubz1 especially on the compression side, with little increase in the corresponding force P1
(Fig. 7.6b), all relative to the uncoupled nonlinear model (Fig. 7.5b).
For the final example – the response of the system using the variable-strength model,
kbz (t) has been replaced by the strength variation Q(t) summed over both bearings and divided
by the nominal strength Q (Fig. 7.7). Since an occurrence of tension that induces a complete
loss of bearing strength is balanced by compression on the other side of the building, Q/Qo
is not likely to fall below 0.5, and indeed does not in this example (Fig. 7.7). Relative to the
constant-strength model (Fig. 7.6), inclusion of the strength variation has a minimal effect on
the response. It causes an increase in the lateral displacement uy due to diminished energy
dissipation, and an increase in the peak axial deformation ubz1 but a decrease in the peak axial
For further understanding, the frequency of the harmonic excitation is varied over a wide
range. The peak lateral uyo and rocking displacement rθxo , and the maximum and minimum
bearing axial deformations ubz± /ust and forces P± /Pst are plotted against the excitation fre-
quency for the two linear (Fig. 7.8) and three nonlinear (Fig. 7.10) bearing models. By extending
the duration of the excitation to two minutes, any resonances that may exist will be apparent.
The duration is irrelevant at non-resonance where the system quickly reaches steady state.
With the uncoupled linear bearing model, the system has a resonance at its first mode
frequency, approximately equal to the isolation frequency ωb (ω/ωb = 1, Fig. 7.8). Including
axial-load effects (coupled linear bearing model) appears to shift the resonance (more dominant
in rocking rx θx ) to a lower frequency (ω/ωb = 0.8), unsurprising since both the lateral and
vertical stiffnesses of the bearings tend to decrease. However, it will be shown that with axial-
load effects, the system does not reach resonance, but instead arrives at a steady state response.
To begin, the displacements uy and rx θx of this system with ω/ωb = 0.8 (Fig. 7.9a and
157
(a)
1
6 uy
uy (cm)
0.8 rx θ x
0
Power spectra
−6 0.6
rx θx (cm)
0.15
0.4
0
−0.15 0.2
0
1.2 0 20 40 60 80 100
uz /ust
Frequency ω (rad/s)
1
kb (t)/kb
2
ubz1 /ust
1
0.9
0
1
kbz (t)/kbz
2
P1 /Pst
0.66
1
0.33
0
0 5 10 15 20 0 5 10 15 20
Time (sec) (b) Time (sec)
1
20 uy
uy (cm)
0.8 rx θ x
0
Power spectra
−20 0.6
rx θx (cm)
0.4
0.4
0
0.2
−0.4
0
3 0 20 40 60 80 100
uz /ust
Frequency ω (rad/s)
1
1
kb (t)/kb
5
ubz1 /ust
2.5
0.9
0
1
kbz (t)/kbz
2
P1 /Pst
0.66
1
0.33
0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Time (sec) Time (sec)
Figure 7.5: Various response histories of the isolated system with rocking, and power spectra
for the lateral and rocking displacements uy and rx θx ; due to lateral excitations: (a) harmonic
with ω/ωb = 4 and (b) LMSR Record No. 20. The system uses the uncoupled nonlinear bearing
model, and has properties Tb = 2 seconds, µ = 0.1, S = 15, S2 = 3, h/d = 2, nby = 2, and no
viscous damping.
158
(a)
1
6 uy
uy (cm)
0.8 rx θ x
0
Power spectra
−6 0.6
rx θx (cm)
0.15
0.4
0
−0.15 0.2
0
1.2 0 20 40 60 80 100
uz /ust
Frequency ω (rad/s)
1
kb (t)/kb
2
ubz1 /ust
1
0.9
0
1
kbz (t)/kbz
2
P1 /Pst
0.66
1
0.33
0
0 5 10 15 20 0 5 10 15 20
Time (sec) (b) Time (sec)
1
20 uy
uy (cm)
0.8 rx θ x
0
Power spectra
−20 0.6
rx θx (cm)
0.4
0.4
0
0.2
−0.4
0
3 0 20 40 60 80 100
uz /ust
Frequency ω (rad/s)
1
1
kb (t)/kb
5
ubz1 /ust
2.5
0.9
0
1
kbz (t)/kbz
2
P1 /Pst
0.66
1
0.33
0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Time (sec) Time (sec)
Figure 7.6: Various response histories of the isolated system with rocking, and power spectra
for the lateral and rocking displacements uy and rx θx ; due to lateral excitations: (a) harmonic
with ω/ωb = 4 and (b) LMSR Record No. 20. The system uses the nonlinear constant-strength
bearing model, and has properties Tb = 2 seconds, µ = 0.1, S = 15, S2 = 3, h/d = 2, nby = 2,
and no viscous damping.
159
(a)
1
6 uy
uy (cm)
0.8 rx θ x
0
Power spectra
−6 0.6
rx θx (cm)
0.15
0.4
0
−0.15 0.2
0
1.2 0 20 40 60 80 100
uz /ust
Frequency ω (rad/s)
1
kb (t)/kb
2
ubz1 /ust
1
0.9
0
2 1
Q(t)/Q
P1 /Pst
1
0.75
0
0 5 10 15 20 0 5 10 15 20
Time (sec) (b) Time (sec)
1
20 uy
uy (cm)
0.8 rx θ x
0
Power spectra
−20 0.6
rx θx (cm)
0.4
0.4
0
0.2
−0.4
0
3 0 20 40 60 80 100
uz /ust
Frequency ω (rad/s)
1
1
kb (t)/kb
5
ubz1 /ust
2.5
0.9
0
2 1
Q(t)/Q
P1 /Pst
1
0.75
0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Time (sec) Time (sec)
Figure 7.7: Various response histories of the isolated system with rocking, and power spectra
for the lateral and rocking displacements uy and rx θx ; due to lateral excitations: (a) harmonic
with ω/ωb = 4 and (b) LMSR Record No. 20. The system uses the nonlinear variable-strength
bearing model, and has properties Tb = 2 seconds, µ = 0.1, S = 15, S2 = 3, h/d = 2, nby = 2,
Pst /Po = 3, and no viscous damping.
160
3000 600
∞ Uncoupled linear
2500 Coupled linear
400
200
1500
0
1000
500 −200
0 −400
−1 0 1 2 −1 0 1 2
10 10 10 10 10 10 10 10
80 60
40
60
P+ /Pst , P− /Pst
20
rθxo (cm)
40 0
−20
20
−40
0 −60
−1 0 1 2 −1 0 1 2
10 10 10 10 10 10 10 10
Excitation to isolation frequency (ω/ωb ) Excitation to isolation frequency (ω/ωb )
Figure 7.8: Various peak responses of system with linear bearing models (uncoupled and
coupled) due to harmonic excitation frequency sweep. System properties are Tb = 2 seconds,
S = 15, S2 = 3, h/d = 2, nby = 2, and no viscous damping.
b) do not grow without bound as would a resonant response, but instead appear to oscillate
between two states. Multiple spikes in the power spectrum of uy (Fig. 7.9c) indicate that
distinct frequencies near the so-called resonance frequency are being excited. To understand
this behavior, the displacement history uy has been split into subintervals at points where its
frequency appears to undergo small changes (Fig. 7.9a). The power spectra of subintervals 1
and 3, and subintervals 2 and 4, respectively, are shown in Fig. 7.9d and e, which reflect distinct
frequencies (albeit only slightly different) for each. Thus, power spectra of the displacement
broken into appropriate subintervals show that a single frequency is excited in each subinterval.
In this sense, the steady state response of this system is defined by the continual transition
between two states defined by the distinct frequencies. Our present understanding is that this
161
(a) (b)
500 100
int 1 int 2
50
rθx (cm)
uy (cm)
0 0
−50
int 3 int 4
−500 −100
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time (s) Time (s)
ωp = 2.22
0.6 0.6 0.6
0 0 0
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
Frequency ω (rad/s) Frequency ω (rad/s) Frequency ω (rad/s)
Figure 7.9: (a,b) Response histories of lateral displacement uy and rocking rx θx of system with
coupled linear model for isolation bearings to harmonic excitation with frequency ω = 0.8ωb ;
(c) power spectrum of displacement uy , and (d,e) power spectra of displacement uy broken into
subintervals 1 (11.5 - 28.4 sec) and 3 (59 - 76.1 sec), and 2 (31.1 - 56.5 sec) and 4 (78.5 - 103.9
sec), with frequency ωp of peak indicated.
shifting between states is caused by changes in stiffness with response amplitude induced by
axial-load effects. The most important point is that, although composed entirely of linear
The peak (over all frequencies) rocking displacement rθxo is two orders of magnitude
smaller for the uncoupled bearing model but the same order of magnitude for the coupled
bearing model as the lateral displacement uyo (Fig. 7.8), indicating that axial-load effects greatly
increase the tendency of the system to rock. While the corresponding peaks of axial deformation
(maximum ubz+ and ubz− ) are much larger for the coupled bearing model, the corresponding
axial forces are larger for the uncoupled bearing model. This implies that the peak forces in
162
100 60
Uncoupled nonlinear
Constant-strength
80 Variable-strength 40
60 20
40 0
20 −20
0 −40
−1 0 1 2 −1 0 1 2
10 10 10 10 10 10 10 10
6 20
5
10
P+ /Pst , P− /Pst
4
rθxo (cm)
3 0
2
−10
1
0 −20
−1 0 1 2 −1 0 1 2
10 10 10 10 10 10 10 10
Excitation to isolation frequency (ω/ωb ) Excitation to isolation frequency (ω/ωb )
Figure 7.10: Various peak responses of system with nonlinear bearing models (uncoupled
nonlinear, constant-strength and variable-strength) due to harmonic excitation frequency sweep.
System properties are Tb = 2 seconds, µ = 0.1, S = 15, S2 = 3, h/d = 2, nby = 2, Pst /Po = 3,
and no viscous damping.
the bearings may be lower than those predicted by a typical uncoupled linear model, but this
was not true for the earlier example (Figs. 7.3 and 7.4). Notice that for the uncoupled linear
model, the maximum and minimum axial deformations are symmetric about ust (ubz /ust = 1),
likewise with axial forces. But for coupled bearings, vertical displacement of the CM disturbs
the symmetry, and peak deformations and forces in compression tend to be comparatively larger
The system with the uncoupled nonlinear bearing model again has a resonance very
close to the isolation frequency ωb (Fig. 7.10). This resonance occurs only if the excitation is
sufficiently strong to overcome the hysteretic energy dissipation, which becomes negligible as the
response amplitude grows large. Again, it can be shown that use of either the constant-strength
163
or variable-strength models eliminates this resonance. However, with the uncoupled nonlinear
bearing model, the resonance builds up so slowly that its component responses (even lateral
displacement uyo ) are exceeded by the corresponding non-resonant responses of the system
exists that the typical uncoupled nonlinear bearing model could significantly underestimate the
peak response of the system. The various spikes in response that occur at frequencies between
The normalization procedure of Chapters 3-5 is extended to the current system, coupled in its
lateral, rocking and vertical DOFs. The normalized equations of motion are:
⎧ ⎫ ⎡ ⎤⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪
⎪
⎪ ¨ y ⎪ ⎪ ⎪ ˙ ⎪ ⎪ ⎪ ⎪ ¨g ⎪
⎪
⎪
⎨
ū ⎪
⎬ ⎢
0 0 0 ⎥⎪⎪
⎨
ū y ⎪ ⎪
⎬
⎪
⎪
⎨
f
i bi ⎪
⎪
⎬
⎪ ū
⎨ ⎪
⎪ ⎪
⎬
⎢ ⎥ ω 2
ωb2
¨ + ⎢ y
cbzi i
2
⎥ ˙ + b h
− yi = − 0 ⎪ (7.17)
⎪ r θ̄ ⎪ ⎢ 0 0 ⎥ r θ̄ f P
⎦⎪ ⎪ Q ⎪ ⎪ ⎪
x x i m rx 2 x x i 2rx bi i rx i
⎪ ⎪ ⎣ ⎪ ⎪ ⎪ ⎪ ηωd ⎪ ⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎩ ū ¨z ⎭ 0 0 c bz ⎩ ū˙ ⎭ ⎩ − i Pi ⎭ ⎩−ḡ⎪⎭
m z
They were obtained by first dividing Eq. 7.7 by u∗y = Q/kb (Eq. 3.6) such that all degrees of
freedom are normalized – e.g. ūy = uy /u∗y – and then substituting the normalized strength
η = a∗y /ωd u̇go (Eq. 3.8) into the right-hand side. This normalizes the ground acceleration by
¨g = üg /u̇go ), such that the corresponding ground velocity varies between
its peak velocity (ū
-1 and +1. Gravity, applied in the vertical direction, is also normalized (ḡ = g/u̇go ). This
normalization allows the normalized strength η to replace the strength coefficient µ as one of
The benefit of normalization is not obvious for this system, because in contrast to Chaps. 3
(Eq. 3.9) and 4 (Eq. 4.17), the strength Q does not drop out of the normalized equations
(Eq. 7.17), and ḡ varies from excitation to excitation. However, as will be seen later, the
normalization reduces the dispersion in the response over a ground motion ensemble, and is
therefore effective.
164
7.4.2 Computation of Median and Dispersion
The median and dispersion of a system response r over the ground motion ensemble are deter-
mined by: (1) computing the peak normalized response by nonlinear RHA of Eq. 7.17 to each
ground motion in the ensemble, (2) computing its median (Eq. 2.1) and dispersion (Eq. 2.2)
over the ensemble, and (3) multiplying the median of the normalized response by u∗y to obtain
the median value of the actual response as a function of η and median ground velocity u̇go of
the ensemble.
The responses of interest are the peak bearing deformation ubo , and the maximum and
minimum axial bearing forces relative to the static force (P+ /Pst and P− /Pst ), since actual
forces vary widely depending on the weight of the structure. Because the axial force oscillates
around the static force rather than zero, the positive and negative force increments, defined as
the maximum deviation of force in either direction from the static force:
are more meaningful for statistical analysis. Also, using force increments avoids the compli-
cation of a negative value of P− in case of bearing tension, since the equations for median
and dispersion are only meaningful for positive values. The dispersion measure of the force
increments ∆P± represents the variation of the actual forces P± , whose dispersion cannot be
computed, while the median forces are computed from the median force increments (Eq. 7.18).
Compared in Fig. 7.11 are the dispersion measures for actual and normalized response quantities
for a rigid block supported on bearings represented by the variable-strength model. System
parameters Tb and η are varied but the others are held constant, as their variation had little
Normalization clearly reduces the dispersion of the lateral bearing deformation, which
varies from 0.3 to 0.6 for normalized ūbo and from 0.4 to 1.0 for actual ubo (Fig. 7.11a); these
values are very consistent with the values observed in Chap. 3. The normalized force increments
165
Dispersion of Normalized Responses
1 1 1
(a) (b) η = 0.25 (c)
0.8 0.8 0.5 0.8
0.75
1.0
∆P̄+
∆P̄−
0.6 0.6 0.6
1.5
ūbo
0 0 0
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
∆P−
0.6 0.6 0.6
ubo
0 0 0
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec) Tb (sec)
Figure 7.11: Dispersion over the LMSR ensemble compared for normalized and actual re-
sponses: (a) lateral deformation ubo , (b) positive force increment ∆P+ , (c) negative force in-
crement ∆P− . System parameters are: Ωz = 30.4, Ωθx = 37.2.
∆P̄+ and ∆P̄− (Fig. 7.11b,c) have been reduced to dispersion values varying between 0.1 and
0.5, compared to values varying between 0.4 and 0.8 for the actual force increments. Fortunately,
the added complexity of the system does not nullify the benefit of normalization.
Investigated next is the variation of the peak bearing responses (lateral deformation ubo and
axial forces P± /Pst ) with the parameters identified previously: isolation period Tb , normalized
strength η, and vertical-to-lateral and rocking-to-lateral frequency ratios Ωz and Ωθx . Median
values due to the strong-component ground motion ensemble (Sec. 2.1) with median ground
velocity = 35 cm/s are computed by the method of Sec. 7.4.2. The axial force increment
∆P+ /Pst is maximized over the median values computed for the exterior bearings on each
side of the building; likewise for the minimum force increment ∆P− /Pst . Unless otherwise
indicated, the bearings are represented by the nonlinear variable-strength model, which is the
166
(a) Ωθx = 37.2 (b) Ωθx = 14.6
15 15
η = 0.25 η = 0.25
10 10
ubo (cm)
0.5 0.5
0.75 0.75
1.0 1.0
5 5
1.5 1.5
Rocking
No Rocking
0 0
0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec)
Figure 7.12: Median response spectra for bearing deformation ubo : (a) Ωθx = 37.2 and (b)
Ωθx = 14.6; Ωz = 30.4.
Rocking of the structure and axial-load effects in bearings have little influence on the lateral
deformation of the bearings, even for taller isolated buildings with significant rocking. Response
spectra of the lateral deformation ubo for a squat building (Ωθx = 37.2, Fig. 7.12a) and a more
slender building (Ωθx = 14.6, Fig. 7.12b) are nearly identical, suggesting that the rocking
properties are irrelevant for determining lateral deformation. Furthermore, these spectra are
very similar to the spectrum determined previously from analysis of the system permitting
lateral motion only, i.e., no rocking (rx θx ) or vertical (uz ) motions (Fig. 3.8); this spectrum is
The deformation increases moderately with an increase in isolation period Tb and de-
creases with an increase in normalized strength η (or energy dissipation), but the maximum
and minimum axial forces (P± /Pst ) have a more complex dependence on Tb and η (Fig. 7.13).
The maximum forces for different normalized strengths are similar when the isolation period Tb
is small (< 2 seconds), but become more disparate as Tb increases, tending to decrease for small
strengths and increase for large strengths. The force spectra also vary a lot for two different
167
(a) Ωθx = 37.2 (b) Ωθx = 14.6
2 3
η = 0.25
0.5
1.8 0.75 2.5
1.0
1.6 1.5
P+ /Pst
2
1.4
1.5
1.2
1 1
0 1 2 3 4 5 0 1 2 3 4 5
1 1
0.8
0.5
0.6
P− /Pst
0
0.4
−0.5
0.2
0 −1
0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec)
Figure 7.13: Median response spectra for axial forces P+ /Pst and P− /Pst : (a) Ωθx = 37.2 and
(b) Ωθx = 14.6; Ωz = 30.4.
For given parameters, the maximum and minimum axial force increments ∆P+ /Pst and
∆P− /Pst are nearly identical (but not exactly identical due to vertical motion of the CM and
varying stiffness of the bearings caused by axial-load effects), which causes the minimum axial
forces to appear as a mirror image of the maximum across the line P/Pst = 1 in Fig. 7.13.
Because of their similarity, the maximum and minimum forces may be referred to hereafter
in conjunction as the force extrema. The fact that the force extrema do not always increase
(maximum force increases while minimum force decreases) with increasing isolation period
seems to contradict the expectation that overturning effects are greater in longer period (taller)
buildings. However, because the structure has been modeled as rigid, the building slenderness
is independent of the isolation period, which may be a limitation of this analysis (Sec. 7.2.5).
168
Although increasing the yield strength of the bearings (increasing η) is known to control
lateral bearing deformation (Fig. 7.12), it does so at the expense of increasing the axial force
extrema (Fig. 7.13). For instance, a large normalized strength (η = 1.5) reduces the deformation
ubo by more than half compared to η = 0.25 (Fig. 7.12), but increasing η above 0.8 causes bearing
tension in the more slender building (P− /Pst < 0, Fig. 7.13b). Thus, limiting the normalized
The responses examined are essentially independent of the vertical frequency ratio Ωz , which
depends on the bearing shape factor S and therefore reflects the relative bearing dimensions
(diameter and rubber thickness). This means that the current practice of using high shape
factor bearings to avoid stability problems may be unnecessary, and the benefits of smaller shape
factors – vertical isolation and accommodation of larger deformations – can be realized. Both
lateral bearing deformations and axial forces show almost no variation with vertical frequency
ratio for slenderness ratios within the practical limit of h/d ≤ 2 (Fig. 7.14). Thus, although
some vertical bouncing of the building is predicted by the variable-strength model (Sec. 7.3),
it appears not to influence the axial forces (due to vertical softening, only axial deformations
are affected). For the rare more slender building (h/d = 4, L3 and L6 in Fig. 7.14), the vertical
frequency ratio Ωz begins to influence the bearing axial forces, although with no systematic
trends.
The bearing axial force extrema (P± /Pst ) are influenced by the rocking frequency ratio
Ωθx but not independently by the slenderness ratio h/d or number of bearings across the plan. In
other words, the force extrema are nearly the same in systems with different slenderness ratios
and bearing distributions so long as their rocking frequency ratios are the same. Evidence
is shown by the limited scatter in the forces when plotted against rocking frequency ratio
(Fig. 7.15a), which includes data for many different combinations of slenderness ratio and
number of bearings.
Due to large values and rapid variation of the axial force extrema in the range of low
frequency ratios, this range (say Ωθx < 15) is best avoided. Buildings with lower frequency
169
(a) (b) (c)
15 3.5 1
L1 L4
L2 L5
L3 L6 3 0.5
10
ubo (cm)
2.5 0
P+ /Pst
P− /Pst
2 −0.5
5
1.5 −1
0 1 −1.5
15 20 25 30 35 40 15 20 25 30 35 40 15 20 25 30 35 40
Vertical frequency ratio Ωz Vertical frequency ratio Ωz Vertical frequency ratio Ωz
Figure 7.14: Median values of (a) bearing lateral deformation ubo and (b,c) axial forces P+ /Pst
and P− /Pst plotted against the vertical frequency ratio Ωz . Other parameters are: Tb = 2 sec
and η = 0.5 (L1-L3), or Tb = 4 sec and η = 1.0 (L4-L6); Ωθx , which varies with Ωz , is determined
by h/d = 2 and nby = 2 (L1,L4), h/d = 2 and nby = 15 (L2,L5), or h/d = 4 and nby = 15
(L3,L6).
ratios (flexible in rocking relative to lateral motion) include very slender buildings that are
susceptible to overturning, and/or buildings with many bearings distributed across the plan,
which reduces their resistance to rocking. Rapid variation of the force extrema in this low
frequency range (Fig. 7.15a) makes them difficult to predict, whereas only a small change
in rocking frequency ratio marks the difference between compression and significant tension.
Overall, the force extrema increase approximately quadratically as the rocking frequency ratio
decreases (Fig. 7.15a). When building slenderness h/d and number of bearings nby are varied
independently, the force extrema vary linearly with h/d and increase rapidly with nby when nby
A simplified model with only two bearings – at the building edges – should not be used
to estimate the axial forces in edge bearings of realistic systems with many bearings. The axial
force extrema for a two-bearing system are much lower than the others considered (5, 9, and
15, Fig. 7.15b). In fact, the data outliers in Fig. 7.15a are for two bearing systems, and hence
can be disregarded.
As demonstrated earlier (Sec. 7.5.1), the lateral deformation ubo of the bearings is essen-
tially unaffected by rocking properties except in the range of rocking frequency ratios Ωθx < 15
(Fig. 7.15a), or slenderness ratios h/d > 2 (Fig. 7.15b), unimportant since such slender buildings
170
(a) (b)
15 15
10 10
ubo (cm)
5 5
0 0
0 10 20 30 40 50 0 1 2 3 4 5
3.5 3.5
nby =2
3 3 5
9
15
P+ /Pst
2.5 2.5
2 2
1.5 1.5
1 1
0 10 20 30 40 50 0 1 2 3 4 5
1 1
0.5 0.5
P− /Pst
0 0
−0.5 −0.5
−1 −1
−1.5 −1.5
0 10 20 30 40 50 0 1 2 3 4 5
Rocking frequency ratio Ωθx Slenderness ratio h/d
Figure 7.15: Median bearing deformation ubo and axial forces P+ /Pst and P− /Pst plotted
against: (a) rocking frequency ratio Ωθx and (b) slenderness ratio h/d for several values of nby ;
Tb = 2 sec, η = 0.5, and Ωz = 30.4.
We have seen that axial force extrema vary with the rocking frequency ratio Ωθx due
to changes in building slenderness or number of bearings (Fig. 7.15a), but do not vary with
vertical frequency ratio Ωz (Fig. 7.14). Because the rocking frequency is not independent of the
vertical frequency, the force extrema plotted against rocking frequency ratio show a wider band
of scatter when the vertical frequency ratio is varied (Fig. 7.16a), implying that the rocking
frequency ratio alone does not adequately represent the variation of forces. Instead, consider
171
(a) (b)
3 3
Ωz =20.7
26.7
2.5 30.4 2.5
32.7
34.3
P+ /Pst
2 2
1.5 1.5
1 1
0 10 20 30 40 50 60 0 0.5 1 1.5 2
1 1
0.5 0.5
P− /Pst
0 0
−0.5 −0.5
−1 −1
0 10 20 30 40 50 60 0 0.5 1 1.5 2
Rocking frequency ratio Ωθx Rocking-to-vertical frequency ratio Υx
Figure 7.16: For systems with different vertical frequency ratios Ωz , variation of median
axial forces P+ /Pst and P− /Pst with (a) rocking frequency ratio Ωθx and (b) rocking-to-vertical
frequency ratio Υx ; Tb = 2 seconds and η = 0.5.
ωbθx Ωθx
Υx = ≡ (7.19)
ωbz Ωz
(Fig. 7.16b). The new parameter Υx is less sensitive than Ωθx to changes in Ωz , such that the
Comparison of the different bearing models demonstrate that the axial-load effects included in
the constant-strength model are unimportant, and only occasional circumstances merit the use
of the most rigorous variable-strength model. Thus, in most cases, the uncoupled nonlinear
bearing model can be used to estimate seismic demands with no loss of accuracy.
172
The peak lateral deformation of the bearing and force extrema are primarily unaffected
by the time variation of lateral and vertical bearing stiffness resulting from axial-load effects.
This claim is shown to be true in Fig. 7.17a and b, which compares the responses given by the
constant-strength and uncoupled nonlinear models for two rocking-to-vertical frequency ratios:
Υx = 1.22 (Fig. 7.17a) and Υx = 0.48 (Fig. 7.17b); and can be extended to all values of Υx
with confidence.
The time variation of bearing strength, which distinguishes the variable-strength model
from the constant-strength model, has only a minor effect on the lateral deformation and
axial force extrema over a wide range of system parameters; however, the variable-strength
model with its additional refinement is recommended for certain combinations of normalized
strength and rocking-to-vertical frequency ratio. Variation in bearing strength causes the lateral
deformation to increase slightly and the force extrema to decrease (Fig. 7.17c,d). The reduction
in the force extrema is greater for large values of the normalized strength η and for smaller
frequency ratios [compare Υx = 0.48 (Fig. 7.17d) to Υx = 1.22 (Fig. 7.17c)]. The increase in
The axial-load effects in the bearing model increase, thus the uncoupled bearing model be-
comes increasingly inaccurate, as the isolation system gains strength and the building increases
its tendency to rock (η increases and Υx decreases). Figure 7.18 demonstrates that for small
Υx the uncoupled model tends to underestimate the bearing deformation and overestimate the
axial force increments, and hence the force extrema, compared to the variable-strength model
(a positive percent discrepancy indicates the uncoupled model is conservative). The discrep-
ancy in both the lateral deformation and axial force increments is usually less than 10%, which
is not very significant given the other approximations and assumptions: a rigid model of the
Over the practical range of slenderness ratio, it does not appear necessary to use a bearing
model that includes the axial-load effects. However, time-variation of the strength – included in
the variable-strength model – should be considered if η ≥ 0.75 and Υx ≤ 0.75, which is intended
to include the approximate region where the error in the force increments exceeds 10%. On
173
(a) Υx = 1.22
15 2 1 η=.25
0.5
η=.25 1.8 0.8 0.75
1.0
10 0.5
ubo (cm)
1.5
P+ /Pst
P− /Pst
0.75 1.6 0.6
1.0
1.5 1.4 1.5 0.4
5
1.0
Constant-strength 1.2 0.75 0.2
0.5
Uncoupled η=.25
0 1 0
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
(b) Υx = 0.48
15 3 1
0.5
P+ /Pst
P− /Pst
0.75 1.0
1.0 2 0.75 0 0.75
1.5 1.0
5 0.5
1.5 η=.25 −0.5
Constant-strength 1.5
Uncoupled
0 1 −1
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
(c) Υx = 1.22
15 2 1 η=.25
0.5
η=.25 1.8 0.8 0.75
0.5 1.0
10
ubo (cm)
1.5
P+ /Pst
P− /Pst
P+ /Pst
P− /Pst
0.75 1.0
1.0 2 0 0.75
0.75
1.5 1.0
5 0.5
1.5 −0.5 1.5
Variable-strength η=.25
Constant-strength
0 1 −1
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
Tb (sec) Tb (sec) Tb (sec)
Figure 7.17: Median bearing deformation ubo and axial forces P+ /Pst and P− /Pst for (a,c)
Υx = 1.22 and (b,d) Υx = 0.48; comparison between (a,b) constant-strength model and uncou-
pled nonlinear model (c,d) variable-strength model and constant-strength model.
174
(a) (b) (c)
0 25 25
η=0.25
20 0.5 20
% Discrepancy in ∆P+
% Discrepancy in ∆P−
% Discrepancy in ubo
0.75
15 1.0 15
−5 1.5
10 10
5 5
0 0
−10
−5 −5
−10 −10
−15 −15 −15
0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2
Rocking-to-vertical frequency ratio Υx
Figure 7.18: Percent discrepancy of (a) the median bearing deformation ubo and (b,c) ax-
ial force increments ∆P+ /Pst and ∆P− /Pst with the uncoupled nonlinear model relative to
variable-strength model; Tb = 2 seconds.
the other hand, the variable-strength (or constant-strength) model is needed to accurately
determine the bearing axial deformations (Sec. 7.3), which, usually of lesser concern, have not
As in previous chapters, regression equations are fit to the peak deformation ubo and peak axial
forces P± /Pst , which can be used to estimate their values in preliminary design. Regression
analysis utilizes the normalized deformation ūbo and forces P̄± /Pst , which were computed by
RHA of Eq. 7.17 to each ground motion in the LMSR ensemble. The response data varies
over the parameters Tb from 2 to 4.5 seconds, η from 0.25 to 1.5, and Υx from 0.45 to 1.55,
based on Ωθx ranging from 9.5 to 53 and Ωz ranging from 20 to 35. This data set is limited
to the practical range of building slenderness ratios h/d ≤ 2; data for more slender buildings
that display less consistent behavior (Fig. 7.14) are ignored. The regression equations are based
on the responses computed using the variable-strength model, although previous observations
(Fig. 7.17) suggest that similar equations are likely to be produced by data from the constant-
Extending methods from Chapters 3-5, the regression equation relating the normalized
175
By testing various possibilities, the best fit to the normalized deformation ūbo was found to be:
which, when multiplied by u∗y (Eq. 3.6) – a function of Tb , η and median (or design) ground
velocity u̇go – results in the following equation for the peak deformation ubo :
Not appearing in this equation, the rocking-to-vertical frequency ratio Υx was found to be
insignificant in regression analysis. This is not surprising since variation in the peak deformation
ubo with either rocking (Ωθx ) or vertical (Ωz ) frequency ratio was limited to instances where the
value of Ωθx was low (Fig. 7.14, Fig. 7.15a), which have been omitted here. For three different
values of Υx , Fig. 7.19 demonstrates that Eq. 7.21 is a good approximation to the exact median
deformation ubo .
The axial force extrema have been shown to depend on isolation period Tb , normalized
strength η, and rocking-to-vertical frequency ratio Υx . As discussed in Sec. 7.4.2, force incre-
ments are preferred for statistical analysis over actual forces, thus, equations are presented for
the force increments ∆P+ /Pst and ∆P− /Pst . Equations were first fit to the normalized force
increment ∆P̄ /Pst , and multiplied by u∗y (Sec. 7.4.1) to obtain the equation for the actual force
increment:
∆P 0.416 (−1.07+0.53 ln Tb ) (0.09+0.18 ln η+0.60 ln Tb ) −2.23
= T η Υx u̇go (7.22)
Pst 4π 2 b
Finally, the axial force extrema P+ /Pst and P− /Pst are computed by:
Although the positive and negative force increments can differ due to vertical displacement of
the CM, the two were not statistically distinct when averaged over many ground motions, and
176
(a) Υx = 1.22 (b) Υx = 0.68 (c) Υx = 0.50
15 15 15
1.5 3 3
1.4
2.5 2.5
1.5
P+ /Pst
1.3
1.5 2 2 1.0
1.2 1.0 1.5 0.75
0.75 1.0 0.5
1.5 0.75 1.5
1.1 0.5 0.5 η=0.25
η=0.25 η=0.25
1 1 1
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
1 1 1
η=0.25 η=0.25
0.9 0.5 0.5
0.75 0.5 0.5 η=0.25
0.75 0.5
1.0 1.0
P− /Pst
0.8 0.75
1.5 1.5
0 0 1.0
0.7 1.5
−0.5 −0.5
0.6
0.5 −1 −1
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
Tb (sec) Tb (sec) Tb (sec)
Figure 7.19: Design equation for bearing deformation ubo (Eq. 7.21) and axial forces P+ /Pst
and P− /Pst (Eqs. 7.22 and 7.23), compared to the exact median by RHA to the strong-
component ensemble, for (a) Υx = 1.22 (b) Υx = 0.68 and (c) Υx = 0.50; median/design
ground velocity u̇go =35 cm/s.
More complex than any of the previous equations for lateral deformation, Eq. 7.22 includes
second order coefficients for both Tb and η (i.e., Tbln Tb ) and a coefficient for interdependence of
Tb and η (i.e., η ln Tb ). The variation of the bearing forces with isolation period Tb depends on
the normalized strength η (Fig. 7.13, Fig. 7.19). If Tb and η were independent, the lines for
different values of η would be approximately parallel like they are for the deformation ubo (see
also Fig. 7.12). Interdependence between other parameters – Tb and Υx , η and Υx – was tested
177
1.5
1.25
Normalized strength η
1
0.75
0.5
0.25
0.2 0.4 0.6 0.8 1
Rocking-to-vertical frequency ratio Υx
Figure 7.20: Conditions that induce bearing tension for a system with isolation period Tb = 3
seconds and PGV u̇go = 35 cm/s.
For three rocking frequency ratios, Eq. 7.23a and b are compared to the exact median
axial force extrema determined by nonlinear RHA in Fig. 7.19 (note different scales in Fig. 7.19a
vs. Fig. 7.19b,c). These design equations are quite accurate for higher frequency ratios, but
less when Υx is as low as 0.5 (Fig. 7.19c). The shape of the spectrum (variation with Tb and η)
appears to depend somewhat on Υx , but we were unable to include this effect without further
The design equations can conveniently predict the occurrence of tension in the bearings.
For example, for design ground velocity u̇go = 35 cm/s and isolation period Tb = 3 seconds,
the shaded portion of Fig. 7.20 indicates the combinations of η and Υx for which Eq. 7.22
predicts at least one exterior bearing to go into tension (P− /Pst < 0 or ∆P/Pst > 1). For
building slenderness ratio h/d ≤ 2, the rocking-to-vertical frequency ratio Υx ≥ 0.45 and
tension is predicted only for normalized strength greater than about 0.75. The particular
combinations that produce tension will vary for other choices of u̇go and Tb . Used in this
manner, Eq. 7.22 provides a valuable tool to identify the potential for bearing tension, which
can then be avoided by modifying the design of the isolation system to ensure that the bearings
remain in compression.
178
7.7 Conclusions
Building codes require that the isolation system be modeled with sufficient detail to assess peak
forces in individual isolator units. Developed in the previous chapter for this purpose, the non-
linear constant-strength and variable-strength models for lead-rubber bearings, which include
axial-load effects, have been implemented in dynamic analysis of the structure. This investi-
gation of the coupled lateral-rocking response of isolated buildings including bearing axial-load
effects, subjected to an ensemble of ground motions, has led to the following conclusions:
1. Rocking of the structure and bearing axial-load effects, i.e., variation of the stiffness and
strength of the bearings with varying axial loads, have little influence on the peak lateral
bearing deformation. The error in neglecting axial-load effects is less than 5% typically
and less than 10% for isolated systems with large strength. Even if rocking is neglected
entirely, response spectra of the lateral deformation are still within 10% of those when
rocking and axial-load effects are included. In these spectra, the deformation increases
slightly with increasing isolation period Tb and decreases with increasing yield strength
2. Unlike lateral deformation, increasing bearing yield strength η generally causes the axial
force extrema to increase – maximum force increases and minimum force decreases – ex-
cept perhaps for small isolation periods Tb where η has little effect. The force extrema
are essentially independent of the vertical-to-lateral frequency ratio Ωz for a wide range
of systems, implying that use of high shape factor bearings to avoid instability is unnec-
essary and not otherwise beneficial. On the other hand, decreasing the rocking-to-lateral
increases the force extrema. Because Ωθx and Ωz are interrelated, Ωθx alone cannot ad-
equately predict axial forces when Ωz is varied. The rocking-to-vertical frequency ratio
Υx = Ωθx /Ωz is less sensitive to changes in Ωz , and is thus a better parameter to predict
179
3. Bearing axial-load effects are usually small enough that they can be neglected in de-
termining the maximum and minimum bearing axial forces. Of these effects, the time-
variation of the lateral and vertical stiffness, as included in the constant-strength model,
cause negligible change to the axial forces. The time-variation of strength included in
the variable-strength model, may, for certain parameter combinations, cause the forces to
vary by more than 10% compared to a typical uncoupled nonlinear model; in this regard
the variable-strength model is recommended when η > 0.75 and Υx < 0.75. Further
investigation is required to determine if axial-load effects are more significant when (1) a
4. By regression analysis of the response computed by nonlinear RHA of the system to the
LMSR ground motion ensemble, equations were developed to estimate the peak lateral
deformation (Eq. 7.21) and axial force extrema (Eqs. 7.22 and 7.23) in the bearings. The
design equation for the force extrema (Eq. 7.22) can predict the occurrence of tension in
the bearings, allowing the design to be modified early to eliminate the possibility of such
tension. The simplest way to reduce the force extrema and eliminate tension is to decrease
the normalized strength of the system, but at the expense of larger lateral deformation
in the bearings.
180
Appendix 7A: Parameter Selection
• System parameters (Eq. 7.7): h/rx , yi /rx for each bearing, and total mass m.
• Bearing properties (Eq. 7.1 and 7.2): initial stiffness kI , post-yield stiffness kbi , strength
Qi , and force Po (variable-strength model only) to determine shear spring force fsi ; Euler
buckling load PE , height hb , vertical stiffness kbzi and damping coefficient cbzi .
We varied the building slenderness ratio h/d from 0.5 to 4, and the number of (in-plane)
bearings nby across the plan from 2 to 15. These two properties uniquely determine h/rx and the
bearing positions yi /rx , by assumption that the mass is uniformly distributed and the bearings
are evenly-spaced.
The linear or bilinear force-deformation relations of the individual bearing shear springs
fsi (si ) (Eq. 7.1a) were determined from the isolation period Tb (kb = ωb2 m with ωb = 2π/Tb ),
The total stiffnesses (kb , kI ) and strength Q were distributed evenly among all bearings. For the
variable-strength model, the additional property Po (Eq. 7.3) was determined from Pst /Po = 3.
These parameters also suffice for the lateral force-deformation of the uncoupled bearing models
(Eq. 7.8).
For the remaining bearing properties, recall that the Euler buckling load PE is related to
√
the bearing buckling load Pcr by Pcr = PE PS (Eq. 6.6), where PS = GA(tr /hb ). Although the
cross-sectional area A can vary widely, isolation bearings are typically used at design pressures
pst ranging from 5 to 7 MPa (700 to 1000 psi) [36], with shear modulus G between 0.35 and
1.4 MPa (50 and 200 psi) [36]. Based on these typical values, pst /G, equivalent to Pst /GA, was
set to 10. The ratio of bearing height to total rubber thickness, hb /tr , was estimated as 4/3.
Together, these assumptions led to the following for PS and Pcr (Eq. 6.16):
181
where Pst was the total weight of the rigid block; and finally PE :
2
Pcr 7.5Pst (πSS2 )2
PE = = (7A.2)
PS 800(1 + Ks S 2 )
which was divided by nby to get PE of a single bearing. The bearing height hb was computed:
kbi
hb = (7A.3)
PS
The nominal vertical bearing stiffness kbzi (Eq. 6.2) was determined from the post-yield stiffness
kbi (Eq. 6.1) and the shape factor S (Eqs. 6.11 and 6.15):
6S 2 kbi
kbzi = (7A.4)
(1 + Kc S 2 )
The damping coefficient cbzi was defined from an assumed damping ratio ζz of 5% (Eqs. 7.12
and 7.13). In the above equations, the shape factor S was varied from 10 to 30 (as previously
stated), the second shape factor (bearing aspect ratio) S2 = 3, and Kc = 0.004 and Ks = 0.0015
To determine a feasible range of slenderness ratio for which structures can be approximated as
rigid, we investigated the variation of structure period with slenderness ratio. Natural periods
of existing structures vibrating in their linear range have been measured from their vibration
responses recorded during earthquakes, and this data has been assembled for 106 buildings
in California [56]. The observed natural period of these buildings, whose structural systems
include concrete shear wall, R/C and steel moment resisting frames, dual systems, masonry,
and braced frames, are plotted against slenderness ratio in Fig. 7B.1a. The huge scatter in the
data might imply little correlation between slenderness ratio and structure period. However, as
the slenderness ratio increases, it becomes more unlikely that these buildings, if isolated, could
The data for the three main structural systems has been separated: concrete shear walls
(Fig. 7B.1b), R/C moment frames (Fig. 7B.1c), and steel moment frames (Fig. 7B.1d). For
each system, the median and 16th percentile (median-σ) curves are also shown, which have
182
(a)
8
concrete SW
R/C MRF
Structure period Ts
6 steel MRF
other
0
0 1 2 3 4 5 6
Slenderness h/d
median
4 16th pct 4 4
3 3 3
2 2 2
1 1 1
0 0 0
0 2 4 6 0 2 4 6 0 2 4 6
Slenderness h/d Slenderness h/d Slenderness h/d
Figure 7B.1: Observed fundamental period Ts of existing buildings varying with slenderness
ratio h/d; median and 16th percentile (median-σ) of observed period for (a) all building types,
(b)concrete shear wall, (c) R/C moment resisting frames, and (d) steel moment resisting frame
buildings.
been fit to the associated data by linear regression analysis assuming Ts = α ∗ (h/d)β . The 16th
percentile curve, lower than the median by one standard deviation, may be more representative
of isolated buildings, which are required to be stronger and thus have shorter natural periods Ts
than comparable fixed-base buildings. Allowing the fixed-base structure period to be no greater
than 1.66 seconds to assume rigidity, the maximum slenderness ratios are ∞, 3.41, and 1.64 for
concrete shear walls, R/C moment frames and steel moment frames, respectively, as computed
by the 16th percentile equation. According to the equations, it is most feasible for concrete
shear wall systems to be both slender and sufficiently stiff; however, no data is available for
these buildings with slenderness ratios above 2. Thus, the selection of upper limiting slenderness
183
ratio is a judgment call. For practical application we consider the slenderness ratio for isolated
buildings to be limited to h/d < 2, but for completeness higher values of slenderness ratio are
considered.
cbz coefficient of damping for vertical motion, summed over all bearings
D bearing diameter
ing
fsi shear spring force (in two-spring model) corresponding to ith bearing
h building height
ings
184
m mass of isolated block
PE Euler buckling load, neglects shear stiffness of the bearing: π 2 EIs /h2b
∆P+ , ∆P− maximum and minimum axial force increments; maximum force de-
bearings
ith bearing
tion system
185
∆ubz− , ∆ubz+ maximum deviation of vertical deformation in either direction from
µg/(ωd u̇go )
ness
tral regions
186
8: ESTIMATING BEARING DEFORMATIONS AND FORCES IN
SYMMETRIC- AND ASYMMETRIC-PLAN SYSTEMS INCLUDING
TORSION AND ROCKING
8.1 Introduction
Previous chapters have emphasized that reliable estimates of the controlling lateral deformation
and axial forces in the isolation system are key ingredients to an efficient design. For rubber
bearings, correct prediction of the lateral deformation helps the designer balance the tradeoff
between bearing flexibility, stability, and energy dissipation. Axial forces are of particular inter-
est to fulfill testing requirements and achieve an isolation system that remains in compression
under design loads. Thus, initial estimation of pertinent design quantities greatly simplifies the
overall procedure.
For this purpose, Chapter 3 developed a design equation based on nonlinear response
history analysis (RHA) to estimate the peak, or absolute maximum, isolator deformation due
to unidirectional – single component – excitation, which was shown to be much superior to the
equivalent-linear analysis upon which the static lateral response procedure of the International
Building Code (IBC) [1] is based. Subsequently, the design equation was extended to include
the increase in deformation due to bidirectional excitation – two components of ground motion
applied in orthogonal lateral directions. In Chapter 4, a factor that depends on the eccentricity
and the distance from the CM to the outlying isolator was developed to extend the design
response, known as axial-load effects, motivated the development of improved models for lead-
rubber bearings (Chapter 6). With these models, analysis was extended to lateral-rocking
response due to unidirectional excitation, i.e., unidirectional analysis, and the overall scope
of investigation was expanded to determine bearing axial forces. New design equations for the
peak lateral deformations and maximum and minimum axial forces were based on the improved
models, although the influence of these models was ultimately shown to be small (Chapter 7).
187
Not considered previously, axial-load effects may cause accidental torsion in symmetric-
plan buildings. Time variation of the stiffnesses and strengths of individual bearings, caused by
variation in axial loads, can induce a time-varying eccentricity. Although it can be computed
with suitable analytical models, torsion from this source is accidental because it occurs in
model [57, 58, 59] was used to identify accidental torsion that may occur from varying axial
loads on the isolators [60]. While accidental torsion was found to be insignificant in the isolation
system unless the superstructure, permitted to uplift, reached incipient overturning, it was
To consider accidental torsion, among other things, the investigations of previous chapters
buildings. Using the tools that have been developed, the following questions can be answered
with minimal additional effort: Does rocking influence the factors by which the lateral deforma-
tion increases due to bidirectional excitation and asymmetry of plan? How does rocking about
multiple axes or plan asymmetry influence the maximum and minimum axial forces?
This final chapter simultaneously considers rocking and torsion due to bidirectional exci-
tation in symmetric and asymmetric-plan buildings with the following objectives: (1) to inves-
tigate the possibility of accidental torsion due to axial-load effects and overturning in buildings
isolated with lead-rubber bearings, and (2) to improve previous design equations for estimating
bearings, such that the total system is either nominally symmetric (Fig. 8.1a) or asymmetric
(Fig. 8.1b) in plan. When subjected to orthogonal components of lateral ground excitation
188
z, uz
θz
z, uz
CM x, ux
CM θx x, ux h -y, -uy
h θy
-y, -uy
Figure 8.1: (a) Symmetric-plan and (b) asymmetric-plan representative systems, for analysis
of lateral-torsional-rocking response due to bidirectional excitation
ügx and ügy , the complete motion of the isolated block is described by six degrees-of-freedom
(DOFs), selected as translations (ux , uy and uz ) and rotations (θx , θy and θz ) about each CM
axis. Specifically, θx and θy refer to rocking about the x and y-axes, while θz refers to twisting
or torsion about the vertical (z) axis. Each isolation bearing is located at xi , yi , −h/2 from
In this study, symmetric buildings are assumed to be rectangular with plan dimensions
b and d (b/d ≥ 1) and height h, wherein the radii of gyration about their respective axes
are easily computed as: rx = (d2 + h2 )/12, ry = (b2 + h2 )/12, and rz = (b2 + d2 )/12.
For this investigation, the variable-strength model of the previous two chapters is extended to
bidirectional motion. To achieve this, the two-spring model (Fig. 6.2), which was the basis for
the variable-strength model, is modified with a shear spring that deforms bidirectionally and a
rotational spring that rotates about any axis in the x − y plane. Equilibrium equations for the
189
ith bearing, representative of the modified spring assemblage are:
Pi − kbzi vi = 0 (8.1e)
Equations 8.1c-e are essentially identical to Eq. 6.4 for unidirectional analysis, and represent
the balance of lateral forces (Eq. 8.1c) and moments (Eq. 8.1d) for motion in the y-direction,
along with the balance of vertical forces (Eq. 8.1e). Analogous equations 8.1a-b are added for
balance of lateral forces and moments in the x-direction. Resultant lateral forces fbxi and fbyi
and axial forces Pi on a bearing are balanced by (a) the force in the nonlinear shear spring
with components fsxi and fsyi acting through deformations sxi and syi , (b) the moment of the
rotational spring with stiffness PE hb acting through rotations θxi and θyi about the respective
axes, and (c) the force of an additional vertical spring with stiffness kbzi acting through axial
deformation vi . With stiffness PE hb – the Euler buckling load times the bearing height – the
rotational spring resists an Euler buckling type of instability; spring rotations θxi and θyi are
The bilinear force-deformation relation of the shear spring is represented by (Eq. 4.1):
⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎨fsxi ⎪
⎪ ⎬ ⎪
⎨sxi ⎪
⎬ ⎪
⎨zxi ⎪⎬
= kbi + Qi (8.2)
⎩fsyi ⎪
⎪ ⎭ ⎪
⎩syi ⎪
⎭ ⎪
⎩zyi ⎪⎭
where kbi and Qi are the nominal postyield stiffness and strength for motion in any lateral
direction. The components zxi and zyi of the yield function interact, with magnitude zi =
2 + z 2 )1/2 ≤ 1, representative of a circular yield surface. The yield deformation of the spring
(zxi yi
is assumed to be 1 cm at the nominal yield strength (Sec. 3.3.1). The actual yield strength
190
where Po , typically about 1/3 of the bearing static load, can be determined from characteristic
directions (ubxi , ubyi and ubzi ) to the local shear and axial deformations and rotations are:
respective direction is increased by the bearing rotation acting through its height (Eqs. 8.4a,b).
The total vertical deformation ubzi includes the axial deformation vi and additional vertical dis-
placement incurred in the laterally deformed configuration (Eq. 8.4c). In contrast to the global
sign convention, the vertical deformation ubzi and force Pi are positive for bearing compression.
This bearing model incorporates axial load effects, first identified in Chapter 6, into
the bidirectional response of the bearing. For instance, a variation of the resultant lateral
stiffness with axial load and variation of the resultant vertical stiffness with lateral deformation
(Sec. 6.2.1), are caused by the interaction of shear and rotational springs in the modified spring
assemblage. Variation of the yield strength with axial load (Eq. 8.3) is incorporated directly into
the numerical algorithm for the bilinear shear spring. It is clear from the equilibrium equations
(Eq. 8.1) that the three resultant forces fbxi , fbyi and Pi are coupled, wherein force-deformation
The equations of motion of the system neglecting bearing axial-load effects will be exam-
ined to identify influential parameters. When axial-load effects are neglected, Eq. 8.1 simplifies
to the following:
191
where the lateral force-deformation equations (Eqs. 8.5a and b) and the linear vertical force-
deformation (Eq. 8.5c) are now uncoupled. This lateral force-deformation relation is simply
Eq. 8.2, where the shear deformation now represents the total lateral deformation, and the
Viscous damping in the vertical direction is included by an additional axial force com-
ponent for each bearing, fDi = cbzi u̇bzi where cbzi is the damping constant and u̇bzi the axial
bearing velocity.
The equations of motion of the system including bearing axial-load effects, subjected to lateral
müx + fbxi = −mügx (8.6a)
i
mrz2 θ̈z − yi fbxi + xi fbyi = 0 (8.6b)
i i
müy + fbyi = −mügy (8.6c)
i
h
mrx2 θ̈x − yi cbzi u̇bzi + fbyi − yi Pi = 0 (8.6d)
2
i i i
h
2
mry θ̈y + xi cbzi u̇bzi − fbxi + xi Pi = 0 (8.6e)
2
i i i
müz − cbzi u̇bzi − Pi = mg (8.6f)
i i
The lateral (fbxi , fbyi ) and axial (Pi ) forces are implicit functions of the bearing deformations
ubxi , ubyi and ubzi (Eqs. 8.1 and 8.4), which can be written in terms of the displacements ux ,
h
ubxi = ux − θy − yi θz (8.7a)
2
h
ubyi = uy + θx + xi θz (8.7b)
2
ubzi = −uz − yi θx + xi θy (8.7c)
Substituting the time derivative of Eq. 8.7c for u̇bzi into Eq. 8.6 and dividing by radii of
192
gyration rx , ry , and rz where appropriate leads to
⎧ ⎫ ⎧ ⎞ ⎧ ⎫ ⎧ ⎫
⎪ ⎪ ⎪ ⎪ ⎪ ⎪
⎪
⎪ ü x ⎪ ⎪ ⎪
⎪ u̇ ⎪
⎪ f ⎪
⎪ ⎪
⎪ ügx ⎪
⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪
x
⎟ ⎪
⎪
i bxi ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎟ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ r θ̈ ⎪
⎪ ⎪
⎪ r θ̇ ⎟ ⎪
⎪ − yi
f + xi
f ⎪
⎪ ⎪
⎪ 0 ⎪
⎪
⎪
⎪ z z ⎪
⎪ ⎪
⎪ z z ⎟ ⎪
⎪ i r z
bxi i r z
byi ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎟ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎨ üy ⎬ ⎪ ⎪
⎨ u̇y ⎟ ⎨ ⎟ ⎪ ⎪
⎬ ⎪
⎨ ⎪
fbyi ügy ⎬
m +C ⎟ i
= −m
⎪ ⎪ ⎪ ⎟+⎪ h yi ⎪ ⎪ ⎪
(8.8)
⎪
⎪ r θ̈ ⎪
⎪ ⎪
⎪ r θ̇ ⎟ ⎪
⎪ f − P ⎪
⎪ ⎪
⎪ 0 ⎪⎪
⎪
⎪
x x⎪
⎪ ⎪
⎪
x x⎟ ⎪
⎪ 2rx i byi i rx i ⎪ ⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎟ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎟ ⎪
⎪ − h xi ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ r y θ̈ y ⎪
⎪ ⎪
⎪ r y θ̇ y ⎟ ⎪
⎪ f bxi + P i ⎪
⎪ ⎪
⎪ 0 ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎠ ⎪
⎪
2r y i i ry
⎪
⎪ ⎪ ⎪
⎪ ⎪
⎪
⎩ ü ⎭ ⎪ ⎪
⎩ u̇ ⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
z z − i Pi −g ⎭
where
⎡ ⎤
⎡ ⎤ yi2 x i yi yi
⎢ i rx2 cbzi − i rx ry cbzi i rx cbzi ⎥
⎢0 0 ⎥ ⎢ ⎥
C=⎣ ⎦ and Cz = ⎢ x i yi
⎢− i rx ry cbzi
x2i
i ry2 cbzi − xi ⎥
i ry cbzi ⎥
(8.9)
0 Cz ⎣ ⎦
yi
i rx cbzi − xi
i ry cbzi i cbzi
Because of the influence of bearing forces at multiple DOFs in the equations of motion
(Eq. 8.8) and their relation to deformations at the bearing level (Eqs. 8.1 and 8.4), all six global
DOFs of the system are coupled. Thus, a single component of ground excitation can induce
a combined lateral, torsional, rocking, and vertical (bouncing) response. In its most general
form, the damping matrix Cz implies coupling of damping forces in the two rocking and vertical
DOFs. However, if the isolated building is symmetric with stiffness-proportional damping, all
response analysis of the isolated block, leading to the following normalized equations:
⎧ ⎫ ⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪
⎪
⎪ ¨
ū x ⎪ ⎪ ⎪
⎪ ˙
ū x ⎪ ⎪ ⎪
⎪ f ⎪
⎪ ⎪
⎪ ¨gx ⎪
ū ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
i bxi ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ r ¨
θ̄ ⎪
⎪ ⎪
⎪ r ˙
θ̄ ⎪
⎪ ⎪
⎪− yi
f + xi
f ⎪
⎪ ⎪
⎪ 0 ⎪
⎪
⎪
⎪ z z ⎪
⎪ ⎪
⎪ z z ⎪
⎪ ⎪
⎪ i rz bxi i rz byi ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎨ ū ⎪ ⎪
¨y ⎬ C ⎨ ū˙ y ⎬ ω 2 ⎨⎪ ⎪ ⎪
⎬ ⎪
⎨ ¨ ⎪
⎬
i f byi ω 2 ū gy
+ + b
= − b
(8.10)
⎪
⎪ ⎪ m⎪ ⎪ Q ⎪ h ⎪ ηωd ⎪ ⎪
⎪
⎪ rx θ̄¨x ⎪⎪
⎪
⎪
⎪
⎪ rx θ̄˙x ⎪⎪
⎪
⎪
⎪
⎪ fbyi − i ryxi Pi ⎪ ⎪
⎪
⎪
⎪
⎪ 0 ⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
2r x i ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ r ¨
θ̄ ⎪
⎪ ⎪
⎪ r ˙
θ̄ ⎪
⎪ ⎪
⎪ − h
f + xi
P ⎪
⎪ ⎪
⎪ 0 ⎪
⎪
⎪
⎪
y y ⎪
⎪ ⎪
⎪
y y ⎪
⎪ ⎪
⎪ 2r y i bxi i ry i ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎩ ū ¨z ⎭ ⎩ ū˙ z ⎭ ⎩ − i Pi ⎭ ⎩ −ḡ ⎭
193
They were obtained by first dividing Eq. 8.8 by u∗y = Q/kb (Eq. 3.6) such that all DOFs are
normalized – e.g. ūx = ux /u∗y – and then substituting the normalized strength η = a∗y /ωd u̇gyo
(Eq. 3.8) into the right-hand side. This normalizes the ground acceleration by the peak
between -1 and +1. Gravity, applied in the vertical direction, is also normalized (ḡ = g/u̇gyo ).
Normalization of the equations of motion (Eq. 8.10) parallels that in previous chapters.
Without proof, then, prior assertions regarding the peak normalized responses of the system
are repeated: (1) for a given set of parameters that suitably define the system, the median
normalized response over an ensemble of ground motions is only weakly affected by ground
motion intensity, and (2) the dispersion of the normalized response over the ensemble is small.
Ultimately, this allows meaningful prediction of the peak forces or deformations for varying
For given parameters, the median of a response over an ensemble of ground motions,
with weak and strong-component ensembles applied in the x and y-directions, respectively, is
determined by: (1) computing the peak normalized response by nonlinear RHA of Eq. 8.10 to
each ground motion in the ensemble, (2) computing the median (Eq. 2.1) over the ensemble,
and (3) multiplying the normalized median by u∗y to obtain the actual median as a function of
wide variety of systems, and not limited to the plans considered. Formerly, plan asymmetry
was introduced by varying the stiffnesses and strengths of individual bearings in an idealized,
rectangular plan (Sec. 4.3). This approach was justified because the idealized system with key
parameters identical to the actual asymmetric-plan system computed its response with less than
1% error (Sec. 4.3.2). Furthermore, it was shown that a two-way asymmetric system could be
194
Table 8.1: Idealized Plan Layouts for Bidirectional Analysis
represented by a one-way asymmetric system, since the two are equivalent under a simple axis
Several difficulties hinder the use of this approach when rocking of the system is included.
First, because the vertical bearing stiffnesses in the idealized system are varied to introduce
eccentricity, the gravity or static loads Pst on the individual bearings, which are proportional to
these stiffnesses, vary nonuniformly. This particular variation of static loading is unnatural, and
whether it affects the dynamic response factor is unknown. Second, when rocking is included
rectangular system, preventing direct comparison of their responses to verify accuracy. Third,
Nevertheless, the idealized system approach is pursued because of the tremendous ben-
efit provided if, despite the difficulties mentioned, the results can still be extended to a wide
range of asymmetric-plans with reasonable accuracy. The effectiveness of this approach will
be determined by comparing the final design equations to the response of several systems with
legitimate asymmetry-of-plan.
Nine different rectangular plan layouts have been selected to characterize a wide range
of symmetric and asymmetric buildings. Labeled S1-S9, these 9 plans are defined by their
plan aspect ratios b/d and the number of bearings nbx and nby along the x and y-axes. These
195
properties, as well as the torsional frequency ratio Ωθz , which depends only on the plan layout,
are listed in Table 8.1 for each plan. The rocking frequency ratios Ωθx and Ωθy depend also on
the slenderness – or height-to-width – ratio h/d, which will be varied in analysis. To represent
symmetric buildings, the stiffnesses and strengths of all bearings are modeled as identical,
while they are varied to impose a one-way eccentricity in asymmetric buildings. Plans S1-S3
are limited to symmetric-plan systems because they have unusually large torsional-to-lateral
Although the analyses presented here are limited to the variable-strength model, the equations
of motion of the system with the uncoupled nonlinear bearing model help to identify several
parameters that characterize the system even when axial-load effects are present. Substituting
Eq. 8.5 into Eq. 8.10 specializes the normalized equations of motion for the simpler uncoupled
model:
⎧ ⎫ ⎧ ⎫ ⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎪
⎪
⎪ ūx ⎪
¨ ⎪
⎪
⎪
⎪
⎪ ˙
ūx ⎪ ⎪
⎪ ⎪
⎪
⎪ ūx ⎪ ⎪
⎪ ⎪
⎪
⎪ i γi zxi
⎪
⎪
⎪
⎪
⎪
⎪ ¨gx ⎪
ū ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪ ⎪
⎪ ⎪
⎪
⎪ r ¨
z z⎪
θ̄ ⎪ ⎪
⎪ r ˙
θ̄ ⎪
⎪ ⎪
⎪ r θ̄ ⎪
⎪ ⎪
⎪ γ (− yi
z + xi
z ⎪
yi ⎪
) ⎪
⎪ 0 ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ z z⎪
⎪ ⎪
⎪ z z⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
i i r z
xi r z
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪ ⎪
⎪
⎪ ¨y ⎪
⎨ ū ⎬ ⎪
⎨ ˙ ⎪
⎬ ⎪
⎨ ⎪
⎬ ⎪
⎨ ⎪
⎬ ⎪ ¨gy ⎪
⎨ū ⎬
C ūy ūy i γ i zyi ωb2
+ +K 2
+ ωb = − (8.11)
⎪
⎪ ⎪ m⎪ ⎪ ⎪ ⎪ ⎪ h ⎪ ηωd ⎪ ⎪
⎪rx θ̄¨x ⎪
⎪ ⎪
⎪
⎪
⎪
⎪ rx θ̄˙x ⎪
⎪
⎪
⎪
⎪
⎪ rx θ̄x ⎪
⎪
⎪
⎪
⎪
⎪ i γi zyi
⎪
⎪
⎪
⎪
⎪
⎪ 0 ⎪
⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
2r x ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ¨⎪ ⎪ ⎪
⎪ ˙ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ h
⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ r θ̄ ⎪
⎪ ⎪
⎪ r θ̄ ⎪
⎪ ⎪
⎪ r θ̄ ⎪
⎪ ⎪
⎪ − γ z ⎪
⎪ ⎪
⎪ 0 ⎪
⎪
⎪
⎪
y y⎪
⎪ ⎪
⎪
y y⎪
⎪ ⎪
⎪
y y⎪
⎪ ⎪
⎪ 2r y i i xi ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎩ ū¨z ⎭ ⎩ ˙ūz ⎭ ⎩ ūz ⎭ ⎩ 0 ⎭ ⎩ −g ⎪
⎭
The symbol γi = Qi /Q was introduced, allowing Q to be factored out of the nonlinear terms
196
Equation 8.11 has been written in terms of the following parameters of a linear system
(η = 0):
1. vertical (Ωz ), rocking (Ωθx and Ωθy ) and torsional (Ωθz ) – all relative to lateral – frequency
ratios:
ωbz ωbθx ωbθy ωbθz
Ωz = Ωθx = Ωθy = Ωθz = (8.13)
ωb ωb ωb ωb
where the lateral (ωb ), vertical (ωbz ), two rocking (ωbθx and ωbθy ), and torsional (ωbθz )
frequencies are
1/2 1/2 $ %1/2
h2
i kbi ikbzi 4 kb + i yi2 kbzi
ωb = ωbz = ωbθx =
m m mrx2
$ %1/2 1/2
h2
4 kb + i x2i kbzi i kbi (x2i + yi2 )
ωbθy = ωbθz = (8.14)
mry2 mrz2
2. x and y-direction eccentricities ebx and eby , and a related term ebxy :
1 1 1 1
ebx = kbi xi ≡ kbzi xi eby = kbi yi ≡ kbzi yi (8.15a)
kb kbz kb kbz
i i i i
1
ebxy = kbzi xi yi (8.15b)
kbz
i
Most of the above definitions (Eqs. 8.13-8.15) were introduced earlier: Eq. 7.10 (Ωz and Ωθx ),
Eq. 7.11 (ωbz and ωbθx ), and Eq. 4.4b (ebx and eby ).
In the traditional sense, nonzero eccentricities ebx and eby represent the distance between
the CM and center of rigidity, and cause a lateral-torsional coupling of the system in response
to a horizontal force. For this particular system, the same eccentricities are also associated
with coupling of the rocking and vertical DOFs, by assumption that the vertical stiffness of
each bearing is proportional to its lateral stiffness. The eccentricities are zero for a nominally
symmetric system (Sec. 4.2.2), causing many off-diagonal terms to drop out of the stiffness ma-
trix K (Eq. 8.12), which eliminates the torsional rz θz and vertical uz deformations (Eq. 8.11).
This conclusion does not apply, however, when axial-load effects are included.
197
The term ebxy (Eq. 8.15) refers to a direct coupling between the two rocking deformations
rx θx and ry θy . When ebxy is nonzero, rocking occurs about both x and y-axes even if the
excitation is unidirectional. For every system, however, there exists an orientation of the x and
y-axes for which ebxy = 0. Thus, the value of ebxy does not convey meaningful information
The properties that must be defined to solve the normalized equations of motion (Eq. 8.10)
with Eqs. 8.1 and 8.4 for the bearing model are numerous. Yet the list of parameters that
significantly influence the system response is short. This list combines the parameters identified
previously for unidirectional lateral-rocking analysis (Sec. 7.2.4) and lateral-torsional analysis
(Sec. 4.2.3). However, Ωθx and Ωz have been replaced by the single parameter Υx ; and one
The isolation period Tb and the normalized strength η are by now very familiar; the remaining
The rocking-to-vertical frequency ratios Υx and Υy are related to the amount of rocking about
the x and y-axes, respectively. Repeating and extending the definition for unidirectional analysis
(Eq. 7.19):
ωbθx Ωθx ωbθy Ωθy
Υx = ≡ Υy = ≡ (8.16)
ωbz Ωz ωbz Ωz
198
Although Υx and Υy do not appear directly in Eq. 8.10, Sec. 7.5.2 demonstrated that for
undirectional analysis, Υx – the ratio of Ωθx to Ωz – characterized the system response better
To determine the range of Υx (or Υy ), consider the parameters Ωθx (or Ωθy ) and Ωz that
enter into the calculation (Eq. 8.16). The vertical frequency ratio Ωz has been shown to be a
function of the bearing shape factor S (Eq. 7.14). Since for unidirectional analysis the variation
of Ωz had little influence on the response of the system, Ωz is selected as constant hereafter
In general, the rocking-to-lateral frequency ratios Ωθx and Ωθy depend on the location and
stiffness of every bearing. However, as was shown in Sec. 7.2.4 (Eq. 7.16), their computation
may be simplified if the bearings are identical and evenly spaced on a grid in plan, and the CM
Using the value of Ωz specified above, along with Eq. 8.17 for Ωθx and Ωθy , Υx and Υy
have been computed as functions of building slenderness ratio h/d for symmetric Plans S1-S9
(Table 8.1). The values of Υx for these plans are identified on a general plot of Υx as a function
of nby for slenderness ratios h/d ranging from 0.5 to 2.0 (Fig. 8.2a), which is an extension
of data that was already shown (Fig. 7.2). While Υx is independent of the plan aspect ratio
b/d, Υy is a function of both b and h (Eq. 8.17), and therefore depends indirectly on both b/d
and the slenderness ratio h/d. Therefore, Υy as a function of nbx is plotted independently for
diferent building aspect ratios b/d=1, 2, or 3 (Figs. 8.2a-c), where values for Plans S1-S9 are
For square plans with equal number of bearings in each direction (b = d and nbx = nby ),
Υx = Υy , implying that the rocking properties are the same about the x and y-axes. This
applies to Plans S1, S4, S7, and S8 (Fig. 8.2a). For non-square plans (S2, S3, S5, S6, S9,
Fig. 8.2b,c), the rocking frequency ratios differ about the x and y-axes, and typically Υx < Υy .
That is, the building is typically more flexible in rocking about the longer x-axis (Fig. 8.1a).
199
(a) (b) (c)
2 2 2
Plans S1,2,3 Plan S2 Plan S3
Plans S4,5,6 Plan S5 Plan S6
Plans S7,9 Plan S9
1.5 Plan S8 1.5 1.5
Υx (or Υy )
h/d=0.5 h/d=0.5
Υy
Υy
1 h/d=0.5 1 1
1.0 1.51.0
1.5 2.0
1.0
2.0
1.5
0.5 0.5 0.5
2.0
0 0 0
0 1 2 0 1 2 0 1 2
10 10 10 10 10 10 10 10 10
nby (or nbx ) nbx nbx
Hence, maximum rocking angles are expected to coincide with the direction of application of the
strong component of ground motion. However, Υx may exceed Υy in squat buildings (compare
Υx and Υy for h/d = 0.5, Plans S2, S3, S5, and S6 in Fig. 8.2). In this case, calculations of
Υy relative to Υx are dominated by the increased number of bearings along the x-axis relative
to the y-axis. In general, increasing the number of bearings for given dimensions reduces the
rocking resistance by moving stiffness away from the edge of the plan.
With a maximum of 7 bearings in any direction, Plans S1-S9 do not envelop the complete
ranges of Υx and Υy , which begin to approach lower bounds at say, 20 bearings, but still continue
to decrease as the number of bearings increases up to 100 (Fig. 8.2). The computational expense
prohibits the number of bearings from being further increased for this type of parametric
analysis, but should not be a limitation since variation of the building slenderness ratio allows
asymmetric-plan buildings is similar to that described above, but does not follow any strict
rules.
200
Torsional-to-Lateral Frequency Ratio Ωθz
Only asymmetric-plan systems or symmetric systems with significant accidental torsion are
affected by variation of the torsional frequency ratio Ωθz . For systems with at least 10 bearings
(most systems), Ωθz < 1.3, and Ωθz is bounded below by 1 as the number of bearings increases
(Sec. 4.3.1, Fig. 4.4). The values of Ωθz for Plans S4-S9, used in the analysis of asymmetric-plan
As applied to the idealized systems, the eccentricity is limited to the x-direction, where ebx /rz is
varied from 0 to 0.15 (Sec. 4.3.1). Although not strictly applicable to axial forces, the direction of
(Sec. 4.3.3). Furthermore, the alignment of eccentricity with the strong-component of ground
This section presents the combined lateral-torsional-rocking response of the symmetric base-
isolated block. Response quantities presented are: (1) the peak lateral deformation in any
direction measured at the CM (ubo ) (2) the peak lateral deformation in any direction maximized
over the bearings (ucorn ), thus occurring in a corner bearing, and (3) axial forces normalized
by their static forces (maximum P+ /Pst and minimum P− /Pst ), which also occur in corner
bearings. Median values of these response quantities at each corner due to the LMSR ensemble
(median PGV = 35 cm/s) are determined by the method outlined in Sec. 8.2.4, where the
weak component of excitation is applied in the x-direction and the strong component in the
y-direction. Force increments ∆P+ /Pst or ∆P− /Pst (Eq. 7.18) are also computed for statistical
analysis, where median values of maximum and minimum axial forces are derived from the
median force increments. The median deformations and force increments are maximized over
the four corners to obtain absolute peak values. Unless otherwise noted, the variable-strength
201
8.4.1 Median Response Spectra
Many trends that were observed for unidirectional excitation of the system (Sec. 7.5) extend
to bidirectional excitation. These trends include, but are not limited to the following, stated
without proof: The peak deformation is not affected by rocking-to-vertical frequency ratios Υx
and Υy . Positive and negative axial force increments vary nearly identically. In general, the
force increments increase with increasing normalized strength, but unsystematically at longer
isolation periods. The uncoupled nonlinear bearing model can generally be used for the bearings
in lieu of the variable strength model with little loss of accuracy, where discrepancy between
the two models is usually less than 10%. For the smallest values of η in combination with the
largest values of Υx , the uncoupled model underestimates the peak force increments by more
than 20%, but the force increment magnitude in such instances are small.
Axial forces on the bearings vary in time due to rocking, which, together with axial-load effects
in the variable-strength model, cause the stiffnesses and strengths of individual bearings to
vary in time. This leads to the possibility of torsion in nominally symmetric buildings – usually
known as accidental torsion – when the excitation is bidirectional. The stiffness and strength
variation can induce a dynamically-varying eccentricity in the direction of ground motion, thus
in Fig. 8.3 where the peak deformations compared at the corner, ucorn , and CM, ubo , are
essentially identical. This conclusion is seen to be true for all combinations of Tb and η, even
though the buildings considered are slender, making them susceptible to overturning and larger
stiffness variations. With Υx and Υy lowest of all systems considered, Plan S8 (Fig. 8.3b) is
most likely to be affected by overturning and hence accidental torsion, but is not. This parallels
the conclusion for FP systems [60] that accidental torsion in the isolation system as a result of
202
(a) (b)
15 15
η=0.25 η=0.25
Lateral deformation (cm)
10 0.5 10 0.5
0.75 0.75
1.0 1.0
1.5 1.5
5 5
ubo
ucorn
0 0
1 2 3 4 5 1 2 3 4 5
Tb (sec) Tb (sec)
Figure 8.3: Median response spectra for peak corner deformation ucorn compared to peak
deformation at CM ubo for (a) Plan S1 and (b) Plan S8 with slenderness ratio h/d = 2; leads
to Υx = 0.78 and 0.55, Υy = 0.78 and 0.55, respectively.
The axial force extrema (P± /Pst ) are influenced by the rocking-to-vertical frequency ratios
Υx and Υy , but not independently by other rocking-related parameters such as the building
slenderness ratio h/d, the plan aspect ratio b/d, and the number of bearings nb . The single
parameter most closely correlated to the force extrema is Υx , and only minor scatter appears
in the maximum axial force P+ /Pst when plotted against Υx (Fig. 8.4a and c). The scatter
occurs among systems with the same or similar values of Υx , but different h/d, b/d, nb , etc.
Some of this scatter can be accounted for by the additional parameter Υy , which varies
with plan aspect ratio b/d when the remaining properties are held constant. Although the
same axial force data considered above shows considerable more scatter when plotted against
Υy instead of Υx (Fig. 8.4b and d), Υy does provide meaningful information. Consider that for
similar values of Υx , the maximum axial forces tend to be largest for square plans (b/d = 1)
and smallest for plans with b/d = 3 (Fig. 8.4a and c). Since Υy is closely related to plan aspect
ratio, it follows that for Υx held constant the axial forces should have a one-to-one relation to
Υy . Lines of constant Υx in Fig. 8.4b and d show that this is sometimes but not always the
case. As will be seen, Υx and Υy , taken together, provide adequate information to reasonably
203
(a) (b)
2 2
b/d=1 b/d=1
b/d=2 b/d=2
1.8 1.8
b/d=3 b/d=3
Υx const.
1.6 1.6
P+ /Pst
1.4 1.4
1.2 1.2
1 1
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Υx Υy
(c) (d)
2.4 2.4
b/d=1 b/d=1
2.2 b/d=2 2.2 b/d=2
b/d=3 b/d=3
2 2 Υx const.
P+ /Pst
1.8 1.8
1.6 1.6
1.4 1.4
1.2 1.2
1 1
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Υx Υy
Figure 8.4: Maximum axial force P+ /Pst plotted against rocking-to-vertical frequency ratios
(a,c) Υx and (b,d) Υy , where (a,b) Tb = 2 sec, η = 0.5 and (c,d) Tb = 4 sec, η = 1.0. Variations
in Υx and Υy are comprised by data from Plans S1-S9 and varying slenderness ratios.
The complete response of the system subjected to bidirectional excitation, which includes rock-
ing about both lateral axes, is compared to the response when the same system is subjected
to unidirectional excitation. The peak deformation of the bearings in the y-direction, in which
component of excitation (Fig. 8.5a), which was true when rocking was neglected (Sec. 3.7). The
peak deformation in any direction, on the other hand, experiences a 10-20% increase when the
204
(a) (b)
15 15
η=0.25 η=0.25
10 10
ucorn (cm)
ubyo (cm)
0.5 0.5
0.75 0.75
1.0 1.0
5 1.5 5 1.5
unidirectional unidirectional
bidirectional bidirectional
0 0
1 2 3 4 5 1 2 3 4 5
Tb (sec) Tb (sec)
Figure 8.5: Median deformations (a) ubyo in the y-direction and (b) ucorn – peak in any direc-
tion – for bidirectional analysis compared to the median deformation ubo from unidirectional
analysis; Υx = Υy = 0.78.
Unlike deformations, increases in bearing axial force extrema due to bidirectional excita-
tion depend on plan layout, and are greatest in square plans (b/d = 1) and minimal in narrow
plans (b/d → ∞). These trends are demonstrated in Fig. 8.6, where the maximum axial force
P+ /Pst is seen to increase – relative to unidirectional excitation – most for Plan S1, which is
square, less for Plan S2 (b/d = 2), and only slightly for Plan S3 (b/d = 3). This reflects the
tendency of a narrow building to rock mostly about its longer axis corresponding to its lower
rocking frequency ratio with minimal influence from the second component of excitation, while
a square plan with equal rocking frequencies about each axis (Υx = Υy ) is influenced by both
components of excitation.
Comprehensive design equations are developed to estimate the peak responses – deformation
ucorn and maximum and minimum axial forces P± /Pst – in symmetric isolation systems that
account for bidirectional excitation and rocking about both axes. These equations were devel-
oped by regression analysis of the normalized deformations (ūcorn ) and forces (P̄± /Pst ), which
were computed by RHA of Eq. 8.10 to each pair of ground motions in the LMSR ensemble.
Equation 8.10 is solved for building Plans S1-S9 with the variable-strength model for the bear-
205
(a) Plan S1 (b) Plan S2 (c) Plan S3
2 2 2
unidirectional unidirectional unidirectional
bidirectional bidirectional bidirectional
1.8 1.8 1.8
1 1 1
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
Tb (sec) Tb (sec) Tb (sec)
Figure 8.6: Maximum axial force P+ /Pst compared for bidirectional and unidirectional analy-
ses. For slenderness ratio h/d = 2 and Υx = 0.78, results shown for (a) Plan S1 (b/d = 1,
Υy = 0.78), (b) Plan S2 (b/d = 2, Υy = 1.00), and (c) Plan S3 (b/d = 3, Υy = 1.07).
ings, over the ranges Tb from 2 to 4.5 seconds, η from 0.25 to 1.5, building slenderness ratio
h/d from 0.5 to 2.0, and vertical frequency ratio Ωz = 30.4, which leads to rocking properties
Υx , Υy and Ωθz (Fig. 8.2, Table 8.1). Because the deformations and axial forces are essentially
independent of Ωz (Sec. 7.5.2), the design equations are also applicable to other values of Ωz .
Using methods that have been fine-tuned in previous chapters, the following equation
and is multiplied by u∗y (Eq. 3.6) – a function of Tb , η and median or design ground velocity
The regression coefficients for Tb and η determined from bidirectional response analysis were
found to be nearly identical to those determined earlier from unidirectional analysis (Sec. 7.6),
and were thus constrained to be the same. As a result, Eq. 8.19 varies from its one-directional
counterpart (Eq. 7.21) by a constant factor of about 1.13, which is identical to the result ignoring
rocking (Sec. 3.7). Despite the complexities of the system model and the possible variations in
parameters, the peak deformation can be described by a relatively simple equation (Eq. 8.19),
206
and the values for bidirectional and unidirectional analysis vary by not only a constant, but the
same constant as when rocking of the system was neglected entirely! Demonstrated for three
different plan layouts (Fig. 8.8a-c), Eq. 8.19 is a good approximation to the median deformation
ucorn .
Equations to estimate both the positive and negative force increments ∆P+ /Pst and
∆P− /Pst (deviation of force extrema from the static force, Eq. 7.18), which were not found to
be statistically distinct when averaged over many ground motions, were developed by methods
presented previously. That is, force increments rather than force extrema were used for statis-
tical analysis (Sec. 7.4.2); and equations were fit to the normalized force increment ∆P̄ /Pst ,
and multiplied by u∗y (Sec. 7.4.1) to obtain the actual force increment.
Two equations to estimate the force increment ∆P/Pst are presented. The first is given
as:
∆P 0.54
= f (Tb , η)Υx−1.66 Υy−0.54 u̇gyo (8.20)
Pst 4π 2
where
(−1.23+0.59 ln Tb )
f (Tb , η) = Tb η (0.05+0.24 ln η+0.65 ln Tb ) (8.21)
and the axial force extrema P+ /Pst and P− /Pst are computed by:
Equation 8.20, substituted into Eq. 8.22, is adequate to estimate the force extrema in symmetric
systems, as demonstrated in Fig. 8.7 for three different values of Υx and Υy . If this were the
final goal, no further improvement would be necessary. However, Eq. 8.20, when extended
to include the effect of eccentricity, was found to give poor estimates of the force extrema in
asymmetric-plan systems. For this reason the following improved equation is presented for
symmetric-plan systems:
0.48 0.28
∆P 0.46 ymax dyz xmax dxz
= f (Tb , η)Υx−1.42 Υy−0.68 u̇gyo (8.23)
Pst 4π 2 rx rx ry ry
where f (Tb , η) is given in Eq. 8.21, xmax and ymax are the x and y-direction components of
the distance from the CM to the outlying bearing; similarly dyz = (h/2)2 + e2by and dxz =
207
(h/2)2 + e2bx are the components in the y-z and x-z planes, respectively, of the distance from
the CM to the CR of the system. Recall also that h is the height of the building, and rx and
ry are radii of gyration about the x and y-axes. For symmetric-plan systems, the improvement
in Eq. 8.23 over Eq. 8.20 is marginal, as demonstrated in Figs. 8.8 and 8.7, which depict the
accuracy of the respective design equations by comparison with the exact median.
dyz
8.4.6 Interpretation of Regression Parameters ymax rx rx and xmax dxz
ry ry
The improvement in Eq. 8.23 for asymmetric-plan systems relative to Eq. 8.20 is due to the
additional regressors (explanatory variables): (ymax /rx · dyz /rx ) and (xmax /ry · dxz /ry ). These
“rocking” regressors can be shown to be analogous to the torsional regressor (1 + c/rz · eb /rz ),
which was applied to account for the increase in bearing deformation in an asymmetric-plan
system relative to a symmetric system (Sec. 4.5, Appendix 4.1). From a simplistic point-of-
view, the above regressor in the torsional problem related to motion or rotation about the
z-axis; hence similar regressors relating to rotation about the x or y-axes, could be expected
to contribute in predicting the rocking response. Because bearing axial force, the quantity to
a component of the rotation about either the x or y-axis, any parameter that can help predict
the rotation of the system about these axes will have some value in predicting bearing axial
forces.
An obvious difference between the two rocking regressors and the torsional regressor is
the additive factor of 1 in the torsional regressor. This factor is not needed for rocking; while
the torsional regressor relates to the increase in deformation contributed by torsion, rocking is
assumed to be the only contributor to the axial force increments. We now select one of the
rocking regressors (ymax /rx · dyz /rx ) and demonstrate its analogy to c/rz · eb /rz . First, eb is the
well-known eccentricity or distance from the CM to CR in the x-y plane, and dyz represents the
same distance in the y-z plane, because the structure CM is vertically h/2 from the isolation
system CR. Second, c is the absolute distance from the CM to the outlying bearing in the x-y
plane, while ymax is the y-component of this distance in the y-z plane. (The vertical component
208
(a) (b) (c)
20 20 20
15 15 η=.25 15
η=.25 η=.25
ucorn (cm)
0 0 0
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
1.5 1.5
1.5 1.0 1.0
1.0 0.75 0.75
1.5 0.75 1.5 0.5 1.5 0.5
0.5 η=.25 η=.25
η=.25
1 1 1
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
1 1 1
η=.25
0.5 η=.25 η=.25
0.5 0.75 0.5 0.5 0.5 0.5
P− /Pst
1.0 0.75
1.5 1.0 0.75
1.5 1.0
0 0 0 1.5
Figure 8.7: Design equation for bearing deformation ucorn (Eq. 8.19) and axial forces P+ /Pst
and P− /Pst (Eqs. 8.20 and 8.22), compared to the exact median by RHA to the LMSR ensemble,
for: (a) Plan S1, Υx = Υy = 0.78, (b) Plan S5, Υx = 0.63, Υy = 0.87 and (c) Plan S9, Υx = 0.58,
Υy = 0.82; all based on slenderness ratio h/d = 2.
209
(a) (b) (c)
2.5 2.5 2.5
Exact median
Design equation
2 2 2
P+ /Pst
1.5 1.5
1.5 1.0 1.0
1.0 0.75 0.75
1.5 0.75 1.5 0.5 1.5 0.5
0.5 η=.25 η=.25
η=.25
1 1 1
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
1 1 1
η=.25
0.5 η=.25 η=.25
0.5 0.75 0.5 0.5 0.5 0.5
P− /Pst
1.0 0.75
1.5 1.0 0.75
1.5 1.0
0 0 0 1.5
Figure 8.8: Improved design equation for axial forces P+ /Pst and P− /Pst (Eqs. 8.23 and 8.22),
compared to the exact median by RHA to the LMSR ensemble, for (a) Plan S1, Υx = Υy = 0.78,
(b) Plan S5, Υx = 0.63, Υy = 0.87 and (c) Plan S9, Υx = 0.58, Υy = 0.82 based on slenderness
ratio h/d = 2.
is dropped from this factor since it is parallel to the axial deformation that is relevant.) Finally,
since both eb and c in the torsional regressor are normalized by the applicable radius of gyration
rz , normalizations by the comparable radius rx are applied to the rocking regressor. The same
analogy extends to the second regressor (xmax /ry · h/2ry ) for rocking about the y-axis.
For symmetric buildings, it can be shown that the rocking regressors vary over a fairly
narrow range, i.e., with values ranging from 1.2 to 1.5. Furthermore, their values are not
independent of, but closely related to, the frequency ratios Υx and Υy , respectively (Eq. 8.17).
For these reasons, the additional regressors were not identified in Sec. 8.3.3 as parameters that
influence the response of the system. However, as already mentioned, their inclusion in Eq. 8.23
led to a substantial improvement in the accuracy of estimated axial forces for asymmetric-plan
buildings.
210
8.5 Asymmetric-Plan Systems
Next, response trends are studied and design equations are developed for asymmetric-plan sys-
symmetric systems. For this purpose, the corresponding symmetric system, first introduced
in Sec. 4.2.2, has properties Tb , η, Υx , Υy , and Ωθz identical to the asymmetric system, but
eccentricity eb /rz = 0. This definition does not uniquely determine a corresponding symmetric
Recall that plan-asymmetry is studied here by varying the stiffnesses and strengths of in-
dividual bearings to introduce eccentricity into a rectangular-plan system; this idealized system
approach was implemented in Plans S4-S9. The corresponding symmetric system to an ideal-
ized asymmetric system has the same plan layout without the variation of individual bearing
properties that induced the eccentricity; this simple means of comparison can be seen as an
The median peak corner deformation ucorn and force extrema (P+ /Pst and P− /Pst ) over
all bearings are computed as described in Sec. 8.4 for symmetric systems. For comparison
of asymmetric-plan and symmetric systems, the deformation and force-increment ratios are
defined as:
where the subscripts A and S refer to the asymmetric and corresponding symmetric systems, re-
spectively. Consistent with Sec. 7.4.2, force-increment ratios were found to be more meaningful
Median values of both deformation and force-increment ratios appear to increase linearly with
increasing eccentricity ebx /rz (Fig. 8.9). The slope of increase in these ratios varies widely, as
observed in the spreading of data in Fig. 8.9 with increasing eccentricity, but this variation was
the only parameter that has been linked to increases in both deformations and force extrema in
211
(a) (b) (c)
1.4 1.4 1.4
∆P̂+
∆P̂−
1.1 1.1 1.1
1 1 1
Figure 8.9: (a) Deformation ratio ûcorn and (b,c) force increment-ratios ∆P̂+ and ∆P̂− with
varying eccentricity ebx /rz . Not meant to be distinguishable, the lines represent different values
of Tb , η, Υx , and Υy applied to Plans S4-S9.
asymmetric-plan systems. This linear increasing relation breaks down slightly for the negative
force-increment ratio ∆P̂− (Fig. 8.9c), as values of this ratio can fall below unity in the small
eccentricity range (eb /rz 0.05) before resuming the increasing trend.
Actual peak deformations and force extrema, rather than their ratios, are compared for
a symmetric system (Plan S5) and idealized systems with several different values of ebx /rz in
Fig. 8.10. As depicted here, the increase in the deformation or force extrema for the largest
eccentricity considered, typically 10-20% (Fig. 8.9), does not seem that significant. Adjusting
the normalized strength η, for instance, from 0.5 to 1.0 has much greater consequence for the
response of the isolation system (Fig. 8.10). Nevertheless, plan asymmetry provides no benefit
Since deformation and force-increment ratios were found to vary only with eccentricity (Fig. 8.9),
Eqs. 8.19 and 8.23, which estimate lateral deformations and axial force increments in symmetric
for eccentricity.
bearings due to torsion. The normalized distance c/rz from the CM to the outlying bearing
212
(a) (b) (c)
15 2 1
1.6 0.6
P+ /Pst
P− /Pst
η=0.5
η=0.5
1.4 0.4
ebx /rz = 0
5 η=1.0
0.025
0.05
0.075 1.2 0.2
0.1
0.15
0 1 0
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
Tb (sec) Tb (sec) Tb (sec)
Figure 8.10: For a symmetric system and several different eccentricities ebx /rz , median re-
sponse spectra for (a) peak corner deformation ucorn , and axial forces (b) P+ /Pst and (c)
P− /Pst ; Plan 5 (Υx = 0.63, Υy = 0.87, Ωθz = 1.27) based on slenderness h/d = 2.
varies in actual asymmetric-plan systems and should be accounted for in the design equation.
When lateral-torsional coupling alone was considered, the peak deformation maximized over a
circle of radius c was shown to be a good upper bound estimate to the peak deformation at
some exterior bearing of distance c from the CM (Sec. 4.5.2), and a similar assumption is made
here to extend to systems with rocking. The peak deformation sampled at varying distances
from the CM in the idealized system serves as data to develop a design equation applicable to
general asymmetric plans. To this end, equations were fit to the deformation ratio ûco :
uco
ûco = (8.25)
(ucorn )S
symmetric system – maximized over a circle of radius c – and an upper bound to the peak
Regression analysis of data from the idealized systems (Plans S4-S9) and corresponding
symmetric systems led to an equation for ûco and hence the following design equation for the
213
total peak deformation in an asymmetric-plan system:
with u(corn )S computed from Eq. 8.19. The regression model in Eq. 8.26 was chosen to be the
same as when rocking was omitted (Sec. 4.5.2), and as required, reduces to the deformation in
a symmetric system when eb /rz = 0. Even with all the improvements to the model, Eq. 8.26
differs from the original estimate of deformation in asymmetric-plan systems excluding rocking
and vertical deformations (Eq. 4.25) only slightly. The largest discrepancy is in the coefficient
for the eccentricity term, 0.7 here compared to 0.88 in Eq. 4.25, which has only a small effect
Because the distance to outlying bearings has already been accounted for in symmetric
systems (Eq. 8.23, Sec. 8.4.6), regression analysis of the forces in the idealized asymmetric
and corresponding symmetric systems led to the following design equation for the peak force
with (∆P/Pst )S defined in Eq. 8.23. The regression coefficient for (1 + eb /rz ) was rounded from
1.01 to 1, which is consistent with the observed linear variation of force-increment ratios with
eb /rz (Fig. 8.9). As required, the force increments (Eq. 8.27) reduce to those of a symmetric
system when eb /rz = 0. (Actual forces are obtained by Eq. 8.22 with Eq. 8.27 for ∆P/Pst .)
Although analysis of the data for the force increment ratio ∆P̂ resulted in statistically distinct
coefficients for ∆P̂+ and ∆P̂− , separate coefficients were not adopted, as they would have
resulted in only inconsequential changes to the total estimated force increments (Eq. 8.27).
Also, possible correlation of ∆P̂ and other parameters (Tb , η, Υx , Υy and Ωθz ) was considered
by testing different regression models, but the influence of these additional parameters was
found to be negligible.
214
Table 8.2: Properties of Asymmetric-Plan Systems
Plan No. A1 A2 A3
ebx/rz 0.049 0.108 0
eby /rz 0.049 0 0.057
Ωθz 1.180 1.280 1.172
h 3t 6t d 2d 0.5d d
Υx 0.95 0.66 0.80 0.47 1.09 0.79
Υy 0.95 0.66 1.08 0.80 1.05 0.90
For asymmetric-plan systems, Eqs. 8.26 and 8.27 represent the best estimates of the
peak bearing deformation and axial-force increments (maximum or minimum) that consider
all modeling complexities introduced in later chapters, including rocking with bidirectional
excitation and axial-load effects. Because Eq. 8.26 for peak lateral deformation differs from
the equation that neglects rocking entirely (Eq. 4.25) only slightly, it is typically unnecessary
to include rocking and axial-load effects in the isolated-structure model to obtain a reasonable
estimate of the peak bearing deformation. On the other hand, rocking must be included to
The accuracy of these design equations (Eqs. 8.26 and 8.27), developed from response data
for idealized systems, is investigated for three asymmetric-plan systems, defined formerly in
Sec. 4.3 (Fig. 4.5a), whose properties are listed in Table 8.2. The eccentricities (ebx /rz , eby /rz )
and torsional frequency ratios (Ωθz ) of these plans are repeated here. The values listed for Υx
and Υy are based on two different assumed building heights h for each plan, which correspond
For strength coefficients µ = 0.05 and 0.1, each system in Table 8.2 was subjected to the
LMSR ensemble with median PGV of 35 cm/s and ωd = 3.05, which led to normalized strengths
η = 0.46 and 0.92 (Sec. 4.5.3). The deformation of each corner bearing was computed two ways:
(1) as the median (determined by nonlinear RHA) over the LMSR ensemble and (2) estimated
from Eq. 8.26; and then maximized over all corners for both. Forces were treated similarly,
215
except since the system can rock about any axis, bearings located around the perimeter were
tested for occurrence of the largest axial forces. For each asymmetric plan, the exact median
deformation and maximum axial force are compared to the corresponding design equation in
Fig. 8.11. The minimum axial force, which has been omitted in Fig. 8.11, showed similar trends.
Equation 8.26, which differs for each plan only due to differing values of eccentricity, is
a good estimate of the exact (median) peak deformation over all corners (Fig 8.11a). Equa-
tion 8.26, based on the deformation maximized over a circle of radius rz and hence an upper
bound to the corner deformation, is not necessarily conservative as it was for systems without
rocking (Table 4.1.) Nevertheless, the discrepancy between the exact median and the design
On the other hand, Eq. 8.27 is not always a good estimate of the exact axial forces
(Fig. 8.11b,c). While this design equation can provides reasonable force estimates (Fig. 8.11b),
for smaller values of Υx and Υy typical of more slender buildings (Fig. 8.11c) it noticeably
underestimates the positive axial force for two of the three plans (A1 and A3). This lack-of-fit
also seems to be worse for η = 0.92 than for η = 0.46. The visual impression of Fig. 8.11b,c is
validated by Fig. 8.12, where the percent discrepancy in Eq. 8.27 relative to the exact median
is plotted for the positive force increment. This discrepancy ranges in value from -25% to 25%,
and is often greater than 10% in magnitude. Also, increases in Υx , Υy do not necessarily lead
to reduced percent error; Fig. 8.11 may be deceptive since the force increments corresponding
Unfortunately, errors as large as 20%, especially on the unconservative side, imply that
Eq. 8.27 has limited value in estimating the axial forces in bearings. Goodness of fit measures
and error analysis (Fig. 8.13) together indicate that the design equations can reasonably predict
forces in the idealized systems by which they were developed. Specifically, the design equations
appear to estimate the maximum axial force pretty accurately for a range of idealized systems
(Fig. 8.13a-b), and percent errors are typically held within 10% (Fig. 8.13c). Thus, it would
appear that the difficulties mentioned in Sec. 8.3.1 prevent successful application of the idealized
system approach; and a good fit cannot be found because idealized systems are no longer
216
(a)
15 15 15
Plan A1 Plan A2 Plan A3
η=0.46 η=0.46 η=0.46
10 10 10
uco (cm)
5 5 5
Υx =0.95 Υx =0.80 Υx =1.09
Υx =0.66 Υx =0.47 Υx =0.79
Design equation Design equation Design equation
0 0 0
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
(b)
2.5 2.5 2.5
Plan A1, Υx =0.95 Plan A2, Υx =0.80 Plan A3, Υx =1.09
2 η=0.46 2 2
P+ /Pst
η=0.92
1 1 1
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
(c)
2.5 2.5 2.5
Plan A1, Υx =0.66 Plan A2, Υx =0.47 Plan A3, Υx =0.79
η=0.46
2 2 2 η=0.92
P+ /Pst
1 1 1
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
Tb (sec) Tb (sec) Tb (sec)
Figure 8.11: For asymmetric plans A1-A3 (Table 8.2), comparison of exact median and design
equation for (a) peak bearing deformation uco (Eq. 8.26); and (b,c) maximum axial forces
(Eq. 8.27), separated for different Υx . In each case, a dotted line for the design equation is
shown next to a solid line for the exact median.
217
(a) Plan A1 (b) Plan A2 (c) Plan A3
30 30 30
Υx =0.95, η=0.46 Υx =1.09, η=0.46
Υx =0.95, η=0.92 Υx =1.09, η=0.92
20 20 20
Υx =0.66, η=0.46 Υx =0.79, η=0.46
% Discrepancy in ∆P+
0 0 0
Figure 8.12: Percent discrepancy in the design equation (Eq. 8.27) for positive axial force
increment ∆P/ Pst compared to its exact median for asymmetric plans (a) A1, (b) A2, and (c)
A3.
fully representative of general asymmetric-plan systems when rocking and axial-load effects are
included in the analysis. These results seem to suggest that the best way to interpret rocking
basis, and applicability of any simplified code procedure should reflect this.
8.6 Conclusions
To accurately determine the controlling response of the isolation system, ground excitation in
tion) of the system should be considered. For this purpose, a previously developed model for
lead-rubber bearings that includes axial-load effects, or lateral vertical coupling, was enhanced
to accommodate motion in two lateral directions. The improved bearing model was incorpo-
rated into symmetric and asymmetric-plan base-isolated building models, and relevant peak
responses were determined by response history analysis to the LMSR strong ground motion en-
semble. Design equations were developed to estimate both peak lateral deformation (Eqs. 8.19
and 8.26) and peak axial forces (Eqs. 8.23 and 8.27) in the bearings for both symmetric and
eral asymmetric-plan systems, an idealized asymmetric system was analyzed, where eccentricity
218
(a)
2.5 2.5 2.5
η=.25 Plan S5 Plan S7 Plan S9
0.5
2 0.75 2 2
P+ /Pst
1.0
1.5 1.5 1.5 1.5
1 1 1
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
(b)
2.5 2.5 2.5
Plan S5 Plan S9
2 2 2
P+ /Pst
(c)
% Discrepancy in ∆P+
30 30 30
20 Plan S5 Plan S7 20
Plan S9
20
10 10 10
0 0 0
−10 −10 −10
−20 −20 −20
−30 −30 −30
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
Tb (sec) Tb (sec) Tb (sec)
Figure 8.13: Design equation for positive axial forces P+ /Pst (Eqs. 8.27 and 8.22) compared
to the exact median by RHA to the LMSR ensemble, for idealized plans S5, S7 and S9, based
on: (a) h/d=1 and (b) h/d=2; (c) percent discrepancy in the design equation (Eq. 8.27) for the
above cases, with h/d=1 (solid lines) and h/d=2 (dashed lines).
was introduced by varying the stiffnesses and strengths of individual bearings in a rectangular
plan. This investigation, the culmination of work in previous chapters, supports the following
conclusions:
symmetric buildings. Rocking of the structure causes variation of the axial loads, which
in turn cause the stiffnesses and strengths of individual bearings to vary and can induce
a time-varying eccentricity. However, accidental torsion in the isolation system from this
219
2. The peak isolator deformation can generally be determined to sufficient accuracy using
a single degree-of-freedom model of the system, as this deformation was only modestly
influenced by refinements to the global model of the structure or individual bearing mod-
els. It was already concluded in Chapter 7 that even if rocking is neglected entirely,
response spectra for lateral deformation are within 10% of those when rocking and axial-
load effects are included. In extending these results to consider bidirectional excitation
and plan-asymmetry, remarkable parallels are drawn between systems with and without
rocking. In both cases, the design equation for deformation is amplified by 13% for bidi-
rectional excitation. Similar functions of eccentricity were used to account for effects of
plan-asymmetry in systems with and without rocking. This ability to simply and ac-
curately predict the isolator deformation may be lost if structural flexibility, neglected
conservative.
3. Unlike lateral deformation, the variation of axial forces in the isolators is directly related
and Υy for rocking about the x and y-axes. The maximum and minimum axial forces
in symmetric systems are influenced by these frequency ratios, but not independently by
other rocking-related parameters such as the building slenderness ratio, the plan aspect
ratio, or distribution of bearings over the plan. The increase in axial force extrema for
bidirectional excitation is largest in square plans and negligible in narrow plans, and is
4. An anology was drawn between rocking and torsion, where in both cases, rotation about
the center of mass axis contributes to the response of interest, which can be computed
as a function of radial distance from this center of mass. Using this comparison to tor-
sion, additional regression parameters were identified for variation of axial forces due to
rocking, and inclusion of these parameters led to some improvement in the design equa-
tions, particularly for asymmetric-plan buildings. However, the majority of the torsional
220
effects were accounted for by the eccentricity while the majority of the rocking effects
5. The design equations for axial forces are complex with many factors and coefficients that
some of the attraction of the method as originally applied to deformations has been lost.
Furthermore, while the design equations to estimate axial forces are accurate for sym-
metric systems and for the idealized asymmetric systems for which they were developed,
they are shown to err by up to 25% when applied to actual asymmetric-plan systems.
Thus, the idealized system approach seems to have reached its limit of application, i.e.,
when rocking and axial-load effects are included. Alternatively, design axial forces in
221
Appendix 8A: Notation
rocking deformations
h building height
ings
222
nbx , nby number of bearings across the plan in the x and y-directions, for
symmetric systems
PE Euler buckling load, neglects shear stiffness of the bearing: π 2 EIs /h2b
∆P+ , ∆P− maximum and minimum axial force increments; maximum force de-
bearings
sxi , syi deformations of shear spring (in two-spring model of ith bearing) in
x and y-directions
tion system
µg/(ωd u̇gyo )
ness
tral regions
224
REFERENCES
[1] International Code Council, Falls Church, VA. International Building Code, 2000.
[2] Naeim F and Kelly JM. Design of Seismic Isolated Structures: From Theory to Practice.
[3] Skinner RI, Robinson WH, and McVerry GH. An Introduction to Seismic Isolation. John
[4] Kawamura S et.al. Seismic isolation retrofit in Japan. Proc., 12th World Conference on
Earthquake Engineering, Paper No. 2523. New Zealand Society for Earthquake Engineer-
[5] Saito H et.al. Structural design of base-isolated high-rise buildings with high-damping
[6] Griffith MC, Aiken ID, and Kelly JM. Displacement control and uplift restraint for base-
[7] Logiadis I, Zilch K, and Meskouris K. Prestressed bearings in the seismic isolation of
structures. Proc., 11th World Conference on Earthquake Engineering, Paper No. 1895.
[8] Nagarajaiah S, Reinhorn AM, and Constantinou MC. Experimental study of slid-
ing isolated structures with uplift restraint. ASCE Journal of Structural Engineering,
118(6):1666–1683, 1992.
[9] Wang X-F and Gould PL. Dynamics of structures with uplift and sliding. Earthquake
225
[10] Kelly JM. Dynamic and Failure Characteristics of Bridgestone Isolation Bearings, Rep.
[11] Clark PW, Aiken ID, and Kelly JM. Experimental Studies of the Ultimate Behavior of
[12] Youssef N et.al. Building case study: Los Angeles City Hall. The Structural Design of Tall
[13] Kelly JM. Tension buckling in multilayer elastomeric bearings. ASCE Journal of Engi-
[15] Kelly JM, Buckle IG, and Koh C-G. Mechanical Characteristics of Base Isolation Bearings
for a Bridge Deck Model Test, Rep. No. UCB/EERC-86/11. Earthquake Engineering
[16] Aiken ID, Kelly JM, and Tajirian FF. Mechanics of Low Shape Factor Elastomeric Seismic
[17] Tyler RG and Robinson WH. High-strain tests on lead-rubber bearings for earthquake load-
ings. Bulletin of the New Zealand National Society for Earthquake Engineering, 17(2):90–
105, 1984.
[18] Hwang J-S and Hsu T-Y. Experimental study of isolated building under triaxial ground
226
[20] Krawinkler H. Personal Communication, 2001.
[21] Chopra AK. Dynamics of Structures: Theory and Applications to Earthquake Engineering,
[22] Huang W-H. Bi-directional Testing, Modeling, and System Response of Seismically Isolated
[23] Makris N and Chang S-P. Effect of viscous, viscoplastic, and friction damping on the
29(1):85–107, 2000.
[24] Mahin SA and Lin J. Construction of Inelastic Response Spectra for Single-Degree-of-
[25] Sehayek S. Effect of Ductility on Response Spectra for Elasto-Plastic Systems, Pub. R76-42.
[26] Murakami M and Penzien J. Nonlinear Response Spectra for Probabilistic Seismic Design
[27] Fenves GL et.al. Analysis and testing of seismically isolated bridges under bi-axial excita-
tion. Proc., Fifth Caltrans Seismic Research Workshop. California Dept. of Transportation,
[28] Chopra AK and Goel RK. Evaluation of NSP to estimate seismic deformation: SDF
[29] Anderson E and Mahin SA. Displacement-based design of seismically isolated bridges.
227
[30] Fajfar P. Capacity spectrum method based on inelastic demand spectra. Earthquake
[31] Wadsworth HM, editor. Handbook of Statistical Methods for Engineers and Scientists, 2nd
[32] Montgomery DC, Runger GC, and Hubele NF. Engineering Statistics. John Wiley & Sons,
1998.
[33] Metcalfe AV. Statistics in Civil Engineering. Arnold Applications of Statistics Series.
[34] Mosqueda G, Whittaker AS, and Fenves GL. Characterization and modeling of friction
[35] Pan TC and Kelly JM. Seismic response of torsionally coupled base-isolated structures.
[37] Jangid RS and Kelly JM. Torsional displacements in base-isolated buildings. Earthquake
[38] Lee DM. Base-isolation for torsion reduction in asymmetric structures under earthquake
[39] Eisenberger M and Rutenberg J. Seismic base isolation of asymmetric shear buildings.
[40] Nagarajaiah S, Reinhorn AM, and Constantinou MC. Torsion in base-isolated structures
2951, 1993.
228
[41] Jangid RS and Datta TK. Nonlinear response of torsionally coupled base isolated structure.
[42] Scheller J and Constantinou MC. Response History Analysis of Structures with Seismic
Isolation and Energy Dissipation Systems: Verification Examples for Program SAP2000,
Rep. No. MCEER 99-02. Multidisciplanary Center for Earthquake Engineering Research,
[43] Constantinou MC, Mokha A, and Reinhorn AM. Teflon bearings in base isolation.
[44] Bondonet G and Filiatrault A. Frictional response of PTFE sliding bearings at high
[45] Constantinou MC, Tsopelas P, Kasalanati A, and Wolff ED. Property Modification Factors
for Seismic Isolation Bearings, Rep. No. MCEER 99-12. Multidisciplanary Center for
[46] Morgan TA. Characterization and Seismic Performance of High-Damping Rubber Isolation
[47] Koh C-G and Kelly JM. Effects of Axial Load on Elastomeric Isolation Bearings, Rep.
[48] Chalhoub MS and Kelly JM. Reduction of the Stiffness of Rubber Bearings due to Com-
[49] Aiken ID et.al. Experimental studies of the mechanical characteristics of three types
[50] Simo JC and Hughes TJR. Computational Inelasticity. Springer, New York, NY, 1998.
229
[51] Pan TC and Kelly JM. Seismic response of base-isolated structures with vertical-rocking
[52] Lobo RF, Naeim F, and Hart GC. 3D-nonlinear analysis of multistory base isolated build-
ings with significant uplift. Proc., Sixth US National Conference on Earthquake Engineer-
[53] Ryan K and Hall JF. Aspects of building response to near-source ground motions. Proc.,
Structural Engineering World Wide, Paper T162–5. Oxford, England, Elsevier Science
Ltd, 1998.
[54] Hall JF and Ryan KL. Isolated buildings and the 1997 UBC near-source factors. Earthquake
[55] Koh C-G and Balendra T. Seismic response of base-isolated buildings including P -δ effects
1989.
[56] Goel RK and Chopra AK. Vibration Properties of Buildings Determined from Recorded
[57] Almazan JL and de la Llera JC. Lateral torsional coupling in structures isolated with the
Paper No. 1534/6/A. New Zealand Society for Earthquake Engineering, Upper Hut, New
Zealand, 2000.
[58] Almazan JL and de la Llera JC. Analytical model of structures with frictional pendulum
[59] Almazan JL and de la Llera JC. Physical model for dynamic analysis of structures with
230
[60] Almazan JL and de la Llera JC. Accidental torsion due to overturning in nominally sym-
metric structures isolated with the FPS. Earthquake Engineering and Structural Dynamics,
32(6):919–948, 2003.
231