MODELING TRIPLE FRICTION PENDULUM

ISOLATORS IN PROGRAM SAP2000

A. A. S. Sarlis1 and M.C. Constantinou2

June 27, 2010

1
Graduate Student, Department of Civil, Structural and Environmental Engineering, University at Buffalo, State
University of New York, Buffalo, NY 14260
2
Professor, Department of Civil, Structural and Environmental Engineering, University at Buffalo, State University
of New York, Buffalo, NY 14260.

1
 1. INTRODUCTION

This document presents guidelines on modeling the Triple Friction Pendulum isolator in program
SAP2000. Particularly, this document presents the following:

1) A summary description of the Triple Friction Pendulum isolator behavior.

2) Presentation of the parallel model for modeling Triple Friction Pendulum isolators.
3) Implementation of parallel model in SAP2000.
4) Use of direct integration and fast nonlinear analysis in SAP2000.
5) Modeling of global damping in seismically isolated structures in SAP2000.
6) Significance of Ritz vector modes in capturing correctly the response of seismically
isolated structures in SAP2000.
7) Consideration of isolator P- effects in direct integration and in fast nonlinear analysis in
SAP2000.
8) Verification examples including input files for modeling Triple Friction Pendulum
isolators using the parallel and the series models.

While the document concentrates on the Triple Friction Pendulum isolator, the issues discussed
on damping specification, Ritz vector modes, and P- analysis also apply to other isolation
systems. Moreover, while the presentation is restricted to program SAP2000, the concepts are
transferable to program ETABS that shares the same features as SAP2000 but limited to fast
nonlinear analysis.

2. DESCRIPTION OF TRIPLE FRICTION PENDULUM ISOLATOR

The Triple Friction Pendulum (FP) isolator exhibits multiple changes in stiffness and strength
with increasing amplitude of displacement. Details on the behavior of the isolator, including test
data, shake table testing and comparison of analytical and experimental data may be found in
Morgan (2007) and Fenz and Constantinou (2008a to 2008e) .

The construction of the force-displacement loop is complex as it may contain several transition
points which depend on the geometric and frictional properties. Figure 1 shows the geometry of
a Triple FP bearing. Its behavior is characterized by radii R1, R2, R3 and R4 (typically R1=R4 and
R2=R3), heights h1, h2, h3 and h4 (typically h1=h4 and h2=h3), distances (related to displacement
capacities) d1, d2, d3 and d4 (typically d2=d3 and d1=d4) and friction coefficients 1 , 2 , 3 and 4
(typically 2 = 3 ). The actual displacement capacities of each sliding interface are given by:
Reffi
d i*  d , i  1...4 (1)
Ri i
Quantity Reffi is the effective radius for surface i given by:

Reffi  Ri  hi , i  1...4 (2)

2
 Figure 1 Triple FP Bearing Definition of Dimensional and Frictional Properties

The lateral force-displacement relation of the Triple FP isolator is illustrated in Figure 2. Five
different loops are shown in Figure 2, each one valid in one of five different regimes of
displacement. The parameters in the loops relate to the geometry of the bearing, the friction
coefficient values and the gravity load W carried by the isolator as described in Fenz and
Constantinou (2008a and 2008b). Triple FP isolators are typically designed to operate in
regimes I to IV, whereas regime V is reserved to act as a displacement restrainer. In regime V
the isolator has consumed its displacement capacities d1 and d4 and only slides on surfaces 2 and
3 (see Figure 1).

Figure 2 Force-Displacement Loops of Triple FP Bearing

3
Table 1 (adopted from Fenz and Constantinou, 2008a) presents a summary of the force-
displacement relationships of the Triple FP bearing in the five regimes of operation. Note that
Ffi  iW is the friction force at interface i and W is the axial compressive load on the bearing.
Consider the special case in which Reff 1  Reff 4 , Reff 2  Reff 3 , d1  d 4 , d 2  d 4 , 1   4 and
* * * *

 2   3 . Furthermore, consider that the bearing does not reach regime V. The result is an
isolator with tri-linear hysteretic behavior as illustrated in Figure 3. Note this special case
represents a typical case of configuration of Triple FP isolators.

The force-displacement relation of the special Triple FP isolator is described by:

W  R 
F u  1  (1   2 ) eff 2  W (3)
2 Reff 1  Reff 1 

Force

2μ2W
μ1W
μW W/2Reff1

W/2Reff2
μ2W

u*=2(μ1-μ2 )Reff2 Displacement

2u*

Figure 3 Force-Displacement Loop of Special Triple FP Isolator

The relation given by equation (3) is valid until the lateral force and displacement reach the
values given by the following equations:

W *
F d1  F f 1 (4)
Reff 1

u  u*  2d1*  2(1  2 ) Reff 2  2d1* (5)

4
 u*  u**  2( 1  2 ) Reff 2 (6)

The force at zero displacement is given by

 Reff 2 
W   1  ( 1  2 ) W (7)
 Reff 1 

5
 Table 1 Summary of Triple FP Bearing Behavior (Nomenclature Refers to Figure 1)

Regime Description Force-Displacement Relationship

W Ff 2 Reff 2  Ff 3 Reff 3
F u
I
Sliding on surfaces 2 Reff 2  Reff 3 Reff 2  Reff 3
and 3 only
Valid until: F  Ff 1 , u  u    1   2  Reff 2   1  3  Reff 3
Motion stops on
II surface 2; Sliding on
surfaces 1 and 3 Valid until: 
F  Ff 4 , u  u  u   4  1  Reff 1  Reff 3 
W Ff 1  Reff 1  Reff 2   Ff 2 Reff 2  Ff 3 Reff 3  Ff 4  Reff 4  Reff 3 
Motion is stopped on F u
surfaces 2 and 3; Reff 1  Reff 4 Reff 1  Reff 4
III
Sliding on surfaces 1  Reff 4 
   4  1   Reff 1  Reff 4 
W *
and 4 Valid until: F  Fdr1  d1  F f 1 , u  udr1  u  d1*  1 
Reff 1  Reff 1 

Slider contacts W W *
restrainer on surface F  u  udr1   d1  F f 1
Reff 2  Reff 4 Reff 1
1; Motion remains
IV
stopped on surface 3;  d *   d* 
d 4  F f 4 , u  udr 4  udr1   4   4    1  1    Reff 2  Reff 4 
W *
Sliding on surface 2 Valid until: F  Fdr 4 
   Reff 1 
and 4 Reff 4  Reff 4    
Slider bears on
W W *
V
restrainer of surface 1 F  u  udr 4   d 4  Ff 4
and 4; Sliding on Reff 2  Reff 3 Reff 4
surfaces 2 and 3
Assumptions: (1) Reff 1  Reff 4  Reff 2  Reff 3 , (2)  2  3  1   4 , (3) d1*    4  1  Reff 1 , (4), d 2*   1   2  Reff 2
(5) d 3*    4   3  Reff 3

6
 3. MODELING OF TRIPLE FP ISOLATOR IN SAP2000
3.1 Introduction

The series model has been developed by Fenz and Constantinou (2008d, e) in order to model
behavior of the Triple FP bearings in all five regimes of operations. The series model, although
unable to provide information on the motion of the internal components, is an exact
representation of the triple FP bearing which indeed behaves as a series arrangement of single FP
elements. However, the series model requires the introduction of artificial masses at the
connecting joints of the FP elements and the use of a large number of degrees of freedom per
bearing. These requirements typically have the following effects:

1) An increased computational effort for analysis.

2) The analysis produces results that are sensitive to the selection of values for secondary
parameters used in the model, such as link element masses, effective stiffness and the
elastic stiffness.
3) The required number of Ritz vector modes for use in fast nonlinear analysis is increased.

The parallel model is a much simpler model capable of describing the behavior of the special
case Triple FP bearing, for which Reff 1  Reff 4 , Reff 2  Reff 3 , d1  d 4 , d 2  d 4 , 1   4 ,  2   3
* * * *

and the bearing does not enter the final regime of operation (stiffening). The parallel model was
originally described in Sarlis et al (2009), where it was implemented in computer program 3D-
BASIS (Nagarajaiah et al, 1989 and Tsopelas et al, 1994 and 2005). This report describes the
implementation of the parallel model in program SAP2000 (and, through its similarity, to
program ETABS).

The parallel model consists of a parallel arrangement of FP elements instead of a series

arrangement of such elements. It has the following limitations:

1) Is incapable of exactly capturing the motion of the internal components of the triple FP
bearing. However, the motion may be derived on the basis of the theory described in
Fenz and Constantinou (2008a, 2008b), except for very small displacements.
2) The above limitation results in inability to exactly describe the velocity dependence of
the friction coefficient. This limitation is bypassed by using the theory in Fenz and
Constantinou (2008a, 2008b) to partition the total bearing motion to the two main sliding
surfaces. Accordingly, the model describes well the velocity dependence of the friction
coefficient except for the case of very small velocities.
3) The model cannot describe the behavior of the bearing in regime V as shown in Figure 2
(stiffening range).

While the parallel model has been extended to describe the more general configuration of the
triple FP where different friction coefficients are used at each sliding interface, this report
specifically focus on the special case isolator due to its simplicity and wide practical use. The
extended, so-called bundle model, will be described in another report.

7
 3.2 Description of Parallel Model

The parallel model consists of two FP elements arranged in parallel and connected to two shared
nodes. The first element (called FP1) represents a flat slider. The second element (called FP2) is
a single Friction Pendulum element. The two elements share the same joints (nodes) and they
overlap each other. The assembly is shown in Figure 4. The top joint is connected to the structure
above, while the bottom joint is either fixed on the ground or rigidly connected to the structure
below. Due to their arrangement, and provided the two elements are assigned equal axial
stiffness, the elements equally share the total load on the isolator. The properties of the FP
elements are selected as schematically depicted in Figure 5 to represent the global behavior of
the Triple FP isolator (details of the parameters are provided in the sequel).

Figure 4 Parallel Model of Triple FP in SAP2000

*
     2W
F F
 

 2W
K1  1W
K2

u u

Y1 W
0 
FP1 R2 FP2

Figure 5 Force Displacement Loops of Elements FP1 and FP2 of the Parallel Model

The use of overlapping elements results in zero rotational stiffness of the element, which
correctly represents the actual behavior. Note that earlier descriptions of the parallel element
(Sarlis et al, 2009) placed the two elements of the model at a distance between them that created
an artificial rotational stiffness. The rotational stiffness may affect the accuracy of the results of
analysis.

8
3.3 Parameters of Parallel Model

3.3.1 Limit on Displacement

The parallel model cannot capture the stiffening behavior of the triple FP bearing and therefore
should not be used when the calculated displacement from response history analysis exceeds the
limit (see Table 1):

udr 4  2  1  2  Reff 2  d1*  (8)

3.3.2 Friction Coefficient

The friction coefficients of elements FP1 and FP2 (see Figure 5) are calculated so the sum of the
forces in the assembly of FP1 and FP2 elements equals the actual forces shown in Figure 3:

 2W   1W  K 2 Y 1
(9)
1W   1W   2 W (10)

In the above equations,

1 is the friction coefficient of the FP1 element
 2 is the friction coefficient of the FP2 element
W is the axial load in the FP1 and FP2 elements, which equals to W/2.
K 1 is the elastic stiffness of the FP1 element.
K 2 is the elastic stiffness of the FP2 element
R 2 is the radius of curvature of the FP2 element.
Y 1 is the yield displacement of the FP1 element. This is a small quantity, typically assumed to be
about 0.01inch or larger (although see example of calculation later in this document).
i is the friction coefficient of i-th surface of the triple FP isolator, where i=1,2,3 or 4.
W is the axial load carried by the bearing.

Assuming that the yield displacement of FP1 in SAP2000 is a small number and using W  W / 2

 1  2 2 (11)

 2  21   2  (12)

The friction coefficients are defined at fast velocities (related to parameter FAST in SAP2000).
For slow velocities, the friction coefficient (related to parameter SLOW in SAP2000) may be
assumed to be half of the fast velocity value.

9
3.3.3 Radii of Curvature

The FP1 is a flat slider (in SAP2000 this condition is represented as one with radius equal to
zero). The post-elastic stiffness of element FP2 is equal to the post-elastic stiffness of the
bearing, so that:

W W W (13)
 
2 Reff 1 R 2 2 R 2

Therefore, R 2  Reff 1 (14)

3.3.4 Elastic Stiffness

The elastic stiffness of the bearing when motion in regime II starts is equal to the elastic stiffness
of the FP2 element of the parallel model.

W (15)
K2 
2 Reff 2

The elastic stiffness of element FP1 can be calculated as follows

 2W W (17)
K1  
2Y 1 2 Reff 2

3.3.5 Force Intercept at Zero Displacement

The force intercept at zero displacement normalized by the axial load (or coefficient of friction at
zero displacement), is given by the following expressions (see Figures 3 and 5):

 2W Reff 1  Reff 2
 2  1  2 
*
2  2  (18)
K 2 R2 Reff 1
*
 1   1  2 2 (19)

3.3.6 Rate Parameter

The rate parameter specified for coefficients of friction  1 and  2 should be half the values of
the rate parameter for the coefficients of friction 2 and 1 , respectively of surfaces 2 (or 3) and
1 (or 4) of the bearing. Note that this is due to fact that the parallel model is limited to cases in

10
which 2  3 and 1  4 for which the sliding displacements and sliding velocities on surfaces
1 and 4 are equal to half of the total displacement and velocity (this is only an approximation for
small displacements).

3.3.7 Effective Stiffness

The effective stiffness should be specified to be low in order to minimize damping leakage in the
isolation system. In fast nonlinear analysis (FNA), SAP2000 constructs a global damping matrix
using the specified effective stiffness of the link elements in the nonlinear isolation system.
Accordingly, artificial viscous damping elements are introduced in the isolation system.
Specification of a small effective stiffness ensures that the artificial damping is low to affect the
calculated response and prevents magnification of damping in nonlinear dynamic analysis as the
element develops an instantaneous effective stiffness that is less than the initial specified value.
It is advised that an appropriate value for the effective stiffness to use is the value of the post
elastic stiffness (or less) of each element, although some trial and error investigation may be
needed to confirm that that damping leakage is minimal. Note that the effective stiffness is also
used to calculate Ritz vector modes and, therefore, the values specified have some additional
effect in FNA. As it will be discussed later in this report, the results of FNA can be exact if an
appropriate number of Ritz vector modes are used regardless of the specified value of effective
stiffness.

3.3.8 Vertical Stiffness

It is important that the FP1 and FP2 elements in the parallel model have the same vertical
stiffness so that they equally share the axial load on the bearing. Otherwise, the model is invalid.

For the triple FP bearing, it is appropriate to calculate the vertical stiffness as the axial stiffness
of a column having the height of the bearing, diameter of the inner slider and modulus of
elasticity equal to about one half the modulus of elasticity of steel in order to account for the
some limited flexibility in the bearing. The calculated vertical stiffness is then equally divided to
the FP1 and FP2 elements.

3.3.9 Shear Deformation Location

The shear deformation location should be specified at the mid-point of the two nodes of the FP1
and FP2 elements. This parameter has significance only in the calculation of the moment
transferred by the bearing to the structure above and below (including P- moments, if such
effects are included). The use of the mid-point for the shear deformation location is strictly valid
for the special case isolator considered herein. (When friction on the two main sliding surfaces
is different, the shear deformation location should be selected on the basis of the partition of the
isolator displacement-theory in Fenz and Constantinou, 2008a and 2008c. It will be closer to the
surface of the largest friction).

11
3.3.10 Element Height

The element height should be the same as the total height of the isolator including the thickness
of the top and bottom concave plates in order to correctly calculate the moment acting at the
bottom of the bearing concave plates (including P- moments, if such effects are included).

3.3.11 Rotational Stiffness (degrees of freedom R2, R3)

The rotational stiffness of elements FP1 and FP2 should be zero. Given that the two elements
are at the same location, the combined rotational (or bending) stiffness of the model is zero. This
is very close to the actual behavior of the isolator, which has minor resistance to rotation (the
resistance results from friction tractions on the inner spherical sliding surfaces).

3.3.12 Torsional Stiffness (degree of freedom R1)

The torsional stiffness of elements FP1 and FP2 should be zero. The isolator has insignificant
rotational resistance.

3.3.13 Link Element Mass

For each of the FP1 and FP2 elements, we recommend the use of a value equal to half of the total
mass of the isolator. Use of larger values for the link element mass may speed up execution of
analysis although results may be slightly affected.

3.3.14 Link Element Rotational Mass Moment of Inertia

A rotational mass moment of inertia must be assigned for the FP1 and FP2 elements for proper
execution of the program. We recommend the approximate calculation of a mass moment of
inertia for the bearing using equation (20) below and then dividing the moment equally to the
two elements. Use of larger values for the link element mass moment of inertia may speed up
execution of analysis although results may be slightly affected.

misolator  ( DR 2  h 2 )
I FP  (20)
12

In this equation, misolator is the mass of the bearing, DR is the diameter of the rigid slider of the
bearing and h is the bearing height.

3.3.16 Addition of Gap Element to Parallel Model

It would appear as straight forward exercise to add gap elements in parallel to the Triple FP
isolator parallel model to capture the stiffening behavior of the isolator in regime V. However,
the behavior produced is not exactly correct. This is due to the fact that a gap element in
SAP2000 engages and disengages at specified displacements (the gap displacement) while this is

12
not true for the Triple FP bearing. Use of a gap element will produce correct stiffening behavior
but the unloading branch of the hysteresis loop will be incorrect. Note that the same problem
exists when other approximations are used to model the Triple FP model (e.g., the bilinear
hysteretic model) together with gap elements. Only the series model (Fenz and Constantinou,
2008d) of the Triple FP bearing is capable of correctly modeling this behavior. However, a gap
element added in parallel to the parallel element of the Triple FP bearing produces correct results
for the very special case of 1  2  3  4 .

A summary of element properties for the parallel model is presented in Table 2.

Table 2 FP1 and FP2 Element Properties in Parallel Model of Triple FP Bearing

FP1 FP2
W W
Supported Weight W
2 2
 2W W W
Elastic Stiffness K 1 , K 2 
2Y 1 2 Reff 2 2Reff 2
Friction Coefficient  1 ,  2 2 2 21   2 
Normalized Force Intercept at Reff 1  Reff 2 )
2 2 2(1   2 )(
Zero Displacement  1 ,  2
* *
Reff 1
a1 a1
Rate Parameter a 2 2
Radius of Curvature R1 , R 2 0(flat) Reff 1
Element Height h h
h h
Shear Deformation Location
2 2
Rotational Stiffness (R2,R3) 0 0
Torsional Stiffness (R1) 0 0
Link Element Mass1 misolator / 2 misolator / 2
Rotational Mass Moment of I FP 

misolator DR  h 2
2
 I FP 
 2
misolator DR  h 2 
Inertia1
24 24
1 Minimum value. A larger value may be specified.
DR=diameter of rigid slider, h=bearing height, Y 1 =yield displacement equal to about 0.01inch or larger

13
3.4 Global Analysis Parameters in SAP2000

3.4.1 Damping Specification in Fast Nonlinear Analysis (FNA)

The specification of structural damping in the analysis of seismically isolated structures (with
nonlinear isolator elements) may appear to be of secondary importance. However, if improperly
specified, damping may have significant effects.

When modeling an isolated structure, the inherent damping of the structure exclusive of the
isolation system is assigned based on the type structure (height, materials, construction method)
and expected level of deformation. For a building, the damping ratio is only approximately
known for the superstructure when fixed at the isolator locations. For example, the data
presented in Satake et al (2003) may be used to select the values of damping ratio for the
superstructure of isolated buildings. It should be noted that when data on damping for
conventional structures are used in the modeling of seismically isolated structures, each mode of
vibration of the fixed-based structure corresponds to two modes of vibration of the
corresponding seismically isolated structure. For example, both the first and second modes, in
say the longitudinal direction, of a seismically isolated exhibit a shape for the superstructure part
that is basically identical to that of the first mode of the non-isolated structure. Accordingly,
when a value of damping ratio is selected for the first mode of the non-isolated structure, that
value should be used for the first six modes of the corresponding three-dimensional isolated
structure (first and second modes in each of the two principal directions and torsion). Note that
this is applicable when diaphragm constraints are used but it may be more complex otherwise
(see discussion on Purely Isolated Modes below).

Program SAP2000 constructs a global damping matrix which involves the degrees of freedom of
the isolation system. The calculation of the global damping matrix makes use of the specified
effective stiffness of each isolator element and of the mass and moment of inertia of the part of
the structure between the superstructure and the isolator supports (e.g., basemat above isolators).
This leads to the introduction of artificial viscous elements in the isolation system. We term this
phenomenon “damping leakage in the isolation system”. This may significantly affect the
calculation of isolator displacements by artificially damping its response as if viscous dampers
were present in the isolation system. For example, consider that the isolation system has bi-
linear hysteretic behavior with yield displacement of 1inch. The damping is specified as 5% of
critical in each mode of vibration and the effective stiffness of each isolator is specified as the
elastic stiffness (that is, the 5% damping ratio is anchored on the elastic stiffness). Say that upon
yielding of the isolator in dynamic analysis the displacement reaches 20inch amplitude. The
effective damping is then equal to 0.0520/1=0.224! The magnification of effective damping by
the square root of the ductility ratio is a well understood effect (for example see Chapter 18 of
ASCE-7, Minimum Design Loads for Buildings and Other Structures, 2010).

In Fast Nonlinear Analysis (FNA), the stiffness matrix remains always the same and is based on
the specified effective stiffness of the isolator elements. Nonlinear forces from the isolator
elements are treated as loads on the right hand side of the equations of motion. This allows the
user to have control on the amount of damping that is leaked into the isolation system by

14
specifying a small effective stiffness of the isolators. When a small effective stiffness is
specified, one can distinguish the following Ritz vectors modes:

 Purely Isolated: An example of purely isolated Ritz vector mode is shown in Figure 6. In
this mode, the superstructure responds nearly as a rigid block. The number of purely
isolated modes depends on the extent of discretization of the structure. When a
diaphragm constraint is specified for all floor levels and all masses are specified as point
masses, the number of such modes is reduced to one for two-dimensional analysis and to
three for three-dimensional analysis.

Figure 6 Example of Purely Isolated Ritz Vector Mode

 Mixed Modes: These modes are characterized by the mixed response of the superstructure
and the isolation system. When a diaphragm constraint is specified for all floor levels
and all masses are specified as point masses, the number of such modes is reduced to a
number equal to the number of floors for two-dimensional analysis and to three times the
number of floors for three-dimensional analysis. Figure 7 shows examples of mixed
modes.

Figure 7 Examples of Mixed Ritz Vector Modes

 Local or vertical Modes: These are high frequency modes that mostly involve local
response of members or the vertical response of the structure on the isolators. These

15
 modes typically create numerical difficulties and negatively affect the computational
effort. However, they are important in the calculation of quantities such as accelerations,
member internal forces and isolator uplift. Figure 8 shows an example of local mode.

Figure 8 Example of Local Ritz Vector Mode

The distinction between modes becomes complicated when vertical modes are of relatively low
frequency so that they appear in between the mixed modes or when combined horizontal and
vertical responses exist in a mode. In such cases, the modes should be classified as mixed rather
than local.

Damping may be specified in SAP2000 as follows:

 Constant Damping for All Modes

Specifying constant damping for all modes together with small effective stiffness is the simplest
approach for reducing, but not eliminating, damping leakage. Note that in this approach the
specified damping ratio of each mode is the same.

 Constant Damping with Override

This method is similar to the previous one but combined with zero damping override for the
purely isolated modes in order to minimize damping leakage in the isolation system. This
approach significantly affects the computational speed of the analysis so that, practically, may be
used when the number of isolators is small. However, the user may want to use override in order
to specify different (typically larger) damping ratio values for the higher modes.

 Rayleigh Damping

Rayleigh damping is based on the specification of damping ratio values at two selected
frequencies. A simple approach is to select the two frequencies such that they bound the
frequencies of the purely isolated and mixed modes as illustrated in Figure 9.

16
 Purely Isolated + Mixed Local or vertical
Damping Ratio Modes Modes

ζ
Selected
Frequencies

ωi ωj
Frequency
Figure 9 Selection of Frequencies for Rayleigh Damping Assignment

Two important considerations related to Rayleigh damping specifications are:

1) The two specified frequencies should be the first isolated mode frequency and the last
mixed mode frequency. Accordingly, frequencies  i and  j are significantly different
(the difference is even larger when low effective element stiffness is specified).
Therefore, modes that lie in between these two frequencies will have significantly lower
damping than the specified value, as illustrated in Figure 9. This typically leads in over-
estimation of the superstructure response (accelerations, drifts and isolator uplift).
2) Rayleigh damping provides the advantage of having higher damping for the higher
modes. This is both correct (e.g., see Satake et al, 2003) and also computationally
efficient as it reduces numerical difficulties and increases the speed of computation.

 Rayleigh Damping with Override

This method is similar to the Rayleigh Damping approach but allows for more control. It is the
most effective way of specifying damping in a seismically isolated structure when the number of
isolators is large and the speed of analysis is a concern.

The procedure requires identification of the purely isolated modes, mixed modes and local
modes. This is relatively easy when low values for the isolator element effective stiffness are
used. The Rayleigh coefficients are calculated using the lowest and highest frequency of the
mixed modes. Modes that have a frequency lower than the lowest mixed mode (these are the
purely isolated modes) should be assigned zero (or very low) damping using the damping
override option in SAP2000. Figure 10 illustrates the assigned damping ratio based on this
approach.

17
 Purely Local or vertical
Isolated Mixed Modes
Damping Ratio Modes
Modes
ζ

ωi ωj
Frequency
Figure 10 Damping Ratio in Rayleigh Damping with Override

An advantage of the Rayleigh Damping with Override approach is that frequencies  i and  j are
closely spaced so that the damping ratio in the modes with frequencies between the two selected
frequencies are not substantially less than the specified value of damping ratio. The analysis
speed with this approach is similar to the Rayleigh Damping case but damping leakage is better
controlled due to the damping override.

 Interpolated Damping

In this method, a different damping ratio may be specified for various frequency ranges.
Therefore the user can implement zero damping or nearly so for the frequency range of the
purely isolated modes, a specified damping ratio for the mixed modes, and a higher damping
ratio for local modes. For the local modes, either a high constant damping ratio can be specified
or a linearly varying damping ratio may be specified up to a maximum value of damping ratio.
This is illustrated in Figure 11.

In terms of minimizing damping leakage and accuracy of superstructure response, this damping
specification approach can produce results similar to the constant damping with override and
Rayleigh damping with override but with increased speed of analysis.

18
 Purely Local or vertical
Damping Ratio Isolated Modes
Modes
ζ

ωi ωj
Frequency
Figure 11 Damping Ratio in Interpolated Damping

3.4.2 Ritz Vector Modes

Fast nonlinear analysis requires the specification of the number of Ritz vectors modes for use in
the analysis. A sufficient number of Ritz vector modes are needed to accurately capture the
response of the analyzed structure. When the number of isolators is large, Ritz vector modes
associated with high frequencies may be included which often creates numerical difficulties. The
use of Rayleigh damping with override for the damping specification may be most useful in such
cases.

An important consideration in the selection of the number of Ritz vector modes has to do with
the development of gravity load on the isolators (which is very important for sliding isolators).
In FNA and prior to start of the seismic analysis, the gravity load is applied through analysis in
which the structure is subjected to vertical ground acceleration that slowly increases with time.
If the number of Ritz vector modes is not sufficiently large to fully develop the gravity load
effects, the analysis with sliding isolators is equivalent to one in which the structure is subjected
a vertical constant downward acceleration. This results in over-prediction of isolator
displacements. Figure 12 provides an example from analysis of a bridge structure (bridge is the
example in Constantinou et al, 2007). The figure shows the sum of axial forces on the bridge
supports (which equal to the weight of the bridge) as function of time and the number of Ritz
vector modes considered. The history of axial load evolves as the vertical acceleration is slowly
imposed to develop the gravity load prior to starting the dynamic response history analysis. The
weight is nearly equal to 6000kip. It requires at least 33 Ritz vector modes to develop correctly
the load. The typically utilized minimum number of Ritz vector modes equal to 3 times the
number of isolators plus rigid masses would have resulted in this case a requirement for use of
30 vectors, which should be sufficient for this example. Note that when the number of Ritz
vector modes is insufficient for the development of the isolator gravity load, member forces are
also underestimated which affects the adequacy assessment of the analyzed structure. It should
also be noted that the required number of Ritz vector modes may depend on the form of the ramp

19
function used to develop the gravity load. That is, the cut-off time of 10 second in Figure 12
may affect the required number of Ritz vector modes. Moreover, it is important to first
determine the number of Ritz vector modes needed to correctly develop the load and then
conduct analyses with that number and with larger number in order to observe the sensitivity of
the calculated response on the number of modes used. It is possible to correctly develop the load
with a small number of Ritz vector modes but require many more modes to obtain correct results
in dynamic analysis.

6000
Total Bridge Weight (kips)

4000
FNA-129 MODES

FNA-5 MODES

FNA-15 MODES
2000
FNA-30 MODES

FNA- 50 MODES

FNA-33 MODES
0
0 10 20

Time (sec)

Figure 12 History of Axial Load on Isolators as Function of Number of Ritz Vectors

3.4.3 Direct Integration

Program SAP2000 has the option of Direct Integration for dynamic analysis. In this approach,
the effect of any nonlinear elements is accounted for by the use of instantaneous stiffness and
damping matrices. This prevents the program user from controlling damping leakage in the
isolation system. Actually, artificial damping in the isolation system may substantially fluctuate
depending on the instantaneous stiffness of the nonlinear elements. We do not recommend use
of the direct integration method unless damping is specified as zero or nearly so.

20
 4. ISOLATOR P-Δ MOMENT ANALYSIS IN SAP2000

4.1 Introduction

Isolator P- effects can have important contributions to structural drift and acceleration. Figure
13 illustrates the transfer of forces at the sliding interfaces of a Double or a Triple FP bearing (a
similar situation exists in elastomeric bearings). The moment P1will cause rotation of the
column on top of the deformed isolator that, in turn, will result in increased first story drift.

Figure 13 Transfer of Force in Double or Triple FP Bearings

P- effects can be explicitly accounted for in dynamic analysis in SAP2000 by use of the direct
integration method of analysis. There is no explicit way of accounting for these effects in FNA
analysis.

4.2 Approximate P-Δ Analysis when using Fast Nonlinear Analysis

In FNA analysis, P-Δ effects cannot be explicitly analyzed. However, it is possible to

approximately account for these effects by performing two analyses as described below:

1) First a dynamic response history analysis is performed, ignoring P-Δ effects, so that
histories of the bearing axial loads and the displacements for each isolator are calculated.
The P- moment at the top of each isolator is calculated (for the Triple FP isolator or any
elastomeric isolator, the moment is equal to one half of the axial load times the
displacement-note that the component of moment resulting from the shear force times the

21
 height is accounted for directly by the program and need not be added to the axial force
component).

2) Dynamic response history analysis is performed with the calculated P- moment applied
as time history function at the top and bottom joints of each isolator and ensuring that the
direction of the moments is such that it causes a shear force in the same direction as the
isolator displacement. Note that the P- moment at the bottom joints is not needed if the
isolators are directly supported on the ground.

The procedure described above is one of a number of options for approximately accounting for
isolator P- effects. Engineers have already tried other approximate ways for accounting for
these effects such as first performing the response history analysis and then performing static
analysis to only calculate the P- moment effects. Such a procedure can provide information on
the effects on the structural drift but not on structural accelerations.

22
 5. VERIFICATION OF ACCURACY OF PARALLEL MODEL

5.1 Description of Analyzed Two-Dimensional Structure and Verification Study

A four story isolated structure is used herein to demonstrate aspects of modeling and analysis
procedures described above. It is two-dimensional slice of an actual building. It is a moment
frame with all connections being rigid. The structure is supported by six isolators. It is assumed
that each isolator carries a load of 900kip, for a total weight on the isolators equal to 5400kip. In
the analysis model, the mass is lumped at the joints as shown in Figure 14 with each joint
assigned the same mass.

Table 3 presents the periods of vibration of the first four modes of the superstructure when
assumed fixed to the ground.

Table 3 Periods of Vibration of Superstructure when Fixed at the Base

Mode Period (sec)

1 0.776
2 0.253
3 0.138
4 0.093

Figure 14 Schematic of Four Story Planar Structure (dimensions in inch)

In modeling the isolated structure, structural damping of 2% of critical for all modes is assumed
(constant damping model). The effective stiffness of each isolator is specified as equal to the
post-elastic stiffness. Analysis is performed using the TCU-065-E component of the 1999 Chi-
Chi, Taiwan earthquake.

23
The isolator properties used in the analysis are presented in Table 4.

Table 4 Properties of Triple FP Isolators

Reff1  Reff4 (inch) 82.5

Reff2  Reff3 (inch) 7.5

d1*  d 4* (inch) 17.8

d 2*  d 3* (inch) 0.94
1   4 0.108
 2  3 0.030
a1  a 2  a 3  a 4
2.54
(sec/in)
All friction values are for high speed conditions=fmax

In order to verify the accuracy of the parallel model, analysis is performed using the parallel
model described herein and the series model described in Fenz and Constantinou, 2008d. It is
presumed that the series model provides an exact description of behavior of the Triple FP
isolator. The series model as implemented in program SAP2000 is depicted in Figure 15.

Figure 15 Assembly of Friction Pendulum Link Elements, Gap Elements and Rigid Beam
Elements Used in the Series Model of Triple FP Bearing in SAP2000

Table 5 presents the parameters of the series model of Triple FP bearings in SAP2000.

24
 Table 5 Parameters of the Series Model of Triple FP Bearing in SAP2000
Friction Radii of Gap Elastic Rate
Element
Coefficient Curvature Displacement Stiffness Parameter
2W 1  a2  a3 
FP1  2  3 Reff 1  Reff 2  Reff 3 NA K
2Y 2 2
Reff 1  Reff 2 1W Reff 1
FP2 1 Reff 2  Reff 1  Reff 2 d1 K a1
Reff 1 2Y Reff 1  Reff 2

Reff 3  Reff 4  Reff 3 Reff 4  Reff 3 4W Reff 4

FP3 4 d4 K a4
Reff 4 2Y Reff 4  Reff 3

Yield displacement Y should be about equal to 0.01inch

Table 6 presents numerical values of the properties implemented in SAP2000 for the series and
parallel models of each bearing.

Results of analysis obtained with the two models are compared in Figures 16 to 20. Analysis
was performed using the FNA method and utilizing 53 Ritz vector modes for the parallel model
and 77 Ritz vector modes for the series model. These correspond to the maximum Ritz vector
modes calculated by SAP2000. Evidently, the two models produce identical results for all
practical purposes. Some small differences in the results are attributed to small differences in
structural damping modeling and in small differences in modeling velocity dependence of the
friction coefficient in the two models.

25
Table 6 Values of Properties of the Series and Parallel Models of Bearings in SAP2000

Series Model Parallel Model

FP1 FP2 FP3 FP1 FP2

Element Height (inch) 8 2 8 16 16

Shear Deformation Location

(in)-(distance from top joint 8 2 0 8 8
of FP element)
Element Mass (kip-s2/in) 0.001 0.001 0.001 0.0025 0.0025

Supported Weight (kip) 900 900 900 450 450

Vertical Stiffness1 (kip/in) 213530 213530 213530 35588.4 35588.4

Elastic Stiffness1 (kip/in) 1350 4860 4860 807.9 54.55

Yield Displacement1 (inch) 0.01 0.01 0.01 0.0155 -

Effective Stiffness1 (kip/in) 60 12 12 0 5.45

Friction Coefficient SLOW 0.015 0.054 0.054 0.030 0.0709

Friction Coefficient FAST 0.030 0.108 0.108 0.060 0.1418

Radius (inch) 15 75 75 0 82.5

Rate Parameter1 (in/sec) 1.27 2.794 2.794 1.27 1.27

Rotational/Torsional
0 Fixed Fixed 0 0
Stiffness (R1,R2,R3)

Rotational Moment of
1.0 0 0 0.5 0.5
Inertia (kip-in-sec2)
1. See Appendix A for details of calculation

26
 Figure 16 Comparison of Isolator Displacements

Figure 17 Comparison of Base Shear-Isolator Displacement Loops

27
Figure 18 Comparison of First Story Drift Ratio

Figure 19 Comparison of Roof Acceleration

28
 Figure 20 Comparison of Moments at Top of Isolators

5.2 Description of Analyzed Three-Dimensional Structure and Verification Study

A three-dimensional structure is generated using three frames identical to the two-dimensional

frame of Figure 14, placed at distance of 384inch, center to center. All connections are assumed
rigid. Figure 21 illustrates the three-dimensional model. The TCU-065-E and TCU-065-N
components of the 1999 Chi-Chi, Taiwan earthquake are used as seismic excitation along the
principal building directions. Structural damping is again specified as 2% of critical in each
mode (constant damping model) with the effective stiffness of each of the18 isolators specified
equal to the post-elastic stiffness. Each isolator is modeled using the data in Tables 4, 5 and 6.
Analysis was performed using the FNA method and utilizing 196 Ritz vector modes for the
parallel model and 200 Ritz vector modes for the series model. The results are shown in Figures
22 to 24. Evidently, the two models produce identical results for all practical purposes.

Figure 21 Schematic of Four Story Three-Dimensional Structure

29
Figure 22 Comparison of Isolator displacement Orbits and Histories

30
Figure 23 Comparison of Isolator Force-Displacement Loops in Orthogonal Directions

31
Figure 23 Comparison of First Story Drift Ratio

Figure 24 Comparison of Roof Accelerations

32
5.3 Effect of Ritz Vector Modes

The number of Ritz vector modes affects the accuracy of analysis through (a) its effect on the
development of the axial load on the isolators and the development of actions in the structural
members, and (b) the incomplete description of the structural response when an insufficient
number of modes are used.

The two-dimensional, 4-story structure example of Section 5.1 with constant 2% damping (and
effective isolator stiffness specified equal to the post-elastic isolator stiffness) was analyzed
using the parallel Triple FP model in fast nonlinear analysis with varying number of Ritz vector
modes. Selected peak response results are presented in Table 7.

Table 7 Effect of Number of Ritz Vector Modes on Peak Response Results for Two-
Dimensional Structure
No. of Ritz Isolator First Story Roof CPU
Base
Vector Displacement Drift Ratio Acceleration Time
Shear/Weight
Modes (inch) (%) (g) (sec)
2 0.234 22.18 0.18 0.24 14
5 0.235 22.22 0.33 0.34 17
10 0.236 22.38 0.34 0.43 84
20 0.235 22.31 0.32 1.55 158
30 0.235 22.29 0.31 0.43 188
40 0.235 22.29 0.31 0.42 229
50 0.235 22.29 0.31 0.42 257

Note that stable response prediction in this example is obtained when the number of modes
exceeds 30. Also, note that the prediction of acceleration response is very sensitive to the
number of modes. It is apparent that some experimentation is needed to determine a sufficient
number of modes for use in the analysis.

5.4 Effect of Gap Element

As described in Section 3.3.16, the addition of a gap element in parallel with the parallel model
for the Triple FP bearing does not correctly capture the behavior of the isolator. To demonstrate
the effects of such modeling, the two-dimensional structure of Section 5.1 was analyzed with the
TCU-065-E component of the 1999 Chi-Chi earthquake scaled up by a factor equal to 1.4 so that
the bearing enters the stiffening regime V. Each isolator was modeled using the parallel model
described in Section 5.1 and with a gap element added. The gap opening was calculated using
equation (8) and gap element stiffness after closing was selected in accordance with equation
(21) below. Note that the gap element stiffness is such that the sum of the post elastic stiffness
of the parallel model and the gap element stiffness equals to the stiffness of the FP bearing when
sliding occurs simultaneously on surfaces 2 and 3.

33
 W W
K gap   (21)
2 Reff 2 2 Reff 1

Moreover, and just for the purpose of demonstrating the gap element effects, the displacement
capacities of surfaces 2 and 3 have been considered unlimited so that the parallel model can be
used.

Figure 25 compares the isolation system force-displacement loops for the case of the series
model (exact) to the approximate parallel model with added gap element. The error introduced
by the incorrect use of the gap element in the parallel model is seen in the enlarged figure.
Nevertheless, the calculated displacement of the isolation system is practically unaffected.

Figure 25 Comparison of Results Obtained with the Exact Series Model and the
Approximate Parallel Model with Gap Element

34
The enlarged version of Figure 25 demonstrates the error introduced by the gap element in the
parallel model. The gap element engages and disengages at the same displacement as specified
in the gap element parameters. The stiffening behavior of the Triple FP bearing, however, starts
and ends at different displacements, which depend on the geometric and frictional properties of
the bearing. Moreover, the fluctuating loop seen in Figure 25 for the series model is a result of
numerical difficulty in the integration process.

35
 6. STUDY OF DAMPING SPECIFICATION EFFECTS

The two-dimensional model of Figure 14 is used to demonstrate the damping specification

effects discussed in Section 3.4.1.

6.1 Demonstration of Damping Leakage

Figure 26 compares the isolator displacement in the two-dimensional example of Section 5.1
when modeling damping using the constant damping model with 2% damping ratio and when
specifying zero damping. All other parameters in the analysis model are identical for the two
analyses. Note the effective stiffness of each isolator has been specified as equal to the post-
elastic stiffness.

The difference in the peak isolator displacement observed in Figure 26 is primarily attributed to
leakage of damping in the isolation system when the 2% constant damping specification is used.
While the difference in peak isolator displacement is small in this case, it is noticeably larger
when 5% damping is specified.

Figure 26 Comparison of Isolator Force-Displacement in Analysis with Different Damping

Specification

6.2 Effect of Effective Stiffness Value on Damping Leakage

The effect of the specified value of effective stiffness of each isolator used in analysis is
demonstrated by analyzing the two-dimensional example of Section 5.1 using the constant
damping model with various values of the effective isolator stiffness. Results are compared in
Figure 27 when the damping ratio is specified as 2% in each mode and in Figure 28 when the
damping ratio is specified as 5% in each mode. The effective stiffness is specified as either
equal to the post-elastic stiffness (case KEFF=KPE) or equal to the elastic stiffness (case
KEFF=KEL).

36
 Figure 27 Comparison of Isolator Force-Displacement Loops in Case of 2% Constant
Damping and with Different Values of Effective Stiffness (equal to the post-elastic stiffness
or equal to the elastic stiffness)

The results in Figure 27 demonstrate a significant effect when a large value of stiffness is used.
As discussed earlier, the observed substantial difference in peak displacement is the result of
magnification of effective damping that leaked into the isolation system and is magnified as
inelastic action occurs in the isolation system.

Figure 28 Comparison of Isolator Force-Displacement Loops in Case of Zero Damping

and 5% Damping and with Effective Stiffness Equal to the Elastic Stiffness

The results of Figure 28 show that an incorrect specification of damping (although seemingly
appropriate) may lead to totally erroneous results. Incorrect specification of the effective
stiffness results in so much damping leakage in the isolation system that the predicted peak
isolator displacement is reduced to less than half of the actual value.

37
6.3 Use of Constant Damping with Override

Figure 29 compares isolator force-displacement loops in the example of Section 5.1 when
damping is specified either as zero or as 2% constant damping with a 0% damping override for
the first mode and with effective stiffness equal to the post-elastic stiffness. The results of the
two models of damping are identical for all practical purposes. Note that constant damping with
override, as described in Section 3.4.1, is an appropriate damping specification for seismically
isolated structures. In this example, non-zero damping was specified but prevented from leaking
into the isolation system.

Figure 29 Comparison of Isolator Force-Displacement Loops in Case of Zero Damping and

2% Constant Damping with Override and with Effective Stiffness Equal to the Post-Elastic
Stiffness

6.4 Use of Rayleigh Damping

Table 8 presents information on the calculated modes of the 4-story isolated structure. The
identification was based on observation of the shapes as discussed in Section 3.4.1.

Table 8 Properties of Modes of Four Story Isolated Structure

Mode No. Mode Identification Period (sec)
1 Purely Isolated 4.167
2 0.462
3 0.216
Mixed Modes
4 0.131
5 0.091
6 0.0885
Local or vertical
7 0.0884

Damping was modeled either as constant in all modes with value equal to 2% or as Rayleigh
damping with values of 2% at frequencies of 0.24 and 11.1Hz (which correspond to the first and

38
sixth modes as shown in Table 8). The effective stiffness of the isolators was specified equal to
the post-elastic stiffness for both analyses. Results for the two cases are compared in Figure 30.
Evidently the two methods of specifying damping produce practically identical results.

Figure 30 Comparison of Isolator Force-Displacement Loops in Case of 2% Constant

Damping and 2% Rayleigh Damping and with Effective Stiffness Equal to the Post-Elastic
Stiffness

Nevertheless, the use of 2% Rayleigh damping as described above still had some damping
leakage in the isolation system. This is demonstrated in Figure 31 where the zero damping case
is compared to the 2% Rayleigh damping case.

Figure 31 Comparison of Isolator Force-Displacement Loops in Case of Zero Damping and

2% Rayleigh Damping and with Effective Stiffness Equal to the Post-Elastic Stiffness

Use of Rayleigh damping as described in Section 3.4.1 may significantly increase the speed of
the analysis. Figure 32 compares the CPU time required for the analysis of the two cases of 2%

39
constant damping and 2% Rayleigh damping described above as function of the number of Ritz
vector modes used in the analysis.

Figure 32 Comparison of CPU Time for Analysis Using the Constant Damping and
Rayleigh Damping Methods

Small CPU time is achieved in the Rayleigh damping specification method because the method
assigns large damping ratio values to very high frequency modes which effectively removes their
effects and reduces numerical difficulties associated with high frequency response. Despite the
differences in computational speed, the two methods produce practically identical results as
demonstrated in Figure 30.

6.5 Use of Rayleigh Damping with Override

Figure 33 compares results of analysis for the two-dimensional structure of Section 5.1 when
damping is specified as (a) 2% using the Rayleigh method for frequencies f i  2.16 Hz and
f j  11.1Hz (bounding frequencies of mixed modes) with 0% damping override for the first
mode (purely isolated mode) and (b) as zero. For both cases the effective stiffness was specified
equal to the post-elastic stiffness.

Note that the case of zero damping is used herein to provide the most accurate results for the
isolation system displacement as it does not have damping leakage in the isolation system.
However, the specification of zero damping affects the computation of accelerations and bearing
uplift displacements, which are then typically overestimated.

40
Figure 33 Comparison of Isolator Force-Displacement Loops in Case of Zero Damping and
2% Rayleigh Damping at f i  2.16 Hz , f j  11.1Hz with 0% Override for the First Mode

6.6 Use of Interpolated Damping

In this example, damping was specified in three ranges as follows: (a) 0% for frequencies less
than 2.13Hz, (b) 2% for frequencies in the range of 2.13 to 11.1 Hz and (c) linearly varying from
2% at frequency of 11.1Hz to a value of 40% at frequency of 1515Hz. The latter frequency
corresponds to the last mode included in the analysis. Results are compared in Figure 34. The
two methods of damping specification produce practically identical results.

6.7 Comparison of Results for all Analyzed Cases

Table 9 compares peak response quantities in the analysis of the two-dimensional structure of
Section 5.1 using the five methods of damping specification studied. Figure 35 presents a
comparison of damping ratio values as function of frequency for the five cases.

On the basis of these results we conclude that Rayleigh damping with override and interpolated
damping are most appropriate for use in the analysis of seismically isolated structures.

41
Figure 34 Comparison of Isolator Force-Displacement Loops in Case of Zero Damping and
Interpolated Damping

0.05
Rayleigh
Rayleigh with Override
Interpolated
0.04 Constant Damping
Constant Damping with Override
Damping Ratio

0.03

0.02

0.01

0.00
0 10 20 30 40 50 60 70 80 90 100 110 120

Frequency (Hz)

Figure 35 Damping Ratio as Function of Frequency for Studied Cases

42
Table 9 Comparison of Peak Response Results in Studied Cases of Damping Specification
2%
2%
2% Constant 2%
Rayleigh Interpolated Zero
Constant Damping Rayleigh
with damping Damping
Damping with Damping
Override
Override
Base Shear/
0.235 0.242 0.235 0.241 0.241 0.242
Weight
Isolator
Displacement 22.3 23.4 22.2 23.3 23.3 23.4
(inch)
First Story
Drift Ratio 0.31 0.32 0.32 0.32 0.32 0.34
(%)
Roof
Acceleration 0.42 0.42 0.48 0.43 0.42 0.51
(g)

43
 7. DIRECT INTEGRATION METHOD OF ANALYSIS

The Direct Integration (DI) method of analysis is presumed to provide the most accurate results
and accordingly often used to verify results of fast nonlinear analysis. Yet it is known, and will
be demonstrated herein too, that the direct integration and the fast nonlinear method of analysis
do not, in general, produce the same results. We will show herein that actually the fast nonlinear
analysis method allows the user for better control of the analysis procedures and most often
produces results of higher fidelity than direct integration.

In the direct integration method the damping matrix is continuously updated during the
integration process, whereas in the fast nonlinear analysis method the damping matrix remains
constant. It is apparent that the two analysis methods are comparable in terms of results only
when damping is zero. Figures 36 and 37 compare results of analysis of the two-dimensional
structure of Section 5.1 when damping is specified as zero and analysis is performed (a) with
direct integration using a time step equal to 1/10th of the excitation time step (integration time
step equal to 0.0005 sec ), and (b) with fast nonlinear analysis using 53 Ritz vector modes. The
two methods produce identical results.

Figure 36 Comparison of Isolator Force-displacement Loops Obtained with Direct

Integration and with Fast Nonlinear Analysis in Zero-Damped Structure

44
Figure 37 Comparison of First Story Drift Histories Obtained with Direct Integration and
with Fast Nonlinear Analysis in Zero-Damped Structure

Figure 38 compares the isolator force-displacement loops of the same structure with non-zero
damping when analyzed with direct integration (integration time step equal to 0.0005 sec ) and
fast nonlinear analysis (53 Ritz vector modes). Damping was specified as 2% using the Rayleigh
damping method with limit frequencies equal to 0.24 and 11.1Hz (first and sixth modes). Direct
integration in this case produces a lesser isolator displacement. This is likely due to more
leakage of damping in the isolation system in the direct integration method rather than errors in
the fast nonlinear analysis. On the basis of these examples, the authors of this report believe that

1) The fast nonlinear analysis method, when properly implemented, is sufficiently accurate.

2) Comparison of fast nonlinear analysis results to direct integration analysis results for
verification of the analysis model is not warranted nor recommended.

3) When verification of the analysis model is required, other programs should be used. For
example, program 3D-BASIS (Nagarajaiah et al, 1989; Tsopelas et al, 1994 and 2005;
Sarlis et al, 2009) may be used.

45
Figure 38 Comparison of Isolator Force-displacement Loops Obtained with Direct
Integration and with Fast Nonlinear Analysis in 2%-Damped Structure

46
 8. EXAMPLE OF ANALYSIS WITH ISOLATOR P-Δ EFFECTS IN SAP2000

8.1 Introduction

The two-dimensional structure of Section 5.1 is used for demonstration of analysis with isolator
P- effects. However, in order to magnify the P- effects for the demonstration, the sections of
the girders above the isolators (see Figure 14) were changed from W36X300 to W27X217 and
the earthquake was scaled up by a factor equal to 1.2.

Exact P- analysis requires use of the direct integration method. Approximate P- analysis is
performed using the fast nonlinear analysis. Due to the differences in modeling damping in the
two methods of analysis, we utilize zero damping in order to have comparable results.

8.2 Comparison of Response with Exact and Approximate Isolator P- Effects

Analysis was performed in SAP2000 with direct integration and with the P- effects first de-
activated and then activated. When P-Δ effects are activated, the calculated P-Δ moment was
equally distributed as moments at ends I and J of each element. A comparison of the calculated
first story drift ratio history in the two cases is presented in Figure 39. In this case, isolator P-
has some effect on drift.

Analysis was performed in two steps for approximate isolator P- effects as described in
Section 4. In this case the isolator displacements was first calculated in dynamic analysis
without the P- effects, and then used to calculate P- moment histories. The calculated moment
histories were added as time history functions, acting on top of each isolator. This requires the
introduction of separate load patterns, separate time history and a separate point moment for each
isolator. Dynamic analysis was then repeated with both the P- moment and the ground
acceleration histories as input.

Figure 40 compares the calculated first story drift ratio history, with P- effects accounted for
either by the exact method in direct integration or by the approximate 2-step method in fast
nonlinear analysis. The results are identical.

47
Figure 39 Isolator P- Effect on First Story Drift

48
Figure 40 Comparison of First Story Drift History Calculated with Exact and Approximate
Isolator P- Effects

49
 9. CONCLUSIONS

This document presented the following:

1) A summary description of the Triple Friction Pendulum isolator behavior.

2) Presentation of the parallel model for modeling Triple Friction Pendulum isolators.
3) Information on the implementation of the parallel model in SAP2000.
4) Information on the use of direct integration and fast nonlinear analysis in SAP2000.
5) Information on modeling of global damping in seismically isolated structures in
SAP2000.
6) Information on the significance of Ritz vector modes in capturing correctly the response
of seismically isolated structures in SAP2000.
7) Information on how isolator P- effects can be accounted for in fast nonlinear analysis in
SAP2000.

On the basis of the results presented in this report, the authors have the following conclusions:

1) Fast nonlinear analysis in SAP2000 (and by similarity in ETABS), when properly

implemented, produces results of acceptable accuracy in the analysis of seismically
isolated structures.
2) Direct integration should not be used as means of verification of the analysis model in
fast nonlinear analysis unless structural damping is specified to be zero.
3) Important considerations in the application of the fast nonlinear analysis are the number
of Ritz vector modes and the specification for damping.
4) A criterion for determining the minimum required number of Ritz vector modes is the
development of the correct vertical load on the isolators in the initial application of
vertical acceleration prior to staring the dynamic response history analysis. However, a
larger number of modes may be needed to correctly perform dynamic analysis.
5) Damping may be specified in a variety of ways of which the constant damping with
override and the Rayleigh damping with override result in control of the damping leakage
problem in the isolation system. The Rayleigh damping with override approach typically
results in least computational time.
6) Regardless of the damping specification method used, the effective stiffness of the
isolators needs to specified low for preventing or reducing damping leakage in the
isolation system. We recommend use of the post-elastic isolator stiffness value (or a
value less than that) for the effective stiffness together with low damping ratio values.

50
 10. REFERENCES

1. Computers and Structures Inc. (2007), “SAP2000: INTEGRATED FINITE ELEMENT

ANALYSIS AND DESIGN OF STRUCTURES”, Version 11.0.8, Berkeley, CA.

2. Constantinou, M.C., Whittaker, A.S., Fenz, D.M. and Apostolakis, G. (2007b), “Seismic
Isolation of Bridges”, University at Buffalo, Report to Caltrans for contract 65A0174, June.

3. Fenz, D.M. and Constantinou, M.C. (2008a),”MECHANICAL BEHAVIOR OF MULTI-

SPHERICAL SLIDING BEARINGS”, Report No. MCEER-08-0007, Multidisciplinary
Center for Earthquake Engineering Research, Buffalo, NY.

4. Fenz, D.M. and Constantinou, M.C. (2008b), “SPHERICAL SLIDING ISOLATION

BEARINGS WITH ADAPTIVE BEHAVIOR: THEORY,” Earthquake Engineering and
Structural Dynamics, Vol. 37, No. 2, 163-183.

5. Fenz, D.M. and Constantinou, M.C. (2008c), “SPHERICAL SLIDING ISOLATION

BEARINGS WITH ADAPTIVE BEHAVIOR: EXPERIMENTAL VERIFICATION”,
Earthquake Engineering and Structural Dynamics, Vol. 37, No. 2, 185-205.

6. Fenz, D.M. and Constantinou, M.C., (2008d), “MODELING TRIPLE FRICTION

PENDULUM BEARINGS FOR RESPONSE-HISTORY ANALYSIS”, Earthquake Spectra,
Vol. 24, No. 4, 1011-1028.

7. Fenz, D.M. and Constantinou, M.C. (2008e),”DEVELOPMENT, IMPLEMENTATION

AND VERIFICATION OF DYNAMIC ANALYSIS MODELS FOR MULTI-
SPHERICAL SLIDING BEARINGS”, Report No. MCEER-08-0018, Multidisciplinary
Center for Earthquake Engineering Research, Buffalo, NY.

8. Morgan, T. A. (2007), “THE USE OF INNOVATIVE BASE ISOLATION SYSTEMS TO

ACHIEVE COMPLEX SEISMIC PERFORMANCE OBJECTIVES”, Ph.D. Dissertation,
Department of Civil and Environmental Engineering, University of California, Berkeley.

9. Nagarajaiah, S., Reinhorn, A.M., and Constantinou, M.C. (1989), “NONLINEAR

DYNAMIC ANALYSIS OF THREE DIMENSIONAL BASE ISOLATED STRUCTURES
(3D-BASIS)” Report NCEER-89-0019, National Center for Earthquake Engineering
Research, State University of New York, Buffalo, NY.

10. Satake, N., Suda, K., Arakawa, T., Sasaki, A and Tamura, Y. (2003), “DAMPING
EVALUATION USING FULL-SCALE DATA OF BUILDINGS IN JAPAN”, ASCE, J.
Structural Engineering, 129 (4).

11. Sarlis, A.A., Tsopelas, P.C., Constantinou, M.C. and Reinhorn, A.M., “3D-BASIS-ME-MB:
Computer Program for Nonlinear Dynamic Analysis of Seismically Isolated Structures,
Element for Triple Pendulum Isolator and Verification Examples”(2009), supplement to
MCEER Report 05-009, document distributed to the engineering community together with
executable version of program and example files, University at Buffalo.
51
12. Tsopelas, P.C., Constantinou, M.C., and Reinhorn, A.M. (1994), “3D-BASIS-ME:
COMPUTER PROGRAM FOR NONLINEAR DYNAMIC ANALYSIS OF
SEISMICALLY ISOLATED SINGLE AND MULTIPLE STRUCTURES AND LIQUID
STORAGE TANKS”, Report NCEER-94-0010, National Center for Earthquake Engineering
Research, State University of New York, Buffalo, NY.

13. Tsopelas, P.C., Roussis, P.C., Constantinou, M.C., Buchanan, R. and Reinhorn, A.M. (2005),
“3D-BASIS-ME-MB: COMPUTER PROGRAM FOR NONLINEAR DYNAMIC
ANALYSIS OF SEISMICALLY ISOLATED STRUCTURES,” Report MCEER-05-0009,
Multidisciplinary Center for Earthquake Engineering Research, State University of New
York, Buffalo, NY.

52
 APPENDIX A
CALCULATION OF MODEL PARAMETERS

SERIES MODEL

Vertical Stiffness

The bearing vertical stiffness is approximately calculated as AE/h, where A is the area of the
center slider (herein a circle of 10inch diameter), E is a representative modulus (typically
assumed about half of that of steel to account for flexibilities in the bearing assembly, herein
14500ksi) and h is the height (herein 16inch). The vertical stiffness is thus equal to
x52x14500/16=71176.7kip/in.

The three elements of the series model each have vertical stiffness K so that the combined
stiffness equals 7176.7kip/in. That is, (1/K+1/K+1/K)-1=71176.7kip/in or K=213530kip/in.

Elastic Stiffness (see Table 5)

2W 0.03x900
Element FP1 Elastic Stiffness: K1    1350kip / in
2Y 2 x0.01

1W 0.108 x900

Element FP2 Elastic Stiffness: K2    4860kip / in
2Y 2 x0.01

4W 0.108 x900

Element FP3 Elastic Stiffness: K3    4860kip / in
2Y 2 x 0.01

Effective Stiffness (see Table 5)

W W 900
Element FP1 Effective Stiffness: K1     60kip / in
Reff 1 Reff 2  Reff 3 7.5  7.5

W W 900
Element FP2 Effective Stiffness: K2     12kip / in
Reff 2 Reff 1  Reff 2 82.5  7.5

W W 900
Element FP3 Effective Stiffness: K3     12kip / in
Reff 3 Reff 4  Reff 3 82.5  7.5

Rate Parameter (see Table 5)

1  a2  a3  (2.54  2.54)
Element FP1 Rate Parameter:   1.27 sec/ inch
2 2 4
Reff 1 82.5
Element FP2 Rate Parameter: a1  x 2.54  2.794 sec/ inch
Reff 1  Reff 2 82.5  7.5

53
 Reff 4 82.5
Element FP3 Rate Parameter: a1  x 2.54  2.794sec/ inch
Reff 4  Reff 3 82.5  7.5

PARALLEL MODEL

Vertical Stiffness

The vertical stiffness of 71176.7kip/in is equally divided to elements FP1 and FP2 of the parallel
model. That is, each element is assigned vertical stiffness of 35588.4kip/in.

Elastic Stiffness (see Table 2)

Element FP1 Elastic Stiffness:

 2W W 0.03  900 900

K1      807.9kip / in
2Y 1 2 Reff 2 2  0.0155 2  7.5

Note that the yield displacement Y 1 was assumed to be equal to 0.0155inch so that the elastic
stiffness of the parallel model is the same as the elastic stiffness of the series model (see
demonstration below). In general, the value of the yield displacement Y 1 is arbitrarily assumed
to be about or larger than 0.01inch.

W 900
Element FP2 Elastic Stiffness: K 2    60kip / in
2 Reff 2 2 x 7.5

The value to be specified in SAP2000 is not 60kip/in but a value equal to 60kip/in minus the
post-elastic stiffness of the element (SAP2000 adds the specified value of elastic stiffness to the
post-elastic stiffness and uses that as the elastic stiffness):

W W W  Reff 1  Reff 2 
K2     54.55kip / in
2 Reff 2 2 Reff 1 2 Reff 2 Reff 1

Note that the elastic stiffness of the parallel model of the bearing is equal to the elastic stiffness
of the series model as a result of the choice of the yield displacement Y 1 . Specifically, the two
1
 1 1 1 
models satisfy the condition K1  K 2      . In this equation, K1, K2 and K3 are the
 1
K K 2 K 3 

elastic stiffness of the three elements of the series model.

1 1
 1 1 1   1 1 1 
         867.9kip / in
 K1 K 2 K3   1350 4860 4860 

K1  K 2  807.9+60=867.9kip/in

54
Note that SAP2000 also performs the operation mentioned above (SAP2000 adds the specified
value of elastic stiffness to the post-elastic stiffness and uses that as the elastic stiffness) for the
elastic stiffness in each of the three elements of the series model. However, since the values of
elastic stiffness of each of the elements in the series model are very large, the effect of the
modification is insignificant and neglected. For example the specified value of elastic stiffness
for element FP3 of the series model should be specified in SAP2000 as

4W W 0.108 x 900 900

K3      4860  12  4848kip / in .
2Y Reff 4  Reff 3 2 x 0.01 82.5  7.5

The difference between the value of 4848 and 4860kip/in is insignificant, and neglected.

Effective Stiffness

The effective stiffness is specified to be the post-elastic stiffness of each element. For the FP1
W 900
element is zero (flat slider). For the FP2 element it is equal to   5.45kip / in .
2 Reff 1 2 x82.5
Note that these values are consistent with the values specified for the series model. For the
parallel model, the total effective stiffness is 0+5.45=5.45kip/in. For the series model, the total
1 1
 1 1 1   1 1 1
effective stiffness is           5.45kip / in .
 K1 K2 K3   60 12 12 
Rate Parameter

For each element, the parameter is specified as 2.54/2=1.27sec/in. This is based on the
assumption that the velocity is equally partitioned between sliding surfaces (true for surfaces 1
and 4).

55