Methods For Earthquake Analysis

Methods for Earthquake Analysis
Helge Drheim
Civil and Environmental Engineering

Submission date: June 2012
Supervisor:
Anders Rnnquist, KT
Co-supervisor:
Ragnar Sigbjrnsson, KT
Norwegian University of Science and Technology

Department of Structural Engineering

Faculty of Engineering Science and Technology
NTNU- Norwegian University of Science and Technology
ACCESSIBILITY
Open
MASTER THESIS 2012

SUBJECT AREA: Earthquake
Engineering
DATE: June 2012
NO. OF PAGES: 90
TITLE:
Methods for Earthquake Analysis
BY:
Helge Drheim
SUMMARY:
Based on the need for effective solutions of structures response to seismic loading, this thesis
includes the theoretical background, and explanation of several methods. There are many theories
in the community, but the ones chosen to be looked closer at here are Modal response spectrum
analysis for symmetric and asymmetric plan structures, Modal Pushover Analysis for symmetric
and asymmetric plan structures and Linear and Non-Linear Response History Analysis. The
theory has been adapted from published papers and books on the topic.
The report is divided into three main sections, numerical methods for solving of dynamic problems,
elastic analysis and inelastic analysis. The chapters on Elastic and Inelastic analysis are further
divided into sections on symmetric and asymmetric plan.
Elastic analyses by hand calculations and SAP2000 have been performed on a 2D-frame, a
SAP2000 Elastic Response spectrum analysis has been performed on a 3D-frame, and an
inelastic modal pushover analysis in SAP2000 has been performed on a 2D frame.
The target for this inquiry is to discover advantages and disadvantages in the different methods
and comparing them.
RESPONSIBLE TEACHER: Anders Rnnquist

SUPERVISOR(S): Anders Rnnquist, Ragnar Sigbjrnsson
CARRIED OUT AT: Department for Structural Engineering, NTNU
Norwegian University of Science and Technology

NTNU
Figur
Error! No
Masters Thesis for Stud techn. Helge Drheim, Spring 2012
text of
specified
Methods for Earhquake Analysis
style in
document..
Earthquake design was introduced to land based Norwegian structures throughError!
NS 3491-12.
This code was implemented as a transition into the new NS-EN 1998 EurocodeMain
8. This code
Document
demands that, with a few exceptions, all structures have to be controlled for earthquake load.
Only.
It is common practice to use elastic design even for seismic loading, and this may cause an
unnecessary stiff building, and larger material costs. Using more advanced analysis methods
allows for better use of the inelastic regions energy dissipating properties, giving more
effective structures.
This thesis will investigate different analysis methods that are relevant for Norwegian
structures, and go into the theory behind them.
Literature Study
Elastic analysis of symmetric plan structures
Elastic analysis of asymmetric plan structures
Modal Pushover Analysis, symmetric plan structures
Modal Pushover Analysis, asymmetric plan structures
Numerical time integration methods
Time History Analysis
Case study:
Numerical models in SAP2000 for comparison of methods
Compare with hand calculations
Supervisors:
Anders Rnnquist and Ragnar Sigbjrnsson NTNU
The thesis is to be submitted to the Department of Structural Engineering by 11th of

June 2012
Preface
This report is written as a Masters Thesis at the Department of Structural Engineering at
the Norwegian University of Science and Technology. The report has been written over the
course of 20 weeks during the spring semester of 2012 and should cover 30 credits.
In addition to the content of this report much time has been used on studying NS-EN-19981:2004, Eurocode 8, and work in CSi SAP2000, MATLAB, LATEX and Microsoft Excel. Many
hours were also spent reading papers and narrowing the scope of the thesis. Non-linear finite
element methods have also been studied to understand and use the non-linear possibilities in
CSi SAP2000 effectively. It has been a rewarding semester where I have learned a lot about
earthquake analysis, individual work and myself.
I would like to present a thanks to my supervisor Anders Rnnquist for his help, and to my
fellow students at room 2-66 for input and discussion. As a final remark I would also like to
express my gratitude to my lecturers, Dr. Tam Larkin, Associate Professor Nawawi Chouw
and Dr. Rolando Orense, at the University of Auckland who opened the door to earthquake
engineering for me.
Helge Drheim
Trondheim, Juni 2012
vii
Abstract
Based on the need for effective solutions of structures response to seismic loading, this thesis
includes the theoretical background, and explanation of several methods. There are many
theories in the community, but the ones chosen to be looked closer at here are Modal response
spectrum analysis for symmetric and asymmetric plan structures, Modal Pushover Analysis
for symmetric and asymmetric plan structures and Linear and Non-Linear Response History
Analysis. The theory has been adapted from published papers and books on the topic.
The report is divided into three main sections, numerical methods for solving of dynamic
problems, elastic analysis and inelastic analysis. The chapters on Elastic and Inelastic analysis are further divided into sections on symmetric and asymmetric plan.
Elastic analyses by hand calculations and SAP2000 have been performed on a 2D-frame, a
SAP2000 Elastic Response spectrum analysis has been performed on a 3D-frame, and an
inelastic modal pushover analysis in SAP2000 has been performed on a 2D frame.
The target for this inquiry is to discover advantages and disadvantages in the different
methods and comparing them.
ix
Sammendrag
Med utgangspunkt i behovet for en effektiv mte lse konstruksjoners respons til seismisk
last har denne rapporten utforsket flere metoder for lsning. I tidligfasen ble det sett p flere
analysemetoder, fr de ble begrenset ned til modal responsspektrumanalyse for symmetriske
og usymmetriske plan, modal pushover analyse for symmetriske og usymmetriske plan, samt
liner og ikke-liner tidshistorie. Underliggende teori og forklaring av de utvalgte metodene
er inkludert. Teorien er bearbeidet fra bker og publiserte artikler om emnet.
Rapporten er delt opp i tre hoveddeler, den frste beskriver numeriske metoder for lsning av
dynamiske problemer, den andre elastisk analyse og den tredje inelastisk analyse. Kaptilene
om elastisk og inelastisk analyse er videre delt mellom symmetrisk og usymmetriske plan.
Elastisk analyse gjort ved hndberegninger og i SAP2000 har blitt utforrt p en 2D-ramme,
en SAP2000 elastisk responsspektrumanalyse har blitt utfrt p en 3D-ramme og en inelastisk modal pushover analyse har blitt utfrt p en 2D-ramme.
Mlet med oppgaven er se p fordeler og ulemper med de forskjellige metodene, og dermed
gi et forslag p hvilke metoder som br brukes i gitte situasjoner.
xi
Contents
1 Introduction
2 Numerical Methods to Solve Dynamic Problems
2.1 Direct Integration . . . . . . . . . . . . . . . . . .
2.1.1 Newmark Method . . . . . . . . . . . . . .
2.1.2 Hilber-Hughes-Taylor . . . . . . . . . . . .
2.1.3 Non-linear Newmarks Method . . . . . . .
2.2 Numerical Damping . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3 Elastic Analysis
3.1 Symmetric Plan Buildings . . . . . . . . . . . . . . . . . . . . . . . .
3.1.1 Modal Analysis . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.2 Modal Response Contributions . . . . . . . . . . . . . . . . .
3.1.3 Response Spectrum Analysis Procedure . . . . . . . . . . . . .
3.1.4 Elastic Analysis of a 2D-frame . . . . . . . . . . . . . . . . . .
3.2 Asymmetric Plan Buildings . . . . . . . . . . . . . . . . . . . . . . .
3.2.1 One Story, Two Way Asymmetric System . . . . . . . . . . . .
3.2.2 Multi-story One Way Asymmetric System . . . . . . . . . . .
3.2.3 Response Spectrum Analysis of an Asymmetric Plan Structure
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4 Inelastic Analysis
4.1 Response History Analysis . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Uncoupled Modal Response History Analysis . . . . . . . . . . . . . . .
4.3 Properties of the nth Mode Inelastic SDOF-system and Pushover Curve
4.4 Modal Pushover Analysis . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4.1 Modal Pushover Analysis of a 2D Frame . . . . . . . . . . . . .
4.5 Modal Pushover Analysis for Asymmetric Plan Systems . . . . . . . . .
4.5.1 Elastic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.2 Inelastic Systems . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
3
4
9
9
12
.
.
.
.
.
.
.
.
.
15
15
15
17
21
22
27
28
30
34
.
.
.
.
.
.
.
.
39
42
42
44
45
46
54
54
54
5 Eurocode 8
5.1 Non-Linear Static Analysis (Pushover Analysis) . . . . . . . . . . . . . . . .
5.1.1 Determination of the Target Displacement for Non-linear Static Analysis (Pushover Analysis) . . . . . . . . . . . . . . . . . . . . . . . . .
59
59
6 Conclusion
63
7 Further Work
65
60
xiii
A Response Spectrum
67
B Elastic Analysis - Matlab
69
C Correlation Coefficient
73
xiv
List of Figures
2.1
2.2
2.3
2.4
2.5
2.6
2.7
Comparison of implicit and explicit methods [1] . . . . . . . . .

Constant Acceleration Method and Linear Acceleration Method
El Centro Ground Motion [4] . . . . . . . . . . . . . . . . . . .
Displacement history . . . . . . . . . . . . . . . . . . . . . . . .
Illustration of secant and tangent stiffness [3] . . . . . . . . . . .
Illustration of numerical errors [3] . . . . . . . . . . . . . . . . .
Illustration of iterations [3] . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4
5
8
8
10
11
11
Schematic description of modal response [3] . . . . . . . . . . . . . . . . . .

Correlation Factor [3] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Multi-story Frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Normalized mode shapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Response spectrum for El Centro ground motion for 5% damping . . . . . .
Comparison of time history of base shear for Linear Modal History and Linear
Direct Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.7 Illustration of asymmetric system . . . . . . . . . . . . . . . . . . . . . . . .
3.8 Imposed unit displacements to construct k [3] . . . . . . . . . . . . . . . . .
3.9 SAP2000 System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.10 Correlation coefficient for 5% damping . . . . . . . . . . . . . . . . . . . . .
3.11 Correlation coefficient for 20 % damping . . . . . . . . . . . . . . . . . . . .
3.12 Envelope of deformation for Linear Modal History Analysis in Y-direction . .
19
21
23
24
24
4.1
4.2
4.3
4.4
4.5
4.6
40
40
41
41
43
3.1
3.2
3.3
3.4
3.5
3.6
4.7
4.8
4.9
4.10
4.11
4.12
4.13
4.14
Global Pushover Curve of 20 Story Building LA, with and without P-delta [8]
Effects of P-delta effects [3] . . . . . . . . . . . . . . . . . . . . . . . . . . .
Influence on story drifts by modelling [3] . . . . . . . . . . . . . . . . . . . .
Statistical values of story drift demands for LA-Structure M2 Model [10] . .
Modal Decomposition of the roof displacement [9] . . . . . . . . . . . . . . .
Conceptual explanation of uncoupled modal RHA of inelastic MDOF systems
[9] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Hinge Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Mode shapes 1 through 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pushover-curves for mode 1,2 and 3 . . . . . . . . . . . . . . . . . . . . . . .
Plastic hinge formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Force Displacement Curves for mode 1,2 and 3 . . . . . . . . . . . . . . . . .
Design Spectrum from EC8 for ground acceleration 0.3188g and 0.4782g . . .
Floor Displacement comparison NL-RHA and MPA . . . . . . . . . . . . . .
Development of plastic hinges, NL-RHA to the left, and MPA-mode 1 to the
right . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
28
29
35
37
37
38
44
47
47
49
49
50
51
52
53
xv
4.15 Plan of asymmetric plan building [12] . . . . . . . . . . . . . . . . . . . . . .

4.16 Mode shapes of systems [12] . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.17 Floor Displacements and story drift demands for symmetric plan, U1, U2 and
U3 [12] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.18 Floor Displacement and story drift comparison of CQC and ABSSUM [12] .
5.1
xvi
Determination of the idealized elasto-perfectly plastic force displacement relationship [11] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
55
56
57
58
61
List of Tables
3.1
3.2
3.3
3.4
3.5
Comparison of modal periods . . . . . . . . . .

Comparison of base shear and roof displacement
Modal Static Responses [3] . . . . . . . . . . . .
Modal Periods . . . . . . . . . . . . . . . . . . .
Comparison of CQC and SRSS . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
26
26
33
36
36
4.1
4.2
4.3
4.4
Modal Participation Factors . . . . . . . .

Comparison of elastic and inelastic periods
Modal Pushover Analysis Results . . . . .
Base Shear comparison MPA and NL-RHA
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
48
50
51
52
.
.
.
.
.
.
.
.
.
.
.
.
xvii
Chapter 1
Introduction
When the new NS 1998 Eurocode 8 was introduced, it meant that most Norwegian structures
had to be designed for earthquake loading. Historically these kind of designs were all done in
the elastic range, but this is very ineffective when it comes to seismic loading. The reason for
this is the large energy dissipation that can be taken advantage of when structural members
enter the inelastic range.
The use of elastic analysis can cause the building to become unnecessary stiff, giving two
main drawbacks, number one: that more, and stronger, materials are needed, and number
two: the use of more and stronger materials increases weight, again causing more forces,
again demanding a stronger structure. The consequence for this is of course a higher price.
When allowing the structure to enter the inelastic region in the right areas, money can be
saved because of lower material costs and lower weight, again reducing forces. There are
obviously large challenges to this method of design. It demands a lot from the model used,
and if the wrong place yields, it could cause major damage.
This thesis will investigate several methods for analysis of seismic loading on structure, both
elastic and inelastic, discussing their advantages and disadvantages.
Chapter 2
Numerical Methods to Solve Dynamic
Problems
This chapter contains the theory used to numerically solve the response to an acceleration
time history. It will describe Newmarks method, which is used to create the displacement
time history from El Centro ground motion through a MATLAB-script. The Hilber-HughesTaylor will be presented, which is used in SAP2000 for time history analysis. Mass- and
stiffness-proportional damping, also used by SAP2000 will be described.
2.1
Direct Integration
This section will present the difference between implicit and explicit methods and general
theory behind the Newmark method and Hilber-Hughes-Taylor method. The theory in this
section is obtained from lecture notes from TKT4197 Non-linear Finite Element Analysis
by Kjell Magne Mathisen [1] and Civil 720 Earthquake Engineering by Tam Larkin [2] and
Dynamics of Structures by Chopra [3].
Direct integration methods are used to solve an initial value problem using step-by-step
integration in time. This means that the displacements, U0 and velocities u 0 , are assumed
to be known at a given time, t = 0. The time period where a result is wanted is then divided
into time increments and the integration method solves approximate solutions at each of
these time steps. In earthquake analysis, the acceleration is assumed to vary in a given way
during the time interval, and then integrated to find velocities and displacements at the next
time step.
The direct integration methods are divided into two main groups, implicit and explicit.
Explicit Methods: The displacement at time tn+1 , Un+1 , is obtained explicitly from
the equilibrium conditions at one or more of the preceding steps, without solving an
equation system. Hence: Unknown values are found based on the information already
known. If data from only one time-step is used, it is a single step method, while a
method using data from two steps back is called a two step method. Explicit methods
are conditionally stable, meaning that there is a critical time step tcr that causes the
process to become unstable if exceeded. This means that a large number of time-steps
3
Helge Drheim
Figure 2.1: Comparison of implicit and explicit methods [1]

are needed, but a lumped mass matrix will increases the speed of which each step is
executed.
Implicit methods: These methods finds the displacement un+1 implicitly from the
equilibrium conditions at time tn+1 , hence needing equation solving. Implicit methods
are unconditionally stable, meaning that they stay stable for any t. The accuracy
will nonetheless decrease with increasing t.
As a summary it can be said that an explicit method requires many steps, but at low cost
per step, while an implicit method requires fewer steps, but the cost per step is higher. This
means that for responses dominated by high frequency modes, with a need for small time
steps, like wave propagation problems from blasts or impacts calls for an explicit method,
while structural dynamic problems with a limited time range and no need for small time
steps calls for an implicit method. Figure 2.1 compares the computational effort for implicit
and explicit methods.
2.1.1
Newmark Method
In the Journal of the Engineering Mechanics Division in 1959 N.M. Newmark developed a
group of time stepping method based on the assumptions in equations (2.1a) and (2.1b)
u n+1 = u + t[ un+1 + (1 )
un ]
2
t
[2 un+1 + (1 2)
un ]
un+1 = un + tu n +
2
(2.1a)
(2.1b)
The two special cases, constant average and linear acceleration methods will be looked at
here. The methods are illustrated in figure 2.2 and the coefficients are derived as follows
4
Chapter 2. Numerical Methods to Solve Dynamic Problems
Figure 2.2: Constant Acceleration Method and Linear Acceleration Method

Constant Acceleration
1
un + un+1 )
u( ) = (
2
u n+1 = u n +
(2.2)
t+t
Z
u(t)dt
(2.3)
u(
) = u n + (
un + un+1 )
2
t
u n+1 = u n +
(
un + un+1 )
2
un+1 = un +
(2.4)
(2.5)
t+t
Z
u(t)dt
(2.6)
2
(
un + un+1 )
4
(t)2
= un + u n t +
(
un + un+1 )
4
u( ) = un + u n +
(2.7)
un+1
(2.8)
This shows that Constant Average Acceleration gives
= 12 ,
1
4
Helge Drheim
Linear Acceleration
u( ) = un +
u n+1 = u n +
(
un+1 un )
t
t+
Z
(2.9)
u( )d
(2.10)
u(
) = u n + un +
u n+1 = u n +
un+1 = un +
2
(
un+1 un )
2t
t
(
un+1 + un )
2
(2.11)
(2.12)
t+t
Z
u(t)dt
(2.13)
2
3
+
(
un+1 un )
2
6t
(t)2 (t)2
+
(
un+1 un )
= un + ut
+ un
2
6

1
1
= un + ut
+ (t)2 un+1 + un
6
3
u( ) = un + u n + un
(2.14)
un+1
(2.15)
un+1
Linear Acceleration gives = 21 ,
(2.16)
1
6
These parameters define the stability, variation of acceleration over a time step, amount of
algorithmic damping and the accuracy of the method.
The incremental quantities shown in equation (2.17) are not necessary for linear problems,
but provides a practical extension to non-linear systems.
un un+1 un
u n u n+1 u n
pn pn+1 pn
un
un+1 un
(2.17)
Combining equation (2.1a), equation (2.1b) and equation (2.17) gives
u n = (t)
un + (t)
un
un = (t)u n +
(t)2
un + (t)2
un
2
(2.18)
Solving the second for

un provides
un =
1
1
1
u
un
n
n
(t)2
t
2
(2.19)
When equation (2.19) is substituted into the first part of equation (2.18) the expression in
equation (2.20) is obtained
u n =
6

un u n + t 1
un
t
(2.20)
This is then substituted into the incremental equation of motion

m
un + cu n + kun = pn
(2.21)
This provides
1
k+
c+
m un =
t
(t)2
!
"
#

1
1
m + c u n +
m + t
1 c un
pn +
t
2
2
(2.22)
Then the notations k and pn are introduced as
c+
m
k = k +
t
(t)2
!
"
#

1
1
pn =pn +
m + c u n +
m + t
1 c un
t
2
2
(2.23)
The incremental displacement is then calculated

un =
pn
k
(2.24)
When un is found equations (2.20) and (2.19) can compute u n and

un respectively.
These values are then put into equation (2.17) to find un+1 , u n+1 and un+1 .
This method has been used to compute the response of a SDOF system with period T=2
seconds to the El Centro Ground Motion. The ground motion and system displacements are
shown in figures 2.3 and 2.4, respectively.
Helge Drheim
Figure 2.3: El Centro Ground Motion [4]
Figure 2.4: Displacement history
2.1.2
Hilber-Hughes-Taylor
Theory presented here is taken from Berkeley OpenSees Wiki [5] and the lecture notes
by Mathisen [1]. Hilber-Hughes-Taylor is an implicit method that can handle numerical
damping, without degrading the order of accuracy. This is convenient because introducing
Rayleigh proportional damping in the Newmark methods mostly damps just the middle
modes, and barely affects the higher and lower modes. To overcome these limitations, one
can introduce algorithmic damping in the Newmark method by assigning with a value
larger than 0.5. The problem with doing that, is a reduction of accuracy from O(t2 ) to
O(t).
In the Hilber-Hughes-Taylor-method the approximations of the Newmark method, shown
in equations (2.1a) and (2.1b), are used. The time discrete momentum equation is then
modified giving equation (2.25)
m
un+1 + (1 + H )cu n+1 H cu n + (1 + H )kun+1 H kdn = Rext
(2.25)
When loads vary linearly over a time step the load vector may be written as seen in equation
(2.26).
ext
Rext = (1 + H )Rn+1
H Rnext
(2.26)
Hilber-Hughes-Taylor is more effective than Newmark to suppress high frequency noise, and
decreasing the parameter H keeps th level of accuracy while increasing the amount of
numerical dissipation.
In the following use of Hilber-Hughes-Taylor in SAP2000, H will be set to zero, practically
making Hilbert-Hughes-Taylor equal to Newmarks average acceleration method. The reason
for this choice is the lack of high frequency noise in the models, keeping the accuracy on a
good level with the choice of H = 0.
2.1.3
Non-linear Newmarks Method
The theory presented in this section is adapted from Chopra [3]. When doing non-linear
time history direct integration, the non-linear Newmark method is a popular choice, and
this subsection will present the underlying theory. When doing non-linear analysis the
incremental equilibrium equation is as shown in equation (2.27)
m
un + cu n + (fS )n = pn
(2.27)
The incremental resisting force is
(fS )n = (kn )sec un
(2.28)
9
Helge Drheim
Figure 2.5: Illustration of secant and tangent stiffness [3]

(kn )sec is the secant stiffness, illustrated in figure 2.5. This can not be determined due to
the fact that un+1 is not known. Because of this, the tangent stiffness, (kn )T can be used
instead. Putting it into equation (2.27) gives equation (2.29)
m
un + cu n + (kn )T = pn
(2.29)
This is similar to equation (2.21). The only change is the replacement of k with (kn )T at
each time step.
Using constant time steps can cause large errors for two main reasons
Use of constant time steps delays the detection of transitions in the
force-deformation relations
Tangent Stiffness is used instead of secant stiffness
These errors are illustrated in figure 2.6, and can be reduced by using an iterative method.
The important equation solved at each time step is
kT u = p
(2.30)
c+
kT = (kn )T +
t
(t)2
(2.31)
Where
The non-linear stiffness is not constant. Equation (2.30) is shown in figure 2.7, and the
iterative procedure illustrated is described in more detail here: The first step is to determine
u(1) by using equation (2.30). The force f (1) which is connected to u(1) is less than
p and
the residual force becomes R(2) =
p f (1) . Additional displacement u(2) from R(2)
is found by
10
Figure 2.6: Illustration of numerical errors [3]
Figure 2.7: Illustration of iterations [3]
11
Helge Drheim
kT u(2) = R(2) =
p f (1)
(2.32)
This displacement is used to find a new magnitude of the residual force, and repeated until
it converges. This iterative process is listed below
1. Initialize data
(0)
(a) un+1 = un
(0)
fS = (fS )n
R(1) =
pn
kT = kn
2. Calculations for each iteration

(a) Solve kT u(j) = R(j) u(j)
(j)
(j1)
(b) un+1 = un+1 + u(j)

(j)
(j1)
(c) f (j) = fS fS
+ (kT kT )u(j)
(d) (j+1) = R(j) f (j)

3. Repeat for next iteration, replace j with j + 1
The iterations are repeated until the incremental displacement u(l) becomes small enough
in comparison to the current estimate of u.
u(l)
l
P
<
(2.33)
u(j)
j=1
When un is found, the process continues as shown in section 2.1.1, except that
un+1 =
2.2
pn+1 cu n+1 kun+1

m
(2.34)
Numerical Damping
The information in this section is obtained from CSI Berkeleys online wiki [6]. The damping
used by SAP2000 for direct integration time history analysis is called mass- and stiffnessproportional damping. Another word for this damping is Rayleigh Damping, and it is commonly used when running non-linear dynamic analysis. When formulated, the damping
matrix is said to be proportional to the stiffness and mass matrix as shown in equation
(2.35)
c = m + k
(2.35)
The mass proportional damping coefficient is and the stiffness damping coefficient is .
They are related through orthogonality and modal equation as follows in equation (2.36)
n =
12
n
1
+
2n
2
(2.36)
n is the critical damping ratio, and n is the natural frequency of the system. This shows
that the damping ratio is not constant with the natural frequency. The variables and
are given at two frequencies where the damping is known, or assumed. The critical damping
ratio will then be smaller between the two frequencies chosen, and larger outside.
In SAP2000 the coefficients and can be designated directly, or a given critical damping
ratio can be given at two different frequencies or periods. If the critical damping for two
periods or frequencies are set equal, the proportionality factors simplify as shown in equation
(2.37).
i = j = hence =
2
and = i j
i + j
(2.37)
13
Chapter 3
Elastic Analysis
This chapter will present the basic theory on which modal analysis of elastic buildings are
based on. A calculation using this theory will be performed, and compared with a similar
analysis done with the program SAP2000. The theory is adapted from Tam Larkin [2],
Nawawi Chouw [7] and Chopra [3]
3.1
3.1.1
Symmetric Plan Buildings

Modal Analysis
The equations of motion for a MDOF-system are

m
u + cu + ku = p(t)
(3.1)
For an undamped system, the modes are uncoupled. This is not always correct if the
system has damping, but for some types that are justifiable idealizations for many structures,
uncoupled damping can be used as well. The dynamic response of a system can then be
expressed as equation (3.2)
N
X
u(t) =
r qr (t) = q(t)
(3.2)
r=1
If equation (3.2) is substituted into equation (3.1), equation (3.3) is obtained

N
X
mr qr (t) +
r=1
N
X
cr qr (t) +
r=1
N
X
kr qr (t) = p(t)
(3.3)
r=1
To use the orthogonality properties of modes, all of terms are pre-multiplied by Tn

N
X
r=1
Tn mr qr (t) +
N
X
r=1
Tn cr qr (t) +
N
X
Tn kr qr (t) = tn p(t)
(3.4)
r=1
15
Helge Drheim
Orthogonality causes all terms of the summation, except r=n, to disappear, reducing equation (3.4) to
(Tn mn )
qn (t) + (Tn cn )q(t)
+ (Tn kn )qn (t) = Tn p(t)
(3.5)
The following notation is then adapted

Mn = Tn mn
Cn = Tn cn
Kn = Tn kn
Pn (t) = Tn p(t)
(3.6)
Using this and classical damping, equation (3.5) can be written like this
Mn qn (t) + Cn qn (t) + Kn qn (t) = Pn (t)
(3.7)
This equation exists for all N modes, and can be written on matrix form:
M
q + Cq + Kq = P(t)
(3.8)
This is used to solve N SDOF systems with the parameters Mn , Kn , Cn and Pn , and
combining them gives the total response. Dividing equation (3.7) by Mn renders
qn + 2n n qn + n2 qn =
Pn (t)
Mn
(3.9)
Here n is the damping ratio for the nth mode.

When each of the SDOF systems are solved, the contribution to nodal displacement un (t)
from mode n is given by this
un (t) = n qn (t)
(3.10)
And combining all the modal contributions to find total displacements
u(t) =
N
X
n=1
un (t) =
N
X
n qn (t)
(3.11)
n=1
This procedure has many names, for example classical modal analysis, classical mode
superposition method, or more precisely classical mode displacement superposition
method. Its short name is just modal analysis, and it is restricted only to linear system.
This because superposition can only be used in a linear systems. The damping also has to
be of the classical form to obtain uncoupled modal equations.
The procedure to solve the dynamic response of a MDOF system can be summarized like
Chopra [3] has done:
1. Define structural properties
(a) Determine the stiffness matrix k and mass matrix m
(b) Estimate modal damping ratios n
16
Chapter 3. Elastic Analysis
2. Calculate the natural frequencies n and modes n

3. Compute the response in each mode by the following steps
(a) Set up equation (3.7) and solve for qn (t)
(b) Compute nodal displacement un (t) from equation (3.10)
(c) Compute element forces associated with the nodal displacements
4. Combine the contributions of all the modes to determine the total response, using
equation (3.11)
3.1.2
Modal Response Contributions
Modal Expansion of Excitation Vector

The expansion of the excitation vector is useful because of two main reasons
sn p(t) only produces response in the nth mode, and no response in any other mode
The dynamic response in the nth mode is due entirely to the partial force vector sn p(t)
To expand the vector p(t), it is divided into a time variation part, p(t), and a spatial
distribution part s
p(t) = sp(t)
It can be helpful to expand s like this
s=
N
X
sr =
r=1
N
X
r mr
(3.12)
r=1
Pre-multiplying this equation with Tn and using the orthogonal properties of modes gives
n =
Tn s
Mn
(3.13)
This gives the contribution of the nth mode to s
sn = n mn
(3.14)
Equation (3.12) can also be viewed as an expansion of the distribution s of applied forces
in terms of inertia force distribution sn associated with natural modes. This interpretation
can be observed by considering the structure vibrating in its nth mode with accelerations
u
(t) = qn (t)n . The associated inertia forces are
(fI )n = m
un (t) = mn qn (t)
(3.15)
17
Helge Drheim
Their spatial distribution mn is the same as that of sn To use this for dynamic analysis
the generalized force Pn (t) = Mn p(t) for the nth mode is substituted into equation (3.9) to
obtain the modal equation
qn + 2n n qn + n2 qn = n p(t)
(3.16)
The solution of qn (t) can be written in terms of the response of a SDOF system with m = 1
and = n . For this special case we use D instead of q.
n + 2n n D n + 2 Dn = p(t)
D
n
(3.17)
Comparing equation (3.16) and (3.17) it seen that
qn (t) = n Dn (t)
(3.18)
This shows that qn (t) is easily obtainable when equation (3.17) has been solved for Dn (t),
utilizing regular SDOF solution methods. The contribution to the nodal displacement u(t)
from the nth mode is then
un (t) = n n Dn (t)
(3.19)
And the equivalent static forces become

fn (t) = sn [n2 Dn (t)]
(3.20)
The nth mode contribution to rn (t) to any response quantity r(t) is determined by static
analysis of the structure subjected to forces fn (t). The static value of r due to external forces
sn is noted as rnst giving the following expression
rn (t) = rnst [n2 Dn (t)]
(3.21)
When all the modes are combined, the total response is given as
r(t) =
N
X
n=1
rn (t) =
N
X
rnst [n2 Dn (t)]
(3.22)
n=1
This means that the contribution of rn (t) of the nth mode to the dynamic response is the
product of two analyses. First the static analysis of the structure excited by the external
force sn and secondly the dynamic analysis of the nth mode SDOF system excited by the
force p(t), as is shown schematically in figure 3.1
18
Figure 3.1: Schematic description of modal response [3]
19
Helge Drheim
Modal Contribution Factors

Equation (3.21) can be expressed as
rn (t) = rst rn [n2 Dn (t)]
(3.23)
where rn is the static value of r due to external forces s, an the nth modal contribution
factor
rn =
rnst
rst
(3.24)
These modal contribution factors rn have three important properties

They are dimensionless
They are independent of how the modes are normalized
The sum of the modal contribution factors over all modes is unity,
N
P
n=1
rn = 1
Modal Combination Rules

This part has been adapted from my project from the fall of 2011.
The peak modal response rno for mode n is found like this rno = rnst An , where rnst is the modal
static response, and An is the pseudo acceleration ordinate A(Tn , n ). It is not possible to
find the exact value of ro from the modal responses, because their maximum does not occur
at the same time. If all the maximum modal responses are added together, ro =
N
P
n=1
|rno |,
an upper bound value will be acquired, and this is usually very conservative. This happens
because there is a very low probability for all modal maximums to happen at the same time.
To get around this problem E. Rosenblueth in 1951 developed the square-root-of-sum-ofsquares (SRSS) rule for modal combination. It states that thes
peak response for each mode
is squared, summed and then taken the square root of: ro '
N
P
n=1
2 . This provides very

rno
good estimates, but has a limitation when the modes natural periods are close. Another
method is then used to avoid this problem. It
s is named the complete quadratic combination
(CQC), and it is formulated like this ro '
N P
N
P
i=1 n=1
in rio rno . in varies from 0 to 1 when
i = n. It can also be written as follows to show that the first summation is identical to the
SRSS rule. The CQC can turn out both larger and smaller than SRSS.
ro '
N
X
n=1
2
rno
+
N X
N
X
i=1 n=1
| {z }
i6=n
20
in rio rno
Figure 3.2: Correlation Factor [3]

There are also several formulations for the correlation coefficient in , and the most used one
now, according to Chopra is the one found by A. Der Kiureghian in 1981. It is, for equal
modal damping like this:
3
2
8 2 (1 + in )in
in =
2 2
2 2
(1 in
) + 4 2 in (1 + in
)
(3.25)
Where is the damping ratio and in = ni . Figure 3.2 shows how in varies with in , and
how it gets large if there is close modal periods. In a 5% damped structure in > 0, 1 only
1
when 1,35
in 1, 13. This shows how CQC turns in to SRSS if the modal periods are
wide enough apart.
Both SRSS and CQC are derived from random vibration theory, and are therefore most
accurate for loading with a wide frequency content and long phases of strong shaking. It
is less accurate for short impulsive loads or for many cycles of harmonic excitations. These
methods are best used with smooth response spectra based, and they tend to have errors
on the non-conservative side. This means that the response spectrum has to be on the
conservative side. Errors up to 25% have been observed for local response quantities.
3.1.3
Response Spectrum Analysis Procedure
There are two main ways to do earthquake analysis of linear systems. Namely Response
History Analysis (RHA) and Response Spectrum Analysis (RSA). RHA is a more labourintensive analysis because it provides a structural response r(t) as a function of time over
21
Helge Drheim
the duration of a shaking event. For a SDOF RSA will provide the same result, but that is
not the case for a MDOF-situation. But it does provide a good estimate.
For a N -story building with a plan symmetric about two axes, you can compute the peak
response as follows from Chopra [3]:
1. Define the structural properties
(a) Determine the mass matrix m and the lateral stiffness matrix k
(b) Estimate the modal damping ratios n
2. Determine the natural frequencies n and natural modes n of vibration
3. Compute the peak response in the nth mode
(a) Corresponding to natural period Tn and damping ratio n , read Dn and An from
the response or design spectrum
(b) Compute the floor displacement and story drifts with ujn = n jn Dn and jn =
n (jn j1,n )Dn
(c) Compute equivalent static forces fn from fjn = n mj jn An
(d) Compute the story forces, shear and overturning moment, and element forces,
bending moments and shear, by static analysis of the structure subjected to lateral
forces fn .
4. Determine an estimate for the peak value r of any response quantity by combining
the peak modal values rn according to SRSS or CQC depending on the spacing of the
modal frequencies.
When using this method it is important to be aware that it is wrong to compute the combined peak value of a response quantity from the combined peak values of other response
quantities. The correct procedure is to combine the peak modal values, and then calculating
the combined peak of this.
3.1.4
Elastic Analysis of a 2D-frame
In this section a double bay, five story shear frame will be analysed by the use of an elastic
modal response spectrum analysis. The structure consists of HE120A beams and columns
and is simplified to work as a shear structure. The model is shown in figure 3.3. Each bay
is 6 meters wide, and each story is 3 meters high.
The system has the following K-matrix:
3 393 600 1 696 800

0
0
1 696 800 3 393 600 1 696 800
0
0
[N/m]
0
1
696
800
3
393
600
1
696
800
0
K=
0
0
1 696 800 3 393 600 1 696 800
0
0
0
1 696 800 1 696 800
(3.26)
The mass of the frame is put as distributed loads on each floor equal to 5 kN/m, giving the
22
Figure 3.3: Multi-story Frame

following mass matrix
6116.2
0
0
0
0
0
6116.2
0
0
0
0
6116.2
0
0
M =
0
[kg]
0
0
6116.2
0
0
0
0
0
0
6116.2
(3.27)
Solving the eigenvalue problem gives following periods
1.3253
0.4540
0.2880 [s]
0.2242
0.1966
(3.28)
And the normalized mode shapes shown in figure 3.4

The modal participation factors are found by
n =
Tn m
Tn mn
(3.29)
Giving
1.2517
0.3621
=
0.1586
0.0632
0.0150
(3.30)
23
Helge Drheim
Figure 3.4: Normalized mode shapes
At this point the information needed to perform a modal analysis is found. In this example
the pseudo acceleration response spectrum from El Centro will be used to calculate the
different responses. The ground motion and spectrum is included in figure 2.3 and 3.5,
respectively.
Accelerations are found from the spectrum, and spectral displacements are found by
Dn =
San
n2
Figure 3.5: Response spectrum for El Centro ground motion for 5% damping
24
Giving
1.99
8.10 " #
m
Sa =
7.70 2
s
5.60
7.59
0.0866
0.0423
Sd =
0.0162 [m]
0.0071
0.0074
(3.31)
The modal story displacements are then found by

u = Sd
(3.32)
Giving
0.0316 0.0127
0.0034
0.0008
0.0002
0.0606 0.0167
0.0010 0.0006 0.0004
u=
0.0847 0.0091 0.0031 0.0002 0.0004 [m]
0.1020 0.0047 0.0018 0.0008 0.0003

0.1109 0.0153 0.0025 0.0005 0.0001
(3.33)
Combined with SRSS the floor displacements are estimated as
usrss
0.0342
0.0629
=
0.0852 [m]
0.1021
0.1120
(3.34)
Similar calculations are done for base shear providing the following result
53.5
21.6
Vb =
5.70 [kN]
1.29
0.36
Vsrss =
N q
X
2
Vbn
= 58.06kN
(3.35)
n=1
SAP2000 Calculations
The structure has also been analysed in SAP2000 for a more accurate solution of the problem.
This is important because the simplification of a shear building with no bending of the beams
is unrealistic.
The Model SAP2000 has calculated the response via several methods that will be compared with the hand calculation. These include response spectrum, linear modal history
and linear direct integration history. The model is fixed to the ground and consists of the
same HE120A beams and columns as said in the beginning of section 3.1.4. The mass is
implemented as a distributed 5 kN/m load on each floor. Plastic hinges are assigned, but
will not be affecting the results in this section due to the linearity of the analysis performed
here. All joints are restrained for movement in the y-direction, out of the plane . The bays
are still 6 meters wide, and stories 3 meters high. The load will be applied as a response
spectrum, linear modal history and linear direct integration history. The ground motion and
response spectrum applied are shown in figure 2.3 and 3.5, respectively.
25
Helge Drheim
Linear modal history used the method of modal superposition and solves it for each
time step, modal damping is kept fixed at 0.05. The time step will be 0.02 seconds,
same as the ground motion record, and to cover the 30 second long ground motion
1500 time steps have been chosen.
Linear direct integration solves the whole structure for each time step. The time step
will be 0.02 seconds, same as the ground motion record, and to cover the 30 second
long ground motion 1500 time steps have been chosen. It has proportional damping
specified by period that gives a mass proportional coefficient of 0.5984 and a stiffness
proportional coefficient of 7.599 104 when the damping ratio is 0.05. The time
integration is done by Hilber-Hughes-Taylor with standard coefficients of Gamma=0.5,
Beta=0.25 and Alpha=0.
Response spectrum finds the maximum response for each mode and combines it with
SRSS. 0.05 damping is also used here. (see section 3.1.2)
Table 3.1 compares the period of the modes of the SAP2000-model, and the hand calculated
one.
Table 3.1: Comparison of modal periods
Mode
Shear building
SAP2000 Model
1.3253 sec
2.4937 sec
0.4540 sec
0.7801 sec
0.2880 sec
0.4247 sec
0.2242 sec
0.2801 sec
0.1966 sec
0.2166 sec
The difference between the two result are because a shear structure assumes that the stories does not deform, making the system stiffer, hence reducing the period. The model in
SAP2000 does include the stiffness of the beams hence softening the system. The modes
does have the same shape in each instance.
In table 3.2 maximum base shear and roof displacement are compared
Table 3.2: Comparison of base shear and roof displacement
Case
Base shear
Roof Displacement
Shear Structure
58.06 kN (Eq (3.35))
0.1120 m (Eq (3.34))
Linear Modal History
51.35 kN
0.3684 m
Linear Direct Integration
52.63 kN
0.3774 m
Response Spectrum
50.96 kN
0.3534 m
The increased stiffness has a low impact on the base shear. This makes sense since the
mass accelerated by the quake is the same in all systems. This is not valid for the roof
26
Figure 3.6: Comparison of time history of base shear for Linear Modal History and Linear
Direct Integration
displacement. Here the difference in stiffness causes a large difference between the shear
story model and the SAP2000 model. This is to be expected. The difference between Linear
Modal History and Linear Direct integration is caused by the way damping is implemented
in the two cases. This causes a small deviation. The difference between response spectrum
and the two others comes from the resolution of the response spectrum.
Figure 3.6 shows the time history of the base shear for the Linear Modal History and the
Linear Direct Integration.
The time history verifies what was shown in table 3.2 occurs over the whole time history.
This section has shown that there are several shortcomings when you model a structure as
a shear building. Using linear analysis the computational power demanded by the different
analysis is quite low no matter which one is chosen, and hence the accurate direct integration
can be used just as well as linear modal history. The problem with using time histories is
that several relevant time histories has to be analysed, since no quake is the same. This is
where the response spectrum method has its advantage, since a design spectrum can easily
be implemented.
3.2
Asymmetric Plan Buildings
This section will present the theory behind linear analysis of asymmetric plan buildings.
First a one story system will be described, then a multi-story system. In the end a SAP2000
analysis of a steel moment resisting frame structure with asymmetric plan will be analysed
using response spectrum analysis with both CQC and SRSS to compare the effect of the
modal combination factor.
27
Helge Drheim
Figure 3.7: Illustration of asymmetric system
3.2.1
One Story, Two Way Asymmetric System
The system and theory from Chopra [3] will be presented here. The system is a one story
building consisting of three frames, A,B and C and a roof diaphragm that is rigid in its own
plane. Figure 3.7 illustrates the system. Frame B and C are oriented in the x-direction while
Frame A is oriented in the y-direction.
The force displacement relation can be expressed as follows
fSx
kxx kxy kx ux

fSy = kyx kyy ky uy
fS
kx ky k u
or fS = ku
(3.36)
The elements in the stiffness matrix k can be determined by the direct equilibrium method.
The lateral stiffness of each frame is defined as follows
fSA = ky uA
fSB = kxB uB
fSC = kxC uC
(3.37)
By imposing unit displacements for each DOF successively as shown in figure 3.8 the following
k is acquired
kxB + kxC
0
(d/2)(kxC kxB )
0
ky
eky
k=
2
2
(d/2)(kxC kxB ) eky e ky + (d /4)(kxB + kxC )
(3.38)
Direct stiffness method could also be used to acquire this matrix. The inertia forces on the
mass components are, since the DOFs are located at centre of mass O, as follows
fIx = m
utx
28
fIy = m
uty
fI = IO ut
(3.39)
Figure 3.8: Imposed unit displacements to construct k [3]
29
Helge Drheim
m is the diaphragm mass distributed uniformly over the plan, the moment of inertia about
the z-axis going through O is IO = m(b2 +d2 )/12 and utx ,
uty and ut are the components of the
total acceleration of the CM. The relationship between the inertia forces and accelerations
are related through the mass matrix as shown here
utx
fIx
m 0 0
t
y
fIy = 0 m 0 u
fI
0 0 IO ut
or fI = m
ut
(3.40)
Using the information presented above, the equation of motion can be assembled. The total
acceleration ut can be expanded to ut = u + ug . This gives
m 0 0
ux
kxB + kxC
0
(d/2)(kxC kxB )
ux

y +
0
ky
eky
0 m 0 u
uy
0 0 IO u
(d/2)(kxC kxB ) eky e2 ky + (d2 /4)(kxB + kxC ) u
m
ugx (t)
ugy (t)
=
m
IO ug (t)
(3.41)
Here it is easy to spot that the governing DOFs are coupled through the stiffness matrix.
This makes the system respond in DOFs different from the one where the load is applied.
For example an acceleration in the y-direction will also cause lateral displacement both in
the x and y direction, and also a torsional response.
3.2.2
Multi-story One Way Asymmetric System
In this section a system, from Chopra [3], with N floors that are symmetric about the x-axis
but not the y-axis and where the jth floor has three DOFs, namely ujx , ujy and uj , will be
presented. Since the x-motion is uncoupled to the two other DOFs, it can be solved alone.
This gives a coupled lateral motion described by 2N DOFs. The displacement vector u has
the size 2N 1.
"
u
u= y
u
(3.42)
where
u1y
u2y
uy = ..
.
uN y
u1
u2
u = ..
(3.43)
uN
To obtain the stiffness matrix of such a system the direct stiffness method can be used, going
through four steps listed in Chopra [3].
30
Step 1 Determine the lateral stiffness matrix for each frame. For the ith frame it is
determined as follows. Define the DOF for the ith frame: lateral displacement at floor levels,
ui , and vertical displacements and rotation for each node. Then the complete stiffness matrix
for frame i with reference to the frames DOF. Then using statical condensation to condense
all the rotational and vertical DOF to obtain the N N lateral stiffness matrix of the ith
frame, denoted kxi or kyi for a frame in the x or y-direction respectively.
Step 2 Determine the displacement transformation matrix relating the lateral DOF ui
defined in step 1 for the ith frame to get the global DOF u for the building. This N
2N matrix is denoted by axi or ayi . Thus:
ui = axi u or ui = ayi u
(3.44)
These transformation matrices are

h
axi = 0 yi I
or ayi I xi I
(3.45)
Where xi and yi define the location of the ith frame oriented in the y and x directions
respectively. I is an identity matrix of order N and 0 is a square matrix of N elements, all
equal to zero.
Step 3 Transform the lateral stiffness matrix for the ith frame to the building DOF u to
obtain
T
ki = axi
kxi axi
T
or ki = ayi
kyi ayi
(3.46)
Step 4 Add the stiffness matrices for all frames to obtain the stiffness matrix for the
building
k=
ki
(3.47)
Substituting equation (3.45) into equation (3.46) and the latter into equation (3.47) leads to
"
k
k
k = yy y
ky k
(3.48)
where
kyy =
X
i
kyi
T
=
ky = ky
X
i
xi kyi
k =
(x2i kyi + y 2 kxi )
(3.49)
31
Helge Drheim
The equation of undamped motion of the structure subjected to ground acceleration ugy (t)
can be developed as follows
"
m 0
0 IO
#"
"
u
y
k
k
+ yy y
u
ky k
#"
"
uy
m 0
=
u
0 IO
#" #
1
u (t)
0 gy
(3.50)
In equation (3.50) m is a diagonal matrix of order N , with mjj = mj the mass lumped
at the jth floor diaphragm, IO is a diagonal matrix of the N th order with Ijj = IOj , the
moment of inertia of the jth floor diaphragm about the vertical axis through the center of
mass. 1 and 0 are vectors of dimension N with elements equal to 1 and 0, respectively.
If all diaphragms are of the same size and weight, the equation simplifies to this
"
m 0
0 IO
#"
"
u
y
k
k
+ yy y
u
ky k
#"
"
uy
m 0
=
u
0 r2 m
#" #
1
u (t)
0 gy
(3.51)
To perform a modal analysis of the system, a modal expansion of the effective earthquake
forces have to be done, similar to the symmetric plan structure. The effective forces pef f (t)
are defined by the right side of equation (3.51)
"
m1
pef f (t) =
ugy (t) s
ugy (t)
0
(3.52)
The summation of of modal inertia force distributions sn are the expansion of the spatial
distribution s
"
2N
2N
X
X
m
m1
n 2 yn
sn =
=
r mn
0
n=1
n=1
"
(3.53)
Here yn and n contains the translations and rotations, respectively. of the N floors about
a vertical axis in the nth mode.
The modal participation factor is then found
n =
Lhn
Mn
(3.54)
where
N
X
m1
Tn ]
= Tyn m1 =
mj jyn
0
j=1
"
Lhn = [Tyn
(3.55)
and
"
Mn =
32
[Tyn
Tn ]
m 0
0 r2 m
#"
yn
n
(3.56)
or
Mn = Tyn myn + r2 Tn mn =
N
X
mj 2jyn + r2
j=1
N
X
mj 2jn
(3.57)
j=1
j denotes the floor number and mj the floor mass.

The lateral force, sjyn , and torque, sjn , at the jth floor level is expressed as follows
sjn = n r2 mj jn
sjyn = n mj jyn
(3.58)
The lateral and torsional displacements are

uyn (t) = n yn Dn (t) un (t) = n n Dn (t)
(3.59)
The equivalent static forces fn as a result of the displacements un (t) include lateral and
torsional forces.
"
"
fyn (t)
s
= yn An (t)
fn (t)
sn
(3.60)
The response to the nth mode is given as mentioned in equation (3.21) and repeated here
rn (t) = rnst An (t)
(3.61)
Selected modal static responses are presented in table 3.3

Table 3.3: Modal Static Responses [3]
Response, r
Vi
Mi
Ti
Modal Static Response, rnst

N
P
Vinst =
st
Min
=
st
Tin
=
sjyn
j=i
N
P
(hj hi )sjyn
j=i
N
P
sjn
j=i
N
P
sjyn = n Lhn = Mn
Vb
st
Vbn
=
Mb
st
Mbn
Tb
st
Tbn
ujy
ust
jyn
uj
2
ust
jn = (n /n )jn
j=1
N
P
j=1
N
P
j=1
hj sjyn = n Ln = hn Mn
sjn IOn
= (n /n2 )jyn
33
Helge Drheim
To get the forces of the elements in the structure the lateral displacements of each frame in
the building is connected to the floor displacement through equation (3.44). Substituting
T
T
equation (3.45) for axi and ayi , unT = [uyn
un
] and equation (3.59) for uyn and un provides
uin (t) = n (yi n )Dn (t)
uin (t) = n (yn + xi n )Dn (t)
(3.62)
The expressions in equation (3.62) are for frames in the x- and y-direction, respectively. The
total response can then be found by summing up the modal contributions.
r(t) =
2N
X
2N
X
rn (t) =
n=1
rnst An (t)
(3.63)
n=1
A summary of how to adapt this to response spectrum analysis was made by Chopra [3] and
quoted here
1. Define the structural properties
(a) Determine the mass and stiffness matrices from equation (3.51) and (3.49)
(b) Estimate the modal damping ratios n
2. Determine the natural frequencies n and natural modes of vibration, n
3. Compute the peak response in the nth mode by the following steps, repeated for all
modes, n = 1, 2, ..., 2N
(a) Corresponding to the natural period Tn and damping ratio n , read Dn and An
from the earthquake response spectrum, or the design spectrum.
(b) Compute the lateral displacements and rotations of the floors from equation (3.59)
(c) Compute the equivalent static forces: Lateral forces fyn and torques fn from
equation (3.60)
(d) Compute the story forces by three dimensional static analysis of the structure
subjected to external forces fyn and fn
4. Determine an estimate for the peak value r of any response quantity by combining the
peak modal values rn . The CQC-rule is explained in section 3.1.2
3.2.3
Response Spectrum Analysis of an Asymmetric Plan Structure
In this section a model with asymmetric plans has been modelled in SAP2000. The influence
of the choice of modal combination rule on the results from a response spectrum analysis
will be controlled.
The system investigated is shown in figure 3.9
The left frame consists of HE200A beams and columns, while the two to the right consists
of HE120A. The floor is C35 concrete. The total weight of the structure is 352 422 kg. It
is subjected to the El Centro ground motion shown in figure 2.3 and the response spectrum
34
Figure 3.9: SAP2000 System

shown in figure 3.5. The response spectrum results have been found using both SRSS and
CQC to investigate the difference between the two. The load has been applied in the xdirection, which is parallel to the short side of the system.
The modal period of the system is listed in table 3.4
In table 3.5 base shear and displacement of joint 35, the top left joint in figure 3.9, has been
listed.
35
Helge Drheim
Table 3.4: Modal Periods

Mode Period
1
3.450248 sec
2.97132 sec
1.509088 sec
1.16982 sec
1.01603 sec
0.728498 sec
0.640462 sec
0.5561 sec
0.514713 sec
10
0.49744 sec
11
0.480185 sec
12
0.435423 sec
Table 3.5: Comparison of CQC and SRSS
36
Modal Combination Rule
Base Shear
Joint Disp X Joint Disp Y
SRSS (5% Damping)
349.958 kN
0.1299 m
0.3667 m
CQC (5% Damping)
364.678 kN
0.1313 m
0.3671 m
SRSS (20% Damping)
229.416 kN
0.0851 m
0.2404 m
CQC (20% Damping)
290.581 kN
0.0866 m
0.2411 m
Figure 3.10: Correlation coefficient for 5% damping
Figure 3.11: Correlation coefficient for 20 % damping

The results show that there is a slight difference in the results. They are especially pronounced in the base shear. As shown in figure 3.2 the damping influences the correlation
coefficient. This is illustrated with the results found when the analysis is run with 20%
modal damping. In figure 3.10 and 3.11 the correlation coefficient is shown graphically for
the 5% and 20% modal damping case, respectively. The wider curves in figure 3.11 illustrates
how the different modes influences over a wider area of the spectrum, compared to the 5%
case. It also illustrates that the closely spaced last modes affect each other more, than the
more spread early ones, as listed in table 3.4. Equation (3.25) was used to calculate the
coefficients.
37
Helge Drheim
Figure 3.12: Envelope of deformation for Linear Modal History Analysis in Y-direction
The symmetry in the y-direction causes, as the theory claims, no rotation or translation in
the orthogonal direction, an envelope of the linear modal history analysis in the y-direction
is shown in figure 3.12
38
Chapter 4
Inelastic Analysis
This chapter will present the procedure to perform inelastic modal pushover analysis both
for symmetric and asymmetric plan systems. Factors that needs to be included in an analysis
will also be discussed and a modal pushover analysis will be performed on a 2D frame. The
theory is adapted from Chopra [3], Gupta and Krawinkler [8], [10], Chopra and Goul [9],
[12].
For an inelastic system the relationship between lateral forces s and displacements u are not
single-valued, but depend on the history of the displacement.
fs = fs (u, signu)
(4.1)
The loading and unloading curves differs from the initial loading branch, and experiments
on structural elements have provided force-deformation relations appropriate for different
elements and materials. With the generalization in equation (4.1) the equation of motion
becomes
m
u + cu + fs (u, u)
= m
ug (t)
(4.2)
This matrix equation represents N non-linear differential equations for N floor displacements
uj (t), j = 1, 2, ..., N . With a given mass- and damping matrix combined with the inelastic
force-deformation relationship, fs (u, u)
and ground acceleration ug , a numerical solution of
equation (4.2) is needed to acquire the displacement response of the structure.
Solving this equation is very computationally demanding. The stiffness matrix has to be
reformulated at each time step, according to the deformation and state of each structural
element. For large structures this process has to be repeated for a very large number of
elements. In addition non-linear geometry has to be considered, since structures subjected
to intense ground motions can undergo large displacement, causing P --effects.
To obtain usable result for inelastic response of a structure, a number of factors should
be included. In addition to the previously mentioned P --effects, there are idealization
assumptions and ground motion characteristics.
P --effects is the name for second-order effects of downwards gravity caused by lateral
deformation of the structure. As shown by Gupta and Krawinkler [8] in figure 4.1 the
inclusion of this effect greatly changes the pushover curve. The initial stiffness is almost
equal, showing that it is negligible in elastic analysis, but when it moves into the inelastic
range the differences starts to show. While the post-yield stiffness remains positive without
39
Helge Drheim
Figure 4.1: Global Pushover Curve of 20 Story Building LA, with and without P-delta [8]
Figure 4.2: Effects of P-delta effects [3]

the P --effects, the opposite happens when they are included, and after a short constant
strength plateau, the stiffness reduces quickly, getting to zero when the roof displacement
is 4% of the building height when used on a SAC-LA 20-story building. This can also be
shown with a response to a time history load, in figure 4.2
The way in which a structure is modelled and the idealizations used in the process can
greatly influence the inelastic response of a structure. To demonstrate this effect the work
of Gupta and Krawinkler will be used again [8]. In their work on the SAC-Los Angeles
20-story building they demonstrated the difference in response from three different planar
idealization of the frames considered. The first model, named M1 is a basic centreline model
where the panel zones strength, size and stiffness are ignored. The second model (M2)
explicitly includes the strength and stiffness properties of the panel zones, and the third
model (M2A) is a further enhanced version of model two, including interior gravity columns,
shear connections and floor slabs.
These modelling differences will cause a large difference in the response to an earthquake
load. Figure 4.3 shows the drift demands for each story with model M2 and M2A. Model
40
Chapter 4. Inelastic Analysis
Figure 4.3: Influence on story drifts by modelling [3]
M1 went to failure, so that one is not included in the figure.

There is also variation in the different ground motions. This causes a statistical variation.
Figure 4.4 shows the statistical drift demands for three ensembles of ground motion with a
return period of 2475 years (2/50), 475 years (10/50) and 72 years (50/50). It is observed
that with increasing intensity of the ground motion, a larger dispersion of the demand is
obtained.
Figure 4.4: Statistical values of story drift demands for LA-Structure M2 Model [10]
41
Helge Drheim
4.1
Response History Analysis
This section presents the equation of motion for a non-linear MDOF system by theory
adapted from Chopra [3] and Chopra and Goel [9]. When doing the response history analysis
equation (4.2) is expanded using the natural vibration modes of the corresponding linear
system getting
u(t) =
N
X
n qn (t)
(4.3)
n=1
Equation (4.3) is then substituted into equation (4.2) and pre-multiplied with Tn . Using
mass and classical damping orthogonality the following equation is obtained
qn + 2n n qn +
Fsn
= n ug (t),
Mn
n = 1, 2, ..., N
(4.4)
Here the only thing different from equation (3.16) is

= Tn fs (u, signu)
Fsn = Fsn (q, signq)
(4.5)
The resisting force now depends on all modal coordinates qn (t), implying the coupling of
modal coordinates because of yielding of the structure. Unlike equation (3.16) equation
(4.4) consists of N coupled equations.
4.2
Uncoupled Modal Response History Analysis
When neglecting the coupling of modes the uncoupled modal response history analysis can
be obtained. This simplification is the basis from where modal pushover analysis has been
derived. This theory is adapted from Chopra and Goel [9]
The spatial distribution of the effective earthquake forces, s, is expanded into modal contribution sn according to equation (3.12), repeated here for convenience
s=
N
X
n=1
sn =
N
X
n mn
n=1
n is still the modes of the corresponding linear system and the governing equation of motion
for this inelastic system is given by this
m
u + cu + fs (u, signu)
= sn ug (t)
(4.6)
The solution of equation (4.6) can not exactly be solved by the help of equation (3.19)
because the inelastic properties will cause other modes than the nth mode to influence the
response.
This is where one of the simplifications in this method comes in, and the coupling is assumed
to be negligible. The effect of this action has been illustrated numerically by Chopra and
Goel [9] on a 9 story SAC-building. They solved the problem numerically with non-linear
42
Figure 4.5: Modal Decomposition of the roof displacement [9]

RHA and decomposed the resulting roof displacement history into its "modal" components.
The model was subjected to 1.5 times El Centro ground motion and all but two beams
yielded.
The second and third mode started responding to peff,1 (t) the moment the structure yielded,
but their contribution to roof displacement were a mere 7% for mode two and 1% for mode
three. The response to peff,2 (t) yielded a contribution from mode one at 12% and 7% for
mode three. Figure 4.5 illustrates this.
This approximation changes Fsn from equation (4.5) so that it only depends on one modal
coordinate, qn .
Fsn = Fsn (qn , signqn ) = Tn fs (qn , signqn )
(4.7)
By using this, the solution of equation (4.4) can be expressed with equation (3.18) yielding
n + 2n n D n + Fsn =
ug (t)
D
Ln
(4.8)
Fsn = Fsn (Dn , signD n ) = Tn fs (Dn , signD n )
(4.9)
and
Equation (4.8) can be interpreted as the governing equation of the nth "mode" inelastic
SDOF-system, this system has the damping ratio n and natural frequency n of the corresponding linear MDOF-systems nth mode, and the Fsn /Ln Dn relation between resisting
forces Fsn /Ln and modal coordinate Dn defined by equation (4.9). The Fsn /Ln -curve is
found using the procedure showed in section 4.3
43
Helge Drheim
Figure 4.6: Conceptual explanation of uncoupled modal RHA of inelastic MDOF systems
[9]
The use of equation (4.8) is preferable compared to equation (4.4) because it is on the same
form as a standard equation for a SDOF-system, and the peak value can be estimated from
the inelastic response, or design, spectrum.
Solution of equation (4.8) written like this provides Dn (t), which when substituted into
equation (3.19) gives the floor displacements associated with the nth "mode" inelastic SDOFsystem. Other deformation quantities rn (t) is given by the following equations
rn (t) = rnst An (t)
An (t) = n2 Dn (t)
(4.10)
(4.11)
An (t) is now the pseudo-acceleration response of the nth "mode" inelastic SDOF-system.
The two analysis leading to this have been described schematically by Chopra and Goel and
is included as figure 4.6
Equation (4.10) and (4.11) now represent the response of inelastic MDOF system to peff,n (t),
also known as the nth mode contribution to peff (t)
The errors in UMRHA gets contribution from the three following assumptions and approximations.
The coupling of modal coordinates is ignored.
The superposition of responses to peff,n (t) is invalid for inelastic systems.
The Fsn /Ln Dn relationship is approximated via a bilinear curve.
4.3
Properties of the nth Mode Inelastic SDOF-system

and Pushover Curve
To obtain the Fsn /Ln Dn -curve a non-linear static analysis where the structure undergoes
the displacement u = Dn n with increasing Dn has to be performed, but most software
can not implement such a displacement controlled analysis. To work around this problem a
44
force-controlled non-linear static analysis with invariant distribution of lateral forces can be
used. The distribution fitting best to the purpose would be n . When this is done in most
software the result provided is the pushover curve, not the Fsn /Ln Dn -curve. To convert
this into the correct terms, the following equations have been derive
Fsn =
Vbn
,
n
Dn =
urn
n rn
(4.12)
Equation (4.12) enables the conversion to the Fsn /Ln Dn -curve from the pushover curve.
The yield points in the pushover curve are
Fsny
Vbny
=
,
Ln
Mn
Dny =
urny
n rn
(4.13)
Mn = Ln n is the effective modal mass. The two are related through

Fsny
= n2 Dny
Ln
(4.14)
This implies that the slope of the initial line in the Fsn /Ln Dn -curve is n2 . From the values
defined the elastic period Tn of the nth "mode" inelastic system is calculated like this
Ln Dny
Tn = 2
Fsny
4.4
!1/2
(4.15)
Modal Pushover Analysis
While the Uncoupled Modal Response History Analysis is exact for elastic systems its weaknesses makes it less precise for inelastic system, and opening the use of Modal Pushover
Analysis instead. The theory is adapted from Chopra [3] and Chopra and Goel [9]
The response rn of an inelastic system subjected to effective earthquake forces peff,n (t). Now
a non-linear static analysis of the structure subjected to lateral forces distributed over the
building height according to sn with the structure pushed to the roof displacement urno . The
value of roof displacement is given by equation (3.19) where Dn is determined by solving
equation (4.8) and finding the maximum, or determining it from the inelastic design or
response spectrum. Pushed to this roof displacement, the analysis provides estimated peak
values of any response, for example floor displacements, story drifts, joint rotations and
plastic hinge rotations. These are combined for the different modes using modal combination
rules.
The difference between UMRHA and MPA is contained in their assumptions. UMRHA is
based on the assumption that modal displacements are uncoupled. In the MPA analysis the
displacements are calculated through a non-linear static analysis using the force distribution
sn . This causes the displacements not to be proportional to the mode shape, causing MPA
to be more accurate for inelastic cases. MPA does contain another approximation not found
in UMRHA which is the combination of responses through a modal combination rule of the
modal responses, as done in the linear case.
45
Helge Drheim
The method can be summarized for a symmetric plan building on a step by step form as
done by Chopra [3].
1. Compute the natural frequencies, n , and modes, n for linearly elastic vibration of
the system
2. Develop the base shear-roof displacement pushover curve for mode n, Vbn urn by
non-linear static analysis of the building using the force distribution sn
3. Utilizing equation (4.13) to convert the vbn urn pushover curve to the force deformation relation for the nth mode inelastic SDOF-system, Fsn /Ln Dn
4. Idealize the F-D-relationship for the nth mode SDOF-system. For example as a bilinear
curve.
5. Compute the peak deformation Dn of the nth mode SDOF-system. This is done by
estimating the damping ratio and using equation (4.15) to find the inelastic vibration
period. Dn is then determined using non-linear RHA or response/design spectrum.
6. Calculate the roof displacement urn associated with the nth mode inelastic SDOF
system with equation (3.19)
7. From the pushover curve in step 2, extract values of desired responses rn+g de to the
combined effects of gravity and lateral loads at roof displacement equal to urn + urg
8. Repeat step 3 to 7 for as many modes as required for sufficient accuracy
9. Compute the dynamic response due to the nth mode: rn = rn+g rg , where rg is the
contribution to gravity loads alone
10. Determine the total dynamic response rd by combining the peak modal responses using
an appropriate modal contribution rule 3.1.2
11. Determine the total seismic demand by combining the initial static responses due to
gravity loads and dynamic response.
4.4.1
Modal Pushover Analysis of a 2D Frame
This section will contain a modal pushover analysis of the frame analysed in section 3.1.4.
The model is the same, and it will be subjected to a EC8 design spectrum with ag equivalent
to the peak acceleration of El Centro and 1.5xEl Centro. It will then be compared with
direct integration time history analysis to control the accuracy.
The plastic hinges are assigned to both ends of all column and beams, according to FEMA356
Table 5-6 for columns and beams respectively. The hinges have the properties shown in figure
4.7.
The mode shapes of the system is shown in figure 4.8.
46
Figure 4.7: Hinge Properties
Figure 4.8: Mode shapes 1 through 3
47
Helge Drheim
Table 4.1: Modal Participation Factors

Mode
Modal Participation Factor, n
Mode 1
163.664
Mode 2
59.943
Mode 3
37.491
Mode 4
25.896
Mode 5
14.468
The structure is then subjected to 3 different loads, each distributed spatially equally to the
mode shapes, giving the pushover curves and deformations displayed in figure 4.9 and 4.10
respectively. The colour codes for the plastic hinges are as follows:
Pink: Yield
Blue: Immediate occupancy
Teal: Life Safety
Green: Collapse prevention
Yellow: Loss of strength
Orange: Tearing
Red: Fracture
The colours pink, yellow, orange and red can also be found in figure 4.7.
The roof displacement of mode 2 is very low, this is due to the shape of the mode. The modal
coefficient on level 3 is almost equal to the modal coefficient on the roof, but in the opposite
direction. When a load with that distribution is applied, they will counteract each-other,
limiting the absolute roof displacement, while causing large story drifts. Only three modes
have been included due to the low participation factor of the higher modes. As can be seen
in table 4.1.
48
Figure 4.9: Pushover-curves for mode 1,2 and 3
Figure 4.10: Plastic hinge formation
49
Helge Drheim
Figure 4.11: Force Displacement Curves for mode 1,2 and 3
Table 4.2: Comparison of elastic and inelastic periods

Mode
Elastic Period
Inelastic Period
Mode 1
2.4937 sec
2.7252 sec
Mode 2
0.7801 sec
0.7754 sec
Mode 3
0.4248 sec
0.3970 sec
Equation (4.13) is then used to transform the curves to the Fsn /Ln Dn relationship, and
then equation (4.15) is used to find the inelastic period of the mode. The Fsn /Ln Dn -curves
are shown in figure 4.11 and the periods in table 4.2. The reason for the large difference
between mode 3 and the other two modes is the fact that in the transformation, the base
shear is divided by the modal participation factor squared. Since it is more than four times
larger for mode 1, than mode 3, combined with the larger base shear to start with, a large
difference is seen.
The periods acquired from this transformation are now used to find the SDOF displacement
from the Eurocode 8 design spectrum. The spectrum is created using ground type B, with
a ductility factor of 4. Ground acceleration is set to 0.3188 g and 0.4782 g, which is the
maximum of the El Centro and El Centro x1.5 ground motion used earlier. The design
spectrum is shown in figure 4.12. The different sections in the spectrum is defined as follows
in equation (4.16) through (4.19) [11].
50
Figure 4.12: Design Spectrum from EC8 for ground acceleration 0.3188g and 0.4782g
Table 4.3: Modal Pushover Analysis Results
Mode
Period
PSA
Dn
Ur
Vbn
Mode 1
2.7252
0.06376 g
0.4471 m
0.6001 m
52.39 kN
Mode 2
0.7754
0.09642 g
0.0576 m
0.0245 m
13.98 kN
Mode 3
0.3970
0.19662 g
0.0308 m
0.0063 m
10.27 kN
"
2
T 2.5 2
0 T 0.1 : Sd (T ) = ag S +
3 0.1 q
3
2.5
0.1 T 0.3 : Sd (T ) = ag S
q
0.3 T 1.5 : Sd (T )
1.5 T : Sd (T )
"
= ag S 2.5
q
0.2ag
=
0.2ag
ag S 2.5
q
(4.16)
(4.17)
#
0.3
T
"
!#
(4.18)
#
0.31.5
T2
(4.19)
Using the design spectrum from figure 4.12 and periods from table 4.2, combined with
the relationship between pseudo acceleration and displacement, Dn = An /(n2 ) and Ur =
n rn Dn . Table 4.3 is found. It can be noted that it is only mode 1 that goes into the
inelastic range in both cases.
Using the modal combination method of SRSS to find the total, base shear turns out as
55.18 kN. Roof displacement turns out as 0.601 meters. Figure 4.13 shows the comparison
51
Helge Drheim
Table 4.4: Base Shear comparison MPA and NL-RHA

Method
Base Shear
MPA (3 modes), El Centro x1
55.10 kN
NL-RHA, El Centro x1
45.20 kN
MPA (3 modes), El Centro x1,5
60.39 kN
NL-RHA, El Centro x1,5
60.18 kN
Figure 4.13: Floor Displacement comparison NL-RHA and MPA

of floor displacement between Non-Linear RHA and MPA. MPA overestimates the floor
displacement, just as it overestimates the base shear. as shown in table 4.4.
Figure 4.14 shows the development of plastic hinges for the modal pushover analysis for
mode 1 for the roof displacement found according to the 1.5x El Centro Design Spectrum
and NL-RHA envelope. The NL-RHA has some more yielding in the top floors, while the
MPA has some higher rotations in the lower part. This is due to the shape of mode 1. Mode
2 and 3 does not make any plastic hinges.
The results show that MPA combined with a design spectrum and NL-RHA gives similar
values for base shear. The difference is 22 % for 1x El Centro and almost nothing for
the 1.5x El Centro. There are larger differences in the floor displacement. This can be
due to the fact that a design spectrum usually has higher spectral values than the response
spectrum, causing higher acceleration values, and displacements. The errors are not to large,
and considering the time saving that one will get with larger systems, it causes acceptable
52
Figure 4.14: Development of plastic hinges, NL-RHA to the left, and MPA-mode 1 to the
right
results. This is similar to the results found by Chopra and Goel in [9].
53
Helge Drheim
4.5
Modal Pushover Analysis for Asymmetric Plan

Systems
This section will present the procedure developed by Chopra and Goel [12] to use modal
pushover analysis to estimate the seismic demands for asymmetric buildings. The method
will first be presented for elastic systems, and then for inelastic systems.
4.5.1
Elastic Systems
The lateral forces and torques are defined as follows

fxn = sxn An
fyn = syn An
fn = sn An
(4.20)
The factors sxn , syn and sn are given by the following relationship
mxn
sxn
sn = syn = n myn
sn
I0 n
(4.21)
An is given as n2 Dn where Dn is the peak deformation of the nth mode elastic SDOF system.
Static analysis of the system subjected to the forces listed in equation (4.20) will provide the
peak value rn of the nth mode contribution to the total response. The modal responses has
then to be combined with a modal combination rule, preferably the CQC for asymmetric
plan buildings. This is equal to standard RSA procedure.
4.5.2
Inelastic Systems
The procedure for asymmetric plans are quite similar to the symmetric method, with some
differences. The structure is subjected to lateral forces and torques distributed over the
structures height according to
mxn
sn = myn
I0 n
(4.22)
These forces are used to develop the base shear-roof displacement, Vbn urn . Chose the
pushover curve in the direction of the dominant direction of motion in the mode. Gravity
loads have to be applied before the pushover analysis. This curve is then idealized as a
bilinear curve and converted to the force displacement relation Fsn /Ln Dn as shown in
equation (4.13), where rn is the value in the direction of the selected pushover curve, and
Mn and n is in the direction of the ground motion.
The SDOF-system defined by the force deformation relation derived above, equivalent to
the nth mode, is analysed to find the peak deformation Dn . The period is calculated as
in equation (4.15). Dn can now be found by non-linear RHA or from a design/response
54
Figure 4.15: Plan of asymmetric plan building [12]

spectrum. The peak roof displacement is then found from urn = n rn Dn . This is repeated
for as many mode as decided necessary, and then combined with CQC as described in section
3.1.2.
Chopra and Goel [12] also investigated the accuracy of this method on a 9-story SAC structure. They moved the centre of mass to make it asymmetric while maintaining its stiffness
properties, a cross section is shown in figure 4.15.
3 different systems were chosen, to investigate different effects. The first one kept the Ioj /mratio equal to the symmetric plan building called U1. This caused a distinct difference
between the period of the main torsional mode and lateral mode, as seen in figure 4.16.
This is representative for perimeter moment resisting frames. The second model, U2, chose
a Ioj /m that achieved very close modal periods, that can easily be seen in 4.16. The last
system, U3, has an even larger Ioj /m-ratio making torsional rotations dominate the first
mode, and lateral displacements in the 2nd mode, opposite of system one.
The system were analysed using the first 3 modal pairs. It is apparent that one modal pair
is not enough, especially when it comes to story drift. Figure 4.17 shows how more modes
improves the accuracy of the method for a symmetric plan, and the three system analysed
in this section, compared to rigorous non linear response history analysis.
55
Helge Drheim
Figure 4.16: Mode shapes of systems [12]
56
Figure 4.17: Floor Displacements and story drift demands for symmetric plan, U1, U2 and
U3 [12]
57
Helge Drheim
Figure 4.18: Floor Displacement and story drift comparison of CQC and ABSSUM [12]
An important observation of these results are the reduced accuracy of the 2nd system.
Chopra and Goel assumes this is due to the fact that the maximums actually occurs at
almost the same time, making ABSSUM a better modal combination method. This can be
seen in figure 4.18
58
Chapter 5
Eurocode 8
This section will present demands that Eurocode 8 sets for Non-Linear Static analysis.
Section 4.3.3.4.1 states that the general rules demands that a bilinear force displacement
relationship has to be used on every structural part. The elastic part of this bilinear relationship should be used as the secant stiffness to the yield point. The code allows for the post
yield stiffness to be zero, or a trilinear relationship can also be used. The material parameters can be used as middle values calculated from characteristic values given in NS-EN 1992
to NS-EN 1996. The gravity load are to be included, but moments caused by P -effects
can be ignored unless it has a large influence on the behaviour of the structure. The seismic
loads will be added in both positive and negative direction, and the largest results should
be used.
5.1
Non-Linear Static Analysis (Pushover Analysis)
Section 4.3.3.4.2 contains the rules for this kind of analysis. The code states that if a
structure fulfils the criteria for regularity in plan from section 4.2.3.2 or 4.3.3.1(8) a)-d),
it can be analysed through two planar models, one for each horizontal direction. If these
criteria are not fulfilled, a spatial model has to be adapted, where two independent analysis
can be performed, one in each direction.
Section 4.3.3.4.2.2 demands that at least two vertical distributions of the lateral load has to
be applied, namely a uniform pattern that is based on lateral forces which is proportional
to the mass, independent of elevation, and a "modal" pattern which is proportional to the
lateral force distribution found from elastic analysis. The loads are to be applied in the mass
centre of the model, and accidental eccentricity, as presented in 4.3.2(1)P, should be taken
into account.
When constructing the capacity curve, a displacement equal to 150% of the displacement
defined in 4.3.3.4.2.6, and the overstrength factor is chosen as the lowest of the two lateral
load cases. The determination of the displacement value follows in the subsection.
59
Helge Drheim
5.1.1
Determination of the Target Displacement for Non-linear

Static Analysis (Pushover Analysis)
Appendix B in Eurocode 8 is an informative appendix that describes the codes requirements

to find the target displacement. The notation in Eurocode 8 is different from the one used
in the rest of the thesis, but in this section, the notation from EC8 will be used.
Section B.1 General states that the relationship between normalised lateral forces Fi and
normalised displacements i is
Fi = mi i
(5.1)
Where mi is the mass of the ith story. The displacements are normalised so n = 1, where
n is the control node, usually the roof, hence Fn = mn .
Transformation to an Equivalent SDOF-system

The mass of a SDOF system, m is determined as shown in equation (5.2)
m =
mi i =
Fi
(5.2)
and the transformational factor is given as
= P
m
mi 2i
(5.3)
This equation is equivalent to the modal participation factor, given in equation (3.29).
The displacement d and force F for the equivalent SDOF system is calculated as shown in
equation (5.4)
dn
F
b
F =
d =
(5.4)
Fb and dn is base shear and displacement of control node respectively.
Determination of the Idealized Elasto-perfectly Plastic Force-displacement Relationship

The yield force Fy that represents the ultimate strength of the idealised system is equal to
the base shear force when the plastic mechanism is formed. The idealised systems stiffness is
determined so that the area under the deformation curve from both the actual and idealized
60
Chapter 5. Eurocode 8
Figure 5.1: Determination of the idealized elasto-perfectly plastic force displacement relationship [11]
force are equal. Using this assumption, the yield displacement of the SDOF system dy is
given by equation (5.5)
dy
=2
dm
E
m
Fy
(5.5)
Where Em
is the actual deformation energy needed to create the plastic mechanism. This
is illustrated in figure 5.1
Determination of the Period of the Idealized Equivalent SDOF-system

The period T is defined in equation (5.6)
T =
v
u
u m dy
2 t
Fy
(5.6)
This is equivalent to equation (4.15)

Determination of the Target Displacement for the Equivalent SDOF-system
The target displacement for a structure with unlimited elastic behaviour and period T is
given by
det = Se (T )
2
T
(5.7)
Where Se (T ) is the spectral value of the elastic acceleration spectrum at period T .

The calculation of dt depends if the period is short or long, where periods higher than TC
are deemed as long.
1. If T < TC
If Fy /m Se (T ) the response is elastic, and dt = det
61
Helge Drheim
If
Fy /m
Se (T ) the response is non-linear and
dt
det
qu
1 + (qu 1) TTC
det ,
(T )m
where qu = Se F , the relationship between the acceleration in a infinitely

y
elastic structure and a structure with limited strength.
2. If T > TC , dt = det
Determination of the Target Displacement for the MDOF-system
The target displacement is then given by
dt = dt
(5.8)
Which is equivalent to (3.19), with n = 1.

This target displacement is then used to acquire the structural responses according to the
capacity curve. The responses from each mode are then added together as described earlier to
find the total response of the structure. This method is an expansion of the modal pushover
analysis presented in section 4.4.
62
Chapter 6
Conclusion
It is clear from the information presented in this report, and in the community in general, that
if a very accurate and precise result is wanted, there is no way around the non-linear response
history analysis. This method does nonetheless bring with it some large complications,
especially in the demand for computational power. The need to find a new stiffness, k, for
each element in the system for each time step causes even a simple model as the one in
section 3.2.3 to use over 45 minutes on one run, depending on how many members yielding.
Considering a larger structure, and a need for the analysis to be run with 7 time series
according to 4.3.3.4.3 in EC8, the computational resources needed are huge. Finding relevant
time histories for the location chosen, can also be a challenge.
This is the reason alternative methods are needed. Modal pushover analysis saves a lot of
computational time. Instead of solving each member at each time step of a acceleration time
history, the structure is statically pushed to a given roof displacement with forces spatially
distributed as the mode shapes. It also allows for the use of a design spectrum to provide the
roof displacements, again saving time by not having to run 7 time histories. It is obviously
not as accurate as NL-RHA, but for most uses accurate enough. It also discovers where
plastic hinges are found, so it will be easy to control if the structure yields in the wrong
places.
Both non-linear methods demands a lot of the model used. As described, a wrong model,
simplified in the wrong places, can cause very different results compared to the real building.
This is especially important in seismic loading, because when a section is designed to yield,
and it turns out to be stronger than designed, it may cause the wrong part to yield, putting
the whole structure in jeopardy.
In smaller structures it may not be worth the effort needed to construct a proper detailed
model to investigate the effects of seismic loading, and elastic response spectrum analysis
can be used with little effort.
63
Chapter 7
Further Work
Ideas for further work could be:
Create hinge models that are more adapted to Norwegian situations than the FEMA
356
Run the modal pushover analysis with a inelastic response spectrum instead of the
Design Spectrum
Further investigate the effects of modelling assumptions
Investigate damping properties of different structures
Investigate other method described in the community
65
Bibliography
[1] Mathisen, Kjell Magne Lecture Notes, TKT4197 Non-linear Finite Element Analysis,
Fall 2011.
[2] Larkin, Tam, Lecture notes, CIVIL720 Earthquake Engineering, University of Auckland,
New Zealand
[3] Chopra, Anil K. Dynamics Of Structures. Pearson Prentice Hall, 2007, Print
[4] El Centro Acceleration History. http://www.vibrationdata.com/elcentro.dat
[5] OpenSees Berkeley,
http://opensees.berkeley.edu/wiki/index.php/Hilber-Hughes-Taylor_Method
[6] CSIBerkeley, https://wiki.csiberkeley.com/display/kb/Damping+coefficients
[7] Chouw, Nawawi , Lecture Notes, Civil 314 Structural Dynamics Fall 2010, University
of Auckland, New Zealand
[8] Gupta, A and Krawinkler, H. "Dynamic P-delta effects for flexible inelastic steel structures, Journal of Structural Engineering, ASCE, 126: 145-154, 2000
[9] Chopra, Anil K. and Goel, Rakesh K. "A modal Pushover analysis for estimating seismic
demands for buildings", Earthquake Engineering And Structural Dynamics, 31:561-582,
2002
[10] Gupta, A. and Krawinkler, H. Behaviour of Ductile SMRFs at Various Seismic Hazard
Levels", Journal of Structural Engineering, ASCE, 126: 98-107, 2000
[11] Eurocode 8: Design of structures for earthquake resistance, Part 1: General rules,
seismic actions and rules for buildings, Standard Norge 2005
[12] Chopra, Anil K. and Goel, Rakesh K. "A modal Pushover analysis procedure to estimate seismic demands for unsymmetric-plan buildings, Earthquake Engineering And
Structural Dynamics, 33:903-927, 2004
66
Appendix A
Response Spectrum
67
Ag=importdata('elcentro.txt');
for j=1:725
T(1,j)=0.02*j;
m=10;
k(1,j)=((2*pi)^2*m)/(T(1,j)^2);
c(1,j)=0.05*2*sqrt(m*k(1,j));
a0=0;
deltaT=0.02;
gamma=0.5;
beta=1/6;
khat(1,j)=k(1,j)+(gamma/(beta*deltaT))*c(1,j)+(1/(beta*(deltaT)^2))*m;
a(1,j)=(1/(beta*deltaT))*m+(gamma/beta)*c(j);
b(1,j)=(1/(2*beta))*m+deltaT*((gamma/(2*beta))-1)*c(1,j);
dD=zeros(length(Ag),1);
dV=zeros(length(Ag),1);
dA=zeros(length(Ag),1);
D=zeros(length(Ag),1);
V=zeros(length(Ag),1);
A=zeros(length(Ag),1);
dP=zeros(length(Ag),1);
dPhat=zeros(length(Ag),1);
for i=1:(length(Ag)-1)
dP(i)=((Ag(i+1)*m)-(Ag(i)*m))*9.81;
dPhat(i)=dP(i)+a(1,j)*V(i)+b(1,j)*A(i);
dD(i)=dPhat(i)/khat(1,j);
dV(i)=(gamma/(beta*deltaT))*dD(i)-(gamma/beta)*V(i)+deltaT*(1(gamma/(2*beta)))*A(i);
dA(i)=(1/(beta*(deltaT^2)))*dD(i)-(1/(beta*deltaT))*V(i)-(1/
(2*beta))*A(i);
D(i+1)=D(i)+dD(i);
V(i+1)=V(i)+dV(i);
A(i+1)=A(i)+dA(i);
end
Dmax(1,j)=max(abs(D));
Vmax(1,j)=max(abs(V));
Amax(1,j)=max(abs(A));
end
Appendix B
Elastic Analysis - Matlab
69
N=5; %Number of stories

g=9.81; %Gravity
h=3 %Height of stories
%M-Matrix
m=diag([6116.2 6116.2 6116.2 6116.2 6116.2]);
m
%K-Matrix
for n= 1:(N-1)
k(n,n)=(2*3*12*210*10^9*6.06*10^-6)/3^3;
end
for n=N
k(n,n)=(3*12*210*10^9*6.06*10^-6)/3^3;
end
for n=1:(N-1)
k(n,n+1)=-(3*12*210*10^9*6.06*10^-6)/3^3;
end
for n=1:(N-1)
k(n+1,n)=-(3*12*210*10^9*6.06*10^-6)/3^3;
end
k
%Eigenvalues
[modes wsquare]=eig(k,m)
T=zeros(N,1);
for n=1:N
T(n)=(sqrt(wsquare(n,n))*(1/(2*pi)))^-1;
end
T
Modes=zeros(N,N);
for i= 1:N
for j = 1:N
Modes(j,i)=modes(j,i)/modes(N,i);
end
end
Modes
%Generalized mass
Mn=zeros(N,1);
for i = 1:N
for j = 1:N
Temp=m(j,j)*(Modes(j,i))^2;
Mn(i,1)=Mn(i,1)+Temp;
end
end
Mn
Lhn=zeros(N,1);
for i = 1:N
for j = 1:N
Temp=m(j,j)*(Modes(j,i));
Lhn(i,1)=Lhn(i,1)+Temp;
end
end
Lhn
%Modal Participation Factor
GamN=zeros(N,1);
for i = 1:N
temp=Lhn(i,1)/Mn(i,1);
GamN(i,1)=temp;
end
GamN
Lnteta=zeros(N,1);
for i = 1:N
for j=1:N
temp=144*j*m(j,j)*Modes(i,j);
Lnteta(i,1)=temp+Lnteta(i,1);
end
end
Lnteta
%Insert Displacement from Response Spectrum
D=[(1.992)/wsquare(1,1) (8.1)/wsquare(2,2) (7.7)/wsquare(3,3) (5.6)/wsquare
(4,4) (7.59)/wsquare(5,5)]
%Story Displacements
for i= 1:N
for j = 1:N
u(j,i)=GamN(i)*Modes(j,i)*D(i);
end
end
u
%SRSS Displacement
usum=zeros(N,1);
for i = 1:N
for j = 1:N
temp=(u(i,j))^2;
usum(i,1)=usum(i,1)+temp;
end
usrss(i,1)=sqrt(usum(i,1));
end
usrss
%Acceleration from Response Spectrum

A=[1.992 8.1 7.7 5.6 7.59]
%Story Forces
for i=1:N
for j = 1:N
f(j,i)= GamN(i)*m(j,j)*Modes(j,i)*A(i);
end
end
f
%Base Shear
Vb=zeros(N,1);
for i= 1:N
for j = 1:N
Temp=f(j,i);
Vb(i,1)=Vb(i)+Temp;
end
end
Vb
%Base Shear SRSS
Vbs=0;
for i = 1:N
Temp=(Vb(i,1))^2;
Vbs=Vbs+Temp;
end
Vbsrss=sqrt(Vbs)
Mb=zeros(N,1);
for i= 1:N
for j = 1:N
Temp = f(j,i)*h*j;
Mb(i,1)=Mb(i,1)+Temp;
end
end
Mb
Appendix C
Correlation Coefficient
73
omega=[3.450248;2.97132;1.509088;1.16982;1.01603;0.728498;0.640462;0.5561;
0.514713;0.49744;0.480185;0.435423;]
zeta=0.05
betain=zeros(12,12);
for i=1:12
for j=1:12
betain(i,j)=sqrt(omega(i)^2)/sqrt(omega(j)^2);
end
end
betain
roin=zeros(12,12);
for i=1:12
for j=1:12
roin(i,j)=(8*zeta^2*(1+betain(i,j))*betain(i,j)^(3/2))/((1-betain
(i,j)^2)^2+4*zeta^2*betain(i,j)*(1+betain(i,j))^2);
end
end
roin
zeta2=0.2
roin2=zeros(12,12);
for i=1:12
for j=1:12
roin2(i,j)=(8*zeta2^2*(1+betain(i,j))*betain(i,j)^(3/2))/((1-betain
(i,j)^2)^2+4*zeta2^2*betain(i,j)*(1+betain(i,j))^2);
end
end
roin2

Methods For Earthquake Analysis

Uploaded by

Copyright:

Available Formats

Methods For Earthquake Analysis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Methods For Earthquake Analysis

Uploaded by

Copyright:

Available Formats

Methods for Earthquake Analysis

Civil and Environmental Engineering

Norwegian University of Science and Technology

Department of Structural Engineering

MASTER THESIS 2012

DATE: June 2012

Methods for Earthquake Analysis

RESPONSIBLE TEACHER: Anders Rnnquist

Norwegian University of Science and Technology

Anders Rnnquist and Ragnar Sigbjrnsson NTNU

The thesis is to be submitted to the Department of Structural Engineering by 11th of

B Elastic Analysis - Matlab

Comparison of implicit and explicit methods [1] . . . . . . . . .

Schematic description of modal response [3] . . . . . . . . . . . . . . . . . .

4.15 Plan of asymmetric plan building [12] . . . . . . . . . . . . . . . . . . . . . .

Determination of the idealized elasto-perfectly plastic force displacement relationship [11] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Comparison of modal periods . . . . . . . . . .

Modal Participation Factors . . . . . . . .

Figure 2.1: Comparison of implicit and explicit methods [1]

Chapter 2. Numerical Methods to Solve Dynamic Problems

Figure 2.2: Constant Acceleration Method and Linear Acceleration Method

This shows that Constant Average Acceleration gives

Linear Acceleration gives = 21 ,

Combining equation (2.1a), equation (2.1b) and equation (2.17) gives

Solving the second for

Chapter 2. Numerical Methods to Solve Dynamic Problems

This is then substituted into the incremental equation of motion

Then the notations k and pn are introduced as

The incremental displacement is then calculated

When un is found equations (2.20) and (2.19) can compute u n and

Figure 2.3: El Centro Ground Motion [4]

Figure 2.4: Displacement history

Chapter 2. Numerical Methods to Solve Dynamic Problems

Non-linear Newmarks Method

The incremental resisting force is

(fS )n = (kn )sec un

Figure 2.5: Illustration of secant and tangent stiffness [3]

Chapter 2. Numerical Methods to Solve Dynamic Problems

Figure 2.6: Illustration of numerical errors [3]

Figure 2.7: Illustration of iterations [3]

2. Calculations for each iteration

(b) un+1 = un+1 + u(j)

(d) (j+1) = R(j) f (j)

pn+1 cu n+1 kun+1

Chapter 2. Numerical Methods to Solve Dynamic Problems

Symmetric Plan Buildings

The equations of motion for a MDOF-system are

If equation (3.2) is substituted into equation (3.1), equation (3.3) is obtained

To use the orthogonality properties of modes, all of terms are pre-multiplied by Tn

The following notation is then adapted

Here n is the damping ratio for the nth mode.

And combining all the modal contributions to find total displacements

Chapter 3. Elastic Analysis

2. Calculate the natural frequencies n and modes n

Modal Response Contributions

Modal Expansion of Excitation Vector

This gives the contribution of the nth mode to s

Comparing equation (3.16) and (3.17) it seen that

And the equivalent static forces become