Barna Suzanna Honors Project

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

Comparison of the Eigenvector and Ritz Vector Analysis

Approaches for Design-Oriented Seismic Analysis of Bridge


Structures

Suzanna R. Barna
Nov. 2021
University of Florida

Abstract
Design-oriented response spectrum analysis (RSA) is imperative to seismic design of bridge
structures. This study focuses on the potential benefits (robustness and computational efficiency)
of using the Ritz vector approach, in lieu of the traditional eigenvector analysis, while
conducting RSA. To accomplish this objective, the study focuses on three cases: 1) a 2-DOF
spring-mass system analyzed using the eigenvector analysis approach; 2) a five-story shear frame
analyzed by employing the Ritz vector approach; and 3) a pile bent analyzed using design-
oriented software with RSA for four scenarios (one eigenvector analysis and three loading
variations of the Ritz vector analysis). The first two illustration cases verify that these two modal
analysis approaches were correctly implemented in the selected software by comparing
analytical solutions to computational results. The third configuration investigated involves a case
study that provides measurable, comparative results for the robustness and the computational
efficiency of the eigenvector and Ritz vector approaches. The results of the study suggest that the
Ritz vector approach employed in RSA directly improves the incorporation of nonlinear
phenomena. Further, the uptake of cumulative mass participation is greater for the Ritz vector
approach than the eigen vector approach across the same number of modes. In addition, for the
case study, use of the Ritz vector approach more efficiently satisfies the commonly accepted
minimum threshold of 90% mass participation for design of bridge systems.

I
Table of Contents
Abstract ............................................................................................................................................ I
Introduction ..................................................................................................................................... 1
Purpose: ....................................................................................................................................... 1
Objectives: ................................................................................................................................... 1
Scope of Work ................................................................................................................................ 1
Background ..................................................................................................................................... 2
Overview of Response Spectrum Analysis (RSA)...................................................................... 2
Definition:................................................................................................................................ 2
Process: ................................................................................................................................... 2
Limitations & Assumptions: .................................................................................................... 4
Overview of Eigenvector Analysis ............................................................................................. 4
Definition ................................................................................................................................. 4
Process..................................................................................................................................... 4
Overview of Force-dependent Ritz vectors ................................................................................. 5
Definition ................................................................................................................................. 5
Process..................................................................................................................................... 5
Limitations & Assumptions ...................................................................................................... 6
Data and Results ............................................................................................................................. 7
Enhancements to the Modal Analysis Feature Set in FB-MultiPier ........................................... 7
Illustration Cases ......................................................................................................................... 8
Overview .................................................................................................................................. 8
Eigenvector Analysis: 2-DOF System ..................................................................................... 8
Ritz Vector Analysis: Five-story Shear Frame ...................................................................... 10
Case Study: Pile Bent Subjected to Longitudinal Seismic Excitation ................................... 12
Computational Efficiency ............................................................................................................. 16
Conclusion .................................................................................................................................... 17
References ..................................................................................................................................... 19
Appendix A: Python coding.......................................................................................................... 20

II
Table of Figures
Figure 1. Spectral acceleration response spectrum example: a) Multiple seismic excitations
(average acceleration is drawn using a thick black line) [5]; b) Example acceleration
spectrum for the El Centro earthquake. ................................................................................... 3
Figure 2. Pseudo-code for RSA with use of eigenvalue analysis ................................................... 5
Figure 3. Approximate solution of the eigenproblem using force-dependent Ritz vectors [4] ...... 6
Figure 4. FB-MultiPier model data view window .......................................................................... 7
Figure 5 FB-MultiPier Dynamics - Advanced dialog ..................................................................... 8
Figure 6. 2-DOF spring-mass system for illustration of eigenvector analysis ............................... 9
Figure 7. Five-story shear frame from [3], for use in illustration of Ritz vector analysis ............ 11
Figure 8. Pile bent structural configuration and soil conditions (adapted from [4])..................... 13
Figure 9. Northridge earthquake response spectra scaled to 1.4g (max) ...................................... 13
Figure 10. Placement of static longitudinal loads for generation of pushover curve (adapted from
[4]) ......................................................................................................................................... 14
Figure 11. Pushover curve for the pile bent model under uniform longitudinal loading.............. 14
Figure 12. Profile plot of deflection for pile 5 when subjected to high-intensity longitudinal
seismic excitation .................................................................................................................. 15
Figure 13. Profile plot of bending moment for pile 5 when subjected to high-intensity
longitudinal seismic excitation .............................................................................................. 16
Figure 14. Improved cumulative mass participation (%) with use of the Ritz vector approach... 17

III
Introduction
In seismic design of bridge systems for a wide range of loading scenarios, specifications such as
the ASCE 7 Standard [1] and the AASHTO LRFD Bridge Design Specifications [2], indicate the
need to perform Response Spectrum Analysis (RSA). In the context of structural design,
provisions typically call for use of eigenvector analysis to perform RSA with inclusion of a
specified minimum number of modes and/or achievement of a minimum threshold of cumulative
mass participation (e.g., 90%) [1]. However, design-oriented numerical models of physical
infrastructure systems are often complex and comprised of thousands of nodes (and associated
degrees of freedom, DOFs). Specific to the design of bridge infrastructure, both improved
incorporation of nonlinear phenomena and computational efficiency remain of interest to the
current industry design practices.
Given these two areas of interest, the present study aims to review and demonstrate the potential
benefits of a selected analytical modal analysis technique, when employed as part of RSA.
Namely, an analysis technique investigated in this study is the force-dependent Ritz vector
(approximate) solution to the eigenproblem. The Ritz vector approach is an alternative to the
more traditional eigenvector analysis method and provides more flexibility in seismic design
applications where computational efficiency and increased incorporation of nonlinear
phenomena are of interest.
Purpose:
The purpose of this applied research effort is to highlight the benefits of the Ritz vector analysis
method, with respect to the potential increases in computational efficiency and robustness when
utilized as part of RSA. The research focuses on bridge design scenarios where it is of interest to
more prominently capture nonlinear phenomena under seismic loadings.
Objectives:
To investigate the potential advantages associated with the Ritz vector analysis methodology as a
means of approximating the solution to the eigenproblem, the objectives of this study are to: 1)
document the salient aspects of both the eigenvector analysis and the Ritz vector approaches, as
employed within RSA; 2) verify correct implementation of these two approaches with a selected
software tool; and, 3) perform a case study, where computed results obtained from use of the
eigenvector analysis and Ritz vector approaches are presented and compared.

Scope of Work
To accomplish the objectives of the study, the scope of work includes delineation of both the
eigenvector analysis and Ritz vector analysis approaches, along with contextualization of the
roles of these techniques within RSA. For both approaches, highly idealized illustration cases are
selected (e.g., from a structural dynamics textbook [3]) and manual frequency-based solutions of
these (idealized) systems are obtained under seismic loading. Subsequently, a case study is
carried out on a pile bent model adapted from a real-world configuration. The case study serves
to further elucidate the merits of both approaches. Computational efforts carried out as part of the
present research involve use of design-oriented bridge finite element (FEA) software. The
software FB-MultiPier (developed at the University of Florida) is utilized in the study for
computing the solutions to the eigenproblem and Ritz vector analysis for the aforementioned

1
(idealized) illustration cases, and also, for obtaining solutions to RSA performed as part of the
case study.
The first illustration case presented is a 2-DOF spring-mass system, consisting of two springs
and attaching masses, subjected to an excitation (in spectral form) comparable to the El Centro
earthquake. The analysis of this 2-DOF system includes manual and computed solutions,
obtained through use of the eigenvector analysis method. For this illustration case, eigenvector
analysis is first performed manually (by hand; as well as via use of Python coding). Then, the 2-
DOF system is modeled and analyzed using the FB-MultiPier software. Pertinent response
quantities obtained from the solutions (manual, computed) are then compared.
For illustration of the Ritz vector approach, an example from a structural dynamics text [3] is
selected. The example is a five-story shear-frame building subjected to lateral loading and given
initial loading conditions. The solution to the eigenproblem is approximated using Ritz vectors as
is done in the text [3]. Then, the example building is modeled and analyzed using the FB-
MultiPier software. In this second illustration case, focus is dedicated to the Ritz vector
(approximate) solution to the eigenproblem, rather than that associated with employing the full
RSA procedure. Comparisons are made between the manually produced and the computed mode
shapes of the five-story building.
As the final component of this study, a pile bent, adapted from [4], is modeled and analyzed with
both eigenvector analysis and Ritz vector analysis. More specifically, RSA is performed upon a
pile bent model subjected to the Northridge earthquake (and associated spectral accelerations).
One eigenvector analysis and several Ritz vector analyses—the latter utilizing a range of lateral
load levels—are conducted while employing RSA for this case study. Computed profiles of
demands (e.g., displacements, internal moments) are then plotted and compared for each of the
analyses conducted. The relative computational efficiencies associated with use of the
eigenvector analysis and Ritz vector approaches are also examined. Further, the required number
of modes needed to achieve generally recommended thresholds of, for example cumulative mass
participation, are compared across the various solution sets.

Background
Overview of Response Spectrum Analysis (RSA)
Definition:
Response Spectrum Analysis (RSA) estimates the maximum response of a system due to the
base excitation from a dynamic event. This linear stochastic analysis is often used for seismic
events (earthquakes). RSA utilizes the information found from modal analysis and response
spectra plots to combine responses, using modal superposition, into an average maximum
response. Overall, the purpose of RSA is to promote conservative predictability for system
responses to otherwise dynamic and random events.
Process:
The inputs for RSA are mass and stiffness matrices, as well as the chosen base excitation
response spectra. This analysis involves three main steps: modal analysis, response spectrum
interpolation, and modal superposition. For the first step, modal analysis, eigenvector analysis is
typically conducted to find the natural frequencies (eigenvalues) and periods. Modes of vibration
are also determined, and are commonly referred to as mode shapes or eigenvectors. Eigenvector
analysis is further discussed later in this study. Once completed, the spectral acceleration (𝑆𝑆𝑎𝑎 ),
2
defined as the peak linear (and absolute) acceleration of a structure at a particular natural
frequency, is found on a response spectrum plot. A response spectrum can represent either a
particular seismic event, or an average magnitude for different seismic events as a function of
period (Fig. 1). Using 𝑆𝑆𝑎𝑎 , the coefficient of seismic response (𝐶𝐶𝑠𝑠 ) can also be determined for the
analysis.

a)
Acceleration Spectrum for El Centro
4.5
4.0
3.5
Acceleration (g)

3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.065
0.077
0.091
0.107
0.127
0.150
0.178
0.210
0.248
0.294
0.347
0.411
0.486
0.574
0.679
0.803
0.950
1.123
1.328
1.570
1.857
2.195
2.596
3.069
3.630
4.292
5.075

Period (s)

b)
Figure 1. Spectral acceleration response spectrum example: a) Multiple seismic excitations (average acceleration is
drawn using a thick black line) [5]; b) Example acceleration spectrum for the El Centro earthquake.

The next step in the RSA procedure is to solve for participation factors that indicate the
proportional distribution of each mode shape contribution. Modal superposition is then used to
combine the responses of multiple modes to result in a single predicted displacement response.
Square Root Sum of Squares (SRSS) and Complete Quadratic Combination (CQC) are two
common methods for combining modal responses. Note that the FB-MultiPier software makes

3
use of CQC for this purpose, as the CQC method is potentially more robust to a greater quantity
of modes and unevenly spaced (non-uniform intervals between) modes [3].
The final (combined) system response is linear and can be analyzed in the frequency domain,
where the set of displacements associated with the combined mode shapes constitutes a
conservative prediction of the system response given the (for example) acceleration spectrum.
Absolute values of internal forces are also computed from this overall process, based on the
(single) combined set of displacements and the distributions of system mass and acceleration
quantities.
Limitations & Assumptions:
The RSA methodology provides an acceptable estimate for maximum responses for “short,
nondeterministic transient dynamic events” [6]. Moreover, RSA is valid “for single point, single
component excitation classically damped linear systems” [7]. As a limitation, only indirect,
approximate means of incorporating nonlinearity is available for routine design-oriented RSA, as
in there is a constant stiffness and mass matrix for a single RSA. For bridge models with
increased DOFs, the number of modes required to achieve generally accepted thresholds of
quantities such as mass participation (e.g., 90%) can render the traditional solution to the
eigenproblem intractable.
Overview of Eigenvector Analysis
Definition
As mentioned above, eigenvector analysis is an integral part of RSA of linear undamped and
damped systems. The eigenvector analysis uses mass and stiffness matrices as input quantities, to
result in eigenvalues (natural frequencies) and eigenvectors (mode shapes of vibration). The
mode shapes determined from eigenvector analysis pertain to undamped free-vibration
conditions, which in turn, provide valuable information about the behavior of physical systems
[8].
Process
To aid in contextualizing how systems are analyzed utilizing this modal analysis method, the
process of conducting response spectrum analysis—with use of eigenvector analysis—is
delineated in Fig. 2. Consistent with the discussion above, the overall RSA methodology is
divided into three parts. The analysis begins with a given stiffness [K] and mass [M] matrix of
the system. The eigenvalues (𝜆𝜆) and eigenvectors (ϕ) are solved for by first determining the
system matrix [H], which is the inverse of [M] multiplied by [K]. With the known system matrix,
either manual solution or a function available in various coding languages (for example,
np.linalg.eig() in Python) can produce 𝜆𝜆 and ϕ; Fig. 2. depicts a Python function. Next, the
natural frequency (𝜔𝜔) for each eigenvalue can be determined and the natural frequencies can be
traced along the abscissa of the acceleration spectrum, where the corresponding spectral
acceleration (𝑆𝑆𝑎𝑎 ) can be plotted and found. From this, the seismic response coefficient (𝐶𝐶𝑠𝑠 ) is
calculated with the given equation. The final part of RSA utilizes the information obtained thus
far to solve for the effective modal mass for each mode �𝑚𝑚 � 𝑒𝑒𝑟𝑟 �, and subsequently, the
displacements for each node in each mode (𝑥𝑥𝑖𝑖𝑖𝑖 ) (Fig. 2). For the single system response, the
modal responses are combined.

4
Figure 2. Pseudo-code for RSA with use of eigenvalue analysis

Overview of Force-dependent Ritz vectors


Definition
In the context of structural analysis techniques, force-dependent Ritz vectors allow for
approximation to the solution of the eigenproblem. The resulting quantities obtained from the
Ritz vector approach are the response frequencies (eigenvalues) and mode shapes (eigenvectors)
for a desired number of response modes. After solution to the eigenproblem has been achieved,
the overall RSA procedure is carried out in the usual manner. However, since Ritz vectors are
force-dependent and are influenced by a particular, chosen loading scenario [8], it is possible that
the use of a Ritz vector approach (as opposed to the eigenvector analysis approach) may enable
more effective estimation of structural response to seismic excitation through computation and
combination of relatively fewer vibration modes. For large systems with many degrees of
freedom, this reduction in calculation and computing streamlines programs and codes for greater
efficiency and allows for design-relevant response quantities to be obtained with relatively
smaller computing burden. In turn, the Ritz vector approach may lead to (for example) more
efficient uptake of cumulative mass participation to reach the commonly required 90% mass
participation for RSA analysis (e.g., per ASCE 7 [1]).
Process
The procedure for Ritz vector analysis, in contrast to eigenvector analysis, utilizes supplied
(approximated) desired displacements to estimate eigenvectors and mode shapes. Paired with the
stiffness and mass matrices, these quantities (displacement, stiffness, and mass) are used to solve
for the Ritz vectors in a recursive manner, as delineated in Fig. 3 [4]. With the initial, desired
lateral displacements chosen the Ritz vector approach utilizes given system stiffness [K], mass
[M], and displacements {𝑦𝑦1 } to form the first Ritz vector {𝜓𝜓1 }. Once completed, the other Ritz

5
vectors {𝜓𝜓𝑖𝑖 } can be found utilizing the stiffness, mass, displacements, and the previous Ritz
vector (or previous mode of vibration). Once all Ritz vectors are recursively produced, the
eigenproblem is solved to calculate eigenvalues (𝜆𝜆𝑟𝑟𝑟𝑟𝑟𝑟 ) and eigenvectors (ϕ𝑟𝑟𝑟𝑟𝑟𝑟 ), obtained with
use of the reduced-problem stiffness [Kred] and the reduced-problem mass [Mred]. In this context,
‘reduced’ signifies that the degrees of freedom have been simplified down from the number of
DOF in the actual numerical model to that of the number of Ritz vectors of interest. As
aforementioned, the RSA procedure then proceeds in the usual manner (recall the bottom
portions of Fig. 2).

Figure 3. Approximate solution of the eigenproblem using force-dependent Ritz vectors [4]

As emphasis, the use of reduced stiffness [Kred] and mass [Mred] matrices as part of the force-
dependent Ritz vector approach can facilitate the estimation of eigenvectors utilizing fewer
modes than traditional eigenvector analysis, potentially saving engineers in industry valuable
time and resources.
Limitations & Assumptions
When the Ritz vector approach is utilized to approximate the solution to the eigenproblem, such
use makes it prone to similar benefits and drawbacks are associated with ‘ordinary’ eigenvector
analysis. However, because the Ritz vector approach incorporates a set of system displacements
(under a given, initial loading) into the approximate solution to the eigenproblem, this approach
is better equipped to incorporate nonlinear phenomena into the overall RSA process.

6
Simultaneously, a limitation of this analysis approach arises due to the engineering judgement
necessary to choose reasonable displacements associated with the desired initial loading.

Data and Results


Enhancements to the Modal Analysis Feature Set in FB-MultiPier
The Ritz vector modal analysis procedures discussed above have been implemented in FB-
MultiPier v5.9.0. To make use of this feature within the program user interface (UI), while
editing any dynamics model, the engineer can navigate to the Dynamics page and open the
Advanced dialog (Fig. 4). Within the Advanced dialog, the feature is then activated by checking
the option (Fig. 5) labeled as: “Use Force-dependent Ritz Vectors for Eigenanalysis.”

Figure 4. FB-MultiPier model data view window

7
Figure 5 FB-MultiPier Dynamics - Advanced dialog

Illustration Cases
Overview
The following illustration cases are presented to: 1) provide additional context for how the
eigenvector analysis and the Ritz vector analysis approaches factor into the overall RSA
procedure, particularly as implemented in design-oriented bridge FEA software; and, 2) establish
idealized verification cases for both analysis approaches.
Eigenvector Analysis: 2-DOF System
Consider the 2-DOF system depicted in Fig. 6, which consists of two masses and two springs
connected in series. For this illustration case, the 2-DOF system is subjected to seismic excitation
in the form of an acceleration spectrum adapted from the El Centro earthquake. This system is
analyzed using manual (hand) solution, Python coding (see Appendix A for additional details),
and FB-MultiPier modeling.

8
Figure 6. 2-DOF spring-mass system for illustration of eigenvector analysis

Note that the analytical Python coding solution makes use of the SRSS technique to combine the
modal responses; recall, on the other hand, that the CQC method is utilized in FB-MultiPier.
However, for the 2-DOF system, there is negligible difference between the two modal
combination methods, as attributable to the system consisting of only two degrees of freedom.
The mass and stiffness values chosen for this calculation selected purely to facilitate
straightforward manual calculations. Namely, the stiffnesses (k1, k2) and masses (m1, m2) are
defined as:
𝒌𝒌𝒌𝒌𝒌𝒌
𝒌𝒌𝟏𝟏 = 𝟏𝟏 Eqn. (1)
𝒊𝒊𝒊𝒊
𝒌𝒌𝒌𝒌𝒌𝒌
𝒌𝒌𝟐𝟐 = 𝟐𝟐 Eqn. (2)
𝒊𝒊𝒊𝒊

𝒌𝒌𝒌𝒌𝒌𝒌 − 𝒊𝒊𝒊𝒊
𝒎𝒎𝟏𝟏 = 𝟏𝟏 Eqn. (3)
𝒔𝒔𝟐𝟐

𝒌𝒌𝒌𝒌𝒌𝒌 − 𝒊𝒊𝒊𝒊
𝒎𝒎𝟐𝟐 = 𝟏𝟏 Eqn. (4)
𝒔𝒔𝟐𝟐

With the given inputs, the resulting natural frequencies (ω1, ω2) are calculated to be:
𝝎𝝎𝟏𝟏 = 𝟎𝟎. 𝟔𝟔𝟔𝟔𝟔𝟔𝟔𝟔
Eqn. (5)
𝝎𝝎𝟐𝟐 = 𝟐𝟐. 𝟏𝟏𝟏𝟏𝟏𝟏𝟏𝟏𝟏𝟏
Eqn. (6)
When plotted on the El Centro acceleration spectrum plot, the coefficients of seismic response
(Cs1, Cs2) are found to be:
𝑪𝑪𝒔𝒔𝒔𝒔 = 𝟎𝟎. 𝟎𝟎𝟎𝟎𝒈𝒈
Eqn. (7)
𝑪𝑪𝒔𝒔𝒔𝒔 = 𝟎𝟎. 𝟎𝟎𝟎𝟎𝟎𝟎𝟎𝟎
Eqn. (8)

9
From these quantities, the resulting displacements, participation factors, and internal (spring)
forces are solved for (Table 1).
Displacement (in.)
Node Analytical FB-MultiPier Percent Difference
1 9.74 9.74 0.00%
2 7.61 7.61 0.00%

Participation Factors
Mode Analytical FB-MultiPier Percent Difference
1 12.36 12.36 0.00%
2 0.16 0.16 0.04%

Spring Forces (kips)


Node Analytical FB-MultiPier Percent Difference
1 4.27 4.30 0.69%
2 7.61 7.61 0.03%
Table 1. Comparison of analytical and computed results for the 2-DOF illustration case

The results listed in Table 1 serve to verify that the FB-MultiPier software consistently produces
results that are in agreement with the analytical results. For example, the percent difference
between the resulting displacements obtained from computational versus analytical versus
approaches is approximately 0%. For the participation factors, the greatest percent difference is
0.04%, which can be considered negligible and again constitutes suitable agreement between the
analytical solution and the computed results. Moreover, the spring forces fall within 1% percent
of one another, demonstrating excellent agreement. Given the consistently strong levels of
agreement observed in association with the 2-DOF illustration case, the FB-MultiPier software is
judged to be (within the scope of testing conducted) verified for use in carrying out RSA, in
conjunction with the traditional eigenvector analysis.
Ritz Vector Analysis: Five-story Shear Frame
The second illustration case considered is based on a five-story shear frame building, as detailed
in Ch. 14 of [3]. For this five-story shear frame (Fig. 7), the objective is to determine the
vibration properties using the Ritz vector approach, through use of both a manual solution and
the FB-MultiPier software. Accordingly, the first two natural frequencies and modes of vibration
are determined using two-force dependent vectors. For this example, each story of the shear
frame possesses a lateral translation DOF (u1 through u5). Also, mass (m) and inter-story stiffness
(k) quantities are assigned as shown in (Fig. 7). Furthermore, to ‘prime’ the Ritz vector approach,
initial horizontal forces (denoted, s) are applied to each story at the respective DOF. For
simplicity, the story forces (s) are defined as the product of the story mass (m) and 1g.

10
Story mass (m) = 0.26 k-s/in.2
Story stiffness (k) = 36.6 k/in.
Story force (s) = mg

Figure 7. Five-story shear frame from [3], for use in illustration of Ritz vector analysis

To perform RSA with the Ritz vector approach (Fig. 3), the first step is the determination of the
first Ritz vector. As aforementioned, the first Ritz vector is simply the orthogonalized (and mass
normalized) set of system displacements that were obtained due to the initial loading (i.e., the
horizontal force s applied at each story). Subsequent to calculation of the first Ritz vector, the
second Ritz vector (and any other Ritz vector of interest beyond the second) is quantified via the
recursive process delineated in the center portion of Fig. 3.
With the two Ritz vectors for the five-story illustration case determined, reduced-stiffness and
reduced-mass matrices are subsequently solved for and submitted for eigenvector analysis. Here,
‘reduced’ denotes a reduced space, from the number of DOF in the actual numerical model down
to a space that is commensurate with the number of Ritz vectors of interest. For example,
because the five-story shear building possesses five DOF and two Ritz vectors are solved for, the
reduced-stiffness and reduced-mass quantities resolve to 2x2 matrices. In turn, further operations
involving use of the reduced stiffness and mass matrices promote computational efficiency in
approximating the natural frequencies (eigenvalues) and mode shapes (eigenvectors) of the
system.
Comparative listings of the manual solution, as sourced from [3], and computed solution, as
obtained from FB-MultiPier analysis, are given in Table 2. Both the manual solution and the FB-
MultiPier results indicate natural frequencies to be 0.5 Hz and 1.53 Hz for the first and second
modes, respectively. Furthermore, results produced from modeling the five-story system in FB-
MultiPier are accurate to within approximately 0.05% percent difference relative to the manual
solution. Summarily, the solution approaches indicate excellent agreement.

11
Displacement (in.) Percent Difference
Mode 1 (0.5Hz) Mode 2 (1.53 Hz)

Node Analytical [3] FB-MultiPier Analytical [3] FB-MultiPier Mode 1 Mode 2


1 1.17 1.17 0.89 0.89 0.02% 0.01%

2 1.08 1.08 0.40 0.40 0.04% 0.02%

3 0.90 0.90 -0.39 -0.39 0.01% -0.02%

4 0.64 0.64 -1.12 -1.12 0.01% -0.03%

5 0.33 0.33 -1.23 -1.23 0.00% -0.03%


Table 2. Comparison of analytical [3] and computed results for the five-story illustration case

Case Study: Pile Bent Subjected to Longitudinal Seismic Excitation


Having delineated two established techniques for producing eigenvectors and eigenvalues of
spring-mass systems and having verified the associated computations in a design-oriented
software tool, a case study is undertaken in the remainder of the present study. More specifically,
a pile bent is modeled in FB-MultiPier and subjected to seismic excitation (within the frequency
domain). The structural configuration, underlying soil conditions, and excitation are adapted
from [4]. The case study consists of performing RSA with use of, one at a time, the eigenvector
analysis and the Ritz vector approaches. In the following, computed responses from the
collection of analyses are compared and the relative computational efficiencies of the two
approaches are assessed.
The pile bent structural configuration and soil conditions, as adapted from [4], are shown in Fig.
8. For the in-service bridge, the bent cap is positioned at the terminating end of an 80-ft span.
However, for the sake of simplicity in the present study, the superstructure is not included in the
analyses conducted. As an additional simplifying measure, and to offset the absence of the
superstructure in the numerical model, the pile heads are assumed as being under fixed-head
conditions.

12
Figure 8. Pile bent structural configuration and soil conditions (adapted from [4])

Also shown in Fig. 8, the pile bent cap consists of a 3-ft square cross-section, supported by ten
HP 350x79 piles. The bent cap concrete has a compressive strength of 4.5 ksi and an elastic
modulus of 3829 ksi. Longitudinal reinforcement in the bent cap is of mild steel and holds a
yield stress of 60 ksi. The reinforcement has an elastic modulus of 29000 ksi. The h-piles (HP
350x79) have a yield strength of 5.5 ksi and an elastic modulus of 29000 ksi. Spacing between
the centerlines of each of the ten h-piles is 7.87-ft. All piles are embedded 16.4-ft into the soil
below; the soil consists of a medium-dense sand, underlain by a rock layer; additional
configuration details are given in [4], and detailed properties for the soil are listed in Ch. 13 of
[9].
For the pile bent, the model is subjected to a base acceleration, derived from the 1994 Northridge
earthquake, but where the base acceleration is scaled purely to mimic a high-intensity seismic
event. As a result, the scaled base acceleration has a maximum acceleration of 0.4g and Arias
intensity of 4.4 ft/s [10], [11]. The corresponding response spectrum that is actually used for the
analyses conducted herein (Fig. 9) possesses a maximum ordinate of 1.4g.
1.6
1.4
1.2
Acceleration (g)

1
0.8
0.6
0.4
0.2
0
0.001 0.01 0.1 1 10
Period (sec)

Figure 9. Northridge earthquake response spectra scaled to 1.4g (max)

13
The analyses discussed in the following comprise a targeted parametric study. Four analyses are
conducted in total. In each scenario considered, the pile bent model is analyzed using RSA and
the response spectrum shown in Fig. 9, when applied in the longitudinal direction (perpendicular
to the bent cap long dimension). For one of the analyses, RSA is conducted with use of
eigenvector analysis. For the three additional analyses, RSA is conducted using the Ritz vector
approach, and across a range of initial lateral (longitudinal) loadings.
To select initial lateral loadings, a pushover curve is first generated for the pile bent model to
determine how much longitudinal load the system can withstand before catastrophic failure
occurs (Fig. 11). Here, the top of each pile (Node 1-10) is subjected to a load that equals the
percentage of pushover load divided by the number of piles (10) as seen in Table 3 and
illustrated in Figure 10. Through placement of uniform static lateral loads, a maximum pushover
load for the pile bent is determined to be 300 kips. The load levels selected for the variations of
the Ritz vector method are 25%, 50% and 75% of the maximum pushover load (Table 3).

Figure 10. Placement of static longitudinal loads for generation of pushover curve (adapted from [4])

300
Longitudinal Load (kips)

250
Load 75%
200
Load 50%
150

100 Load 25%


Pushover Curve
50
Load Levels
0
0 0.5 1 1.5 2 2.5 3
Longitudinal Displacement (in)

Figure 11. Pushover curve for the pile bent model under uniform longitudinal loading

14
Applied longitudinal
load as a percentage Total load applied Load applied at each
of the pushover load (kips) pile head (kips)
25% 75 7.5
50% 150 15
75% 225 22.5

Table 3. Maximum pushover load for the model and at each pile for the three Ritz vector variations (Load 25%,
Load 50%, and Load 75%).

The load levels allow the system to be ‘primed’ with incrementally increasing displacement,
highlighting a key advantage of Ritz vector analysis to produce pertinent estimations of
eigenvalues and eigenvectors under variable loading conditions. Alternatively stated, the Ritz
vector approach grants the engineer freedom to ‘prime’ the system being analyzed with any
chosen set of static loads, and thereby facilitate the capture of some measure of nonlinearity in
the system when performing RSA.
Demand profile plots obtained from four uses of the RSA procedure (one eigenvector analysis;
three variations of the Ritz vector analysis) are illustrated in Fig. 12 and Fig. 13. These demand
profiles are based on one of the two piles nearest to the center of the bent cap (i.e., the fifth pile
from the left from Fig. 10).
Consistent with the initial loading scenarios, deflections obtained from the four RSA scenarios
indicate increases in maximum displacements for all Ritz vector analyses, as compared to that
obtained with use of eigenvector analysis (Fig. 12). Furthermore, for the 50% and 75% initial
loadings (associated with Ritz vector analysis), the most severe profiles of displacement are
predicted. Overall, profiles of bending moments indicate a consistent trend with the displacement
profiles from one analysis to another; the eigenvector analysis produces moments of relatively
smaller magnitude (relative to those associated with the Ritz vector approach). Further, for the
50% and 75% initial loading, the most severe moment profiles are produced (Fig. 13).

584
582
580
578
Elevation (ft)

576
574
572
Eigen
570
568 Ritz (Load 25%)
566 Ritz (Load 50%)
564 Ritz (Load 75%)
562
0 0.2 0.4 0.6 0.8
Deflection (in.)

Figure 12. Profile plot of deflection for pile 5 when subjected to high-intensity longitudinal seismic excitation

15
584
582
580
Elevation (ft) 578
576
574
572
Eigen
570
568 Ritz (Load 25%)
566 Ritz (Load 50%)
564 Ritz (Load 75%)
562
0 10 20 30 40 50
Bending Moment (kip-ft)

Figure 13. Profile plot of bending moment for pile 5 when subjected to high-intensity longitudinal seismic excitation

Summarily, for both types of demand (displacement, moment), the eigenvector analysis results in
the lowest maximum demands among the four analyses conducted. Furthermore, a variety of
responses can be obtained—with generally increasing severity under increasingly severe initial
loadings—with use of the Ritz vector approach. The ability to bring about a wider array of
responses constitutes a key advantage of the Ritz vector approach, as opposed to the eigenvector
approach. Simultaneously, this same advantage brings about additional responsibility to
practicing engineers when deciding on representative initial loadings for ‘priming’ the system
models when employing the Ritz vector approach for conducting RSA.

Computational Efficiency
For the case study of the pile bent subjected to seismic excitation, the computational efficiencies
of the analysis scenarios (one eigenvector analysis and the three variations of the Ritz vector
analysis) are compared. The cumulative mass participation uptake was recorded over the five
modes for the pile bent under High intensity excitation (Fig. 9), where the uptakes are plotted in
Fig. 14. Among the first five modes, the eigenvector analysis exhibits nearly zero accumulation
of additional mass participation after that of the first mode; on the other hand, the Ritz vector
analyses accumulate mass participation over the first few modes.
The eigenvector analysis does not meet the ASCE 7 [1] threshold of 90% cumulative mass
participation within the 5 modes included in the FB-MultiPier analysis; the three variations of the
Ritz vector approach meet this requirement. The cumulative mass participation associated with
the Ritz (Load 75%) reaches the 90% threshold with only one mode. Moreover, both the Ritz
(Load 25%) and Ritz (Load 50%) pass the threshold once three modes are included in the
analysis (Fig. 14). The improved uptake of cumulative mass participation for the Ritz vector
analyses consistently demonstrates increases in efficiency relative to use of eigenvector analysis.

16
100
98

Cumulative Mass Participation (%)


96
94
92
90
88
86
84 Eigen Ritz (Load 25%)
82 Ritz (Load 50%) Ritz (Load 75%)
80
1 2 3 4 5
Number of Modes Included

Figure 14. Improved cumulative mass participation (%) with use of the Ritz vector approach

On an ordinary laptop, the analysis duration for the model was between two to three seconds for
all four runs. For this simple pile bent, the analysis does not take significant time to run, but for
more complex systems, the Ritz vector approach is typically associated with lower analysis
durations than eigenvector analyses [4]. Therefore, the Ritz vector approach can conserve
valuable time and resources.
The Ritz vector approaches exhibits more efficient uptake of mass participation than the eigen
vector analysis and, therefore, proves to be a viable alternative when conducting RSA to the
traditional eigenproblem solution.

Conclusion
In the seismic design of bridge systems, two areas of interest remain with respect to frequency-
based analysis techniques: 1) improved incorporation of nonlinear phenomena; and, 2)
computational efficiency. The present study was therefore undertaken assess the potential
benefits of the Ritz vector approach (as another option besides the traditional eigenvector
analysis) for solving the eigenproblem as part of RSA. Summaries were presented for each of the
overall RSA procedure, the eigenvector analysis approach, and the Ritz vector approach. Highly
idealized illustration cases were developed and analyzed to further contextualize the eigenvector
and Ritz vector methodologies, and also, the illustration cases were utilized in verifying correct
implementations in design-oriented bridge finite element analysis software.
After, a case study was carried out based on a pile bent model adapted from a real-world
configuration. Seismic analyses were conducted using both the eigenvector analysis and several
scenarios with the Ritz vector approach, used in conjunction with RSA. Various aspects of the
computed solutions were then examined in a comparative manner. Based on the pile bent case
study, the Ritz vector analysis method can be acknowledged as a useful design tool for engineers
as an alternative to the traditional eigenvector approach. The flexibility to ‘prime’ a system with
a desired, representative static loading chosen by the engineer can lend to a more practical and
customized design for bridge structures.

17
The study also explored the computational efficiency of the two approaches. The Ritz vector
approach for RSA has more efficient uptake of mass participation (Fig. 14); this suggests that
this approximation approach is a potentially more robust alternative to eigenvector analysis. The
Ritz vector analysis method can be used as an acceptable and practical analysis method for
bridge design when subjected to seismic excitation.

18
References
[1] ASCE, Minimum Design Loads and Associated Criteria for Buildings and Other Structures,
ASCE/SEI 7-16. American Society of Civil Engineers, 2017. doi: 10.1061/9780784414248.
[2] AASHTO, LRFD Bridge Design Specifications, Ninth. Washington, DC: American
Association of State Highway and Transportation Officials, 2020.
[3] A. Chopra, Dynamics of Structures: Theory and Applications to Earthquake Engineering,
2nd ed. Upper Saddle River, New Jersey: Prentice Hall, 2000.
[4] M. Davidson, S. Patil, A. Rosenfeld, and Z. Zhu, “Comparison of eigenanalysis and Ritz
vector approaches for response spectrum analysis of soil-pile-bridge systems,” Bridge
Structures, 17:3, 2021 (in press).
[5] K. D. Harris, N. D. Robinson, and E. L. Sammarco, “Seismic Time Histories,” STRUCTURE
magazine, Mar. 2013. Accessed: Nov. 04, 2021. [Online]. Available:
https://www.structuremag.org/?p=886
[6] “Response Spectrum Analysis,” COMSOL, Jan. 28, 2019.
https://www.comsol.com/multiphysics/response-spectrum-
analysis#:~:text=What%20Is%20Response%20Spectrum%20Analysis%3F%20Response%2
0spectrum%20analysis,it%20is%20difficult%20to%20perform%20a%20time-
dependent%20analysis. (accessed Oct. 23, 2021).
[7] A. A. Kasar, “Response Spectrum Method of Analysis - with simplified examples,”
civildigital.com, Sep. 10, 2013. https://civildigital.com/response-spectrum-method-analysis-
learn-examples/ (accessed Sep. 26, 2021).
[8] Riaz, “Ritz vs. Eigen vectors - Technical Knowledge Base - Computers and Structures, Inc. -
Technical Knowledge Base (csiamerica.com),” wiki.csiamerica.com, Apr. 04, 2014.
https://wiki.csiamerica.com/display/kb/Ritz+vs.+Eigen+vectors (accessed Aug. 01, 2021).
[9] Bridge Software Institute, “FB-MultiPier Help Manual v. 5.8.1,” University of Florida,
Gainesville, FL, 2021.
[10] V. Grazier, “Effect of Low-Pass Filtering and Re-Sampling on Spectral and Peak Ground
Acceleration in Strong-Motion 500 Records, 15th World Conference on Earthquake
Engineering,” Libson, Portugal, 24-28 September, 2012.
[11] University of California, “PEER Ground Motion Database,” University of California
Berkeley, Berkeley, CA, 2019. Accessed: Mar. 22, 2021. [Online]. Available:
https://ngawest2.berkeley.edu/site

19
Appendix A: Python coding

20
21
Output Results:

22

You might also like