University of Rhode Island

DigitalCommons@URI

Open Access Master's Theses

2016

Dynamic Analysis of Multi-Degree-Of-Freedom Systems Using a

Pole-Residue Method
Kevin A. Goodell
University of Rhode Island, [email protected]

Follow this and additional works at: https://digitalcommons.uri.edu/theses

Recommended Citation
Goodell, Kevin A., "Dynamic Analysis of Multi-Degree-Of-Freedom Systems Using a Pole-Residue Method"
(2016). Open Access Master's Theses. Paper 876.
https://digitalcommons.uri.edu/theses/876

This Thesis is brought to you for free and open access by DigitalCommons@URI. It has been accepted for inclusion
in Open Access Master's Theses by an authorized administrator of DigitalCommons@URI. For more information,
please contact [email protected].
DYNAMIC ANALYSIS OF MULTI-DEGREE-OF-FREEDOM SYSTEMS

USING A POLE-RESIDUE METHOD

BY

KEVIN A. GOODELL

A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE

REQUIREMENTS FOR THE DEGREE OF

MASTER OF SCIENCE

IN

OCEAN ENGINEERING

UNIVERSITY OF RHODE ISLAND

2016
 MASTER OF SCIENCE THESIS

OF

KEVIN A. GOODELL

APPROVED:
Thesis Committee:

Major Professor James Hu

Harold Vincent

George Tsiatas

Nasser H. Zawia
DEAN OF THE GRADUATE SCHOOL

UNIVERSITY OF RHODE ISLAND

2016
 ABSTRACT

This dissertation is concerned with the development of a robust pole-residue

method for the dynamic analysis of multiple-degree-of-freedom structures. The

utilization of this new pole-residue method requires additional analysis and a de-

termination of its effectiveness in calculating a structure’s response to an external

excitation. The components of the computational framework are a deterministic

model of the dynamic loading and a well defined structure. The finite element

method will be used to describe the system, simplify the problem, and develop

the mathematical expressions to describe the system’s behavior. A commercial

finite element software program was used to verify the natural frequencies of the

structures and ensure that the governing equations were defined correctly. A pole-

residue method was implemented and compared to the frequency domain method

and a time-domain method. The system’s total response was found using the time-

domain and pole-residue methods, while the frequency-domain method found the

steady-state response.
 ACKNOWLEDGMENTS

Since 2009 there have been several people who have assisted me in achieving

this Master’s Degree. I owe a great deal of gratitude to everyone who has helped

me reach this point in my life. I have a deep appreciation for all their support and

time spent nurturing and fostering my development. I have been fortunate to have

Professor Hu as my advisor throughout this endeavor. He has been more than

understanding of the difficulties with balancing work, life, and school. Without

his assistance, this would not be possible. I am grateful for all the support and

understanding my co-workers, supervisors, and colleagues have shown as I have

trekked from course to course and finally during these last two months as I closed

out and defended my thesis. No matter how many times I said I wanted to quit,

you guys never let me. Instead, you played devil’s advocate and beat me up a bit.

Okay, fine, a lot. Finally to my parents, for answering the phone and accepting

that there could be a temper tantrum on the other end regardless of the time of

day. I hate admitting when I am wrong, but you guys were right about a lot of

things.

iii
 TABLE OF CONTENTS

ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii

ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . iii

TABLE OF CONTENTS . . . . . . . . . . . . . . . . . . . . . . . . . . iv

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

CHAPTER

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

List of References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Structural Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.1 Single Degree of Freedom Systems . . . . . . . . . . . . . . . . . 6

2.2 Newton’s Laws to Lumped-Parameter Models . . . . . . . . . . 7

2.2.1 Newton’s Laws . . . . . . . . . . . . . . . . . . . . . . . 7

2.2.2 Assumed-Modes Method . . . . . . . . . . . . . . . . . . 10

2.3 Finite Element Modeling of Structures . . . . . . . . . . . . . . 11

2.3.1 Axial Motion . . . . . . . . . . . . . . . . . . . . . . . . 12

2.3.2 Transverse Motion . . . . . . . . . . . . . . . . . . . . . 14

2.3.3 Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3.4 Three Dimensional Frame Element . . . . . . . . . . . . 16

2.4 Vibration Properties of MDOF Systems . . . . . . . . . . . . . . 18

2.4.1 Eigensolution of Undamped Free-Vibration . . . . . . . . 19

iv
 Page

2.4.2 Eigensolution of Damped MDOF Systems . . . . . . . . 19

2.5 Response of Systems to Periodic Excitation . . . . . . . . . . . . 21

2.5.1 Complex Fourier Series . . . . . . . . . . . . . . . . . . . 21

2.5.2 Complex Frequency Response Function . . . . . . . . . . 22

2.5.3 Fourier Integral Transform . . . . . . . . . . . . . . . . . 23

2.5.4 Frequency Response Function . . . . . . . . . . . . . . . 24

2.5.5 System Response Function . . . . . . . . . . . . . . . . . 25

2.5.6 Mode-Superposition for Undamped Systems . . . . . . . 25

2.5.7 Mode-Superposition for Damped MDOF Systems . . . . 27

2.5.8 Frequency Response of MDOF Systems based on Com-

plex Modes . . . . . . . . . . . . . . . . . . . . . . . 28

2.6 Abaqus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.6.1 Abaqus Spring-Dashpot Commands . . . . . . . . . . . . 31

2.6.2 Three Dimensional Frame Element . . . . . . . . . . . . 33

2.6.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.6.4 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.7 Matlab State-Space Methods . . . . . . . . . . . . . . . . . . . . 36

2.7.1 MatLab Function ss . . . . . . . . . . . . . . . . . . . . . 36

2.7.2 MatLab Function lsim . . . . . . . . . . . . . . . . . . . 37

3 A Pole Residue Method for Dynamic Analysis of Systems . . 38

3.1 System Response by the Laplace Transform . . . . . . . . . . . 38

3.2 Utilizing a Prony’s Method for Developing the Poles and Residues 39

3.3 Differential/Difference Equation for Prony Series . . . . . . . . . 40

3.3.1 Implemented Prony’s Method . . . . . . . . . . . . . . . 41

v
 Page

3.3.2 Prony’s Method to Laplace . . . . . . . . . . . . . . . . . 43

3.4 Integration of Pole and Residues with Structural Systems . . . . 43

3.4.1 Equation of Motion . . . . . . . . . . . . . . . . . . . . . 43

3.4.2 Symmetric System Response . . . . . . . . . . . . . . . . 44

3.4.3 Asymmetric System Response . . . . . . . . . . . . . . . 46

3.4.4 Displacement Response . . . . . . . . . . . . . . . . . . . 47

4 Six Degree of Freedom Problem . . . . . . . . . . . . . . . . . . 50

4.1 Dynamical Solution . . . . . . . . . . . . . . . . . . . . . . . . . 50

4.2 Problem Overview . . . . . . . . . . . . . . . . . . . . . . . . . . 50

4.2.1 Free Body Diagram . . . . . . . . . . . . . . . . . . . . . 52

4.2.2 Equation of Motions . . . . . . . . . . . . . . . . . . . . 52

4.2.3 Relation to Displacements and Velocities . . . . . . . . . 53

4.2.4 State Space Model . . . . . . . . . . . . . . . . . . . . . 56

4.3 Abaqus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

4.4 Impact and Measurement Locations . . . . . . . . . . . . . . . . 56

4.5 Excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

4.6 Frequency Domain . . . . . . . . . . . . . . . . . . . . . . . . . 58

4.7 MatLab Time-Domain Analysis . . . . . . . . . . . . . . . . . . 59

4.8 Pole-Residue Method . . . . . . . . . . . . . . . . . . . . . . . . 60

4.8.1 Steady-State Response through the Poles and Residues . 61

4.9 Comparison of Results . . . . . . . . . . . . . . . . . . . . . . . 62

4.10 Function Completion Times . . . . . . . . . . . . . . . . . . . . 64

5 Three Dimensional Frame Structure . . . . . . . . . . . . . . . . 65

vi
 Page

5.1 Problem Overview . . . . . . . . . . . . . . . . . . . . . . . . . . 65

5.2 Matlab Finite Element Analysis Solution . . . . . . . . . . . . . 66

5.2.1 State Space Model . . . . . . . . . . . . . . . . . . . . . 66

5.3 Impact and Measurement Locations . . . . . . . . . . . . . . . . 67

5.4 Excitation Signal . . . . . . . . . . . . . . . . . . . . . . . . . . 67

5.5 Frequency-Domain Analysis . . . . . . . . . . . . . . . . . . . . 68

5.6 MatLab Time-Domain Analysis . . . . . . . . . . . . . . . . . . 69

5.7 Pole-Residue Method . . . . . . . . . . . . . . . . . . . . . . . . 70

5.7.1 Steady-State Response through the Poles and Residues . 71

5.8 Comparison of Results . . . . . . . . . . . . . . . . . . . . . . . 72

5.9 Function Completion Times . . . . . . . . . . . . . . . . . . . . 73

6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

vii
 LIST OF TABLES

Table Page

1 Railway Vehicle Parameters and Values . . . . . . . . . . . . . . 51

2 Parameter Values for the Excitation p (t) . . . . . . . . . . . . 57

3 Poles and Residues of the Excitation . . . . . . . . . . . . . . . 60

4 Poles and Residues of the Transfer Function . . . . . . . . . . . 61

5 Poles and Residues of the Displacement Response . . . . . . . . 61

6 Parameter Values for the Excitation p (t) . . . . . . . . . . . . 67

7 Poles and Residues of the Excitation . . . . . . . . . . . . . . . 70

8 Poles and Residues of the Transfer Function . . . . . . . . . . . 71

9 Poles and Residues of the Displacement Response . . . . . . . . 71

viii
 LIST OF FIGURES

Figure Page

1 Single-Degree-of-Freedom Lumped Parameter Model . . . . . . 6

2 Multiple-degree-of-freedom lumped parameter model . . . . . . 8

3 Multiple-degree-of-freedom lumped parameter model free-body

diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

4 Notation for a three dimensional frame element . . . . . . . . . 17

5 Fourier Transform Process . . . . . . . . . . . . . . . . . . . . . 22

6 Abaqus Representation of the Six Degree of Freedom Lumped

Parameter Model . . . . . . . . . . . . . . . . . . . . . . . . . . 32

7 Zoomed in view of the Abaqus Spring-Dashpot Model . . . . . . 33

8 Abaqus Representation of a Five Story Framed Structure . . . . 34

9 Simplified 6 Degree of Freedom Model of One-Half of a Railway

Vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

10 Free Body Diagram for the One-Half of a Railway Vehicle . . . 52

11 Excitation p (t) . . . . . . . . . . . . . . . . . . . . . . . . . . 58

12 Results utilizing the Frequency-Domain Method . . . . . . . . . 59

13 Results from the use of MatLab functions ss and lsim . . . . . . 60

14 Results from the Pole-Residue Method . . . . . . . . . . . . . . 62

15 Calculated Displacement Response from the Pole-Residue,

Time-Domain, and Frequency-Domain Methods . . . . . . . . . 63

16 Time required to complete twenty iterations for the Pole-

Residue, Analytical, and Fourier Methods for the one-half of
a railway vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . 64

17 Five Story Three Dimensional Frame Structure . . . . . . . . . 66

ix
Figure Page

18 Excitation p (t) . . . . . . . . . . . . . . . . . . . . . . . . . . 68

19 Results from the Frequency-Domain Method . . . . . . . . . . . 69

20 Results from the use of MatLab functions ss and lsim . . . . . . 70

21 Results from the Pole-Residue Method . . . . . . . . . . . . . . 72

22 Calculated Displacement Response from the Pole-Residue,

Time-Domain, and Frequency-Domain Methods . . . . . . . . . 73

23 Execution time for the Pole-Residue, Time-Domain, and

Frequency-Domain Methods . . . . . . . . . . . . . . . . . . . . 74

x
 CHAPTER 1

Introduction

This thesis is concerned with the development of a robust pole-residue method

for the dynamic analysis of multiple-degree-of-freedom structures. The strategy de-

veloped involves utilizing finite element analysis to determine the system response

resulting from time-varying loads. A commercial finite element analysis software

package is used to validate the obtained numerical results. [1]

The study of structures is performed in the initial design phase to ensure

that public safety, finished product reliability, and determination of the required

construction material is well understood. In order to accomplish this feat, the dy-

namic loading needs to be characterized. A dynamic load is one whose magnitude,

direction, or point of application varies with time. The resulting time-varying

displacements and stresses constitute the dynamic response of the structure. [2]

The essential characteristics of a dynamic problem differ from a static problem

not only in the time-varying nature of the load, but also in the complexity of the

solution for related system characteristics. A static problem has a single solution

since there is no time-varying change. However, since a dynamic problem is time-

varying there is a single solution per snapshot of time. [3]

A static problem when subjected to a static load can be solved through using

established principles of force equilibrium equations. The internal moments and

shears of the deflected shape rely solely upon the static load. In a dynamic problem

the resulting displacements of the system not only depend upon the load but also

upon the opposing inertial forces. Thus, the resulting internal forces must be solved

in equilibrium with the internal moments and shears, and not rely solely upon the

dynamic loading. [4]

1
 The two basic approaches for evaluating dynamic structural response involve

deterministic and nondeterministic methods. The selection of which approach to

use depends on how the loading is defined. If the loading is fully known, the

time-varying magnitude is understood even if the loading is highly oscillatory or

irregular. The loading is then referenced as being prescribed dynamic loading.

The analysis of any system’s response to prescribed dynamic loading is performed

using deterministic methods. However, if the time-varying nature of the load is

not known but can be described through the use of statistics, it is referenced as

being random dynamic loading. The analysis of a system’s response to random

dynamic loading is performed using nondeterministic methods. [5]

The structural response to any dynamic loading is expressed through the

structure’s displacement. Deterministic analysis allows for the calculation of the

time-varying displacements, as well as other useful related time-varying response

quantities. A nondeterministic analysis provides only statistical information about

the displacements resulting from the statistically defined loading, and thus related

statistical response quantities.

There are two categories of prescribed or deterministic loading, periodic and

nonperiodic. Periodic loading can be described as a simple continuous width si-

nusoidal pulse, also known as a simple harmonic. Complex periodic loads can be

described by a summation of a series of simple harmonic components through the

incorporation of the fourier transform. This allows the same general procedure to

be used in the analysis of any periodic loading. [6]

Nonperiodic loadings are created from impulsive loadings or long-duration

general loads. Impulsive responses are more commonly observed in Shock testing,

typically the result of the test apparatus being hit by a large hammer or weight.

Long-duration general loads are observed during vibration testing, resulting in the

2
shaking of the structure.

The analysis of dynamic loading encompasses three major steps: defining the

analytical model, creating and applying the corresponding mathematical model,

and solving for the dynamical response. The most demanding step in any structural

analysis is the creation of the mathematical model used to represent the structure.

A mathematical model is developed from the idealized model of the structural

system to be studied similar to the real system. This process reduces the dynamical

problem to an analytical model, thus reducing the complexity of the problem

and making it easier to analyze mathematically. Once the analytical model has

been developed, the application of physical laws (e.g. Newton’s laws, stress-strain

relationships) are used to obtain the differential equations of motions representing

the corresponding mathematical model. The differential equations are then solved

to obtain the predicted dynamical response. [7]

There are three methods for solving dynamic loading problems: lumped-mass

procedure, generalized displacements, and finite element method. Lumped-mass

procedure utilizes Newton’s laws dealing with force and acceleration in vector quan-

tities, to derive the equation of motion for a single particle or rigid body. A simple

spring-mass model is a common example of a lumped-mass model. When using

the lumped-mass model to analyze a structure, the physical components for that

structure are related to displacement, velocity, and acceleration. The generalized

displacements method utilizes a generalized parameter model to approximate the

system’s deformation. This model is used for a continuous system, whose defor-

mation is described through one or more functions of space and time. In order to

be solvable, the model requires that a geometric boundary condition be specified,

constraining the kinematic displacement on a portion of the structure. The finite

element method utilizes key components of both lumped-mass and generalized dis-

3
placements methods. This method discretizes the structure in multiple segments or

elements and solves for multiple highly coupled interpolation functions describing

the structure’s geometry. [8]

This thesis will implement a pole-residue method to determine the system

response. In order to understand the effectiveness of any new method, the results

must be compared against a well known and understood method. This comparison

will be performed using the assumed-modes method which combines the lumped-

mass procedure with Newton’s laws for the analysis of a six-degree-of-freedom

mass-spring-dashpot system, and the finite element method to investigate a simple

five story frame structure. In order to depict conceptual understanding of the topic

at hand, this method will first be implemented in Matlab and later verified in a

commercially available finite element analysis software application, such as Abaqus.

[9, 10]

List of References

[1] S.-L. J. Hu, F. Liu, B. Gao, and H. Li, “Pole-residue method for numerical
dynamic analysis,” Journal of Engineering Mechanics, p. 04016045, 2016.

[2] R. R. Craig and A. J. Kurdila, Fundamentals of structural dynamics. John

Wiley & Sons, 2006.

[3] Y. W. Kwon and H. Bang, The finite element method using MATLAB. CRC
press, 2000.

[4] R. C. Hibbeler, Statics and Dynamics. Upper Saddle River, New Jersey,
United States of America: Pearson Prentice Hall, 2004.

[5] K. Petsounis and S. Fassois, “Parametric time-domain methods for the iden-
tification of vibrating structuresa critical comparison and assessment,” Me-
chanical Systems and Signal Processing, vol. 15, no. 6, pp. 1031–1060, 2001.

[6] A. Cunha and E. Caetano, “Experimental modal analysis of civil engineering

structures,” Sound and Vibration, vol. 40, no. 6, pp. 12–20, 2006.

[7] A. De Vivo, C. Brutti, and J. Leofanti, “Modal shape identification of large

structure exposed to wind excitation by operational modal analysis tech-

4
 nique,” Mechanical Systems and Signal Processing, vol. 39, no. 1, pp. 195–206,
2013.

[8] S. Modak, “Separation of structural modes and harmonic frequencies in op-

erational modal analysis using random decrement,” Mechanical Systems and
Signal Processing, vol. 41, no. 1, pp. 366–379, 2013.

[9] S.-L. J. Hu, W.-L. Yang, and H.-J. Li, “Signal decomposition and recon-
struction using complex exponential models,” Mechanical Systems and Signal
Processing, vol. 40, no. 2, pp. 421–438, 2013.

[10] Abaqus/CAE User’s Guide, Dassault Systemes, Providence, Rhode Island,

United States of America, 2013.

5
 CHAPTER 2

Structural Dynamics

In order to perform dynamic analysis, several concepts need to be understood.

There are a number of factors that influence which model is implemented. The

complexity of the system and the number of influencing parameters need to be

determined to appropriately decide on which model to use. This section will cover

the assumed-modes method, the finite element method, complex analysis using

the Fourier transform, as well as the commercial finite element analysis software

Abaqus.

2.1 Single Degree of Freedom Systems

The fundamental equation in structural dynamics and linear vibration theory

is expressed as second-order differential equation which relates force to displace-

ment, velocity, and acceleration

mẍ + cẋ + kx = p(t) (1)

This equation of motion represents a single lumped-parameter model and is

derived using Newton’s laws, (Figure 1).

Figure 1. Single-Degree-of-Freedom Lumped Parameter Model

6
 The elements which comprise this model include the mass, the spring, and

the damping component. According to the model, an object is placed into motion

through the displacement of its mass at a specified distance from a reference. The

spring element then goes into tension due to being stretched, and it will have the

desire to compress and restore the system to equilibrium. The damping component,

which occurs when impacted by the spring’s restoring motion, will dissipate system

energy decreasing the time required to reach equilibrium.

2.2 Newton’s Laws to Lumped-Parameter Models

By connecting several lumped-parameter models together, a simple single-

degree-of-freedom model can be constructed into a more complex multiple-degree-

of-freedom model. Depending on the complexity of this model, the correct math-

ematical models can then be developed through the application of Newton’s laws

and/or the assumed-modes method. Newton’s laws are best used on simpler mod-

els comprised of a few lumped parameter models described by a mass, spring,

and damping values. For more complex systems, the assumed-modes method is

preferred due to the influence of additional parameters on the system.

2.2.1 Newton’s Laws

In this section, the mathematical model for a multiple-degree-of-freedom

model will be developed using Newton’s laws, (Figure 2). The notation mi repre-

sents mass, ci represents a dashpot, and ki represents a spring. The force related

to the dashpots are represented by Fci and the force related to the springs are

represented by Fsi .

7
 Figure 2. Multiple-degree-of-freedom lumped parameter model

The first step in setting up this solution is to break the model into a free-body

diagram for each mass, (Figure 3).

Figure 3. Multiple-degree-of-freedom lumped parameter model free-body diagrams

After each free-body diagram has been drawn and annotated, the equation of

motion for each particle must be defined per Newton’s second law

X
→ F1 = m1 u¨1 = −Fs1 + Fs2 − Fc1 + Fc2 + P1 (2)

X
→ F2 = m2 u¨2 = −Fs2 + Fs3 − Fc2 + Fc3 + P2 (3)

X
→ F3 = m3 u¨3 = −Fs3 − Fc3 + P3 (4)

The mass element can easily be plugged into the equation of motion. However

the linear elastic spring and viscous damping forces still need to be related to the

8
displacements ui for each mass element. This relationship is illustrated for the

three springs by

Fs1 = k1 u1 (5)

Fs2 = k2 (u2 − u1 ) (6)

Fs3 = k3 (u3 − u2 ) (7)

This relationship is described for the three dashpots through

Fc1 = c1 u˙1 (8)

Fc2 = c2 (u˙2 − u˙1 ) (9)

Fc3 = c3 (u˙3 − u˙2 ) (10)

Incorporating the defined spring and dashpot equations of motions with re-

spect to the displacement ui with the previously defined governing equations of

motions, the mathematical model for the multiple-degree-of-freedom system is de-

fined through the use of matrix notation. This will allow for the evaluation and

determination of the exact solution. These matrices define the mass M, stiffness

K, damping C. The corresponding excitation vector is defined as p.

The mass matrix M is

 
m1 0 0
M =  0 m2 0  (11)
0 0 m3

The damping matrix C is

 
c1 + c2 −c2 0
C =  −c2 c2 + c3 −c3  (12)
0 −c3 c3

The stiffness matrix K is

9
  
k1 + k2 −k2 0
K =  −k2 k2 + k3 −k3  (13)
0 −k3 k3
The excitation vector p is

 
p1 (t)
p = p2 (t) (14)
p3 (t)
 

2.2.2 Assumed-Modes Method

The assumed-modes method incorporates an extension of the virtual displace-

ment method to produce a generalized-parameter model of a continuous system in

order to approximate the deformation of that system. In order to complete this

transformation, several definitions are required to define the generalized-parameter

single-degree-of-freedom model. These definitions are:

1. A continuous system is a system whose deformation is described by multiple

spatial variables and time

2. A geometric boundary condition is a specified kinematical constraint placed

on the displacement of the body

3. A virtual displacement is an infinitesimal, imaginary change in the displace-

ment functions

4. An admissible function that satisfies the geometric boundary conditions of

the system and possesses a derivative of order appearing in the strain energy

expression

To create the generalized-parameter single-degree-of-freedom model of a con-

tinuous system, a single assumed mode is used

v(x, t) = ψ(x)qv (t) (15)

10
The shape function should represent a single deformation experienced by the struc-

ture. The generalized displacement coordinate is defined as qv (t) with the subscript

v denoting the generalized coordinate in relation to the physical displacement

v(x, t). The shape function ψ(x) can be represented by any function, however

a shape representing the deforming structure should be chosen. To generate a

multiple-degree-of-freedom model of a continuous system, Eq 15 is expanded to

include N shape functions, thus allowing the continuous displacement to be ap-

proximated through a finite sum

N
X
v(x, t) = ψi (x)qvi (t) (16)
i=1

The assumed-modes method consists of substituting this equation into the

expressions for kinetic energy, T , and strain energy, V, through representation of

the material by its density ρ, area A, length L, and modulus of elasticity E, as

shown below

Z L
1
T = ρA(u̇)2 dx (17)
2 0
Z L
1
V= EA(v 0 )2 dx (18)
2 0

The analyst defines the N -degree-of-freedom assumed-modes through the se-

lection of the shape function. The shape functions must form a linear independent

set and must possess derivatives up to the order appearing in the strain energy V.

The shape function must also satisfy all the prescribed boundary conditions.

2.3 Finite Element Modeling of Structures

The finite element method is a far more powerful approximation method than

the assumed-modes method. Through the approximation of the displacement func-

tion by using a continuous function, the deflected shape of the entire structure is

11
then described by each shape function ψi
N
X
v(x, t) = ψi (x)vi (t) (19)
i=1

The N -degree-of-freedom mathematical model of the structure is obtained,

using the generalized stiffness, mass coefficients, and generalized forces obtained

from integrals involving the ψi ’s and their derivatives. While using the computer to

solve difficult engineering problems has facilitated the evaluation of these integrals,

the draw backs in the assumed modes method:

1. It is difficult for an engineer to choose a set of ψi ’s for a structure that has

complex geometry

2. The equations that result from the assumed mode method are coupled

3. There is little reuse from one problem to the next

The finite element method overcomes these difficulties and has become the

main computational tool for structural dynamics. The finite element method uti-

lizes a set of ψi ’s to represent a portion of the structure through the assembling

of elements to form the structure. The elements are linked together at nodes and

the displacement compatibility is enforced at these linkages.

2.3.1 Axial Motion

Axial motion is defined through the study of a uniform bar element of length

L, mass density ρ, modulus of elasticity E, and a cross-sectional area A. The

approximation of the axial displacement u(x, t) within the element employs linear

interpolation between the displacements u1 (t) and u2 (t) at the two ends and shape

functions ψ1 (x) and ψ2 (x)

u(x, t) = ψ1 (x)u1 (t) + ψ2 (x)u2 (t) (20)

12
The shape functions ψ1 (x) and ψ2 (x) must satisfy the boundary conditions, which

are derived by considering axial deformation under static loads and are used to

approximate axial displacement along an element of length L.

When the axial motion of the plane cross section at x is represented by u(x, t)

where the generalized coordinates are labeled ui (t), Eq. 16 can be written as

N
X
u(x, t) = ψi (x)ui (t) (21)
i=1

When Eq. 21 is subsituted in Eq. 18 the resulting expression for strain energy

is

N N
1 XX
V= kij ui uj (22)
2 i=1 j=1

Where the stiffness coefficients for axial deformation is given by

Z L
kij = EAψi0 ψj0 dx (23)
0

In a similiar fashion, the combination of Eq. 21 with Eq. 17 produces the

following quadratic expression for the kinetic energy of a bar

N N
1 XX
T = mij u̇i u̇j (24)
2 i=1 j=1

Where the mass coefficient is

Z L
mij = ρAψi ψj dx (25)
0

When the mass matrix and stiffness matrix are formed through an equation

like Eqs. 23 and 25, they are defined as being consistent, as they were formed

through a similiar manner using the same shape functions.

13
 If the bar experiences an external excitation force, then the generalized forces

are determined by employing virtual work δW

Z L N
X
δW = p(x, t)δu(x, t) dx = pi (t)δui (26)
0 i=1
Thus, when the virtual work equation, Eq. 26, is integrated with the displacement

function
N
X
δu(x, t) = ψi (x)δui (27)
i=1
The generalized force for each shape function is obtained from

Z L
pi (t) = p(x, t)ψi (x) dx (28)
0

Expressions for the stiffness coefficients kij , the mass coefficients mij , and the

generalized forces pi (t) for axial motion were defined earlier in Eqs. 23, 25, and

28. Through substituting ψ1 and ψ2 into those equations, the stiffness and mass

matrices for a uniform element undergoing axial deformation is defined by

 
AE 1 −1
k= (29)
L −1 1

 
ρAL 2 1
m= (30)
6 1 2

2.3.2 Transverse Motion

Transverse motion is defined through the study of a uniform beam element of

length L, mass density ρ, elastic modulus E, cross sectional-area A, and moment

of inertia I. The approximation of the transverse displacement v(x, t) with the

element utilizes linear interpolation between the displacements v1 (t) and v2 (t) at

the two ends and shape functions ψ1 (x) and ψ2 (x)

v(x, t) = ψ1 (x)v1 (t) + ψ2 (x)v2 (t) (31)

14
 The shape functions ψ1 (x) and ψ2 (x) must satisfy the boundary conditions.

The shape functions are derived by considering the beam element to be loaded

statically by end shears and bending moments to produce the various static deflec-

tion shapes. Expressions for the stiffness coefficients kij , the mass coefficients mij ,

and the generalized forces pi (t) for axial motion was defined earlier in Eqs. 23, 25,

and 28. Through the substitution the boundary conditions for ψ1 through ψ4 into

those equations, the stiffness and mass matrices for a uniform element undergoing

transverse deformation is

 
6L −12 6L
12
EI  4L2 −6L 2L2 
k= 3   (32)
L  12 −6L
symm. 4L2
 
156 22L 54 −13L
ρAL  4L2 13L −3L2 
m=   (33)
420  156 −22L
symm. 4L2

2.3.3 Torsion

Torsional deformation is defined through the study of the rotation along the

element of a uniform rod of length L, mass density ρ, polar moment of inertia Ip ,

and torsional stiffness GJ. The strain V and kinetic energies T for pure torsion

are

Z L
1
V= GJ(θ0 )2 dx (34)
2 0

Z L
1
T = ρIp (θ̇)2 dx (35)
2 0

The rotation θ(x, t) along the element is given by the assumed-modes form where

the torsion members are represented by θ1 (t) and θ2 (t) and shape functions ψ1 (x)

and ψ2 (x) is

15
 θ(x, t) = ψ1 (x)θ1 (t) + ψ2 (x)θ2 (t) (36)

The shape functions ψ1 and ψ2 must satisfy the boundary conditions. The

shape functions are derived by considering a torsion member in equilibrium but

under static end torques. Combining Eqs. 34, 35, and 36 leads to the development

of the mass, stiffness, and force expressions shown below

Z L
kij = GJψi0 ψj0 dx (37)
0

Z L
mij = ρIp ψi ψj dx (38)
0

Z L
pi (t) = tθ (x, t)ψi dx (39)
0

Through the incorporation of shape functions, the stiffness matrix and mass matrix

for uniform torsion can be determined for an element by

 
GJ 1 −1
k= (40)
L −1 1

 
ρIp L 2 1
m= (41)
6 1 2

2.3.4 Three Dimensional Frame Element

The stiffness and mass matrices for a three-dimensional frame element can

be obtained from the axial, bending, and torsion elements. Bernoulli-Euler beam

theory was used in deriving the bending elements. A three dimensional frame

element, (Figure 4), shows the local references and the displacement coordinates

ui . The x axis follows along the centroids of the beam, while the y and z axis are

principal to the cross section.

16
 Figure 4. Notation for a three dimensional frame element

The area moments of inertia for the cross section are Iy and Iz , while Ip

is the area polar moment of inertia about the x-axis. The stiffness and mass

coefficients for u1 and u7 are associated with axial displacements, Eqs. 29 and 30.

Displacements u2 , u6 , u8 , and u12 are based on bending in the x-y plane, Eqs 32

and 33. While displacements u9 , u11 , u3 , and u5 are for bending in the x-z plane.

Finally, the coefficients associated with torsional degrees of freedom, u4 and u10

are obtained from Eqs. 40 and 41. Combining axial, transverse, and torsion, a

three dimensional stiffness matrix and mass matrix for a uniform three-dimensional

element can be determined

17
k=
−EA
 EA

L L
 12EIz 6EIz −12EIz 6EIz 
 L3 L2 L3 L2 
 12EIy −6EIy −12EIy −6EIy 
 L3 L2 L3 L2 
 GJ −GJ 
 L L 
 −6EIy 4EIy 6EIy 2EIy 
 L2 L L2 L 
 6EIz 4EIz −6EIz 2EIz 

 −EA L2 L L2 L 
EA 
 L L 
 −12EIz −6EIz 12EIz −6EIz 
 L3 L2 L3 L2 
 −12EIy 6EIy 12EIy 6EIy 
L3 L2 L3 L2
 

 −GJ GJ 

L L
 −6EIy 2EIy 6EIy 4EIy 
L2 L2
 L L

6EIz 2EIz −6EIz 4EIz
L2 L L2 L

(42)

ρAL
m=
420
 
140 70

 156 22L 54 −13L

 156 −22L 54 13L 

 140Ip 70Ip 

 A A 

 −22L 4L2 −13L −3L2 


 22L 4L2 −13L 2
−3L 
 70 140 
 

 54 13L 156 −22L


 54 −13L 156 22L 

 70Ip 140Ip 
 A A 
 13L −3L2 22L 4L2 
−13L −3L2 −22L 4L2
(43)

2.4 Vibration Properties of MDOF Systems

Mass matrix M and Stiffness matrix K are related to strain energy and kinetic

energy, respectively. M and K are positive definite matrices for most structures.

This implies that any arbitrary displacement of a system with a positive definite

K strain energy will be positive and that any arbitrary velocity displacement of a

18
system with positive definite M will result in a positive kinetic energy.

2.4.1 Eigensolution of Undamped Free-Vibration

The free vibration equation of motion for a multiple-degree-of-freedom system

is written in symbolic matrix form

Mü + Ku = 0 (44)

M and K are N × N matrices and u (t) is the corresponding N × 1 vector of

physical or generalized displacement coordinates. When harmonic motion defined

by displacement U and natural frequency ω

u(t) = U cos(ωt − α) (45)

When harmonic motion is then incorporated into Eq. 44 the N th order eigen value

problem is formed

[K − ω 2 M]U = 0 (46)

For the solution to be nontrivial, the determinant must equal zero

det(K − ω 2 M) = 0 (47)

This is known as a characteristic equation. When the determinant is expanded,

the corresponding values for the squared natural frequencies, ωn2 , are determined

from the roots of the polynomial. Corresponding to each eigenvalue, ωn2 , there will

be an eigenvector, or natural mode Un .

2.4.2 Eigensolution of Damped MDOF Systems

When a damping component is incorporated into the undamped N -degree-of-

freedom matrix equation of motion, Eq. 44, through the inclusion of the viscous

dampening component C then the equation of motion becomes

19
 Mü + Cu̇ + Ku = p(t) (48)

A procedure for determining the natural modes, ωn2 , requires the transforma-

tion of the N second-order equations of motion into a 2N first-order equations.

The generalized state-space form is developed when the displacement vector is

transformed into the state vector

Aż + Bz = F(t) (49)

Which is comprised of the displacement vector and an effective velocity vector

 
u(t)
z(t) ≡ (50)
v(t)

The effective velocity vector v could also be defined as u̇. Where A and B are the

two 2N × 2N constant coefficient matrices and F is the 2N state forcing vector

shown below

 
C M
A= (51)
M 0

 
K 0
B= (52)
0 −M

 
p(t)
F= (53)
0
Through the use of the generalized state space equation of motion, Eq. 49, it

is possible to solve for the eigen vectors and values for systems with general viscous

damping. Through the substitution of A and B into Eq. 44 it becomes

Aż + Bz = 0 (54)

20
 Since this is a homogeneous set of ordinary differential equations with constant

coefficients, the solution is comprised of the eigen values, λ, and the eigen vectors,

 
λt θu
z(t) = θe = eλt (55)
λθu
Through the division of Eq. 55 by eλt the generalized algebraic eigenvalue equation

is obtained by

[λA − B]θ = 0 (56)

The solution for both the eigenvectors and eigenvalues consist of i = 1, 2, ..., 2N

and must satisfy the characteristic equation

det(λA − B)θ = 0 (57)

2.5 Response of Systems to Periodic Excitation

2.5.1 Complex Fourier Series

A common and reliable method used in structural analysis to separate the

harmonic components found in these excitations is the Fourier series

∞
X
p(t) = P̄n ei(nΩ1 t) (58)
n=−∞

The fundamental frequency Ω1 (rad/s) is related to the period of the function

by Ω1 T1 = 2π. The bar over a parameter symbolizes that the coefficients may be

complex. When the excitation is not periodic, it can be represented by a Fourier

integral. In developing expressions for the Fourier integral transform pair, it is

convenient to use complex notation by allowing the period, T1 , approach infinity.

Determining the structural response requires the multiplication of the excitation

and the impulse response in the frequency domain and then performing the inverse

Fourier transform to return to the time domain (Figure 5).

21
 Figure 5. Fourier Transform Process

The complex Fourier expansion of p(t) is achieved by multiplying Eq. 58 by

ei(mΩ1 t) and integrating over one period

Z τ +T1
1
P̄n = p(t)e−i(nΩ1 t) dt, n = 0, ±1, ±2, .... (59)
T1 τ

The complex conjugate pair of P̄n is represented by P̄−n = P̄n∗ and that the

average value of p(t) can be found through

Z t+T1
1
Po = p(t)dt (60)
T1 t

2.5.2 Complex Frequency Response Function

The steady state response, ū(t) shown below is in complex form

ū(t) = Ū (Ω)eiΩt = H̄po eiΩt (61)

The frequency response H̄ is defined by Ω the forcing frequency and ωn the

undamped natural frequency of the system is

1
k
H̄(Ω) ≡ H̄ up (Ω) = 2 (62)
[1 − ( ωΩn )] + i[2ζ( ωΩn )]

When the excitation is periodic as shown in Eq. 58, the steady state response

can be written in complex notation as

∞
X
u(t) = Ūn ei(nΩ1 t) (63)
n=−∞

22
 The corresponding expression for periodic response becomes

Ūn = H̄n P̄n (64)

The process for determining the response through the use of the Fourier series is

shown in Figure 5. The method utilizes Eq. 59 to convert the forcing excitation

such that it relates the magnitudes to their respective frequencies. Eq. 64 is used

to couple the harmonic excitation P̄n and the frequency response H̄n .

2.5.3 Fourier Integral Transform

The Fourier coefficient P̄n in relation to p(t) was shown in Eq. 59, providing

that the integral exists. If you allow T1 → ∞ then it is convenient to allow the
2π
following notation: Ω1 = ∆Ω, Ωn = nΩ1 , P̄ (Ωn ) = T1 P̄n = P̄
∆Ω n
then Eq. 58

becomes

∞
1 X
p(t) = P̄ (Ωn )ei(Ωn t) ∆Ω (65)
2π n=−∞
Through the same notation the inverse is

Z T1
2
P̄ (Ωn ) = p(t)e−iΩn t dt (66)
−T1
2

The limits of integration will include the entire time history of p(t) when

T1 → ∞. As T1 → ∞, Ωn can be represented as a continuous variable Ω and

∆Ω becomes the differential notation dΩ. Thus allowing Eqs. 65 and 66 to be

represented respectively by

Z ∞
1
p(t) = P̄ (Ω)ei(Ωt) dΩ (67)
2π −∞

Z ∞
P̄ (Ω) = p(t)e−i(Ωt) dt (68)
−∞

23
 These two equations are the Fourier transform pairs. P̄ (Ω) is known as the

Fourier transform of p(t) and p(t) is called the inverse Fourier transform of P̄ (Ω).

Finally, the Fourier Transform pair can be written in terms of frequency, when
Ω
the relationship f = 2π
. Then Eqs. 67 and 68 can be written to illustrate the

transform to and from the frequency domain

Z ∞
−1
p(t) ≡ F [P̄ (f )] = P̄ (f )ei(2πf t) df (69)
−∞

Z ∞
P̄ (f ) ≡ F[p(t)] = p(t)e−i(2πf t) dt (70)
−∞

2.5.4 Frequency Response Function

The periodic excitation can be expressed as Eq. 63 and the relationship be-

tween the excitation, the periodic response, and the system response can be ex-

pressed as Eq. 64. However, if the relationships used to convert the Fourier

transform for a periodic excitation function are incorporated into the response,

then we obtain the Fourier transform pair for u(t) and Ū (f )

Z ∞
u(t) = Ū (f )ei((2πf t) df (71)
−∞

Z ∞
Ū (f ) = u(t)e−i(2πf t) dt (72)
−∞

The relationship Eq. 64 thus becomes

Ū (f ) = H̄(f )P̄ (f ) (73)

Which is the product of the system frequency response function H̄(f ) and the

Fourier transform of the excitation P̄ (f ).

24
 Therefore, the response can be expressed as the following inverse Fourier trans-

form
Z ∞
u(t) = H̄(f )P̄ (f )ei(2πf t) df (74)
−∞

2.5.5 System Response Function

The system response function, H̄(f ) can be obtained from the Fourier trans-

forms of the measured time histories p(t) and response u(t). The system parameters

for natural frequency ωn and the damping factor ζ can be extracted from the sys-

tem frequency response function. However, other analytical methods are available

for estimating the system response if either p(t) or u(t) is unknown, such as the

mode-superposition method.

2.5.6 Mode-Superposition for Undamped Systems

The mode-superposition method is used to obtain the response u(t) through

the use of the natural frequencies and natural modes of the system. The natural

frequencies and modes must satisfy the algebraic eigen problem, shown earlier in

Eq. 57. Given the natural frequency, ωr2 , and the modes, φr , for r = 1, 2, ..., N ,

the modal masses, Mr , and modal stiffness, Kr are calculated through

Mr = φTr Mφr (75)

Kr = φTr Kφr = ωr2 Mr (76)

Any natural frequencies that are redundant or repeated have been orthogo-

nalized by

φTr Mφs = φTr Kφs = 0 (77)

This is satisfied when r 6= s. The modes are then gathered to create the modal

matrix, Φ = [φ1 φ2 ... φN ]. Using the modal matrix we can rewrite Eqs. 75 and

25
76 for the modal mass and modal stiffness to be

M = ΦT MΦ (78)

K = ΦT KΦ (79)

The key step of the mode-superposition method is the coordinate transform

shown by
N
X
u(t) = Φη(t) = φr ηr (t) (80)
r=1

The coordinates ηr (t) is referred to as a principal coordinates or modal coordinates

and can be defined by

Mη̈ + Cη̇ + Kη = ΦT p(t) (81)

where:

The generalized modal mass matrix is

M = ΦT MΦ (82)

The generalized modal damping matrix is

C = ΦT CΦ (83)

The generalized modal stiffness matrix is

K = ΦT KΦ (84)

The generalized modal force vector is

ΦT p(t) (85)

26
2.5.7 Mode-Superposition for Damped MDOF Systems

When a viscous damped system satisfies the orthoganality requirements for

Eq. 81, then the uncoupled modal equation of motion can be written as

1 1 T
η̈r + 2ζωr η̇r + ωr2 ηr = fr (t) = φ p(t) for r = 1, 2, ..., N (86)
Mr Mr r
If the system is under harmonic excitation, p(t) = P cos(Ωt), then the mode-

superposition solution for the steady-state response becomes

1
η̈r + 2ζr η̇r + ωr2 ηr = Fr cos(Ωt) (87)
Mr
Where Fr = φTr P. Using the complex-frequency response techniques shown earlier,

Eq. 87 can be defined as

Fr iΩt
η̄¨r + 2ζr ωr η̄˙ r + ωr2 η̄r = ωr2 e (88)
Kr
The steady-state solution for η̄r becomes
1/Kr
H̄ηr /Fr (Ω) = (89)
(1 − rr2 )
+ i(2ζr rr )
Ω
Where rr is the modal frequency ratio, rr = ωr
. The magnitude and phase of Eq.

89 can be found through

Fr /Kr
ηr (t) = p cos(Ωt − αr ) (90)
(1 − rr2 )2 + (2ζr rr )2

2ζr rr
tan(αr ) = (91)
1 − rr2
The complex frequency response in generalized coordinates can be obtained by

writing ū(t) in the complex form

N
X
ū(t) = Φη̄(t) = φr η̄r (t) (92)
r=1

27
 Combining the above equations a solution for ū(t) is shown in to be

N
X φr φT P r 1
ū(t) = eiΩt (93)
r=1
Kr (1 − rr2 ) + i(2ζr rr )

The complex frequency response function in physical coordinates, H̄ij (Ω) gives

the response at coordinate ui due to harmonic excitation at pj

N
X φir φjr 1
H̄ij (Ω) ≡ H̄ui /pj (Ω) = (94)
r=1
Kr (1 − rr2 ) + i(2ζr rr )
Where φir is the element in row i of the r-th mode φr , that is the element in

row i and column r of the modal matrix Φ. The steady-state response u(t) can

then be obtained from

N
X φr φT P r 1
u(t) = cos(Ωt − αr ) (95)
r=1
Kr (1 − rr2 ) + i(2ζr rr )

2.5.8 Frequency Response of MDOF Systems based on Complex

Modes

The mode-superposition method also applies when a state space form which is

used to transform N second order differential equations to 2N first order differential

equations. The complex modes from the state-space problem have the following

form

   
θu θu
θr ≡ = (96)
θv r λθu r
λr is the complex scalar eigenvalue and θr is the corresponding 2N state

eigenvector. The subscripts u and v reference the displacement and velocity, re-

spectively. The eigenvalue, λr can be written in the following forms

p
λr = µr + jνr = −ηr ωr + iωr 1 − ζr2 = −ζr ωr + iωdr (97)

28
 The natural frequency, ωr , and the dampening factor, ζr are given by

p
ωr = µ2r + νr2 (98)

−µr
ζr = (99)
ωr
The modal matrix containing the state eigenvectors is

Θ = [θ1 θ2 ... θ2N ] (100)

The orthogonality equations can be combined with definitions of diagonal

terms modal ar and modal br

ΘT AΘ = diag(ar ) (101)

ΘT BΘ = diag(br ) (102)

−br
Such that λr = ar
. The harmonic excitation at the forcing frequency, ω, will

have the steady state response z = Zeiωt . Where Z can be expressed in the mode-

superposition form

2N
θrT Pθr
  X
U
Z≡ = (103)
iωU αr (iω − λr )
r=1
The displacement partition can be extracted by

2N
 T 
X θu r p θu r
U= (104)
r=1
αr (iω − λr )
The complex receptance frequency response function is

2N
X (θui )r (θuj )r
Hui /pj (ω) = (105)
α (iω − λr )
r=1 r

29
(θui )r is the i-th element in the displacement partition of the r-th eigenvector. Since

the eigenvectors and eigenvalues occur in complex conjugate pairs, Eq. 105 can be

written as

2N
X (θui )r (θuj )r (θui )∗ (θuj )∗r
Hui /pj (ω) = + ∗ r (106)
α (iω − λr ) αr (iω − λ∗r )
r=1 r

2.6 Abaqus

Abaqus is a commercial finite element analysis software application that can

be purchased from Dassault Systemes. Abaqus provides product simulation to

engineers in order to vary and simulate various design attributes. The Abaqus

Unified FEA product suite offers powerful and complete solutions for both rou-

tine and sophisticated engineering problems covering a vast spectrum of industrial

applications. Abaqus allows engineering working groups to consider full loads,

dynamic vibration, multibody stems, impact/crash, nonlinear static, thermal cou-

pling, and acoustic-structural coupling using a common model data structure and

integrated solver technology. Abaqus provides four analaysis packages to the user.

They include:

1. Abaqus/Standard is the general purpose finite element module.

2. Abaqus/Explicit is an explicit dynamics finite element module.

3. Abaqus/CFD is for modeling Computational Fluid Dynamics for offshore

fluid-structure loading

4. Abaqus/Electromagnetic provides a software application which solved ad-

vanced computational electromagnetic problems.

The user can gain access to these analysis tools through Abaqus/CAE. CAE

stands for Complete ABAQUS Environment and provides the modeling, managing,

30
and monitoring for performing analysis and visualization of results. The CAE user

interface allows for the creation of application specific systems through a graphical

user interface and dialog boxes.

Abaqus utilizes three steps to develop, perform, and present the results for

each finite element model. The first step is the pre-processing or modeling stage.

This involves the creation of an input file which contains the design of the system,

the system parameters, and the analysis to be performed. The second step is the

processing of the model which is performed behind the scenes. The final step

is post-processing where a report, image, animation, data table, etc is provided

from an output file. Abaqus CAE provides the capability for pre-processing, post-

processing, and monitoring of the processing stage.

2.6.1 Abaqus Spring-Dashpot Commands

Abaqus provides a graphical user interface for the design and evaluation of

mechanical systems. These systems are typically defined by the component(s) be-

ing drawn or integrated and their specific material properties. However, Abaqus

does provide the capability to evaluate mass-spring-dashpot systems through the

use of specific definitions and commands. In order to incorporate spring and dash-

pots to a system these specific functions must be called. In the example provided

in this paper, these commands are implemented behind the scenes in the Abaqus

input file, (Figure 6).

31
Figure 6. Abaqus Representation of the Six Degree of Freedom Lumped Parameter
Model

A zoomed in view of the spring-dashpot implementation is shown, (Figure 7).

32
 Figure 7. Zoomed in view of the Abaqus Spring-Dashpot Model

The specific commands used to generate this model are:

1. ∗M ass allows for the introduction of a concentrated mass at a point and the

ability to associate with the three translational degrees of freedom at a node.

2. ∗Spring allows for a spring element to be defined. This spring can couple a

force with a relative displacement.

3. ∗Dashpot allows for a dashpot element to be defined. This dashpot can

couple a force with a relative velocity.

4. ∗RotaryInertia allows for rotary inertia to be included at a node. This

inertia can be associated with the three rotational degrees of freedom at a

node and can be paired with a mass element.

2.6.2 Three Dimensional Frame Element

Abaqus provides a complex simulation environment for the design of struc-

tures. Through varying specific attributes of the structure an engineer can de-

termine the most feasible option for their client through material selection and

the method used for joining linkages. In the example provided in this paper a

33
3-Dimensional Framed Structure is evaluated, (Figure 8).

Figure 8. Abaqus Representation of a Five Story Framed Structure

Similar to the mass-spring-dashpot system, an input file was generated using

the required commands. This model is shown as a simple wireframe, due to the

fact that there was no need to describe the beam selection. If a pipe, I-beam,

or a square beam were used the model could be updated to depict this design

parameter. Frame elements are 2-node slender beam elements intended for use

in the elastic or elastic-plastic fatigue analysis of frame-like structures. These

elements have elastic responses that follow Euler-Bernoulli beam theory with the

plastic responses concentrated at the element end. They can be used in static,

implicit dynamic, and eigen frequency extraction analyses only.

The specific commands used to generate this model are:

1. ∗N ode allows for the user to define the nodes by directly specifying its coor-

dinates. Nodal coordinates are given in a local system.

34
 2. ∗Element allows for the definition of an element by specifying its nodes.

3. ∗F rameSection allows for the definition of the cross-section for frame ele-

ments.

2.6.3 Analysis

Abaqus provides an extensive suite of analysis tools to the user to evaluate

the system being considered. For the examples utilized in this paper the analysis

focused on finding the natural frequencies. The command used to perform this task

is the ∗f requency. This command tasks Abaqus to perform eigenvalue extraction

to calculate the natural frequencies and the corresponding mode shapes of a system.

The analysis will include initial stress and load stiffness effects due to preloads and

initial conditions if geometric nonlinearity is accounted for in the base state.

2.6.4 Output

Abaqus has several reporting features which can be used to extract solutions

for the model, system, or at the declared nodes. The ∗Output command is used

to write contact, element, energy, nodal, or diagnostic output. There are various

commands that can be used in parallel to efficiently extract the desired parameters.

The specific output commands that are available are:

1. ∗Contact Output allows for the extraction of variables associated with

surfaces in contact, coupled temperature-displacement, coupled thermal-

electrical, and crack propagation problems.

2. ∗Element Output allows the request of element variables such as stresses,

strains, section forces, element energies, etc.

3. ∗Energy Output allows for the extraction of the total energy of the model or

of a specific element.

35
 4. ∗Modal Output allows for the output in generalized coordinates the modal

amplitude and phase values.

5. ∗Node Output allows for the displacements, reaction forces, etc. that de-

scribe the nodal variables.

6. ∗Radiation Output allows for the extraction of cavity radiation variables.

2.7 Matlab State-Space Methods

MatLab provides several built in functions which allow for the modeling of

complex systems through the use of the appropriate mathematical models derived

from physical laws. Dynamic systems change in time per the set mathematical

model. The functions to be reviewed in this section are the ss and lsim functions.

When utilizing these MatLab functions for a mass-spring-dashpot system it is best

to construct a free body diagram and follow the procedures shown earlier.

2.7.1 MatLab Function ss

The MatLab function ss is used to create state-space models with real or

complex matrices. This function takes the following inputs:

1. a: a state matrix that is square real or complex with as many rows as states

such that it is X by X

2. b: an input state matrix that is real or complex with as many rows as states

and as many columns as inputs and its size is X by U

3. c: a real or complex state to output matrix that has as many rows as outputs

and as many columns as states and its size is Y by X

4. d: a real or complex feedthrough matrix that has as many rows as outputs

and as many columns as inputs and its size is Y by U

36
Where X is equal to the number of states, U is equal to the number of inputs, and

Y is equal to the number of outputs. For dynamic systems, governing equations

are determined through the summation of forces following Newtons second law.

A state-space representation of the system is then developed in order to reduce

the N -second order differential equations to 2N -first order differential equations.

Once the state space representation is gained, the information for a, b, and c is

available as an input. Since feedback is not being studied, d is set to zero.

2.7.2 MatLab Function lsim

The MatLab function lsim is used to simulate the time response of continuous

or discrete linear systems to arbitrary inputs. This function returns the system

response, sampled at the same rate as the time history. The output will have the

same number of elements as the time history and the same number of columns

as the system outputs. Lsim uses a linear interpolation (first-order hold) method

based on the smoothness of the signal u. This function takes the following inputs:

1. sys: dynamic system model, output of ss

2. u: input history

3. t: time history of the input

37
 CHAPTER 3

A Pole Residue Method for Dynamic Analysis of Systems

The numerical methods used to solve the system response of multiple-degree

of freedom systems primarily operate in the time or frequency domain. While more

complex problems used the Fourier transform or numerical integration procedures

in the time domain, the Laplace transform has been used for some simple problems

which require the use of look-up tables. This section will detail how to implement

a semi-numerical pole residue method to solve more complex problems.

3.1 System Response by the Laplace Transform

Transforms are used to simplify the solution of certain dynamics problems.

Their purpose is to shift a complex problem from one domain to another, allowing

a solution to be found more readily. Once the solution is obtained, these results

can be transferred back to the original domain. The Fourier transform allows the

problem to be shifted from the time domain to the frequency domain and back.

The Laplace transform shifts the problem from the time-domain to the s-domain.

The Laplace transform is particularly useful if the forward and inverse transform

can be identified in a lookup table. However, this is typically not the case with

the vast majority of engineering problems. In this section a new pole and residue

method is to be discussed that falls within the category of Laplace domain method.

Thus far, this method has proven to be applicable to arbitrary input functions, but

it has yet to be utilized in the determination of the system response of a structure.

This method focuses on obtaining the poles and residues of the response function.

Knowledge of these parameters will allow one to express the response function in

both the time and s domain. The following steps are required for this method:

1. Preparing the poles and residues of input and system transfer functions

38
 2. Conducting algebraic computation to obtain the poles and residues of the

response function based on those of the input and system transfer functions

3. Providing a solution in the time domain from the poles and residues of the

response

3.2 Utilizing a Prony’s Method for Developing the Poles and Residues

Structural excitations can be represented as an arbritrary signal y(t). This

signal can be decomposed into a finite number of exponential components

N
X
y(t) = α n e λn t (107)
n=1

In this equation, N is the number of terms, αn and λn are constants. When

excitation y(t) is a real-valued signal, λn must be real or form complex conju-

gate pair. When λn is equivalent to −δn + iΩn , then δn is the damping factor in

seconds−1 and Ωn represents damped frequency in radians per seconds. The coef-

ficients αn corresponds to the complex exponents, thus λn must appear in complex

conjugate pairs. If αn is equivalent to An eiθn , then An represents the amplitude,

and θn is the sinusoidal initial phase in radians in association with eλn t .

When digital signal analysis is performed the continuous y(t) its components

are not known and the captured signal is usually represented by equally spaced

samples. The Prony series corresponding to this discrete signal is

p
X
yk = γn znk (108)
n=1

A Prony series has its components being complex exponentials, which could be

purely harmonic, damped harmonic, or purely damped exponential components.

A Prony series is comprised of a smaller subset of terms, p, in comparison to the

sampled signal length, N . Most often p is much smaller than N . Finally, zn

39
must be estimated first from the discrete signal yk before computing the complex

coefficients γn .

3.3 Differential/Difference Equation for Prony Series

The arbitrary signal y(t) contains the exponenetial functions eλn t , for n =

1, 2, ...N which forms a basis on the open internval 0 ≤ t < T . Thus allowing the

Prony series, Eq. 107, to be viewed as the general p-th order homogenous linear

ordinary differential equation with constant real valued coefficents an

p
X
an y n (t) = 0, for 0≤t<T (109)
n=0

In Eq. 109, y n (t) = dn y(t)/dt, the exponents λn for n = 1, ..., N are the roots

of the characteristic equation

p
X
an λn = 0 (110)
n=0

If the sampled data, is equally spaced yk where k = 0, 1, ...N − 1 then Eq.

109 can be expressed as a pth-order difference equation where bn are real valued

constants

p
X
bn γk+n = 0 for k = 0, 1, ..., N − p − 1 (111)
n=0

Without losing generality by allowing bp = 1, and through the use of Eq. 108

and Eq. 111, the characteristic equation can then be obtained

p
X
bn z n = 0 (112)
n=0

40
3.3.1 Implemented Prony’s Method

A realization of A is possible through the substitution of the discrete signal

yk into the Hankel matrix

 
yk yk+1 · · · yk+η−1
 yk+1 yk+2 · · · yk+η 
H(k) =  (113)
 
.. .. .. .. 
 . . . . 
yk+ε−1 yk+ε · · · yk+ε+η−2

Upon the substitution of yk into H(k), it can be shown that H(k) is the

product of three matrices

H(k) = Pε Ak Qn (114)

Where

 

 cT  
 cT A 

 

Pε = cT A2 ∈ Rξxp (115)
··· 

 

 
 T ε−1 
c A


and

Qη = q0 Aq0 A2 q0 · · · Aη−1 q0
 
(116)

Through the introduction of weighting matrices W1 and W2 and applying

them to the Hankel matrix H(k), Eq. 117, an extension is performed mathemat-

ically to illustrate that the algorithm is applicable to intermittent or truncated

signals, Eq. 118.

H̃(k) = W1 H(k)W2 (117)

H̃(k) = P̃ξ Ak Q̃η (118)

41
 When all the elements of a particular row of W1 are zero and multiplied with

the corresponding row from H(k) the result amounts to deleting that corresponding

row. Similarly, when this process is performed with W2 . A realization of A can

be obtained from H(0) and H(1) by

H(0) = Pξ Qη (119)

H(1) = Pξ AQη (120)

Applying singular value decomposition of H(0) yields

 S1 0 VT1
  
= U1 S1 VT1

H(0) = U1 U2 (121)
0 0 VT2
The model order of the dynamic system, p, is equal to the number of non-zero

singular values. The singular values should go to zero when the rank of the matrix

is exceeded. However, when performed on measured data, the singular values will

not become zero due to random errors and inconsistencies and will instead become

very small. When choosing the size of H(0) both ξ and η must be greater than p.

A direct comparison between Eq. 119 and Eq. 121 suggests that Pξ is related to

U1 and Qη to VT1 . Substituting these relationships into Eq. 120 leads to

1 1
H(1) = U1 S12 AS12 VT1 (122)
1
Rearranging Eq. 122 and premultiplying by S12 UT1 and post multiplying by
−1
V1 S12 , yields to a realization of A

−1 −1
A = S12 UT1 H(1)V1 S12 (123)

The computed eigenvalues of A are zn , for n = 1, ..., p corresponding to λn .

42
Finally the complex coefficient γn , for n = 1, ..., p can be computed in the same

way as that used in the original third step of Prony’s method.

3.3.2 Prony’s Method to Laplace

α
It is evident that the exponential function αeλt and the pole-residue form s−λ

are a Laplace transform pair. Thus, after signal decomposition has been performed

using the modified Prony’s method, the Laplace transform yields the following

equation in the s-domain

L
X αl
ỹ(s) = (124)
l=1
s − λl

3.4 Integration of Pole and Residues with Structural Systems

In the previous sections, background information on the laplace transform and

the pole-residue method was presented. Through the coupling of this information

with material already presented in Chapter 2, another method capable of solving

complex structural problems will be presented.

3.4.1 Equation of Motion

The governing equations for a N -degree-of-freedom system is

Mẍ + Cẋ + Kx = f(t) (125)

This second order differential equation is comprised of matrices for M, C,

and K representing the mass, damping, and stiffness. These matrices are N × N .

The displacement vector, x(t), and the force vector, f(t) are N × 1. The initial

conditions are assumed to be at rest, such that the displacement, x(0), and the

velocity, ẋ(0), equal zero. This equation can then be transferred to the s-domain

resulting in

43
 (Ms2 + Cs + K)x̃(s) = f(s) (126)

Eq. 126 can then be re-arranged to represent the transfer function, Eq. 127,

where Z(s) is defined in Eq. 128 creating the transform from input f̃(s) to output

x̃(s), Eq. 129.

f̃(s) = Z(s)x̃(s) (127)

Z(s) = Ms2 + Cs + K (128)

x̃(s) = T(s)f̃(s) (129)

Through the annotation of T(s) with (j, k), one can determine the appropriate

transfer function for a response at coordinate, j, caused by the impact of an input

force at coordinate, k. There are two scenarios to be considered when deriving

the pole-residue form of Tjk (s). The first is when the problem is composed of a

symmetric M, C, and K matrices and the second is the asymmetric case.

3.4.2 Symmetric System Response

The majority of structural dynamic problems are composed of symmetric M,

C, and K. This is especially true when the matrices are developed through a finite

element procedure. In order to determine the system response, the N -second order

differential equations will need to be simplified into 2N -first order equations. This

can be performed through the procedure mentioned earlier in Section 2.4. Since M,

C, and K are symmetric, the state space model maintains this property through

coefficient matrices A and B. From Eq. 49, the generalized eigenvalue equation is

44
 [µA + B]θ = 0 (130)

The solution consists of 2N eigenvalues φn , which are real or occurs in complex

conjugate pairs
 
φn
θn = (131)
µn φn
The first N modes can be extracted from the first N elements of the eigenvector

θn . Since A and B are symmetric the orthogonality property, θrT Aθs = 0 and

θrT Bθs = 0 must hold true for µr 6= µs . Through the utilization of the modal

matrix Θ

Θ = [θ1 θ2 · · · θ2N ] (132)

The physical-modal coordinates transformation relationship is

z = Θy (133)

One can then convert, Eq. 49 into 2N uncoupled modal equations of motion

where an = θnT Aθn , bn = θnT Bθn , and gn = θnT g

an ẏn + bn yn = gn (134)

From Eq. 134 the transfer function of the n-th modal equation is found to be

1 1
Tn (s) = = (135)
an s + b n an (s − µn )
−bn
The pole, µn is defined as being equal to an
. One can express the transfer

function Tn (s) as Tjk (s) through the use of mode super-position as

2N
X βjk,n
Tjk (s) = (136)
n=1
s − µn

45
 Where the residue βjk,n is

θk,n θj,n
βjk,n = (137)
an
While poles are global parameters and are independent of the input and output

coordinates, j and k, the residues, βjk,n are dependent on j and k.

3.4.3 Asymmetric System Response

When M, C, and K, are not symmetric, another approach is required to

determine the Transfer Function, Tjk without the use of coordinate transformation.

An intermediate step is required to gain the zero-pole form, prior to obtaining the

pole-residue form Tjk (s). Since the Transform, Tjk (s), contains the inverse of Z,

the explicit formula for the inverse of a nonsingular square matrix, Z

 
C11 C21 · · · Cn1
1  C12 C22 · · · Cn2 
Z−1 = =  .. (138)
 
.. . . .. 
det(Z)  . . . . 
C1n C2n · · · Cnn
The cofactor, Cjk is equal to (−1)j+k det(Zjk (s)), where Zjk denotes the sub-

matrix obtained from Z by deleting the j-th row and k-th column from Z. From

Eq. 138 the (j, k) elements of T can be written as

det(Zkj (s))
Tjk (s) = (−1)j+k (139)
det(Z(s))
Note the subscript for T and Z are reversed. Since Eq. 128, is an N x

N matrix, for a non-singular matrix Z (s) the determinate must be a 2N -order

polynomial in the s-domain. The determinant of the non-singular matrix for Zjk (s)

must be a 2(N − 1) order polynmial in the s-domain. The zero-pole form for Tjk is

Q2(N −1)
n=1 (s − τjk,n )
Tjk (s) = κjk Q 2N
(140)
n=1 (s − µn )

46
 The remaining task is to compute poles, µn , the jk-dependent zeroes τjk,n and

leading coefficient κjk from M, C, and K. The poles, µn are computed by carrying

out eigen analysis of a state space model. Since M, C, and K are real-valued

matrices, the solution to the generalized eigenvalue problem yields 2N eigenvalues.

Computing the zeros, τjk,n follows the same procedure as computing the poles by

replacing M, C, and K by Mjk , Cjk , and Kjk , respectively. The coefficient for κjk

can be computed using Eq. 141, through the multiplicative rule of the determinants

Eqs. 142 and 143.

det(Mjk )
κjk = (−1)j+k (141)
det(M)

det(Z) = det(M) det(Is2 + M−1 Cs + M−1 K) (142)

det(Zkj ) = det(Mkj ) det(Is2 + M−1 −1

kj Ckj s + Mkj Kkj ) (143)

Once the zero-pole form has been obtained through Eq. 140, the corresponding

pole-residue form can be found to be

2N
X βjk,n
Tjk (s) = (144)
n=1
s − µn
The jk dependent residues of βjk,n is

βjk,n = lim (s − µn )Tjk (s) (145)

s→µn

3.4.4 Displacement Response

The displacement response at coordinate j to a force at k is given by the

convolution integral of the system and force

47
 Z t
xjk (t) = hjk (t − τ )fk (τ )dτ (146)
0

The total response at coordinate j produced by a force involving all compo-

nents of the load vector is obtained through the summation all load components

and thus capturing their individual contributions

N
X
xj (t) = xjk (t) (147)
k=1

The pole-residue analysis is similar to the frequency and analytical procedures

in that superposition is performed. The resulting s-domain x̃jk (s) response to

the displacement at the j-th coordinate due to a force at the k-th coordinate is

expressed as

x̃jk (s) = Tjk (s)f˜k (s) (148)

The total response at coordinate j produced by a general loading at all k can be

obtained to be

N
!
X
xj (t) = L−1 x̃jk (s) (149)
k=1

Eq. 149 forms the general solution to the coupled equation of motion, assum-

ing zero initial conditions. In order to successfully perform Laplace analysis, one

must be able to efficiently compute the pole-residue form of the response, x̃jk (s).

In order to implement Eq. 148 and compute the pole-residue of the response, f˜k (s)

and Tjk (s) can be expressed as

L
X αl
f˜k (s) = (150)
l=1
s − λl

2N
X βn
Tjk (s) = (151)
n=1
s − µn

48
 When Eqs. 150 and 151 is substituted into Eq. 148 the following form for

x̃jk (s) is obtained

2N
! L
!
X βn X αl
x̃jk (s) = (152)
n=1
s − µn l=1
s − λl
When Eq. 152 is transferred into the pole and residue form, it is evident that

there are a total of 2N + L poles resulting from the L excitation poles and 2N

system poles

2N +L
X γm
x̃jk (s) = (153)
m=1
s − νm
The first L poles correspond to the excitation poles such that νm = λm , where

m = 1, 2, ..., L. The remaining 2N poles corresponding to the system poles such

that, νm+L = µm , where m = 1, 2, ..., 2N . The corresponding residue can be found

for each response pole, νm by

γm = lim (s − νm )x̃jk (s) = lim (s − νm )f˜k (s)Tjk (s) (154)

s→νm s→νn

Through Eq. 154, the residues corresponding to the first L response poles can

be found using

L
!
X αl
γm = lim (s − νm ) Tjk (s) = αm Tjk (λm ) for m = 1, ..., L (155)
s→νm
l=1
s − λl

The residues corresponding to the last 2N response poles can be found through

2N
!
X βn
γm+L = lim (s − νm+L )f˜k (s) = βm f˜k (µm ) for m = 1, ..., 2N
s→νm+L
n=1
s − µn
(156)

49
 CHAPTER 4

Six Degree of Freedom Problem

4.1 Dynamical Solution

The problem being examined in this chapter was found in the Journal of Me-

chanical Systems and Signal Processing titled, ”Parametric Time-Domain Methods

for the Identification of Vibrating Structures - A Critical Comparison and Assess-

ment”. The authors, K. A. Petsounis and S.D. Fassois reviewed four stochastic

and three deterministic methods for the parametric time-domain identification of

vibrating structures and random excitation and noise-corrupted response signals.

The paper analyzed a real world six degree of freedom problem that is comprised

of two closely spaced modes, two weak modes, and a wide range of modal damping.

The authors determined that the sensitivity of implemented methods was a direct

result of the model order, design parameters, and user expertise.

4.2 Problem Overview

The structure presented is a simplified model of one-half of a railway vehicle,

(Figure 9). The railway car is represented as a rigid body with two degrees of

freedom, vertical displacement u3 and pitch angle u4 . The railway car is connected

to two nodes located at the front and rear end of the car representing the secondary

suspension. These nodes are modeled as a rigid body with a single degree of

freedom, comprised of a vertical displacement u2 or u4 . The secondary suspension

nodes are connected through the primary suspension to a wheelset modelled as an

unsprung mass with a single degree of freedom, comprised of a vertical displacement

u1 or u6 . The interaction between the wheel set and the track is modelled by

linearized springs. The track is assumed to be fixed and rigid. As one-half of a

railway vehicle the values for the system parameters are nearly symmetrical with

50
respect to the vertical axis through the car’s center of mass.

Figure 9. Simplified 6 Degree of Freedom Model of One-Half of a Railway Vehicle

Table 1. Railway Vehicle Parameters and Values

Property Symbol Value Units
Vehicle body mass B 4125 kg
Vehicle Body Inertia Ib 1.25 ∗ 105 kg ∗ m2
Leading/Trailing Element Mass M2 /M4 850 kg
Leading/Trailing Wheelset Mass M1 /M6 1120 kg
Leading/Trailing Secondary Suspension Stiffness k3 /k4 6 ∗ 106 N/m
Leading/Trailing Secondary Suspension Damping c3 /c4 1.8 ∗ 106 N s/m
Leading/Trailing Primary Suspension Stiffness k2 /k5 106 N/m
Leading/Trailing Primary Suspension Damping c2 /c5 0.6 ∗ 104 N s/m
Leading/Trailing Hertzian Contact Spring k1 /k6 3 ∗ 107 N/m
Leading/Trailing Ground Contact Dashpot c1 /c6 0 N s/m
Distance Between Center of Rail Cart and Bogies L 25.6/3 m

One solution to this problem utilizes the assumed modes method to develop

the Newtonian equations representing the mathematical model. The first step

is to develop the free body diagrams of the system, followed by writing the basic

equation of motion for each free body diagram. Then use of the equations of motion

to relate the elastic forces to displacements and the dampening forces to velocities.

Through the combination of the related forces to their respective displacement

51
and velocities, a mathematical model can be written in matrix notation. The

mode shapes, modal frequencies, and modal dampings, can be determined after

transitioning the problem to a state space model.

4.2.1 Free Body Diagram

A free body diagram is developed utilizing the process depicted in Section 2.2.

Through the completion of this activity the equation of motion for each particle can

then defined, thus allowing for the creation of the mathematical model representing

this problem, (Figure 10).

Figure 10. Free Body Diagram for the One-Half of a Railway Vehicle

4.2.2 Equation of Motions

Now that the free body diagram for each element has been drawn and anno-

tated, we can now define the equation of motions as shown below.

X
↑ F1 = m1 u¨1 = Fs2 + Fc2 − Fs1 − Fc1 (157)

X
↑ F2 = m2 u¨2 = Fs3 + Fc3 − Fs2 − Fc2 (158)

X
↑ F3 = m3 u¨3 = −Fs3 − Fc3 − Fs4 − Fc4 (159)

52
 X
y F4 = m4 u¨4 = −Fs3 L − Fc3 L + Fs4 L + Fc4 L (160)

X
↑ F5 = m5 u¨5 = +Fs4 + Fc4 − Fs5 − Fc5 (161)

X
↑ F6 = m6 u¨6 = +Fs5 + Fc5 − Fs6 − Fc6 (162)

4.2.3 Relation to Displacements and Velocities

As mentioned earlier in Section 2.2 the mass element can be easily plugged

into the equations above. However, the linear elastic spring and viscous damping

forces still need to be related to the displacements ui for each mass element. This

relationship is illustrated in the equations below for the six elastic spring forces.

Fs1 = k1 u1 (163)

Fs2 = k2 (u2 − u1 ) (164)

Fs3 = k3 (u3 − u2 + Lu4 ) (165)

Fs4 = k4 (u3 − u5 − Lu4 ) (166)

Fs5 = k5 (u5 − u6 ) (167)

Fs6 = k6 u6 (168)

53
 This relationship is illustrated in the equations below for the six viscous damp-

ing forces.

Fc1 = c1 u˙1 (169)

Fc2 = c2 (u˙2 − u˙1 ) (170)

Fc3 = c3 (u˙3 − u˙2 + Lu˙4 ) (171)

Fc4 = c4 (u˙3 − u˙5 − Lu˙4 ) (172)

Fc5 = c5 (u˙5 − u˙6 ) (173)

Fc6 = c6 u˙6 (174)

Now that we have defined the relationship between the linear elastic spring

and viscous damping forces to the displacements, we can complete the equations

of motion and complete the mathematical model.

m1 u¨1 + (c1 + c2 )u˙1 − c2 u˙2 + (k1 + k2 )u1 − k2 u2 = 0 (175)

m2 u¨2 −c2 u˙1 +(c2 +c3 )u˙2 −c3 u˙3 −c3 Lu˙4 −k2 u1 +(k2 +k3 )u2 −k3 u3 −k3 Lu4 = 0 (176)

m3 u¨3 − c3 u˙2 + (c3 + c4 )u˙3 + (c3 L − c4 L)u˙4 − c4 u˙5

(177)
−k3 u2 + (k3 + k4 )u3 + (k3 L − k4 L)u4 − k4 u5 = 0

54
 m4 u¨4 − c3 u˙2 + (c3 + c4 )u˙3 + (c3 − c4 )L2 u˙4 + c4 Lu˙5
(178)
−k3 u2 + (k3 + k4 )u3 + (k3 − k4 )L2 u4 + k4 Lu5 = 0

m5 u¨5 −c4 u˙3 +c4 Lu̇4 +(c4 +c5 )u̇5 −c5 u̇6 −k4 u3 +k4 Lu4 +(k4 +k5 )u5 −k5 u6 = 0 (179)

m6 u¨6 − c5 u˙5 + (c5 + c6 )u˙6 − k5 u5 + (k5 + k6 )u6 = 0 (180)

Through the incorporation of the above defined equations of motion in respect

to the displacements ui , the mathematical model for the multiple-degree-of-freedom

model can be expressed in matrix notation for mass M, stiffness K, damping C,

and excitation p.

 
m1 0 0 0 0 0
 0 m2 0 0 0 0
 
0 0 m3 0 0 0 
M=
0
 (181)
 0 0 m4 0 0
0 0 0 0 m5 0 
0 0 0 0 0 m6

 
c1 + c2 −c2 0 0 0 0
 −c2 c2 + c3 −c3 −c3 L 0 0 
 
 0 −c 3 c 3 + c 4 c 3 L − c 4 L −c 4 0 
C= 2 2
 (182)
 0
 −c3 L c3 L − c4 L c3 L + c4 L c4 L 0  
 0 0 −c4 c4 L c4 + c5 −c5 
0 0 0 0 −c5 c5 + c6

 
k1 + k2 −k2 0 0 0 0
 −k2 k2 + k3 −k3 −k3 L 0 0 
 
 0 −k 3 k 3 + k4 k 3 L − k4 L −k 4 0 
K= 2 2
 (183)
 0
 −k3 L k3 L − k4 L k3 L + k4 L k4 L 0  
 0 0 −k4 k4 L k4 + k5 −k5 
0 0 0 0 −k5 k5 + k6

55
  

 0

0

 


 
 
0

p= (184)

 0

0

 


 
0
 

4.2.4 State Space Model

Through the use of the eigensolution of damped multiple-degree-of-freedom

systems, the modal frequencies and modal dampings were found through the use

of a generalized state space model.

 

 2.7874 

3.7105

 


 

16.5460
 
ω= (185)

 19.2725
25.3552

 


 

25.5660
 

4.3 Abaqus

In order to ensure that the modal frequencies that were calculated in Matlab

are correct, this six degree-of-freedom model was implemented in Abaqus and

analyzed through the command prompt. An input file representing the model was

developed and analysis was performed through the command prompt was launched.

 

 2.7866

3.7082

 


 

16.524
 
ω= (186)

 19.239

25.401

 

 

25.611
 

4.4 Impact and Measurement Locations

In order to analyze a multiple degree of freedom problem and determine the

response, the two key parameters are the input, p, and the output, q. In this

56
example the input location was chosen to be node 2 and the output location was

node 4. The displacement response is to be represented as x42 .

4.5 Excitation

An excitation, p(t) was developed utilizing the parameters for the frequencies

components, amplitudes, phase angle, and decay factor, (Table 6).

Table 2. Parameter Values for the Excitation p (t)

Amplitude An Decay Factor δn Frequency fn [Hz] Phase Angle θn
−π
1.2 -0.04 0.3 4
π
0.8 -0.030 0.22 8
0.2 0.003 0.0 0

The excitation was generated using

N
X
p(t) = An eδn t cos(2πfn + θn ) (187)
n=1

The three component signals are shown in red and the excitation is shown in black,

(Figure 11).

57
 Figure 11. Excitation p (t)

4.6 Frequency Domain

Analysis in the frequency domain was performed utilizing the fast Fourier

transform provided in Matlab. The fast fourier transform provides the ability to

shift the excitation, p(t), from the time-domain to the frequency domain. Then

one half of the complex frequency response was calculated utilizing the modes

computed earlier. In order to perform the multiplication of the these two complex

frequency coefficients, the mirrored image of the response needs to be filled in, in

order to maintain the symmetric properties of the fast fourier transform. Through

multiplication of these two complex frequency coefficients, the periodic response

has been calculated. The steady-state displacement can then be found by taking

the inverse fourier transform of the periodic response in order to translate the

solution from the frequency domain to the time domain, (Figure 12).

58
 Figure 12. Results utilizing the Frequency-Domain Method

4.7 MatLab Time-Domain Analysis

A time-domain method was performed using the built-in MatLab functions

ss and lsim. The function ss generates a state space representation of the system,

while lsim simulates the time response, (Figure 13)

59
 Figure 13. Results from the use of MatLab functions ss and lsim

4.8 Pole-Residue Method

The poles and residues components of the excitation were found using the

modified Prony’s method , (Table 3).

Table 3. Poles and Residues of the Excitation

Excitation
Poles Residues
0.0030 + 0.0000 0.2000 + 0.0000i
−0.0300 ± 1.2566i 0.3696 ± 0.1531i
−0.0400 ± 1.8850i 0.4342 ∓ 0.4242i

The poles and residues for the system transfer function were found following

the symmetric process. The poles and residues were found by constructing the

state space model from M, C, and K and then solving the generalized eigenvalue

problem, (Table 4).

60
 Table 4. Poles and Residues of the Transfer Function
System Transfer Function
Poles (102 ) Residues (10−9 )
−0.0282 ± 1.6061i −0.1678 ± 0.5031i
−0.0278 ± 1.5929i 0.0304 ± 0.0251i
−0.2274 ± 1.1892i 0.6031 ∓ 144.3058i
−01730 ± 1.0251i 0.0190 ∓ 0.0012i
−0.0154 ± 0.2326i −0.4841 ± 734.7525i
−0.0086 ± 0.1749 −0.0007 ∓ 0.0038i

Table 5. Poles and Residues of the Displacement Response

Output
Poles (102 ) Residues (10−9 )
−0.0282 ∓ 1.6061i −0.0000556 ∓ 0.0000199i
−0.0277 ∓ 1.5929i −0.0000028 ± 0.0000033i
−0.2274 ∓ 1.1892i 0.0208709 ± 0.0041965i
−0.1730 ∓ 1.0251i −0.0000004 ± 0.000003i
−0.0154 ∓ 0.2326i −0.562568 ∓ 0.053815i
−0.0086 ∓ 0.1749i 0.000000319 ∓ 0.00000042i
0.00003 + 0.0i 0.121138 − 0.0i
−0.0003 ∓ 0.0126i 0.2251826 ∓ 0.091387i
−0.0004 ∓ 0.0188i 0.256000 ± 0.261559i

4.8.1 Steady-State Response through the Poles and Residues

Once the poles and residues of the excitation and the transfer function are

known, then determining the poles and residues of the time history for ũ4 can be

accomplished, (Table 5).

From these corresponding poles and residues the time history u42 (t) can be

obtained, (Figure 14).

61
 Figure 14. Results from the Pole-Residue Method

4.9 Comparison of Results

The displacement response x4 due to the excitation at node 2 was found

through the use of a frequency-domain, time-domain, and pole-residues methods.

The displacement response from these three methods do agree, (Figure 15).

62
Figure 15. Calculated Displacement Response from the Pole-Residue, Time-
Domain, and Frequency-Domain Methods

63
 The time-domain and pole-residue methods agree strongly since both method

depict the steady-state portion of the response, but also the transient component.

The frequency-domain result is composed of only the steady-state solution.

4.10 Function Completion Times

The time to complete the execution of the above methods for the one-half of

a railway cart was determined through the use of the profile function in MatLab,

(Figure 16).

Figure 16. Time required to complete twenty iterations for the Pole-Residue,
Analytical, and Fourier Methods for the one-half of a railway vehicle

The algorithms were executed twenty times. The frequency-domain method

was the fastest and completed under 0.005 seconds. The time-domain method was

second with completion times just above the 0.005 seconds mark. The pole-residue

method was last with completion times between the 0.02 and 0.03 seconds.

64
 CHAPTER 5

Three Dimensional Frame Structure

The problem being examined in this chapter is a three dimensional frame

structure. This problem represents a real world multiple-degree-of-freedom sys-

tem depicting a five story building with four fixed points representing the earth-

structure interface. There are four fixed nodes, twenty nodes that are not fixed, and

forty linkages. This structure will be analyzed through the use of Bernoulli-Euler

beam theory.

5.1 Problem Overview

The structure is a simplified five story building, (Figure 17). The five story

building is represented as a three dimensional frame element with six degrees of

freedom. The building will suffer from axial motion, transverse motion, and tor-

sion. The axial motion will be defined by the structures cross sectional area A,

Elastic Modulus E, and mass density ρ. The transverse motion is governed by the

structure’s mass density ρ, elastic modulus E, cross sectional area A, and moment

of inertia I. This structure will undergo torsion which is governed by the shear

modulus of elasticity G, elasticity J, and polar moment of inertia Ip . The linkages

are modelled as a three dimensional frame element with 12 degree of freedoms, six

displacements, and six moments.

65
 Figure 17. Five Story Three Dimensional Frame Structure

5.2 Matlab Finite Element Analysis Solution

One solution to this problem is to implement the finite element method.

Through the representation of the structure as an assembly of elements the dis-

placement can then be found. In order to perform this task, the structure needs

to be defined in such a way that the following tasks could be accomplished:

1. The structure can be modeled

2. The parameters for each beam and the corresponding beam elements can be

defined

3. This finite element problem was implemented in Matlab utilizing the three

dimensional 12x12 matrices for mass and stiffness

4. The eigenvalues were found using the undamped free-vibration eigensolution

5.2.1 State Space Model

The M and K matrices were generated utilizing the finite element method.

The undamped eigensolution was then used to determine the natural frequencies

66
  

 43.42010862

58.82131769

 


 

76.11925025

 


 

144.7518675

 


 

184.4992609
 
ω= (188)

 239.0917646

265.4024803

 


 

282.0018361

 


 

325.2904703

 


 

329.0605147
 

5.3 Impact and Measurement Locations

In order to analyze a multiple-degree-of-freedom determine the response, the

two key parameters are the input, p, and the output, q. In this example the input

was chosen to be node 20 and the output location was node 14. These two nodes

were selected randomly. The displacement response is to be represeneted as x14 20 .

5.4 Excitation Signal

An excitation, p(t) was developed utilizing the parameters for the frequencies

components, amplitudes, phase angle, and decay factor, (Table 6).

Table 6. Parameter Values for the Excitation p (t)

Amplitude An Decay Factor δn Frequency fn [Hz] Phase Angle θn
−π
1.2 -0.04 0.3 4
π
0.8 -0.030 0.22 8
0.2 0.003 0.0 0

The excitation was generated by

N
X
p(t) = An eδn t cos(2πfn + θn ) (189)
n=1

The three component signals are shown in red and the encompassing excitation

signal is shown in black, (Figure 18).

67
 Figure 18. Excitation p (t)

5.5 Frequency-Domain Analysis

Analysis in the frequency domain was performed utilizing the fast fourier

transform provided in Matlab. The fast fourier transform provides the ability to

shift the excitation, p(t), from the time-domain to the frequency domain. Then

one half of the complex frequency response was calculated utilizing the modes

computed earlier. In order to perform the multiplication of the these two complex

frequency coefficients, the mirrored image of the response needs to be filled in, in

order to maintain the symmetric properties of the fast fourier transform. Through

multiplication of these two complex frequency coefficients, the periodic response

was calculated. The steady-state displacement can then be found by taking the

inverse fourier transform of the periodic response in order to translate the solution

from the frequency domain to the time domain, (Figure 19).

68
 Figure 19. Results from the Frequency-Domain Method

5.6 MatLab Time-Domain Analysis

A time-domain method was performed using the built-in MatLab functions

ss and lsim. The function ss generates a state space representation of the system,

while lsim simulates the time response, (Figure 20)

69
 Figure 20. Results from the use of MatLab functions ss and lsim

5.7 Pole-Residue Method

The poles and residues components of the excitation were found using the

modified prony’s method , (Table 7).

Table 7. Poles and Residues of the Excitation

Excitation Force
Poles Residues
0.0030 + 0.0000 0.2000 + 0.0000i
−0.0300 ± 1.2566i 0.3696 ± 0.1531i
−0.0400 ± 1.8850i 0.4342 ∓ 0.4242i

The poles and residues for the system transfer function were found following

the symmetric process. The poles and residues were found by constructing the

state space model from M, C, and K and then solving the generalized eigenvalue

problem, (Table 8).

70
 Table 8. Poles and Residues of the Transfer Function
Transfer Function
Poles (104 ) Residues (10−12 )
−0.104564 ± 1.041145i −0.00176 ∓ 943.3013i
−0.104713 ± 1.041882i −0.00092 ∓ 2239.9994i
−0.094454 ± 0.989796i −0.000011 ± 523.346i
−0.097087 ± 1.003432i −0.00010 ± 136.464i
−0.819287 ± 0.922063i 0.00027 ± 391.836i
−0.078647 ± 0.9034513i 0.00055 ± 696.428i
−0.077273 ± 0.895539i 0.00010 ∓ 1411.253i
−0.075774 ± 0.886826i 0.00027 ∓ 2435.710i

5.7.1 Steady-State Response through the Poles and Residues

Once the poles and residues of the excitation and the transfer function are

known, then determining the poles and residues of the time history for ũ14 can be

accomplished, (Table 9).

Table 9. Poles and Residues of the Displacement Response

Output
Poles (10−4 ) Residues (10−14 )
−0.1045 ∓ 1.0411i 1.6034 ± 0.1611i
−0.1047 ∓ 1.0418i 3.8048 ± 0.3826i
−0.0944 ∓ 0.9897i −0.9366 ∓ 0.0894i
−0.0970 ∓ 1.0034i −0.2408 ∓ 0.0233i
−0.0819 ∓ 0.9220i −0.7536 ∓ 0.0670i
−0.0786 ∓ 0.9034i −1.3676 ∓ 0.1191i
−0.0772 ∓ 0.8955i 2.7961 ± 0.2414i
−0.0757 ∓ 0.8868i 4.8741 ± 0.4168i

From these corresponding poles and residues the time history u14 (t) can be

obtained, (Figure 21).

71
 Figure 21. Results from the Pole-Residue Method

5.8 Comparison of Results

The displacement response x14 due to an excitation at node 20 was found

through the use of a frequency-domain, time-domain, and pole-residues methods.

The displacement response from these three methods do agree, (Figure 22).

72
Figure 22. Calculated Displacement Response from the Pole-Residue, Time-
Domain, and Frequency-Domain Methods

The time-domain and pole-residue methods agree strongly as they not only

depict the steady-state portion of the response, but also the transient component.

The frequency-domain results depict just the steady-state portion.

5.9 Function Completion Times

The time to complete the execution of the above methods for the five-story

building was determined through the use of the profile function in MatLab. The

program was implemented twenty times and the execution time was captured for

each method, (Figure 23).

73
Figure 23. Execution time for the Pole-Residue, Time-Domain, and Frequency-
Domain Methods

The time-domain method was the fastest with completion times just above the

0.05 seconds mark. The frequency-domain method was second and had completion

times between 0.15 and 0.20 seconds. The pole-residue method was last with

completion times between the 0.30 and 0.40 seconds.

74
 CHAPTER 6

Conclusions

The displacement response was calculated through the use of a pole-residue

method, frequency-domain method, and a time-domain method. There were two

structures studied in this thesis. The first structure was a six degree-of-freedom

system and the second was a one hundred and twenty degree-of-freedom system.

The calculated displacement response from these three methods were compared.

The first structure studied represented one half of a railway vehicle. The

railway car was represented as a rigid body with two degrees of freedom, vertical

displacement, and pitch angle. The railway car was connected to two nodes located

at the front and rear end of the car representing the secondary suspension. These

nodes were modelled as a rigid body with a single degree of freedom comprised of

a vertical displacement. The secondary suspension nodes were connected through

the primary suspension to a wheelset modelled as an unsprung mass with a single

degree of freedom, comprised of a vertical displacement. The interaction between

the wheel set and the track was modelled by linearized springs. The track was

assumed to be fixed and rigid.

The second structure studied was a five story building. The five story building

was represented as a three dimensional frame element with six degrees of freedom.

The building suffered from axial motion, transverse motion, and torsion. The axial

motion was defined by the structure’s cross sectional area, elastic modulus, and

mass density. The transverse motion was governed by the structure’s mass density,

elastic modulus, cross sectional area, and moment of inertia. This structure under-

went torsion which was governed by the shear modulus of elasticity, elasticity, and

polar moment of inertia. The linkages were modelled as a three dimensional frame

75
element with twelve degree of freedoms, six displacements, and six moments.

A frequency-domain method was implemented through the use of Fourier

transform. The excitation was transferred from the time domain to the frequency

domain through the fast Fourier transform. The complex frequency response was

determined from the system’s modes which were found by implementing a state-

space model to simplify the problem. The complex frequency response was mul-

tiplied with the complex periodic excitation, and then shifted back to the time-

domain through the inverse fourier transform. The resulting displacement response

was found to be steady-state and lacks the transient compnents.

A time-domain method was implemented utilizing the functions ss and lsim

found within MatLab. The ss command transferred the system into the required

state-space model. The lsim command simulated the displacement response of con-

tinuous or discrete linear systems to arbitrary inputs through linear interpolation.

The displacement response was calculated through the implementation of a

pole-residue method. The poles and residues associated with the excitation were

determined from the implementation of a modified Prony’s method. The poles and

residues for the system were found using eigen analysis of the state-space model.

These excitation and system poles and residues were then used to construct the

displacement response.

The different displacements from the frequency-domain, time-domain, and

pole-residue methods were compared. The displacement responses from these three

methods are shown in Figures 15 and 22. From the initial inspection, one can

say that three methods are in agreement. The frequency method provides just

the steady-state response, and does not inlcude the transient portion. The time-

domain and pole-residue method provided the complete displacement response

containing the steady-state and transient portion.

76
 The time required to execute the respective methods was accumulated over

a large number of executions in order to determine execution speed. This was

measured from the first step of rearranging the M, C, and K matrices to the

required state-space form to the calculation of the displacement response. The

execution time for each method to complete twenty iterations is shown in Figures

16 and 23. From the initial inspection, one can see that the frequency-domain

method is the quickest when operating within a simple model. The analytical

method was the fastest when operating on the more complex five-story building.

The pole-residue method was the slowest in both situations.

From the results provided in this paper, the built in analytical functions pro-

vided by MatLab performed the best. The pole-residue method provided the com-

plete response, however it was the slowest. The frequency-domain method was

the fastest for the simple six degree of freedom structure, but it provided just the

steady-state response.

There are still several opportunities available to demonstrate the capabilities

of this pole-residue method, especially since the work presented has been based

on theoretical structures. The pole-residue method is an input output function

and requires knowledge of the input and system. The utilization of this method

in signal processing may allow for gains in signal detection and isolation. The

inclusion of this algorithm in the detection of signals in the undersea environment

may allow for gains in the study of anthropogenic noise and the study of marine

mammals. The accuracy of the method may allow for the replacement of time-

domain methods and allow for the quick iteration through material choices.

77
 BIBLIOGRAPHY

Craig, R. R. and Kurdila, A. J., Fundamentals of structural dynamics. John Wiley

& Sons, 2006.

Cunha, A. and Caetano, E., “Experimental modal analysis of civil engineering

structures,” Sound and Vibration, vol. 40, no. 6, pp. 12–20, 2006.

Abaqus/CAE User’s Guide, Dassault Systemes, Providence, Rhode Island, United

States of America, 2013.

De Vivo, A., Brutti, C., and Leofanti, J., “Modal shape identification of large struc-
ture exposed to wind excitation by operational modal analysis technique,”
Mechanical Systems and Signal Processing, vol. 39, no. 1, pp. 195–206, 2013.

Hibbeler, R. C., Statics and Dynamics. Upper Saddle River, New Jersey, United
States of America: Pearson Prentice Hall, 2004.

Hu, S.-L. J., Liu, F., Gao, B., and Li, H., “Pole-residue method for numerical
dynamic analysis,” Journal of Engineering Mechanics, p. 04016045, 2016.

Hu, S.-L. J., Yang, W.-L., and Li, H.-J., “Signal decomposition and reconstruction
using complex exponential models,” Mechanical Systems and Signal Process-
ing, vol. 40, no. 2, pp. 421–438, 2013.

Kwon, Y. W. and Bang, H., The finite element method using MATLAB. CRC
press, 2000.

MATLAB, version 7.12.0 (R2011a). Natick, Massachusetts: The MathWorks

Inc., 2011.

Modak, S., “Separation of structural modes and harmonic frequencies in oper-

ational modal analysis using random decrement,” Mechanical Systems and
Signal Processing, vol. 41, no. 1, pp. 366–379, 2013.

Petsounis, K. and Fassois, S., “Parametric time-domain methods for the identifica-
tion of vibrating structuresa critical comparison and assessment,” Mechanical
Systems and Signal Processing, vol. 15, no. 6, pp. 1031–1060, 2001.

78