Vibrations - Analytical and Experimental Modal Analysis

+UC-SDRL-RJA
CN-20-263-662
UC-SDRL-CN-20-263-662
VIBRATIONS:
ANALYTICAL AND EXPERIMENTAL MODAL ANALYSIS
Dr. Randall J. Allemang, Professor

Structural Dynamics Research Laboratory
Department of Mechanical, Industrial and Nuclear Engineering
University of Cincinnati
Cincinnati, Ohio 45221-0072
February 1999
Approved for public release

Copyright 1990-1999 UC-SDRL
All Rights Reserved
-i-
Revision: February 19, 1999 +
+UC-SDRL-RJA
CN-20-263-662
PREFACE
An initial version of the following set of notes was originally prepared by Dave Formenti in 1977
for use in a short course at the University of Cincinnati on the subject of "Experimental Modal
Analysis". The notes have since been rewritten several times to standardize the notation and add
further clarification and additional topics by several authors including, R. J. Allemang, R. W.
Rost, T. J. Severyn, and J. M. Leuridan, for use in other short courses and the dual level
Mechanical Vibrations II (20-263-662) course at the University of Cincinnati. Any comments
relative to corrections or improvements will be welcomed.
PRINTING/REVISION HISTORY
In an attempt to improve and correct these notes, several revisions have been made. Please be
certain that you have the latest revision.
March 1990
Revision 1: February 1992
Revision 2: June 1992
Revision 3: July 1993
Revision 4: March 1994
Revision 5: February 1999
-iv-
+UC-SDRL-RJA
CN-20-263-662
NOMENCLATURE
Matrix Notation
{..}
T
braces enclose column vector expressions
[..] H
row vector expressions

brackets enclose matrix expressions
complex conjugate transpose, or Hermitian transpose, of a matrix
[..]T
transpose of a matrix
{..}
[..]
[..]
[..]+
[. . ]q p
..
inverse of a matrix
generalized inverse (pseudoinverse)
size of a matrix: q rows, p columns
diagonal matrix
Operator Notation
A*
F
complex conjugate
Fourier transform
F1
inverse Fourier transform

Hilbert transform
H
H1
ln
L
L1
Re + j Im
x
x
y
y
inverse Hilbert transform

natural logarithm
Laplace transform
inverse Laplace transform
complex number: real part "Re", imaginary part "Im"
first derivative with respect to time of dependent variable x
second derivative with respect to time of dependent variable x
mean value of y
estimated value of y
-v-
+UC-SDRL-RJA
CN-20-263-662
Ai B i
i=1
summation of Ai Bi from i = 1 to n
partial derivative with respect to independent variable "t"
det[. . ]
determinant of a matrix
||. . ||2
Euclidean norm
Roman Alphabet
A pqr
C
e
F
residue for response location p, reference location q, of mode r

damping
base e (2.71828...)
input force
Fq
[I ]
spectrum of q th reference
impulse response function
impulse response function for response location p, reference location q
transfer function
frequency response function, when no ambiguity exist, H is used instead
of H( )
frequency response function for response location p, reference location q,
when no ambiguity exist, H pq is used instead of H pq ( )
identity matrix
j
K
Kr
L
M
Mr
M Ar
M Br
1
stiffness
modal stiffness for mode r
modal participation factor
mass
modal mass for mode r
modal A for mode r
modal B for mode r
h(t)
h pq (t)
H(s)
H( )
H pq ( )
-vi-
+UC-SDRL-RJA
N
Ni
No
p
q
r
RI
RF
s
t
tk
T
x
X
Xp
z
CN-20-263-662
number of modes
number of references (inputs)
number of responses (outputs)
output, or response point (subscript)
input, or reference point (subscript)
mode number (subscript)
residual inertia
residual flexibility
Laplace domain variable
independent variable of time (sec)
discrete value of time (sec)
tk = k t
sample period
displacement in physical coordinates
response
spectrum of pth response
Z domain variable
-vii-
+UC-SDRL-RJA
CN-20-263-662
Greek Alphabet
(t)
f
t
Dirac impulse function

discrete interval of frequency (Hertz or cycles/sec)
discrete interval of sample time (sec)
small number
noise on the output
r th complex eigenvalue, or system pole

r = r + j r
diagonal matrix of poles in Laplace domain
noise on the input

variable of frequency (rad/sec)
imaginary part of the system pole, or damped natural frequency, for mode
r (rad/sec)
r = r
1 r 2
undamped natural frequency (rad/sec)

r =
r 2 + r 2
pr
{ }r
[]
{ }
pr
{ }r
[]
scaled pth response of normal modal vector for mode r

scaled normal modal vector for mode r
scaled normal modal vector matrix
scaled eigenvector
scaled pth response of a complex modal vector for mode r
scaled complex modal vector for mode r
scaled complex modal vector matrix
variable of damping (rad/sec)
real part of the system pole, or damping factor, for mode r
damping ratio
damping ratio for mode r
vector implied by definition of function
(0-8)
TABLE OF CONTENTS
PREFACE
................................................................................................................................
PRINTING/REVISION HISTORY
.........................................................................................
4
4
NOMENCLATURE ................................................................................................................. 5
Matrix Notation .................................................................................................................... 5
Operator Notation ................................................................................................................ 5
Roman Alphabet .................................................................................................................. 6
Greek Alphabet .................................................................................................................... 8
1. INTRODUCTION ............................................................................................................... 1
1.1 Degrees of Freedom .................................................................................................... 2
1.2 Basic Assumptions ...................................................................................................... 5
2. SINGLE DEGREE OF FREEDOM SYSTEM ....................................................................
2.1 Theory .........................................................................................................................
2.2 Laplace Domain Theory .............................................................................................
2.3 Definition Of Terms ....................................................................................................
2.3.1 Critical Damping .............................................................................................
2.3.2 Fraction of Critical Damping - Damping Ratio (Zeta) ...................................
2.4 System Classification ..................................................................................................
2.5 Analytical Model - Scalar Polynomial ......................................................................
2.6 Analytical Model - Partial Fraction ..........................................................................
2.7 Frequency Response Function Representation .........................................................
2.8 Impulse Response Function Representation .............................................................
2.9 Change of Physical Parameters .................................................................................
2.10 Estimating Partial Fraction Parameters .....................................................................
2.10.1 Example ........................................................................................................
2.11 Why Study/Emphasize SDOF Systems? ..................................................................
1
1
3
5
5
5
6
18
18
19
21
23
27
30
35
3. MULTIPLE DEGREE OF FREEDOM SYSTEMS ............................................................

3.1 Theory .........................................................................................................................
3.2 Solution of the Eigenvalue Problem ...........................................................................
3.2.1 Two Degree of Freedom Example: Undamped, Unforced .............................
3.3 Weighted Orthogonality of Modal Vectors .................................................................
3.4 Modal Vector Scaling Example ................................................................................
3.5 Principal Coordinates - Modal Coordinates ..............................................................
3.6 Two Degree of Freedom Example: Undamped, Forced ............................................
3.7 Proportional Damping ...............................................................................................
3.8 Modal Vectors from the System Matrix ....................................................................
1
1
5
6
9
12
13
18
23
24
4. FREQUENCY RESPONSE FUNCTION DEVELOPMENT ............................................. 1

4.1 Theory ......................................................................................................................... 1
4.2 Analytical Model - Scalar/Matrix Polynomial (MDOF)
............................................
Analytical Model - Partial Fraction (Residue) ..........................................................
Modal Vector Example .............................................................................................
Analytical Model - General Partial Fraction (Residue) ............................................
Residue Relationship to Modal Vectors ....................................................................
Residue Relationship to Modal Mass .......................................................................
8
10
14
18
19
20
5. GENERAL DAMPED SYSTEMS ....................................................................................

5.1 Non-Proportionally Damped Systems ........................................................................
5.1.1 Weighted Orthogonality of the Eigenvectors ..................................................
5.2 Proportionally Damped Systems .................................................................................
5.3 Example with Proportional Damping .........................................................................
5.3.1 Frequency Response Function Implications .................................................
5.3.2 Impulse Response Function Considerations .................................................
5.4 Non-Proportional Damping Example .......................................................................
5.4.1 Impulse Response Function Example ..........................................................
5.5 Summary of Modal Parameters (Undamped and Damped Systems) ........................
1
1
3
5
8
11
14
23
27
36
4.3
4.4
4.5
4.6
4.7
6. FREQUENCY RESPONSE FUNCTION SYNTHESIS ..................................................... 1

6.1 Displacement, Velocity and/or Acceleration Data ...................................................... 7
6.2 Skewed Sensor Orientation ......................................................................................... 8
6.3 Simulation of Structural Response .............................................................................. 8
7. MODAL SCALING .............................................................................................................
7.1 Proportionally Damped Systems (Modal Mass) .........................................................
7.1.1 Modal Vector Scaling ......................................................................................
7.1.1.1 Unity Modal Mass ............................................................................
7.1.1.2 Unity Modal Coefficient .................................................................
7.1.1.3 Unity Modal Vector Length ............................................................
7.1.2 Modal Vector Scaling Example ......................................................................
7.2 Non-Proportionally Damped Systems (Modal A) ......................................................
7.3 Modal Mass, Modal A Units Discussion ....................................................................
1
1
4
4
5
5
6
7
9
8. ADVANCED MODAL ANALYSIS CONCEPTS ...............................................................

8.1 Introduction .................................................................................................................
8.2 Measurement Degrees of Freedom .............................................................................
8.3 Mathematical Models ..................................................................................................
8.4 Repeated Modal Frequencies ......................................................................................
8.4.1 Repeated Modal Frequency Example: Residue Synthesis ...........................
8.4.2 Repeated Modal Frequency Example: Modal Vector Solution ....................
8.5 Left and Right Eigenvectors ......................................................................................
8.6 Summary/Conclusions ..............................................................................................
8.7 References .................................................................................................................
1
1
1
2
6
11
16
18
20
20
Appendix A: MEASUREMENT DISPLAY FORMATS

Appendix B: MATLAB Script Files
ii
.......................................................
+UC-SDRL-RJA
CN-20-263-662
1. INTRODUCTION
Todays engineers are faced with many complex noise and vibration problems associated with
the design and troubleshooting of structures. Never before have structures had so many
constraints, such as legislative, cost, and durability, imposed upon them. Because of these
constraints, present day structures are typically more complex in terms of design and materials.
Therefore, the engineering that goes into these structures must be more exact than ever before.
The structural analyst must design with accurate and realistic models and the experimentalist
must be able to accurately define a structures dynamic response to specified input forces.
Recently, tools have been developed to assist the structural engineer in these areas. The analyst
now uses sophisticated finite element programs, such as Nastran, to aid in the understanding and
the design of structures. The experimentalist has sophisticated digital signal analysis equipment
that aids him in quantifying and understanding a structures dynamic and input forces. However,
there is little communication between the analyst and experimentalist. Due to the complexity of
todays problems, the need for communication and exchange of ideas is even more critical.
The major reason for this lack of communication is that the analytical and experimental
engineers do not understand each others terminology and methods of problem solution. The
discussion that will follow will attempt to fill this void.
The major point of the text is to show how frequency response function (FRF) measurements are
related to the structures mode shapes and vibrational frequencies. This overall objective will be
accomplished by building a mathematical foundation from the analytical and experimental point
of view.
The material is divided into two sections: single degree of freedom, and multiple degree of
freedom systems. The single degree of freedom system will be used as a means for defining
some standard terminology. The multiple degree of freedom material is further divided into:
undamped, proportionally damped, and non-proportionally damped systems. Throughout the
discussion emphasis will be placed on the relationship of frequency response measurements to
modal vectors. Furthermore, the following concepts will be presented:
(1-1)
+UC-SDRL-RJA
CN-20-263-662
Modal Frequencies, Eigenvalues, System Poles, Characteristic Roots ( r )
Modal Vectors, Eigenvectors ( { r }, { r } )
Modal Coefficients ( pr ), Residues ( A pqr )
Real Modal Vectors, Complex Modal Vectors
Physical Coordinates, Generalized Coordinates
Principal Coordinates, Modal coordinates
Modal Mass, Damping, and Stiffness ( M r , C r , K r )
Modal A, Modal B ( M Ar , M Br )
1.1 Degrees of Freedom

The development of any theoretical concept in the area of vibrations, including modal analysis,
depends upon an understanding of the concept of the number of degrees of freedom (N ) of a
system. This concept is extremely important to the area of modal analysis since the number of
modes of vibration of a mechanical system is equal to the number of degrees of freedom. From a
practical point of view, the relationship between this theoretical definition of the number of
degrees of freedom and the number of measurement degrees of freedom (N o , N i ) is often
confusing. For this reason, the concept of degree of freedom will be reviewed as a preliminary to
the following modal analysis material.
To begin with, the basic definition that is normally associated with the concept of the number of
degrees of freedom involves the following statement: The number of degrees of freedom for a
mechanical system is equal to the number of independent coordinates (or minimum number of
coordinates) that is required to locate and orient each mass in the mechanical system at any
instant in time. As this definition is applied to a point mass, three degrees of freedom are
required since the location of the point mass involves knowing the x, y, and z translations of the
center of gravity of the point mass. As this definition is applied to a rigid body mass, six degrees
of freedom are required since x , y , and z rotations are required in addition to the x, y, and z
translations in order to define both the orientation and location of the rigid body mass at any
instant in time. This concept is represented in Figure 1-1. As this definition is extended to any
general deformable body, it should be obvious that the number of degrees of freedom can now be
(1-2)
+UC-SDRL-RJA
CN-20-263-662
considered as infinite. While this is theoretically true, it is quite common, particularly with
respect to finite element methods, to view the general deformable body in terms of a large
number of physical points of interest with six degrees of freedom for each of the physical points.
In this way, the infinite number of degrees of freedom can be reduced to a large but finite
number.
y, y
x, x
z, z
Figure 1-1. Degrees of Freedom of a Rigid Body
When measurement limitations are imposed upon this theoretical concept of the number of
degrees of freedom of a mechanical system, the difference between the theoretical number of
degrees of freedom (N ) and the number of measurement degrees of freedom (N o , N i ) begins to
evolve. Initially, for a general deformable body, the number of degrees of freedom (N ) can be
considered to be infinite or equal to some large finite number if a limited set of physical points of
interest is considered as discussed in the previous paragraph. The first measurement limitation
that needs to be considered is that there is normally a limited frequency range that is of interest to
(1-3)
+UC-SDRL-RJA
CN-20-263-662
the analysis. For example, most dominant structural modes of vibration for an automobile would
be located between 0 and 200 Hertz. As this limitation is considered, the number of degrees of
freedom of this system that are of interest is now reduced from infinity to a reasonable finite
number. The next measurement limitation that needs to be considered involves the physical
limitation of the measurement system in terms of amplitude. A common limitation of
transducers, signal conditioning and data acquisition systems results in a dynamic range of 80 to
100 db (104 to 105 ) in the measurement. This means that the number of degrees of freedom is
reduced further due to the dynamic range limitations of the measurement instrumentation.
Finally, since few rotational transducers exist at this time, the normal measurements that are
made involve only translational quantities (displacement, velocity, acceleration, force) and thus
do not include rotational effects, or RDOF. In summary, even for the general deformable body,
the theoretical number of degrees of freedom that are of interest are limited to a very reasonable
finite value (N = 1 50). Therefore, this number of degrees of freedom (N ) is the number of
modes of vibration that are of interest.
Finally, then, the number of measurement degrees of freedom (N o , N i ) can be defined as the
number of physical locations at which measurements are made times the number of
measurements made at each physical location. For example, if x, y, and z accelerations are
measured at each of 100 physical locations on a general deformable body, the number of
measurement degrees of freedom would be equal to 300. It should be obvious that since the
physical locations are chosen somewhat arbitrarily, and certainly without exact knowledge of the
modes of vibration that are of interest, that there is no specific relationship between the number
of degrees of freedom (N ) and the number of measurement degrees of freedom (N o , N i ). In
general, in order to define N modes of vibration of a mechanical system, N o , N i must be equal to
or larger than N . Note also that even though N o , N i is larger than N , this is not a guarantee that
N modes of vibration can be found from N o , N i measurement degrees of freedom. The N o , N i
measurement degrees of freedom must include physical locations that allow a unique
determination of the N modes of vibration. For example, if none of the measurement degrees of
freedom are located on a portion of the mechanical system that is active in one of the N modes of
vibration, portions of the modal parameters for this mode of vibration can not be found.
In the development of the single and multiple degree of freedom information in the following
Sections, the assumption is made that a set of N measurement degrees of freedom (N o , N i = N )
exist that will allow for N modes of vibration to be determined. In reality, N o , N i is always
(1-4)
+UC-SDRL-RJA
CN-20-263-662
chosen much larger than N since a prior knowledge of the modes of vibration is not available. If
the set of N o , N i measurement degrees of freedom is large enough and if the N o , N i measurement
degrees of freedom are distributed uniformly over the general deformable body, the N modes of
vibration will normally be found.
1.2 Basic Assumptions

Before proceeding, the basic assumptions must be established before the theory can be
developed. The first assumption is that the structure is a linear system whose dynamics may be
represented by a set of linear, second order, differential equations. The second assumption is that
the structure during the test can be considered as time invariant. This assumption implies that the
coefficients in the linear, second order, differential equations are constants and do not vary with
time. The third assumption is that the structure is observable. While this may seem trivial, this
means that the system characteristics that are affecting the dynamics can be measured and that
there are sufficient sensors to adequately describe the input-output characteristics of the system.
Another assumption that is often made is that the structure obeys Maxwells reciprocity theorem.
Maxwells reciprocity theorem, in terms of frequency response function measurements, implies
the following: if one measures the frequency response function between points p and q by
exciting at p and measuring the response at q, the same frequency response function will be
measured by exciting at q and measuring the response at p H pq = H qp .
(1-5)
+UC-SDRL-RJA
CN-20-263-662
2. SINGLE DEGREE OF FREEDOM SYSTEM
2.1 Theory
The general mathematical representation of a single degree of freedom system is expressed using
Newtons second law in Equation 2.1:
M x(t)
+ C x(t)
+ K x(t) = f (t)
(2.1)
and is represented schematically in Figure (2-1).
x(t)
f (t)
K
M
C
Figure 2-1. Single Degree of Freedom System
Equation 2.1 is a linear, time invariant, second order differential equation. The total solution to
this problem involves two parts as follows:
(2-1)
+UC-SDRL-RJA
CN-20-263-662
x(t) = x c (t) + x p (t)

where:
x c (t) = Transient portion
x p (t) = Steady state portion
By setting f (t) = 0, the homogeneous (transient) form of Equation 2.1 can be solved.
M x (t) + C x (t) + K x (t) = 0
(2.2)
From differential equation theory, the solution can be assumed to be of the form x c (t) = Xe st ,
where s is a constant to be determined. Taking appropriate derivatives and substituting into
Equation 2.2 yields:
(Ms2 + C s + K ) X(s) e s t = 0
Thus, for a non-trivial solution X(s) e s t 0 :
s2 + ( C / M ) s + ( K / M ) = 0
(2.3)
Equation 2.3 is the systems characteristic equation, whose roots 1 and 2 ( = system pole )
are:
1,2
=
2M
C
K
2M
M
Thus the homogeneous solution of Equation 2.1 is:

x c (t) = X 1 e 1 t + X 2 e 2 t
where X 1 and X 2 are constants determined from the initial conditions imposed on the system at
t = 0.
The particular solution (steady state) is a function of the form of the forcing function. If the
(2-2)
+UC-SDRL-RJA
CN-20-263-662
forcing function is a pure sine wave of a single frequency, the response will also be a sign wave
of the same frequency. If the forcing function is random in form, the response is also random.
2.2 Laplace Domain Theory
Equation 2.1 is the time domain representation of the system in Figure (2-1). An equivalent
equation of motion may be determined for the Laplace or s domain. This representation has the
advantage of converting a differential equation to an algebraic equation. This is accomplished
by taking the Laplace transform of Equation 2.1, thus:
L { M x + C x + K x } = M ( s2 X(s) s x(0) x(0)
)
+ C ( s X(s) x(0) ) + K X(s)
L { M x + C x + K x } = ( M s2 + C s + K ) X(s) Ms x(0) M x(0)
C x(0)
L { f (t) } = F(s)
Thus Equation 2.1 becomes:
[M s2 + C s + K ] X(s) = F(s) + (M s + C) x(0) + M x(0)
(2.4)
where:
x(0) is the initial displacement at time t = 0.
x(0)
is the initial velocity at time t = 0.
If the initial conditions are zero, Equation 2.4 becomes:

[M s2 + C s + K ] X(s) = F(s)
(2.5)
Let B(s) = M s2 + C s + K . B(s) is referred to as the system impedance. Then Equation 2.5
becomes:
B(s) X(s) = F(s)
(2-3)
(2.6)
+UC-SDRL-RJA
CN-20-263-662
Equation 2.6 is an equivalent representation of Equation 2.1 in the Laplace domain. The Laplace
domain (s domain) can be thought of as complex frequency (s = + j ). Therefore, the
quantities in Equation 2.6 can be thought of as follows:
F (s) - the Laplace domain (complex frequency) representation of the forcing function
f (t)
X (s) - the Laplace domain (complex frequency) representation of the system response
x (t)
Equation 2.6 states that the system response X(s) is directly related to the system forcing
function F(s) through the quantity B(s). If the system forcing function F(s) and its response
X(s) are known, B(s) can be calculated. That is:
B(s) =
F(s)
X(s)
More frequently one would like to know what the system response is going to be due to a known
input F(s), or:
X(s) =
By defining H(s) =
F(s)
B(s)
(2.7)
1
, Equation 2.7 becomes:
B(s)
X(s) = H(s) F(s)
(2.8)
The quantity H(s) is known as the system transfer function. In other words, a transfer function
relates the Laplace transform of the system input to the Laplace transform of the system
response. From Equations 2.5 and 2.8, the transfer function can be defined as:
H(s) =
1/M
X(s)
= 2
F(s) s + ( C / M ) s + ( K / M )
assuming initial conditions are zero.
(2-4)
(2.9)
+UC-SDRL-RJA
CN-20-263-662
The denominator term is referred to as the system characteristic equation. The roots of the
characteristic equation are:
1,2 = ( C / 2M )
( C / 2M )2 ( K / M )
(2.10)
Note that the above definition of the transfer function establishes a form of an analytical model
that can be used to describe the transfer function. This analytical model involves a numerator
and denominator polynomial with scalar coefficients. For the single degree of freedom case, the
numerator polynomial is zeroeth order and the denominator polynomial is second order.
2.3 Definition Of Terms
2.3.1 Critical Damping
Critical damping C c is defined as being the damping which reduces the radical in the
characteristic equation to zero.
( C c / 2M )2 (K / M ) = 0
( C c / 2M ) =
K / M = 1
C c = 2M 1 = critical damping coefficient

1 = undamped natural frequency ( rad / sec )
2.3.2 Fraction of Critical Damping - Damping Ratio (Zeta)
The fraction of critical damping or damping ratio, , is the ratio of the actual system damping to
the critical system damping.
1 = C / Cc
The roots of characteristic Equation 2.10 can now be written as:
(2-5)
+UC-SDRL-RJA
CN-20-263-662
1,2 = 1
12 1 1
1,2 = 1 j
1 12 1
(2.11a)
(2.11b)
2.4 System Classification
Systems can be classified depending on their damping ratios. That is:
Overdamped system:
1 > 1
Critically damped system:
1 = 1
Underdamped system:
1 < 1
Figures (2-2) through (2-4) illustrate typical the time domain response of these 3 different cases.
The following plots illustrate the location of the roots of the characteristic equation in the splane. Figure (2-2) (overdamped) shows two real roots that lie on the axis, if damping were to
increase the roots would move apart. For Figure (2-3) (critically damped), there are two identical
real roots. For Figure (2-4) (underdamped), there are two complex roots, complex conjugates of
each other. As the damping and/or the frequency changes, these roots stay in the second and
third quadrant of the graph. It should be pointed out that if any roots of the characteristic
equation lie to the right of the j axis, the system would be unstable.
(2-6)
+UC-SDRL-RJA
CN-20-263-662
Overdamped SDOF System

20
18
16
Displacement
14
12
10
8
6
4
2
0
0
0.5
1.5
2.5
3.5
Time(sec.)
Figure 2-2. Overdamped SDOF System Response: Initial Displacement
(2-7)
+UC-SDRL-RJA
CN-20-263-662
Critically Damped SDOF System

20
18
16
Magnitude
14
12
10
8
6
4
2
0
0
0.5
1.5
2.5
3.5
Time(sec.)
Figure 2-3. Critically Damped SDOF System Response: Initial Displacement
(2-8)
+UC-SDRL-RJA
CN-20-263-662
Underdamped SDOF System

20
Magnitude
15
10
-5
0.5
1.5
2.5
3.5
Time(sec.)
Figure 2-4. Underdamped SDOF System Response: Initial Displacement
Figures (2-5) through (2-7) show the location of the roots, in the s or Laplace plane, of the
characteristic equation for each of the three cases.
(2-9)
+UC-SDRL-RJA
CN-20-263-662
Overdamped SDOF System
Imaginary Part of Complex Frequency
15
10
-5
-10
-15
-15
-10
-5
Real Part of Complex Frequency
Figure 2-5. Overdamped SDOF System
(2-10)
10
15
+UC-SDRL-RJA
CN-20-263-662
Critically Damped SDOF System
15
10
-5
-10
-15
-15
-10
-5
Figure 2-6. Critically Damped SDOF System
(2-11)
10
15
+UC-SDRL-RJA
CN-20-263-662
Underdamped SDOF System

15
10
-5
-10
-15
-15
-10
-5
10
15
Figure 2-7. Underdamped SDOF System
For most real structures, unless active damping systems are present, the damping ratio is rarely
greater than ten percent. For this reason, all further discussion will be restricted to underdamped
systems 1 < 1. For an underdamped system, the roots of the characteristic equation can be
written as:
1,2 = 1 j 1
where:
1
= damping factor (units of rad/sec)
(2-12)
+UC-SDRL-RJA
CN-20-263-662
= damped natural frequency
Note that for this case 2 is always the complex conjugate of 1 . Therefore, the 2 notation will
be replaced in further equations by *1 .
Using Equation 2.11 the above parameters can be related to the damping ratio ( 1 ) and the
undamped natural frequency (1 ) as follows:
1 =
1
12 + 12
1 = 1 1
1 =
12 + 12
The transfer function H(s) can now be rewritten as a product of the roots (in factored) form as
follows:
H(s) =
1/M
( s 1 ) (s *1 )
(2.12)
where:

= pole of the transfer function
= 1 + j 1
= 1 j 1
The poles of the single degree of freedom system can also be viewed, looking down on the splane as shown in Figure (2-8) :
(2-13)
+UC-SDRL-RJA
CN-20-263-662
j
pole location
r
r
s = r + j r
r
conjugate pole
X
r - damping coefficient
r - damped natural frequency
r - resonant (undamped natural frequency)
r = cos r - damping factor (or percent of critical damping)
Figure 2-8. Laplace Plane Pole Location
Figures (2-9) through (2-11) illustrate a 3-dimensional plot of Equation 2.12. Figure (2-9) views
the surface in real/imaginary format, Figure (2-10) represents the same data in a magnitude/phase
format and Figure (2-11) uses a log magnitude/phase format. Remember that the variable s in
Equation 2.12 is a complex variable, that is, it has a real part and an imaginary part. Therefore, it
can be viewed as a function of two variables which represent a surface. Note also that the FRF
measurement that is typically estimated is the slice through these surfaces where s = j .
(2-14)
+UC-SDRL-RJA
CN-20-263-662
10
-5
-10
-0.2
-0.15
1
-0.1
0.5
0
-0.05
-0.5
0
-1
6
4
2
0
-2
-4
-6
-0.2
-0.15
1
-0.1
0.5
0
-0.05
-0.5
0
-1
Figure 2-9. Transfer Function (Real-Imaginary), Surface Representation
(2-15)
+UC-SDRL-RJA
CN-20-263-662
10
8
6
4
2
0
-0.2
-0.15
1
-0.1
0.5
0
-0.05
-0.5
0
-1
-2
-4
-0.2
-0.15
1
-0.1
0.5
0
-0.05
-0.5
0
-1
Figure 2-10. Transfer Function (Magnitude-Phase), Surface Representation
(2-16)
+UC-SDRL-RJA
CN-20-263-662
3
2
1
0
-1
-2
-3
-0.2
-0.15
1
-0.1
0.5
0
-0.05
-0.5
0
-1
-2
-4
-0.2
-0.15
1
-0.1
0.5
0
-0.05
-0.5
0
-1
Figure 2-11. Transfer Function (Log Magnitude-Phase), Surface Representation
(2-17)
+UC-SDRL-RJA
CN-20-263-662
2.5 Analytical Model - Scalar Polynomial
One popular method of representing the transfer function involves a scalar polynomial
representation in the numerator and denominator. For the single degree of freedom case, this is
very simple concept that is directly based upon the physical characteristics (M,C,K) of the
system. Generalizing Equation (2.9) yields:
H(s) =
0
2
(s)2
+ 1 (s)1 + 0 (s)0
(2.13)
This can be rewritten:
H(s) =
0
2
(2.14)
(s)k
k=0
This model serves as the basis for many modal parameter estimation methods and is a common
formulation utilzed in control theory applications..
2.6 Analytical Model - Partial Fraction
The concept of residues can now be discussed in terms of the partial fraction expansion of the
transfer function equation. This is just one common approach to determining the residues.
Another popular method involves a polynomial representation in the numerator and denominator.
Equation 2.12 can be expressed in terms of partial fractions:
H(s) =
c1
c2
1/M
=
+
(s 1 ) (s *1 ) (s 1 ) (s *1 )
(2.15)
The residues of the transfer function are defined as being the constants c 1 and c 2 . The
(2-18)
+UC-SDRL-RJA
CN-20-263-662
terminology and development of residues comes from the evaluation of analytic functions in
complex analysis. As will be shown later, the residues of the transfer function are directly
related to the amplitude of the impulse response function. The constants c 1 and c 2 (residues) can
be found by multiplying both sides of Equation 2.15 by (s 1 ) and evaluating the result at
s = 1 . Thus:
c (s 1 )
1/M
|s = 1 = c 1 + 2
|
*
* s = 1
(s
)
(s 1 )
1
1/M
= c1
( 1 *1 )
Thus:
c1 =
1/M
1/M
= A1
=
( 1 + j 1 ) ( 1 j 1 )
j 2 1
Similarly:
c2 =
1/M
= A*1
j2 1
In general, for a multiple degree of freedom system, the residue A1 can be a complex quantity.
But, as shown for a single degree of freedom system A1 is purely imaginary.
Therefore:
H(s) =
A1
A*1
+
( s 1 )
( s *1 )
(2.16)
2.7 Frequency Response Function Representation

The frequency response function is the transfer function (surface) evaluated along the j
(frequency) axis. Thus, from the previously dervied equations:
(2-19)
+UC-SDRL-RJA
CN-20-263-662
Polynomial Model
H(s)|s= j = H( ) =
0
2
( j )2
+ 1 ( j )1 + 0 ( j ) 0
(2.17a)
Partial Fraction Model

H(s)|s= j = H( ) =
H( ) =
H( ) =
A1
A*1
+
( j 1 ) ( j *1 )
A1
( j 1 j 1 )
A1
j( 1 ) 1
A*1
( j 1 + j 1 )
A*1
j( + 1 ) 1
(2.17b)
From an experimental point of view, when one talks about measuring a transfer function, the
frequency response function is actually being measured.
The value of the frequency response function at the damped natural frequency of the system is:
H( 1 ) =
A1
1
A*1
j 2 1 1
(2.18)
which can be approximated as:

H( 1 ) =
A1
1
The second term on the right of Equation 2.18 approaches zero as 1 gets large. In other words,
the contribution of the negative frequency portion of the frequency response function is
negligible.
Therefore, many single degree of freedom models are represented as:
H( )
A1
( j 1 )
(2-20)
(2.19)
+UC-SDRL-RJA
CN-20-263-662
Another way of interpreting Equation 2.17 is that the value of the transfer function, for a single
degree of freedom system, at a particular frequency ( ) is a function of the residue, damping,
and damped natural frequency.
2.8 Impulse Response Function Representation
The impulse response function of the single degree of freedom system can be determined from
Equation 2.17 assuming that the initial conditions are zero and that F(s) = 1 for a system
impulse. Thus:
A*1
A1
X(s) =
+
(s 1 ) (s *1 )
x(t) = L1 {X(s) }
x(t) = A1 e 1 t + A*1 e 1 t = h(t) = impulse response
*
x(t) = e 1 t A1 e j 1 t + A*1 e j 1 t
Thus, using Eulers formula for e j 1 t and e j 1 t , the residue A1 controls the initial amplitude of
the impulse response, the real part of the pole is the decay rate and the imaginary part of the pole
is the frequency of oscillation. Figures (2-12) and (2-13) illustrate the frequency response and
impulse response functions respectively, for a single degree of freedom system.
(2-21)
+UC-SDRL-RJA
CN-20-263-662
1
0.8
0.6
0.4
Real
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
0
0.5
1.5
2.5
2.5
Frequency (Rad/Sec)
1
0.8
0.6
Imaginary
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
0
0.5
1.5
Frequency (Rad/Sec)
Figure 2-12. Frequency Response Function: SDOF System
(2-22)
+UC-SDRL-RJA
CN-20-263-662
0.1
0.08
0.06
Amplitude
0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
10
20
30
40
50
60
Time (Sec.)
Figure 2-13. Impulse Response Function: SDOF System
2.9 Change of Physical Parameters
While it is not always possible to alter the physical parameter (mass, stiffness and/or damping) of
a system and is very difficult to practically alter one physical parameter (mass, for example)
without altering another physical parameter (stiffness, for example), it is still important to
understand how a change in physical parameter will affect the system characteristics. Figures
(2-14), (2-15) and (2-16) show how the frequency response function will be affected due to a
change in one physical parameter at a time.
(2-23)
+UC-SDRL-RJA
CN-20-263-662
-2
10
-3
Log Magnitude
10
-4
10
-5
10
10
20
30
Frequency (Rad/Sec)
40
50
10
20
30
Frequency (Rad/Sec)
40
50
0
-20
-40
Phase (Deg)
-60
-80
-100
-120
-140
-160
-180
0
Figure 2-14. Change of Stiffness: SDOF System
Note that a change in stiffness affects both the resonant frequency as well as the system
characteristic at low frequency. This dominance of stiffness at low frequency is the reason that
this region of the frequency response function is known as the stiffness, or more accurately,
compliance line.
(2-24)
+UC-SDRL-RJA
CN-20-263-662
Effect of Changing Damping
-1
10
-2
Log Magnitude
10
-3
10
-4
10
-5
10
10
20
30
Frequency (Rad/Sec)
40
50
10
20
30
Frequency (Rad/Sec)
40
50
0
-20
-40
Phase (Deg)
-60
-80
-100
-120
-140
-160
-180
0
Figure 2-15. Change of Damping: SDOF System
Note that a change in damping has no apparant effect on the resonant frequency. The only
noticable change involves a change in frequency response function in the region of the resonant
frequency.
(2-25)
+UC-SDRL-RJA
CN-20-263-662
-2
10
-3
Log Magnitude
10
-4
10
-5
10
10
20
30
Frequency (Rad/Sec)
40
50
10
20
30
Frequency (Rad/Sec)
40
50
0
-20
-40
Phase (Deg)
-60
-80
-100
-120
-140
-160
-180
0
Figure 2-16. Change of Mass: SDOF System
Note that a change in mass affects both the resonant frequency as well as the system
characteristic at high frequency. This dominance of mass at high frequency is the reason that this
region of the frequency response function is known as the mass line. Also note that as the mass
changes, the apparant damping (sharpness of the resonant frequency) changes accordingly. A
change in mass affects both the resonant frequency, the system characteristic at high frequency as
(2-26)
+UC-SDRL-RJA
CN-20-263-662
well as the fraction of critical damping ( 1 =
C
C
=
).
C c 2 M 1
2.10 Estimating Partial Fraction Parameters
Assuming that a lightly damped single degree of freedom (SDOF) system is being evaluated, the
parameters needed for a partial fraction model can be quickly estimated directly from the
measured frequency response function. While this approach is based upon a SDOF system, as
long as the modal frequencies are not too close together, the method can be used for multiple
degree of freedom (MDOF) systems as well.
Starting with the partial fraction model formulation of a SDOF frequency response function
H( ) =
A1
A*1
+
( j 1 )
( j *1 )
(2.20)
only the constants 1 and A1 must be estimated. Since 1 = 1 + j 1 , the estimation process
begins by estimating 1 . The damped natural frequency 1 is estimated in one of three ways:
The frequency where the magnitude of the FRF reaches a maximum.
The frequency where the real part of the FRF crosses zero.
The frequency where the imaginary part of the FRF reaches a relative minima (or
maxima).
Of these three methods, the last approach gives the most reliable results under all conditions.
Once the damped natural frequency 1 has been estimated, the real part of the modal frequency,
the damping factor 1 , can be estimated. The damping factor 1 can be estimated by using the
half-power bandwidth method. This method uses the data from the FRF in the region of the
resonance frequency to estimate the fraction of critical damping from the following formula:
1 =
b2 a2
(2 1 )2
(2-27)
(2.21)
+UC-SDRL-RJA
CN-20-263-662
In the above equation, 1 is the damped natural frequency as previously estimated. a is the
frequency, below 1 , where the magnitude is 0.707 of the peak magnitude of the FRF. This
corresponds to a half power point. b is the frequency, above 1 , where the magnitude is 0.707
of the peak magnitude of the FRF. This also corresponds to a half power point.
For lightly damped systems, the above equation can be approximated by the following:
1
b a
2 1
(2.22)
Once 1 is estimated, the damping factor 1 can be estimated from the following equation.
1 = 1 1
(2.23)
Again, assuming that the system is lightly damped, 1 1 , the damping factor can be estimated
from the following equation:
1 1 1
(2.24)
Once the modal frequency 1 has been estimated, the residue A1 can be estimated by evaluating
the partial fraction model at a specific frequency. If the specific frequency is chosen to be 1 , the
following result is obtained.
H( 1 ) =
A*1
A1
+
( j 1 ( 1 + j 1 )
( j 1 ( 1 j 1 )
(2.25)
A1
A*1
+
( 1 )
( 2 j 1 1 )
(2.26)
H( 1 ) =
As long as 1 is not too small, the above equation can be approximated by:
H( 1 )
A1
( 1 )
Therefore, the residue A1 can be estimated from the following relationship:
(2-28)
(2.27)
+UC-SDRL-RJA
CN-20-263-662
A1 ( 1 ) H( 1 )
(2.28)
In the above relationship, H( 1 ) is very close to being a purely imaginary value for the
displacement over force FRF. This means that the residue A1 will be very close to a purely
imaginary value as well.
Figure 2-17. Modal Vectors from the Imaginary Part of the FRF
(2-29)
+UC-SDRL-RJA
CN-20-263-662
2.10.1 Example
For the following SDOF case, (M=10, K=1600, C=2), the data can be estimated from the FRF as
shown in the following plots. The exact answers are 1 = -0.1000 + j 12.6487 and A1 = -j
0.0040. Note that the digitized data, in the neighborhood of the damped natural frequency, is
tabulated in Table 2-1.
(2-30)
+UC-SDRL-RJA
CN-20-263-662
0.02
0.015
0.01
Real
0.005
0
-0.005
-0.01
-0.015
-0.02
10
15
20
25
30
20
25
30
Frequency (Rad/Sec)
-0.005
-0.01
Imaginary
-0.015
-0.02
-0.025
-0.03
-0.035
-0.04
10
15
Frequency (Rad/Sec)
(2-31)
+UC-SDRL-RJA
CN-20-263-662
0.04
0.035
0.03
Magnitude
0.025
0.02
0.015
0.01
0.005
0
10
15
20
25
30
20
25
30
Frequency (Rad/Sec)
0
-20
-40
Phase (Deg)
-60
-80
o
-100
-120
-140
-160
-180
10
15
Frequency (Rad/Sec)
(2-32)
+UC-SDRL-RJA
CN-20-263-662
0.04
0.035
0.03
Magnitude
0.025
0.02
0.015
0.01
0.005
0
11.5
12
12.5
13
Frequency (Rad/Sec)
(2-33)
13.5
+UC-SDRL-RJA
CN-20-263-662
Frequency (rad/sec)
12.2705
12.3205
12.3706
12.4207
12.4708
12.5209
12.5710
12.6210
12.6711
12.7212
12.7713
12.8214
12.8715
12.9215
12.9716
Real
0.0099
0.0112
0.0127
0.0147
0.0170
0.0193
0.0193
0.0103
-0.0083
-0.0186
-0.0192
-0.0170
-0.0146
-0.0126
-0.0110
Imag
-0.0026
-0.0034
-0.0045
-0.0064
-0.0095
-0.0150
-0.0246
-0.0367
-0.0376
-0.0259
-0.0158
-0.0099
-0.0066
-0.0047
-0.0035
Magnitude
0.0103
0.0117
0.0135
0.0160
0.0195
0.0245
0.0313
0.0381
0.0385
0.0319
0.0249
0.0197
0.0160
0.0135
0.0115
Phase (deg)
-14.578
-16.717
-19.549
-23.452
-29.109
-37.804
-51.903
-74.301
-102.401
-125.711
-140.565
-149.696
-155.597
-159.646
-162.569
TABLE 2-1. Discrete SDOF Data from Plot
This example also illustrates a common problem with simplified modal parameter estimation. In
this example, referring to Figure 2-20, it is apparent that the damped natural frequency occurs
between two of the measured frequencies in the frequency respone function. In Figure 2-20, the
apparent truncation, or clip, of the frequency response function near the peak frequency is a
result of this lack of resolution. For a lightly damped situation, the true magnitude at the
resonance may be 2 to 20 times higher. This means that finding the half-power frequencies in
order to estimate damping will be impossible since the true magnitude of the resonance is
unknown. In this situation, the damping estimate will be in error (too high) which will cause the
residue to be in error (too high).
Note that there are a number of other more robust SDOF, and MDOF, modal parameter
estimation algorithms that do not require knowledge of the half-power frequencies. These
techniques depend only upon the data being accurate at the measured frequencies in order for an
accurate estimate of the damping estimate. These methods do not have the accuracy problem of
the simplified SDOF case utilized in this example.
(2-34)
+UC-SDRL-RJA
CN-20-263-662
2.11 Why Study/Emphasize SDOF Systems?
Frequently, there is some concern that the amount of time studying SDOF systems is not
warranted. Many multiple degree-of-freedom (MDOF) systems can be simplified as single
degree-of-freedom systems. More often, the MDOF system can be broken down, on a frequency
range basis, into frequency regions that are dominated by only one degree-of-freedom. This is
the situation with SDOF modal parameter esimation algorithms.
Even though a continuous beam has an infinite number of modes, the evaluation of these modes
(estimation of frequency, damping, modal vector and modal scaling) can often be accomplished
with essentially single degree-of-freedom (SDOF) concepts. The primary assumption is that
each mode of vibration is well separated in frequency from the other modes. This is often the
case for lightly damped structures. Different modes of vibration of the beam can be visulaized in
Figure 2-17 by noting the solid black line connecting the peaks of the imaginary parts of each
frequency response. Normally, these modes are plotted in a wireframe model showing the
extrema of the modal vector so that the motion can be easily understood. Figures 2-21 and 2-22
show the first two bending modes of a uniform beam that is pinned at each end.
Figure 2-21. First Bending Mode at First Damped Natural Frequency ( 1 )
(2-35)
+UC-SDRL-RJA
CN-20-263-662
Figure 2-22. Second Bending Mode at Second Damped Natural Frequency ( 2 )

In order to understand why this information can be determined from the imaginary part of the
frequency response functions, SDOF theory must be reviewed and extended slightly, primarily
from a notational point of view.
The general mathematical representation of a single degree of freedom system is expressed using
Newtons second law in Equation 2-29:
M x(t)
+ C x(t)
+ K x(t) = f (t)
(2-29)
For the general case with a forcing function that can be represented as a summation of sin and
cosine terms, the forcing function can be represented as:
f (t) =
F( ) e j t
=0
(2-30)
Assuming that the system is underdamped and that enough time has passed that any transient
response of the system due to initial condition or startup of the excitation has decayed to zero,
the response of the system can be represented as:
x(t) =
X( ) e j t
=0
(2-36)
(2-31)
+UC-SDRL-RJA
CN-20-263-662
Note that, while x(t) and f (t) are real valued functions, X( ) and F( ) are complex valued.
Working with any arbitrary frequency term in Equations 2-30 and 2-31, Equation 2-30, Equation
2-31 and the derivatives of Equation 2-31 can substituted into Equation 1 yielding the following
frequency response function (FRF) relationship for a SDOF system:
H( ) =
X( )
1
=
2
F( ) M + C j + K
(2-32)
Note the characteristic of the above frequency response function when it is evaluated (measure)
at the undamped natural frequency. At the undamped natural frequency, the mass and stiffness
terms cancel each other and the FRF is purely imaginary valued.
The first extension that is necessary provides a description for the case where x(t) and f (t) are
not located at the same point. On a single degree-of-freedom system, this would provide
redundant information (no new information) but it becomes important as the extension to multple
degrees-of-freedom occurs. For example, assume that the particular point (and direction) on the
mass where the force is applied is referred to as DOF p and the particular point (and direction)
on the mass where the response is measured is referred to as DOF q. Equation 2-32 now can be
written as follows to note this information.
H qp ( ) =
X q ( )
1
=
2
F p ( ) M + C j + K
(2-33)
The system is still a SDOF system so H qp = H pp = H qq = H qs = . . . but the input and output
location can now be described. This clearly demonstrates that the number of modes (one in this
case) is unrelated to the number of input and output sensors that are used to measure the system.
The second extension that is necessary provides a way to indicate that the modal characteristics
(modal coefficients) of both the input and output are represented in the frequency response
function model. The modal frequency is already represented by noting that the denominator is
related to the characteristic equation. A form of modal scaling is already represented by noting
the the mass term in the denominator scales the equation. Modal coefficient information, which
is relative not absolute information, can be added by changing the numerator to reflect this.
H qp ( ) =
q p
X q ( )
=
F p ( ) M 2 + C j + K
(2-37)
(2-34)
+UC-SDRL-RJA
CN-20-263-662
Note that, since the system is still a SDOF system, the relative motion at each DOF would be
normalized to 1 such that p = q = s = 1 which shows that Equation 2-33 and 2-34 still
represent the same information. Note as before, if the FRF is evaluated (measured) at the
undamped natural frequency), the FRF is once again imaginary valued and is a function of the
modal coefficients and damping. Assuming the damping is unknown but constant means that the
product of the modal coefficients is proportional to the imaginary part of the FRF.
Finally, the third extension that is necessary provides for the change from SDOF to MDOF. Note
that for a linear system, linear superposition can be used in the frequency domain to add the
information associated with each mode together to represent the frequency response function of a
MDOF system. To desribe this, every term in Equation 2-34 will need a subcript (r) to indicate
which mode the information is associated with. The final form of the frequency response
function is:
H qp ( ) =
qr pr
X q ( )
=
F p ( ) r=1 M r 2 + C r j + K r
(2-35)
Equation 2-35 is one common representation of the FRF of a MDOF system. Note that the M r ,
C r and K r terms in the denominator are the modal or generalized mass, damping and stifness
parameters, not the physical mass, damping and stiffness parameters. The modal or generalized
parameters can be found analytically from the physical mass, damping and stiffness parameters
or experimentally using more complicated parameter estimation algorithms.
Note that, as long as the modes are well separated in frequency, the information in the
neighborhood of the undamped natural frequency for a given mode can be found from:
H qp ( ) =
qr pr
X q ( )
F p ( ) M r 2 + C r j + K r
(2-36)
Note that, if the output DOF (point and location) is held fixed while the input DOF is moved, the
only information that changes in Equation 2-36 as different FRFs are measured is the
information relative to the modal coefficient for the particular mode of interest. If Equation 2-36
is evaluated (measured) near the undamped natural frequency, this means that the imaginary part
of the FRF will be proportional to the modal coefficient. The proportionality constant r ) is:
(2-38)
+UC-SDRL-RJA
CN-20-263-662
pr
M r 2 + C r j + K r
(2-37)
Since mode shapes are relative patterns, not absolute motions, the value of the constant is not
important unless damping or modal scaling is required.
Therefore, modal vectors can be estimated from the imaginary part of the frequency response
functions at the damped natural frequencies (or from the magnitude and phase information of the
frequency response functions at the damped natural frequencies). This will be reasonably
accurate as long as the undamped natural frequencies are well separated and the damping is
small (undamped and damped natural frequencies nearly equal).
This result is consistent with the expansion theorem concept (the response of the system at any
instant in time or at any frequency is a linear combination of the modal vectors):
Expansion Theorem - Time Domain:
{ x(t i ) } =
r { r }
(2-38)
r=1
Expansion Theorem - Frequency Domain:

{ X( i ) } =
r { r }
(2-39)
r=1
Using the frequency domain form of the expansion theorem, if the response is evaluated at the
undamped natural frequency of mode r, the expansion coefficient r will dominate and be
approximately equal to alpha defined in Equation 2-37.
(2-39)
+UC-SDRL-RJA
CN-20-263-662
3. MULTIPLE DEGREE OF FREEDOM SYSTEMS

3.1 Theory
Generally, most structures are more complicated than the single mass, spring, and damper system
discussed in the previous section. The general case for a multiple degree of freedom system will
be used to show how the frequency response functions of a structure are related to the modal
vectors of that structure. Throughout the following section the following two degree of freedom
system will be used to illustrate the concepts discussed.
K1
x 1 (t)
x 2 (t)
f 1 (t)
f 2 (t)
K2
M1
C1
K3
M2
C2
Figure 3-1. Two Degree of Freedom System
The equations of motion for the above system are:
(3-1)
C3
+UC-SDRL-RJA
CN-20-263-662
M 1 x 1 (t) + (C 1 + C 2 ) x 1 (t) C 2 x 2 (t) + (K 1 + K 2 ) x 1 (t) K 2 x 2 (t) = f 1 (t)

M 2 x 2 (t) + (C 2 + C 3 ) x 2 (t) C 2 x 1 (t) + (K 2 + K 3 ) x 2 (t) K 2 x 1 (t) = f 2 (t)
In matrix notation:
0 x 1 (t) ( C 1 + C 2 )
+
M 2 x 2 (t)
C 2
M1
( K1 + K2 )
+
K 2
K 2
( K2 + K3 )
C 2
( C2 + C3 )
x 1 (t)
x 2 (t)
x 1 (t) f 1 (t)
x 2 (t) f 2 (t)
(3.1)
The above equations are still second order, linear, time invariant, differential equations, but are
now coupled by the coordinate choice. Therefore, this system of equations must be solved
simultaneously. The process of solving the set of equations in Equation 3.1 will now be
reviewed in an analytical sense. The modal vectors and frequencies will result as the solution to
the homogeneous portion of the differential equations summarized in Equation 3.1.
The solution of the above system of second order differential equations is first obtained for the
undamped system. Assuming that C 1 = C 2 = C 3 = 0.
} + [ K ] { x(t) } = { f (t) }
[ M ] { x(t)
where:
M
[ M ]= 1
0
( K1 + K2 )
[ K ]=
K 2
f (t)
{ f }= 1
= Forcing Vector
f 2 (t)
0
M2
= Mass Matrix
K 2
= Stiffness Matrix
( K2 + K3 )
(3-2)
(3.2)
+UC-SDRL-RJA
CN-20-263-662
x (t)
{ x }= 1
= Response Vector
x
(t)
2
Since the forcing and response vectors are always functions of time, the functional notation (t)
will be dropped in further equations.
The system of equations represented by Equation 3.2 has the general solution of:
{ x } = { X } es t
Thus:
{ x } = s { X } e s t = s x
{ x } = s2 { X } e s t = s2 x
(3.3)
where:
s = + j = complex valued frequency
Substituting Equation 3.3 into Equation 3.2 yields:

s2 [ M ] { X } + [ K ] { X } = { f }
If there are no forcing function so that { f } = { 0 }, then:
s2 [ M ] { X } + [ K ] { X } = { 0
2
s [ M ]+[ K ] { X }={ 0
}
}
(3.4)
Equation 3.4 is nothing more than a set of simultaneous algebraic equations in X i . The
unknowns are the Xs and the ss. From the theory of differential equations, in order for
Equation 3.4 to have other than the trivial solution, { X } = { 0 }, the determinant of the
coefficients must equal zero. The determinant of the coefficients will be a polynominal in s2 .
The roots of this polynominal are called eigenvalues.
(3-3)
+UC-SDRL-RJA
CN-20-263-662
In order to manipulate Equation 3.4 into a standard eigenvalue-eigenvector form, Equation 3.4
can be reformulated in a couple of different ways. First, divide Equation 3.4 by s2 and
premultiply by [K ]1 .
1
1
[ K ] [ M ]+ 2 [ I
s
] { X }={ 0
(3.5)
A different way of formulating the eigenvalue problem would be to premultiply Equation 3.4 by
[M]1 . Note that by doing this, the resulting dynamic matrix, [ K ]1 [ M ] in Equation 3.5 or
[ M ]1 [ K ] in Equation 3.6, is no longer symmetric.
[ M ]1 [ K ] + s2 [ I
] { X }={ 0
(3.6)
1
and in Equation 3.6 the eigenvalues are s2 . Equations 3.5
2
s
and 3.6 are really just the inverse of each other. In Equation 3.5 or Equation 3.6, the matrix on
the left hand side of the equation is often referred to as the dynamic matrix. Note that the
multiplication of Equation 3.4 by a matrix to obtain Equation 3.5 or 3.6 amounts to a coordinate
transformation.
In Equation 3.5 the eigenvalues are
The frequency of a mode of vibration is defined in terms of the eigenvalue. The solution vector
{X } of Equation 3.5 or 3.6 corresponding to a particular eigenvalue is called an eigenvector,
characteristic vector, mode shape, or modal vector. The Xs represent a deformation pattern of
the structure for a particular frequency of vibration. Since Equations 3.5 or 3.6 are homogeneous
there is not a unique solution for the Xs; only a relative pattern or ratio among the Xs can be
obtained. In other words, the Xs can only be solved for in terms of one of the Xs, which in turn
can be given any arbitrary value. Mathematically, the rank of the equation systems represented
by Equation 3.5 or 3.6 is always one less than the number of equations.
Therefore, the deflected deformation of a structure, which describes a natural mode of vibration,
is defined by known ratios of the amplitude of motion at the various points on the structure.
Thus, the actual amplitude of vibration of a structure is a combination of the modal vector and
the level, location, and characteristic of excitation forces and not a direct property of a natural
mode of vibration. The amplitude of vibration is really dependent on the placement and
(3-4)
+UC-SDRL-RJA
CN-20-263-662
amplitude of the systems forcing functions along with any initial conditions of the system
together with the properties of the structure described by the eigenvalues and eigenvectors.
3.2 Solution of the Eigenvalue Problem

The solution of either Equation 3.5 or 3.6 is obtained by recognizing that these equations are a
set of homogeneous equations. Therefore, for a non-trivial solution, the determinant of the
coefficients must equal zero.
[ K ] + s2 [ M ] = 0.
(3.7a)
[ M ]1 [ K ] + s2 [ I
] = 0.
(3.7b)
1
1
[ K ] [ M ]+ 2 [ I
s
] = 0.
(3.7c)
The determinant in Equation 3.7 is referred to as the characteristic determinant. The expansion
of the characteristic determinant results in the characteristic equation or the
frequency equation.
Equation 3.7 may be rewritten as:
n + a1 n1 + a2 n2 + . . . + a n = 0.
(3.8)
Equation 3.8 is the characteristic equation of a N -degree of freedom system, where = s2 for
1
Equation 3.7a or 3.7b or = 2 for Equation 3.7c. The roots of Equation 3.8 are the eigenvalues
s
of the system. Note that the values of s corresponding to the roots of Equation 3.8 are the
complex-valued modal frequencies ( r = r + j r ).
(3-5)
+UC-SDRL-RJA
CN-20-263-662
3.2.1 Two Degree of Freedom Example: Undamped, Unforced

Given a two degree of freedom system (Equation 3.1), find its eigenvalues (undamped natural
frequencies) and the respective eigenvectors (modal vectors) for the undamped system.
Referring to Figure (3-1), let:
M1 = 5
M 2 = 10
K1 = 2
K2 = 2
K3 = 4
Substituting into Equation 3.1:

5
0 x 1 4
+
10 x 2 2
2 x 1 0
=
6 x 2 0
The eigenvalue problem then becomes (Equation 3.5):

4
2
3
10
1
10
1
10
1
5
X 1 = 0
X2 0
0 1 1
+
10 s2 0
0 1 1
+
10 s2 0
X1 0
0

=
1 X2 0
or:
1
3
2 + s2
1
1
2+ 2
s
X1
={ 0
X
2
(3-6)
(3.9)
+UC-SDRL-RJA
CN-20-263-662
The determinant of the coefficient matrix of Equation 3.9 must equal zero for a non-trivial
solution.
(
Using =
1
1
1
3
+ 2 ) ( 2 + 2 ) = 0.
2 s
s
2
1
as a change of variable, the characteristic equation becomes:
s2
2 +
7
5
+ = 0.
2
2
(3.10)
The roots of Equation 3.10 are:

7
49/4 10 7
9/4
= 2
=
2
4
2
1,2
1 =
5
2
2 = 1
Noting the change of variable =
1
:
s2
1 =
2 =
21
1
22
Since r = r j r , the complex-valued modal frequencies are:

1 = 1 j 1 = j 1 = j
2/5
2 = 2 j 2 = j 2 = j 1
Now the frequencies 1 and 2 can be used in Equation 3.9 to determine the modal vectors.
The modal vector for 1 = j 1 is determined using the following equations:
(3-7)
+UC-SDRL-RJA
CN-20-263-662
1
1
1 X1
={ 0
1
X 2
2
or
X1 + X2 = 0
X2 = X1
Thus, the modal vector corresponding to the natural frequency 1 is:
X
{ }1 = 1
X 1 1
where:
X 1 is arbitrary (depends on scaling method)
Similarly for 2 = j 2 , the modal vector is:
1
2
1
2
1 X
1
={ 0
X
2
1
or
1
X + X2 = 0
2 1
X2 =
1
X
2 1
or:
X1
{ }2 = X 1
2
(3-8)
+UC-SDRL-RJA
CN-20-263-662
If the deformation of X 1 = 1, which is an arbitrary choice depending on the scaling method, then:
2/5 :
For 1 =
1
{ }1 =
1 1
For 2 = 1 :
1
{ }2 = 1
2
3.3 Weighted Orthogonality of Modal Vectors

The solution of the eigenvalue problem as formulated in Equation 3.5 yields N natural
frequencies, r , and N modal vectors { }r where N is the number of degrees of freedom of the
system.
Note that any particular undamped natural frequency and the associated modal vector { }r
satisfy Equation 3.4. Thus, substituting into Equation 3.4 s = r and {X } = { }r yields:
2r [ M ]{ }r = [ K ] { }r
(3.11)
Now pre-multiply Equation 3.11 by a different modal vector, { }s T , thus:

2r { }Ts [ M ] { }r = { }Ts [ K ] { }r
where the superscript T denotes a matrix transpose.

Using a rule of matrix algebra for the transpose of a product of matrices:
[ [ C ] [ D ] ]T = [ D ]T [ C ]T
(3-9)
(3.12)
+UC-SDRL-RJA
CN-20-263-662
Taking the transpose of both sides of Equation 3.12 yields:

2r { }Tr [ M ] { }s = { }Tr [ K ] { }s
(3.13)
where:
[ M ]T = [ M ] since [ M ] is a symmetric matrix.
[ K ]T = [ K ] since [ K ] is a symmetric matrix.
Next, substitute s = s and {X } = { }s into Equation 3.4 and pre-multiply both sides by { }s T .
This yields:
2s { }Tr [ M ] { }s = { }Tr [ K ] { }s
(3.14)
Subtracting Equation 3.14 from Equation 3.13 gives:

( 2r 2s ) { }Tr [ M ] { }s = 0.
(3.15)
If r s (implying two different frequencies), it follows that:

{ }Tr [ M ] { }s = 0.
(3.16)
From Equation 3.14, it follows that:

{ }Tr [ K ] { }s = 0.
(3.17)
Equations 3.16 and 3.17 are statements of the weighted orthogonality properties of the modal
vectors with respect to the system mass and stiffness matrices. The concept of orthogonality can
be looked at from a vector analysis standpoint. In vector analysis, two vectors are orthogonal if
their dot product equals zero. This means that the projection of one vector on the other is zero.
Therefore, the two vectors are perpendicular to each other. An obvious example is the
3-dimensional cartesian coordinate system. The i, j, and k unit vectors for the cartesian
coordinate system are orthogonal to each other. Modal vectors of an n-degree of freedom system
can be viewed as being just a vector in n-dimensional space, which unfortunately cannot be
(3-10)
+UC-SDRL-RJA
CN-20-263-662
visualized. In order for modal vectors to be orthogonal, though, a simple dot product will not
suffice. The concept of a weighted dot product, where the weighting matrix is the theoretical
mass or stiffness matrix, must be used. If, for instance, the mass matrix in Equation 3.16 was the
identity matrix, the weighted dot product would reduce to the simple dot product and result in a
direct analog of the orthogonality condition for the unit vectors in the cartesian coordinate
system. Because the mass and stiffness matrices in Equation 3.16 and 3.17 are not generally the
identity matrix, the orthogonality relationships in Equation 3.16 and Equation 3.17 are generally
referred to as weighted orthogonality.
If two modal vectors happen to have the same frequency r = s (Equation 3.15) their
corresponding modal vectors are not necessarily orthogonal to one another. This condition is
known as a repeated root or repeated pole and will be discussed further in a later section. For
this condition, the modal vectors associated with the repeated roots will be orthogonal to the
other modal vectors and independent of one another.
In Equation 3.15, if the same modal vector is used to pre- and post-multiply the mass matrix,
then Equation 3.16 is equal to some scalar constant other than zero, commonly noted as M r .
Thus:
{ }Tr [ M ] { }r = M r = Modal Mass
(3.18)
Similarly, Equation 3.14 yields:

{ }Tr [ K ] { }r = r2 M r = K r = Modal Stiffness
(3.19)
Since, as previously shown, the amplitude of any particular modal vector (eigenvector) is
completely arbitrary, the modal vector can be normalized in an arbitrary way. This means that
M r is not unique.
For instance, one common criteria used to normalize the modal vector is to scale the modal
vector such that M r in Equation 3.18 is equal to unity.
The resulting scaled modal vectors normalized in this manner are generally referred to as
orthonormal modal vectors (eigenvectors).
(3-11)
+UC-SDRL-RJA
CN-20-263-662
3.4 Modal Vector Scaling Example

Using the previous two degree of freedom example, normalize the modal vectors { }1 and { }2
such that:
{ }T1 [ M ] { }1 = M 1 = 1.
(3.20)
{ }T2 [ M ] { }2 = M 2 = 1.
(3.21)
and:
From the previous example:
X1
{ }2 = X 1
2
X
{ }1 = 1
X 1 1
Substituting { }1 into Equation 3.20 yields:

T
X1 5

X1 0
0 X1
=1
10 X 1
T
5 X1 X1

=1
10
X
X
1
1
5 X 12 + 10 X 12 = 1
X 12 =
1
15
1/15
X1 =
Using the positive root, the modal vector { }1 , normalized for unity modal mass, results:
X
1/15
{ }1 = 1 =
1/15
X 1 1
(3-12)
+UC-SDRL-RJA
CN-20-263-662
Similarly for { }2 :
X1
X1
2
5

0
X1
0
X1
10
2
T
5 X1 X1
X1
5 X 1 2
5 X 12 +
=1
=1
5 2
X =1
2 1
X 12 =
2
15
2/15
X1 =
Thus, { }2 normalized to unity modal mass is:
2/15

{ }2 =
2/15
The normalized modal vectors will give M r = 1 for all the modes of vibration. The significance
of this normalization will be obvious later.
3.5 Principal Coordinates - Modal Coordinates

With reference to the equations of motion for an undamped system (Equation 3.2), the major
obstacle encountered when trying to solve for the system response {x}, due to a particular set of
forcing functions and initial conditions, is the coupling between the equations. In terms of the
systems mass and stiffness matrices, coupling is represented in terms of non-zero off diagonal
elements. Generally two types of coupling can exist for an undamped system; (1) Static coupling
(3-13)
+UC-SDRL-RJA
CN-20-263-662
(non-diagonal stiffness matrix); or (2) dynamic coupling (non-diagonal mass matrix). Equation
3.2 represents a system which is only statically coupled. If the system of equations in Equation
3.2 could be uncoupled, that is diagonal mass and stiffness matrices, then each equation in
Equation 3.2 could be solved independent of the other equations. Another way of looking at this
would be that each uncoupled equation would look just like the equation for a single degree of
freedom, whose solution can very easily be obtained. Therefore, if a set of coupled system
equations could be reduced to an uncoupled system, the solution would become
straightforward. Indeed, from an analytical sense, this is the whole point of what has become
known as modal analysis.
The procedure used to uncouple a set of coupled system equations is basically a coordinate
transformation. In other words, the goal is to find a coordinate transformation that transforms the
original coordinates {x} into another equivalent set of coordinates {q } that renders the system
statically and inertially decoupled. This new set of coordinates {q } is typically referred to as
principal coordinates, normal coordinates or modal coordinates.
A similar benefit of a coordinate transformation occurs in many other engineering problems.
One example of this situation is in the calculation of moments and products of inertia when the
inertia properties of a complex structure need to be defined. The first step in the calculation of
the inertia properties is to choose a set of axis to base the inertia properties on. Then, the
following properties would be measured or calculated: I xx , I yy , I zz , I xy , I yz , I xz . In general, both
moments of inertia and products of inertia are required. However, if a different set of axis with
respect to the structure were defined such that these axis happened to coincide with the structures
principle axis, the result would be moments of inertia I x , I y , and I z but the products of inertia
would all be zero (I xy = I yz = I xz = 0). Therefore, by changing the coordinate system, the
products of inertia have been eliminated.
Another example of the benefit of a coordinate transformation is noticed when computing
principle strains at a point on a structure. Typically, a strain gage rosette is used to determine the
normal and shearing strains at a point of interest. From this information, a new coordinate
system can be determined (strain element orientation) such that only principal normal strains
exist; the shear strains are equal to zero for the new coordinate system. To determine the
orientation of this new coordinate system that renders the shearing strain to zero, MOHRs circle
techniques are commonly used. Once again, a simple coordinate transformation is used to
(3-14)
+UC-SDRL-RJA
CN-20-263-662
eliminate the shearing strains.

The problem of finding a coordinate transformation that uncouples our original equations of
motion is very straightforward. It turns out that, due to the unique orthogonality properties of the
modal vectors, the required coordinate transformation is already available. Referring to
Equations 3.16-3.19, if either the mass or stiffness matrix is pre- and post- multiplied by
different modal vectors (Equations 3.16 and 3.17), the result is zero. However, if the same modal
vector is used to pre- and post- multiply the mass or stiffness matrix (Equations 3.18 and 3.19),
the result is a constant.
Therefore, the new coordinate system can be defined by the following transformation:
{ x }= { q }
(3.22)
where:
[ ]
is the transformation matrix (matrix whose columns are the modal vectors of the
original system).
This matrix is generally referred to as the modal matrix or matrix of modal vectors. Recall the
general form of the undamped system of equations with forcing functions:
[ M ] { x } + [ K ] { x } = { f }
(3.23)
Substituting Equation 3.22 into Equation 3.23 gives:

[ M ] { q } + [ K ] { q } = { f }
(3.24)
Pre-multiplying by yields:

T
[ M ] { q } + [ K ] { q } = { f }
Equation 3.25 is the equivalent of Equation 3.23 but in a different coordinate system. Analyzing
(3-15)
+UC-SDRL-RJA
CN-20-263-662
Equation 3.25, noting the orthogonality properties of the modal vectors (Equation 3.16-3.19):
T
[ M ] = M
and:
T
[ K ] = K
where:
M is a diagonal matrix.
K is a diagonal matrix.
Therefore, Equation 3.25 becomes:

T
M { q } + K { q } = { f }
(3.26)
From inspection, since both the new mass and stiffness matrices are diagonal, the coordinate
transformation {x} = [ ] {q } has completely uncoupled the set of equations. Now each equation
in Equation 3.26 is an equation for a single degree of freedom oscillator which is easily solved.
The r-th Equation of Equation 3.26 is:
M r q r + K r q r = { }Tr { f } = f r
This is the equation of motion for the single degree of freedom system shown below.
(3-16)
(3.27)
+UC-SDRL-RJA
CN-20-263-662
q r (t)
f (t)
Kr
Mr
Figure 3-2. Single Degree of Freedom System
The quantity M r is called the modal mass or generalized mass for the r-th mode of vibration.
The quantity K r is called the modal stiffness or generalized stiffness for the r-th modal vector of
vibration. While these quantities are viewed as mass and stiffness related, it is important to
remember that the magnitude of these quantities depends upon the scaling of the modal vectors.
Therefore, although both the modal vectors and the modal mass/stiffness quantities are computed
in a relative manner, only the combination of a modal vector together with the associated modal
mass represents a unique absolute characteristic concerning the system being described.
It has been shown previously that the modal vectors may be normalized such that M r = 1. If this
has been done, then Equation 3.27 can be rewritten for the r-th modal vector as:
(3-17)
+UC-SDRL-RJA
CN-20-263-662
q r + 2r q r = f
(3.28)
where:
M r = 1. 0
K r = r 2
Once the solution (time responses) of Equation 3.26 for all qs has been computed, the solution
in terms of the original coordinates can then be obtained through the use of the coordinate
transformation equation in Equation 3.22.
3.6 Two Degree of Freedom Example: Undamped, Forced

Referring to the previous example, some forcing functions can now be included in the system of
equations.
5
4
0
{ x } +
10
2
2
{ x } = { f }
6
The natural frequencies and normalized modal vectors of the above system are:
2/5 :
For 1 =

1/15
{ }1 =
1/15
For 2 = 1 :
2/15

{ }2 =
2/15
(3-18)
(3.29)
+UC-SDRL-RJA
CN-20-263-662
Forming the modal matrix:

= { }
1

1/15
=

1/15
{ }2
2/15
2/15
Now make the following coordinate transformation:

{ x }= { q }
1/15
x 1
=
1/15
x 2
2/15
2/15
(3.30)
q1
q2
Substituting Equation 3.30 into Equation 3.29 and pre-multiplying by yields:
[ M ] { q } + [ K ] { q } = { f (t) }

T
1/15
[ M ] =
2/15
1/15
2/15
5
0
T
[ M ] = 1
(3-19)

1/15
0
10
1/15
2/15
2/15
+UC-SDRL-RJA
CN-20-263-662

T
1/15
[ K ] =

2/15
1/15
2/15
4
2
[ K ] = 2/5

1/15
2
6
1/15
2/15
2/15
Therefore, the new equations of motion are:

1
0 q 1 2/5
+
1 q 2 0
1/15 f 1 +
1/15 f 2
0 q1
=

2/15
1 q 2
f2
2/15 f 1
2
f 1
=
f2
The matrix equation of Equation 3.31 can now be written in terms of algebraic differential
equations:
2
q = f 1
5 1
(3.32)
q 2 + q 2 = f 2
(3.33)
q 1 +
Hence, the system equations have been uncoupled by using the modal matrix as a coordinate
transformation.
The original system looked like:
(3-20)
+UC-SDRL-RJA
CN-20-263-662
x 1 (t)
x 2 (t)
f 1 (t)
f 2 (t)
10
Figure 3-3. Original System
The transformed system can be pictured as:
(3-21)
+UC-SDRL-RJA
CN-20-263-662
2/5
q 1 (t)
q 2 (t)
f 1 (t)
f 2 (t)
Figure 3-4. Transformed System
Once q 1 (t) and q 2 (t) are known, Equation 3.30 can be used to compute x 1 (t) and x 2 (t). Thus:
x 1 (t) =
1/15 q 1 (t) +
2/15 q 2 (t)
x 2 (t) =
1/15 q 1 (t)
2/15
q 2 (t)
2
Many points should be emphasized from the previous discussion. Modal vectors, along with
their frequencies, are a dynamic property of a structure. The amplitudes of a modal vector are
completely arbitrary; that is, only the ratios between the components of a particular modal vector
are unique. Because of the orthogonality properties of the modal vectors, with respect to the
systems mass and stiffness matrices, modal mass and modal stiffness can be defined. These
quantities depend upon the scaling of the modal vectors, so that the absolute magnitudes of these
quantities are also arbitrary. Finally, a simple coordinate transformation (modal matrix) can be
(3-22)
+UC-SDRL-RJA
CN-20-263-662
used to represent a complicated interconnection of springs and masses as a collection of single

degree of freedom oscillators.
3.7 Proportional Damping

In order to evaluate multiple degree of freedom systems that are present in the real world, the
effect of damping on the complex frequencies and modal vectors must be considered. Many
physical mechanisms are needed to describe all of the possible forms of damping that may be
present in a particular structure or system. Some of the classical types are:
Structural Damping
Viscous Damping
Coulomb Damping
Hysteretic Damping
It is generally difficult to ascertain which type of damping is present in any particular structure.
Indeed most structures exhibit damping characteristics that result from a combination of all the
above, plus others that have not been described here.
It will suffice to say that whenever a structure is modeled with a particular form of damping, for
example, viscous, that the damping model is an equivalent model to whatever type of damping
that may actually be present.
Rather than consider the many, different physical mechanisms, the probable location of each
mechanism, and the particular mathematical representation of the mechanism of damping that is
needed to describe the dissipative energy of the system, a model will be used that is only
concerned with the resultant mathematical form. This model will represent a hypothetical form
of damping, that is proportional to the system mass or stiffness matrix. Therefore:
[ C ] = [ M ]
or :
[ C ] = [ K ]
(3-23)
+UC-SDRL-RJA
CN-20-263-662
The most common formulation for proportional damping is:

[ C ] = [ M ] + [ K ]
where:
[ C ] = damping matrix
, = constants
Note that the case of no damping is the trivial proportional damped case with both coefficients
equal to zero. While the above definition is sufficient for most cases, the theoretical relationship
between mass, stiffness and damping matrices can be somewhat more complicated and still
qualify as proportional damping. Theoretically, any damping matrix that satisfies the following
relationship will yield proportional damping with all the qualifications (normal modes) involved
in subsequent discussion.
s
[M]1 [C] [M]1 [K ] = [M]1 [K ] [M]1 [C]
where:
r and s = integers.
For the purposes of most practical problems, the simpler relationship will be sufficient.
3.8 Modal Vectors from the System Matrix

The modal vectors can be determined in a somewhat more direct manner through a manipulation
of the system matrix. Understanding this approach to the evaluation of modal vectors is very
useful in relating measured frequency response function data to the system modal vectors.
Starting with Equation 3.4:
[ M ]s2 + [ C ]s + [ K ] { X } = { 0
(3-24)
(3.34)
+UC-SDRL-RJA
CN-20-263-662
Define:
[ B ( s ) ] = [ M ]s2 + [ C ]s + [ K ]
where:
[ B(S) ] = System Impedance Matrix
From matrix algebra:

[ B ( s ) ] [ B ( s ) ]1 = [ I
[ B ( s ) ]1 =
(3.35)
[ B ( s ) ]A
|[B(s)]|
(3.36)
where:
[ B(s) ] A is the adjoint of matrix [ B(s) ].
Substituting Equation 3.36 into Equation 3.35 yields:

[ B ( s ) ] [ B ( s ) ] A = |[ B ( s ) ]| [ I
(3.37)
If r is a root of the characteristic equation from Equation 3.34, then | [ B( r ) ] | = 0.

Evaluating Equation 3.37 at s = r gives:
A
B( ) B( ) =[ 0
r
r
(3.38)
Equation 3.38 can be rewritten using any column of [ B( r ) ] A , the i th column for example
{ B( r ) }i A . Therefore:
B( ) { B( )
r
r
(3-25)
}iA = { 0 }
(3.39)
+UC-SDRL-RJA
CN-20-263-662
Equation 3.39 represents a set of homogeneous equations in { B( r ) }i A which determines each

element of { B( r ) }i A to within an arbitrary constant. Note that the constant will be different
depending upon the column that is used.
Evaluating Equation 3.34 at one of the eigenvalues of the system ( r ):
B( ) { X } ={ 0 }
r
r
(3.40)
Equation 3.40 (formerly Equation 3.34), just like Equation 3.39, represents a set of homogeneous
equations in { X } Equation 3.34 is evaluated at a specific eigenvalue, the resulting solution is the
eigenvector coresponding to the specific eigenvalue. This eigenvector is determined to within an
arbitrary constant. Therefore, from Equation 3.39 and Equation 3.40, { B( r ) }i A and {X }r are
proportional and both represent the eigenvector corresponding to the eigenvalue r . Recall that
{X }r (Equation 3.40) has been previously shown to be the r-th modal vector of the system.
Therefore:
{ X }r = ir { B ( r )
}iA
where:
ir
is a proportionality constant.
Note: One of the major points is that the columns of the adjoint matrix [ B( r ) ] A are all
proportional to the r-th modal vector.
Since the mass, damping and stiffness matrices are assumed to be symmetric when absolute
coordinates are used (and proportional damping is present), the system impedance matrix
[ B(s) ] is symmetric. Therefore, the adjoint matrix of [ B( r ) ] is also symmetric. Thus, the
rows of the adjoint matrix are also proportional to the modal vector. The adjoint matrix can now
be expressed for the r-th mode in terms of the r-th modal vector.
A
B ( ) = { } { }T
r
r
r
r
(3-26)
(3.41)
+UC-SDRL-RJA
CN-20-263-662
1 1 1 2

2 2
2 1
A
B ( ) =
.
.
r
r
.
.
N 1 N 2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1 N
.
.
.
N N
where:
= constant associated with the scaling of { }r relative to the absolute scaling (units) of
the adjoint matrix.
Note that the adjoint matrix is not the same as the modal matrix since each column of the adjoint
matrix is proportional to the same modal vector. Therefore, the adjoint matrix needs to be
evaluated for each of the N eigenvalues to determine the N eigenvectors. Also note that, due to
the symmetry of the adjoint matrix, if one of the modal coefficients is zero, then a complete row
and column of the corresponding adjoint matrix will be zero. This phenomenon is normal and
corresponds to physically trying to excite (force) the system at the node (modal coefficient equal
to zero) of one of the modal vectors of the system. Theoretically, the corresponding mode of
vibration will not be observed in this situation. Analytically, this problem can be overcome by
evaluating a different row or column of the adjoint matrix. Experimentally, the configuration of
the input and/or output sensors may have to be altered to detect this case.
Equation 3.41 is extremely important and will be used in the next section to show that the
residues of a frequency response function for a particular pole ( r ) are directly related to the
elements of a modal vector. Also, the symmetry of the adjoint matrix is the justification for not
needing to evaluate the complete frequency response function matrix.
(3-27)
+UC-SDRL-RJA
CN-20-263-662
4. FREQUENCY RESPONSE FUNCTION DEVELOPMENT

4.1 Theory
All of the techniques discussed previously are useful if an analytical model of the system
already exists. From an experimental point, this is rarely the case. Typically, solving problems
on real systems or pieces of systems must be accomplished without the aid of a theoretical model
or in order to verify a theoretical model.
In this chapter, frequency response function measurements will begin to be used as the basis for
defining modal frequencies and damping values, modal vectors, modal mass, modal stiffness, and
modal damping of real life structures. To accomplish this task, an analytical model will be
developed to represent the transfer function between any possible measurement locations on the
structure. Frequency response functions will be directly related to the transfer functions that
have been theoretically developed.
The transfer function representation of an undamped multiple degree of freedom system can be
formulated by starting with the differential equations of motion in terms of mass, stiffness, and
damping matrices.
[ M ] { x } + [ K ] { x } = { f }
(4.1)
Taking the Laplace transform of Equation 4.1, assuming all initial conditions are zero, yields:
s2 [ M ] + [ K ] { X (s) } = { F (s) }
(4.2)
Let:
[ B (s) ] = s2 [ M ] + [ K ]
Then Equation 4.2 becomes:

[ B (s) ] { X (s) } = { F (s) }
(4-1)
(4.3)
+UC-SDRL-RJA
CN-20-263-662
where [ B(s) ] is referred to as the system impedance matrix or just the system matrix.
Pre-multiplying Equation 4.3 by [ B(s) ]1 yields:
[ B (s) ]1 { F (s) } = { X (s) }
Defining:
[ H (s) ] = [ B (s) ]1
Then:
[ H (s) ] { F (s) } = { X (s) }
(4.4)
Equation 4.4 relates the system response { X(s) } to the system forcing functions { F(s) }
through the matrix [ H(s) ]. The matrix [ H(s) ] is generally referred to as the transfer function
matrix.
Figure 4-1. Two Input-Output Model
Equations 4.2-4.4 can be expanded for a two degree of freedom system in order to view in detail
the components of each position in the system matrix.
(4-2)
+UC-SDRL-RJA
CN-20-263-662
M1 1
M2 1
M 1 2 x 1 K 1 1
+
M 2 2 x 2 K 2 1
(4-3)
K1 2 x1 f1
K2 2 x2 f2
(4.5)
+UC-SDRL-RJA
CN-20-263-662
Taking the Laplace transform of Equation 4.5 yields:

M1 1

M2 1
M1 2 2 K1 1
s +
M2 2
K2 1
K 1 2 X 1 (s) F 1 (s)

=
K 2 2 X 2 (s) F 2 (s)
Thus,
[ B (s) ] { X (s) } = { F (s) }
where:
M s2 + K 1 1
[ B (s) ] = 1 1 2
M2 1 s + K2 1
M 1 2 s2 + K 1 2
M 2 2 s2 + K 2 2
Defining the transfer function as the inverse of the impedance matrix:

[ B (s) ]1 = [ H (s) ]
The inverse of the impedance matrix can be found for this analytical case as the adjoint of the
impedance matrix divided by the determinant of the impedance matrix as follows:
[ B (s) ]1 =
(M 1 1
M 2 2 s2 + K 2 2
( M 1 2 s2 + K 1 2 )
2
M 1 1 s2 + K 1 1
( M 2 1 s + K 2 1 )
s2 + K 1 1 ) (M 2 2 s2 + K 2 2 ) (M 2 1 s2 + K 2 1 ) (M 1 2 s2 + K 1 2 )
(4.6)
Note that the denominator of Equation 4.6 is | B(s) | which is the characteristic or frequency
equation for the system. This highlights the fact that the complex-valued modal frequencies ( r )
are global properties of the system since this characteristic equation appears in every term of
[ H(s) ]. This characteristic equation can be expressed as a product of its roots, thus:
| B (s) | = E ( s 1 ) ( s 2 ) ( s 3 ) ( s 4 )
where:
(4-4)
+UC-SDRL-RJA
CN-20-263-662
E = Constant coeficient of the highest order term in the polynomial (Product/Sum of the
mass terms)
1, 2, 3
and 4 are the 4 roots of the characteristic equation.
Equation 4.6 for a two degree of freedom system can then be written as:
M 2 2 s2 + K 2 2
( M 1 2 s2 + K 1 2 )
2
M 1 1 s2 + K 1 1
X 1 (s) ( M 2 1 s + K 2 1 )
=
X
(s)
E ( s 1 ) ( s 2 ) ( s 3 ) ( s 4
2
F 1 (s)
) F 2 (s)
(4.7)
[ H(s) ], the transfer function matrix, can be defined as:

H (s)
[ H (s) ] = 1 1
H 2 1 (s)
H 1 2 (s)
H 2 2 (s)
where, for instance:

M 2 2 s2 + K 2 2
H 1 1 (s) =
E ( s 1 ) ( s 2 ) ( s 3 ) ( s 4 )
(4.8)
Equation 4.4 can now be expressed as:

H 1 1 (s)
H 2 1 (s)
H 1 2 (s) F 1 (s) X 1 (s)
H 2 2 (s) F 2 (s) X 2 (s)
(4.9)
Multiplying out Equation 4.9 yields:

H 1 1 (s) F 1 (s) + H 1 2 (s) F 2 (s) = X 1 (s)
(4.10)
H 2 1 (s) F 1 (s) + H 2 2 (s) F 2 (s) = X 2 (s)

If, in Equation 4.10, F 2 (s) = 0, then:
(4-5)
+UC-SDRL-RJA
CN-20-263-662
H 1 1 (s) F 1 (s) = X 1 (s)

H 2 1 (s) F 1 (s) = X 2 (s)
(4-6)
+UC-SDRL-RJA
CN-20-263-662
This results in the familiar relationships for the transfer function (output over input):
H 1 1 (s) =
X 1 (s)
F 1 (s)
H 2 1 (s) =
X 2 (s)
F 1 (s)
In general
H pq =
Xp
Fq
where :
p is the output degree of freedom (physical location and orientation).
q is the input degree of freedom (physical location and orientation).
Therefore, measuring a column of the [ H ] matrix is accomplished by using a single, fixed

input (excitor system) with a roving response and measuring a row is accomplished by using a
roving input (hammer) and a single fixed response. It should be reiterated that the subscript
notation of p or q refers to both a physical location and also direction or orientation.
Thus, H 1 1 (s) is the transfer function measured by exciting the system with F 1 (s) and measuring
the response X 1 (s). Similarly H 2 1 (s) is the transfer function measured by exciting the system
with F 1 (s) and measuring the response at X 2 (s). Likewise, H 1 2 (s) and H 2 2 (s) can be measured
by exciting the system with F 2 (s), letting F 1 (s) = 0, and measuring the responses X 1 (s) and
X 2 (s).
As in the single degree of freedom case, the denominator polynominal in Equation 4.8 is called
the characteristic equation. Notice that all of the transfer functions that are represented in
Equation 4.6 have the same denominator polynomial. The roots of this denominator polynomial
(characteristic equation) are the modal frequencies of the system.
Since the coefficients of the characteristic equation are real, the roots will appear as complex
conjugate pairs. Rewriting Equation 4.8 with this in mind gives the following result:
(4-7)
+UC-SDRL-RJA
CN-20-263-662
M 2 2 s2 + K 2 2
H 1 1 (s) =
E ( s 1 ) ( s *1 ) ( s 2 ) ( s *2 )
(4.11)
where 1 , *1 , 2 , and *2 are the roots of the characteristic equation. The roots of the
characteristic equation are also referred to as the poles of the transfer function H 1 1 (s).
4.2 Analytical Model - Scalar/Matrix Polynomial (MDOF)

The general formulation of Equation 4.11 or of any of the transfer functions defined by an
anlytical mass, damping and stiffness matrix model can be written as a numerator polynomial of
the independent variable s divided by a denominator polynomial of the independent variable s.
Both polynomials involve coefficients that are different numerical combinations of the discrete
values of mass, damping, stiffness of the system. The roots of the denominator polynomial are
the modal frequencies of the system and are considered global properties of the system. The
roots of the numerator polynomial are the zeroes of the system and are local properties of the
system that depend upon the specific input-output relationship of the transfer function. The
general polynomial model for a single transfer function can be written as follows using scalar
coefficients:
X p (s)
(s)n + n1 (s)n1 + . . . + 1 (s)1 + 0 (s)0
= H pq (s) = n m
m (s) + m1 (s)m1 + . . . + 1 (s)1 + 0 (s)0
F q (s)
(4.12)
The previous model can be rewritten in a more concise form as follows:

n
X p (s)
=
F q (s)
k (s)k
k=0
m
k (s)k
k=0
Further rearrranging yields the following equation that is linear in the unknown and terms:
m
k (s)k X p (s) =
k=0
k (s)k F q (s)
k=0
(4.13)
This model can be generalized to represent the general multiple input, multiple output case as
(4-8)
+UC-SDRL-RJA
CN-20-263-662
follows using a matrix polynomial formulation:

m
[ ] (s)k {X(s)} =
k
k=0
[ ] (s)k {F(s)}
k
k=0
(4.14)
The previous models can be used to represent frequency response function (FRF) data by
limiting the s variable (s = j ) and applying a few matrix operations. If both sides of the above
equation are post multiplied by the hermitian (complex conjugate transpose) of the force vector
({F(s)} H ), the following equation results.
m
[ ] (s)k {X(s)}{F(s)} H =
k
k=0
[ ] (s)k {F(s)}{F(s)} H
k
k=0
In the above equation note that the vector products are the definition of the cross power
([G xf (s)] = {X(s)}{F(s)} H ) and auto power spectrum ([G ff (s)] = {F(s)}{F(s)} H ) when s = j
(after averaging).
m
[ ] (s)k G (s) =
[ ] (s)k G (s)
k
xf
k
ff
k=0
k=0
The above equation can be post multipled by the inverse of the auto power spectrum matrix.
m
[ ] (s)k G (s) G (s)

k
xf ff
k=0
[ ] (s)k [I ]
k
k=0
Finally, the above equation can be put in final form by noting that the product of the cross
spectrum matrix and the inverse of the auto spectrum matrix ([G xf (s)] [G ff (s)]1 ) is the definition
of the FRF matrix ([H(s)]) for the multiple input, multiple output case when s = j .
m
[ ] ( j )k [H( )] =
[ ] ( j )k [I ]
k
k
k=0
k=0
(4.15)
For a single input, single output case the above equation yields:
m
k ( j )k H pq ( ) =
k=0
(4-9)
k ( j ) k
k=0
(4.16)
+UC-SDRL-RJA
CN-20-263-662
4.3 Analytical Model - Partial Fraction (Residue)

Equation 4.11 can be represented very generally by expansion in terms of its partial fractions.
That is:
H 1 1 (s) =
X 1 (s)
c1
c2
c3
c4
=
+
+
+
*
F 1 (s) ( s 1 ) ( s 1 ) ( s 2 ) ( s *2 )
(4.17)
The constants c 1 c 4 can be found in a similar fashion as in the single degree of freedom case.
Equating Equation 4.11 and Equation 4.17 yields:
M 2 2 s2 + K 2 2
c1
c2
c3
c4
(4.18)
=
+
+
+
*
*
*
E (s 1 ) (s 1 ) (s 2 ) (s 2 ) (s 1 ) (s 1 ) (s 2 ) (s *2 )
Note that c 1 can be evaluated by multiplying Equation 4.18 by s 1 and evaluating the
expression at s = 1 . Thus:
M 2 2 21 + K 2 2
c1 =
= A111
E ( 1 *1 ) ( 1 2 ) ( 1 *2 )
(4.19)
In a similar fashion:
2
M 2 2 *1 + K 2 2
= c *1 = A*111
c2 =
E ( *1 1 ) ( *1 2 ) ( *1 *2 )
c3 =
M 2 2 22 + K 2 2
= A112
E ( 2 1 ) ( 2 *1 ) ( 2 *2 )
2
M 2 2 *2 + K 2 2
c4 =
= c *3 = A*112
*
*
*
*
E ( 2 1 ) ( 2 1 ) ( 2 2 )
Equation 4.17 becomes:
A1 1 1
A*1 1 1
A1 1 2
A*1 1 2
H 1 1 (s) =
+
+
+
( s 1 ) ( s *1 ) ( s 2 ) ( s *2 )
(4-10)
(4.20)
+UC-SDRL-RJA
CN-20-263-662
Thus, the transfer function of a two degree of freedom system has been represented by the sum
of two single degree of freedom systems (Section 2.5). This result can now be extrapolated to
apply to any number of degrees of freedom.
Rewriting Equation 4.20 in terms of this summation:
H 1 1 (s) =
A1 1 r
A*1 1 r
+
s *r
r=1 s r
2
(4.21)
As in a single degree of freedom case, the A1 1 r s are again referred to as the residues associated
with the poles r .
The rest of the transfer functions of the system can be expressed, using the same logic as the
development of Equation 4.21.
H 2 1 (s) =
A2 1 r
A*2 1 r
+
s *r
r=1 s r
(4.22)
H 1 2 (s) =
A1 2 r
A*1 2 r
+
s *r
r=1 s r
(4.23)
H 2 2 (s) =
A2 2 r
A*2 2 r
+
s *r
r=1 s r
(4.24)
Equation 4.9 can now be rewritten in terms of the above partial fraction expansion.
X (s)

1
=
X 2 (s)

A1 1 r
A*1 1 r
+
s *r
r=1 s r
2
A2 1 r
+
*
s r s r
r=1
A*2 1 r
A1 2 r
A*1 2 r
+
s *r F 1 (s)
r=1 s r
*
2 A
A
F (s)
22r
+ 2 2 r* 2
s r
r=1 s r
(4.25)
The transfer function matrix [ H(s) ] can also be rewritten in terms of partial fractions combining
Equations 4.21 through 4.24.
(4-11)
+UC-SDRL-RJA
CN-20-263-662
A1 1 1 A1 2 1 A*1 1 1 A*1 2 1 A1 1 2 A1 2 2 A*1 1 2 A*1 2 2
A2 1 1 A2 2 1 A*2 1 1 A*2 2 1 A2 1 2 A2 2 2 A*2 1 2 A*2 2 2
[ H (s) ] =
+
+
+
( s 1 )
( s *1 )
( s 2 )
( s *2 )
.
(4.26)
The numerator matrices in each term of the above equation are called the residue matrices. Note
that there is a separate matrix associated with each of the modal frequencies (poles) of the
system. Note also that the residue matrices associated with complex conjugate modal
frequencies are also complex conjugates.
Previously, the constant A1 1 1 has been evaluated (Equation 4.19). The constants A2 1 1 , A1 2 1 , and
A2 2 1 can now be evaluated in a similar fashion. Once this is done, the form of the first residue
matrix can be evaluated for pole 1 .
A2 1 1 =
( M 2 1 21 + K 2 1 )
E ( 1 *1 ) i( 1 2 ) ( 1 *2 )
A1 2 1 =
( M 1 2 21 + K 1 2 )
E ( 1 *1 ) ( 1 2 ) ( 1 *2 )
A2 2 1 =
( M 1 1 21 + K 1 1 )
E ( 1 *1 ) ( 1 2 ) ( 1 *2 )
The first term on the right of Equation 4.27 can be rewritten as follows:
M 2 2 21 + K 2 2
A1 1 1 A1 2 1
( M 1 2 21 + K 1 2 )
2
M 1 1 21 + K 1 1
A2 1 1 A2 2 1 = ( M 2 1 1 + K 2 1 )
*
*
( s 1 )
E ( 1 1 ) ( 1 2 ) ( 1 2 ) ( s 1 )
(4.27)
Note that the numerator matrix on the right side of Equation 4.27 is the adjoint of the system
matrix discussed in Section 3.8. Therefore, from Equation 3.41:
M 2 2 21 + K 2 2
2
( M 2 1 1 + K 2 1 )
1 1
( M 1 2 21 + K 1 2 )
=
1
M 1 1 21 + K 1 1
2 1
(4-12)
1 2
2 2
1
(4.28)
+UC-SDRL-RJA
CN-20-263-662
Plugging Equation 4.28 into Equation 4.27 gives:

1 1
A1 1 1 A1 2 1
1 2
1
2 2
2 1
A2 1 1 A2 2 1 =
1
*
*
( s 1 )
E ( 1 1 ) ( 1 2 ) ( 1 2 ) ( s 1 )
(4.29)
Finally, the relationship between the transfer function data (via the residue matrices) and the
modal vectors of the system can be established. Equation 4.29 shows that the residue matrix has
the same structure as the adjoint matrix: each column contains a redundant estimate of the same
modal vector, different only by a constant. This relationship is normally stated as follows:
For pole 1 :
A1 1 1
A2 1 1
1 1
A1 2 1
= Q1
A2 2 1
2 1
1 2
2 2
1
In general, for pole r :

A1 1 r
A2 1 r
1 1
1 1
A1 2 r
= Qr

= Qr
A2 2 r
2 r 2 r
2 1
1 2
2 2
r
(4.30)
where:
Q1 is a constant that is a function of the modal vector scaling and the absolute units of the
residue matrix.
1
mode shape for pole 1
Equation 4.30 indicates that, with respect to the residue matrix, every row and column of the
residue matrix contains the same modal vector multiplied by a component of the modal vector
and the scaling constant. This is an important result for the experimental case. Therefore,
assuming that all of the residues within a row or column are not perfectly zero, the modal vector
can be estimated from only one row or column of the residue matrix. This will completely define
(4-13)
+UC-SDRL-RJA
CN-20-263-662
the modal vector for that particular pole. In terms of the experimental requirements, in order to
estimate a particular element in the residue matrix, a frequency response function must be
measured. Under the assumption that every element in a particular row or column is not zero,
only N measurements will be required rather than N N . Once this information is measured, all
other terms in the residue matrix could be synthesized.
This previous discussion represents the MINIMUM requirements in terms of measurements. In
order to be certain that modal vectors are not missed or to utilize the redundant information in
different rows or columns, more measurements than one row or column are typically taken.
It should be pointed out that the constant of proportionality Q r in Equation 4.30 is not unique
since the constant will depend upon the choice of how the modal vector is scaled. This is
consistent with the concept that modal vectors represent relative motion between the degrees of
freedom. Note however that the residues are scaled quantities that depend upon the units of the
frequency response function(s).
4.4 Modal Vector Example

The transfer function matrix for the two degree of freedom system example used previously can
now be used to determine the modal vectors.
Substituting mass and stiffness values into Equation 4.6 gives:
( 10 s2 + 6 )
2
2
2
(5s +4)
[ H (s) ] =
( 5 s2 + 4 ) ( 10 s2 + 6 ) (2) (2)
( 10 s2 + 6 )
2
( 5 s2 + 4 )
[ H (s) ] =
50 ( s4 + 7/5 s2 + 2/5 )
The roots of the characteristic equation have previously been calculated as:
1 = j
2/5 ( rad/sec )
*1 = j
2/5 ( rad/sec )
(4-14)
+UC-SDRL-RJA
CN-20-263-662
2 = j ( rad/sec )
*2 = j ( rad/sec )
Therefore:
( 10 s2 + 6 )
2
2
2
(5s +4)
[ H (s) ] =
50 ( s j
2/5 ) ( s + j
2/5 ) ( s j ) ( s + j )
The transfer function H 1 1 (s) can now be represented in terms of its partial fraction expansion:
H 1 1 (s) =
H 1 1 (s) =
10 s2 + 6
50 ( s j
2/5 ) ( s + j
2/5 ) ( s j ) ( s + j )
A1 1 1
(s j
2/5 )
A1 1 1 =
A*1 1 1
(s+ j
2/5 )
j
2/5
12
A1 1 2 =
A1 1 2
A*1 1 2
+
(s j) (s j)
A*1 1 1 =
j
15
A*1 1 2 =
j
2/5
12
j
15
In a similar fashion, the rest of the residues for the remaining transfer functions can also be
determined.
(4-15)
+UC-SDRL-RJA
CN-20-263-662
The system transfer function matrix [ H(s) ] can now be expressed in terms of partial fractions.
j
2/5
j
2/5
12
12
j
2/5
j
2/5
12
12
[ H (s) ] =
2/5 )
(s j
j
j
15
30
j
j
30
60
+
(s j)
j
2/5
j
2/5
12
12
j
2/5
j
2/5
12
12
2/5 )
( s + j
j
j
15
30
j
j
60
30
(s+ j)
Recall that the modal vector associated with the pole frequency 1 is proportional to the residue
matrix for pole 1 . Notice that the first two residue matrices are just the complex conjugate of
each other. Therefore, the modal vector for pole *1 is just the complex conjugate of the modal
vector for pole 1 . The same is true for the last two residue matrices. This will always be the
case for conjugate pairs of poles.
2/5 can be extracted from the first residue matrix.
The modal vector for the pole 1 =
Using Equations 4.29 and 4.30, the modal vectors can be related to the residue matrix for the first
pole.
1 1
Q1 { }1 { }T1 = Q1
2 1
j
2/5
12
Q1 { }1 { }T1 =
j
2/5

12
The constant Q1 =
1 2
2 2
1
j
2/5
12
j
2/5
12
j
2/5
can now be factored out of the residue matrix. The choice of this
12
(4-16)
+UC-SDRL-RJA
CN-20-263-662
constant at this point is purely arbitrary by virtue of the fact that the absolute amplitudes of the
modal vectors are purely arbitrary.
1 1 1 2 1
=
2 1 2 2 1 1
1
1
Hence, the modal vector for the first mode is:

1 1
=
2
1 1 1
Similarly, the modal vector for the second mode:

1 1
Q2 { }2 { }T2 = Q2
2 1
The constant Q2 =
j
15
= j
2
30
1 2
2 2
j
30
j
60
j
can now be factored from this residue matrix in the same arbitrary
60
manner as before.
1 1
2 1
1 2
2 2
4
=
2 2
1
2
Equating elements of the two matrices yields:

1 1 = 12 = 4
2 1 = 2
1 = 2
2 = 1
Thus:
1 2
=
2 2 1 2
(4-17)
+UC-SDRL-RJA
CN-20-263-662
Both of these modal vectors could have been obtained directly by just using either a row or a
column of the residue matrix as the modal vector directly.
4.5 Analytical Model - General Partial Fraction (Residue)

Recalling Equation 4.21 for a two degree of freedom system:
H 1 1 (s) =
r=1
A1 1 r
A*1 1 r
+
( s r ) ( s *r )
Equation 4.21 can now be generalized for an N -degree of freedom system as the following:
H 1 1 (s) =
r=1
A1 1 r
A*1 1 r
+
( s r ) ( s *r )
A
A*
N
pqr + pqr
H (s) =
( s r ) ( s * )
pq
r=1
r
(4.31)
(4.32)
Furthermore, the entire system transfer function matrix [ H(s) ] can be generalized as:
[ H (s) ] =
r=1
[ A ] r [ A* ] r
+
s r
s *r
(4.33)
In terms of the modal vectors of the system directly:

[ H(s) ] =
r=1
Q r { }r { }Tr
Q* { }*r { }*T
r
+ r
*
( s r )
( s r )
(4.34)
Equation 4.34 is the general form of the system transfer function matrix. As will be shown later,
this form does not change when a system with damping is considered. Remember, though, that
the complete transfer function can not be measured. Not once again that the frequency response
function measurement, which is just Equation 4.34 evaluated at s = j , is actually what is
measured.
(4-18)
+UC-SDRL-RJA
CN-20-263-662
4.6 Residue Relationship to Modal Vectors

The relationship established in Section 4.3 between the residue matrix and the modal vector can
be developed in a more formal or rigorous manner. It is important to understand that the residue
defined in the previous section is the key to the relationship between modal vectors and modal
scaling (modal mass) for the experimental case. At the present time, the following discussion is
limited to the undamped and/or proportionally damped cases. The relationship for the case of
general damping is discussed in a later section.
The development of the relationship between the residue matrix and the modal vectors proceeds
along a similar path as the development of the relationship between the residue matrix and the
adjoint of the system matrix (Section 3.8). Beginning with the definition of the impedance
matrix:
[B(s)] = [ M ]s2 + [ C ]s + [ K ]
(4.35)
where:
[ B(s) ] = System Impedance Matrix
From matrix algebra:

[B(s)] [B(s)]1 = [I ]
(4.36)
Noting that the inverse of the impedance matrix is the transfer function matrix gives:
[B(s)] [H(s)] = [I ]
(4.37)
Replacing the transfer function matrix with the equivalent partial fraction representation from
Equation 4.33 yields:
[I ] =
r=1
[ B(s) ] [ A ] r [B(s)] [ A* ] r
+
s r
s *r
(4.38)
Premultiplying each term of the above equation by (s r ) and evaluating the equation at s = r
(4-19)
+UC-SDRL-RJA
CN-20-263-662
for any specific mode r (all of the terms drop out except the term associated with r ):
[ 0 ] = B( r ) [ A ] r
(4.39)
Note that the above equation proves that each column of the residue matrix [ A]r must be
proportional to the modal vector associated with r just as it did in the adjoint matrix case
(Section 3.8). Likewise the structure of the residue matrix must be the same as the structure of
the adjoint matrix.
[ A]r = Q r { }r { }Tr
1 1 1 2

2 2
2 1
[ A]r = Q r
.
.
.
.
N 1 N 2
.
.
.
.
.
.
.
.
.
.
(4.40)
.
.
.
.
.
.
.
.
.
.
1 N
.
.
.
N N
where:
Q r = constant associated with the scaling of { }r relative to the absolute scaling (units) of
the residue matrix.
Q r = is proportional, but not generally equal to, the proportionality constant r defined in
Section 3.8.
4.7 Residue Relationship to Modal Mass

For the proportionally damped case, which includes the undamped case as a trivial form, the
relationship between the residue and the modal mass can also be established consistent with the
modal mass found analytically from the mass matrix. Starting with Equation 4.38 and 4.40:
[I ] =
r=1
[ B(s) ] Q r { }r { }Tr
+
s r
(4-20)
[B(s)]
*

r r
Q*r *
s *r
(4.41)
+UC-SDRL-RJA
CN-20-263-662
Note that, for proportionally damped systems, the modal vectors are always real (normal) modes.

Therefore, the conjugate of a modal vector is the same as the modal vector ({ }r = * ).
r
T
Making this substitution and premultiplying both sides of Equation 4.41 by { }t :
{ }Tt =
r=1
{ }Tt [ B(s) ] Q r { }r { }Tr

{ }Tt [B(s)] Q*r { }r { }Tr
+
s r
s *r
(4.42)
Substitute Equation 4.35 into Equation 4.42:
{ }Tt =
r=1
Q r { }Tt [ M ] s2 + [ C ] s + [ K ] { }r { }Tr
+
s r
Q*r { }Tt [ M ] s2 + [ C ] s + [ K ] { }r { }Tr
*
s r
(4.43)
Applying the orthogonality relationships between the modal vectors and the mass, damping and
stiffness matrices eliminates all terms except for those associated with mode t:
{ }Tt =
Q t ( M t s2 + C t s + K t ){ }Tt
Q* ( M t s2 + C t s + K t ){ }Tt
+ t
s t
s *t
(4.44)
Eliminating { }Tt from each term of Equation 4.44 leaves the following scalar equation:
1=
Q t ( M t s2 + C t s + K t ) Q*t ( M t s2 + C t s + K t )
+
s t
s *t
(4.45)
Clearing the fractions from the previous equation:

(s t ) (s *t ) = Q t ( M t s2 + C t s + K t ) (s *t ) + Q*t ( M t s2 + C t s + K t )(s t )
.
(4-21)
(4.46)
+UC-SDRL-RJA
CN-20-263-662
Note that, by definition:

( M t s2 + C t s + K t ) = M t (s t ) (s *t )
(4.47)
Substituting Equation 4.47 into Equation 4.46 and eliminating the common terms on both sides
of the equation (s t ) (s *t ):
1 = Q t M t (s *t ) + Q*t M t (s t )
(4.48)
Evaluating Equation 4.49 at s = t gives the final relationship:

1 = Q t M t ( t *t )
(4.49)
1 = Q t M t (2 j t )
(4.50)
Mt =
1
2 j t Qt
(4.51)
Equation 4.51 represents the relationship between modal mass and the scaling involved between
the residues and the modal vectors (Recall Equation 4.40). Therefore, once the residue
information is found for mode t and some convenient form of modal vector scaling is chosen for
mode t, the scaling constant Q t can be determined. Equation 4.51 can then be used to determine
modal mass for mode t consistent with the modal vector scaling. This means that as long as the
modal vector is chosen consistently, modal mass can be compared between different solution
approaches (analytical versus experimental, for example).
(4-22)
5. GENERAL DAMPED SYSTEMS

5.1 Non-Proportionally Damped Systems
In reality, physical structures or systems are generally comprised of many substructures tied
together in various fashions. These substructures can be made-up of a variety of materials, i.e.
metals, plastics, and wood. Furthermore, these substructures may be connected to one another
by rivets, bolts, screws, dampers, springs, weldments, friction, etc. Also, the spatial geometry of
the structure may be very complicated, such as an exhaust system of an automobile. All of these
factors influence the inherent dynamical properties of the structure. For these structures, mass,
damping, and stiffness distribution (matrices) of the system are rather complicated. In general,
for real life structures, the damping matrix for such a system will not always be proportional to
the mass and/or stiffness matrix. Therefore, the damping of this system can be classified as nonproportional.
In an analytical sense, modal analysis of this general type of damped system cannot be described
using the formulation of the eigenvalue problem as discussed previously for an undamped
system. Remember, that in an analytical sense, the purpose of modal analysis is to find a
coordinate transformation that uncoupled the original equations of motion. This coordinate
transformation turned out to be a matrix comprised of the modal vectors of the system. These
modal vectors were determined from the solution of the eigenvalue problem for that system. The
coordinate transformation diagonalized the system mass, damping, and stiffness matrices, for an
undamped or proportionally damped system. When a system contains non-proportional
damping, the previously used formulation of the eigenvalue problem will not yield modal vectors
(eigenvectors) that uncouple the equations of motion of the system. A technique used to
circumvent this problem was first documented by Duncan and Collar and involves the
reformulation of the original equations of motion, for an N -degree of freedom system, into an
equivalent set of 2 N first order differential equations known as Hamiltons Canonical
Equations. The solution of these equations can be carried out in a similar manner that has been
discussed previously.
For nonproportional damping, the coordinate transformation discussed previously, that
diagonalizes the system mass and stiffness matrices, will not diagonalize the system damping
matrix. Therefore, when a system with nonproportional damping exists, the equations of motion
are coupled when formulated in N dimension physical space. Fortunately, the equations of
motion can be uncoupled when formulated in 2N dimension state space. This is accomplished
(5-1)
+UC-SDRL-RJA
CN-20-263-662
Revision: March 2, 2000 +
by augmenting the original N dimension physical space equation by a N dimension identity as

follows.
Assume a viscous, nonproportionally damped system can be represented by Equation 5.1.
[ M ] { x } + [ C ] { x } + [ K ] { x } = { f }
(5.1)
This system of equations can be augmented by the identity shown in Equation 5.2:
[ M ] { x } [ M ] { x } = { 0 }
(5.2)
Equations 5.1 and 5.2 can be combined as follows to yield a new system of 2N equations. Note
that all the matrices in Equation 5.3 are symmetric and Equation 5.3 is now in a classical
eigenvalue solution form. The notation used in Equation 5.3 is consistent with the notation used
in many mathmatics and/or controls textbooks.
[ A ] { y } + [ B ] { y } = f
(5.3)
where:
[0]
[A ] =
[ M ]
[M]
[C ]
{ x}

{ y } =

{ x}
[ M ]
[B ] =
[0]
{ x}

{ y } =
{x}
[0]
[K]

{0}
f =
{f}
Forming the homogeneous equation from Equation 5.3 yields:

[ A ] { y } + [ B ] { y } = { 0 }
(5.4)
The solution of Equation 5.4 yields the complex-valued natural frequencies (eigenvalues) and
complex-valued modal vectors (eigenvectors) for the augmented 2N equation system. Note that
in this mathematical form, the eigenvalues will be found directly (not the square of the
(5-2)
+UC-SDRL-RJA
CN-20-263-662
eigenvalue) and the 2N eigenvectors will be 2N in length. The exact form of the eigenvectors
can be seen from the eigenvector matrix (state space modal matrix) for the 2N equation system.
Note that the notation { } is used for an eigenvector in the 2N equation system and that the
notation { } is used for an eigenvector of the original N equation system.
The state space modal matrix [ ] for this nonproportionally damped system can now be
assembled.
= { }
1
{ }2
{ }3
{. . . }
{ }r
{. . . }
{ }2N
(5.5a)
Based upon the change of coordinate applied in Equation 5.3, each column of the eigenvector
matrix (each eigenvector) is made up of the derivative of the desired modal vector above the
desired modal vector. This structure is shown in Equation 5.5b.
= 1 { }1

{ }1
2 { }2
3 { }3
{ }2
{ }3
...
...
r { }r
{ }r
...
...
2N { }2N
{ }2N
(5.5b)
5.1.1 Weighted Orthogonality of the Eigenvectors

Similar to the case for undamped systems, a set of weighted orthogonality relationships are valid
for the system matrices [ A] and [B].
{ }r T [ A ] { }s = 0
(5.6)
{ }r T [B ] { }s = 0
(5.7)
The terms modal A and modal B can now be defined as follows. Note that these quantities have
the same properties as modal mass and modal stiffness for the undamped and proportionally
damped cases.
{ }r T [ A ] { }r = M Ar
(5.8)
{ }r T [B ] { }r = M Br
(5.9)
(5-3)
+UC-SDRL-RJA
CN-20-263-662
The terms modal A and modal B are modal scaling factors for the nonproportional case just as
modal mass and modal stiffness can be used for the undamped and proportionally damped cases.
Whenever complex modal vectors are present, modal A and modal B should be used to provide
the modal scaling. Note that modal A and modal B could be used to provide modal scaling even
for the undamped and proportionally damped cases.
Note that the eigenvector matrix (state space modal matrix) provides a coordinate transformation
from the physical state space coordinate system to the uncoupled principl state space coordinate
system.
[ A ] { y } + [ B ] { y } = f
(5.10)
[ A ] { q } + [ B ] { q } = f
(5.11)
T
T
T
[ A ] { q } + [ B ] { q } = f
(5.12)
For the r th eigenvalue/eigenvector:

M Ar q r + M Br = f r
(5.13)
This uncoupled equation has a characteristic equation of the form:

M Ar s + M B r = 0
(5.14)
This means that Modal A and Modal B for each mode are related.
M Ar r + M B r = 0
(5.15)
M Ar r = M B r
(5.16)
This concept will not be pursued further here. The interested reader is referred to Chapter 6 of
Mechanical Vibrations by Tse, Morse and Hinkle or Chapter 9 of Analytical Methods in
(5-4)
+UC-SDRL-RJA
CN-20-263-662
Vibrations by Leonard Meirovitch.

In an experimental sense, the approach to the problem is the same for non-proportionally damped
system as for an undamped or a proportionally damped system. No matter what type of damping
a structure has, proportional or non-proportional, the frequency response functions of the system
can be measured. For this reason, the modal vectors of the system can be found by using the
residues determined from the frequency response function measurements. While the approach is
the same as before, the results in terms of modal vectors will be somewhat more complicated for
a system with non-proportional damping. As an example of the differences that result for the
non-proportional case, the same two degree of freedom system, used in previous examples, will
be used again except that the damping matrix will be made non-proportional to the mass and/or
stiffness matrices.
5.2 Proportionally Damped Systems

For the class of physical damping mechanisms that can be mathematically represented by the
proportional damping concept, the coordinate transformation discussed previously for the
undamped case, that diagonalizes the system mass and stiffness matrices, will also diagonalize
the system damping matrix. Therefore, when a system with proportional damping exists, that
system of coupled equations of motion can be transformed as before to a system of equations that
represent an uncoupled system of single degree of freedom systems that are easily solved. The
procedure to accomplish this follows.
Assume a viscously damped system can be represented by Equation 5.17.
[ M ] { x } + [ C ] { x } + [ K ] { x } = { f }
(5.17)
The eigenvalue problem associated with the undamped system can be solved as a first step to
understanding the problem.
[ M ] { x } + [ K ] { x } = { 0 }
This yields the systems natural frequencies (eigenvalues) and modal vectors (eigenvectors).
(5-5)
+UC-SDRL-RJA
CN-20-263-662
The modal matrix [ ] for this undamped system can now be assembled.
= { } { } ... { }
1
2
N
The coordinate transformation can now be applied to Equation 5.10:

[ x ] = { q }
[ M ] { q } + [ C ] { q } + [ K ] { q } = { f }
Now pre-multiply Equation 5.18 by [ ]T .

T
[ M ] { q } + [ C ] { q }
+ [ K ] { q } = { f }
(5.19)
Due to the orthogonality properties of the modal vectors :

T
[ M ] = M
[ K ] = K

Since the assumed form of the damping matrix is proportional to the mass and/or stiffness
matrix, the damping matrix will also be diagonalized.
[ C ] = [ M ] + [ K ]
The application of the orthogonality condition yields:
T
[ C ] = { [ M ] + [ K ] }
(5-6)
+UC-SDRL-RJA
CN-20-263-662
[ C ] = [ M ] + [ K ]
[ C ] = M + K

Therefore:
T
[ C ] = C

where:
C is a diagonal matrix.
Therefore, Equation 5.19 becomes:

T
M { q } + C { q } + K { q } = { f (t) }
(5.20)
Equation 5.20 represents an uncoupled set of damped single degree of freedom systems. The rth equation of Equation 5.20 is:
M r q r + C r q r + K r q r = f r (t)
Equation 5.21 is the equation of motion for a system represented below.
(5-7)
(5.21)
+UC-SDRL-RJA
CN-20-263-662
q r (t)
f r (t)
Kr
Mr
Cr
Figure 5-1. Proportional Damped SDOF Equivalent Model
The solution of this damped single degree of freedom system has been discussed previously.
5.3 Example with Proportional Damping

In order to understand the concept of proportional damping, the same two degree of freedom
example, worked previously for the undamped case, can be reworked for the proportionally
damped case.
(5-8)
+UC-SDRL-RJA
CN-20-263-662
x 1 (t)
x 2 (t)
f 1 (t)
f 2 (t)
2
5
4
10
Figure 5-2. Two Degree of Freedom Model with Proportional Damping

5
2
0 x 1
+
10 x 2
1
4
1 x 1
+
3 x 2
2
f1
2 x 1
6 x2
f2
Note that the damping matrix [ C ] is proportional to the stiffness matrix [ K ]:

[ C ] =
1
[ K ]
2
While this form of the damping matrix is quite simple, the solution that will result will yield a
general characteristic that is common to all problems that can be described by the concept of
proportional damping.
Using the previously calculated modal vectors and natural frequencies for the undamped system,
a coordinate transformation can be performed:
(5-9)
+UC-SDRL-RJA
CN-20-263-662
{ x } = { q }
Noting the results of the previous example in Section 3.6, the uncoupling of the equations of
motion for the mass and stiffness matrices has already been shown. Therefore, for the damping
matrix:
T
1 / 15
[ C ] =
2 / 15
1 / 15
2
1
2 / 15 1

1 / 15
2 / 15
1
1
3
1 / 15
2 / 15
T
0
[ C ] = 1/5
1/2
0
The transformed system in terms of modal coordinates would be:

1
1/5
0 q 1
+
1 q 2
0
2/5
0 q 1
+
1 / 2 q 2
0
or pictorially:
(5-10)
f 1 (t)
0 q1
1 q2
f 2 (t)
+UC-SDRL-RJA
CN-20-263-662
q 1 (t)
q 2 (t)
f 1 (t)
f 2 (t)
2/5
1
1
1/5
1/2
Figure 5-3. Proportionally Damped MDOF Equivalent Model
5.3.1 Frequency Response Function Implications

This same problem can be evaluated from a transfer function point of view by determining the
modal frequencies and modal vectors. The transfer function matrix of the system is:
(M 22 s2 + C 22 s + K 22 ) (M 12 s2 + C 12 s + K 12 )
(M 21 s2 + C 21 s + K 21 ) (M 11 s2 + C 11 s + K 11 )
[ H(s) ] =
| B (s) |
| B (s) | = M 11 s2 + C 11 s + K 11 M 22 s2 + C 22 s + K 22
M 21 s2 + C 21 s + K 21 M 12 s2 + C 12 s + K 12
(5-11)
+UC-SDRL-RJA
CN-20-263-662
Upon substituting the values for the mass, damping, and stiffness matrices from the example
problem:
10 s2 + 3 s + 6
(s + 2)

(s + 2)
5s + 2s+ 4
[ H(s) ] =
4
3
2
50 s + 35 s + 75 s + 20 s + 20
The roots of the characteristic equation are:

1 =
1
39
+ j
( rad/sec )
10
10
*1 =
1
39
j
( rad/sec )
10
10
1
15
+ j
( rad/sec )
4
4
*2 =
1
15
j
( rad/sec )
4
4
2 =
Note that the poles of a damped system now contain a real part. The pole of a transfer function
has been previously defined (Chapter 2) as:
r = r + j r
where:
r
= damping factor
= damped natural frequency.
Remember that in the previous undamped case, the poles could be written in the following form:
r = j r
where:
r
= damped natural frequency = undamped natural frequency
The system modal vectors can be determined by just using the minimum transfer function data,
assuming that all elements of the row or column are not perfectly zero. If this is true, then the
(5-12)
+UC-SDRL-RJA
CN-20-263-662
modal vectors can be found by using one row or column out of the system transfer function
matrix. For example, choose H 11 (s) and H 21 (s), which amounts to the first column.
First, H 11 (s) and H 21 (s) can be expanded in terms of partial fractions.
H 11 (s) =
10 s2 + 3 s + 6
50 ( s 1 ) ( s *1 ) ( s 2 ) ( s *2 )
After some work:

4
4
15
15
39
j
j
225 +
225
117
(s) =
+
+
( s 1 )
( s *1 )
( s 2 )
( s *2 )
j
H 11
H 21 (s) =
s + 2
50 ( s 1 ) ( s *1 ) ( s 2 ) ( s *2 )
2
2
15
15
39
j
j
225
225
117
(s) =
+
+
+
( s 2 )
( s *2 )
( s 1 )
( s *1 )
j
H 21
39
117
39
117
Recall that the residues have been shown to be proportional to the modal vectors. Since only a
column of the transfer function has been used from the transfer function matrix, only a column of
the residue matrix for each modal frequency has been calculated.
Thus, by looking at the residues for the first pole 1 , the modal vector that results is:
2 1
39
j
117
1
j
39
1 1
117
1
(5.23)
For the modal frequency 2 , the modal vector that results is:
2 2
15
j 4
225

2 j
15
225
(5-13)
=
1/2 2
(5.24)
+UC-SDRL-RJA
CN-20-263-662
Thus, the modal vectors calculated from the transfer function matrix of this proportionally
damped system are the same as for the undamped system. This will always be the case as long
as the system under consideration exhibits proportional type of damping.
Notice that for the undamped and proportionally damped systems studied, not only were their
modal vectors the same, but their modal vectors, when normalized, are real valued. This may not
be the case for real systems. Since the modal vectors are real valued, they are typically referred
to as real modes or normal modes.
5.3.2 Impulse Response Function Considerations

The impulse response function of a proportionally damped multiple degree of freedom system
can now be discussed. Remember that the impulse response function is the time domain
equivalent to the transfer function. Once the impulse response function for a multiple degree of
freedom system has been formulated, it can be compared to the previous single degree of
freedom system. Recall Equation 4.24.
H pq (s) =
r=1
A*pqr
A pqr
+
( s r )
( s *r )
(5.25)
Recall the definition of H pq (s):

H pq (s) =
X p (s)
F q (s)
Therefore:
X p (s) = F q (s)
r=1
A*pqr
A pqr
+
( s r )
( s *r )
(5.26)
If the force used to excite the DOF (q) is an impulse, then the Laplace transform will be unity.
While this concept is used to define the impulse response function, once the transfer function is
formulated and the system is considered to be linear, this type of assumption can be made with
complete generality.
(5-14)
+UC-SDRL-RJA
CN-20-263-662
F q (s) = 1
Thus the system impulse response at DOF ( p) is the inverse Laplace transform of Equation 5.19.
1
h pq (t) = L
1
X p (s) = L
A*pqr
A pqr
+
*
( s r )
r = 1 ( s r )
N
If p = q, this would be the driving point impulse response function. For simplicity, Equation
5.27 can be expanded for the two degree of freedom case (N = 2).
A*pp1
A*pp2
App1
App2
X p (s) =
+
+
+
( s 1 )
( s *1 )
( s 2 )
( s *2 )
(5.28)
*
*
h pp (t) = L1 X p (s) = App1 e 1 t + A*pp1 e 1 t + App2 e 2 t + App2 e 2 t
(5.29)
Recall that:
r = r + j r
*r = r j r
Also, since the system is proportionally damped, the residues are all purely imaginary.
Therefore, the following definition can be made arbitrarily. This definition will make certain
trigonometric identities obvious in a later equation.
App1 =
Rpp1
2 j
App2 =
Rpp2
2 j
(5-15)
+UC-SDRL-RJA
CN-20-263-662
Using Eulers formula, Equation 5.22 then becomes:

h pp (t) =
Rpp1 ( 1
e
2 j
+
h pp (t) = Rpp1 e 1 t
( e j 1 t e
2 j
+ j 1 ) t
Rpp2 ( 2
e
2 j
j 1 t
e( 1
j 1 ) t
+ j 2 ) t
+ Rpp2 e 2 t
e( 2
j 2 ) t
( e j 2 t e
2 j
j 2 t
h pp (t) = Rpp1 e 1 t sin ( 1 t ) + Rpp2 e 2 t sin ( 2 t )

Finally:
h pp (t) = Rpp1 e 1 t sin( 1 t ) + Rpp2 e 2 t sin( 2 t )
(5.30)
Comparing this to the single degree of freedom case done previously, note that the impulse
response function is nothing more than the summation of two single degree of freedom
responses. Note also that the amplitude of the impulse response function is directly related to the
residues for the two modal vectors. This means that the impulse response function amplitude is
directly related to the modal vectors of the two modes, since the residues have been shown to be
directly proportional to the modal vectors.
In summary, a transfer function, and therefore a frequency response function, can be expressed as
a sum of single degree of freedom systems (Equation 4.24). A typical frequency response
function is illustrated in Figure (5-4), in terms of its real and imaginary parts. This frequency
response function represents the response of the system at degree of freedom 1 due to a force
applied to the system at degree of freedom 2 H 1 2 ( ). Figure (5-5) is the equivalent impulse
response function of Figure (5-4). Likewise, this impulse response function represents the
response of the system at degree of freedom 1 due to a unit impulse applied to the system at
degree of freedom 2 h1 2 (t). Notice that the impulse response function starts out at zero for t = 0
as it should for a system with real modes of vibration. Since the transfer function is the sum of
single degree of freedom systems, the frequency response functions for each of the single degree
of freedom systems can be plotted independently as is illustrated in Figures (5-6) and (5-8).
(5-16)
+UC-SDRL-RJA
CN-20-263-662
Note that by adding together Figures (5-6) and (5-8), the same plot as in Figure (5-4) would
result. In Figures (5-6) and (5-8) the imaginary parts peak where the real parts cross zero, where
as, in Figure (5-4) this is not exactly true. Similarly, the impulse response functions for each
single degree of freedom system can be plotted separately as in Figures (5-7) and (5-9). The sum
of these two figures would be the same as Figure (5-5). Therefore, the impulse response of a two
degree of freedom system is just the sum of two damped sinusoids.
(5-17)
+UC-SDRL-RJA
CN-20-263-662
0.6
0.4
Real
0.2
-0.2
-0.4
-0.6
0.5
1.5
2.5
2.5
Frequency (Rad/Sec)
0.6
0.4
Imaginary
0.2
-0.2
-0.4
-0.6
0.5
1.5
Frequency (Rad/Sec)
Figure 5-4. FRF, 2 DOF System, Proportionally Damped
(5-18)
+UC-SDRL-RJA
CN-20-263-662
0.1
0.08
0.06
Amplitude
0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
0
10
15
20
25
Time (Sec.)
Figure 5-5. IRF, 2 DOF System, Proportionally Damped
(5-19)
30
+UC-SDRL-RJA
CN-20-263-662
0.6
0.4
Real
0.2
-0.2
-0.4
-0.6
0.5
1.5
2.5
2.5
Frequency (Rad/Sec)
0.6
0.4
Imaginary
0.2
-0.2
-0.4
-0.6
0.5
1.5
Frequency (Rad/Sec)
Figure 5-6. FRF, 2 DOF System, Proportionally Damped, First Mode
(5-20)
+UC-SDRL-RJA
CN-20-263-662
0.1
0.08
0.06
Amplitude
0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
0
10
15
20
25
Time (Sec.)
Figure 5-7. IRF, 2 DOF System, Proportionally Damped, First Mode
(5-21)
30
+UC-SDRL-RJA
CN-20-263-662
0.2
0.15
0.1
Real
0.05
0
-0.05
-0.1
-0.15
-0.2
0
0.5
1.5
2.5
2.5
Frequency (Rad/Sec)
0.2
0.15
0.1
Imaginary
0.05
0
-0.05
-0.1
-0.15
-0.2
0
0.5
1.5
Frequency (Rad/Sec)
Figure 5-8. FRF, 2 DOF System, Proportionally Damped, Second Mode
(5-22)
+UC-SDRL-RJA
CN-20-263-662
0.1
0.08
0.06
Amplitude
0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
0
10
15
20
25
30
Time (Sec.)
Figure 5-9. IRF, 2 DOF System, Proportionally Damped, Second Mode
5.4 Non-Proportional Damping Example

The same two degree of freedom system, in terms of mass and stiffness, as before will now be
used to explain the effect of non-proportional damping.
(5-23)
+UC-SDRL-RJA
CN-20-263-662
x 1 (t)
x 2 (t)
f 1 (t)
f 2 (t)
2
5
2
4
10
Figure 5-10. Two Degree of Freedom System
The damping coefficients have been chosen such that they are not proportional to the mass and/or
stiffness. Using Equation 5.22, the transfer function matrix of the resulting system is:
10s2 + 5s + 6
4s + 2
2
4s + 2
5s + 6s + 4
[ H(s) ] =
50[s4 + (17/10) s3 + (42/25) s2 + (4/5) s + 2/5]
(5.31)
The poles of the transfer function are the roots of the characteristic equation:
s4 +
17 3 42 2 4
2
s +
s + s+ =0
10
25
5
5
(5-24)
(5.32)
+UC-SDRL-RJA
CN-20-263-662
The roots are:

1 = 0. 095363 + j 0. 629494 ( rad/sec )
*1 = 0. 095363 j 0. 629494 ( rad/sec )
2 = 0. 754635 + j 0. 645996 ( rad/sec )
*2 = 0. 754635 j 0. 645996 ( rad/sec )
These poles are the damping and frequency parameters for the systems two modes of vibration.
Recall: r = r + j r .
As before, the modal vectors of the system can be determined by evaluating one row or column
of the transfer function matrix in terms of partial fractions. In a previous example, the first
column of the transfer function matrix was utilized in order to determine the modal vectors. This
time, to illustrate the pervasiveness of the modal vectors with respect to all elements of [ H (s) ],
the last row of the transfer function matrix will be utilized. That is:
H 21 (s) and H 22 (s)
H 21 (s) can be measured by exciting the system at mass 1 and measuring the response at mass 2.
Similarly, H 22 (s) can be measured by exciting the system at mass 2 and measuring the response
at mass 2. H 22 (s) and H 21 (s) can now be expanded in terms of partial fractions.
H 22 (s) =
5s2 + 6s + 4
50 (s 1 ) (s *1 ) (s 2 ) (s *2 )
After some work:

H 22 (s) =
0. 003707 j 0. 058760 0. 003707 + j 0. 058760

+
(s 1 )
(s *1 )
0. 003707 j 0. 016358 0. 003702 + j 0. 016358

+
(s *2 )
(s 2 )
Also:
(5-25)
+UC-SDRL-RJA
CN-20-263-662
H 21
4s2 + 2
(s) =
50 (s 1 ) (s *1 ) (s 2 ) (s *2 )
After some more work:

H 21 (s) =
0. 003473 j 0. 050100 0. 003473 + j 0. 050100

+
(s *1 )
(s 1 )
0. 003473 + j 0. 045270 0. 003473 j 0. 045280

+
(s *2 )
(s 2 )
Using the residues as the modal vectors directly (without any normalization or scaling), the
following vectors result:
Mode 1:
1 = - 0.095363 + j 0.629494
1 0. 003473 j 0. 050100
=
2 1 + 0. 003707 j 0. 058760
Mode 2:
2 = - 0.754635 + j 0.645996
1 + 0. 003473 + j 0. 045270
=
2 2 0. 003707 j 0. 016358
The modal vectors are obviously no longer purely real. In both modal vectors, regardless of how
the modal vectors are normalized, the description of the modal vector will now require a vector
of complex numbers. This will in general be true for any system with non-proportional damping.
The above modal vectors can be converted to amplitude and phase.
Thus, for modal vector one:
1 0. 050220,
=
2
1 0. 058877,
(5-26)
93. 96
86. 39
1
+UC-SDRL-RJA
CN-20-263-662
For modal vector two:

1 0. 045403,
=
2
2 0. 016765,
+ 85. 61
102. 77
2
For mode one, as in the previous examples, the two masses are moving approximately in phase.
Likewise, for mode two, as in the previous examples, the two masses are moving approximately
out of phase. Thus, the non-proportionally damped system, which resulted in complex mode
shapes, has phase terms which are not generally 0 or 180.
Another way of comparing the differences between real modal vectors and complex modal
vectors is as follows: A real modal vector represents a mode of vibration where all the points on
the structure pass through their equilibrium positions at the same time. A complex modal vector
represents a mode of vibration where the points tested do not pass through their equilibrium
positions at the same time. In other words, a real mode appears as a standing wave and a
complex mode appears as a traveling wave with respect to the structure. The nodal point or line
for real modes will be stationary with respect to the structure. Conversely, the nodal point or line
for a complex mode will be non-stationary with respect to the structure.
Many structures tested will have nearly real mode shapes, that is, the phase term will be very
close to being 0 or 180. Structures in general can have both real and complex mode shapes. In
the previous example, the first modal vector is within 4 of being totally real, whereas the
second modal vector is within 13 of being real. Thus, the second modal vector is more
complex than the first modal vector.
5.4.1 Impulse Response Function Example

The impulse response function is computed in a similar fashion as for the proportionally damped
system. From Equation 5.27, the basic definition of the impulse response function is:
1
H pq (t) = L
r=1
A*pqr
A pqr
+
(s r ) (s *r )
(5-27)
(5.33)
+UC-SDRL-RJA
CN-20-263-662
The only difference now is that the residues A pqr are no longer purely imaginary but are, in
general, complex. For a two-degree of freedom system, the impulse response function becomes:
h pq (t) = A pq1 e
1 t
+ A*pq1 e 1 t + A pq2 e 2 t + A*pq2 e 2 t

*
(5.34)
The only difference between the impulse response function of a system with complex modal
vectors (Equation 5.34) and a system with real modal vectors (Equation 5.29 and 5.30) is that the
individual contributions from each mode of vibration to the total impulse response function do
not begin with a value of zero response at time zero. Note that the sum of these individual
contributions must total zero at time zero for the system to be causal. Again note that the
residues, which are proportional to the product of the modal coefficients of the input and
response degrees of freedom, appear as the amplitude of the damped sinusoids.
A typical frequency response function is once again illustrated in Figure (5-11), in terms of its
real and imaginary parts. This frequency response function represents the response of the system
at degree of freedom 1 due to a force applied to the system at degree of freedom 2 H 1 2 ( ).
Figure (5-12) is the equivalent impulse response function of Figure (5-11). Likewise, this
impulse response function represents the response of the system at degree of freedom 1 due to a
unit impulse applied to the system at degree of freedom 2 h1 2 (t). Figure (5-11) depicts a
frequency response function of a two degree of freedom system that has complex modal vectors.
The only difference between this plot and that of Figure (5-4) is that the residues in this case
have a phase angle of other than 0 or 180. The impulse response function of this system is
shown in Figure (5-11). Notice that the impulse response function does start at zero for t=0. This
is always the case for any causal system with or without complex modal vectors. Notice, though,
that the impulse response function contribution from each individual mode does not start at zero
for this case of complex modal vectors. Nevertheless, for a causal system, the sum of the
residues, and therefore the combination of magnitudes and phase angles in Equation 5.27, will
always yield an impulse response function that is zero at time zero. (The sum of the residues
will be equal to zero for any system where the order of the numerator polynomial is two or more
less than the order of the denominator polynomial.) As before, the frequency response function
can be represented as (Figure (5-11)) is the sum of two single degree of freedom systems
(Figures (5-13) and (5-15)). Another indication of a complex modal vector is the asymmetric
characteristic of the imaginary part of the frequency response function for such a simple system.
In Figure (5-6) and (5-8), note that the imaginary parts are symmetrical, whereas in Figures
(5-28)
+UC-SDRL-RJA
CN-20-263-662
(5-13) and (5-15) the imaginary parts are not symmetrical. The degree to which they are not
symmetrical is an indication of how complex that particular component of the modal vector is.
Figures (5-14) and (5-16) are the impulse response functions of Figures (5-13) and (5-15).
The modal vectors and modal frequencies of the undamped, proportionally damped, and nonproportionally damped system can now be compared for the simple two degree of freedom
system used in all the examples. Remember that the stiffnesses and masses are the same
throughout the examples. Therefore, the modal vectors and modal frequencies can only be
different as a result of the damping elements. For comparison purposes, the unscaled residues
from the standard form of the transfer function will be used as the modal vectors for each case.
(5-29)
+UC-SDRL-RJA
CN-20-263-662
0.6
0.4
Real
0.2
-0.2
-0.4
-0.6
0.5
1.5
2.5
2.5
Frequency (Rad/Sec)
0.6
0.4
Imaginary
0.2
-0.2
-0.4
-0.6
0.5
1.5
Frequency (Rad/Sec)
Figure 5-11. FRF, 2 DOF System, Non-Proportionally Damped
(5-30)
+UC-SDRL-RJA
CN-20-263-662
0.1
0.08
0.06
Amplitude
0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
0
10
15
20
25
Time (Sec.)
Figure 5-12. IRF, 2 DOF System, Non-Proportionally Damped
(5-31)
30
+UC-SDRL-RJA
CN-20-263-662
0.6
0.4
Real
0.2
-0.2
-0.4
-0.6
0.5
1.5
2.5
2.5
Frequency (Rad/Sec)
0.6
0.4
Imaginary
0.2
-0.2
-0.4
-0.6
0.5
1.5
Frequency (Rad/Sec)
Figure 5-13. FRF, 2 DOF System, Non-Proportionally Damped, First Mode
(5-32)
+UC-SDRL-RJA
CN-20-263-662
0.1
0.08
0.06
Amplitude
0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
0
10
15
20
25
Time (Sec.)
Figure 5-14. IRF, 2 DOF System, Non-Proportionally Damped, First Mode
(5-33)
30
+UC-SDRL-RJA
CN-20-263-662
0.6
0.4
Real
0.2
-0.2
-0.4
-0.6
0.5
1.5
2.5
2.5
Frequency (Rad/Sec)
0.6
0.4
Imaginary
0.2
-0.2
-0.4
-0.6
0.5
1.5
Frequency (Rad/Sec)
Figure 5-15. FRF, 2 DOF System, Non-Proportionally Damped, Second Mode
(5-34)
+UC-SDRL-RJA
CN-20-263-662
0.1
0.08
0.06
Amplitude
0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
0
10
15
20
25
Time (Sec.)
Figure 5-16. IRF, 2 DOF System, Non-Proportionally Damped, Second Mode
(5-35)
30
+UC-SDRL-RJA
CN-20-263-662
5.5 Summary of Modal Parameters (Undamped and Damped Systems)

Recalling the modal vectors from the previous examples:
Mode 1
Mode 2
Undamped Case:
1 = j
2/5
2/5
12
1 =
2/5
j
12
2 = j
j 15
2 =
1
j
30
Proportionally Damped Case:
1 =
1
39
+ j(
)
10
10
39
j 117
1 =
39
117
2 =
1
15
+ j(
)
4
4
4
15
j 225
2 =
2
15
j
225
Non-proportionally Damped Case:

1 = 0. 095363 + j 0. 629494
2 = 0. 754635 + j 0. 645996
0. 008569 j 0. 041970
0. 008569 j 0. 122700
2 =
0. 003473 j 0. 050100
0. 003473 + j 0. 045270
1 =
(5-36)
+UC-SDRL-RJA
CN-20-263-662
Note that for the non-proportional damped case, the residues given in the table are for the same
frequency response functions used in the undamped and proportionally damped cases (the first
column of the frequency response function matrix). This required the computation of the
residues for H 11 (s).
Recall that a pole ( r ) may be written as:
r = r + j r = ( r + j
1 r2 ) n
(5.35)
where:
r
= damping ratio
r = undamped natural frequency
= damping factor
= damped natural frequency
From Equation 5.35, the following relationships can also be noted:

r =
r
r2 + r2
r =
r
r
The first modal vector for the 3 different cases can now be compared. Note that the modal
vectors have been converted to amplitude and phase.
(5-37)
+UC-SDRL-RJA
CN-20-263-662
For modal vector one ( 1 ):

Undamped
Prop. Damped
Non-prop. Damped
0. 052705, 90
0. 052705, 90 1
0. 053376, 90
0. 053376, 90 1
0. 042836, 101. 53
0. 050220, 93. 96 1
1 = 0. 632456
1 = 0%
1 = 0. 632456
1 = 15.8114%
1 = 0. 63667
1 = 14.9782%
Undamped
Prop. Damped
Non-prop. Damped
0. 066667, 90
0. 033333, +90 2
0. 068853, 90
0. 034427, +90 2
0. 122999, 86. 00
0. 045403, +85. 61 2
2 = 1.0
2 = 0%
2 = 1.0
2 = 25%
2 = 0.99337
2 = 75.9671%
For modal vector two ( 2 ):
In conclusion, note that the addition of damping (proportional and non-proportional) has not
affected the undamped natural frequencies for either modal vector. This will always be the case.
Furthermore, the overall behavior of the modal vectors remained the same. That is, in the first
mode the 2 masses are moving in the same direction and in the second mode they are moving in
opposite directions. The major difference arises in the phase of the mode shapes. Note that for
the non-proportionally damped system, the phase of the components is other than 0 or 180.
These types of modes are again referred to as complex modes. How much the phase differs from
0 or 180 is directly related to the systems mass, damping, and stiffness distributions. Note that
as the phase approaches 0 or 180 for a particular mode, the modal vector can be approximated
as a real or normal mode by setting the phase equal to 0 or 180.
(5-38)
6. FREQUENCY RESPONSE FUNCTION SYNTHESIS

In this section, the approach for constructing the entire frequency response function matrix
[ H( ) ] will be explained. The frequency response function matrix [ H( ) ], which is made up
of N 0 N i frequency response functions, can be synthesized by using the data from N o
frequency response functions where N o is the number of measured response degrees of freedom
(physical response points times number of direction(s) at each physical point) on the structure.
For the following discussion, N o is assumed to be larger than N i and N o is assumed to include
the N i measurement degrees of freedom. For the case of fixed response degrees of freedom with
a large number of applied inputs (impact testing, for example), N i will be much larger than N o
but since the [ H( ) ] is reciprocal, the same assumption can be made concerning N i . These are
reasonable assumptions for all testing situations. There are two restrictions that apply to being
able to synthesize the complete matrix accurately from only N o or N i frequency response
functions. First of all, the N o or N i frequency response functions must consist of either a
complete row or complete column of frequency response function measurements from the
frequency response function matrix. Secondly, the N o or N i frequency response functions must
contain non-zero residue information for every modal vector present in the structure within the
frequency range of interest. This means that if the modal vector is entirely zero (due to
excitation at the node of a modal vector) proper frequency response function synthesis will not
be possible.
Recall the general form of the frequency response function matrix [ H( ) ] (Equation 4.25) in
terms of partial fractions.
H 11 ( )
H ( )
21
[ H( ) ] = .
.
H m1 ( )
H 12 ( )
.
.
.
.
(6-1)
.
.
.
.
.
.
.
.
.
.
H 1m ( )
.
.
.
H mm ( )
(6.1)
+UC-SDRL-RJA
CN-20-263-662
Assuming that only the k-th column of [ H( ) ] has been measured by way of frequency
response functions, the k-th column can be represented as follows:
H 1k ( )
H 2k ( )
.
.
H ik ( )
.
.
H mk ( )
N
{ Ar } k
{ A*r }k
+
=
*
r=1 ( j r ) ( j r )
(6.2)
where:
m = the number of measurements (N o or N i ).
Equation 6.2 is the mathematical description of the N o frequency response function

measurements that were obtained during the test of the structure. For a particular mode of
vibration, the N modal frequencies and the N residues for each of the N o frequency response
functions can be determined through the use of a modal parameter estimation algorithm.
Once these modal parameters are known, frequency response functions from column k or any
other column in the frequency response function can be synthesized using a partial fraction
model with the correct residues for the particular input and output degrees of freedom desired.
The synthesis is formulated on a frequency by frequency basis for each mode - the response for
each mode is summed together to get the total frequency response function as is shown in the
following figures:
(6-2)
+UC-SDRL-RJA
CN-20-263-662
-40
Log Magnitude (dB)
-50
-60
-70
-80
-90
-100
50
100
150
200
250
300
350
400
450
500
350
400
450
500
Frequency (Hertz)
0
-20
-40
Phase (Deg)
-60
-80
-100
-120
-140
-160
-180
50
100
150
200
250
300
Frequency (Hertz)
Figure 6-1. Frequency Response Function Synthesis
(6-3)
+UC-SDRL-RJA
CN-20-263-662
-40
Log Magnitude (dB)
-50
-60
-70
-80
-90
-100
50
100
150
200
250
300
350
250
300
350
Frequency (Hertz)
0
-20
-40
Phase (Deg)
-60
-80
-100
-120
-140
-160
-180
50
100
150
200
Frequency (Hertz)
(6-4)
+UC-SDRL-RJA
CN-20-263-662
40
50
Log Magnitude (dB)
60
70
80
90
100
110
500
400
300
200
100
0
100
Frequency (Hertz)
200
300
400
500
400
300
200
100
0
100
Frequency (Hertz)
200
300
400
500
200
150
100
Phase (Deg)
50
50
100
150
200
500
(6-5)
+UC-SDRL-RJA
CN-20-263-662
In order to construct all of the elements that make up [ H( ) ], all of the elements of [ A]r for
each mode will be required. Previously, the residue matrix [ A]r for a particular mode has been
shown to be directly related to the modal vector according to the following equation:
A = Q { } { }T = Q
r
r
r
r
r
1 1
1 2
2 1
2 2
.
.
.
.
m 1
m 2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1 m
.
.
.
m m
(6.3)
When the k-th column of [ H( ) ] has been measured, the k-th column of each residue matrix
[ A]r can be defined. Equation 6.3 can now be rewritten for only the k-th column of of the r-th
mode of vibration.
A1k
A2k
.
.
Aik
.
.
Amk
= Qr
1 k
2 k
= Q r kr
i k
.
m k r
.
.
1
2
.
.
i
.
.
m
(6.4)
where:
m = Number of measurements (N o or N i ).
r = Mode number
In order to compute the residue for any position in the residue matrix for each mode of vibration,
Equation 6.4 can be used to compute the values of Q r and { }r . In order to do this, either Q r or
kr must be chosen according to a scaling criteria. At this point the individual elements of { }r
can now be found. Once these individual elements are known, any residue for a given mode can
be computed using the following equation:
A pqr = Q r pr qr
(6-6)
(6.5)
+UC-SDRL-RJA
CN-20-263-662
If modal mass will be calculated, the individual values of Q r and { }r will be required.
Therefore, Equation 6.5 can be used easily to synthesize any arbitrary residue. In reality, there is
no need to know the individual values of Q r and { }r in order to synthesize another residue. If
the k-th column of the residue matrix [ A] r is again used to synthesize any arbitrary residue from
another location in the residue matrix, the following equation is all that is required:
A pqr =
A pkr Aqkr Q r pr kr Q r qr kr
=
Akkr
Q r kr kr
(6.6)
Note that with Equation 6.6, any arbitrary residue can be synthesized if the proper elements of
the frequency response function matrix and, therefore, the proper elements of each residue
matrix [ A]r have been measured. For example, the driving point information for the k-th column
of any residue matrix need not be measured directly if the following elements have been
measured:
Akkr =
A pkr Aqkr
A pqr
(6.7)
When the errors involved with measuring the driving point frequency response function are taken
into account, Equation 6.7 may be a preferable way to estimate the driving point information
even though this method requires information from another column of the residue matrix. Note
that this will require that another excitation location be taken during the test of the structure.
In summary, one row or column of the frequency response function matrix, including driving
point information, must be measured in order to synthesize any arbitrary measurement in the
frequency response function matrix. A modal parameter estimation algorithm is then used on
these measurements to determine the modal parameters. With the driving point residue
information for a particular mode of vibration, any other required residue information can be
synthesized by Equation 6.5 or Equation 6.6.
6.1 Displacement, Velocity and/or Acceleration Data

In practice, frequency response functions are often measured using an accelerometer to measure
the response of the system and a load cell to measure the input to the system. Velocity or
(6-7)
+UC-SDRL-RJA
CN-20-263-662
displacement transducers could also be used to measure the response but have disadvantages with
respect to size, configuration or datum. Regardless, the mathematical model that has been
developed is a representation of displacement normalized to the force that caused the
displacement. Note that if velocity normalized to force is actually measured, this data can be
synthetically integrated to displacement over force by dividing the velocity over force
measurement by j . Likewise, if acceleration normalized to force is actually measured, this data
can be synthetically integrated to displacement over force by dividing the acceleration over force
measurement by ( j )2 . This approach has numerical problems near zero frequency but is
generally satisfactory elsewhere.
6.2 Skewed Sensor Orientation

While it is common and somewhat desirable to align all sensors (input and output) in directions
that are colinear with a set of orthogonal axes (x, y, z), this is not a requirement. Directional
cosine information can be used to reorient the sensor information according to any desirable set
of axes. Note that the spatial subscripts used to this point ( pq for example) do not imply any
specific constraint in this sense.
However, if modal scaling or the ability to synthsize arbitrary functions that may or may not have
been measured is desired, the measurment of the driving point information is very important. If
a skewed excitation is used with a set of essentially orthogonal response sensors, the simplest
solution to this problem is to add a skewed response sensor, at the excitation location, so that a
true driving point measurement can be estimated. With this in mind, the Equations 6.1 through
6.7 still apply when synthesis of residues and frequency response functions from data taken from
a skewed input is the data represented by column k.
6.3 Simulation of Structural Response

Once all of the frequency response functions (transfer functions) for a particular system have
been measured or synthesized, the response of the structure to any arbitrary set of inputs can be
simulated or predicted. To illustrate the process, the two degree of freedom system that is
represented by the following equation can be used as an example.
(6-8)
+UC-SDRL-RJA
CN-20-263-662
X 1 ( ) H 1 1 ( )
=
X
(
)
2
H 2 1 ( )
H 1 2 ( ) F 1 ( )
H 2 2 ( ) F 2 ( )
(6.8)
Equation 6.8 is the frequency response function form of Equation 4.9 ( s = j ). The matrix
equations in Equation 6.8 can be expanded as follows:
X 1 ( ) = H 1 1 ( ) F 1 ( ) + H 1 2 ( ) F 2 ( )
X 2 ( ) = H 2 1 ( ) F 1 ( ) + H 2 2 ( ) F 2 ( )
Note that X 1 ( ) and X 2 ( ) are the response of the system (in the frequency domain) due to the
forces F 1 ( ) and F 2 ( ). Thus, X 1 ( ) and X 2 ( ) can be defined by the process of multiplication
and addition of the known frequency response function measurements and forcing functions.
The time domain response of the system, x 1 (t) and x 2 (t), can then be computed by taking the
inverse Fourier transform of X 1 ( ) and X 2 ( ).
This synthesis procedure can be extended to systems with more than two degrees of freedom. In
general, for an N -degree of freedom system:
X 1 ( )
X 2 ( )
.
.
.
X N ( )
H 1 1 ( ) H 1 2 ( )
H 2 1 ( ) H 2 2 ( )
.
.
.
.
.
.
H N 1 ( ) H N 2 ( )
. . . . H 1 N ( )
. . . . H 2 N ( )
.... .
.... .
.... .
. . . . H N N ( )
F 1 ( )
F 2 ( )
.
.
.
F N ( )
(6.9)
While Equation 6.9 is written in terms of the number of degrees of freedom N , this is the
limiting theoretical consideration. In general, Equation 6.9 can be written in terms of any
arbitrary number of inputs and responses such that [ H(s) ] is no longer a square matrix.
For instance, the time domain response x 1 (t) is just the inverse Fourier transform of X 1 ( )
where:
X 1 ( ) = H 1 1 F 1 ( ) + H 1 1 F 2 ( ) + . . . .
(6-9)
+ H 1N F N ( )
+UC-SDRL-RJA
CN-20-263-662
Theoretically, the transfer function matrix [ H(s) ] consists of N N transfer functions for an N degree of freedom system. From a practical point of view, the measured portion of the transfer
function matrix (frequency response function matrix) will consist of N o N i frequency response
functions regardless of the number of modes N .
(6-10)
+UC-SDRL-RJA
CN-20-263-662
7. MODAL SCALING
7.1 Proportionally Damped Systems (Modal Mass)
The modal matrix (matrix of modal vectors) has been previously used as a coordinate
transformation in order to diagonalize the mass, damping (if proportional), and stiffness matrices.
The diagonalization of these matrices leads to the analytical definition of modal mass, modal
damping, and modal stiffness. From an experimental view point, the mass, damping, and
stiffness matrices are generally not known. Therefore, this theoretical approach to the definition
of modal mass, modal damping, and modal stiffness is not useful. Even so, the modal mass,
modal damping, and modal stiffness can be computed directly from the measured frequency
response functions without the benefit of prior knowledge of the mass, damping, and stiffness
matrices. Note that the modal mass, modal damping, and modal stiffness are, in general, not
physical properties but are generalized, or normalized, properties related to the physical
properties.
Recall that, analytically, the value of the modal mass is completely dependent on the scaling
chosen for the modal vectors. Any development from the frequency response function
(experimental) must be consistent with this concept. The development in Section 4.7 (Equation
4.51) satisfies this constraint. Equation 4.51 is repeated here as Equation 7.1.
Mr =
1
j 2 Qr r
(7.1)
The scaling coefficient Q r in the above definition can only be found after the choice of modal
vector scaling is made.
Recal that for the r-th mode of an N degree of freedom system:
[ A ]r = Q r { }r { }Tr
1 1

2 1
[ A ]r = Q r .
.
m 1
1 2
2 2
.
.
.
.
.
.
.
.
.
.
.
.
.
(7-1)
.
.
.
.
.
(7.2)
.
.
.
.
.
.
.
.
.
.
1 m
.
.
.
m m
(7.3)
+UC-SDRL-RJA
CN-20-263-662
where:
Q r is the scaling constant that is a function of the scaling of the modal vectors.
Using only the q-th column of the residue matrix:
A1q
A2q
.
.
Aiq
.
.
Amq
= Qr
1 q
2 q
= Q r qr
i q
.
.
m q r
.
.
1
2
.
.
i
.
.
m r
(7.4)
The relationship between the residue, the scaling constant and the scaling of the modal vector is
most clear when the FRF measurement H qq ( ) is utilized. Note that H qq ( ) is obtained by
exciting at point q and measuring the response at point q. This is typically referred to as the
driving point frequency response function. Therefore, the Aqqr residue for all modes r = 1 N
can be determined from the H qq ( ) measurement.
Now, note the q-th element of Equation 7.4:
2
Aqqr = Q r qr qr = Q r qr
(7.5)
Therefore, the r-th modal mass of a multi-degree of freedom system is defined as:
Modal Mass
Mr =
1
j 2 Qr r
(7.6)
In general:
Mr =
pr qr
j 2 A pqr r
(7-2)
(7.7)
+UC-SDRL-RJA
CN-20-263-662
where:
M r = Modal mass
Q r = Modal scaling constant
= Damped natural frequency
While this concept of modal mass being a relative quantity (relative to scaling) is at odds with the
engineering view of mass being an absolute quantity, this is consistent with the analytical
definition of modal mass.
While many choices of modal vector scaling exist, choosing to scale the modal vector such that
the largest modal coefficient will be equal to 1.0 gives a result where the modal mass will always
be bounded between zero and the physical mass of the system. For the case where the modal
vector of the system describes the rigid body translation (bounce) of the system, this particular
choice will give the mass of the system as the modal mass. Therefore, if the largest scaled modal
coefficient is equal to unity, Equation 7.7 will compute a quantity of modal mass that has
physical significance. The physical significance is that the quantity of modal mass computed
under these conditions will be a number between zero and the total mass of the system.
Therefore, under this scaling condition, the modal mass can be viewed as the amount of mass
that is participating in each mode of vibration.
Note that the modal mass defined in Equation 7.6 or 7.7 is developed in terms of displacement
over force units. If measurements, and therefore residues are developed in terms of any other
units ( velocity over force or acceleration over force), either the measurements or Equation 7.6
and 7.7 will have to be altered accordingly.
Once the modal mass is known, the modal damping and stiffness can be obtained through the
following SDOF relationships:
Modal Damping
Cr = 2 r Mr
(7-3)
(7.8)
+UC-SDRL-RJA
CN-20-263-662
Modal Stiffness
K r = ( r2 + r2 ) M r = 2r M r
(7.9)
7.1.1 Modal Vector Scaling

There are many ways to scale the modal vectors, however the following approaches are typically
used:
Unity Modal Mass
Unity Modal Coefficient
Unity Modal Vector Length
The scaling value, Q r , can be calculated for each of the three cases.
7.1.1.1 Unity Modal Mass

Qr =
1
1
=
j 2 Mr r
j 2 r
The scaled modal coefficient at the driving point can now be computed as follows:
Q r qr qr = Aqqr
Dividing both sides by Qr :
qr qr =
Aqqr
Qr
The scaled modal coefficient at the driving point can now be found. For the proportionally
damped case, this is obviously trivial since the square root of both sides of the above equation
involves the square root of a real-valued number. For the general case, the modal coefficient can
be complex, and the above equation must be solved for the complex modal coefficient. In this
case the square root cannot be used.
(7-4)
+UC-SDRL-RJA
CN-20-263-662
Thus, the scaled modal vector is:

{ }r =
1
Q r qr
{ A }r
7.1.1.2 Unity Modal Coefficient

Assume that the i-th component i of a modal vector must be set equal to unity ( i = 1.0). Then:
{ }r =
1
{ A }r
Aiq
Thus:
Qr =
Aiqr
ir qr
7.1.1.3 Unity Modal Vector Length

{ }r =
1
{ A }r
||{ A}r ||2
where:
||{ A}r ||2
= vector norm of { A}r
||{ A}r ||2
Aiqr A*iqr
i=1
Thus:
Qr =
Aiqr
ir qr
(7-5)
+UC-SDRL-RJA
CN-20-263-662
7.1.2 Modal Vector Scaling Example

Using the modal vectors from the example in Chapter 5, the modal vector scaling required for
unity modal mass can be determined.
The modal vectors for modes 1 and 2 are:
A11
A21
39
j

117
=
39
1 j
117
A11
A21
4
15
j

225
=
2
15
2 + j
225
1 =
1
39
+ j
10
10
2 =
1
15
+ j
4
4
For unity modal mass, the scaling factor can be evaluated from Equation 7.6.
Qr =
1
j 2 r
Thus, the scaled modal vector for mode 1 for a modal mass of unity is:
1
A11
= Q1 1
2 1
A21 1
1
1/15
1/15
2 1
1
The scaled modal vector for mode 2 for a modal mass of unity is:
1
2/15
= 1
2/15
2 2 2
1
A11
= Q2 1
2 2
A21 2
Notice that these modal vectors that have been scaled with respect to unity modal mass, are
identical to the modal vectors previously computed (Section 3.2). The important concept,
(7-6)
+UC-SDRL-RJA
CN-20-263-662
though, is that these scaled modal vectors were determined directly from the frequency response
functions. In other words, the mass, damping, and stiffness of the system did not have to be
known in order to determine the same information.
The modal damping and modal stiffnesses can now be calculated using Equations 7.8.and 7.9.
1
)1
10
C1 = 2 1 M1
=2(
C2 = 2 2 M2
1
=2( )1
4
1
5
1
5
2
5
K 1 = ( 12 + 12 ) M 1 = (
39
1
+
)1
100 100
K 2 = ( 22 + 22 ) M 2 = (
15
1
+
)1
16 16
=1
These are exactly the same parameters as those calculated previously.
7.2 Non-Proportionally Damped Systems (Modal A)

Consistent with the definition of modal A as the modal scaling factor used for the theoretical case
of nonproportionally damped systems, the modal A scaling factor is also the basis for the
relationship between the scaled modal vectors and the residues determined from the measured
frequency response functions. In general, for most experimental work, modal A is used as the
default scaling approach. If modal mass needs to be estimated, under the contraint of real
(normal) modes, modal mass can be estimated from modal A. The following development
explains the relationship between modal A and modal mass. Starting with the anlytical definition
of modal mass in 2N space:
[0]
M Ar = { }Tr
[ M ]
T
M Ar
{ }r [ 0 ]
= r

{ }r [ M ]
(7-7)
[ M ]
{ }r
[C ]
(7.10)
[ M ] r { }r
[ C ] { }r
(7.11)
+UC-SDRL-RJA
CN-20-263-662
Multiplying Equation 7.11 out in terms of N space yields:

M Ar = r { }Tr [ M ] { }r + r { }Tr [ M ] { }r + { }Tr [ C ] { }r
(7.12)
If the system is proportionally damped, the weighted orthogonality relationships between the
mass matrix and the modal vectors can now be applied.
M Ar = r M r + r M r + C r = 2 r M r + C r
(7.13)
Applying the SDOF relationship between the modal damping and modal mass (C r = r M r ):
M Ar = 2 r M r 2 r M r
(7.14)
M Ar = 2 ( r + j r ) M r 2 r M r
(7.15)
M Ar = j2 r M r
(7.16)
Equation 7.16 indicates that, for proportionally damped systems, if the modal vectors are scaled
to give real valued normal modes and the modal mass is, therefore, real valued, the associated
modal A for a proportionally damped system is imaginary valued.
While Equation 7.16 is valid only for proportionally damped systems, another form of Equation
7.16 gives a more general result that includes any type of damping. Equation 4.51 is repeated
here for convenience.
Mr =
1
j2 r Q r
(7.17)
Plugging Equation 7.17 into 7.16 yields:

M Ar =
1
Qr
(7.18)
Since the basic development of the relationship between the residue and the modal vector
coefficients and associated modal scaling ( A pqr = Q r pr qr ) did not depend upon the assumption
of proportional damping, Equation 7.18 is valid for any damping condition and is the most
(7-8)
+UC-SDRL-RJA
CN-20-263-662
general form of modal scaling. Note that in general, modal A will be complex valued. Equation
7.18 can also be written in an equivalent form that clearly expresses the dependence of modal A
on the modal vector scaling.
M Ar =
pr qr
1
=
Qr
A pqr
(7.19)
Note that the above development of modal A is in terms of displacement over force units. If
measurements, and therefore residues are developed in terms of any other units ( velocity over
force or acceleration over force), Equation 7.19 or the FRF data will have to be altered
accordingly.
Consistent with the modal B scaling value defined previously from an analytical viewpoint,
modal B can be determined once modal A is known by way of the modal frequency.
M B r = r M Ar
(7.20)
7.3 Modal Mass, Modal A Units Discussion

In general, the modal vectors are considered to be dimensionless since they represent relative
patterns of motion. Therefore, the modal mass or modal A scaling terms are considered to carry
the units of the respective measurement. For example, the development of the frequency
response is based upon displacement over force units. The residue must therfore, have units of
length over force-seconds. Since the modal A scaling coefficient is inversely related to the
residue, modal A will have units of force-seconds over length. This unit combination is the same
as mass over seconds. Likewise since modal mass is related to modal A, for porportionally
damped systems, through a direct relationship involvinng the damped natural frequency, the units
on modal mass are mass units as expected.
The following table summarizes the units of modal A and modal mass for typical consistent unit
applications.
(7-9)
+UC-SDRL-RJA
CN-20-263-662
Consistent Units Relationships

Mass
Force
KG
NT
KG
Length
Modal A (M Ar )
Modal Mass (M r )
M/S
Meter
KG/S
KG
KGf
g-S-S
KG/S
KG
LBm
LBf
g-S-S
LBm/S
LBm
Slug
LBf
Feet
Slug/S
Slug
(7-10)
+UC-SDRL-RJA
CN-20-263-662
8. ADVANCED MODAL ANALYSIS CONCEPTS

8.1 Introduction
As the theoretical basis of expermental modal analysis is extended to real world problems,
several clarifications of the theory developed to the present time must be made. The
development of more general and/or concise models to represent the entire frequency response
function matrix [H( )] or impulse response function matrix [h(t)] is the primary concern. These
models must be consistent with the single reference concepts developed previously but must be
compatible with multiple reference concepts as well. Therefore, the general concept of
measurement degree of freedom must be extended to account for the multiple input, multiple
output nature of the problem. Two other concepts must also be discussed in order to fully
develop the theoretical basis for experimental modal analysis. These concepts include systems
that have repeated modal frequencies (repeated roots) and systems that can not be considered
reciprocal.
8.2 Measurement Degrees of Freedom

For the general situation of a multiple input, multiple output model of a system, the experimental
definition of the mechanical system is generated from the frequency, or impulse, response
function matrix. The size of this matrix is a function of the locations where forces are applied to
the mechanical system (inputs) and a function of the locations where responses of the mechanical
system (outputs) are measured. This general concept is often referred to as measurement degrees
of freedom to distinguish the size of the matrix from the number of modal frequencies N of the
mechanical system. Obviously, since there is no reason to assume that the number of inputs will
be the same as the number of outputs, this general concept of measurement degrees of freedom
needs to be extended to properly reflect that the dimension of the frequency, or impulse, response
function matrix is rectangular. With this in mind, the number of inputs can be defined by N i and
the number of outputs can be defined by N o . Therefore, the dimension of the frequency, or
impulse, response function matrix is N o N i .
(8-1)
+UC-SDRL-RJA
CN-20-263-662
8.3 Mathematical Models

The mathematical model that represents the relationship between the modal parameters and the
measured frequency, or impulse, response functions can be represented as follows:
Frequency Response Function Model
Single Measurement:
k ( j )k H pq ( ) =
k ( j ) k
k=0
(8.1)
A*pqr
A pqr
+
j r
j *r
(8.2)
k=0
H pq ( ) =
r=1
Multiple Measurement:
[ ] ( j )k [H( )] =
k
k=0
[H( ) ] N o Ni =
r=1
[ ] ( j )k [I ]
k
k=0
A*
A
r + r
j r
j *r
(8-2)
(8.3)
(8.4)
+UC-SDRL-RJA
CN-20-263-662
Impulse Response Function Model

Single Measurement:
h pq (t) =
A pqr e t + A*pqr e t
r=1
*
r
(8.5)
Multiple Measurement:
[h(t) ] N o Ni =
A e t + A* e t
r
r=1
r
*
r
(8.6)
where:
s = Laplace variable
s = + j = Angular damping variable (rad/sec)
= Angular frequency variable (rad/sec)
p = Measured degree-of-freedom (response)
q = Measured degree-of-freedom (input)
r = Modal vector number
m = Number of poles or modal frequencies (2N)
n = Number of zeroes (2N-2 or less)
N = Number of positive modal frequencies
A pqr = Residue = Q r pr qr
Q r = Complex modal scaling coefficient for mode r
pr
= Modal coefficient for measured degree-of-freedom p and mode r
[ Ar ] = Residue matrix for mode r ( N o N i )
= System pole = r + j r
While these models are perfectly appropriate for the multiple input, multiple output case, by a
(8-3)
+UC-SDRL-RJA
CN-20-263-662
slight alteration of these models a more appropriate form of the models can be developed which
will facilitate the development of parameter estimation algorithms.
First of all, the summation form of the equations can be simplified from two terms to one term as
follows:
[H( ) ] =
2N
r=1
[h(t) ] =
A
r
j r
(8.7)
2N
A e t
r
r=1
r
(8.8)
If these forms of the mathematical models are used, no assumption is made in the model
concerning the complex conjugate nature of the solution for modal frequencies ( r ) or modal
vectors ( { r } ). When the modal parameters are estimated, the evaluation of modal parameters
can include a comparison of these terms to determine whether the complex conjugate nature of
the solution is found.
Finally, the summation in the mathematical models can be eliminated completely if a different
form of the residue is used. In order to do this, the concept of modal participation factor ( L qr )
is introduced. Physically, the modal participation factor is a relative indication of how well a
particular mode of vibration is excited from a specific measurement degree of freedom. If all of
the modal participation factors for a specific modal vector are represented in a row, this vector is
referred to as the modal participation vector and has dimension of 1 N i . The modal
participation vector is not unique (has properties of an eigenvector) but in combination with the
modal coefficient defines the residue in the following way.
A pqr = Q r pr qr
(8.9)
A pqr = L qr pr
(8.10)
{ A }qr = L qr { }r
(8.11)
L qr = Q r qr
(8.12)
(8-4)
+UC-SDRL-RJA
CN-20-263-662
Equations 8.10 and 8.11, therefore, are the general statements relating the residue, the modal
participation factor, and the modal coefficient. For the general case, the modal coefficient can be
thought of as an element from the right eigenvector of the system; the modal participation factor
can be thought of as an element from the left eigenvector of the system. With this in mind, when
a system obeys Maxwells Reciprocity, the right and left eigenvectors represent the same modal
vector, Equations 8.9 and 8.12 are valid. For the general (nonsymmetric) case, though, only
Equations 8.10 and 8.11 will be valid.
From the above equations it is important to note that only the residue can be absolutely and
uniquely defined. Both the modal coefficient pr and the modal participation factor L qr are a
function of one another and can take on any value; only the combination of the two terms is
unique.
Since the term modal mass can be defined in terms of the modal scaling Qr , the modal mass can
also be defined in terms of the modal participation factor L qr .
Mr =
qr
(8.13)
2 jL qr r
If this simplification is made, the frequency response function model can now be written in the
following form:
[ ]2N 2N [L ]2N Ni
[H( ) ] N o Ni =
N o 2N
(8.14)
where:
[ ]2N 2N
j 1
=
0
0
1
j 2
0
........
0
1
j r
0
........
0
1
j 2N
(8-5)
+UC-SDRL-RJA
CN-20-263-662
Likewise, a similar simplification can be developed for the impulse response function matrix:
e r t
[L ]2N Ni
[h(t)] N o Ni =
N o 2N
2N 2N
where:
e r t
2N 2N
e 1
0
0
=
0
0
0
2t
0
0
0
0
0
0
........
0
0
0
0
0
0
e r t
0
0
0
0
0
0
........
0
(8.15)
e 2N t
0
0
0
0
0
Since the modal participation vector { L }r is unique only to within a complex scaling constant,
several other definitions of the modal participation factors are possible. One common form can
be defined by scaling the modal participation such that a specific element of the vector, L pr , is
unity. If this normalization is used, the following definition of modal participation factor will
apply:
L pqr =
L qr
Q r qr qr
=
=
L pr Q r pr pr
(8.16)
This form of the definition of modal participation factor is used in the development of the theory
for the Polyreference Time Domain modal parameter estimation algorithm.
8.4 Repeated Modal Frequencies

Repeated modal frequencies occur whenever two or more modes of the system occur at exactly
the same modal frequency r . This condition is often also referred to as repeated roots or
repeated poles. Analytically, the presence of repeated roots is determined directly from the
characteristic equation just as in any other case. The modal vectors, associated with the repeated
roots, now come from a system of equations (the homogeneous equations evaluated at a repeated
root) which will be rank deficient by more than one. This means that rather than choosing one of
the physical coordinates, as is the case for the non-repeated root situation, a number of physical
(8-6)
+UC-SDRL-RJA
CN-20-263-662
coordinates, equal to the number of repeated roots, must be assumed. This process is repeated
once for each of the repeated roots. Each of the vectors found in this manner will in general be
independent to one another and orthogonal to all other modal vectors. Each vector within this
set, however, will not necessarily be orthogonal to each other. While this is not a problem
mathematically, it may be more consistent to find a set of vectors that are also orthogonal to one
another. This can be accomplished by additionally imposing the cross-orthogonality constraints
between the vectors. Using any mathematical procedure to determine a set of orthogonal vectors
from a set of independent vectors will also work as long as the weighting matrix (mass and/or
stiffness) is utilized in the procedure. Note that there is an infinite number of modal vector sets
that will satisfy repeated root situation, whether the orthogonality constraint is enforced or not.
Experimentally, it is important to detect the presence of repeated roots in order to build a
complete modal model of the mechanical system that will accurately represent the dynamic
response of the mechanical system to any set of forcing conditions. The most common cause of
this situation is symmetry in the mechanical system. For example, in the case of a flagpole, there
are two modes of vibration occurring at the same frequency for each lateral bending mode of the
flagpole. Any time one or more axes of symmetry exist in the mechanical system, this condition
will exist. The important consideration, though, is that, in order to detect the repeated root
condition, more than one row or column of the frequency, or impulse, response function matrix
must be measured. Therefore, to detect a repeated root of order two, two independent rows or
columns of the frequency, or impulse, response function matrix must be used. Note that it is
possible to choose two rows or columns that are not independent and thus miss the repeated root.
The modal vectors that are associated with repeated modal frequencies are independent of one
another and each is orthogonal (weighted) to the other modal vectors of the set. Even so, the
modal vectors associated with the repeated modal frequencies do not individually have fixed
patterns even in the relative sense that modal vectors associated with nonrepeated modal
frequencies do. Only when the set of modal vectors associated with repeated modal frequencies
are considered as a set, are the modal vectors unique in any way. In the case of a repeated modal
frequency of multiplicity two, two independent modal vectors will be required to describe the
modal space but any two independent modal vectors will do. An analogy in three dimensional
graphics involves using two vectors to define a plane. There is an infinite set of vectors that can
be used to describe the same plane. Any two vectors lying in that plane can be used to uniquely
define the plane and yet the two vectors are not unique but only independent from one another.
(8-7)
+UC-SDRL-RJA
CN-20-263-662
In this case, independent simply means that the two vectors are not scalar multiples of one
another.
Therefore, there are an infinite number of combinations of modal vectors which will serve as the
modal vectors for the repeated modal frequencies. In this case, the exact characteristic of each
modal vector is unimportant since the set of modal vectors will always be considered together.
The modal vectors associated with the repeated modal frequencies often appear to be exactly the
same at first glance. This is due to the symmetry and on close inspection using physical
coordinates and directions, the distinction can be easily detected. Again consider the flagpole
example. Since there is no way to define a unique x and y direction with respect to the circular
cross section of the flagpole, there will not be a single set of modal vectors that will describe the
modal deformation at a repeated modal frequency.
At this point, repeated modal frequencies may seem to be only a theoretical concept that does not
have much impact on real structures that are not symmetric. Actually, due to the discrete nature
of the frequency response function, a very real problem often exists where several modal
frequencies occur between the frequency resolution that is used. Since a priori knowledge of the
modal density is not normally possible, this condition happens quite frequently. This is referred
to as psuedo-repeated modal frequencies. While the modal vectors do not theoretically have the
same attributes as in the repeated root case, for all practical purposes, the result is the same.
For the case of a mechanical system with repeated roots, or psuedo-repeated roots, Leuridan
[1]
has shown that the same mathematical model can be used to represent the relationship between
measured frequency, or impulse, response function data and modal parameters. For a mechanical
system with repeated modal frequencies of order N r the following basic relationship for
frequency response functions will apply.
H pq ( ) =
2N
H pq ( ) =
r=1
A pq1
A pq2
+
+ ........ +
j 1
j 2
Nr
s=1
(8-8)
A pqr
j r
(8.17)
A pqs
A pqr
+ ........ +
+ ........
j s
j r
(8.18)
+UC-SDRL-RJA
CN-20-263-662
With the above model, several poles could be repeated; the multiplicity of any pole can be at
most N .
In order to understand the implications of the repeated modal frequency on the residue
information, the above equation can be rewritten in terms of column q of the frequency response
function matrix.
{H }q =
Q1 q1 { }1 Q2 q2 { }2
+
j 1
j 2
+ ........ +
Nr
s=1
Q s qs { }s
j s
+ ........ +
Q r qr { }r
+ ........
j r
(8.19)
The implication of this representation is that if only one column of the frequency response
function matrix is measured, the residue that will be estimated for the repeated modal
frequencies will be the linear combination represented by the summation in the above equation
[1]
. Note that this linear combination is not unique; if a different column of the frequency
response function is used, the residue column for the repeated modal frequencies will not be the
same. This observation yields the simplest procedure for the detection of repeated or psuedorepeated modal frequencies. If the modal vectors that are estimated from different reference
positions (different inputs) are not the same, the modal vectors are probably the result of repeated
modal frequencies. Note that the modal vectors associated with the repeated modal frequencies
are independent of one another and not simply a scalar multiple of one another as would be the
case for a nonrepeated modal frequency.
Therefore, if the repeated modal frequency condition is not detected, the residue vector,
associated with the repeated modal frequency, that will be estimated from the frequency response
function data taken from only one reference will represent a linear combination of the modal
vectors associated with the repeated modal frequencies. If frequency response function data
from a second reference is observed without knowledge of the first reference (and the repeated
modal frequency condition is still not detected) the residue vector will again represent a linear
(8-9)
+UC-SDRL-RJA
CN-20-263-662
combination of the modal vectors associated with the repeated modal frequencies.
Unfortunately, this residue vector will not represent the same modal vector when compared to the
previous estimate. This means that knowledge of any single reference set of data will be
insufficient to describe the repeated root situation.
Nevertheless, if data from several references is available, a characteristic of the modal vectors
associated with the repeated root is possible. In general, for a repeated root of order N r , at least
N r references will be required to make this determination. Unlike in the non-repeated condition,
though, the individual modal vectors determined by this process will not be unique. Only the
combination of vectors will represent a unique characteristic. For example, if a system contains
a repeated modal frequency s of order 2, the two residue columns that will be estimated for
columns p and q can always be represented as follows:
For column p:
{ A } p = Q11 p1 { }1 + Q22 p2 { }2
(8.20)
{ A }q = Q11 q1 { }1 + Q22 q2 { }2
(8.21)
For column q:
Therefore, in matrix notation:
{ A } { A } = { } { } Q11
p
q
1
2

0
0 p1 q1
Q22 p2 q2
(8.22)
The above equation states that, when the modal vectors, { }1 and { }2 , are excited in
independent combinations in column p and column q of the frequency response function matrix,
then the residue vectors, { A } p and { A }q , are independent and in turn define two independent
modal vectors that always can be normalized such that the modal scaling matrix, [ Q ], is
diagonal.
For the general case of a repeated modal frequency of order N r , there will be a set of N r
(8-10)
+UC-SDRL-RJA
CN-20-263-662
independent modal vectors that will satisfy Eq. 8.22. Note that any set of N r modal vectors that
are independent linear combinations of the modal vectors defined by Eq. 8.22 can be used. In
this case, the modal scaling matrix, [ Q ], will not be diagonal. This case is represented by the
following equation and will be demonstrated in a later example.
{ A } { A } = { } { } [Q ] p1 q1
p
q
1
2

p2 q2
(8.21)
8.4.1 Repeated Modal Frequency Example: Residue Synthesis

In order to fully understand the nature of the problem that can arise due to repeated modal
frequencies, a simple example may be used. Consider a repeated modal frequency, s , of order
two in a three degree of freedom system. The following represents two independent modal
vectors associated with the two repeated modal frequencies.
1
{ }1 = 2
3
{ }2 = 1
3
For this example, assume that the modal vectors given above have been scaled such that the
modal scaling matrix, [ Q ], is diagonal with elements on the diagonal equal to j.
j
[Q ] =
0
For the following discussion, only the portion of the frequency response function matrix, [ H ],
that depends upon the repeated modal frequencies will be synthesized.
To begin with, all of the columns of the frequency response function matrix will be synthesized
(8-11)
+UC-SDRL-RJA
CN-20-263-662
from the theoretical modal data (Case 1, 2, 3). This data represents the true answer or the answer
that would be generated from the measured frequency response functions if the repeated root
condition was not detected.
Case Number One: Column 1 of [H ]
Case One represents the proper synthesis of the repeated root portion of Column 1 of the
frequency response function matrix.
1
2

j1 2
j2 1
3
3
2
1
{H }1 = . . . . . . . . +
+
+ ........
j s
j s
5

j 4
9
+ ........
{H }1 = . . . . . . . . +
j s
4
j
5
5
5
{H }1 = . . . . . . . . +
j s
+ ........
Notice that if only this column of the frequency response function matrix is measured, there is no
reason to suspect that a repeated modal frequency exists. Therefore, the number of modal
frequencies and modal vectors would be reduced accordingly.
(8-12)
+UC-SDRL-RJA
CN-20-263-662
Case Number Two: Column 2 of [H ]

Case Two represents the proper synthesis of the repeated root portion of Column 2 of the
1
2

j2 2
j1 1
3
3
2
1
{H }2 = . . . . . . . . +
+
+ ........
j s
j s
4

j 5
9
+ ........
{H }2 = . . . . . . . . +
j s
j
5
5
9
{H }2 = . . . . . . . . +
j s
+ ........
Once again notice that if only this column of the frequency response function matrix is
measured, there is no reason to suspect that there is a repeated modal frequency. If the modal
vector for the modal frequency s is compared to that found from column 1 of the frequency
response function matrix, it is obvious that a different modal vector has been estimated. This
comparison is the simplest method of detecting a repeated modal frequency.
(8-13)
+UC-SDRL-RJA
CN-20-263-662
Case Number Three: Column 3 of [H ]

Case Three represents the proper synthesis of the repeated root portion of Column 3 of the
1
2

j3 2
j3 1
3
3
2
1
{H }3 = . . . . . . . . +
+
+ ........
j s
j s
9

j 9
18
+ ........
{H }3 = . . . . . . . . +
j s
18
9
j
18
18
18
{H }3 = . . . . . . . . +
j s
+ ........
Notice that the modal vector determined from this column of the frequency response function is
once again different from either of the first two columns. This is a characteristic of the modal
vector resulting from a repeated modal frequency; if the repeated modal frequency is not
detected, the modal vector appears to be different as a function of the reference (input) location.
If only the first column of the frequency response function matrix [H ] has been measured and
the repeated modal frequency has not been identified, the synthesis of column 2 and column 3
can be attempted from the measured first column as follows:
(8-14)
+UC-SDRL-RJA
CN-20-263-662
Case Number Four: Column 2 of [H ]

Case Four represents the improper synthesis Column 2 of the frequency response function if only
frequency response function data from Column 1 is used (repeated root not detected).
4 4
j
5
5
5
{H }2 = . . . . . . . . +
j s
+ ........
4
16
j
5
36
{H }2 = . . . . . . . . +
+ ........
j s
Note that this synthesis does not agree with Case 2 which is the theoretical result. Therefore, if
only column 1 of the frequency response function matrix is measured, the repeated modal
frequency cannot be detected and the proper dynamic characteristics of the system, when excited
at location 2, cannot be predicted.
(8-15)
+UC-SDRL-RJA
CN-20-263-662
Case Number Five: Column 3 of [H ]

Case Five represents the improper synthesis Column 3 of the frequency response function if only
frequency response function data from Column 1 is used (repeated root not detected).
9 4
j
5
5
5
{H }3 = . . . . . . . . +
j s
+ ........
9
36
j
5
81
{H }3 = . . . . . . . . +
+ ........
j s
Once again the result for this case should compare to Case 3 (the theoretical case) but does not.
Therefore, if only column 1 of the frequency response function matrix is measured, the repeated
modal frequency cannot be detected and the proper dynamic characteristics of the system, when
excited at location 3, cannot be predicted.
8.4.2 Repeated Modal Frequency Example: Modal Vector Solution

Using the residue data generated from two independent columns of the frequency response
function matrix, the set of independent modal vectors for the repeated modal frequency can be
determined if the order of the repeated modal frequency is known. This can be estimated via
singular value decomposition techniques. For the previous example, assuming that the
multiplicity of the repeated modal frequency is 2, the results can be determined by using the first
(8-16)
+UC-SDRL-RJA
CN-20-263-662
two columns of the frequency response function matrix (Case 1 and Case 2).
5 4
j 4 5
9
9
+ ........
{H } {H } = . . . . . . . . +
1
2
j s
Since the residue columns found from the two columns are independent, these residue columns
can be used as the [ ] matrix in Eq. 8.21. In this case, though, there is no reason to assume that
the modal scaling matrix [ Q ] is diagonal. With this in mind, Eq. 8.21 can be used in Eq. 8.2 as
follows:
5
j 4
{H } {H } = . . . . . . . . + 9
1
2
4
5
9
5
4
j s
+ ........
While it may not be immediately obvious, the modal scaling matrix in the above equation must
be chosen so that the product of the three matrices and the scalar j yields the proper residues for
the first two columns. The inverse of this matrix is the only matrix that satisfies this condition.
5
j 4
{H } {H } = . . . . . . . . + 9
1
2
4
5
9
5/9
4/9
4/9 5
5/9 4
j s
+ ........
Now using these two residue columns and the modal scaling matrix [ Q ] the synthesis of the
residue for column 3 is possible.
5
j 4
9
{H }3 = . . . . . . . . +
4
5
9
5/9
4/9
4/9 9

5/9 9
j s
9
j 9
18
(8-17)
+ ........
{H }3 = . . . . . . . . +
j s
+ ........
+UC-SDRL-RJA
CN-20-263-662
This is the correct answer as previously defined by Case 3. Note that it is possible to find a set of
modal vectors such that the corresponding modal scaling matrix [ Q ] is diagonal. This is done
by substituting the eigenvalue decomposition of [ Q ] into the above equation. While this is
possible and makes the theory consistent, it is unnecessary as the previous example proved.
Question?
If the modal vectors corresponding to a repeated modal frequency with multiplicity of two are:
1

{ }1 = 2
1

2

{ }2 = 1
2

Is it possible to correctly identify the modal vectors from the first and third columns of the
frequency response function matrix if the multiplicity of the repeated modal frequency is known?
(Hint: Are the first and third columns of the frequency response function matrix independent?)
8.5 Left and Right Eigenvectors

The relationship between the modal participation vector and the modal vector is particularly
interesting when the theoretical matrix equation of motion is examined. For the case of a system
that obeys Maxwells reciprocity, the mass, stiffness, and damping matrices will be symmetric
matrices. With this in mind, the basic eigenvalue-eigenvector problem that is generated is of the
following form:
2 [M ] + [C ] + [K ] { } = {0 }
r
r
(8.22)
In the above equation, note that the dimension of { }r is the same as the dimension of the
(8-18)
+UC-SDRL-RJA
CN-20-263-662
square mass, stiffness, or damping matrix. Since this theoretical problem is of dimension N N ,
there is no difference between N , N o, or N i . Therefore, the modal participation vector is also of
the same dimension.
The eigenvector { }r is also referred to as the right eigenvector of the system. If the
eigenvalue-eigenvector problem is reformulated, the eigenvector is also the left eigenvector of the
system.
{ }r T r 2 [M ] + r [C ] + [K ] = {0 }
(8.23)
Note that in both Eqs. 8.22 and 8.23, the modal vector { }r could be replaced by the modal
participation vector { L }r without loss of generality.
If the system does not satisfy reciprocity (the mass, stiffness, and/or damping matrices are not
symmetric), then the right and left eigenvectors will be different. For the right eigenvector, Eq.
8.24 will be appropriate.
2 [M ] + [C ] + [K ] { } = {0 }
r
r
(8.24)
For the left eigenvector, Eq. 8.25 will now be the appropriate form.
{L }r T r 2 [M ] + r [C ] + [K ] = {0 }
(8.25)
Note that the modal participation vector is the same as the left eigenvector. At this point it is
important to note that the residue for a nonreciprocal system can still be defined by the product
of the modal vector and the modal participation factor as in Eqs. 8.8 and 8.9. For the
nonreciprocal case, the relationship between the modal vector and modal participation vector, as
defined by Eq. 8.10, no longer is true.
This concept of the relationship between left and right eigenvectors and modal vectors and modal
participation vectors is true for systems that have repeated modal frequencies as well.
(8-19)
+UC-SDRL-RJA
CN-20-263-662
8.6 Summary/Conclusions
Whether a system contains repeated modal frequencies or nonreciprocal characteristics, the
general equations relating modal parameters to the measure frequency, or impulse, response
function matrices will properly predict the dynamics of the system as long as sufficient elements
of the matrix are measured. The extension of Eqs. 8.1 and 8.3 to Eqs. 8.12 and 8.13 provide the
generality and flexibility to describe these characteristics.
8.7 References
[1]
Leuridan, J., Some Direct Parameter Model Identification Methods Applicable for
Multiple Input Modal Analysis, Doctoral Dissertation, Department of Mechanical
Engineering, University of Cincinnati, 1984, 384 pp.
(8-20)
+UC-SDRL-RJA
CN-20-263-662
Appendix A: MEASUREMENT DISPLAY FORMATS

The following measurement display formats are all from the same theoretical system and
represent the information associated with either the impulse response function ( h12 (t) ) or the
frequency response function ( H 12 ( ) ). The theoretical system is a two degree of freedom
system with damping and the measurement represents the system characteristics measured by
forcing the system at the second degree of freedom and measuring the response at the first degree
of freedom.
x10 -4
3
Amplitude
-1
-2
-3
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Time (Sec.)
Figure A-1. Impulse Response Function Measurement Format
(A-1)
0.9
+UC-SDRL-RJA
CN-20-263-662
x10 -5
4
3
Amplitude (Real)
2
1
0
-1
-2
-3
-4
10
20
30
40
50
60
70
80
90
100
70
80
90
100
Frequency (Hertz)
x10 -5
8
6
Amplitude (Imag)
4
2
0
-2
-4
-6
-8
10
20
30
40
50
60
Frequency (Hertz)
Figure A-2. Frequency Response Function Measurement Format
(A-2)
+UC-SDRL-RJA
CN-20-263-662
x10 -5
7
Magnitude
0
0
10
20
30
40
50
60
70
80
90
100
70
80
90
100
Frequency (Hertz)
200
150
Phase (Degrees)
100
50
0
-50
-100
-150
-200
0
10
20
30
40
50
60
Frequency (Hertz)
(A-3)
+UC-SDRL-RJA
CN-20-263-662
-80
-90
Log Magnitude (dB)
-100
-110
-120
-130
-140
-150
-160
10
20
30
40
50
60
70
80
90
100
70
80
90
100
Frequency (Hertz)
200
150
Phase (Degrees)
100
50
0
-50
-100
-150
-200
0
10
20
30
40
50
60
Frequency (Hertz)
(A-4)
+UC-SDRL-RJA
CN-20-263-662
x10 -5
8
6
Amplitude (Imag)
4
2
0
-2
-4
-6
-8
Amplitude (Real)
(A-5)
+UC-SDRL-RJA
CN-20-263-662
Appendix B: MATLAB Script Files

The following script files have been used to generate the solutions and plots for the examples
shown in the text.
%
VIBS2 brings up a menu of the available exercises designed
%
for the Vibrations II course at the University of
%
Cincinnati.
%
%**********************************************************************
% Author: Randall J. Allemang
% Date:
18-Apr-94
% Structural Dynamics Research Lab
% University of Cincinnati
% Cincinnati, Ohio 45221-0072
% TEL: 513-556-2725
% FAX: 513-556-3390
% E-MAIL: [email protected]
%*********************************************************************
%
echo off
clear
while 1
vibs2= [v2_001
v2_002
v2_009
v2_011
v2_015
v2_016
v2_017
v2_020
v2_052
v2_055
v2_056
v2_057
v2_090
v2_095
v2_099];
clc
help v2list
n = input(Select a VIBS2 program number: );
if ((n <= 0) | (n > 17))
clear
break
end
vibs2 = vibs2(n,:);
eval(vibs2)
clear
end
clc
(B-1)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
%
CN-20-263-662
Vibrations 2 Scripts
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
v2_001
v2_002
v2_009
v2_011
v2_015
v2_016
v2_017
v2_020
v2_052
v2_055
v2_056
v2_057
v2_090
v2_095
v2_099
0)
EXIT
SDOF System
SDOF System, Under-Critical-Over Damped
SDOF System, Transfer Function (3-D)
SDOF System
Changing Stiffness, SDOF System
Changing Damping, SDOF System
Changing Mass, SDOF System
SDOF System
SDOF System
2DOF System-Proportional Damping
2DOF System-Non-Proportional Damping
2DOF System-Different Plot Formats
MDOF(3) System
MDOF(2-3) Addition of MKC Subsystem
MDOF(3) System, FRF by Inversion
(B-2)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
CN-20-263-662
v2_001.m
This is a script file to solve a sdof system
given the mass, damping and stiffness terms
in dimensionless units. The output includes
poles, residues (modal coefficients) and both
time and frequency domain plots of the impulse
frequency response functions.
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clg, clear
plt=input(Store plots to file (Yes=1): (0));if isempty(plt),plt=0;end;
mass=input(Mass value: (10));if isempty(mass), mass=10;end;
stiff=input(Stiffness value: (16000));if isempty(stiff),stiff=16000;end;
damp=input(Damping value: (10)); if isempty(damp),damp=10;end;
f=linspace(0,150,500);
H=1.0./((-mass.*f.*f+stiff)+j.*damp.*f);
plot(f,real(H))
pause
plot(f,imag(H))
pause
clear f,H;
pause
a=[mass,damp,stiff]
b=[0,0,1]
[r,p,k]=residue(b,a)
residue = r(1)
lambda=p(1)
whos
pause(2)
t=linspace(0,5,500);
xt=residue./2*exp(lambda.*t) + residue./2*exp(lambda.*t);
plot(t,xt)
xlabel(Time (Sec.)),ylabel(Amplitude),grid
if plt==1,print -deps v2_001a.eps, end;
pause
clear t,xt;
xf=residue./(j.*f-lambda) + residue./(j.*f-lambda);
plot(f,abs(xf))
xlabel(Frequency (Hz)),ylabel(Magnitude),grid
if plt==1,print -deps v2_001b.eps, end;
pause
semilogy(f,abs(xf))
xlabel(Frequency (Hz)),ylabel(Log Magnitude),grid
(B-3)
+UC-SDRL-RJA
CN-20-263-662
if plt==1,print -deps v2_001c.eps, end;

pause
plot(f,360./(2.*pi).*angle(xf))
xlabel(Frequency (Hz)),ylabel(Phase (Deg)),grid
if plt==1,print -deps v2_001d.eps, end;
pause
plot(f,360./(2.*pi).*angle(xf))
xlabel(Frequency (Hz)),ylabel(Phase (Deg)),grid
if plt==1,print -deps v2_001e.eps, end;
pause
plot(f,real(xf))
xlabel(Frequency (Hz)),ylabel(Real),grid
if plt==1,print -deps v2_001f.eps, end;
pause
plot(f,imag(xf))
xlabel(Frequency (Hz)),ylabel(Imaginary),grid
if plt==1,print -deps v2_001g.eps, end;
(B-4)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_002.m
This is a script file to illustrate the response
of an underdamped, critically damped and overdamped
SDOF system to an initial displacement.
SDOF System
Underdamped, critically damped overdamped response
Figures for UC-SDRL-CN-20-263-662, Chapter 2
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clear,clg
plt=input(Store plots to file (Yes=1): (0));if isempty(plt),plt=0;end
mag=10;
V=[-15,15,-15,15];
%
%
Overdamped Solution
%
lambda1=-2;
lambda2=-8;
t=linspace(0,4.0,201);
x=mag.*exp(lambda1.*t)+mag.*exp(lambda2.*t);
plot(t,x)
xlabel(Time(sec.)),ylabel(Displacement),grid
title(Overdamped SDOF System)
if plt==1,print -deps v2_002a.eps, end;
pause
clg
axis(V);
plot(real(lambda1),imag(lambda1),*,real(lambda2),imag(lambda2),*)
axis;
xlabel(Real Part of Complex Frequency)
ylabel(Imaginary Part of Complex Frequency),grid
title(Overdamped SDOF System)
if plt==1,print -deps v2_002b.eps, end;
pause
clear x
lambda1=-5;
lambda2=-5;
plot(t,x)
xlabel(Time(sec.)),ylabel(Magnitude),grid
title(Critically Damped SDOF System)
if plt==1,print -deps v2_002c.eps, end;
pause
(B-5)
+UC-SDRL-RJA
CN-20-263-662
clg
axis(V);
axis;
title(Critically Damped SDOF System)
if plt==1,print -deps v2_002d.eps, end;
pause
clg
clear x
lambda1=-5+j*10;
lambda2=-5-j*10;
plot(t,x)
xlabel(Time(sec.)),ylabel(Magnitude),grid
title(Underdamped SDOF System)
if plt==1,print -deps v2_002e.eps, end;
pause
clg
axis(V);
axis;
title(Underdamped SDOF System)
if plt==1,print -deps v2_002f.eps, end;
pause
(B-6)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_009.m
in dimensionless units. The output is a three
dimensional plot in the s domain (complex independent
variable.
SDOF System, Laplace Domain (3D) plot
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clear,clg
mass=10;
a=input(Real Part of Pole: (-0.0625)); if isempty(a),a=-0.0625;end
b=input(Imaginary Part of Pole: (0.51)); if isempty(b),b=0.51;end
rot=input(Rotation View Angle: (60)); if isempty(rot),rot=60;end
lambda(1)=a+j*b;
lambda(2)=a-j*b;
%
%
Set up mesh for only quadrants two and three
%
[sigma,omega]=meshdom(-0.2:0.005:0,-1:0.02:1);
s=sigma+j*omega;
H=(1.0./mass).*(1.0./((s-lambda(1)).*(s-lambda(2))));
view=[rot,30];
mesh(real(H),view)
%
title(Transfer Function (SDOF): Real Part)
if plt==1,print -f1 -deps v2_009a,end;
pause
mesh(imag(H),view)
%
title(Transfer Function (SDOF): Imaginary Part)
if plt==1,print -f1 -deps v2_009b,end;
pause
mesh(abs(H),view)
%
title(Transfer Function (SDOF): Magnitude)
if plt==1,print -f1 -deps v2_009c,end;
pause
mesh(angle(H),view)
%
title(Transfer Function (SDOF): Phase)
if plt==1,print -f1 -deps v2_009d,end;
pause
mesh(log(abs(H)),view)
%
title(Transfer Function (SDOF): Log Magnitude)
(B-7)
+UC-SDRL-RJA
CN-20-263-662
if plt==1,print -f1 -deps v2_009e,end;
(B-8)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_011.m
SDOF System
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clear,clg
mass=input(Mass value: (10)); if isempty(mass),mass=10;end
stiff=input(Stiffness value: (4)); if isempty(stiff),stiff=4;end
damp=input(Damping value: (2)); if isempty(damp),damp=2;end
mass,stiff,damp
a=[mass,damp,stiff];
b=[0,0,1];
[r,p,k]=residue(b,a);
residu=r(1);
lambda=p(1);
xt=residu./2*exp(lambda.*t) + residu./2*exp(lambda.*t);
axis([0,60,-0.1,0.1])
plot(t,xt)
xlabel(Time (Sec.)),ylabel (Amplitude),grid
pause
clg
clear t,xt;
xf=residu./(j.*f-lambda) + residu./(j.*f-lambda);
axis([0,3,-1.0,1.0])
plot(f,real(xf))
xlabel(Frequency (Rad/Sec)),ylabel(Real),grid
pause
clg
plot(f,imag(xf))
xlabel(Frequency (Rad/Sec)),ylabel(Imaginary),grid
axis([1,2,3,4]);axis;
(B-9)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_015.m
in dimensionless units when the stiffness is varied.
The output includes
poles, residues (modal coefficients) and
frequency domain plots of the
SDOF System, Change of Stiffness
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clg,clear
pi=3.14159;
mass=input(Mass value: (10));if isempty(mass),mass=10;end
istiff=input(Initial Stiffness value: (1000));if isempty(istiff),istiff=1000;end
damp=input(Damping value: (10));if isempty(damp),damp=10;end
hold off
for i=1:10
stiff=istiff + (i-1).*200
b=[0,0,1];
residu= r(1);
lambda=p(1);
semilogy(f,abs(xf))
hold on
end
xlabel(Frequency (Rad/Sec)),ylabel(Log Magnitude),grid
pause
hold off
clg;
for i=1:10
stiff=istiff + (i-1).*200
b=[0,0,1];
residu= r(1);
lambda=p(1);
(B-10)
+UC-SDRL-RJA
CN-20-263-662
scale=360.0/(2.0*pi);
plot(f,scale.*angle(xf))
hold on
end
xlabel(Frequency (Rad/Sec)),ylabel(Phase (Deg)),grid
hold off
(B-11)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_016.m
in dimensionless units when the damping is varied.
The output includes
SDOF System, Change of Damping
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clg,clear
pi=3.14159;
mass=input(Mass value: (10));if isempty(mass),mass=10;end
stiff=input(Stiffness value: (1600));if isempty(stiff),stiff=1600;end
idamp=input(Initial Damping value: (2));if isempty(idamp),idamp=2;end
hold off;
for i=1:10
damp=idamp + (i-1).*2
b=[0,0,1];
residu= r(1);
lambda=p(1);
semilogy(f,abs(xf))
hold on;
end
title(Effect of Changing Damping)
pause
hold off;
clg;
for i=1:10
damp=idamp + (i-1).*2
b=[0,0,1];
residu= r(1);
(B-12)
+UC-SDRL-RJA
CN-20-263-662
lambda=p(1);
scale=360.0/(2.0*pi);
hold on;
end
xlabel(Frequency (Rad/Sec)),ylabel(Phase (Deg)),grid
hold off;
(B-13)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_017.m
in dimensionless units when the mass is varied.
The output includes
SDOF System, Change of Mass
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clg,clear
pi=3.14159;
imass=input(Initial Mass value: (2));if isempty(imass),imass=2;end
stiff=input(Stiffness value: (1600));if isempty(stiff),stiff=1600;end
damp=input(Damping value: (10));if isempty(damp),damp=10;end
hold off
for i=1:10
mass=2 + (i-1).*1
b=[0,0,1];
residu= r(1);
lambda=p(1);
semilogy(f,abs(xf))
hold on
end
pause
hold off
for i=1:10
mass=2 + (i-1).*1
b=[0,0,1];
residu= r(1);
lambda=p(1);
(B-14)
+UC-SDRL-RJA
CN-20-263-662
scale=360.0/(2.0*pi);
hold on
end
hold off
(B-15)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_020.m
in dimensionless units. The output is a three
dimensional plot in the s domain (complex independent
variable.
SDOF System, Laplace Domain (3D) plot
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clear,clg
mass=10;
a=input(Real Part of Pole: (-0.0625)); if isempty(a),a=-0.0625;end
b=input(Imaginary Part of Pole: (0.51)); if isempty(b),b=0.51;end
rot=input(Rotation View Angle: (60)); if isempty(rot),rot=60;end
lambda(1)=a+j*b;
lambda(2)=a-j*b;
%
%
Set up mesh for only quadrants two and three
%
[sigma,omega]=meshdom(-0.2:0.005:0,-1:0.02:1);
s=sigma+j*omega;
H=(1.0./mass).*(1.0./((s-lambda(1)).*(s-lambda(2))));
view=[rot,30];
mesh(real(H),view)
%
title(Transfer Function (SDOF): Real Part)
pause
mesh(imag(H),view)
%
title(Transfer Function (SDOF): Imaginary Part)
pause
mesh(abs(H),view)
%
title(Transfer Function (SDOF): Magnitude)
pause
mesh(angle(H),view)
%
title(Transfer Function (SDOF): Phase)
pause
mesh(log(abs(H)),view)
%
title(Transfer Function (SDOF): Log Magnitude)
(B-16)
+UC-SDRL-RJA
CN-20-263-662
(B-17)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_052.m
SDOF Damped System
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clear,clg
mass=input(Mass value: (10)); if isempty(mass), mass=10;end;
stiff=input(Stiffness value: (4)); if isempty(stiff),stiff=4;end;
damp=input(Damping value: (2)); if isempty(damp),damp=2;end;
b=[0,0,1];
residu = r(1)
lambda=p(1)
xt=residu./2*exp(lambda.*t) + residu./2*exp(lambda.*t);
axis([0,60,-0.1,0.1])
plot(t,xt)
pause
clg; clear t; clear xt;
axis([0,3,-1.0,1.0])
plot(f,real(xf))
pause
clg
plot(f,imag(xf))
axis([1,2,3,4]);axis;
(B-18)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_055.m
This is a script file to solve a 2 DOF system
given the mass, damping and stiffness matrices
in dimensionless units. The output includes poles,
residues(modal coefficients) and time and frequency
domain plots of impulse and frequency response
functions.
2 DOF Proportionally Damped System
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clear,clg
pi=3.14159265;
alpha=0.0
beta=0.5
mass=[5,0;0,10];
stiff=[4,-2;-2,6];
damp=alpha*mass + beta*stiff;
null=[0,0;0,0];
%
Form 2N x 2N state space equation.
a=[null,mass;mass,damp];
b=[-mass,null;null,stiff];
[x,d]=eig(-inv(a)*b);
% Sort Modal Frequencies
orig_lambda=diag(d);
[Y,I]=sort(imag(orig_lambda));
lambda=orig_lambda(I);
xx=x(:,I);
%
Normalize x matrix
for ii=1:4
xx(1:4,ii)=xx(1:4,ii)./xx(3,ii);
end
%
Extract modal vectors from state-space formulation
psi(1:2,1)=xx(3:4,1);
psi(1:2,2)=xx(3:4,2);
%
Calculate modal mass matrix
mm=psi.*mass*psi;
%
Calculate modal scaling value (Q)
Q(1)=1./(2*j*imag(lambda(1))*mm(1,1));
Q(2)=1./(2*j*imag(lambda(2))*mm(2,2));
%
Calculate residue matrices
A1=Q(1).*psi(1:2,1)*psi(1:2,1).;
(B-19)
+UC-SDRL-RJA
CN-20-263-662
A2=Q(2).*psi(1:2,2)*psi(1:2,2).;
%
Formulate H(1,2) FRF as Default
resp=input(Enter response (row) DOF: (1));if isempty(resp),resp=1;end
inp=input(Enter input (column) DOF: (2));if isempty(inp),inp=2;end
residu(1) = A1(resp,inp);
A1,A2
pause
magA1=abs(A1);
phaseA1=angle(A1).*360.0./(2.0*pi);
magA2=abs(A2);
phaseA2=angle(A2).*360.0./(2.0*pi);
magA1,phaseA1
magA2,phaseA2
pause
lambda,residu
pause
%
Calculate desired plots
xt1=residu(1).*exp(lambda(1).*t) + residu(1).*exp(lambda(1).*t);
xt=xt1+xt2;
axis([0,30,-0.10,0.10])
plot(t,xt)
pause
clg
plot(t,xt1)
if plt==1,print -f1 -deps v2_055f,end;
pause
clg
plot(t,xt2)
if plt==1,print -f1 -deps v2_055i,end;
pause
clg
clear t; clear xt; clear xt1; clear xt2;
f=linspace(0,2.5,500);
xf1=residu(1)./(j.*f-lambda(1)) + residu(1)./(j.*f-lambda(1));
xf=xf1+xf2;
axis([0,2.5,-0.6,0.6])
plot(f,real(xf))
pause
clg
plot(f,imag(xf))
pause
clg
plot(f,real(xf1))
(B-20)
+UC-SDRL-RJA
CN-20-263-662
pause
clg
plot(f,imag(xf1))
pause
clg
axis([0,2.5,-0.2,0.2])
plot(f,real(xf2))
if plt==1,print -f1 -deps v2_055g,end;
pause
clg
plot(f,imag(xf2))
if plt==1,print -f1 -deps v2_055h,end;
axis([1 2 3 4]);axis;
(B-21)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_056.m
in dimensionless units. The output includes poles,
residues(modal coefficients) and time and frequency
domain plots of impulse and frequency response
functions.
2 DOF Non-Proportionally Damped System
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clear,clg
pi=3.14159265;
mass=[5,0;0,10];
stiff=[4,-2;-2,6];
damp=[6,-4;-4,5];
null=[0,0;0,0];
%
[x,d]=eig(-inv(a)*b);
xx=x(:,I);
xxx=input(Hit any key to continue);
%
Compute modal a and modal b matrix
ma=xx.*a*xx;
mb=xx.*b*xx;
%
psi(1:2,1)=xx(3:4,1);
psi(1:2,2)=xx(3:4,2);
psi(1:2,3)=xx(3:4,3);
psi(1:2,4)=xx(3:4,4);
%
A1=psi(1:2,1)*psi(1:2,1)../ma(1,1);
A2=psi(1:2,2)*psi(1:2,2)../ma(2,2);
A3=psi(1:2,3)*psi(1:2,3)../ma(3,3);
A4=psi(1:2,4)*psi(1:2,4)../ma(4,4);
%
Formulate H(1,2) FRF as Default
resp=input(Enter response (row) DOF: (1));if isempty(resp),resp=1;end;
(B-22)
+UC-SDRL-RJA
CN-20-263-662
inp=input(Enter input (column) DOF: (2));if isempty(inp),inp=2;end;

A1,A2,A3,A4
lambda,residu
%
xt=xt1+xt2;
axis([0,30,-0.10,0.10])
plot(t,xt)
pause
clg
plot(t,xt1)
if plt==1,print -f1 -deps v2_056i,end;
pause
clg
plot(t,xt2)
if plt==1,print -f1 -deps v2_056f,end;
pause
clg
clear t; clear xt; clear xt1; clear xt2;
f=linspace(0,2.5,500);
xf=xf1+xf2;
axis([0,2.5,-0.6,0.6])
plot(f,real(xf))
pause
clg
plot(f,imag(xf))
pause
clg
plot(f,real(xf1))
if plt==1,print -f1 -deps v2_056g,end;
pause
clg
plot(f,imag(xf1))
if plt==1,print -f1 -deps v2_056h,end;
pause
(B-23)
+UC-SDRL-RJA
CN-20-263-662
clg
axis([0,2.5,-0.6,0.6])
plot(f,real(xf2))
pause
clg
plot(f,imag(xf2))
axis([1 2 3 4]);axis;
(B-24)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_057.m
in dimensionless units and plot any desired
frequency response function.
Plot Formats for FRF and IRF Measurements
Figures for UC-SDRL-CN-20-263-662, Appendix A
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clear,clg
pi=3.14159265;
axis(normal);
axis([1 2 3 4]),axis;
pi=3.14159265;
alpha=5;
mass=[8,0;0,10];
stiff=[600000,-200000;-200000,200000];
damp=alpha.*mass;
null=[0,0;0,0];
%
[x,d]=eig(b,-a);
xx=x(:,I);
xx,lambda
%
ma=xx.*a*xx;
mb=xx.*b*xx;
%
psi(1:2,1)=xx(3:4,1);
psi(1:2,2)=xx(3:4,2);
psi(1:2,3)=xx(3:4,3);
psi(1:2,4)=xx(3:4,4);
%
A1=psi(1:2,1)*psi(1:2,1)../ma(1,1);
A2=psi(1:2,2)*psi(1:2,2)../ma(2,2);
A3=psi(1:2,3)*psi(1:2,3)../ma(3,3);
(B-25)
+UC-SDRL-RJA
CN-20-263-662
A4=psi(1:2,4)*psi(1:2,4)../ma(4,4);
lambda,residu
%
xt1=residu(1)./2*exp(lambda(1).*t);
xt=xt1+xt2+xt3+xt4;
plot(t,xt)
%
Reset plot axis for same + and - limits
V=axis;
ymax=max([abs(V(3)),abs(V(4))]);V(3)=-ymax;V(4)=+ymax;axis(V);
%
plot(t,xt)
if plt==1,meta v2_057a,end;
pause
axis([1 2 3 4]),axis;
clear t; clear xt;
om=f.*2.*pi;
H1=residu(1)./(j.*om-lambda(1));
H=H1+H2+H3+H4;
plot(f,real(H))
V=axis;
plot(f,real(H))
xlabel(Frequency (Hertz)),ylabel(Amplitude (Real)),grid
if plt==1,meta v2_057b,end;
pause
axis([1 2 3 4]),axis;
plot(f,imag(H))
V=axis;
plot(f,imag(H))
xlabel(Frequency (Hertz)),ylabel(Amplitude (Imag)),grid
if plt==1,meta v2_057c,end;
pause
axis([1 2 3 4]),axis;
plot(f,abs(H))
xlabel(Frequency (Hertz)),ylabel(Magnitude),grid
if plt==1,meta v2_057d,end;
pause
(B-26)
+UC-SDRL-RJA
CN-20-263-662
axis([1 2 3 4]),axis;
plot(f,20.*log10(abs(H)))
xlabel(Frequency (Hertz)),ylabel(Log Magnitude (dB)),grid
if plt==1,meta v2_057e,end;
pause
plot(f,360./(2.*pi).*angle(H))
V=axis;
plot(f,360./(2.*pi).*angle(H))
xlabel(Frequency (Hertz)),ylabel(Phase (Degrees)),grid
if plt==1,meta v2_057f,end;
pause
axis([1 2 3 4]),axis;
axis(square)
polar(angle(H),abs(H))
xlabel(Amplitude (Real)),ylabel(Amplitude (Imag)),grid
if plt==1,meta v2_057g,end;
(B-27)
+UC-SDRL-RJA
%
%
%
%
%
%
CN-20-263-662
v2_090.m
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clear,clg
pi=3.14159265;
mass=[10,0,0;0,14,0;0,0,12];
stiff=[5000,-3000,0;-3000,5500,-2500;0,-2500,2500];
damp=[50,-30,0;-30,55,-25;0,-25,25];
null=[0,0,0;0,0,0;0,0,0];
%
[x,d]=eig(b,-a);
xx=x(:,I);
%
Normalize x matrix to real vectors if possible
for ii=1:6
xx(1:6,ii)=xx(1:6,ii)./xx(4,ii);
end
%
ma=xx.*a*xx;
mb=xx.*b*xx;
%
psi(1:3,1)=xx(4:6,1);
psi(1:3,2)=xx(4:6,2);
psi(1:3,3)=xx(4:6,3);
psi(1:3,4)=xx(4:6,4);
psi(1:3,5)=xx(4:6,5);
psi(1:3,6)=xx(4:6,6);
lambda
psi
ma
%
(B-28)
+UC-SDRL-RJA
CN-20-263-662
A1=psi(1:3,1)*psi(1:3,1)../ma(1,1);
A2=psi(1:3,2)*psi(1:3,2)../ma(2,2);
A3=psi(1:3,3)*psi(1:3,3)../ma(3,3);
A4=psi(1:3,4)*psi(1:3,4)../ma(4,4);
A5=psi(1:3,5)*psi(1:3,5)../ma(5,5);
A6=psi(1:3,6)*psi(1:3,6)../ma(6,6);
%
om=linspace(0,50,501);
H=H1+H2+H3+H4+H5+H6;
om(151),H(151)
om(251),H(251)
subplot(211),semilogy(om,abs(H),om(151),abs(H(151)),*,om(251),abs(H(251)),*)
xlabel(Frequency (Rad/Sec)),ylabel(Magnitude),grid
subplot(212),plot(om,angle(H),om(151),angle(H(151)),*,om(251),angle(H(251)),*)
xlabel(Frequency (Rad/Sec)),ylabel(Phase),grid
title([FRF
Input: ,num2str(inp),
Response: ,num2str(resp)])
pause
(B-29)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
%
%
%
CN-20-263-662
v2_095.m
frequency response function. A third DOF is
then added to act as a spring-mass-damper to
reduce the magnitude of the first mode of the
original system. The 3 DOF system is then
solved and any frequency response functions
can be plotted.
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clear, clg
pi=3.14159265;
alpha=0.001
mass=[8,0;0,10];
stiff=[6000,-2000;-2000,2000];
damp=alpha.*stiff;
null=[0,0;0,0];
%
[x,d]=eig(b,-a);
xx=x(:,I);
%
for ii=1:4
xx(1:4,ii)=xx(1:4,ii)./xx(3,ii);
end
%
ma=xx.*a*xx;
mb=xx.*b*xx;
%
psi(1:2,1)=xx(3:4,1);
psi(1:2,2)=xx(3:4,2);
psi(1:2,3)=xx(3:4,3);
psi(1:2,4)=xx(3:4,4);
lambda
psi
(B-30)
+UC-SDRL-RJA
CN-20-263-662

ma
%
A1=psi(1:2,1)*psi(1:2,1)../ma(1,1);
A2=psi(1:2,2)*psi(1:2,2)../ma(2,2);
A3=psi(1:2,3)*psi(1:2,3)../ma(3,3);
A4=psi(1:2,4)*psi(1:2,4)../ma(4,4);
A1,A2,A3,A4
lambda,residu
%
HH=H1+H2+H3+H4;
fig1=figure(1);
subplot(211),semilogy(om,abs(HH))
title(Frequency Response Function)
subplot(212),plot(om,angle(HH))
pause
%
% now solve 3 dof system (with spring-mass-damper added)
%
pi=3.14159265;
alpha=0.001
mass=[8,0,0;0,10,0;0,0,2.50];
stiff=[6000,-2000,0;-2000,2300,-300;0,-300,300];
damp=alpha.*stiff;
null=[0,0,0;0,0,0;0,0,0];
%
[x,d]=eig(b,-a);
x,d
%
Pick frequency roots
lambda=diag(d);
%
for ii=1:6
x(1:6,ii)=x(1:6,ii)./x(4,ii);
end
%
(B-31)
+UC-SDRL-RJA
CN-20-263-662
ma=x.*a*x;
mb=x.*b*x;
%
psi(1:3,1)=x(4:6,1);
psi(1:3,2)=x(4:6,2);
psi(1:3,3)=x(4:6,3);
psi(1:3,4)=x(4:6,4);
psi(1:3,5)=x(4:6,5);
psi(1:3,6)=x(4:6,6);
%
A1=psi(1:3,1)*psi(1:3,1)../ma(1,1);
A2=psi(1:3,2)*psi(1:3,2)../ma(2,2);
A3=psi(1:3,3)*psi(1:3,3)../ma(3,3);
A4=psi(1:3,4)*psi(1:3,4)../ma(4,4);
A5=psi(1:3,5)*psi(1:3,5)../ma(5,5);
A6=psi(1:3,6)*psi(1:3,6)../ma(6,6);
A1,A2,A3,A4,A5,A6
lambda,residu
%
H=H1+H2+H3+H4+H5+H6;
fig2=figure(2);
subplot(211),semilogy(om,abs(H),om,abs(HH))
subplot(212),plot(om,angle(H),om,angle(HH))
pause
% Now compute final H22 using impedance method
M3=2.5;
C3=.3;
K3=300;
H33=-M3.*om.*om+j.*C3.*om+K3;
H33=H33./((-M3.*om.*om).*(K3+j.*om.*C3));
HHH=HH.*H33./(HH+H33);
fig3=figure(3);
subplot(211),semilogy(om,abs(H),om,abs(HHH))
subplot(212),plot(om,angle(H),om,angle(HHH))
(B-32)
+UC-SDRL-RJA
CN-20-263-662
pause
(B-33)
+UC-SDRL-RJA
%
%
%
%
%
%
%
%
CN-20-263-662
v2_099.m
Solution for FRF by Inverse of [B]
%**********************************************************************
% Date:
18-Apr-94
% TEL: 513-556-2725
% FAX: 513-556-3390
%*********************************************************************
%
clear,clf,close all;
pi=3.14159265;
%
% solve 3 dof system
%
mass=[10,0,0;0,14,0;0,0,12];
stiff=[5000,-3000,0;-3000,5500,-2500;0,-2500,2500];
damp=[50,-30,0;-30,55,-25;0,-25,25];
null=[0,0,0;0,0,0;0,0,0];
%
[x,d]=eig(b,-a);
xx=x(:,I);
%
for ii=1:6
xx(1:6,ii)=xx(1:6,ii)./xx(4,ii);
end
%
ma=xx.*a*xx;
mb=xx.*b*xx;
%
psi(1:3,1)=xx(4:6,1);
psi(1:3,2)=xx(4:6,2);
psi(1:3,3)=xx(4:6,3);
psi(1:3,4)=xx(4:6,4);
psi(1:3,5)=xx(4:6,5);
psi(1:3,6)=xx(4:6,6);
lambda
(B-34)
+UC-SDRL-RJA
CN-20-263-662
psi
ma
clear a,clear b,clear x,clear d;
omega=linspace(0,100,501);
for ii=1:501;
FRF(:,:,ii)=inv(-mass.*omega(ii).*omega(ii) + damp.*j.*omega(ii) + stiff);
end;
for Ni=1:3;
for No=1:3;
H=squeeze(FRF(No,Ni,:));
figure;
subplot(211),semilogy(omega,abs(H))
scale=360.0/(2.0*pi);
subplot(212),plot(omega,scale.*angle(H))
title(Frequency Response Function: H(x,x));
pause
end
end
pause
(B-35)

Vibrations - Analytical and Experimental Modal Analysis

Uploaded by

Copyright:

Available Formats

Vibrations - Analytical and Experimental Modal Analysis

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Vibrations - Analytical and Experimental Modal Analysis

Uploaded by

Copyright:

Available Formats

+UC-SDRL-RJA

Dr. Randall J. Allemang, Professor

Approved for public release

Revision: February 19, 1999 +

Revision: February 19, 1999 +

Revision: February 19, 1999 +

braces enclose column vector expressions

row vector expressions

inverse Fourier transform

inverse Hilbert transform

Revision: February 19, 1999 +

partial derivative with respect to independent variable "t"

residue for response location p, reference location q, of mode r

Revision: February 19, 1999 +

Revision: February 19, 1999 +

Dirac impulse function

r th complex eigenvalue, or system pole

diagonal matrix of poles in Laplace domain

noise on the input

undamped natural frequency (rad/sec)

scaled pth response of normal modal vector for mode r

3. MULTIPLE DEGREE OF FREEDOM SYSTEMS ............................................................

4. FREQUENCY RESPONSE FUNCTION DEVELOPMENT ............................................. 1

5. GENERAL DAMPED SYSTEMS ....................................................................................

6. FREQUENCY RESPONSE FUNCTION SYNTHESIS ..................................................... 1

8. ADVANCED MODAL ANALYSIS CONCEPTS ...............................................................

Appendix A: MEASUREMENT DISPLAY FORMATS

Revision: February 19, 1999 +

Revision: February 19, 1999 +

Modal Frequencies, Eigenvalues, System Poles, Characteristic Roots ( r )

Modal Vectors, Eigenvectors ( { r }, { r } )

Modal Coefficients ( pr ), Residues ( A pqr )

Real Modal Vectors, Complex Modal Vectors

Physical Coordinates, Generalized Coordinates

Principal Coordinates, Modal coordinates

Modal Mass, Damping, and Stiffness ( M r , C r , K r )

1.1 Degrees of Freedom

Revision: February 19, 1999 +

Figure 1-1. Degrees of Freedom of a Rigid Body

Revision: February 19, 1999 +

Revision: February 19, 1999 +

1.2 Basic Assumptions

measured by exciting at q and measuring the response at p H pq = H qp .

Revision: February 19, 1999 +

2. SINGLE DEGREE OF FREEDOM SYSTEM

and is represented schematically in Figure (2-1).

Figure 2-1. Single Degree of Freedom System

Revision: February 19, 1999 +

x(t) = x c (t) + x p (t)

x c (t) = Transient portion

x p (t) = Steady state portion

Thus, for a non-trivial solution X(s) e s t 0 :

Thus the homogeneous solution of Equation 2.1 is:

Revision: February 19, 1999 +

x(0) is the initial displacement at time t = 0.

If the initial conditions are zero, Equation 2.4 becomes:

Revision: February 19, 1999 +

assuming initial conditions are zero.

Revision: February 19, 1999 +

2.3 Definition Of Terms