Mass, Stiffness, and Damping Matrices From Measured Modal Parameters
Mass, Stiffness, and Damping Matrices From Measured Modal Parameters
Mass, Stiffness, and Damping Matrices From Measured Modal Parameters
Page - 1
MASS, STIFFNESS, AND DAMPING MATRICES
FROM MEASURED MODAL PARAMETERS
Ron Potter
and
Mark Richardson
Digital Signal Analysis
HEWLETT-PACKARD COMPANY
Santa Clara, California
ABSTRACT
The theory of complex mode shapes for damped oscillatory
mechanical systems is explained, using the matrix of
transfer functions in the Laplace domain. These mode
shapes are defined to be the solutions to the homogeneous
system equation. It is shown that a complete transfer
matrix can be constructed once one row or column of it has
been measured, and hence that mass, stiffness, and
damping matrices corresponding to a lumped equivalent
model of the tested structure can also be obtained from the
measured data.
INTRODUCTION
In recent years, there has been considerable activity in the
study of elastic structure dynamics,. in an attempt to design
structures that will function properly in a hostile vibration
environment. Although much of the early work centered
around fatigue and life testing, the latest efforts have been
directed towards analytical modeling and simulation of
mechanical structures. Distributed structures are generally
modeled as networks of lumped mechanical elements, in an
effort to predict failures more reliably and faster than is
afforded by conventional life testing procedures.
With the advent of the inexpensive mini-computer, and
computing techniques such as the Fast Fourier Transform
algorithm, it is now relatively easy to obtain fast, accurate,
and complete measurements of the behavior of mechanical
structures in various vibration environments.
Modal responses of many modes can be measured
simultaneously and complex mode shapes can be directly
identified instead of relying upon and being constrained by
the so called "normal mode" concept. Furthermore, the
entire system response matrix, which comprises the mass,
stiffness, and damping matrices of the lumped equivalent
model, can be measured.
The following material covers the theoretical background
that is needed to understand these new measurement
techniques.
COMPLEX MODES AND THE TRANSFER MATRIX
Let's assume that the motion of a linear physical system can
be described by a set of n simultaneous second order linear
differential equations in the time domain, given by
Mx Cx Kx f && & + + (1)
where the dots denote differentiation with respect to time. f
= f(t) is the applied force vector, and x = x(t) is the
resulting displacement vector, while M, C, and K are the (n
by n) mass, damping, and stiffness matrices respectively.
In this discussion, our attention will be limited to
symmetric matrices, and to real element values in M, C,
and K.
Taking the Laplace transform of the system equations gives
B s X s F s ( ) ( ) ( ) , where (2)
B s Ms Cs K ( ) + +
2
(3)
Here, s is the Laplace variable, and now F(s) is the applied
force vector and X(s) is the resulting displacement vector in
the Laplace domain. B(s) is called the system matrix, and
the transfer matrix H(s) is defined as
H s B s
1
( ) ( )
(4)
which implies that
H s F s X s ( ) ( ) ( ) (5)
Each element of the transfer matrix is a transfer function.
The elements of B are quadratic functions of s, and since
H B
1
, it follows that the elements of H are rational
fractions in s, with det(B) as the denominator. Thus, H(s)
can always be represented in partial fraction form.
If it is assumed that the poles of H, i.e. the roots of det(B) =
0, are of unit multiplicity, then H can be expressed as
H
a
s s
k
k
2n
k 1
(n by n) (6)
ISA 74 International Instrumentation-Automation Conference & Exhibit October, 1974
Page - 2
The poles occur at s s
k
(zeros of det B), and each pole
has an (n x n) residue matrix a
k
describing its spatial
behavior. For an n
th
order oscillatory system, there will
always be 2n poles, but they will appear in complex
conjugate pairs. The poles are complex numbers expressed
as
s
k
+
k k
i (7)
where
k
is the damping coefficient (a negative number for
stable systems), and
k
is the natural frequency of
oscillation. The resonant frequency is given by
rk k k
+
2 2
rad/sec (8)
and the damping factor is
k
rk
(9)
In reference [1], modal vectors were derived in terms of the
eigenvectors of the system matrix B. However, these
eigenvectors were only introduced as an intermediary in the
determination of modal vectors. Here, the modal vectors
are described in terms of the B matrix. Pre-multiplying B
times the expression for H, multiplying by the scalar
( ) s s
k
, and letting s s
k
gives
B
k
a
k
0, where B
k
B s
k
( ) (10)
Similarly, post-multiplication of H by B gives
a
k
B
k
0 (11)
Thus, all rows and columns of a
k
must comprise linear
combination of homogeneous solution vectors u
k
, given by
B
k
u
k
0 (12)
These u
k
homogeneous solution vectors are defined as
modal (or mode shape) vectors associated with the pole at
s s
k
.
Restricting our attention to the special case where only one
u
k
modal vector exists for each pole, it is clear that all
rows and columns of a
k
must be some scalar multiple of
u
k
. Thus, a
k
can be represented by
a
k
A u u
k k k
t
(n by n) (13)
where A
k
is a scalar. In these terms, H can be rewritten
as
H
A
s s
k
k k 1
2n
_
,
u u
k k
t
(14)
and this is easily written in matrix form as
H
1
t
(n by n) (15)
where the columns of comprise the u
k
modal vectors:
1
]
1
1
1
| | |
| | |
u u u
1 2 2n
(n by 2n) (16)
and
1
is a diagonal matrix containing all s dependence:
1
]
1
1
1
1
1
1
1
1
1
A
s s
0
0
A
s s
1
1
2n
2n
,(2n x 2n) (17)
Pre-multiplying H by
t
, equation (5) can be written as
( )( ) ( )
t
1 t t
F X (18)
so that
t
transforms the spatial vectors (F and X) to
vectors
t
F and
t
X in modal coordinates. Similarly
( )
t 1
' is the modal representation of H.
Since B( ) s u
k k
0 it follows that B( ) s u
k k
0 , so
the modal vector associated with the conjugate pole ( s
k
)
is ( u
k
), (the conjugate of u
k
). Thus, the above
matrix always contains conjugate pairs of modal vectors
and the
1
matrix always contains elements
corresponding to conjugate pole pairs along its diagonal. If
poles are purely imaginary (zero damping) then all modal
vectors are real, so only half of is needed. Otherwise
is rectangular (n x 2n) and consequently, even though H
can be written in diagonal form using the matrix, its
inverse B cannot be diagonalized using except in the
special case when damping is zero.
ISA 74 International Instrumentation-Automation Conference & Exhibit October, 1974
Page - 3
IDENTIFICATION OF MODAL PARAMETERS
Because of the form of the a
k
matrix, only one row or
column of the transfer matrix need be measured and
analyzed, since all other rows and columns contain
redundant information. In the process of measuring the
transfer matrix, unknown parameters in equation (14), i.e.
the complex values of s
k
and the complex values of the
elements of one row or column of a
k
are identified.
For example, the q
th
column of a
k
is given by
a
kq
A u u A u u
k k kq k kq k
(19)
where u
kq
is the q
th
element of u
k
. Thus the modal
vector u
k
(whose normalization is arbitrary) can be
recovered once the q
th
column of a
k
is identified. In
addition the complex scalar A
k
can be recovered using the
formula
A
k
a a
u u u
kq
t
kq
kq k
t
k
(20)
It is clear that the numerical value of A
k
depends upon
the normalization of the modal vector. If we choose
u
k
t
k
u 1, then
A
k
1
u
a a
kq
kq
t
kq
(21)
Since complex modal vectors appear in conjugate pairs, H
can always be written in two parts as
H
A
s s
k
k k 1
n
1
]
1
u u
A
s s
u u
k k
t k
k
k k
t
(22)
For the k
th
pair of conjugate poles,
H
A
s s
k
k
k k k
t k
k
k k
t
u u
A
s s
u u
(23)
Each pole of the transfer matrix has a corresponding mode
shape vector and, furthermore, each complex conjugate pair
of poles has a corresponding complex conjugate pair of
mode shapes. The pq
th
element of H
k
is
H
A
s s
k
k
kpq kp kq
k
k
kp kq
u u
A
s s
u u
(24)
The time domain displacement at point p due to an
impulsive force at point q is given by the inverse Laplace
transform of H
kpq
which is
h
kpq
( ) t A u u e A u u e
k kp kq
s t
k kp kq
s t
k k
+
( )
[
2e
k
t
k kp kq k
A u u t Re cos( )
( ) ] Im A
k
u u t
kp kq k
sin( )
+ 2e
k
t
k kp kq k k
A u u t cos( ) (25)
for t 0.
Note that the peak amplitude of the impulse response is
A u u
k kp kq
and the phase angle ( )
k
is the angle of the
complex residue A u u
k kp kq
with respect a cosine. The
magnitude and phase of this complex residue are different,
in general, for each spatial point on the structure.
MASS, STIFFNESS, AND DAMPING MATRICES
Recall that B s Ms Cs K ( ) + +
2
, so it is apparent that
K B H
( ) ( ) 0 0
1
(26)
Now, H(0) is obtained by setting s = 0 in
1
.
A modal compliance
1
can be defined as
1
]
1
1
1
1
1
1
1
1
1 1
1
1
2
2
0
0
0
( )
A
s
A
s
n
n
(2n x 2n ) (27)
So
( ) K H 0
1
( )
1
1
t
(28)
ISA 74 International Instrumentation-Automation Conference & Exhibit October, 1974
Page - 4
Thus, the stiffness matrix is readily obtained from the
measured modal vectors (columns of ), and the identified
A
k
and s
k
complex scalars. Since
HB I (29)
it follows that
HB H B 0 ' ' + (30)
and
HB H B H B '' ' ' '' + + 2 0 (31)
where the prime denotes differentiation with respect to s.
The damping matrix can be computed with the following
relationship
C B B H B KH K '( ) ( ) '( ) ( ) '( ) 0 0 0 0 0
(32)
Or alternatively by defining the modal damping matrix
as
1
]
1
1
1
1
1
1
1
1
1
0
0
1
1
2
A
A
n
(2n x 2n) (33)
then
( ) H
t t
'( ) '( ) 0 0
1 1 1
(34)
and
( ) ( ) C K K
t
1 1
(35)
In a similar manner, we can write
M
1
2
B''( ) 0
B H B B H B ( ) '( ) '( ) ( ) ''( ) ( ) 0 0 0
1
2
0 0 0
KH C KH K '( ) ''( ) 0
1
2
0 (36)
But H"(0) can be written as
H
t
''( ) 0 2
1 1 1
(37)
Finally, the mass matrix can be obtained from
( )( )( ) M K K
t
1 1 1
( ) +
K C
t
1 1
( )( )( ) +
K K CK C
t
1 1 1 1
(38)
Now, from the expression for K, we obtain
( ) K
t t
1 1
1
1
0
(39)
where
t
is the left-handed inverse of
t
defined by
t t
I (40)
To summarize, the stiffness, damping, and mass matrices
are obtained from the expressions
( ) K
t t
1
1
0 0
1
(41)
C
t
0 0
1
(42)
M CK C
t
+
0 0
1 1
(43)
where
1
(diagonal) (44)
All of these quantities are readily obtained from the
measured poles s
k
, scalars A
k
, and modal vectors u
k
.
COMMENTARY
The fundamental nature of complex modes in the real
world of measurement cannot be over-emphasized.
Damping is always present in a structure, and it can always
be observed in measured transfer function data. The fact
that a complex mode implies a complex time waveform
should be of little concern, because the conjugate waveform
is always present to make the observed signal real valued.
A complex mode has the character of a "traveling wave"
across the structure (as opposed to the usual "standing
wave" produced by a normal mode) as indicated by the
changing phase angle of the displacement from point to
point.
Furthermore, it is important to note that a mode shape still
a global (as opposed to local) property of a structure , even
though damping may be heavy. The fact that local motion
near the point of excitation in a heavily damped structure
ISA 74 International Instrumentation-Automation Conference & Exhibit October, 1974
Page - 5
tends to dominate is generally caused by many closely
spaced modes that are excited in phase at this point, but
tend to cancel each other elsewhere.
Historically, considerable emphasis has been placed on
simultaneously diagonalizing the mass and stiffness
matrices, so that the displacement of a particular point
could be easily calculated from an arbitrary excitation force.
Unfortunately, it is not possible to simultaneously
diagonalize more than two symmetric matrices, so this
technique cannot be used when damping is present.
However, when a system is represented by the partial
fraction form of its transfer matrix, a closed form solution
for the displacement at any point, for any combination of
modes, is readily obtainable using simple matrix-vector
multiplication. This is particularly helpful when the
response to only a few modes is of interest.
The definition of a modal vector as a solution to the
homogeneous system equation is also very fundamental,
and removes much of the ambiguity about what modal
vectors really are, and when they exist. This definition also
makes them relatively easy to calculate or measure.
It should be apparent from the equations for obtaining
mass, damping, and stiffness from measured data, that the
order of these matrices is equal to the number of complex
modal pairs that are measured. This means that if n modes
are identified, only n spatial points are needed to represent
the lumped equivalent model of the physical system. This
result can be stated as the following theorem.
SPATIAL SAMPLING THEOREM: A lumped system
modeled with second order elements has exactly 2n poles
an 2n modal vectors. Thus, if only n modal pairs are
found, there is no need for more than n spatial points in
the lumped model.
Finally, the scope of this discussion has been limited to
symmetric matrices having distinct poles (unit multiplicity)
in complex conjugate pairs. Even though these
assumptions are satisfied by a large majority of the linear
systems tested, modes of vibration can still be defined when
all of these assumptions are relaxed, and we have worked
out the theory for this general case.
SUMMARY
We have defined the complex modal (mode shape) vector
associated with each pole (zero of the system determinant)
of an elastic mechanical system as the homogeneous
solution to the system equations. We have also emphasized
that these poles occur in complex conjugate pairs, and that
the two associated modal vectors are also conjugates of one
another.
The transfer matrix (inverse of system matrix) was
introduced, and it was written in partial fraction form,
comprising one term for each pole. We found that the rows
and columns of the residue matrix associated with each
pole are multiples of the corresponding modal vector. The
modal matrix , whose columns are the modal vectors,
was defined, and it was shown that
t
will transform a
vector from spatial coordinates to modal coordinates.
Finally, we derived expressions for mass stiffness, and
damping in terms of the poles and modal vectors, and
indicated how measured parameters can be used to
calculate these matrices.
REFERENCES
1. Richardson, M. and Potter, R., "Identification of the
Modal Properties,of an Elastic Structure from
Measured Transfer Function Data", 20
th
International
Instrumentation Symposium, Albuquerque, New
Mexico, May 21-23, 1974.
2. Thoren, A.R., "Derivation of Mass and Stiffness
Matrices from Dynamic Test Data", AI4A paper 72-
346, 13'Ln Structures, Structural Dynamics, and
Materials Conference, San Antonio, Texas, April 12,
1972.
3. Rubinstein, M.F., "Structural Systems -Statics;
Dynamics and Stability", Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey, 1970.