Modal Analysis Theory

R. Kashani, Ph.D.
www.deicon.com

Modal Analysis is a computationally elegant technique for modeling structural dynamics. It is based on
the use of eigenvalue and Eigenvector information of a system. The elegance and appeal of this technique
is mainly due to its decoupling capability. Moreover, it is the basis for understanding modal test methods
(known also as experimental modal analysis).
Considering the forced response of an underdamped system shown by Equation 1

M ẍ t
 Kx t
 f t
(1)

where M and K are n  n positive definite matrices of mass and stiffness, x t
is the vector of displacement,
and f t
is the vector of applied force. Equation 1 can be solved usinjg eigenvalue expansion. This is done
by first solving the eigenvalue problem for the corresponding homogenous system

Aψ  λψ (2)

where A  M  1 K, λ is the diagonal matirx of eigenvalues, and ψ is the matrix of eigenvectors. The spatial
coordiante x can be changed to a new coordinate η using

η  ψ 1 x (3)

where ψ is an invertible matrix. Premultiplying Equation 3 by ψ results in

x  ψη (4)

Substituting for x in Equation 1 from Equation 4

Mψη̈  Kψη  f t
(5)

Premultiplying Equation 5 by ψ 1

ψ  1 Mψη̈  ψ  1 Kψη  ψ  1
f t
(6)

Using similarity transformation (described in the following section)

ψ 1
 ψT

Moreover, ψT Mψ and ψT Kψ are diagonal matrices.

Mass normalizing the eigenvectors, i.e.,

ψT Mψ  I

1
will result in
ψT Kψ  diag ω2n

Thus the use of mass normalized eigenvectors as the columns of the transformation matrix φ, results
in a set of n decoupled 2nd order differential equations of

η̈i  ω2i ηi  fi t
(7)

where fi denotes the i-th element of the vector ψT f .

Equation 4 indicates that the physical displacement of the structure, x, is the summation of its modal
contributions, i.e., modal displacements ηi ’s scaled by their corresponding eigenvector (mode shape). In
other words
n
x t
 ∑ ψi ηi (8)
i 1

1 Similarity Transformation
Two matrices A and B are similar if they have the same eigenvalues. Their transformation is via a nonsin-
gular matrix P, according to
A  P 1 BP
When the columns of the similarity transformation P consist of the n eigenvectors of B, the similar matrix
of B becomes diagonal, denoted by Λ, i.e.,

Λ  P 1 BP

or
B  PΛP 1

This can be rewritten as

BP  PΛ (9)
The matrix equation 9 is known as eigenvalue problem. If λii denotes the i-th diagonal element of the
diagonal matrix Λ, Equation 9 can be rewritten as n seperate equations

APi  λii Pi i  1 2     n (10)

Equation 10 states that Pi is the i-th eigenvector of the matrix P and λii is the associated eigenvalue.

2 Poles and Zeros

Consider the equation of motion in modal domain, i.e.,

η̈ t
 ω2nη t
 ψT f t
(11)
x  ψη (12)

Taking the Laplace transform of Equations 11 and 12 results in

2
 
s2  ω2n  η  s  ψT f  s  (13)

x  s   ψη  s  ψ s2  ω2n   1 ψT f  s  (14)

leading to the transfer function matrix mapping the input f to the output x

ψ s2  ω2n   1 ψT f  s 

and the frequency response function (FRF) of


α  ω  ψ ω2  ω2n   1 ψT
 
 ψ ω2n ω2   1 ψT (15)
 
Note that the term ω2n ω2   1 in Equation 15 is a diagonal matrix. The FRF of Equation 15 can also be
written in summation notation by considering the ik-th element of α  w  and partitioning the matrix ψ into
columns denoted by ψr . The vectors ψr are the mass normalized modal vectors, i.e., eigenvectors of the
matrix K normalized with respect to the mass matrix M. This yields
n  
α  ω ∑ ω2r ω2   1 ψr ψTr (16)
r 1

The ik-th element of the α  ω  matrix becomes

n   1
αik  ω ∑ ω2r ω2   ψr ψTr  ik (17)
r 1

where the matrix
 elements ψr ψTr  ik is identified as the modal constant or residue for the r-th mode and
the matrix ψr ψr  is called the residue matrix.
T

Partial Fraction Expansion

The 2nd. order system of Equation 18 can be expressed as the sum of partial fractions shown in Equation
19.
1
H  s   M
(18)
s λ1   s λ1 
 1 
c c
  2 (19)
 s λ1   s λ1 

where λ1  σ1  jω1 and λ1  σ1 jω1 are the complex conjugate poles and c1 and c2 defined below
are the complex conjugate residues of the rationoal polynomial transfer function describing the 2nd order
system.
1
c1  M
(20)
j2ω1
 1
c2  M
 c1 (21)
j2ω1

3
 The mathematical representation of a transfer function in terms of a partial fraction expansion in noth-
ing more than a sum of single degree of freedom systems. Therefore the mass of a single dof system,
which is by definition the modal mass, canbe related to the residue for a single dof system.
Note that the in the jargons of multi-dof system, the transfer function of a single dof system, i.e., Equa-
tion 18, is a driving point (collocated) transfer function. Considering that a multi-dof transfer function, or
frequency response function, is represented as H  s  11 or H  w  11 , the residue c1 which is the driving point
residue will be represented as cqq1
The modal mass for a single dof system is
1
M
j2ω1 c1
1
j2ω1 cqq1
j
"! (22)
2ω1 cqq1

Recalling that the residue matrix for a particular pole λr is related to the modal vectors. For an r–th
mode of an N dof system
#
c$ r Qr % u & r % u & Tr
'( +,
( u1 u1 u1 u2 * * * u1 um ,
( ,
u2 u1 u2 u2
Qr .. .. (23)
) . . -
um u1 um um r

where Qr is an arbitrary scaling constant. Now the r-th modal mass of a multi-dof is defined as
1
Mr (24)
j2Qr ωr

Example 2.1: A 2-dof system

Using the differential Equation 25,
. . .
m1 0 ẍ1 c1 c2 ! c2 ẋ1 k1 k2 ! k2 x1
0 m2 /10 ẍ2 243 ! 3 c2 c2 /10 ẋ2 253 ! 3 k2 k2 1
/ 0 x2 2
F
(25)
0 0 2
the model of the 2dof undamped system shown in Figure 1 in the absence of damping is given in
Equation 26
M ẍ  t  Kx  t  u  t  (26)
# # 3
where x x1 x2 $ 6 and u f1 f2$ 6 are the vectors of displacements (outputs) of the two masses and
forces (inputs) exciting the two masses; see Figure ???. The transfer function matrix mapping the input
vector to output vector is evaluated by taking the Laplace transfrom of Equation 26, resulting in
# 2
Ms K $ x  s  u  s  (27)
3

4
 7
F
c1 c2

8m1 8m2

k1 k2

x1 x2

Figure 1: A 2–DOF discrete system

Using the differential Equation 25

x 9 s :;=< Ms2 > K ? @ 1 u 9 s :

m1 s 2 > @ 1
;BADC
0 > C k1 k2 F k2 A
f1
m2 s 2 E EHG f2 G
(28)
0 F k2 k2
m s2 > k1 > k2 F k2 @ 1 f1
; C 1 A
M2 s2 > K2 E f2 G
(29)
F k2
m s2 > k2 k2
C 2
k2 m1 s2 > k1 > k2 E f
; A 1 G (30)
I J K
det L f2
G M sN

where
det ; m1 m2 s4 > 9 m1 k2 > m2 k1 > m2 k2 : s2 > k1 k2
is the determinant of the matrix < Ms2 > K ? . Letting m1 ; 8, m2 ; 15, k1 ; 5, and k2 ; 10, the transfer
function G 9 s : in Equation 30 becomes

G11 9 s : G12 9 s :
G 9 s :;OC
G21 9 s : G22 9 s : E
15s2 > 10 10
C
10 8s > 15 E
2

120s4 > 305s2 > 50

(31)

Looking at the first element of the transfer function matrix 31,

15s2 > 10
G11 9 s :;
120s4 > 305s2 > 50
F j0 P 0315 > j0 P 0315 > F j0 P 0334 > j0 P 0334
> >
s F j1 P 538 s j1 P 538 s F j0 P 4197 s j0 P 4197
(32)

Note that the pairs c111 ; F j0 P 0315, c111

Q ; j0 P 0315 and c112 ; F j0 P 0334, c112
Q ; j0 P 0334 are complex
conjugate residues corresponding to the two complex conjugate pairs of poles.

5
 In a similar fashion the rest of the residues for the remaining transfer functions can be determined. The
system transfer function G R s S can now be expressed in terms of partial fractions, shown in Equation 33.

0 X 0315i 0 X 0124i 0 X 0315i 0 X 0124i

0 X 0124i 0 X 0049i Y 0 X 0124i 0 X 0049i Y
G R s STVU1W U W
s 1 X 5380i Z s 1 X 5380i Z
W W Z
0 X 0334i 0 X 0453i 0 X 0334i 0 X 0453i
W Z Z
0 X 0453i 0 X 0616i Y 0 X 0453i 0 X 0616i Y
U W W U Z Z (33)
s 0 X 4197i Z s 0 X 4197i
W W Z
Note that residues corresponding to a pair W of complex conjugate poles, appear as a complex conjugate pair
themselves. Recall that the modal vector associated with each pole is proportional to the residue matrix
for that pole. This makes the modal vectors corresponding to a pair of complex conjugate poles complex
conjugate of each other, as well.
Relating the modal vector to the residue matrix of the first pole

u1 u1 u1 u2
Q1 [ u \ 1 [ u \ ]1 T Q1
u2 u1 u2 u2 Y
U 1
0 X 0315i 0 X 0124i
T
0 X 0124i 0 X 0049i Y
U W
1
3 X 15i W 1 X 24i
T 100 (34)
1 X 24i 0 X 49i Y
U^W
W
3 Modal Vector Scaling
One of the widely used scaling techniques of modal vectors is “unity modal mass” in which the scaling
factor Qr is decided upon based on having unity modal mass Mr , i.e.,
1
Qr T
j2Mr ωr
1
T
j2ωr
j
T (35)
2ωr
W
Using the unity modal mass scaling factor in conjunction with the diagonal elements (corrensponding
to driving force measurements) of Equation 23, the scaled modal coefficients can be computed as
cqqr
uqr uqr T (36)
Qr
Having the driving point scaled modal coefficients (corresponding to the diagonal elements of coefficient
matrix in Equation 23), the other coefficients (off–diagonal elements of the matrix can be calculated. In
general, the scaled modal vector is
1
[ u\ r T [ c\ r (37)
Qr uqr