20

Modal Analysis
of MDOF Unforced
Undamped Systems

20–1
Lecture 20: MODAL ANALYSIS OF MDOF UNFORCED UNDAMPED SYSTEMS

TABLE OF CONTENTS

Page
§20.1 Introduction . . . . . . . . . . . . . . . . . . . . . 20–3
§20.2 Modal Analysis of Unforced Undamped MDOF System . . . . . 20–3
§20.2.1 Natural Frequencies . . . . . . . . . . . . . . . 20–3
§20.2.2 Vibration Modes . . . . . . . . . . . . . . . . 20–3
§20.2.3 Generalized Mass and Stiffness . . . . . . . . . . . 20–5
§20.2.4 Eigenvector Mass Orthonormalization . . . . . . . . . 20–5
§20.2.5 Generalized Coordinates and Modal Matrix . . . . . . . 20–6
§20.2.6 Modal Equations of Motion . . . . . . . . . . . . 20–6
§20.2.7 Modal Initial Conditions . . . . . . . . . . . . . . 20–7
§20.3 Unforced Response Example . . . . . . . . . . . . . . . 20–7

20–2
 §20.2 MODAL ANALYSIS OF UNFORCED UNDAMPED MDOF SYSTEM

§20.1. Introduction
This subject of this Lecture and of the next one is modal analysis. This is a technique by which the
equations of motion (EOM), which are originally expressed in physical coordinates, are transformed
to modal coordinates using the eigenvalues and eigenvectors gotten by solving the undamped
frequency eigenproblem. The transformed equations are called modal equations. In a mathematical
context, modal coordinates are also called generalized coordinates, or principal coordinates. For
structural dynamics they can be interpreted as response amplitudes of orthonormalized vibration
modes.
The distinguishing feature of modal equations is that for an undamped system they uncouple.
Consequently, each modal equation may be solved independently of the others. Once computed,
modal responses may be transformed back to physical coordinates and superposed to produce the
physical response of the original system.
The method is a particular case of what is known in applied mathematics as orthogonal projection
methods. Instead of tackling the original equations directly, they are projected onto another space
(the ‘modal space” in the case of dynamics) in which equations decouple.

§20.2. Modal Analysis of Unforced Undamped MDOF System

Here we retake the two-DOF example introduced in §19.1. As numerical values we take*
m 1 = 2, m 2 = 1, c1 = c2 = 0, k1 = 6, k2 = 3, p1 = p2 = 0. (20.1)
The resulting undamped and unforced system is displayed in Figure 20.1. The known matrices and
vectors in the general EOM: Mü + Cu̇ + Ku = p, become
       
2 0 0 0 9 −3 0
M= , C= , K= , p= . (20.2)
0 1 0 0 −3 3 0
and the equations become
       
2 0 ü 1 9 −3 u1 0
+ = . (20.3)
0 1 ü 2 −3 3 u2 0

§20.2.1. Natural Frequencies

The free-vibration eigenproblem associated with (20.3) is
          
9 −3 U1 2 2 0 U1 9 − 2ω2 −3 U1 0
=ω , or = . (20.4)
−3 3 U2 0 1 U2 −3 3 − ω2 U2 0
The characteristic polynomial equation is
 
9 − 2ω2 −3
det = 18 − 15ω2 + 2ω4 = (3 − 2ω2 )(6 − ω2 ) = 0. (20.5)
−3 3 − ω2
The roots of (20.5) give the two squared frequencies
3
ω12 = = 1.5, ω22 = 6. (20.6)
2

* Note that no units are specified; being tacitly understood that a consistent set of physical units, SI or English, is used
througout.

20–3
 ;;
Lecture 20: MODAL ANALYSIS OF MDOF UNFORCED UNDAMPED SYSTEMS

(a) (b)

k1 = 6
Static equilibrium
position Fs1 = k1 u1

u 1 = u1(t)
Mass m 1 = 2

..
Fs2 FI1 = m 1 u1
Static equilibrium k2 = 3
position
Fs2 = k2 (u2−u1)

u 2 = u2(t)
Mass m 2 = 1

..
x FI2 = m 2 u 2

Figure 20.1. Two-DOF, unforced, undamped spring-mass example system: (a) configuration, (b) DFBD.

§20.2.2. Vibration Modes

To get U1 , replace ω12 = 3/2 into the second of (20.4), set its first entry U11 to one, and solve for
the second entry U12 :
         
9−2 × (3/2) −3 U11 6 −3 U11 0 1
= = ⇒ U1 = . (20.7)
−3 3−3/2 U12 −3 3/2 U12 0 2
To get U2 , replace ω22 = 6 into the second of (20.5), set its first entry U21 to one, and solve for the
second entry U22 :
         
9−2 × 6 −3 U21 −3 −3 U21 0 1
= = ⇒ U2 = . (20.8)
−3 3−6 U22 −3 3 U22 0 −1
We now apply the first normalization method described in §19.3.3, by forcing the largest entry to
be +1:    
1/2 1
φ1 = 2 U1 =
1
, φ2 = U2 . (20.9)
1 −1
The vibration modes are pictured in Figure 20.2. Note that in the first mode the masses oscillate in
phase while in the second one they are 180◦ out of phase, moving opposite to each other.
It is convenient at this point to verify the orthogonality property: φiT Mφ j = 0 and φiT Kφ j = 0
for i
= j. Here only one combination, namely i = 1 and j = 2, needs to be tested:
     
2 0 1 9 −3 1
φ1 M φ2 = [ 1/2 1 ]
T
= 0, φ1 K φ2 = [ 1/2 1 ]
T
= 0. (20.10)
0 1 −1 −3 3 −1

20–4
 ;; ;;
§20.2 MODAL ANALYSIS OF UNFORCED UNDAMPED MDOF SYSTEM

Mode 1, ω 21 = 3/2 Mode 2, ω22 = 6

1/2 Mass m 1 1

Red lines mark the

static equilibrium
positions of the masses

−1
1 Mass m 2

Figure 20.2. Vibration modes for example system.

Note that there is no need to explicitly check that φ2T M φ1 = 0 and φ2T K φ1 = 0 because
(φ1T M φ2 )T = φ2T MT φ1 = φ2T M φ1 , since M = MT if M is symmetric. Likewise for K.
The vibration modes (20.9) are orthogonal but not orthonormal with respect to the mass matrix.
To achieve that property it is necessary to rescale them so that generalized masses are unity. This
is done next.

§20.2.3. Generalized Mass and Stiffness

Next we compute the generalized masses and stiffnesses associated with the modes (20.10):

 
2 0 1/2
M1 = M φ1 = [ 1/2 1 ]
φ1T = 3/2,
0 1 1
  
2 0 1
M2 = φ2 M φ2 = [ 1 −1 ]
T
= 3,
0 1 −1
   (20.11)
9 −3 1/2
K 1 = φ1 K φ1 = [ 1/2 1 ]
T
= 9/4,
−3 3 1
  
9 −3 1
K 2 = φ2 K φ2 = [ 1 −1 ]
T
= 18.
−3 3 −1

Quick check:
K1 9/4 K2 18
ω12 = = = 3/2, ω22 = = = 6, (20.12)
M1 3/2 M2 3
as expected. Note that should eigenvectors be normalized in a different way, M1 , M2 , K 1 , K 2 will
be generally different, but the ratios ωi2 = K i /Mi remain invariant.

20–5
Lecture 20: MODAL ANALYSIS OF MDOF UNFORCED UNDAMPED SYSTEMS

§20.2.4. Eigenvector Mass Orthonormalization

Next we renormalize φ1 and φ2 to get unit generalized masses. These rescaled vectors will be
initially denoted by φ̃1 and φ̃2 , respectively. After renormalization the tildes are dropped.
T
Suppose φ̃1 = c1 φ1 . To find c1 , insert this into 1 = φ̃1 M φ̃1 =√(c1 φ1T ) M (c1 φ1 ) = c12 φ1T M φ1 =
√ √
c12 × 3/2, whence c1 = 1/ 3/2 = 2/3. Likewise c2 = 1/ 3. Dropping the tildes for brevity,
the mass-orthonormalized vibration modes are
         
2 1/2 0.4088 1 1 0.5773
φ1 = = , φ2 = = (20.13)
3 1 0.8165 3 −1 −0.5773

After this renormalization,

3
φ1T M φ1 = 1, φ2T M φ2 = 1, φ1T K φ1 = = ω12 , φ2T K φ2 = 6 = ω22 . (20.14)
2

§20.2.5. Generalized Coordinates and Modal Matrix

We will express† the displacement vector u(t) in terms of normal coordinates η1 (t) and η2 (t), as
follows:
   
u 1 (t) def η1 (t)
u= = φ1 η1 (t) + φ2 η2 (t) = [ φ1 φ2 ] = Φ η. (20.15)
u 2 (t) η2 (t)

In a mathematical context, the ηi (t)s are called generalized coordinates or principal coordinates.
They represent the amplitude of the orthonormalized mode shapes, or modal amplitudes for short.
Equation (20.15) is an instance of modal superposition, which is a general feature of linear dynamical
systems.
Two new matrix symbols appear in (20.15). The normal coordinate vector η collects η1 (t) and
η2 (t) as its entries. The modal matrix Φ is formed by stacking the mass-orthogonal eigenvectors
as columns:  1 
√ √1  
 6 3  0.4082 0.5773
Φ = [ φ1 φ2 ] = = . (20.16)
√2 − √1 0.8165 −0.5773
6 3
Note that u̇ = Φ η̇ and ü = Φ η̈, because Φ does not depend on time. With the help of Φ, the
orthonormality conditions can be expressed compactly as‡
   2   
1 0 ω1 0 3/2 0
Φ M Φ = Mg = I =
T
, Φ K Φ = Kg = diag[ωi ] =
T 2
= .
0 1 0 ω22 0 6
(20.17)
Here Mg and Kg denote the generalized mass matrix and generalized stiffness matrix, respectively.
If Φ is built with mass-orthogonalized eigenvectors, as in (20.16), Mg reduces to the identity matrix
while Kg becomes a diagonal matrix with squared frequencies stacked along its diagonal.
† This is a consequence of the so-called expansion theorem for MDOF systems; in the applied mathematics context it is
known as the spectral decomposition. See Section 10.1.9 of the Craig-Kurdila textbook for details.
‡ The Craig-Kurdila textbook uses bold italics symbols for generalized mass and stiffness matrices. Since those fonts are
not available to the TEX document processor used for these Lectures, we shall use g subscripts to denote those quantities.

20–6
 §20.3 UNFORCED RESPONSE EXAMPLE

§20.2.6. Modal Equations of Motion

To transform the undamped, unforced EOM: Mü + Ku = 0 to modal coordinates, replace u = Φη
and ü = Φη̈ into it, and premultiply by ΦT :
ΦT M Φ η̈ + ΦT K Φ η = ΦT 0 = 0. (20.18)
Using the generalized matrices introduced in (20.17) we get
η̈ + Kg η = 0, (20.19)

For the example problem,

       
1 0 η̈1 (t) 3/2 0 η1 (t) 0
+ = . (20.20)
0 1 η̈2 (t) 0 6 η2 (t) 0
Because the matrices in (20.20) are diagonal, (20.20) uncouples into two homogeneous, second-
order ODE already in canonical form:
η̈1 (t) + (3/2) η1 (t) = 0, η̈2 (t) + 6 η2 (t) = 0. (20.21)
Each ODE models an unforced, undamped SDOF oscillator. As such it may be treated using the
methods of Lecture 17 as long as initial conditions (IC) are known. Once solutions η1 (t) and η2 (t)
are available, they can be combined via the mode superposition relation (20.15) to get the physical
response u(t) = Φ η(t). But to solve (20.21) we need their IC in modal coordinates.
§20.2.7. Modal Initial Conditions
Suppose that the initial conditions for a MDOF system are
u(0) = u0 , u̇(0) = v0 . (20.22)
Here u0 and v0 denote vectors of initial displacements and velocities, respectively. Because u(t) =
Φ η(t) and u̇(t) = Φ η̇(t) for any time t, setting t = 0 we get
u0 = u(0) = Φ η(0), v0 = u̇(0) = Φ η̇(0). (20.23)
These can be solved for η0 = η(0) and η̇0 = η̇(0) by inverting the modal matrix:
η0 = η(0) = Φ−1 u0 , η̇0 = η̇(0) = Φ−1 v0 . (20.24)
or equivalently, solving the linear systems (20.23). But there is a more elegant scheme that only
requires matrix multiplication. Postmultiplying both sides of ΦT MΦ = I by Φ−1 gives Φ−1 =
ΦT M, whence
η0 = ΦT M u0 , η̇0 = ΦT M v0 . (20.25)

Remark 20.1. Should the eigenvectors φi not be mass-orthonormal, we have the more complicated rela-
tions Mg η0 = ΦT M u0 and Mg η̇0 = ΦT M v0 , in which Mg = ΦT MΦ is the generalized mass matrix.
Consequently
η0 = M−1
g Φ M u0 ,
T
η̇0 = M−1
g Φ M v0 .
T
(20.26)

Since Mg is always diagonal, so is M−1 g . Consequently (20.26) merely amounts to scaling entries of η0 and
η̇0 by reciprocals of the generalized masses.

20–7
Lecture 20: MODAL ANALYSIS OF MDOF UNFORCED UNDAMPED SYSTEMS

1 .
0.6 u1(t)
0.4 u1(t) 0.5
Displacements

0.2

Velocities
0
0
−0.2 −0.5
−0.4 .
u2(t) −1 u2(t)
−0.6
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time t Time t
Figure 20.3. Response histories for example system under initial conditions u̇ 1 (0) = 1, others zero.

§20.3. Unforced Response Example

The following initial conditions (IC) are specified for the 2-DOF mass-spring system of Figure
20.1: mass m 1 is given a unit initial velocity along +x, while all other initial values are zero. In
matrix form:
   
0 1
u0 = , v0 = . (20.27)
0 0

The modal IC are obtained from (20.25) as

 1 
  √ √2     
η10 2 0 0 0
η0 = = Φ M u0 = 
T 6 6  = ,
η20 1
√ −√ 1 0 1 0 0
3 3 (20.28)
 1   2 
  √ √2    √  
η̇10  6 6  2 0 1  6  0.8165
η̇0 = = Φ M v0 =
T
= = .
η̇20 √1 − √1 0 1 0 √2 1.1547
3 3 3

The solution of the modal equations (20.20) for the initial conditions (20.28) are

η̇10 2 √ 2
η1 (t) = sin ω1 t = √ sin( 6 t) = sin( 3/2 t) = 0.6667 sin(1.2247 t),
ω1 3 3 3
√ (20.29)
η̇20 2/ 3 √ 2 √
η2 (t) = sin ω2 t = √ sin( 6t) = √ sin( 6 t) = 0.4714 sin(2.4495 t).
ω2 6 3 2

These can be transformed back to physical coordinates via the modal matrix, as per u(t) = Φη(t).
Here are the calculations for the mass displacement responses, with numerical expressions listed

20–8
 §20.3 UNFORCED RESPONSE EXAMPLE

to 4-place accuracy:
 1 
  √  
  √ √1 2 sin(√3/2 t)  1 2
sin(
√
3/2t) + sin( 6t)
u 1 (t) 3  3
= 6 √  
3 3
2 sin( 6 t) =  √ √
u 2 (t) √2 − √1 √ 1 2
2 sin( 3/2t) − sin( 6t)
6 3 3 2 3 3

 
0.2722 sin(1.2247t) + sin(2.4495t)
= 
0.2722 2 sin(1.2247t) − sin(2.4495t)
(20.30)
Mass velocities are obtained by direct numerical differentiation with respect to t:
 
 
1 cos(√3/2t) + 2 cos(√6t) 
0.3333 cos(1.2247t) + 2 cos(2.4495t) 
u̇ 1 (t) √  = 
3
= √
u̇ 2 (t) 2
cos( 3/2t) − cos( 6t) 0.6667 cos(1.2247t) − cos(2.4495t)
3
(20.31)
The results (20.30) and (20.31) are plotted in Figure 20.3 over 0 ≤ t ≤ 12. Since there is no
damping, the harmonic oscillation patterns visible in the Figure will repeat with no distortions or
amplitude decay. In other words, the free vibrations will go on forever.* Observe that the initial
conditions (20.27) are correctly reproduced.

* The reason behind the repeating oscillation patterns in Figure 20.3 is that frequency ω2 is exactly twice ω1 in the example.

20–9