Chapter 4 Multiple Degree of

Freedom Systems

The Millennium bridge required

many degrees of freedom to model
and design with.

Extending the first 3 chapters to

more then one degree of freedom
© D.J. Inman
1/58 Mechanical Engineering at Virginia Tech
The first step in analyzing multiple
degrees of freedom (DOF) is to look at 2
DOF
• DOF: Minimum number of coordinates to specify the
position of a system
• Many systems have more than 1 DOF
• Examples of 2 DOF systems
– car with sprung and unsprung mass (both heave)
– elastic pendulum (radial and angular)
– motions of a ship (roll and pitch)

© D.J. Inman Fig 4.1

Mechanical Engineering at Virginia Tech
2/58
4.1 Two-Degree-of-Freedom
Model (Undamped)

A 2 degree of freedom system used to base

much of the analysis and conceptual
development of MDOF systems on.
© D.J. Inman
3/58 Mechanical Engineering at Virginia Tech
Free-Body Diagram of each mass
Figure 4.2
k2(x2 -x1)

k1 x1 m1 m2

x1 x2
© D.J. Inman
4/58 Mechanical Engineering at Virginia Tech
Summing forces yields the
equations of motion:
m1 x1 (t )   k1 x1 (t )  k 2  x2 (t )  x1 (t ) 
(4.1)
m2 x2 (t )   k 2  x2 (t )  x1 (t ) 

Rearranging terms:
m1 x1 (t )  ( k1  k 2 ) x1 (t )  k 2 x2 (t )  0
(4.2)
m2 x2 (t )  k 2 x1 (t )  k 2 x2 (t )  0

© D.J. Inman
5/58 Mechanical Engineering at Virginia Tech
Note that it is always the case
that
• A 2 Degree-of-Freedom system has
– Two equations of motion!
– Two natural frequencies (as we shall see)!

© D.J. Inman
6/58 Mechanical Engineering at Virginia Tech
The dynamics of a 2 DOF system consists
of 2 homogeneous and coupled equations
• Free vibrations, so homogeneous eqs.
• Equations are coupled:
– Both have x1 and x2.
– If only one mass moves, the other follows
– Example: pitch and heave of a car model
• In this case the coupling is due to k2.
– Mathematically and Physically
– If k2 = 0, no coupling occurs and can be solved
as two independent SDOF systems
© D.J. Inman
7/58 Mechanical Engineering at Virginia Tech
Initial Conditions
• Two coupled, second -order, ordinary
differential equations with constant
coefficients
• Needs 4 constants of integration to solve
• Thus 4 initial conditions on positions and
velocities
x1 (0)  x10 , x1 (0)  x10 , x2 (0)  x20 , x2 (0)  x20

© D.J. Inman
8/58 Mechanical Engineering at Virginia Tech
Solution by Matrix Methods
The two equations can be written in the form of a
single matrix equation (see pages 272-275 if matrices and
vectors are a struggle for you) :
 x1 (t )   x1 (t )   x1 (t ) 
x(t )    , x (t )    , x (t )    (4.4), (4.5)
 2 
x (t )  2 
x ( t )  2 
x ( t )

m 0 k  k k2 
M  1  K  1 2
k2 
, (4.6), (4.9)
0 m2    k2

Mx  K x  0 m1 x1 (t )  (k1  k 2 ) x1 (t )  k 2 x2 (t )  0
m2 x2 (t )  k 2 x1 (t )  k 2 x2 (t )  0
© D.J. Inman
9/58 Mechanical Engineering at Virginia Tech
Initial Conditions
IC’s can also be written in vector form

 x10   x10 
x(0)    , and x(0)   
 x20   x20 

© D.J. Inman
10/58 Mechanical Engineering at Virginia Tech
The approach to a Solution:
For 1DOF we assumed the scalar solution aelt
Similarly, now we assume the vector form:
jt
Let x(t )  ue (4.15)

j  1, u  0,  , u unknown
  - M  K  ue
2 jt
0 (4.16)

  - M  K  u  0
2
(4.17)

© D.J. Inman
11/58 Mechanical Engineering at Virginia Tech
This changes the differential
equation of motion into algebraic
vector equation:
 M  K u  0
- 2
(4.17)
This is two algebraic equation in 3 uknowns
( 1 vector of two elements and 1 scalar):
 u1 
u =   , and 
u2 

© D.J. Inman
12/58 Mechanical Engineering at Virginia Tech
The condition for solution of this matrix
equation requires that the the matrix
inverse does not exist:
If the inv  - M  K  exists  u  0 : which is the
2

static equilibrium position. For motion to occur

u  0   - M  K 
2 1
does not exist
or det  - 2 M  K   0 (4.19)

The determinant results in 1 equation

in one unknown (called the characteristic equation)
© D.J. Inman
13/58 Mechanical Engineering at Virginia Tech
Back to our specific system: the
characteristic equation is defined as
det- M  K   0 
2

 2 m1  k1  k2 
k2 (4.20)
det  0 
 k2  m2  k2 
2

m1m 2 4  (m1k2  m2 k1  m2 k2 ) 2  k1 k2  0 (4.21)

Eq. (4.21) is quadratic in 2so four solutions result:

 and   1 and  2
2
1
2
2
© D.J. Inman
14/58 Mechanical Engineering at Virginia Tech
Once  is known, use equation (4.17) again to
calculate the corresponding vectors u1 and u2
This yields vector equation for each squared frequency:
(12 M  K )u1  0 (4.22)
and
(22 M  K )u 2  0 (4.23)

Each of these matrix equations represents 2

equations in the 2 unknowns components of the
vector, but the coefficient matrix is singular so
each matrix equation results in only 1 independent
equation. The following examples clarify this.
© D.J. Inman
15/58 Mechanical Engineering at Virginia Tech
Examples 4.1.5 & 4.1.6:calculating u
and 

• m1=9 kg,m2=1kg, k1=24 N/m and k2=3 N/m

• The characteristic equation becomes
4-62+8=(2-2)(2-4)=0
2 = 2 and 2 =4 or

 1,3   2 rad/s,  2,4   2 rad/s

Each value of 2 yields an expression or u:

© D.J. Inman
16/58 Mechanical Engineering at Virginia Tech
 Computing the vectors u
 u11 
For  =2, denote u1    then we have
2
1
u12 
(-12 M  K )u1  0 
 27  9(2) 3   u11  0 
 3    
 3  (2)  u12  0 
9u11  3u12  0 and  3u11  u12  0
2 equations, 2 unknowns but DEPENDENT!
(the 2nd equation is -3 times the first)
© D.J. Inman
17/58 Mechanical Engineering at Virginia Tech
Only the direction of vectors u can be
determined, not the magnitude as it
remains arbitrary
u11 1 1
  u11  u12 results from both equations:
u12 3 3
only the direction, not the magnitude can be determined!
This is because: det( 12 M  K )  0.
The magnitude of the vector is arbitrary. To see this suppose
that u1 satisfies
( 12 M  K )u1  0, so does au1 , a arbitrary. So
(12 M  K )au1  0  (12 M  K )u1  0

© D.J. Inman
18/58 Mechanical Engineering at Virginia Tech
Likewise for the second value of 2:
 u21 
For  = 4, let u 2    then we have
2
2
u22 
(-12 M  K )u  0 
 27  9(4) 3   u21  0 
 3    
 3  (4)  u22   0 
1
9u21  3u22  0 or u21   u22
3
Note that the other equation is the same
© D.J. Inman
19/58 Mechanical Engineering at Virginia Tech
What to do about the magnitude!
Several possibilities, here we just fix one element:

Choose:
 3 1
u12  1  u1   
1
Choose:
 3 1
u22  1  u 2   
1
© D.J. Inman
20/58 Mechanical Engineering at Virginia Tech
 Thus the solution to the algebraic
matrix equation is:
 3 1
 1, 3   2, has mode shape u1   
1
 1 3 
 2,4   2, has mode shape u2   
1
Here we have introduce the name
mode shape to describe the vectors
u1 and u2. The origin of this name comes later
© D.J. Inman
21/58 Mechanical Engineering at Virginia Tech
Return now to the time response:
We have computed four solutions:
x(t )  u1e  j1t , u1e j1t , u 2 e  j2t , u 2 e j2t  (4.24)
Since linear, we can combine as:
x(t )  au1e  j1t  bu1e j1t  cu 2 e  j2t  du 2 e j2t
 x(t )   ae  j1t  be j1t  u1   ce  j2t  de j2t  u 2
 A1 sin(1t  1 )u1  A2 sin(2t  2 )u 2 (4.26)

where A1 , A2 , 1 , and 2 are constants of integration

Note that to go from the exponential
form to to sine requires Euler’s formula
determined by initial conditions. for trig functions and uses up the
+/- sign on omega
© D.J. Inman
22/58 Mechanical Engineering at Virginia Tech
Physical interpretation of all that
math!
• Each of the TWO masses is oscillating at TWO
natural frequencies 1 and 2
• The relative magnitude of each sine term, and
hence of the magnitude of oscillation of m1 and m2
is determined by the value of A1u1 and A2u2
• The vectors u1 and u2 are called mode shapes
because the describe the relative magnitude of
oscillation between the two masses

© D.J. Inman
23/58 Mechanical Engineering at Virginia Tech
What is a mode shape?
• First note that A1, A2, 1 and 2 are determined by
the initial conditions
• Choose them so that A2 = 1 = 2 =0
• Then:
 x1 (t)   u11 
x(t)     A1   sin 1t  A1u1 sin 1t
 x2 (t)  u12 
• Thus each mass oscillates at (one) frequency 1
with magnitudes proportional to u1 the 1st mode
shape

© D.J. Inman
24/58 Mechanical Engineering at Virginia Tech
A graphic look at mode shapes:
If IC’s correspond to mode 1 or 2, then the response is purely in
mode 1 or mode 2.
x1 x2
k1 k2
 13
Mode 1: m1 m2 u1   
1
x1=A/3 x2=A

x1 x2
k1 k2
 1 3 
Mode 2: m1 m2 u2   
1
x1=-A/3 x2=A

© D.J. Inman
25/58 Mechanical Engineering at Virginia Tech
Example 4.1.7 given the initial
conditions compute the time response
1  0 
consider x(0) =   mm, x (0)   
0  0 
 
sin  2t  1   sin 2t  2 
A1 A2
 1 
x ( t )

 x (t )   3 3
 2   A sin  2t     A sin 2t    
1 1 2 2

 
2 cos  2t  1   2 cos 2t  2 
A A
 x1 (t )  1 2

 x (t )   3 3
 2   A 2 cos  2t     A 2 cos 2t    
1 1 2 2

© D.J. Inman
26/58 Mechanical Engineering at Virginia Tech
At t=0 we have

 
1 mm 
 
A1
sin   
A2
sin  
 0 
1 2
3 3
   A1 sin 1   A2 sin 2  
 
2 cos1   2 cos2 
A1 A2
0  

0   3 3

 A1 2 cos1   2 A2 cos2  
 

© D.J. Inman
27/58 Mechanical Engineering at Virginia Tech
4 equations in 4 unknowns:
3  A1 sin 1   A2 sin 2 
0  A1 sin 1   A2 sin 2 
0  A1 2 cos1   A2 2 cos2 
0  A1 2 cos1   A2 2 cos2 
Yields:

A1  1.5 mm , A2  1.5 mm , 1  2  rad
2

© D.J. Inman
28/58 Mechanical Engineering at Virginia Tech
The final solution is:
x1 (t)  0.5 cos 2t  0.5 cos 2t
(4.34)
x2 (t)  1.5 cos 2t  1.5 cos 2t
These initial conditions gives a response that is a combination
of modes. Both harmonic, but their summation is not.

Figure 4.3
© D.J. Inman
29/58 Mechanical Engineering at Virginia Tech
Solution as a sum of modes

x(t)  a1u1 cos 1t  a2u 2 cos  2 t

Determines how the second

frequency contributes to the
Determines how the first response
frequency contributes to the
response

© D.J. Inman
30/58 Mechanical Engineering at Virginia Tech
Things to note
• Two degrees of freedom implies two natural
frequencies
• Each mass oscillates at with these two frequencies
present in the response and beats could result
• Frequencies are not those of two component
systems
k1 k2
1  2  1.63, 2  2   1.732
m1 m2

• The above is not the most efficient way to

calculate frequencies as the following describes
© D.J. Inman
31/58 Mechanical Engineering at Virginia Tech
Some matrix and vector reminders
a b 1 1  d b 
A  A   
 c c  ad  cb   c a 
x x  x1  x2
T 2 2

 m1 0 
M    x Mx  m1 x1  m2 x2
T 2 2

 0 m2 
M  0  x Mx  0 for every value of x except 0
T

Then M is said to be positive definite

© D.J. Inman
32/58 Mechanical Engineering at Virginia Tech
4.2 Eigenvalues and Natural
Frequencies
• Can connect the vibration problem with the
algebraic eigenvalue problem developed in
math
• This will give us some powerful
computational skills
• And some powerful theory
• All the codes have eigen-solvers so these
painful calculations can be automated
© D.J. Inman
33/58 Mechanical Engineering at Virginia Tech
Some matrix results to help us use
available computational tools:
A matrix M is defined to be symmetric if
MM T

A symmetric matrix M is positive definite if

x Mx  0 for all nonzero vectors x
T

A symmetric positive definite matrix M can

be factored M  LL T

Here L is upper triangular, called a Cholesky matrix

© D.J. Inman
34/58 Mechanical Engineering at Virginia Tech
If the matrix L is diagonal, it defines
the matrix square root
The matrix square root is the matrix M 1/ 2 such that
M 1/ 2 M 1/ 2  M
If M is diagonal, then the matrix square root is just the root
of the diagonal elements:
 m1 0 
LM 1/ 2
  (4.35)
 0 m2 

© D.J. Inman
35/58 Mechanical Engineering at Virginia Tech
A change of coordinates is introduced to
capitalize on existing mathematics
For a symmetric, positive definite matrix M:
 m1 0 1  1 m1 0  1/ 2
 1
m1
0 
M   , M  1 
, M  
0 m2  0 m2   0 1
m2 

Let x(t )  M 1/ 2q(t ) and multiply by M 1/ 2 :
M 1/ 2 MM 1/ 2 q (t )  M 1/ 2 KM 1/ 2 q (t )  0 (4.38)
I identity K symmetric

or q(t )  Kq(t )  0 where K  M 1/ 2 KM 1/ 2

k
K is called the mass normalized stiffness and is similar to the scalar
m
used extensively in single degree of freedom analysis. The key here is that
K is a SYMMETRIC matrix allowing the use of many nice properties and
computational tools
© D.J. Inman
36/58 Mechanical Engineering at Virginia Tech
How the vibration problem relates to the
real symmetric eigenvalue problem
j t
Assume q(t )  ve in q(t )  Kq(t )  0
jt jt
 ve 2
 K ve  0, v  0 or
Kv   v 2
 Kv  l v v0
vibration problem real symmetric
eigenvalue problem
(4.40) (4.41)

Note that the martrix K contains the same type of information

as does n2 in the single degree of freedom case.
© D.J. Inman
37/58 Mechanical Engineering at Virginia Tech
Important Properties of the n x n Real
Symmetric Eigenvalue Problem
• There are n eigenvalues and they are all real
valued
• There are n eigenvectors and they are all real
valued
• The set of eigenvectors are orthogonal
• The set of eigenvectors are linearly
independent
• The matrix is similar to a diagonal matrix
Window 4.1 page 285
© D.J. Inman
38/58 Mechanical Engineering at Virginia Tech
Square Matrix Review
• Let aik be the ikth element of A then A is symmetric
if aik = aki denoted AT=A
• A is positive definite if xTAx > 0 for all nonzero x
(also implies each li > 0)
• The stiffness matrix is usually symmetric and
positive semi definite (could have a zero
eigenvalue)
• The mass matrix is positive definite and
symmetric (and so far, its diagonal)

© D.J. Inman
39/58 Mechanical Engineering at Virginia Tech
Normal and orthogonal vectors
 x1   y1  n
   
x   M , y   M , inner product is x y   xi yi
T

i 1
 xn   yn 
x orthogonal to y if x T y  0 

x is normal if x x  1
T

if a the set of vectores is is both orthogonal and normal it
is called an orthonormal set

The norm of x is x  xT x (4.43)

© D.J. Inman
40/58 Mechanical Engineering at Virginia Tech
Normalizing any vector can be
done by dividing it by its norm:
x
has norm of 1 (4.44)
T
x x
To see this compute
x xT x xT x
  T
1
T
x x x xT
x xT x x

© D.J. Inman
41/58 Mechanical Engineering at Virginia Tech
Examples 4.2.2 through 4.2.4
1/ 2 1/ 2  1 0   27  3   1 0
K  M KM 
3 3
  3 3   0 1 
 0 1   
 3 1
so K    which is symmetric.
 1 3 
3-l -1 
det(K  l I )  det    l  6l  8  0
2

 -1 3-l 
which has roots: l1  2  1 and l2  4  2
2 2

© D.J. Inman
42/58 Mechanical Engineering at Virginia Tech
 ( K  l1 I ) v1  0 
3  2 1   v11   0 
 1 3  2   v    0  
   12   
1
v11  v12  0  v1    
1
v1   (1  1)  1   
2 1
2

1 1
v1  1
2  The first normalized eigenvector
© D.J. Inman
43/58 Mechanical Engineering at Virginia Tech
Likewise the second normalized eigenvector is
computed and shown to be orthogonal to the first,
so that the set is orthonormal
1 1 1
v2   1 , v1 v 2  2 (1  1)  0
T

2 
1
v1 v1  (1  1)  1
T

2
1
v 2 v 2  (1  (1)(1))  1
T

2
 v i are orthonormal
© D.J. Inman
44/58 Mechanical Engineering at Virginia Tech
Modes u and Eigenvectors v are
different but related:
u1  v1 and u 2  v 2
1/2 1/2
xM qu M v (4.37)
Note
 3 0   1  1
M u1  
1/2 3
      v1

 0 1   1  1

© D.J. Inman
45/58 Mechanical Engineering at Virginia Tech
This orthonormal set of vectors is
used to form an Orthogonal Matrix
P   v1 v2  called a matrix of eigenvectors (normalized)
 v T
1 v1 v1T v 2  1 0 
P P T
T
T   I
 v 2 v1 v 2 v 2  0 1 
P is called an orthogonal matrix
PT KP  PT  Kv1 Kv 2   PT  l1 v1 l2 v 2 
l1 v1T v1 l2 v1T v 2  l1 0 
 T 
    diag( 2
,  2 )   (4.47)
2

l1 v 2 v1 l2 v 2 v 2   0 l2 
T 1

P is also called a modal matrix

© D.J. Inman
46/58 Mechanical Engineering at Virginia Tech
Example 4.2.3 compute P and show
that it is an orthogonal matrix
From the previous example:
1 1 1 
P  v1 v1   1 1 
2 
1 1 1 1  1 1 
P P
T
1 1 1 1
2 2  
1 1  1 1  1 1  2 0 
      I
2 1  1 1  1 2  0 2 

© D.J. Inman
47/58 Mechanical Engineering at Virginia Tech
Example 4.2.4 Compute the square of
the frequencies by matrix manipulation
1 1 1   3 1 1 1 1 
P KP 
T
1 1  1 3  1 1
2   2 
1 1 1   2 4 
 
2 1 1  2 4 
1 4 0 2 0 12 0 
      2
2 0 8 0 4  0 2 
 1  2 rad/s and  2  2 rad/s
In general:
  PT KP  diag  li   diag(i2 ) (4.48)
© D.J. Inman
48/58 Mechanical Engineering at Virginia Tech
Example 4.2.5

The equations of motion: Figure 4.4

m1 x1  (k1  k2 ) x1  k2 x2  0
(4.49)
m2 x2  k2 x1  (k2  k3 ) x2  0

In matrix form these become:

 m1 0  k1  k2 k2 
0  x  x0 (4.50)
 m2   k2 k 2  k3 

© D.J. Inman
49/58 Mechanical Engineering at Virginia Tech
Next substitute numerical values
and compute P and 
m1  1 kg, m2  4 kg, k1  k3  10 N/m and k2 =2 N/m
1 0  12 2 
M   , K  
 0 4    2 12 
1/ 2 1/ 2 12 1
 K  M KM  
  1 12 
12  l 1 
 det  K  l I   det    l 2
 15l  35  0
 1 12  l 
 l1  2.8902 and l2  12.1098
 1  1.7 rad/s and 2  12.1098 rad/s
© D.J. Inman
50/58 Mechanical Engineering at Virginia Tech
Next compute the eigenvectors
For l1 equation (4.41 ) becomes:
12 - 2.8902 1   v11 
    0
 1 3 - 2.8902   v21 
 9.1089v11  v21
Normalizing v1 yields
1  v1  v112  v21
2
 v112  (9.1089) 2 v112
 v11  0.1091, and v21  0.9940
 0.1091  0.9940 
v1    , likewise v 2   
 0.9940   0.1091 

© D.J. Inman
51/58 Mechanical Engineering at Virginia Tech
Next check the value of P to see
if it behaves as its suppose to:
 0.1091 0.9940 
P   v1 v2    
 0.9940 0.1091 

 0.1091 0.9940  12 1  0.1091 0.9940   2.8402 0 

PT KP     
 
 0.9940 0.1091  1 3  0.9940 0.1091   0 12.1098 

 0.1091 0.9940   0.1091 0.9940  1 0 

P P
T
    
 0.9940 0.1091 0.9940 0.1091   0 1 

Yes!
© D.J. Inman
52/58 Mechanical Engineering at Virginia Tech
A note on eigenvectors
In the previous section, we could have chosed v 2 to be
 0.9940  -0.9940 
v2    instead of v 2   
 0.1091  0.1091 
because one can always multiple an eigenvector by a constant
and if the constant is -1 the result is still a normalized vector.

Does this make any difference?

No! Try it in the previous example

© D.J. Inman
53/58 Mechanical Engineering at Virginia Tech
All of the previous examples can and should
be solved by “hand” to learn the methods
However, they can also be solved on
calculators with matrix functions and
with the codes listed in the last section
In fact, for more then two DOF one must
use a code to solve for the natural
frequencies and mode shapes.
Next we examine 3 other formulations for
solving for modal data
© D.J. Inman
54/58 Mechanical Engineering at Virginia Tech
Matlab commands
• To compute the inverse of the square matrix
A: inv(A) or use A\eye(n) where n is the
size of the matrix
• [P,D]=eig(A) computes the eigenvalues and
normalized eigenvectors (watch the order).
Stores them in the eigenvector matrix P and
the diagonal matrix D (D=)

© D.J. Inman
55/58 Mechanical Engineering at Virginia Tech
More commands
• To compute the matrix square root use sqrtm(A)
• To compute the Cholesky factor: L= chol(M)
• To compute the norm: norm(x)
• To compute the determinant det(A)
• To enter a matrix:
K=[27 -3;-3 3]; M=[9 0;0 1];
• To multiply: K*inv(chol(M))

© D.J. Inman
56/58 Mechanical Engineering at Virginia Tech
An alternate approach to normalizing
mode shapes
From equation (4.17) M 2  K u  0, u0

Now scale the mode shapes by computing  such that

1
 i ui 
T
M  i ui   1   i 
T
u ui
i

w i   i ui is called mass normalized and it satisfies:

i2 Mw i  Kw i  0  i2  wTi Kw i , i  1, 2
(4.53)
© D.J. Inman
57/58 Mechanical Engineering at Virginia Tech
There are 3 approaches to computing
mode shapes and frequencies
1 1
(i)  Mu  Ku (ii)  u  M Ku
2 2 1
(iii)  v  M KM 2 v
2 2

(i) Is the Generalized Symmetric Eigenvalue Problem

easy for hand computations, inefficient for computers

(ii) Is the Asymmetric Eigenvalue Problem

very expensive computationally

(iii) Is the Symmetric Eigenvalue Problem

the cheapest computationally
© D.J. Inman
58/58 Mechanical Engineering at Virginia Tech