Last update: 2010 Ahmed Elgamal

Multi-Degree-Of-Freedom (MDOF)
Systems and Modal Analysis

Ahmed Elgamal

Ahmed Elgamal

SDOF Shear Building (rigid roof)

mu  ku  cu  mu g

m = lumped mass = mroof + 2 (1/2 mcol)

12EI c 24EI c
k  2k col  2 
h3 h3

1
 Ahmed Elgamal
2-Story Shear Building (2-DOF system)
1 
m 2  m 2 roof  2 m col 
2 
u2 1 
m2 m1  m1floor  4 m col 
2 

k2 k1  2k col1 k 2  2k col2

m1 m 2 u 2  k 2 (u 2  u1 )  m 2 u g
u1
m1u1  k 2 (u1  u 2 )  k1u1  m1ug

m1 0   u1  k1  k 2   k 2   u1   m1 

k1  0 m  u     k  u    m u g
 2  2   2 k 2  2   2

or, mu  k u  m1ug

  ku  m1ug
mu
where 1 or 1 is the Identity Matrix 3

3-Story Shear Building (3-DOF system) Ahmed Elgamal

m3
u3

k3

m2
u2

k2 m3u 3  k 3 (u 3  u 2 )  m3u g
m2u 2  k 3 (u 2  u 3 )  k 2 (u 2  u1 )  m2ug
m1
u1 m1u1  k 2 (u1  u 2 )  k1u1  m1ug

m1 0 0   u1  k1  k 2   k2 0   u1   m1 

k1 0 m 0  u 2     k 2 k 2  k 3   k 3  u 2    m 2  u g
  
 2

 0 0 m3  u3   0  k3 k 3   u 3   m3 

2
 For a N-DOF system, Ahmed Elgamal

m1 
 m2 
 
  
 
  
  
m 
 


 mm 
 
  
  
 
 m n 

k1  k 2  - k2 0       0 
 k k 2  k 3   k3 0      0 
 2

 0  
 
 0  
 |  
k 
 | 


 0 0 - km k m  k m1  - k m 1 0 
 
 |  0 
 |   kn 
 
 0 0      0  kn k n  5

Equation of motion for a N-DOF system, Ahmed Elgamal

m1   u1  k1  k 2  - k2 0       0   u1   m1 

 m2   u   k k 2  k 3   k3 0      0   u 2  m 
  2   2  2
   |   0   |   | 
       
   |   0   |   | 
   |   |   |   | 
            u g
   |   |   |   | 
 mk  u   0 0 - km k m  k m1  - k m 1 0  u  m 
 m
  m     m

   |   |  0  |   | 
   |   |   kn |    | 
       
 m n   u n   0 0      0  kn k n   u n   m n 

or (with damping included):

  cu  ku  m1u g
mu

with initial conditions: u  u(t=0)

u  u (t=0) 6

3
Natural Frequencies of a N-DOF system

Similar to the SDOF system, MDOF systems have natural frequencies. A 2-DOF
has 2 natural frequencies w1 and w2, and a n-DOF system has natural frequencies
w1 , w2 , …, wn
Similar to the SDOF, free vibration involves the system response in its
natural frequencies. The corresponding Free Vibration Equation is (with
no damping):

  ku  0
mu
In free vibration, the system will oscillate in a steady-state harmonic
fashion, such that:
  ω2u
u

e.g. u  a  sinωt   b  cosωt  gives u  -ω2 u

7

Ahmed Elgamal

 , we get:
substituting for u

- ω m  k u  0
2

or
k - ω mu  0
2 Equation 1

The above equation represents a classic problem in

Math/Physics, known as the Eigen-value problem.
The trivial solution of this problem is u = 0 (i.e., nothing is
happening, and the system is at rest).
8

4
 Ahmed Elgamal

For a non-trivial solution (which will allow for computing the

natural frequencies during free vibration), the determinant of
k - ω2m must be equal to zero such that:
k - ω2m  0 or k - λm  0 where λ  ω2

For a 2-DOF system for instance (see next page), the

above determinant calculation will result in a quadratic
equation in the unknown term l . If this quadratic equation
is solved (by hand), two roots are found (l1 and l2), which
define w1 and w2 (the natural resonant frequencies of this
2-DOF system).
9

Ahmed Elgamal
2-Story Shear Building (2-DOF system)
m2
m1 0   u1  k1  k 2   k 2   u1  m  u2
 0 m  u     k       1 u g
 2  2   2 k 2  u 2  m 2 

k - lm u  0 k2
k1  k 2   lm1  k 2   u1  0
     0
  k 2 k 2  lm 2   u 2    m1
u1
Set Determinant = 0:
k1  k 2   lm1  k2
0
 k2 k 2  lm 2
k1
or, (k1  k 2   lm1 )(k 2  lm2 )  (k 2 )(k 2 )  0

(m1m2 )l2  ((m1k 2  m2 (k1  k 2 ))l  (k1k 2 )  0

Solve for the l1 and l2 using the standard approach W X

Note: For Y Z 
 b  b  4 ac
2 
al2  bl  c  0 l1 
2 2a Determinant = WZ-XY
10

5
 Ahmed Elgamal

For a general N-DOF system:

Matlab or similar computer program can be used to solve the

determinant equation (of order equal to the NDOF system), defining
NDOF roots or NDOF natural frequencies

w1 , w2 ,…, wNDOF

Note:
These resonant (natural) frequencies w1 , w2 , … are conventionally
ordered lowest to highest (e.g., w1 = 8 radians, w2 = 14 radians, and
so forth).

11

Ahmed Elgamal

Mode Shapes
Steady State vibration at any of the resonant frequencies wn
takes place in the form of a special oscillatory shape, know
as the corresponding mode shape fn

To define these mode shapes (one for each identified wn), go

ahead and substitute the value of wn for w in Eq. 1
k - ω mu  0
2

and solve for the vector u which will define the

corresponding mode shape fn :
k - ω m f
2
n n 0
12

6
2-DOF system ( 2 mode shapes f1 and f2) Ahmed Elgamal

m2
φ 21
m2
f22 φ 21
u2 u2
f 
f1   11 
f21 
m1
φ11
m1
f12 φ11
u1 u1
f 
f 2   12 
f22 
f1 f2
Note: Any mode shape fn only defines relative amplitudes of motion of the
different degrees of freedom in the MDOF system. For instance, if you are solving
a 2-DOF system, you might end up with something like (when solving for the first
mode):
f11 - 2f21 = 0, only defining a ratio between amplitudes of f11 and f21
(for instance, if f11 = 1, then f21 = 0.5, or if you choose f11 = 2, then f21 = 1, and so
forth).
Generally, go ahead and make fmn= 1 (where m is top floor Dof and n is mode
shape number) and solve for the other degrees of freedom in the vector fn 13

Ahmed Elgamal
Note: When you substitute any of the wn values into Eq. 1, the determinant of the matrix (k-
wn2m) automatically becomes = 0, since this wn is a root of the determinant equation (i.e.,
the matrix becomes singular).
The determinant being zero is a necessary condition for obtaining a vector u (the mode shape
fn) that is not equal to zero (i.e., a solution other than the trivial solution of u = 0.

Sample Mode shape Configurations

2-Story shear building
(one node in mode 2)
4-story shear building (4-DOF system)
node Note one node in mode 2, two in mode 3, and 3 in mode 4

3-Story Shear Building

node

node
node

14

7
Properties of fn Ahmed Elgamal

a) Mode shapes are orthogonal such that (for any nr)

φ Tn kφ r  φ Tn mφ r  0 (not φ Tn φ r  0 )
b) For any mode fn , modal mass Mn is defined by:

φ Tn mφ n  M n

c) For any mode fn , modal stiffness Kn is defined by:

φ Tn kφ n  K n  ω2n M n

d) If φ Tn mφ n  1.0 then φ Tn kφ n  ω 2n
1
To do that, multiply each component of mode fn by Mn 15

Ahmed Elgamal

Solution of by Mode superposition

Example of a 2-DOF system ( 2 mode shapes φ 1 and φ 2 )

φ 21 f22
m2 u2

φ11 f12
m1 u1

16

8
 Ahmed Elgamal
Modal Analysis (Solution of MDOF equation of motion by Mode Superposition)
The solution u will be represented by a summation of the mode shapes fn, each
multiplied by a scaling factor qn (known as the generalized coordinate) . For instance,
for the 2-DOF system:
 u1  φ11  φ12  φ11 φ12   q1 
u = u  
   1 φ  2 φ
q q      φ1 φ 2 q  Φq
 2  φ 21   22   21 φ 22  q 2 
In the above, F is known as the modal matrix. As such, changes in the displaced shape
of the structure u with time will be captured by the time histories of the vector q

Substitutingu  Φq in the equation of motion mu

  ku  m1u g
  kΦq  m1u g
Results in mΦq
To benefit from the mode orthogonality property, multiply by FT to get:

  ΦTkΦq  ΦTm1ug
ΦTmΦq

φ1T mφ1 φ1T mφ 2   q1  φ1T kφ1 φ1T kφ 2   q1  φ1T m1

or  T           T u g
φ 2 mφ1 φ T2 mφ 2  q 2  φ T2 kφ1 φ T2 kφ 2  q 2  φ 2 m1
17

Ahmed Elgamal

Due to the orthogonality property of mode shapes (see previous slide), the
matrix equation becomes un-coupled and we get:
φ 21 f22 φ 21
M1q1  M1ω12 q1  L1u g m2 u2 m2 u2

M 2 q 2  M 2 ω 22 q 2  L 2 u g
or, φ11 f12 φ11
m1 u1 m1 u1
L1
q1  ω12 q 1   u g
M1

q 2  ω 22 q 2  
L2
u g
f1 f2
M2
NDOF NDOF
For a diagonal mass matrix: Mi   m jφ 2ji
j1
Li  m φ j ji
j1
The terms L1/M1 and L2/M2 are known as modal participation factors. These terms
control the influence of on the modal response. You may notice that (if both modes
are normalized to 1.0 aturoof
g level for example) L1/M1 > L2/M2 since f11 and f21 are
of the same sign while f12 and f22 are of opposite signs. Therefore, the first mode is
likely to play a more prominent role in the overall response (frequency content of the
input ground motion also affects this issue).
18

9
 Ahmed Elgamal

Note that the original coupled matrix Eq. of motion, has now become
a set of un-coupled equations. You can solve each one separately (as
a SDOF system), and compute histories of q1 and q2 and their time
derivatives. To compute the system response, plug the q vector back
into Equation 2 and get the u vector

u  Φq
(the same for the time derivatives to get relative velocity and
acceleration).

The beauty here is that there is no matrix operations involved, since

the matrix equation of motion has become a set of un-coupled
equation, each including only one generalized coordinate qn.

19

Ahmed Elgamal

Damping in a Modal Solution

Now, you can add any modal damping you wish (which is
another big plus, since you control the damping in each mode
individually). If you choose  i = 0.02 or 0.05, the equations
become:
Li
q i  2 i ωi q i  ωi2 q i   u g , i = 1, 2, … NDOF
Mi

OK, go ahead now and solve for qi(t) in the above uncoupled equations
(using a SDOF-type program), and the final solution is obtained from:
u  Φq
u  Φq
  Φq
u 
 t  u
u   1u g
20

10
 Ahmed Elgamal
Modal Analysis (3-DOF system)
The solution u will be represented by a summation of the mode shapes fn, each
multiplied by a scaling factor qn (known as the generalized coordinate) . For instance,
for the 3-DOF system:

 u1  φ11  φ12  φ13   φ11 φ12 φ13   q1 

 u       
u  2  φ 21  q1  φ 22  q 2  φ 23  q 3  φ 21 φ 22 φ 23  q 2   φ1 φ 2 φ 3 q  Φq
   
u 3  φ 31  φ 32  φ 33  φ 31 φ 32 φ 33  q 3 
In the above, F is known as the modal matrix. As such, changes in the displaced shape
of the structure u with time will be captured by the time histories of the vector q
Note: If a two mode solution is sought, the system above becomes:
 u1  φ11  φ12   φ11 φ12 
q 
u  u 2  φ 21 q1  φ 22  q 2  φ 21 φ 22   1   φ1 φ 2 q  Φq
   
    q 2 
u 3  φ 31  φ 32  φ 31 φ 32 
Note: If a single (1st or fundamental) mode solution is sought, the system above
becomes:
 u1  φ11 
u  u 2   φ 21  q1  φ1 q1
  21
u 3  φ 31 

Ahmed Elgamal

Multi-Degree-Of-Freedom (MDOF) Response Spectrum Procedure

1. Once you have generalized coordinates and uncoupled equations, use response
spectrum to get maximum values of response (ri)max for each mode separately.

rmax   r   root sum

2
Calculate expected max response ( r ) using i max
square formula

where i = 1, 2, … N degrees of freedom of interest (maybe first 4 modes at most) and

r is any quantity of interest such as |umax| or SD

(note that summing the maxima from each mode directly is typically too
conservative and is therefore not popular; because the maxima occur at different
time instants during the earthquake excitation phase)

See A. Chopra “Dynamics of Structures” for improved formulae to estimate rmax.

22

11
 Ahmed Elgamal

Response Spectrum Modal Responses

Max relative displacement |un| or |ujn| (jth floor, nth mode)

Ln
u jn  Sd φ jn (S is S evaluated at frequency ω n or period T )
Mn n dn d n

Estimate of maximum floor displacement

M
uj  u
n 1
2
jn
(M = number of modes of interest)

Maximum Equivalent static force fn or fjn (jth floor, nth mode)

Ln
f jn  Sa m jφ jn
Mn n
23

Ahmed Elgamal

Therefore, modal base shear V0n and moment M0n

# of floors
N N
V0n   f jn M 0n   f jn d j
j1 j1

base
where dj = Distance from floor j to base

Estimate of maximum base shear and moment:

M M
V0   V0n2
n 1
M0  M 2
0n
n 1

24

12
 Ahmed Elgamal

Damping Matrix for MDOF Systems

  cu  ku  m1ug
mu
Mass-proportional damping
c = aom
Stiffness-proportional damping
c = a1k
Classical damping (Rayleigh damping)
c  a 0 m  a 1k

Stiffness proportional damping appeals to intuition

because it generates damping based on story
deformations. However, mass proportional damping may
be needed as will be shown below.
25

Mass-proportional damping: c = ao m Ahmed Elgamal

Defining a0 to obtain a desired modal damping zn in mode n

In any modal equation, we have
M n q n  C n q n  K n q n  0

where, Cn  2ζ nωn Mn

Therefore, ao can be specified to obtain any desired zn for

a given mode n such that Cn = a0 Mn

2ζ n ωn M n  a 0 M n or a 0  2ζ n ωn

(e.g. at w1 = 2 radians/s, z1 = .05)  find a0

26

13
 Ahmed Elgamal

With a0 defined by a0 = 2 zn wn, this form of mass proportional

damping will change with frequency according to z = a0 / 2w as
shown in the figure below.

z
ao
z 
2ω

c  a om

ω1 ω 2 ω 3 ω 4 ωn
27

Stiffness-proportional damping: c = a1 k Ahmed Elgamal

Defining a1 to obtain a desired modal damping zn

In any modal equation, we have
M n q n  C n q n  K n q n  0

where, Cn  2ζ nωn Mn and K n  ω n M n

2

Therefore, ao can be specified to obtain any desired n for

a given mode n such that Cn = a1Kn , or:

2ζ n ωn M n  a1ωn2 M n or a1  2ζ n /ωn

(e.g. at w1 = 2 radians/s, z1 = .05)  find a1

28

14
 Ahmed Elgamal

With a1 defined by a1 = 2 zn / wn , this form of stiffness proportional

damping will change with frequency according to z = a1w / 2 as shown
in the figure below (damping increases linearly with frequency.

z c  a 1k

a 1ω
z 
2

ω1 ω 2 ω3 ω
29

Ahmed Elgamal

Physically, we often observe (in first approximation) a nearly equal

value of damping for the first few modes of structural response
(e.g., first 1- 4 modes or so), and we want to model that. Therefore,
we use (Classical or Rayleigh damping):
c  a 0 m  a 1k
2ζ n ωn Mn  a 0 Mn  a1ωn2 Mn
ζ n  (a 0  a1ωn2 ) / 2ωn
ζ n  (a 0 / 2ωn )  (a1ωn / 2)

Now we choose damping ratios zi and zj for two modes (natural

frequencies wi and wj) and solve for the coefficients a0 and a1 (two
equations in two unknowns).

30

15
Variation of Classical (Rayleigh) Damping with Frequency Ahmed Elgamal

Damping defined by z = (a0/2w)+(a1w/2) results in the variation shown by the

combined curve below, which has the desirable feature of being somewhat uniform
within a frequency range of interest (say 1 Hz to 7 Hz or 2 to 14 in radians/s).
z

Combined
c  a 0 m  a 1k

Frequency range Stiffness-proportional damping

of interest
c  a1k

nearly uniform damping

Mass-proportional damping
c  a 0m
ωi ωj ω
w1 for example 2nd or , 3rd resonance for example 31

Ahmed Elgamal

Notes

1) For a choice of zi = zj = z same damping ratio in the

two modes, we get

2ωiω j 2
, a0  ζ a1  ζ
ωi  ω j ωi  ω j

2) Classical damping and is attractive because of

combination of mass and stiffness, allowing the no-
damping free-vibration mode shapes to un-couple the
matrix equation of motion.
32

16
 Ahmed Elgamal

Caughey damping

The above procedure was generalized by Caughey to allow for more

control over damping in the specified modes of interest (i.e. to be
able to specify z for more than 2 modes i and j)

In this generalization, you can stay within the scope of classical

damping by using

 
N 1
c  m a i m 1k
i

i 0

to find a i coefficients to match zi modal damping ratios, see for

instance “Dynamics of Structures” by A. Chopra.

33

Ahmed Elgamal

Disadvantages:

1. c can become a full matrix instead of being a banded

matrix (if m and k are banded) as with c = a0m + a1k

2. You must check to ensure that you don’t end up with a

negative zi in some mode where you have not specifically
specified damping (because damping variation with
frequency might display sharp oscillations).

In summary, c = a0m + a1k is the usual choice at present

despite the limitations discussed above.

34

17