Sie sind auf Seite 1von 45

ME617 - Handout 7

(Undamped) Modal Analysis of


MDOF Systems
The governing equations of motion for a n-DOF linear
mechanical system with viscous damping are:

 + D U
 +K U =F
MU
(t )
(t )

(1)

 and U
 are the vectors of generalized displacement,
where U, U,
velocity and acceleration, respectively; and F( t ) is the vector of
generalized (external forces) acting on the system.
M, D,K represent the matrices of inertia, viscous damping and
stiffness coefficients, respectively1.
The solution of Eq. (1) is uniquely determined once initial
conditions are specified. That is,

 =U

at t = 0 U (0) = U o , U
(0)
o

(2)

In most cases, i.e. conservative systems, the inertia and stiffness


T
T
matrices are SYMMETRIC, i.e. M = M , K = K . The kinetic
energy (T) and potential energy (V) in a conservative system are

1 T
 , V = 1 UT K U
T= U
MU
2
2

(3)

The matrices are square with n-rows = n columns, while the vectors are nrows.
1

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

In addition, since T > 0, then M is a positive definite matrix2. If V


>0, then K is a positive definite matrix. V=0 denotes the existence
of a rigid body mode, and makes K a semi-positive matrix.
In MDOF systems, a natural state implies a certain
configuration of shape taken by the system during motion.
Moreover a MDOF system does not possess only ONE natural
state but a finite number of states known as natural modes of
vibration. Depending on the initial conditions or external forcing
excitation, the system can vibrate in any of these modes or a
combination of them. To each mode corresponds a unique
frequency knows as a natural frequency. There are as many
natural frequencies as natural modes.
The modeling of a n-DOF mechanical system leads to a set of ncoupled 2nd order ODEs, Hence the motion in the direction of one
DOF, say k, depends on or it is coupled to the motion in the other
degrees of freedom, j=1,2n.
In the analysis below, for a proper choice of generalized
coordinates, known as principal or natural coordinates, the
system of n-ODE describing the system motion is independent of
each other, i.e. uncoupled. The natural coordinates are linear
combinations of the (actual) physical coordinates, and conversely.
Hence, the motion in physical coordinates can be construed or
interpreted as the superposition or combination of the motions in
each natural coordinate.

Positive definite means that the determinant of the matrix is greater than
zero. More importantly, it also means that all the matrix eigenvalues will be
positive. A semi-positive matrix has a zero determinant, with at least an
eigenvalues equaling zero.
2

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

For simplicity, begin the analysis of the system by neglecting


damping, D=0. Hence, Eq.(1) reduces to

 + K U = F
MU
(t )
(t )

(4)

 =U

and at t = 0 U (0) = U o , U
(0)
o
Presently, set the external force F=0, and lets find the free
vibrations response of the system.

 + K U = 0
MU

(5)

The solution to the homogenous Eq. (5) is simply

U = cos( t )

(6)

which denotes a periodic response with a typical frequency .


From Eq. (6),
 = 2 cos( t )
U
(7)
Note that Eq. (6) is a simplification of the more general solution

U = e s t with s = i and where i = 1

(8)

Substitution of Eqs. (6) and (7) into the EOM (5) gives:

 + K U = 0
MU

M 2 cos( t ) + K cos( t ) = 0
M 2 + K cos( t ) = 0
and since cos( t ) 0 for most times, then

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

or

M 2 + K = 0

(9)

2M = K

(10)

Eq. (10) is usually referred as the standard eigenvalue


problem (mathematical jargon):

A =
where =M K and =
1

(11)

Eq.(9) is a set of n-homogenous algebraic equations. A nontrivial


solution, 0 exists if and only if the determinant of the
system of equations is zero, i.e.

= M 2 + K =0

(12)

Eq. (12) is known as the characteristic equation of the system.


2
It is a polynomial in = , i.e.

= 0 = a0 + a1 2 + a2 4 + a3 6 + .... an n
n

= 0 = a0 + ( ai
i =1

(13)

This polynomial or characteristic equations has n-roots, i.e. the


set {k }k =1,2,....n or { k }k =1,2,....n since = .
The s are known as the natural frequencies of the system. In
the MATH jargon, the s are known as the eigenvalues (of matrix
A)

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

Knowledge summary
a)

A n-DOF system has n-natural frequencies.

b)

If M and K are positive definite, then


0 < 1 2 .......n 1 n .

c)

If K is semi-positive definite, then


0 = 1 2 .......n 1 n , i.e. at least one natural frequency
is zero, i.e. motion with infinite period. This is known as
rigid body mode.

Note that each of the natural frequencies satisfies Eq. (9).


Hence, associated to each natural frequency (or eigenvalues)
there is a corresponding natural mode vector (eigenvector) such
that

[ M i + K ] (i )

= 0,

i =1,...n

(14)

The n-elements of an eigenvector are real numbers (for


undamped system), with all entries defined except for a constant.
The eigenvectors are unique in the sense that the ratio between two
elements is constant, i.e.

( k ) j

= constant for any j , i = 1,....n

( k )i

The actual value of the elements in the vector is entirely


arbitrary. Since Eq. (14) is homogenous, if is a solution, so it is
for any arbitrary constant . Hence, one can say that the
SHAPE of a natural mode is UNIQUE but not its amplitude.
For MDOF systems with a large number of degrees of freedom,
n>>3, the eigenvalue problem, Eq. (11), is solved numerically.
MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

Nowadays, PCs and mathematical computation software allow,


with a single (simple) command, the evaluation of all (or some)
eigenvalues and its corresponding eigenvectors in real time, even
for systems with thousands of DOFs.
Long gone are the days when the graduate student or practicing
engineer had to develop his/her own efficient computational
routines to calculate eigenvalues. Handout # 9 discusses briefly
some of the most popular numerical methods to solve the
eigenvalue problem.
A this time, however, lets assume the set of eigenpairs
is known.
i , ( i )

i =1,2...n

Properties of natural modes


The natural modes (or eigenvectors) satisfy important
orthogonality properties. Recall that each eigenpair
satisfies the equation
i , ( i )

i =1,2...n

M i2 + K ( i ) = 0, i =1,...n .

(15)

Consider two different modes, say mode-j and mode-k, each


satisfying

2j M ( j ) = K ( j ) and k2 M ( k ) = K ( k )
T

(16)

Pre-multiply the equations above by ( k ) and ( j ) to obtain

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

2j T( k ) M ( j ) = T( k ) K ( j )
and

(17)

k2 T( j ) M ( k ) = T( j ) K ( k )
Now, perform some matrix manipulations. The products
M and TK are scalars, i.e. not a matrix nor a vector. The
transpose of a scalar is the number itself. Hence,
T

( T( j )K ( k ) ) = ( K ( k ) ) ( T( j ) )
T

= T( k ) K T ( j )
= T( k ) K ( j ) since K = K T
and

T
( j)

M ( k ) ) = T( k ) M ( j ) since M = MT
T

for symmetric systems. Thus, Eqs. (17) are rewritten as

2j T( j ) M ( k ) = T( j ) K ( k ) (a )
and

(18)

k2 T( j ) M ( k ) = T( j ) K ( k ) (b)
Subtract (b) from (a) above to obtain

2
j

k2 ) T( j ) M ( k ) = 0

(19)

if j k , i.e. for TWO different natural frequencies; then it


follows that

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

for j k

T( j ) M ( k ) = 0 and T( j ) K ( k ) = 0

(20)

for j = k T( j ) M ( j ) = M j and T( j ) K ( j ) = K j = 2j M j (20)


where Kj and Mj are known as the j-modal stiffness and j-modal
mass, respectively.
Define a modal matrix has as its columns each of the
eigenvectors, i.e.

= [1 2 .. n ]

(21)

and the modal properties are written as

T M = [ M ] ; T K = [ K ]

(22)

where [M] and [K] are diagonal matrices containing the modal
mass and stiffnesses, respectively.
The eigenvector set

k=1,n is linearly independent. Hence,

any vector (v) in n-dimensional space can be described as a linear


combination of the natural modes, i.e.
n

v = a j ( j ) = a
j =1

v = 1 a1 + 2 a2 + .. + n an = [1 2

(23)

a1
a
.. n ] 2 = a
..

an

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

System Response in Modal Coordinates


The orthogonality property of the natural modes (eigenvectors)
permits the simplification of the analysis for prediction of system
response. Recall that the equations of motion for the undamped
system are
 + K U = F
MU
(4)
(t )
(t )

 =U

and at t = 0 U (0) = U 0 , U
(0)
0
Consider the modal transformation

U(t ) = q(t )

(24)3

 = q
( t ) , then EOM (4) becomes:
And with U
(t )

 + K q = F( t )
M q
which offers no advantage in the analysis. However, premultiply
T
the equation above by to obtain

( M ) q + ( K ) q = F
T

(t )

(25)

and using the properties of the natural modes,


T M = [ M ]; T K = [ K ] , then Eq. (25) becomes
T

M
q
+
K
q
=
Q
=

F( t )
[ ] [ ]

(26)

Eq. (24) sets the physical displacements U as a function of the


modal coordinates q. This transformation merely uses the property
of linear independence of the natural modes.
3

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

And since [M] and [K] are diagonal matrices. Eq. (26) is just a set
of n-uncoupled ODEs. That is,

M 1 q1 + K1 q1 = Q1
M 2 q2 + K 2 q2 = Q2

(27)

.....
M n qn + K n qn = Qn
Or

M j qj + K j q j = Q j with n j =

Kj

Mj

j =1,2...n

(28)

The set of qs are known as modal or natural


coordinates (canonical or principal, too). The vector

Q = T F( t ) is known as the modal force vector.


Thus, the major advantage of the modal transformation (24)
is that in modal space the EOMS are uncoupled. Each equation
describes a mode as a SDOF system.
The unique solution of Eqs. (28) needs of initial
specified in modal space, i.e. {q o , q o } .
Using the modal transformation,
follows

conditions

 = q
U o = q o ; U
o
o


q o = 1U o ; q o = 1 U
o

, it

(28)

However, Eq. (28) requires of the inverse of modal matrix ,


1
i.e. = I . For systems with a large number of DOF, n>> 1,
1
finding the matrix is computationally expensive.

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

10

A more efficient to determine the initial state {q o , q o } in modal


coordinates follows. Start with the fundamental transformation,

U o = q o , and premultiply this relationship by T M

to

obtain,

T M U o = T M q o
= [ M ]qo ,

since [ M ] = M , hence
T

q o = [ M ] T M U o ,
1

q o = [ M ]

(29a)


MU
o
T

or

qok =

1 T
1 T
 )
( k ) ( M U o ) , qok =
( k ) ( M U
o
Mk
Mk

(29b)

Eqs. (29) are much easier to calculate efficiently when n-DOF is


large. Note that finding the inverse of the modal mass matrix [M]-1
is trivial, since this matrix is diagonal.
Comparing eqs. (28) and (29a) it follows that

1= [ M ] T M
1

The solution of ODEs

M j qj + K j q j = Q j

(30)
with initial

conditions qo j , qo j follows an identical procedure as in the


solution of the SDOF response. That is, each modal response adds
the homogeneous solution and the particular solution. The
particular solution clearly depends on the time form of the modal
MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

11

force Q(t), i.e step-load, ramp-load, pulse-load, periodic load, or


arbitrary time form.

Free response in modal coordinates


Without modal forces, Q=0, the modal equations are

M j q H j + K j qH j = 0 = Q j
with solutions, for

an elastic mode

( )

qH j = qo j cos n j t +
; and for a rigid

(31a)

qo j

( )

sin n j t

if n j 0

(31b)

body mode

qH j = qo j + qo j t

if n j = 0

(31c)
j=1,2,.n

Forced response in modal coordinates


For step-loads, QS j , the modal equations are
M j qj + K j q j = QS j
and; for an

(32a)

elastic mode, n 0 ,
j

( )

q j = qo j cos n j t +

qo j

( )

sin n j t +

QS j

( )

1 cos t
nj

Kj

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

(32a)

12

; and for a rigid body mode, n j = 0 ,

1 QS j 2
t
2Mj

q j = qo j + qo j t +

(32c)
j=1,2,.n

For periodic loads,, the modal equations are


M j qj + K j q j = QPj cos(t )
with solutions
for an elastic

mode, n 0 , and n
j

QPj
1
q j = C j cos n j t + S j sin n j t +
K j 1
nj

( )

( )

(33a)

cos ( t )
2

(33b)
Note that if = n j , a resonance appears that will lead to system
destruction.
For a rigid body mode, n j = 0 ,

QPj

q j = qo j + qo j t

M j2

For arbitrary-loads Q j
qj = qjo cos(nj t) +
(34)
for an

q jo

sin(nj t) +

cos(t )

(33c)

, the modal response is

1
M jnj

Q
0

j( )

sin nj (t ) d

elastic mode, n 0 .
j

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

13

System Response in Physical Coordinates


Once the response in modal coordinates is fully determined, the
system response in physical coordinates follows using the modal
transformation

U(t ) = q(t )
U ( t ) = [1 2

q1( t )

q
.. n ] 2 = 1 q1 + 2 q2 + .. + n qn
..

qn
n

U ( t ) = j q j( t )
j =1

(35)

One important question follows: are all the modal

responses important and need be accounted for to


obtain the response in physical coordinates? If not,
savings in computation time are evident. Hence, the physical
response becomes
m

U ( t ) j q j( t ) , m < n
j =1

(36)

If m<n, then how many modes are to be included to ensure the


physical response is accurate? That is, which modes are important
and which others are not?

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

14

Example: Consider the case of force excitation with frequency


n j and acting for very long times. The EOMs in physical

space are

 + K U = F cos ( t )
MU
P

Lets assume there is a little damping; hence, the steady state


periodic response in modal coordinates is (see eq. (33b)):

QPj
1
qj
K j 1
nj

cos ( t )
2

(37a)

And thus,

QPj

U = U P cos(t ) = q = j
Kj
j =1

1
nj

cos ( t )
2

(38)
The physical response is also periodic with same frequency as the
force excitation.
Recall that K j = n j M j = ( j ) K ( j ) and QPj = ( j ) FP
2

However, nowadays the engineer in a hurry prefers to dump the


problem into a super computer; and for U = U P cos(t ) , finds the
solution
1

U P = K M FP
2

(39)

at a fixed excitation frequency . Brute force substitutes beauty


and elegance, time savings in lieu of understanding!
MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

15

ORIGIN := 1

Example: Find natural frequencies and natural


mode shapes of UNDAMPED system.
Given EOMs for a 2DOF - undamped- system:

2K2 X2
M2 0 d2 X2 2 K2
=

+
0 M1 dt2 X1 2 K2 2 K2 + K1 X1



where M2 =mo, M1 =5 m o , K2=k o; K1 =5 ko

mo
0

2 ko
X
d2 2 +
5 mo dt2 X1 2 ko

0
K Z
1

2ko

X2

=
2 ko + 5 ko X1

(1)

0
K Z
1

(a) PROCEDURE TO FIND NATURAL FREQUENCIES AND NATURAL MODES: Assume the
motions are periodic with frequency , ie

X2 = a1 cos ( t)

X1 = a2 cos ( t)

(2)

Set the RHS of Eq. (1) equal to 0. Substitution of (2) into (1) gives

2 ko mo 2

2 ko

a1
cos ( t) = 0

2 a
0
2
7 ko 5 mo

2ko

cancel cos(t) since it


is NOT zero for all times

The homogeneous system of eqns

2 ko mo 2

2 ko

a1
= 0

2 a
2 0
7 ko 5 mo

2ko

(3)

has a non-trivial solution if the determinant of the system of equations equals zero, i.e. if

)(

( ) = 7 ko 5 mo 2 ko mo
Let

=
2

) 4 ko

=0

, and expanding the products in the determinant

0 = 5mo 7 ko mo + 10 ko mo + 14 ko 4 ko
Let

mo
=

ko

Leads to:

( ) 2 + b

+ c

0 = a
with:

a := 5

The roots (eigenvalues) of the characteristic equation are

b := 17

(4)

c := 10

b b2 4 a c
1 :=
2 a

) 0.5

b + b2 4 a c
2 :=
2 a

) 0.5

ko
m
o

0.757
=

2.643

and the natural frequencies are:

1 := ( 1)

2 := ( 2)

0.5

0.5

Find the eigenvectors:

ko

=

1.626 mo
0.87

0.5

The two equations in (3) are linearly dependent. Thus, one cannot solve for a1 and a2. Set

1 := 1

arbitrarily; and from the first equation


for 1

2 ko mo 1 )
(
(2 ko 0.757 ko)
=
=
2

2 ko

2 :=

2 ko

( 2 0.757)
2

2 = 0.621

1 :=

0.621

is the first eigenvector (natural mode)

1 =
(b)

Explanation: DOF1 (X2) and DOF2 (X1) move in phase, with X2>X1

for

2
2

2 :=
(b)

1 := 1

2 ko mo 2 )
(
(2 ko 2.643 ko)
=
=
2

2 ko

0.321

2 =

2 :=

2 ko

( 2 2.643)
2

is the 2nd eigenvector (natural mode)

Explanation:DOF1 (X2) and DOF2 (X1) move 180 deg OUT of phase, with |X2|>|X1|

(c) find the numerical value for each natural frequency:


Since

0.87 ko

:=

1.626 mo

mo :=

0.5

170.947 rad

319.495 sec

=
fn :=

27.207 Hz
fn =

50.849

1000lb

ko := 105

lb
in

0
Note that mass must be
expressed in physical units
consistent with the problem, i.e.

lb sec2
mo = 2.59
in

Perform same work using a calculator


1
0

M :=

Let

mo

Z := M

:= sort ( eigenvals ( Z) )

natural modes:

2 2 k
o
2 7

K :=

Use BUILT IN functions


Not much learning

1 := ( 1)

.5

2 := ( 2)

.5

1 := eigenvec ( Z , 1)

2 := eigenvec ( Z , 2)

2.921 104 1
=
2
5
1.021

10

sec
170.914 rad

319.495 sec

0.849

0.528

1 =

0.952

0.306

2 =

27.202 Hz

50.849

( 1) 2
= 0.622
( 1) 1
( 2) 2
= 0.322
( 2) 1

which are the same ratios as for


the vectors found earlier

Example: Undamped Modal Analysis

mo :=

1000lb

5 lb

k o := 10

Equations of motion:

n :=

sec

M :=

mo 0
0 5 mo

K :=

2 ko 2k o
2 ko 7 k o

170.95 rad

319.5 sec

F :=

3
7.584
8.534 10 lb sec2

in
3
3.925
8.534 10

:=

ORIGIN := 1

1
1

0.621 0.321
Fo := k o Zo

provides a

0 lb
F
o

(a) FIND modal masses and stiffnesses MM := T M

modal matrix (eigenvectors)

Zo := 0.01 in

given:
Define matrices:

MM =

in

natural frequencies,

mo 0 d 2 X2 2 ko 2k o X2 0

=
0 5 mo dt2 X1 2 ko 7 k o X1 k o Z

ft

g = 32.174

constant force

at t=0s, Initial conditions:


system is at REST
T

KM := K

non-diagonal elements are very small= non zero b/c of


roundoff in numerical calculator

modal masses and stiffnesses:

Mode 1

M m := M M
1
1, 1

Mode 2

M m := M M
2
2, 2

Km := n M M
1
1, 1
1
2

Km := n M M
2
2, 2
2

2.216 105 lb

Km =

5 in
4.006 10
(b) Find initial moddal displacements and velocities and modal force vector (Q)
2
7.584 lb sec
Mm =

3.925 in

At time t=0s, the system is at REST at its static equilibrium position, hence the initial conditions are null
displacements and null velocities. Of course, the same applies to modal space, i.e. null initial displacements
and velocities
X1
0
0 ft
for generality, define:
Calculate inverse of A matrix
1
Xo := ft
Vo :=
X
2
inv :=
0
0 sec
q o := inv Xo

and in modal coordinates


(disp & velocities)

qo =
Define modal force

q o_dot := inv Vo

0 ft

0

q o_dot =

Q := F
(c)

Q=

Modal EOMs and modal responses

621 lb

321

where

m =

2.802 10 3

in

4
8.013 10

Both natural modes will be excited

Q
m :=
1

No need for actual calculation


- a knowledge statement suffices

Mm

Using the cheat sheet, and since the Initial conditions


are null, the response in modal coordinates are

0 ft

0 sec

d2
q + Km q = Q
i
2 i
i
i i
dt

The EOMs in modal space are uncoupled


and equal to

q 1 ( t) := m 1 cos n t
1
1

velocity

m :=

Km
1

q 2 ( t) := m 1 cos n t
2
2

i = 1,2

Km
2
where:

n =

are the "static" deflections in modal space. 2 << 1, thus first


modal response is MORE important

170.95 rad

319.5 sec

(d) The response in physical coordinates, X1 and X2, equals (from transformation x=Aq)
with

X1 ( t) := q 1 ( t) + q 2 ( t)

X1 ( t) = m 1 cos n t + m 1 cos n t
1
1
2
2

X2 ( t) := 0.621 q 1 ( t) 0.321 q 2 ( t)

for graph below:

2
Tlarge := 10
n1

X2 ( t) = m 0.621 1 cos n t + m ( 0.321 ) 1 cos n t


1
1
2
2

m 0.621 = 1.74 10
1

1
1

0.621 0.321

m ( 0.321 ) = 2.572 10

in

in

Explanation: Since q1 and q2 are non-zero, then physical motion, X1 &X2, shows excitation of the
TWO fundamental modes of vibration - BUT response for second mode is much less
GRAPHs not needed for exam:

Note that there is no damping or


attenuation of motions.

0.006
displacements (inch)

0.005

Not too complicated physical response. It


shows dominance of first mode (lowest
natural freq or largest period)

0.004
0.003
0.002
0.001

0.001
0.002

= 0.037 sec

2
0

0.046

0.092

0.14

0.18
0.23
time (sec)

0.28

0.32

0.37

= 0.02 sec

X1
X2
terminal value

Terminal condition:
If damping is present and since the applied force is a constant, the system will achieve a new steady state
condition.
In the limit as t approaches very, very large values

And equations of motion reduce to:

X1 0
=
2
dt X2 0
d

; hence ===>

2 ko 2k o X1end 0

=
2 ko 7 k o X2end Fo

And solving this system of equations using Cramer's rule


Fo 2 k o
2 k o Fo
X1end :=
X2end :=

:= 14 k o 4 k o
3

X1end = 2 10

in

Fo = 1 10 lb

determinant of system of eqns.

X2end = 2 10

Note that the graph of undamped periodic motions Z(t) and X(t)
shows oscillatory motions abut these terminal or end values.

OR

X1end
=F
X2end

in

recall

2 10 3
in
K F =

3
2 10
1

Zo = 0.01 in
Zo
X1end

=5

0.006
displacements (inch)

COMPARE actual
response with a
response
neglecting q2.
Indeed mode 2 does
not afffect the physical
response, except for
motion X2 sligthly

0.005
0.004
0.003
0.002
0.001
0
0.001
0.002

0.12

0.25
time (sec)

X1
X2
terminal

0.37

Normalization of eigenvectors (natural modes)

Recall that the components of an eigenvector j are


ARBITRARY but for a multiplicative constant. If one of the
elements of the eigenvector is assigned a certain value, then this
vector becomes unique, since then n-1 remaining elements are
automatically adjusted to keep constant the ratio between any two
elements in the vector.
In practice, the eigenvectors are normalized. The resulting
vectors are called NORMAL MODES.
Some typical NORMS are

( )

L1 norm: q ( j ) =1= max q jk

(39a)

L2 norm: q ( j ) =1= q j1 + q j2 + .... + q jn


2

(39b)

Or making the mass modal matrix equal to the identity matrix,

[M]=I, i.e.
hence

T( j ) M ( j ) = M j = 1

(39c)

T( j )K ( j ) = K j = 2j M j = n 2j

(39d)

This normalization has obvious advantages since it will reduce


the number of operations when conducting the modal analysis.
However, the physical significance of the modal equations is lost.
Note that the modal Eqs. (26) become:

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

16

qj + n2j q j = Q j
Your lecturer recommends this normalization procedure be
conducted only for systems with large number of degrees of
freedom, n>>>1.
Note that the normalization process is a mere convenience,
devoid of any physical significance.

Rayleighs Energy Method


The method is a procedure to determine an approximate value
(from above) for the fundamental natural frequency of a MDOF
system. At times, the full solution of the eigenvalue problem is of
NO particular interest and an estimate of the system lowest natural
frequency suffices.

Recall that the pairs i , ( i )

i =1,2...n

satisfy K ( i ) = i2M ( i )

with properties T M = [ M ] ; T K = [ K ]
i.e. with modal stiffness and masses calculated from:

K i = T( i ) K ( i ) ; M i = T( i ) M ( i ) , and i2 =

That is,

i2 =

Ki

Mi

1
1

Ki

Mi

T( i ) K ( i )

T
2 ( i ) M ( i )

(41)

(42)

Above, the numerator relates to the potential or strain energy of


the system for the i-mode, and the denominator to the kinetic
energy for the same mode.

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

17

Consider an arbitrary vector u and define Rayleighs


quotient R(u) as
T
u
Ku
R (u) = 2 T
1 u Mu
2
1

(43)

R (u) is a scalar whose value depends not only on the matrices M


& K, but also on the choice of the vector u.
Clearly, if the arbitrary vector u coincides with (or is a multiple
of) one of the natural mode vectors, then Rayleighs quotient will
deliver the exact natural frequency for that particular mode. It can
also be shown that the quotient has a stationary value, i.e. a
minimum, in the neighborhood of the system natural modes
(eigenvectors). To show this, since u is an arbitrary vector and the
natural modes are a set of linearly independent vectors, then one
can represent
n

u = j c j = c

(44)

j =1

Where c = {c1 c2 .. cn } is the vector of coefficients in the


expansion. Substitution of the expression above into Rayleighs
quotient gives
T

R (u) =

cT ( T K ) c
2 ( c ) K ( c )
= T T
T
1 c
) M ( c ) c ( M ) c
2(
1

cT [ K ] c
R (u) = T
c [ M ]c

(45)

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

18

Assume the modes have been normalized with respect to the


mass matrix, i.e.
n

c c
=
R (u) =
cT Ic
2
n

c
2
i

i =1

2
ni

(46a)

c
i =1

2
i

Next, consider that the arbitrary vector u (which at this time can be
regarded as an assumed mode vector) differs very little from the
natural mode (eigenvector) ( r ) . This means that in the expansion

of vector u, the coefficients ci << cr ; for i = 1,2,...n and i r


Or

ci = i cr ; i <<1 for i = 1,2,...n and i r

Then, Rayleighs quotient is expressed as


n

c +c
2
r

2
nr

R (u) =

2
r

cr2 + cr2

R (u) =

{ }

i =1, i r
n

1+

i =1, i r

+
2
nr

i =1, i r
n

2
i

i =1, i r

2
i

2
ni

2
i

2
ni

2
i

1+
=

2
nr

i =1, i r

1+

ini

i =1, i r

nr

(46b)

2
i

The quantities i are small, of second order, hence R(u) differs


2

from the natural frequency by a small quantity of second order.

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

19

This implies that R(u)has a stationary value in the vicinity of the


modal vector ( r ) .
The most important property of Rayleighs quotient is that it
shows a minimum value in the neighborhood of the
fundamental mode, i.e. when r=1.
n

R (u) = =
2

2
n1

1+
i =2

ini
n

1 +
i=2

n1

, since

2
i

( ) > 1 (47)
ni

n1

Then each term in the numerator is greater than the corresponding


one in the denominator. Hence, it follows that

R(u) = 2 n21

(48)

i.e., Rayleighs quotient provides an upper bound to the first


(lowest) natural frequency of the undamped MDOF system.
Clearly, the equality holds above if one selects u = c1 (1) ; c1 0 .

Closure
Rayleighs energy method is generally used when one is
interested in a quick (but particularly accurate) estimate of the
fundamental natural frequency of a continuous system, and for
which a solution to the whole eigenvalue problem cannot be
readily obtained. The method is based on the fact that the natural
frequencies have stationary values in the neighborhood of the
natural modes.
In addition, Rayleighs quotient provides an upper bound to the
first (lowest) natural frequency. The engineering value of this
approximation can hardly be overstated. Rayleighs energy
method is the basis for the numerical computing of eigenvectors
and eigenvalues as will be seen later.
MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

20

N 5

number of DOFS

mass 1

ORIGIN 1

stiff 10

Example for estimation of first natural frequency

M identity ( N) mass

2
1

K stiff 0
0

2 1

dt
0

1 2 1 0
0 1 2 1

0 0 1 2

Prepared by
Lecturer Luis San
Andres for ME617
course

U1
K

K
M

U3

U2

U K U = 0

U4
K

U5
K

K
M

Estimate for first natural frequency - Use Rayleigh-Ritz method:


ASSUME

u ( .5 .75 1 .75 .5 ) 1

uT K u1 1
R ( u)
uT Mu1 1

calculate Rayleigh's quotient

n_ap

R ( u)

n_ap 16.903

Actual n 16.369
1

% difference =

Estimate

rad/s

1 n_ap 100 3.262


n

1.5

0.5
0
0

M1
u (approx)

Estimate for second natural frequency - Use Rayleigh-Ritz method:


ASSUME

u ( .5 .5 0 .5 .5 ) 1

uT K u1 1
R ( u)
uT Mu1 1

calculate Rayleigh's quotient

n_ap

R ( u)

n_ap 31.623

Actual n 31.623
2
% difference =

Estimate

rad/s

1 n_ap 100 1.11 10 14


n

1.5

0
0

M2
u (approx)

N 5

number of DOFS

ORIGIN 1

mass 1

stiff 10

Example for estimation of first natural frequency

M identity ( N) mass

2
1

K stiff 0
0

2 1

dt
0

1 2 1 0
0 1 2 1

0 0 1 1

U1
K

K
M

U3

U2

U4
K

U K U = 0

U5
K

Estimate for first natural frequency - Use Rayleigh-Ritz method:


ASSUME

u ( .1 .3 .5 .7 1 ) 1

uT K u1 1
R ( u)
uT Mu1 1

calculate Rayleigh's quotient

n_ap

R ( u)

n_ap 10.935

Actual n 9.001
1

% difference =

Estimate

rad/s

1 n_ap 100 21.485


n

1.5

0.5
0
0

M1
u (approx)

Mode Acceleration Method


Recall that the response in physical coordinates is
m

U ( t ) j q j( t ) , m < n
j =1

(36)

where m<n. The procedure is known as the mode displacement


method.
This method, however, fails to give an accurate solution even
when a static load is applied (See Structural Dynamics, by R.
Craig, J. Wiley Pubs, NY, 1981.).
The difficulty is overcome by using the procedure detailed
below. Recall that the system motion is governed by the set of
equations

 + K U = F
MU
(t )
(t )

(4)

And, if there are no rigid body modes, i.e. all natural frequencies
are greater than zero, then

 )
U ( t ) = K 1 ( F( t ) M U
where K

(51)

is a flexibility matrix. From Eq. (36),


m

 qj , m < n
U
( j ) (t )
j =1

(52)

Hence, Eq, (51) can be written as


MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

21

U ( t ) K 1F( t ) K 1M ( j ) qj( t )
j =1

(53)

Using the fundamental identity,

K ( i ) = i2 M ( i )

=
K
M ( i )
(i )
2

Write Eq. (53) as

( j )
U ( t ) K F( t ) 2 qj( t )

j =1 j
m

Note that

U S = K 1F( t )

(54)

(55)

is the displacement response vector due to a pseudo-static force


F(t), i.e. without the system inertia accounted for. Hence write Eq.
(54), as

( j )
U ( t ) U s ( t ) 2 qj( t )

j =1 j
m

; m<n

(56)

The second term above can be thought as the inertia induced


response.

Example: Consider the case of force excitation with frequency


n j and acting for very long times. The EOMs in physical

space are:

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

22

 + K U = F cos ( t )
MU
P
With a little damping, the steady state periodic response in modal
coordinates is

QPj
1
qj
K j 1
nj

cos ( t )
2

(37a)

Recall that, using the mode displacement method, the


response in physical coordinates is:

QPj

U j
Kj
j =1

1
nj

cos ( t )
2

(38)

From each of the modal responses,

cos ( t ) ;
qj
)
(
2
Kj
1 ( )
j

QPj

QPj 2
1
cos ( t )

2
2
2
j
K j j 1 ( )
j

qj

(57)

Since K j = j M j ; then using the mode acceleration method, the


2

response is

m
QPj

U U SP + j
Kj
j =1


2
1
cos ( t ) (58
2
2
j 1 ( j )

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

23

where the pseudo-static response is U SP = K FP . Now, in the


limit, as the excitation frequency decreases, i.e., as 0 , the
second term in Eq. (58) above disappears, and hence the physical
response becomes:

U = USP = K 1 FP

(59)

which is the exact response, regardless of the number of modes


chosen. Hence, the mode acceleration method is more accurate
than the mode displacement method. Known disadvantages
include more operations.
Finding the flexibility matrix is, in actuality, desirable. In
particular, if derived from measurements, the flexibility matrix is
easier to determine than the stiffness matrix.

MEEN 617 HD#7 Undamped Modal Analysis of MDOF systems. L. San Andrs 2008

24

STEP FORCED RESPONSE of Undamped 2-DOF


mechanical system

ORIGIN := 1

Dr. Luis San Andres (c) MEEN 363, 617 February 2008

The undamped equations of motion are:

d2

dt 2

(1)

X + K X = Fo

where M,K are matrices of inertia and stiffness coefficients, and X, V=dX/dt, d2X/dt2 are the
vectors of physical displacement, velocity and acceleration, respectively.
The FORCED undamped response to the initial conditions, at t=0, Xo,Vo=dX/dt, follows:
========================================================================

The equations of motion are:

M11 M12 d2 x1

2
M
M
22 dt x2
21

K11 K12 x1 F1o


=

K
K
21 22 x2 F2o

(2)

1. Set elements of inertia and stiffness matrices

100
M :=
0
M

Note

and

kg K :=

50
0

DATA FOR problem

2 10 6 1 10 6 N

m
6
6
2 10
1 10

n := 2 # of

DOF

K are symmetric matrices

0
0

Xo := m

initial conditions

0.0 m

0 sec

Vo :=

10000 N

5000

Applied force vector:

Fo :=

2. Find eigenvalues (undamped natural frequencies) and eigenvectors


Set determinant of system of eqns = 0

= K11 M11
4

) (K

22
2

M22

) (K

12

M12

= a + b + c = a + b + c = 0with =
where the
coefficients
are:

) (K

21

M21

) = (2a)
0

(2b)

a := M1 , 1 M2 , 2 M1 , 2 M2 , 1
b := K1 , 2 M2 , 1 K1 , 1 M2 , 2 K2 , 2 M1 , 1 + K2 , 1 M1 , 2
c := K1 , 1 K2 , 2 K1 , 2 K2 , 1

(2c)

The roots of equation (2b) are:

b ( b 2 4 a c) .5 b + ( b 2 4 a c) .5

1 :=
2 :=
2a

(3)

2a

also known as eigenvalues. The natural frequencies follow as:


j := 1 .. n

( ) .5

j := j

( )

f :=

112.6 rad

217.53 sec
17.92 Hz
f=

34.62

( )

Note that: 1 = 2 = 0

(4)

For each eigenvalue, the eigenvectors (natural modes) are


j := 1 .. n
1

a j := K1 , 1 M1 , 1 j
( K1 , 2 M1 , 2 j)

Set arbitrarily first element of vector = 1

1 a = 1
2

0.73
2.73

a1 =
MODAL matrix

j
A := a j

A is the matrix of eigenvectors (undamped


1
modal matrix): each column corresponds to an A =
0.73
eigenvector

Plot the mode shapes:

Aj ,

Aj ,

0.37

2.73

j
DOF

mode 1
mode 2

(5)

2.73
1

3. Modal transformation of physical equations to (natural) modal coordinates


(6)

X = Aq

Using transformation:

EOMs (1) become uncoupled in modal space:


Mm

d2
dt 2

(7)

q + K m q = Qm

(8)

Qm = A Fo

with modal force vector:

and initial conditions (modal displacement=q and modal velocity dq/dt=s)

qo = Mm

A M Xo

so = Mm

A M Vo

(9)

The modal responses are of the form:

qk = qo cos k t +
k

so

k=1....n
Qm

sin k t +
Km
k
k,

))

1 cos k t 0
k

(10a)

k
for an elastic mode

OR

qk = qo + so t +
k

Qm

2 Mm

for k = 0

(10b)

for a rigid body mode

k, k

And, the response in the physical coordinates is given


by the superposition of the modal responses, i.e.

X ( t) = A q ( t)

(5)

=== CHECK ========================================================


Verify the orthogonality properties of the natural mode shapes

Mm := A M A
T

Km := A K A

126.79
2.24 10 14
Mm =
kg

14
473.21
1.58 10

1.61 10 6
Km =
10
3.51 10

3.18 10 10 N
2.24 10 7

112.6 s -1

217.53

=========================================================================================

4. Find Modal and Physical Response for given initial condition and
Constant Force vector

0
0

Recall the vectors of initial conditions

0 m
Vo =
0 s

Xo = m

1 10 4 NDATA FOR problem being analyzed:


Fo =
m m
3
5 10

and Constant forces:

4.a Find initial conditions in modal coordinates (displacement = q, velocity = s)


Set inverse of modal mass matrix

qo := Ainv Xo

Ainv := Mm

A M

so := Ainv Vo

0
0

0
0

qo = m

so = m s

-1

4.b Find Modal forces:


T

6.34 10 3
Qm =
N
4
2.37 10

Qm := A Fo

4.c Build Modal responses:


q1 ( t) := qo cos ( 1 t) +
1

q2 ( t) := qo cos ( 2 t) +
2

so

1
so

sin ( 1 t) +

sin ( 2 t) +

Qm
Km

( 1 cos ( 1 t) )

1 , 1

Qm
Km

( 1 cos ( 2 t) )

2 , 2

for plots:

4.d Build Physical responses:

X ( t) := a1 q1 ( t) + a2 q2 ( t)

4.e Graphs of Modal and Physical responses:


Response in modal coordinates
0.01

Tplot :=

f1

0.005

0.056

0.11

0.17

0.22

0.28

0.33

time (s)

q1
q2

Response in physical coordinates


0.01
0.005
0
0.005
0.01

0.056

0.11

0.17

0.22

0.28

0.33

time (s)

x1
x2

5. Interpret response: analyze results, provide recommendations


S-S displacement

K
Recall natural frequencies & periods

17.92 Hz

34.62

f=

112.6 s -1

217.53

0.056

=
s
f 0.029

5 10 3
Fo =
m
0

1
1

0.73 2.73

A=

RIGID BODY MODE

STEP FORCED RESPONSE of Undamped 2-DOF


mechanical system

ORIGIN := 1

Dr. Luis San Andres (c) MEEN 363, 617 February 2008

The undamped equations of motion are:

d2

dt 2

(1)

X + K X = Fo

where M,K are matrices of inertia and stiffness coefficients, and X, V=dX/dt, d2X/dt2 are the
vectors of physical displacement, velocity and acceleration, respectively.
The FORCED undamped response to the initial conditions, at t=0, Xo,Vo=dX/dt, follows:
========================================================================

WITH RIGID BODY


MODE

The equations of motion are:

M11 M12 d2 x1

2
M
M
22 dt x2
21

K11 K12 x1 F1o


=

K
K
21 22 x2 F2o

(2)

1. Set elements of inertia and stiffness matrices

100
M :=
0
M

Note

and

kg K :=

50
0

DATA FOR problem

1 10 6 1 10 6 N

m
6
6
1 10
1 10

n := 2 # of

DOF

K are symmetric matrices

0
0

Xo := m

initial conditions

0.0 m

0 sec

Vo :=

1000 N

980

Applied force vector:

Fo :=

2. Find eigenvalues (undamped natural frequencies) and eigenvectors


Set determinant of system of eqns = 0

= K11 M11
4

) (K

22
2

M22

) (K

12

M12

= a + b + c = a + b + c = 0with =
where the
coefficients
are:

) (K

21

M21

) = (2a)
0

(2b)

a := M1 , 1 M2 , 2 M1 , 2 M2 , 1
b := K1 , 2 M2 , 1 K1 , 1 M2 , 2 K2 , 2 M1 , 1 + K2 , 1 M1 , 2
c := K1 , 1 K2 , 2 K1 , 2 K2 , 1

(2c)

The roots of equation (2b) are:

b ( b 2 4 a c) .5 b + ( b 2 4 a c) .5

1 :=
2 :=
2a

(3)

2a

also known as eigenvalues. The natural frequencies follow as:


j := 1 .. n

( ) .5

j := j

( )

f :=

rad

173.21 sec
0
f =
Hz
27.57

( )

Note that: 1 = 2 = 0

(4)

For each eigenvalue, the eigenvectors (natural modes) are


j := 1 .. n
1

a j := K1 , 1 M1 , 1 j
( K1 , 2 M1 , 2 j)

MODAL matrix

A j := a j

Set arbitrarily first element of vector = 1

1
a1 =

a2 =

2
1

A is the matrix of eigenvectors (undamped


modal matrix): each column corresponds to an
eigenvector

(5)
A =

2
1

Plot the mode shapes:


2

Aj ,

Aj ,

j
DOF

mode 1
mode 2
3. Modal transformation of physical equations to (natural) modal coordinates
Using transformation:

(6)

(6)

X = Aq

Using transformation:

EOMs (1) become uncoupled in modal space:


Mm

d2
dt

(7)

q + K m q = Qm

(8)

Qm = A Fo

with modal force vector:

and initial conditions (modal displacement=q and modal velocity dq/dt=s)

qo = Mm 1 A M Xo
T

so = Mm 1 A M Vo
T

(9)

The modal responses are of the form:

qk = qo cos k t +
k

so

k=1....n

sin k t +

Qm
Km

k, k

))

1 cos k t 0
k

(10a)

for an elastic mode


OR

Qm

qk = qo + so t +
2 Mm
k
k

for k = 0

t2

(10b)

for a rigid body mode

k, k

And, the response in the physical coordinates is given


by the superposition of the modal responses, i.e.

X ( t) = A q ( t)

(5)

=== CHECK ========================================================


Verify the orthogonality properties of the natural mode shapes

Mm =

Mm := A M A

Km := A K A

Km =

150
0

kg

300
0

9 10 6 m
0

-1
s
173.21
0

=========================================================================================

4. Find Modal and Physical Response for given initial condition and
Constant Force vector
Recall the vectors of initial conditions

0
Xo = m

dC

tf

0 m
Vo =
0 s

and Constant forces:

1 10 3 N
Fo =
m m

980

DATA FOR problem being analyzed:

4.a Find initial conditions in modal coordinates (displacement = q, velocity = s)

Ainv := Mm 1 A M

Set inverse of modal mass matrix

)
0.67 0.33

0.33 0.33

Ainv =
qo := Ainv Xo

so := Ainv Vo

0
qo = m

0
so = m s -1

4.b Find Modal forces:


T

Qm := A Fo

3
2.96

10

Qm =

20

4.c Build Modal responses:


q1 ( t) := qo + so t +
1

Qm

Mm

1 , 1

q2 ( t) := qo cos ( 2 t) +
2

so

t2

response for rigid body mode

sin ( 2 t) +

Qm
Km

( 1 cos ( 2 t) )

2 , 2

for plots:

4.d Build Physical responses:

X ( t) := a1 q1 ( t) + a2 q2 ( t)

Tplot :=

4.e Graphs of Modal and Physical responses:


Response in modal coordinates
0.004
0.003
0.002
0.001
0

0.022 0.044 0.065 0.087 0.11

time (s)

0.13

0.15

0.17

0.2

0.22

f2

time (s)

q1
q2

Response in physical coordinates

0.002

0.002

0.022 0.044 0.065 0.087 0.11

0.13

0.15

0.17

0.2

time (s)

x1
x2

5. Interpret response: analyze results, provide recommendations


S-S displacement - NONE

Recall natural frequencies & periods

f =

Hz

27.57
0

A =

2
1

s -1

173.21
0

0.22