2.

Modal Analysis
2.1 Introduction

Modal analysis (sometimes referred to as normal modes analysis or natural frequency

calculations) provides an understanding of the dynamic behaviour of a component, whether a
structure or an item of equipment. Components may respond in a dynamic manner to a variety
of different loading conditions which include seismic, air blasts, wind, unbalanced machinery,
crowd movement, waves, and so on.
The determination of natural frequency is important in order to avoid resonance. Resonance is
the phenomenon that occurs when the natural frequency of the applied loading corresponds
with the natural frequency of the component, and the amplitude of forced vibrations of the
component is magnified. This magnification is depends on the degree to which energy can be
dissipated from the system (damping), but in practice it can be large. It is therefore important
to attempt to avoid any possible resonance in the design of the component, or at least
minimise resonance to reduce the effects of its response to dynamic loading.
Modal analysis is usually a prerequisite before any dynamic response analysis to a prescribed
loading event is performed, as it provides insight as to how we might expect the structure or
component to behave. It can also enable the dynamic characteristics of a model to be
correlated with experimental data, thereby providing a robust validation of the dynamic model
and it enables the subsequent response analysis to be carried out more efficiently.

2.2 Degrees of Freedom

The Degrees of Freedom (dof) of a dynamic system give a rough guide to it’s complexity. It
describes the number of ways in which a body can move and mathematically, it denotes the
coordinates necessary to specify the motion.

The number of modes of vibration in which the structure can respond is equal to the number
of degrees of freedom of the structure, as shown in Figure 2.1. The most general way to
represent systems with one or only a few degrees of freedom is by making an analogy to
sliding masses (M) connected with springs to a boundary (or interconnected as in the 2 DOF
system shown). For building structures, the majority of the mass is congested at floor levels.
So if we have a building of say 'j' number of storeys, we can assume that each floor has a
mass Mi and is supported on mass-less columns between storeys of stiffness Ki. It is then
common to make the simplifying assumption that each floor can move laterally with degree of
freedom xi .

1
 Mj xj
K1
Kj M3
x x3
M1
K x1 K3 M2 x2
M K2
K2 M1 x1
M2
x2 K1
K3

Single DOF system Two DOF system ‘j’ DOF system

Figure 2.1

2.3 The Equation of Motion

Newton’s second law states that force is the product of mass and acceleration. With x
denoting the displacement, this can be written as:

F = ma or F = mx

All systems have distributed mass and stiffness, however now consider a system where all the
stiffness and mass are lumped into two quantities interacting with each other (SDOF system),
as shown in figure 2.2

x
K
M

Figure 2.2

The motion of this mass is defined by one coordinate only and hence it is a single degree of
freedom system. Let x be the displacement measured from the equilibrium position due to an
applied force P(t). The restoring force due to the stiffness of the spring is:

2
 F(spring) = - kx (1)

Hence, the net force on the mass is given by

F = P(t) – kx (2)

Applying Newton's Second Law of Motion to the system above gives,

P ( t ) − kx = mx (3)

Now, in a real system there will be some damping. This can arise from a combination of
viscous or frictional damping (e.g. movement through a viscous material such as air or oil and
friction between components). Assuming that there is only viscous damping present in the
above system, the mass will be subjected to a damping force which is in direct proportion to
the velocity and opposes the motion of the mass - the viscous damping is conventionally
represented by a dashpot shown in parallel with the spring as shown in figure 2.3.

x
K

M
C
Figure 2.3

The damping force is given by:

F (damping ) = −cx (4)

If Newton's second law of motion is applied to this system with viscous damping, Eq (3)
becomes:

P ( t ) − cx − kx = mx

giving
mx + cx + kx = P ( t ) (5)

3
Equation (5) is generally known as the EQUATION OF MOTION and forms the basis of all
dynamic analyses.

2.4 Single Degree of Freedom System

Equation (5) expresses the motion of a single degree of freedom system subject to a
disturbing force. If there is no disturbing force but the system is subjected to an initial
displacement, free vibrations will occur. These vibrations will gradually die out due to the
presence of damping in the system. This free vibration is given by,

mx + cx + kx = 0 (6)

This is a second order differential equation which has a standard solution of the form:

x=Ae λt

provided that:
mλ 2 + cλ + k = 0

The roots of this equation are:

2
−c ± c 2 − 4mk c ⎛ c ⎞ k
λ= =− ± ⎜ ⎟ − (7)
2m 2m ⎝ 2m ⎠ m

The complete solution of Equation (6) is of the form

x = A1eλ1t + A2 eλ2t (8)

Where λ1 and λ2 are the roots of Equation (7) and the particular form of Equation (8) depends
on the discriminant ⎡( c / 2m ) − k / m ⎤ . There are four different cases for the discriminant as
2

⎣ ⎦
follows:

(i) c=0
(ii) c 2 − 4km > 0
(iii) c 2 − 4km = 0
(iv) c 2 − 4km < 0

4
These cases are now examined:

Case (i), c = 0

When damping is neglected, i.e. c = 0, the particular solution of Equation (6) becomes:

x = A1 sin λ t + A2 cos λ t

This is more commonly written as:

x = A1 sin ωn t + A2 cos ωnt (9)

Where ωn is the circular natural frequency of the system, i.e. its frequency of free undamped
oscillations. ωn is given by:
k
ωn2 =
m
Equation (9) can be written as:

x = A sin (ωnt + α ) where A = A12 − A2 2 tan α = A1 / A2

or x = E cos (ωnt − φ ) where E = A, tan ϕ = A2 / A1 , i.e., φ = π / 2 − α

These terms are shown diagrammatically below in figure 2.4:

Figure 2.4

5
Case (ii), c2-4km > 0

This case is concerned with an overdamped system, where there is a comparatively large
amount of damping. Since c2 - 4km > 0, the roots of the above equation are real and negative.
Therefore equation (8) represents a gradual creeping back of the mass towards the position of
static equilibrium as shown in figure 2.5
x

t
Figure 2.5

For overdamping to occur, the damping has in general to be much greater than that occurring
in practical structures, although damping devices are occasionally included in some systems
to prevent vibration.

Case (iii), c2 – 4km = 0

In this case the motion is said to be critically damped. The associated damping coefficient is
given by,
cc2 = 4km (10)

substituting k = mωn2 in Equation (10) gives:

cc = 2mωn (11)

Critical damping is a particular case of overdarnping. The case is of little practical importance
but the magnitude of the critical damping coefficient gives a useful measure of the damping
capacity of a structure. It is often convenient to express the damping coefficient as a
percentage of the critical damping coefficient.

Case (iv), c2 – 4km < 0

This case represents a motion which is underdamped or simply damped. Since c2 -4km<0, the
roots of the equation are complex and the particular solution to Equation (6) is

6
 ⎡ c ⎤
−⎢ ⎥t
x=e ⎣ 2m ⎦
( A1 sin ωd t + A2 cos ωd t ) (12)

where, ωd is the damped circular natural frequency and is given by

ωd = ⎡( k / m ) − ( c / 2 m ) ⎤ 2
2
(13)
⎣ ⎦

If the damping ratio is defined as ζ = c / cc i.e. the ratio of the actual damping constant to the
critical value, then Equation (11) can be expressed as:

c
= ζωn
2m

Hence equation (12) can be expressed as:

x = Ae−ξωnt sin (ωd t + α )

where α is as defined as before

This expression can be represented diagrammatically as in figure 2.6:

Figure 2.6

7
From Equation (13).

⎡k ⎛ c ⎞
2
⎤
ωd = ⎢ −⎜ ⎟ ⎥
⎢⎣ m ⎝ 2m ⎠ ⎥⎦
Therefore,

ωd = ωn (1 − ξ )

In practice, ωd = ωn since ξ is small for most structures, except where heavy dampers have
been added. The value of ξ depends on the material and on the type of connections at the
joints (any looseness increases the damping). Typical values of ξ are given in table 1:

Table 1
Modal Damping Values (%)
Structure Type Design Stress Level Yield Stress Level
Welded Aluminium 2 4
structures
Welded and friction bolted 2 4
steel structures
Bearing bolted steel 4 7
structures
Prestressed Concrete 2 5
Structures
Reinforced concrete 4 7
structures

8
 Example 1 - Calculation of Natural Frequency of SDOF system

Consider a Portal Frame as shown in figure 2.7 with fixed base columns. The left column 'AB'
has a length L and second moment of area I, and the right column 'CD'has a length of 2L, and
a second moment of area of 3I. The columns can be considered to have no mass, and rigid
floor member BC has a concentrated mass termed 'Mass'.

B C
Mass

L, I
A
2L, 3I

Figure 2.7

For the purposes of illustrating the calculation, we will assume the following parameters:
4
Mass := 5000kg Length L := 3⋅ m I := 0.00005m
⋅
−2
Youngs Modulus E := 200⋅ kN⋅ mm

(Note: Always use compatible units of metres, Newtons, seconds and kilograms)
Also the stiffness of the vertical members against horizontal sway is: 12EI/L3

Hence the stiffness of column AB is:

12⋅ E⋅ I −1
KAB := KAB = 4.444⋅ kN⋅ mm
3
L

Hence the stiffness of column CD is:

12⋅ E⋅ ( 3I) −1
KCD := KCD = 1.667⋅ kN⋅ mm
3
( 2L)

9
The combined stiffness is:
−1
K := KAB + KCD K = 6.111⋅ kN⋅ mm

So for a single degree of freedom system, the circular natural frequency (radians per second)
is given as:
K −1
ωn := ωn = 34.96⋅ rad ⋅ sec
Mass

So the natural frequency in cycles per second is:

1 K
f := ⋅ f = 5.564⋅ Hz
2π Mass

This can also be expressed as a period of vibration which is the reciprocal of the frequency,
i.e.
1
T := T = 0.18s
f

It should be noted that damping has been neglected in the above example. However for steel
structures, ξ can be between 2% and 4%. Therefore from ωd = ωn(1-ξ), it can be seen that the
damped natural frequency will be very similar to the undamped frequency.

2.5 Multi Degree of Freedom Systems

A dynamic system is considered to be a multi degree of freedom system when two or more
independent co-ordinates are required to define it's motion

2.5.1 Two degree of Freedom System

Consider the two degree of freedom system shown in figure 2.8

K1

M1
x1
K2

M2
x2
K3

Figure 2.8

10
For this 2-DOF system, the equations of motion for free undamped vibrations are as follows:

For mass m1 x1 = − k1 x1 + k2 ( x2 − x1 )
m1 

For mass m2 x2 = − k2 ( x2 − x1 ) − k3 x2
m2 

Rearranging, x1 + ( k1 + k2 ) x1 − k2 x2 = 0
m1 

x2 + (k2 + k3 ) x2 − k2 x1 = 0
m2 

As before the solutions for this equation are of the form:

x1 = A1 sin (ωt + α )

and x2 = A2 sin (ωt + α )

Substituting in the equations for x1 and x2 gives

⎡⎣( k1 + k2 ) − m1ω 2 ⎤⎦ A1 − k2 A2 = 0
− k2 A1 + ⎡⎣( k2 + k3 ) − m2ω 2 ⎤⎦ A2 = 0 (14)

Solving these two equations gives:

⎡ ( k1 + k2 ) ( k2 + k3 ) ⎤ ω 2 + ( k1 + k2 )( k2 + k3 ) − k22
ω4 − ⎢ + ⎥ =0 (15)
⎣ m1 m2 ⎦ m1m2

This expression will have two real roots say ω1 and ω2, as expected as the system is a two
degree of freedom system.

Let β1 = A2/A1 be the amplitude ratio for ω1

β2 = A2/A1 be the amplitude ratio for ω2

It can be seen that from substituting these terms in (14) and rearranging that

β1β2 = -m1/m2

11
Therefore one of the amplitude ratios must be negative. This implies that for the above system
the masses are vibrating in phase in the lower mode (ω1) and with a phase angle of 180o for
the higher mode (ω2).

For the special case when all the stiffnesses are equal and all the masses are equal,
i.e. k1 = k2 = k3 = K say, and m1 = m2 = M say, then

2
⎛ 4 K ⎞ 2 3K
ω4 − ⎜ ⎟ω + 2 = 0
⎝M ⎠ M

This provides the two roots:

K
ω1 =
M

3K
ω2 =
M

These are the frequencies of the two modes of vibration of the system shown above with the
simplifying assumptions. A similar procedure would be followed if the masses were all
different of if the stiffnesses were not the same.

Mode Shape

The mode shapes for the two modes of the two degree of freedom system can be determined
as follows:

Let A11 and A12 be the amplitudes of mass m1 in the first and second mode. The notation is
such that the first subscript refers to the degree of freedom in question and the second
subscript refers to the mode in question. For the first mode the mode shape is given by the
expressions:
x1 = A11 sin (ω1t + α1 ) for mass 1

x2 = β1 A11 sin (ω1t + α1 ) for mass 2 (16)

For the second mode:

x1 = A12 sin (ω2t + α 2 ) for mass 1 (17)

x2 = β 2 A12 sin (ω2t + α 2 ) for mass 2

12
The general solution to the equation of motion would contain contributions from both modes
and would be as follows:

x1 = A11 sin (ω1t + α1 ) + A12 sin (ω2t + α 2 )

x2 = β1 A11 sin (ω1t + α1 ) + β 2 A12 sin (ω2t + α 2 ) (18)

Normalisation of Modes

The mode shapes for each mode are usually normalised so that each mode shape is defined
by a unique set of numbers which are independent of the absolute values but dependent on
the relative values. The normalisation employed is arbitrary.
The mode shapes given by Equations (16) and (17) above are dependent on the absolute value of
the amplitudes of mass m1 in the first and second mode, i.e. A11 and A12
The mode shapes can be normalised in a variety of ways. Typical normalising condition
are:

A1n = 1

∑m A
i
i in =1

∑m A
i
i
2
in =1

∑m A = ∑m
i
i
2
in
i

where n = mode shape and i = the degree of freedom.

For a two degree of freedom system the value of i in the above conditions is 2. For example,
for the special 2 DOF case when k1 = k2 = k3 = K, and m1 = m2 = M, using the third criterion
for example, the normalising condition becomes:

(
M A112 + A212 = 1 )
M (A 2
12 + A222 ) =1
The amplitude values can be obtained from equation (14) as follows:

⎡⎣ 2 K − M ω 2 ⎤⎦ A11 − KA21 = 0

13
 K
For the first mode, substituting ω =
M

[ 2 K − K ] A11 − KA21 = 0

Therefore KA11 = KA21 and A11 = A21

3K
Similarly the second mode, substituting ω =
M

[ 2 K − 3K ] A12 = KA22
− KA12 = KA22
Therefore A12 = − A22

Substituting for A21 and A22 in the normalising condition gives:

M ( A112 + A112 ) = 1

( )
M A122 + A122 = 1

Therefore for the first mode, the normalising condition becomes

⎛M ⎞
A 11 = ⎜ ⎟
⎝ 2 ⎠
And for the second mode, the normalising condition becomes

⎛M ⎞
A 12 = ⎜ ⎟
⎝ 2 ⎠
Orthogonal Modes

The modes of vibration of a structure can normally be considered to be independent of each

other. Such modes are referred to as either orthogonal modes or normal modes. This
independence between the modes can be confirmed by examining the orthogonality
condition, as follows:
In general, the orthogonal property of normal modes is given by:
∑m A
i
i in Aim = 0

14
Where m = mode 1, n = mode 2, and i = DOF

The orthogonal property of normal modes in the 2 d.o.f. system considered above is:

m1 A11 A12 + m2 ( β1 A11 )( β 2 A12 ) = 0

If this condition applies, then the inertia forces derived from one mode does not do any work on
the other mode and hence confirms that the modes can be considered to be independent.

For the simplified example where and therefore, A u — A 21 and A n — -/1 3;

Substituting in Equation 19 for ft and ft2
UAuAn + MA2IA22 = 0 Substituting
A2} and A22 gives

MAl}A12 - MAJTA10 = 0

Therefore the two modes are orthogonal for this simplified example and can be
considered independent of each other.

Example 2 - Two Degree of Freedom Shear Frame - Natural Frequency Calculation

Consider a two storey Portal Frame as shown in figure 2.9 with fixed base columns. The mass
of the rigid floors is 5000kg each, and the storey stiffness in sway (combined stiffnesses of all
the columns in each storey are given as 5kN/mm).

x2 C D
M2 = 5000 kg

K2 = 5 kN/mm
x1
B M1 = 5000 kg E

K 1= 5 kN/mm

A F
Figure 2.9

15
Step 1 - Calculate the two natural frequencies

For the 2 degree of freedom system, the equations of motion for free undamped vibrations are
as follows:

For mass m1 x1 = − k1 x1 + k2 ( x2 − x1 )
m1 

Rearrange: x1 + ( k1 + k2 ) x1 − k2 x2 = 0
m1  (A1)

For mass m2 x2 = − k2 ( x2 − x1 )
m2 

Rearrange x2 − k2 x1 + k2 x2 = 0
m2  (A2)

Substituting for m1,m2,k1 and k2 in the preceding equations yields the following equations:

Equation A1 becomes: 5000  ( ) ( )

x1 + 5 ×106 + 5 × 106 x1 − 5 × 106 x2 = 0

Rearrange: x1 + 2000 x1 − 1000 x2 = 0

 (A3)

x2 − 5 ×106 x1 + 5 ×106 x2 = 0
Similarly equation (A2) becomes: 5000 

Rearrange: x2 − 1000 x1 + 1000 x2 = 0

 (A4)

Consider the solution to these equations to be:

x1 = A1 sin (ωt + α )

x2 = A2 sin (ωt + α )

This gives
x1 = − A1ω 2 sin (ωt + α )


x2 = − A2ω 2 sin (ωt + α )



Substituting for x1 , x2 , x1 , and x2 in equations (A3) and (A4) and comparing coefficients of
sin (ωt + α )

16
 − A1ω 2 + 2000 A1 − 1000 A2 = 0

− A2ω 2 + 1000 A1 − 1000 A2 = 0

A solution exists if the above two expressions are satisfied. The expressions can be
rearranged as follows:

( 2000 − ω ) A − 1000 A
2
1 2 =0 (A5)

(
−1000 A1 + 1000 − ω 2 A2 = 0 ) (A6)

These two equations can be solved to give the values of ω2, as follows:

1000 − ω 2
From (A6) A1 = A2 (A7)
1000

⎛ 1000 − ω 2 ⎞
Substituting into (A5) ( 2000 − ω 2 ⎜ ) ⎟ A2 − 1000 A2 = 0
⎝ 1000 ⎠

From which

( 2000 − ω )(1000 − ω ) − 10
2 2 6
=0
Expanding gives
ω 4 − 3000ω 2 + 106 = 0

The roots are:

( −3000 )
2
3000 ± − 4 × 106
ω = 2

Giving the solution

ω 2 = 2618 or 382

Hence ω = 51.2 or 19.5 rads/sec

ω
Since f = then f = 8.15 Hz or 3.10 Hz
2π

These are the two natural frequencies of the portal frame, with the fundamental (lowest)
frequency being 3.10Hz
Step 2 - Determine the Mode Shapes which accompany the natural frequencies

17
Let Ain be the amplitude of the mass mi in the 'n'th mode.

For the first mode of vibration, the mode shape is given by:

x1 = A11 sin (ω1t + α1 ) for mass 1

and x2 = A21 sin (ω1t + α1 ) for mass 2

Similarly for the second mode of vibration, the mode shape is given by

x1 = A12 sin (ω2t + α 2 ) for mass 1

and x2 = A22 sin (ω2t + α 2 ) for mass 2

The maximum value of x1 and x2 will occur when sin (ωnt + α n ) = 1

For the first mode, ω2 = 382

x1 = A11
x2 = A21

From equation (A5)

1000 − ω 2
A11 = A21
1000

A11 = 0.62 A21 (A8)

Similarly for the second mode, ω2 = 2618

A12 = -1.62 A22 (A9)

If we assume a normalisation factor A1n = 1, then

A11 = 1.0, A21 = 1.62 for mode 1

A12 = 1.0 A22 = -0.62 for mode 2

The mode shapes are shown diagrammatically in figure 2.10

18
 C D x2 1.62 0.62
M2

B E x1 1.0 1.0
M1

A F

Figure 2.10

Orthoganality

As discussed previously, the two modes of vibration of the portal frame can be considered to
be independent of each other if they satisfy the orthogonality condition, i.e.

∑m A A
i
i i1 i2 =0

where i = degree of freedom

For the portal frame this expands to:

m1 A11 A12 + m2 A21 A22 = 0

Substituting for m1, m2, A11 and A12 from equations (8) and (9)

5000 ( 0.62 A21 )( −1.62 A22 ) + 5000 A21 A22 = 0

−5000 A21 A22 + 5000 A21 A22 = 0

This implies that the above condition is satisfied and the two modes of vibration of the portal
frame can be considered to be independent.

19
2.5.2 Multi Degree of Freedom Systems
Consider the multi degree of freedom system shown in figure 2.11,

mn xn

mi+1 xi+1
k i+1
mi xi
B D
ki
mi-1 xi-1
A C

k i-1 mi-2 xi-2

Figure 2.11
Note : ki is the combined bending stiffness of AB & CD

The above system needs n coordinates to specify its motion. Therefore the system has n
degrees of freedom giving rise to n equations of motion.

For each mass, there is an equation of motion of the form

xi = − ki ( xi − xi −1 ) + ki +1 ( xi +1 − xi ) − ci ( xi − xi −1 ) + ci +1 ( xi +1 − xi )

mi 

Similar equations can be written for all the other masses. In order to 'simplify' the problem,
the equations of motion of a multi degree of freedom system can be expressed in matrix form,
i.e.
 
MX+CX+KX=0 (20)
where X is a vector of the independent coordinates required to specify the motion of the
system., i.e.

20
 ⎡ x1 ⎤
⎢x ⎥
⎢ 2⎥
⎢x ⎥
X = ⎢ 3⎥
⎢.⎥
⎢.⎥
⎢ ⎥
⎣⎢ xn ⎦⎥

X and X
 are the corresponding velocity and acceleration vectors.
K is the stiffness matrix, which is square and symmetric as follows:

⎡ k11 k12 k13 .... k1n ⎤

⎢k k22 k23 .... k2 n ⎥⎥
K = ⎢ 21
⎢ . . . .... . ⎥
⎢ ⎥
⎣ kn1 kn 2 kn 3 .... knn ⎦

For the above example, the 'i' th row of the stiffness matrix becomes

[0, 0,.............0, − ki , (ki + ki +1 ), −ki +1 , 0,...........0]

C is the damping matrix and is also square and symmetric. It can be defined in a similar way
to the stiffness matrix, as above.
M is the mass matrix. The mass matrix is normally square and symmetric.

⎡ m11 m12 m13 .... m1n ⎤

⎢m m22 m23 .... m2 n ⎥⎥
M = ⎢ 21
⎢ . . . .... . ⎥
⎢ ⎥
⎣ mn1 mn 2 mn 3 .... mnn ⎦

For the example shown above, the mass matrix is diagonal (i.e. mij = 0 for i ≠ j)

21
2.5.2.1 Free Undamped Vibration

In order to simplify the problem, consider the case of free undamped vibration, C = 0. The
equation of motion becomes:
MX + KX = 0 (21)

The solution to this equation is of the form,

X = A sin (ωt + α )

where A is the amplitude of vibration matrix.

Substituting for X in equation 21 and recognising that sin (ωt + α ) is common and not equal
to 0, gives

( K-Mω ) A = 0
2
(22)

For equation (22) to have a solution, other than A = 0, the determinant of the coefficients of A
must be zero, ie

det K-M ω 2 = 0 (23)

k11 − m11ω 2 k12 − m12ω 2 ........

k21 − m21ω 2 k22 − m22ω 2 ........
=0 (24)
k31 − m31ω 2 k32 − m32ω 2 ........
........ ........ ........

In general equation24 gives positive real roots for ω2, say ω12, ω22, ω32,...., ωn2, with
ω12 < ω22 < ω32,...., < ωn2 and ω1, ω2, ω3,...., ωn are the natural frequencies of the multi
degree of freedom system.

22
For any value of frequency ωi, Equation (23) gives the amplitude of vibration matrix.
Eigenvalues and Eigenvectors

The values of ωi2for a multi degree of freedom system are known as the Eigenvalues.
The values of A corresponding to a particular ωi2 are known as the Eigenvectors for that
mode. The eigenvectors can be normalised as described previously.
A typical example for a 3 degree of freedom system is shown in example 3

Example 3 - Three Degree of Freedom Shear Frame - Natural Frequency Calculation

Consider a three storey single bay Portal Frame with fixed base columns as shown in figure
2.12. In this example, the solution will illustrate how matrices can be used for the solution
and hence the general procedure for multi degree of freedom structures is developed.

D E x3
M3 = 2000 kg

K CD = 750 kN/mm K EF = 750 kN/mm

C F x2
M2 = 4000 kg

K BC = 1000 kN/mm K FG = 1000 kN/mm

G x1
B M1 = 4000 kg

K AB = 1500 kN/mm K GH = 1500 kN/mm

A H
Figure 2.12

For each mass, the following equations can be written:

xii = − ki ( xi − xi −1 ) + ki +1 ( xi +1 − xi ) − ci ( xi − xi −1 ) + ci +1 ( xi +1 − xi )

mi 

Assume for this example that damping is negligible and that:

kAB + kGH = k1

23
 kBC + kFG = k2
kCD + kEF = k3
Then the equations of motion will be as follows:

x1 + k1 x1 − k2 ( x2 − x1 ) = 0
m1 
x2 + k2 ( x2 − x1 ) − k3 ( x3 − x2 ) = 0
m2 
x3 + k3 ( x3 − x2 ) = 0
m3 

These can be rearranged as follows:

x1 + (k1 + k2 ) x1
m1  − k2 x2 =0
x2 − k2 x1
m2  + (k2 + k3 ) x2 − k3 x3 = 0
m3 
x3 − k3 x2 + k3 x3 = 0

These three equations can be expressed in matrix form:

MX + KX = 0 (B1)

⎡ m1 0 0⎤ ⎧ x1 ⎫ ⎧ 
x1 ⎫
⎪ ⎪ ⎪ ⎪
where M = ⎢⎢ 0 m2 0 ⎥⎥ X = ⎨ x2 ⎬ X = ⎨ 
x2 ⎬
⎢⎣ 0 m3 ⎥⎦ ⎪ ⎪ ⎪x ⎪
0 ⎩ x3 ⎭ ⎩ 3⎭

⎡ k1 + k2 − k2 0 ⎤
and K = ⎢⎢ −k2 k 2 + k3 − k3 ⎥⎥
⎢⎣ 0 − k3 k3 ⎥⎦

The solution to this equation is:

X = A sin(ωt + α ) (B2)

where A is the amplitude of vibration matrix.

Substituting for X in (B1) and comparing coefficients gives:

(K − M ω 2 ) A = 0

For this equation to have a solution other than A = 0, the determinant of the coeffiicents of A
must be zero.

i.e. K − Mω2 = 0

Using matrix algebra this can be rewritten as :

24
 { }
M M −1 K − ω 2 I = 0 where I is a unit matrix

Since the determinant of the product of 2 matrices is equal to the product of the determinants
of the matrices, then:

M M −1 K − ω 2 I = 0

Since M is a diagonal matrix whose elements are non-zero, M is non singular. hence for the
above equation to hold:

M −1 K − ω 2 I = 0 (B3)

where {M-1K} is known as the dynamic matrix.

Equation (B3) is called the characteristic equation and this represents an a typical eigenvalue
expression. The roots of the equation yield the eignevalues, and these are the squares od the
circular natural frequencies ω2. This is solved as follows:

For the three storey portal frame, the matrices are:

⎡ 4000 0 0 ⎤
⎢
M =⎢ 0 4000 0 ⎥⎥ Kg
⎢⎣ 0 0 2000 ⎥⎦

⎡1 0 0 ⎤
⎢ 4000 ⎥
M −1 = ⎢ 0 1 0 ⎥ Kg −1
⎢ 4000 ⎥
⎢ 0 0 1 ⎥
⎢⎣ 2000 ⎦⎥

Also,

⎡5 −2 0 ⎤
K = ⎢ −2 3.5 −1.5⎥⎥ ×106 N / m
⎢
⎢⎣ 0 −1.5 1.5 ⎥⎦

⎡1250 −500 0 ⎤
M K = ⎢ −500 875 −750 ⎥⎥ N / MKg
−1 ⎢
⎢⎣ 0 −750 750 ⎥⎦

25
Now
1250 − ω 2 −500 0
M K −ω I =
−1 2
−500 875 − ω 2
−375
0 −375 750 − ω 2

The determinant of the above matrix can be calculated giving:

(1250 − ω 2 )[(875 − ω 2 )(750 − ω 2 ) − 281250] − 500[500(750 − ω 2 )] = 0

This can be simplified to:

−ω 6 + 2875ω 4 − 2156250ω 2 + 281250000 = 0

where

⎡167 0 0 ⎤
⎢
ω = ⎢ 0 1000
2
0 ⎥⎥
⎢⎣ 0 0 1710 ⎥⎦

so the eigenvalues (in (rad/sec)2) are: ω12 = 164 ω22 = 1000 ω32 = 1710

ω
With f = , the natural frequencies are: f1 = 2.04Hz, f2 = 5.03Hz, f3 = 6.58Hz
2π

The eigenvectors are given by the amplitude matrix A and represent the mode shapes at the
three different natural frequencies.

(
From the previous expression K − M ω 2 A = 0 )
For any mode, the above equation will give the eigenvector

e.g. Consider mode 2, ω2 = 1000,

⎧⎡ 5 −2 0 ⎤ ⎡4 0 0⎤ ⎫ ⎧ A12 ⎫
⎪⎢ ⎥ ⎢ ⎥ 6⎪ ⎪ ⎪
⎨ ⎢ −2 3.5 −1.5⎥ × 10 − ⎢ 0 4 0 ⎥ × 10 ⎬ × ⎨ A22 ⎬ = 0
6

⎪ ⎢ 0 −1.5 1.5 ⎥ ⎢⎣ 0 0 2 ⎥⎦ ⎪ ⎪A ⎪
⎩⎣ ⎦ ⎭ ⎩ 32 ⎭

This gives:

⎡1 −2 0 ⎤ ⎧ A12 ⎫
⎢ −2 −0.5 −1.5⎥ ×106 ⎪ A ⎪ = 0
⎢ ⎥ ⎨ 22 ⎬
⎢⎣ 0 −1.5 −0.5⎥⎦ ⎪ ⎪
⎩ A32 ⎭

26
Expand the first row => 106 x A12 = 2 x 106 x A22 => A12 = 2 A22

Expand the third row => -1.5 x 106 x A22 = 0.5 x 106 x A32 => A22 = - A32/3

If the mode shape is normalised so that A12 = 1, then the eigenvectors for the 2nd mode are:

⎧ 1 ⎫
⎪ ⎪
⎨ 0.5 ⎬
⎪−1.5⎪
⎩ ⎭

Similarly the eigenvectors for the 1st and 3rd modes are:

⎧ 1 ⎫ ⎧ 1 ⎫
⎪ ⎪ ⎪ ⎪
1st mode = ⎨2.17 ⎬ 2nd mode = ⎨−0.92 ⎬
⎪ 2.78 ⎪ ⎪ 0.72 ⎪
⎩ ⎭ ⎩ ⎭

The mode shapes are shown diagrammatically in figure 2.13

-1.5 0.72
2.78

2.17 0.5
-0.92

1 1 1

1st Mode, freq = 2.04Hz 2nd Mode, freq = 5.04Hz 3rd Mode, freq = 6.58Hz

Figure 2.13

Orthoganality

The modes of vibration of the portal frame can be considered to be independent of each other
if they satisfy the orthogonality condition.

Consider modes 1 and 2 =>

AmT × M × An = 0 where m = mode1 and n = mode2

27
 ⎡ 4000 0 0 ⎤⎧ 1 ⎫
⎢ ⎪ ⎪
Hence {1 2.17 2.78} ⎢ 0 4000 0 ⎥⎥ ⎨ 0.5 ⎬ = 0
⎢⎣ 0 0 2000 ⎥⎦ ⎩⎪−1.5⎭⎪

⎧ 4000 ⎫
i.e. {1 2.17 2.78} ⎪⎨ 2000 ⎪⎬ = 0
⎪−3000 ⎪
⎩ ⎭

giving 4000 + 4340 - 8340 = 0

Therefore these modes can be considered to be independent. If the same calculation is carried
out for mode pairs 1 and 3, then 2 and 3, the results also show orthogonality.

2.5.3 Large Multi Degree of Freedom Systems

For large systems, the computation of eigenvalues and eigenvectors would require a lot of
effort. There are standard methods available (see Dynamics of Structures by Clough and
Penzien),
In some cases the number of eigenvalues and eigenvectors can be reduced which reduces the
amount of computational effort considerably.

2.5.4 Systems with an infinite number of Degrees of Freedom

The above discussion is applicable only to structures with lumped masses. In real structures,
the mass is distributed and not lumped at specific places giving the structure an infinite
number of degrees of freedom.

In some cases, approximations can be made. Consider for example the stepped beam shown in
figure 2.14

28
 1 2 3 4 5 6 7

Figure 2.14

This structure can be idealised as a system of, say, 7 lumped masses supported by a massless
beam and treated as a 7 degree of freedom system.
For hand calculation of natural frequency for simple structures (of either finite or infinite
degrees of freedom), standard formulae exist. In the Reference material given at the end of the
notes, extracts from a publication 'Blevin's - Formulas for natural frequency and mode shape'
are given.

Note:
Extensive use will be made of these formulae in some of the problems we will tackle both in
examples and tutorials, hence it will not be necessary to derive natural frequencies and
mode shapes from first principles each time. In fact we can simplify many complex
structures and use standard formulae in our calculations to derive an approximation of its
dynamic properties, and thus develop a 'feel' for its behaviour

For detailed derivation of formulae and treatment of distributed-mass structures refer to

standard dynamic analysis textbooks.

2.6 Modal Analysis using the Finite Element Method

For structural systems with more than 2 degrees of freedom, it is best to carry out the modal
analysis using computer methods rather than the classical methods previously described. A
commonly used computer method is the Finite Element Method. The theory behind the finite
element approach is outside the scope of these notes and the reader is referred to the many
references on the finite element method. All the major finite element codes, such as ANSYS,
ABAQUS and SAP2000 have the capability to carry out modal analyses.

29
The finite element method idealises the structure or equipment component into a large, but
finite, number of small elements and assembles matrices such as those given in section 2.5.2.
Eigenvalues and eigenvectors are subsequently determined from Equations such as (22) and
(23).

2.6.1 Eigenvalue extraction methods

The extraction of eigenvalues and eigenvectors can be carried out in a number of ways. The
most popular and computational reliable is the Givens/Householder triangulisation method.
However, it requires l/2n2 numbers to be stored (where n = number of equations of motion)
which may cause problems with computer storage limitations. This fact explains why a finite
element model developed for static analysis problems may not be appropriate for use in
dynamic applications. For example a dynamic problem with 500 degrees of freedoms will
typically require the same core storage as a static problem with 10, 000 degrees of freedoms.
Other methods are available for the eigenvalue extraction where the storage requirement is
related to the number of desired modes, so it is possible to obtain a solution for a small
number of modes. However, as the number of modes increases, the problem size becomes
restrictive. In seismic analysis, a significant number of modes may be required up to say 33Hz
(which is a typical frequency cut-off for earthquake motion) and consequently these
techniques are not usually used.
The above problems often lead to the analyst simplifying the model into a simple beams,
masses and springs in order to reduce the numbers of degrees of freedom and subsequently
employ the Given/Householder triangulisation technique.

2.6.2 Dynamic Reduction Techniques

The above limitations with modal analyses were recognised in the mid-1960's and an
approximate method developed to reduce the size of the model for modal analyses purposes.
The method, known as Guyan Reduction (See Bibliography), splits the model up into 'Master'
and 'Slave' degrees of freedoms. The eigenvalue extraction then proceeds with only the master
degrees of freedom, thereby reducing the complexity of the model. However, this presents a
new problem, as the master degrees of freedom have to be selected and clearly the selection
of master degrees of freedom will greatly influence the results. Commonly used practices are
the removal of rotational degrees of freedom and the removal of translational degrees of
freedom near a restrained location.
Guyan Reduction is available in many finite element codes such as ANSYS. Some codes such
as ANSYS include an approximate procedure which automatically selects the degrees of
freedom which are to be retained as masters.
Guyan reduction enables larger models to be run than would be very difficult or even
impossible with no reduction and is normally followed by the extraction of eigenvalues
following the Givens/Householder technique. A generally accepted criterion is that the natural
frequencies obained following Guyan reduction are within 10% of those obtained without
reduction for the first N/3 modes, where N is the number of master degrees of freedom in the
analysis. However, it should be noted that this requires a careful selection of the master
degrees of freedom, as modes can be missed and participation factors may be incorrect if

30
sufficient masters have not been selected. Other reduction techniques are available in specific
finite element programs that lead to more reliable results than Guyan Reduction.

Generally, the best solution is to reduce the complexity of a dynamic model to avoid the need
for any reduction.

2.6.3 Results from Finite Element Modal Analyses

The standard output from FE programs normally comprises the eigenvalues and the
corresponding eigenvectors. The latter are usually normalised but the normalising condition
varies from program to program (refer to the program manuals for details).
There are other details which may be output by the programs or may require some sort of
post-processing. These include the amount of mass participating in each mode and the
participation factor. These two outputs are of importance in assessing the relative importance
of each mode and in checking the results, as will be discussed in the next unit.

31