CEE490b

Chapter 1 - Free Vibration of Multi-Degree-of-Freedom Systems - I

1.1 Free Undamped Vibration

The basic type of response of multi-degree-of-freedom systems is free undamped

vibration. Analogous to single degree of freedom systems the analysis of free
vibration yields the natural frequencies of the system. For the analysis, the elastic
(restoring) properties of the system must be described first. This can either be done
in terms of stiffness or flexibility

Structural Stiffness

Stiffness of a structure is described by the stiffness matrix, whose elements k ij are

defined as the force acting at node i, in order to produce a sole unit displacement at
node j. In “lumped mass” models, the stiffness constants defined above are identical
to the stiffness used in static models

Example - Multi-storey “shear” building

A shear building is one where the resistance to lateral loads is from the bending of
the columns – the floors are infinitely rigid – and the columns are fixed-ended where
connected to the floors.

k31=0 k32 k33

m3 1

l3 I3

k21 k22 k23

m2 1

l2 I2

k11 k12 k13=0

m1 1

l1 I1

Fig. 1.1 Multi Storey Shear Building

Page 1-1
CEE490b

EI
The basic stiffness constant for a column subjected to shear only is k = 12 ,
l3
where EI is the bending stiffness of the column and l is the column length. The
assembly of the stiffness matrix is performed one element at a time, with each floor
of the building sequentially subjected to a unit shear displacement and the
stiffnesses added as appropriate. e.g. the stiffness element for the first floor of the
shear building in Fig. 1.1, due to a unit displacement of the first floor is:

12EI1 12EI 2
k11 = 2 ∗ 3
+ 2∗ 3
l1 l2

(the ‘2’ multiplier is for 2 columns per storey)

The full stiffness matrix for the 3-storey shear building is:

12EI1 12EI 2 ⎛ 12EI 2 ⎞

k11 = 2 ∗ + 2∗ k12 = 2 ∗ ⎜ − ⎟ k13 = 0
l1
3
l2
3 ⎜ 3 ⎟
⎝ l2 ⎠
⎛ 12EI 2 ⎞ 12EI 2 12EI 3 ⎛ 12EI 3 ⎞
k 21 = 2∗⎜− ⎟ k 22 = 2∗ + 2∗ k 23 = 2 ∗ ⎜ − ⎟
⎜ l2 ⎟⎠
3 3 3 ⎜ 3 ⎟
⎝ l2 l3 ⎝ l3 ⎠
⎛ 12EI 3 ⎞ 12EI 3
k 31 = 0 k 32 = 2 ∗ ⎜ − ⎟ k 33 = 2 ∗
⎜ l
3 ⎟ l3
3
⎝ 3 ⎠

Note: - the matrix is diagonally symmetric ( k 12 = k 21 )

Static Condensation

This is the term given to the simplification of a stiffness matrix through the
elimination of degrees of freedom. For example, in most buildings and structures
exposed to lateral loads, there are no significant external moments or mass moment
of inertia acting in the joints. Therefore the joint rotations can be eliminated from the
governing equations, so the deformation of the structure can be expressed in terms
of lateral displacements only.

Considering the full 4x4 stiffness matrix for the column shown below, the elements
can be assembled one degree of freedom at a time. We shall see how this can be
simplified using Static Condensation.

Page 1-2
CEE490b

Recall the stiffness characteristics of a fixed ended beam:

P P
A l B A l B

MA MA
MB MB
P EI P EI

2EIθ
M A = 21 M B =
M A = M B = 6EI∆ / l 2 l
6EIθ
Pl = M A + M B = 2 ∗ 6EI∆ / l 2 Pl = M B + M A =
l
P = 12EI∆ / l 3 6EIθ
P= 2
l
P 6EI
= 2
P 12EI θ l
= 3
∆ l MB 4EI
=
M 6EI θ l
= 2
∆ l M A 2EI
=
θ l

P2 4 2 4
2

EI2 l2

3 3
P1 1 1

EI1 l1

a) b) a) b)
Fig. 5.2 a) Generation of full stiffness matrix (4x4) and b) Condensed (2x2)

Page 1-3
CEE490b

Assembling the elements of the complete stiffness matrix, we obtain:

12EI1 12EI 2
k11 = 3
+ 3
l1 l2
12EI 2
k12 = k 21 = − 3
l2
6EI 6EI
k 13 = k 31 = 2 1 − 2 2
l1 l2
6EI
k 14 = k 41 = − 2 2
l2
12EI 2
k 22 =
l 32
6EI
k 23 = k 32 = 2 2
l2
6EI
k 24 = k 42 = 2 2
l2
4EI1 4EI 2
k 33 = +
l1 l2
2EI 2
k 34 = k 43 =
l2
4EI 2
k 44 =
l2

Now, if only static horizontal forces P1,2 act, the matrix equation relating the input
forces to the output displacements and rotations is:
⎧ P1 ⎫ ⎡ k11 k12 k13 k14 ⎤ ⎧ u1 ⎫
⎪P ⎪ ⎢k ⎥⎪ ⎪
⎪ 2 ⎪ ⎢ 21 k 22 k 23 k 24 ⎥ ⎪u 2 ⎪
⎨ ⎬= ⎨ ⎬
⎪ 0 ⎪ ⎢k 31 k 32 k 33 k 34 ⎥ ⎪ψ 1 ⎪
⎪⎩ 0 ⎪⎭ ⎢⎣k 41 k 42 k 43 k 44 ⎥⎦ ⎪⎩ψ 2 ⎪⎭

or, generally:

⎧P ⎫ ⎡ A B ⎤ ⎧u ⎫
⎨ ⎬=⎢ T ⎨ ⎬
⎩ 0 ⎭ ⎣B C ⎥⎦ ⎩ψ ⎭

note that the lower part of the equation:

{0} = [B ]T {u} + [C ]{ψ }

Page 1-4
CEE490b

yields the rotations:

{ψ } = −[C ]−1 [B ]T {u}

then substituting this expression for the unknown rotations in terms of the
unknown displacements into the upper part of the equation:

{P } = [A]{u} − [B ][C ]−1[B ]T {u}

{P } = ([A] − [B ][C ]−1 [B ]T ){u}
the Condensed Stiffness Matrix is then:

[k ′] = ([A] − [B][C ]−1 [B]T )

In this case, it is a n x n matrix, where n is the number of translational degrees of
freedom (in this case a 2 x 2 matrix) involving translations only. If the rotations
are desired, it is a simple matter to insert the resulting displacements into the
known relationship between displacement and rotation.

Governing Equations for the Solution to the Free Vibration Problem in

n Degrees-of-Freedom

With the stiffness constants defined, the governing equations of motion can be
written using Newton’s Second Law for each of the masses in the system:

i.e. mass ∗ acceleration = ∑ forces acting on the mass

m1u&&1 = −k 11u1 − k12 u 2 ... − k1i u i ... − k1n u n

m2 u&&2 = −k 21u1 − k 22 u 2 ... − k 2 i u i ... − k 2n u n
.
.
mn u&&n = −k n1u1 − k n 2u 2 ... − k ni u i ... − k nn u n

where:

d 2 u i (t )
u&&i =
dt 2

Page 1-5
CEE490b

Then for mass mi , in general for j = 1,2…n where n is the number of the degrees of
freedom:

n
m i u&&i + ∑ k ij u j = 0 (1.1)
j =1

This is a set of simultaneous, ordinary differential equations of the second order.

This can be written in matrix form:

[m]{u&&} + [k ]{u} = {0} (1.2)

⎡m1 0 0 0 ⎤
⎢0 m 0 0 ⎥⎥
where [m ] is the diagonal mass matrix: [m ] = ⎢ 2

⎢0 0 mi 0 ⎥
⎢ ⎥
⎣0 0 0 mn ⎦

The displacement vector is a column matrix:

⎧ u1(t ) ⎫
⎪u (t )⎪
⎪⎪ 2 ⎪⎪
{u(t )} = ⎨ . ⎬ = [u1(t ) u 2 (t ) . . u n (t )]T
⎪ . ⎪
⎪ ⎪
⎪⎩u n (t )⎪⎭
and the symmetrical stiffness matrix [k] is:

⎡ k11 k12 k1 j k1n ⎤

⎢k k k2 j k 2n ⎥⎥
[k ] = ⎢⎢ k21 k 22 k jj k jn ⎥
j1 j2
⎢ ⎥
⎢⎣ n1
k k n2 k nj k nn ⎥⎦

As in the single degree of freedom case, the particular solution is:

u i (t ) = u i sin ωt
u&&i (t ) = −u i ω 2 sin ωt

or, in matrix notation:

{u(t )} = {u}sinωt
(1.3)
{u&&(t )} = −ω 2 {u}sinωt

Page 1-6
CEE490b

In which {u} is the column vector of amplitudes, which are independent of time.
Subsituting equation (1.3) into equation (1.2), yields:

− [m ]ω 2 {u&&}sin ωt + [k ]{u}sin ωt = {0}

The column vector of amplitudes is:

⎧ u1 ⎫
⎪u ⎪
⎪⎪ 2 ⎪⎪
{u} = ⎨ . ⎬ = [u1 u 2 . . u n ]T (1.4)
⎪ . ⎪
⎪ ⎪
⎪⎩u n ⎪⎭

This is a homogeneous algebraic equation for u i , where the frequency, ω is

unknown. This is called the “Eigenvalue Problem” The solution to the Eigenvalue
Problem utilizes the basic properties of homogeneous algebraic equations which
imply that the roots (unknowns) are nontrivial (i.e. not equal to zero) only if the
determinant of the coefficients vanishes.

([k ] − ω [m]){u} = {0}

2

The roots of this equation are non-zero only if the determinant is zero. i.e.:

[k ] − ω 2 [m] =0

if λ = 1/ ω 2 and we pre-multiply by − λ [k ] , then:

−1

(− λ[k ] [k ] + [k ] [m]){u} = {0}

−1 −1

(− λ[I] + [k ] [m]){u} = {0}

−1

where [I] is the Identity Matrix. A non-trivial solution only exists if the
determinant is equal to zero:

− λ [I] + [k ] [m ] = 0
−1

To find the values of λ which satisfy this equation results in a set of unique
Eigenvalues or Natural Frequencies. After the Eigenvalues have been determined,
they are substituted back into the Homogeneous Equation involving {u} . For each
value of λ (recall that λ = 1/ ω 2 ) or natural frequency, a complete set of

Page 1-7
CEE490b

dimensionless displacements are obtained, one for each degree of freedom. There
are the mode shapes associated with each mode of vibration, called Eigenvectors.

In our two-degree of freedom “flagpole” problem, the solution of these equations

results in a closed-form solution for ω 2 , as follows:

Recall that we now have a 2x2 Condensed Stiffness Matrix, [k ′], and the equation
for the characteristic determinant:

[k ′ ] − ω 2
[m ] = 0

becomes:

′ − m1ω 2
k 11 ′
k 12
=0
′
k 21 ′ − m 2ω 2
k 22

the determinant is then a quadratic equation in ω 2 :

′ − m1ω 2 )(k 22
∆ = (k11 ′ − m2ω 2 ) − k12
′ k 21
′ =0
′ + m2 k11
∆ = ω 4 m1m2 − ω 2 (m1k 22 ′ ) + k11
′ k 22
′ − k12
′ k 21
′ =0
⎛ k′ k′ ⎞ k′
ω 4 − ⎜⎜ 11 + 22 ⎟⎟ω 2 − 12 = 0
⎝ m1 m2 ⎠ m1m2

the solution of which is:

′ k 22
1 ⎛ k 11 ′ ⎞ ′ k 22
1 ⎛ k 11 ′ ⎞ k′
ω 1,2 = ⎜⎜ + ⎟⎟ ± ⎜⎜ + ⎟⎟ + 12
2

2 ⎝ m1 m2 ⎠ 4 ⎝ m1 m2 ⎠ m1m2

With each frequency, the amplitude ratios, or mode shapes can be calculated from:

([k ′] − ω [m]){u} = {0}

2

′ k12
⎛ ⎡k11 ′ ⎤ 2 ⎡m1 0 ⎤ ⎞⎧u1 ⎫ ⎧0⎫
⎜⎢ ⎥ − ω j⎢ ⎥ ⎟⎟⎨u ⎬ = ⎨0⎬
⎜ k′ k′
⎝ ⎣ 21 22 ⎦ ⎣ 0 m 2 ⎦ ⎠⎩ 2 ⎭ ⎩ ⎭

⎡k11
′ − ω 2 j m1 ′
k12 ⎤ ⎧u1 ⎫ ⎧0⎫
⎢ ⎥⎨ ⎬ = ⎨ ⎬
⎣ ′
k 21 ′ − ω j m2 ⎦ ⎩u 2 ⎭ ⎩0⎭
k 22 2

Page 1-8
CEE490b

This results in two equations and two unknowns:

(k ′
11 ) ′ u2 = 0
− ω 2j m1 u1 + k12
(
′ u1 + k 22
k 21 )
′ − ω 2j m 2 u 2 = 0

u2
if we define the relative displacement as a j = , then
u1

aj =
u2
=
( ′
ω 2j m1 − k11 )
= 2 12
k
and a1a2 =
m1
u1 k12 ( ′
ω j m2 − k 22 ) m2

Both of these equations must yield the same answer, which acts as a check on the
resulting mode shapes. With more degrees-of-freedom than two, it is desirable to
use a computer to solve the problem.

Page 1-9
CEE490b

Chapter 2 - Free Vibration of Multi-Degree-of-Freedom Systems - II

We can obtain an approximate solution to the fundamental natural frequency

through an approximate formula developed using energy principles by Lord
Rayleigh. As with single-degree-of-freedom systems, MDOF systems can also use
this approximation:

∑Q ui i ∑m ui i
ω 2
=g i
=g i

∑Q u ∑m u
1 2 2
i i i i
i i

where u i = the static deflection under the dead load of the structure Qi , acting in the
direction of motion, and g = the acceleration due to gravity. Thus, the first mode is
approximated in shape by the static deflection under dead load. For a building, this
can be applied to each of the X and Y directions to obtain the estimates of the
fundamental sway modes.

Fig. 2.1a) Deflection for Rayleigh’s Formula Applied to Buildings

Likewise for a bridge, by applying the dead load in each of the vertical and horizontal
directions, the fundamental lift and drag modes can be obtained. The torsional mode
can also be approximated by applying the dead load at the appropriate radius of
gyration and determining the resulting rotation angle.

Page 2-1
CEE490b

Fig. 2.1b) Deflection for Rayleigh’s Formula Applied to Bridges

Even when performing a detailed dynamic analysis using computer software like
SAP, ANSYS or ALGOR, a check using Rayleigh’s method is advisable. Often, for
most preliminary designs, a detailed dynamic analysis is not required and a first-
order analysis using Rayleigh’s method is all that is required.

Generalized Coordinates

Generalized coordinates are a means of simplification of a multi-degree-of-freedom

system into a series of equivalent single-degree-of-freedom systems.

ai cos ω i t
External Load Distribution
P ( z, t ) (kN/m)

y ( z, t ) = ai cos ω i tφ i ( z )

m(z)

Page 2-2
CEE490b

Continuous Structures can be idealized mode-by-mode in Generalized Parameters :

EQUIVALENT SDF SYSTEM

ai cos ω i t
*
K i

M i*
Pi * (t )

Di*

Generalized Mass:

H
M i* = ∫ m( z )φ i2 ( z )dz
0

Generalized Stiffness:

K i* = ω i2 M i*
Di* = Di = not generalized
ω i* = ω i = not generalized

Generalized Force:

H
Pi * (t ) = ∫ P ( z, t )φ i ( z )dz
0

The response of the actual structure in mode “i” is the same as that of its equivalent
SDF system in mode “i” when defined by its Generalized Properties – Stiffness,
Mass and Force.

Page 2-3
CEE490b

Orthogonality of Modes

Orthogonality of modes is a very important relationship between any two modes of

free vibration. It means that each mode is truly independent of another.

Mode 1 Mode 2

Recall that we found that the natural frequencies ω j and corresponding modes can
be determined algebraically:

⎧ u1 ( j ) ⎫
⎪u ( j )⎪
⎪⎪ 2 ⎪⎪
( )
[k ] − ω j [m] {u j } = 0, where {u j } = ⎨ . ⎬ the Eigenvectors or Mode Shapes
2

⎪ . ⎪
⎪ ⎪
⎪⎩u n ( j )⎪⎭
Eigenvalues
or Natural Frequencies

Writing this equation for two modes j and k, (for example the 1st and 3rd mode):

ω 2j [m ]{u j } = [k ]{u j } (2-1)

ω k2 [m ]{u k } = [k ]{u k } (2-2)

Now, transpose equation (2-1), and postmultiply by {u k }

(ω [m]{u }) {u } = ([k ]{u }) {u }

2
j j
T
k k
T
k

Then, because of the “Reversal Law”, ({[a][b]}

T T T
)
= [b ] [a ] , then this is also equal to:

ω 2j {u j } [m ] {u k } = {u j } [k ] {u k }
T T T T
(2-3)

Page 2-4
CEE490b

Matrices [m ] and [k ] are symmetric and so [m ] = [m ] and [k ] = [k ] . If we then

T T

premultiply equation (2-2) by {u j } :

T

ω k2 {u j } [m ]{u k } = {u j } [k ]{u k }
T T
(2-4)

We notice now that the right hand sides of equations (2-3) and (2-4) are equal and
therefore subtracting equation (2-4) from (2-3) yields:

(ω 2
j )
− ω k2 {u j } [m ]{u k } = 0
T

Since ω j ≠ ω k , then {u j } [m ]{u k } = 0 for j ≠ k

T
(2-5)

This is the Orthogonality Condition for mode shapes {u j }and {u k } including the mass
matrix. Then examining equation (2-4) using the orthogonality condition that results
from equation (2-5), we see that:

{u } [k ]{u } = 0 for
j
T
k j ≠k

This is the second Orthogonality Condition including the stiffness matrix.

Equation (2-5) when expanded, is of the form:

⎡m1 0 0 0⎤ ⎧ u1 ⎫
⎢0 m ⎥ ⎪u ⎪
{u1 u2 ... u n }j ⎢ 2 0 0⎥ ⎪⎨ 2 ⎪⎬ = 0
⎢0 0 . 0 ⎥⎪ . ⎪
⎢ ⎥
⎣0 0 0 mn ⎦ ⎪⎩u n ⎪⎭ k

And if one carries out the multiplication, the orthogonality condition involving mass is
obtained in the form:

∑m u u
i =1
i ij ik =0 for j ≠ k (2-6)

Multiplying Equation (2-6) by the natural frequency ω 2j and realize that ω 2j m i u ij is

the inertia force associated with mode j and hence ω 2j m i u ij u ik is force x
displacement or work. Then, equation (2-6) suggests that the total work done by
inertia forces of one mode on displacements of any other mode vanishes.

Page 2-5
CEE490b

In further considerations we will denote the modal displacements by φ ij and all

modes listed as columns of a square matrix [φ ] ,

⎡φ11 φ12 φ13 . φ1n ⎤

⎢φ φ 22 φ 23 . φ 2n ⎥⎥
⎢ 21
[φ ] = ⎢ . . . . . .⎥
⎢ ⎥
⎢ . . . . . ⎥
⎢⎣φ n1 φ n2 φ n3 . φ nn ⎥⎦
mode: 1st 2nd 3rd … nth

The modes are Orthogonal or Independent. We can examine some standard

trigonometric functions and their integrals for an analogy to the Orthogonality
Condition.

∫ cos jx cos kxdx ⎫⎪

⎪
∫ sin jx sin kxdx ⎬ Integrals involving products of harmonic functions.
⎪
∫ sin jx cos kxdx ⎪⎭
The trigonometric identities of sums are:

a) cos( x + y ) = cos x cos y − sin x sin y

b) cos( x − y ) = cos x cos y + sin x sin y

Adding a) and b) we obtain:

cos x cos y = 1
2
(cos( x − y ) + cos( x + y )) , so for x = y,
cos jx cos kx = 1
2
(cos( j − k )x + cos( j + k )x )
and for j ≠ k (analogous to different modes):

2π 2π
1⎡ 1 1 ⎤
∫0 cos jx cos kxdx = 2 ⎢⎣ ( j − k ) sin( j − k )x + j + k sin( j + k )x ⎥⎦ = 0
0
or, orthogonal.

and for j = k (analogous to the same mode):

2π 2π

∫ cos jxdx = ∫ (1 + cos 2 jxdx ≠ 0

2 1
2
or, finite
0 0

Page 2-6
CEE490b

Generalization of Orthogonality Conditions

It was found that between two different modes, where j ≠ k and ω j ≠ ω k that the
Orthogonality Condition is:

{φ } [m]{φ } = 0
j
T
k or ∑m φ φ i ij ik =0 (2-7)
i

∑ m φ ≠ 0 , because m > 0 and φ > 0 . Thus:

2 2
Now for j = k, i ij i ij
i

{φ } [m]{φ } = ∑ m φ = M , the generalized mass of the j th mode.

j
T
j i ij
2 *
j
i

The Generalized Mass, recall, is the equivalent “mass” of mode j if treated as a

single-degree-of-freedom system. More generally, for all modes:

[φ ]T [m][φ ] = [M* ], the diagonal matrix of Generalized Masses (2-8)

This can be verified by re-writing [φ ] in terms of partitioned matrices and treating the
sub-matrices that are created by this partitioning as elements if they are
conformable, as follows:

[φ ]T [m][φ ] symbolizes the following triple matrix product:

1st mode ⎧{φ1}T ⎫
⎪ ⎪
2nd mode ⎪{φ2 }T ⎪
⎪ ⎪
⎨ . ⎬ [m ] [{φ1} {φ2 } . . {φn }]
⎪ . ⎪
⎪ ⎪
nth mode ⎪⎩{φn }T ⎪⎭
nx1 1x1 1xn and is conformable, i.e.: n x n

⎧{φ1}T [m ]⎫ ⎡M1* 0 0 0 0 ⎤
⎪ T ⎪ ⎢ ⎥
⎪{φ2 } [m ]⎪ ⎢0 M 2* 0 0 0 ⎥
⎪ ⎪
⎨ . ⎬[{φ1} {φ2 } . . {φn }] = ⎢ 0 0 . 0 0 ⎥
⎪ ⎪ ⎢ ⎥
⎪ . ⎪ ⎢0 0 0 . 0 ⎥
⎪⎩{φn }T [m ]⎪⎭ ⎢0
⎣ 0 0 0 M n* ⎥⎦

nx1 1xn

Page 2-7
CEE490b

The second Orthogonality Condition involves the stiffness matrix:

{φ } [k ]{φ } = 0,
j
T
k when j ≠ k , or ∑k φ φ i ij ik =0
i
(2-9)

{φ } [k ]{φ } = K
j
T
j
*
j , when j = k , or ∑k φ i ij
2
≠0
i

Where K *j is the Generalized Stiffness (a 1 x 1 matrix or a scalar quantity). Using

Equation (2-9), a relation can be derived involving all modes, written as columns in
[φ ] :
[φ ]T [k ][φ ] = [K * ] = [ω 2j ][M * ] (2-10)

and if we look at equation (2-4), then K * [ ] is the Generalized Stiffness Matrix; ω j is

the j th natural frequency. To prove equation (2-10), [φ ] and [φ ] can be partitioned
T

according to the modes and then the matrices multiplied:

⎧{φ1}T ⎫
⎪ T ⎪
⎪{φ 2 } ⎪
[φ ]T [k ] [φ ] = ⎪⎨ . ⎪⎬ [k ] [{φ1} {φ2 } . . {φn }]
⎪ . ⎪
⎪ ⎪
⎪⎩{φ n }T ⎪⎭

nx1 1x1 1xn

⎧{φ1}T [k ]⎫ ⎡K 1* 0 0 0 0⎤
⎪ ⎪ ⎢ ⎥
⎪{φ 2 } [k ]⎪
T
⎢0 K 2* 0 0 0⎥
⎪
⎨ .
⎪
⎬[{φ1} {φ 2 } . . {φ n }] = ⎢ 0 0 [ ] [ ][ ]
. 0 0 ⎥ = K = ωj M
* 2 *

⎪ . ⎪ ⎢ ⎥
⎪ ⎪ ⎢0 0 0 . 0⎥
⎪⎩{φ n }T [k ]⎪⎭ ⎢0
⎣ 0 0 0 K n* ⎥⎦

nx1 1xn

With respect to equation (2-3), written for j = k:

K *j = {φ j } [k ]{φ j } = ω 2j {φ j } [m ]{φ j } = ω 2j M *j
T T

Page 2-8
CEE490b

Chapter 3 - Forced Vibration of Multi-Degree-of-Freedom Systems

We will be using modal analysis to solve problems involving Forced Vibration of
Multi Degree of Freedom Systems, so the “direct” method which follows is not
generally used. This is because the damping term adds a phase shift which
generally makes this method impractical for real situations where damping is
present.

Forced Undamped Vibration

Equations of motion due to external excitation are

readily obtained from the equations of free vibration
P (t) by adding excitation terms Pi (t ) to the right hand
n mn
side of the MDF equation:

m i u&&i + ∑ k ir u r = Pi (t ) (3.1)
r
P (t) mi
i
where i = mass 1,2,3 …. n and k ir is the term in the
stiffness matrix associated with the force at node r,
generated by a unit displacement at node i.
P (t) m1
1
In matrix notation, the equation of motion is:

[m]{u&&} + [k ]{u} = {P } (3.2)

We assume that the external forces are “harmonic”,

or of the form Pi (t ) = Pi sin ωt which, in matrix notation is {P } = {Po }sin ωt .

The Particular Solution provides the Steady Response and is {u(t )} = {u}sin ωt ,
where {u} is an amplitude vector, describing the amplitude of the individual
displacements. This acceleration is of the following form:

{u&&(t )} = −{u}ω 2 sinωt (3.3)

Substituting these expressions for displacement and acceleration in the equation of

motion yields:

− [m ]ω 2 {u}sin ωt + [k ]{u}sin ωt = {Po }sin ωt

( )
or [k ] − ω 2 [m ] {u} = {Po } (3.4)

Page 3-1
CEE490b

This is a set of nonhomogeneous algebraic equations for the unknown

amplitudes, {u} . The frequency, ω is given, since it is the frequency of excitation,
and so the resulting amplitudes of vibration can be calculated directly as a solution
of simultaneous linear equations, using standard software. A system with n-degrees
of freedom has n resonances. At resonances with the natural frequencies, ω = ω j ,
the amplitudes grow to infinite amplitudes when there is no damping present.

Forced Damped Vibration

In general, there are two types of damping that one has to examine with damped
vibrations.

RELATIVE DAMPING ABSOLUTE DAMPING

- Good for structural damping - Good for aerodynamic damping
- Depends only on inter-storey motion - Depends only on storey motion
- Damping force = c i x relative velocity - Damping force = c i x absolute velocity

The relative velocity is the velocity at station ( i ) - the velocity at station ( i-1)

Since absolute damping depends only on the absolute motion of each mass, the
damping matrix is diagonal. Since relative damping depends on the inter-storey
motion, then off-diagonal terms are present in the damping matrix. Similarly,
stiffness can also be relative, or absolute. In our previous example of the shear
building we had relative stiffness in the inter-storey columns

Page 3-2
CEE490b

c 2 , c 3 - Relative Damping k 2 , k 3 - Relative Stiffness

c1, c 4 , c 5 , c 6 - Absolute Damping k1, k 4 , k 5 , k 6 - Absolute Stiffness

We can expand the 3-storey shear building of last week to include both types of
stiffness and damping components. Recall that the stiffness constant for the
columns of the shear building was of the form:

12EI i
ki = ⋅ N ; where N was the number of columns per storey (3.5)
l3

Applying Newton’s second law to the individual masses (mass x acceleration = sum
of forces), the equations of equilibrium are:

For the first mass:

m1u&&1 = −(k 1 + k 4 + k 2 )u1 + k 2 u 2 − (c1 + c 2 + c 4 )u&1 + c 2u& 2 + P1 (3.6a,b,c)

For the second mass:

m2 u&&2 = k 2 u1 − (k 2 + k 3 + k 5 )u 2 + k 3 u 3 + c 2 u&1 − (c 2 + c 3 + c 5 )u& 2 + c 3 u& 3 + P2

For the third mass:

m3 u&&3 = k 3 u 2 − (k 3 + k 6 )u 3 + c 3 u& 2 − (c 3 + c 6 )u& 3 + P3

Page 3-3
CEE490b

We can use the double subscripted notation as before to further generalize the
equations for each mass:

n n
mi u&&i + ∑ k ir u r + ∑ c ir u& r = Pi (t ) for each of i=1,2,…n (3.7)
r =1 r =1

This in matrix form is:

[m]{u&&} + [c ]{u&} + [k ]{u} = {P} (3.8)

where:

⎡m1 0 0 ⎤ ⎧u1 ⎫ ⎧ P1 ⎫
⎪ ⎪ ⎪ ⎪
[m] = ⎢⎢ 0 m2 ⎥
0 ⎥ , {u} = ⎨u 2 ⎬ , {P } = ⎨P2 ⎬ (3.9)
⎢⎣ 0 0 m3 ⎥⎦ ⎪u ⎪ ⎪P ⎪
⎩ 3⎭ ⎩ 3⎭

⎡c1 + c 2 + c 4 − c2 ⎤ ⎡c11 c12 c13 ⎤

[c ] = ⎢⎢ − c 2 c2 + c3 + c5 − c 3 ⎥⎥ = ⎢⎢c 21 c 22 c 23 ⎥⎥ (3.10)
⎢⎣ − c3 c 3 + c 6 ⎥⎦ ⎢⎣c 31 c 32 c 33 ⎥⎦

⎡k 1 + k 2 + k 4 − k2 0 ⎤ ⎡k 11 k 12 k 13 ⎤
[k ] = ⎢⎢ − k 2 k2 + k3 + k5 − k 3 ⎥⎥ = ⎢⎢k 21 k 22 k 23 ⎥⎥ (3.11)
⎢⎣ 0 − k3 k 3 + k 6 ⎥⎦ ⎢⎣k 31 k 32 k 33 ⎥⎦

Note that both the damping and stiffness matrices are symmetric, since k ij = k ji and
c ij = c ji . The elements of the damping matrix are analogous to those elements of the
stiffness matrix. The element c ij is the force required at mass i (in the direction of
u i to produce a unit velocity at mass j, while the velocities at all other masses are
zero. The equations of motion are established by forming the stiffness, damping and
mass matrices for the whole structure.

Page 3-4
CEE490b

Chapter 4 - Modal Analysis

Modal analysis is a general method for analyzing the response of linear multi-
degree-of-freedom systems. It is particularly suitable for systems whose properties
are frequency independent. The method describes the response in terms of the
modes of free vibration whose orthogonality facilitates the solution. Therefore, the
analysis of the free vibration (the solution of the eigenvalue problem) must be
completed prior to the calculation of the response to external excitation.

The equations of motion can be written in the general form as:

[m]{u&&} + [c ]{u&} + [k ]{u} = {P } (4-1)

in which [m ] , [c ] and [k ] are mass, damping and stiffness matrices respectively;

{u} = the displacement vector and {P} = the vector of excitation.

The solution to equation (4-1) describing the response is sought in the form of a sum
of responses in individual modes of free vibration, Φ ij ,

n
u i (t ) = ∑ Φ ij η j (t ), i = 1,2 … n (4-2)
j =1

in which:
• Φ ij are modal coordinates of the jth mode. These modal coordinates are
independent of time and can be chosen to an arbitrary scale. [In systems with
distributed mass, ui(t) is replaced by u(x,t) and Φ ij by Φ j (x ) ];
• η j (t ) are new variables associated with mode j and depending on time. They
are called generalized coordinates.

To summarize:

The total response at node “i”

Is equal to:

the sum over all modes of:

{the shape function for node “i” and mode “j” x a modal scaling factor (the
generalized coordinate of mode “j”)}

Page 4-1
CEE490b

Equation (4-2) can be interpreted as Fig. 4.1 indicates.

j=1 2 3
Fig. 4.1 Response in terms of modes.

Equation (4-2) represents a coordinate transformation through which one set of n

coordinates can be replaced by another set of n independent coordinates. equation
(4-2) can be rewritten in matrix form to include all nodes, i = 1, 2..., n:

{u} = [Φ ]{η}, {u&} = [Φ ]{η&}, {u&&} = [Φ ]{η&&} (4-3)

where

⎡ Φ 11 Φ 12 Φ 13 ⎤
[Φ ] = ⎢⎢Φ 21 Φ 22 Φ 23 ⎥⎥ (4-4)
⎢⎣Φ 31 Φ 32 Φ 33 ⎥⎦
j=1 2 3

Each column in equation (4-4) represents one mode of free vibrations.

Substitute equation (4-3) into equation (4-1) and premultipiy by the transpose of [Φ ]
which is [Φ ] ; (this is a matrix in which modes are presented in rows)
T

a square matrix
64748
[1
Φ4]2
[m4
T
][3
Φ ]{η&&} + [Φ ] [c ][Φ ]{η&} + [1
T
Φ4]T2
[k4
][3
Φ ]{η} = [Φ ] {P (t )}
T
(4-5)
[M * ] [K * ]

Page 4-2
CEE490b

Equation (4-5) considerably simplifies due to the generalized orthogonality

conditions, discussed previously, according to which

[Φ]T [m][Φ] = [M *] (4-6a)

[Φ]T [k ][Φ] = [K *] = [ω ]2 [M *] (4-6b)

Hence, the two triple products result in two diagonal matrices which is very
advantageous because [M *]{η&&} and [K * ]{η }are column matrices. Therefore, each
equation (4-5), written in ordinary algebraic form, contains only one variable and its
second derivative, η&& j .

If it were not for the presence of damping, the equations would be uncoupled,
however, damping couples these equations. Clearly, it would be most desirable if the
triple matrix product:

[Φ ]T [c ][Φ ] (4-7)

containing the damping constants of the system, resulted in a diagonal matrix

because only then may each equation (line) contain only one derivative η& j . In such a
case, equation (4-5) represents a set of n independent equations for η j , j = 1,2...,n,
that are "uncoupled".

Since [Φ ] and [Φ ] , two multipliers, are the same in equation (4-7) and equation (4-
T

6), the triple matrix product, equation (4-7), can result in a diagonal matrix only when
the damping matrix, [c], is proportional to either the mass matrix [m] or the stiffness
matrix [k]. In the first case, [c] has to be diagonal and proportional to [m],

[c ] = 2α [m] (4-8a)

While the second case occurs if

[c ] = β [k ] (4-8b)

The factor 2 in equation (4-8a), is used for convenience, α , β = constants. (Recall

that in one degree of freedom the viscous damping constant c = 2α m and
α = Dω o ). equation (4-8a) substituted into equation (4-7), yields with respect to
equation (4-6a)

Page 4-3
CEE490b

[Φ ]T [c ][Φ ] = [Φ ]T 2α [m][Φ ] = 2α [M *] (4-9a)

And equation (4-8b) gives

[Φ]T [c ][Φ ] = [Φ]T β [k ][Φ ] = β [Φ]T [ω 2j ][m][Φ] = β [ω 2j ][M *] (4-9b)

Equation (4-8a) implies that only absolute dampers, may be present, every
damping constant, ci, is proportional to the mass mi and, finally, the proportionality
constant α is the same for all dampers.

Equations (4-6a) and (4-9a) substituted into equation (4-5) give

[M *]{η&&} + 2α [M *]{η&} + [1
ω 2j ][M *]{η } = [Φ ] {P (t )}
424 3
T
(4-10)
[K * ]

This is a set of n independent (uncoupled) equations. Each of them has the form

M jη&&j + 2αM jη& j + ω 2j M jη j = {Φ j } {P (t )}

T
(4-1 1)

or

p j (t )
η&&j (t ) + 2αη& j (t ) + ω 2j η j (t ) = , j = 1 ,2,3 … n (4-1 2)
Mj

This is identical to the single degree of freedom equation, in which

p j (t ) = {Φ j } {P (t )} = ∑ Φ ij Pi (t )
n
T
(4-13)
i =1

is the generalized force linked to generalized coordinate η j (i.e the Generalized

Force in mode j). Equation (4-12) are independent and each of them is exactly
equal to the equation of a single degree of freedom system and, therefore, can
easily be solved. Thus, whenever one degree of freedom can be solved, many
degrees of freedom can be solved too; η j are obtained from equation (4-12) and
substituted into equation (4-2), which was an expression for the motion at mass “i”:

n
u i (t ) = ∑ Φ ijη j (t ), i = 1,2 … n
j =1

The uncoupled generalized coordinates η j (t ) are also called normal coordinates.

Page 4-4
CEE490b

The damping constant α occurring in equation (4-12) is calculated for each mode as
α = D j ω j in analogy with one degree of freedom. The modal damping ratio, Dj,. can
be in some cases calculated, e.g. damping due to soil, fluids or air, in other cases
must be estimated.

With the damping matrix proportional to stiffness matrix (equation (4-8b)), uncoupled
equation (4-12) is again obtained with 2α replaced by βω 2j .

The decoupling of equations of motion can be interpreted as a result of the

external loads being in effect replaced by a system of fictitious (generalized)
loads such that each of them excites just one mode only.

Subscript i refers to a component of motion or to mass i; subscript j identifies the

mode.

The approach is general. With the number of masses ⇒ ∞ , a discrete system

changes into a distributed one. The only change in modal analysis is that n = ∞ and
summations in the equations for generalized mass and generalized force change to
integration:

l
M j = ∫ m( x )Φ 2j ( x )dx, the Generalized Mass in mode j,
0
l
p j = ∫ P ( x, t )Φ j ( x )dx, the Generalized Force in mode j,
0

Equations (4-12) and (4-2) remain unchanged.

Further solution depends only on the character of the external forces. The principal
types of excitation are discussed below.

If the damping is not proportional to either [k] or [m], the equations of motion can be
uncoupled using complex vibration modes. (see Novak, M. and El Hifnawy, L.,
"Effect of soil-structure interaction on damping of structures," J. of Earthquake Eng.
and Structural Dynamics, Vol. 11, 1983, pp. 595-621.)

4.1 HARMONIC EXCITATION

Assume harmonic excitation with frequency was in the case of unbalanced masses
of machines, vortex shedding etc. Such forces can be described as:

Pi (t ) = Pi cos ωt , or {P (t )} = {P }cos ωt (4-14)

Page 4-5
CEE490b

Where

{P} = [P1 P2 ... Pi ... Pn ]

T

The generalized forces for mode j, are from equation (4-13)

n
p j (t ) = cos ωt ∑ Φ ij Pi (4-15)
i =1
or

n
p j (t ) = L j cos ωt , L j = ∑ Φ ij Pi , (4-16)
i =1

the Force Participation Factor in mode j

The generalized equation of motion, equation (4-12), is

p j (t ) Lj
η&&j (t ) + 2αη& j (t ) + ω 2j η j (t ) = = cos ωt (4-1 7)
Mj Mj

This is an equation formally identical with that of single degree of freedom systems
whose mass is M j and which is subjected to harmonic excitation with amplitude L j .
This is called the Force Participation Factor because its magnitude is a measure of
the degree to which the excitation forces participate in the excitation of mode j.

The solution of equation (4-17) follows from the SDOF solution found previously,

η j (t ) = η j cos(ωt + φ j ) + η oj e −αt cos(ω j t + φ jo ) (4-18)

1442443 144424443
Steady State Transient

in which, the first term describes the most important steady state part of the motion
(the particular solution). Its amplitude is

Lj 1 Lj
ηj = = εj (4-19)
M jω 2j ⎡ ⎛ 2 ⎤2 2 M j ω 2j
⎞ ⎛ ⎞ 2
⎢1 − ⎜ ω ⎟ ⎥ + 4⎜ ω ⎟ Dj
⎢ ⎜⎝ ω j ⎟ ⎥
⎠ ⎦
⎜ω j
⎝
⎟
⎠
⎣

where ε j = the dynamic magnification factor in one degree of freedom whose natural
frequency is ω j . and damping D j . and M j ω 2j = K j = generalized stiffness. The phase
shift of the steady state component

Page 4-6
CEE490b

ω
2D j
ωj
φ j = − arctan 2
(4-20)
⎛ω ⎞
1− ⎜ ⎟
⎜ω ⎟
⎝ j ⎠

The transient part of equation (4-18) dies out due to damping; constants η oj and φ jo , if
needed, are given by initial conditions.

The real steady motion is from equation (4-2)

n n
u i (t ) = ∑ Φ ijη j = ∑ u ij cos(ωt + φ j ) (4-21)
j =1 j =1

where the amplitude in mode j is

L j Φ ij 1 L j Φ ij
u ij = = ε j = Φ ijη j (4-21a)
M jω 2
⎡ ⎛ω 2
2
⎤ 2 M j ω 2j
j
⎞ ⎛ ⎞ 2
⎢1 − ⎜ ⎟ ⎥ + 4⎜ ω ⎟ Dj
⎢ ⎜⎝ ω j ⎟
⎠ ⎥ ⎜ω
⎝ j
⎟
⎠
⎣ ⎦

Thus, the response in each displacement coordinate i consists of harmonic

components that have the same frequency ω but different amplitudes and phase
shifts. At resonance with mode r, ω = ω j = ω r , ε j = 1/ 2Dr ; the resonant amplitude of
the resonating mode and its phase shift are

L j Φ ir
1
u ir = , φ r = −π / 2 (4-22)
M r ω 2Dr 2
r

With small damping, the resonant amplitude is usually much larger than the non-
resonant amplitudes of the other modes and equation (4-22) is sufficient to estimate
the resonant amplitudes of the system.

With large damping the contribution from all modes may be more significant. At
resonance with the first natural frequency, i.e. for ω = ω1 , the phase shifts,
evaluated from equation (4-20), take on the values φ1 =-90° and φ 2,3,..... ≅ 0 .
Consequently, the first resonance amplitude u i = u i (ω1 ) becomes approximately

Page 4-7
CEE490b

2
⎛ n ⎞
u i = u + ⎜⎜ ∑ u ij ⎟⎟ , the total response at mass “i”, when ω = ω1
2
i1
⎝ j =2 ⎠

Similar approximate expressions can be written for the other resonances, r,

realizing that φ1 ≅ −180° , φ r ≅ −90° and φ i ≅ 0° for i > r (see Fig 4.2).

The superposition of the responses in individual modes is shown in Fig.4.2. Because

of the phase differences of the modal components, the resultant amplitude:
u i ≤ u i 1 + u i 2 + u i 3 + ........

Fiq. 4.2 Superposition of responses in individual modes.

Page 4-8
CEE490b

When using the modal analysis, the coordinates of the modes (eigenvectors) can be
chosen to an arbitrary scale. This can be seen from equation (4-21a) and from the
application of this formula to one degree of freedom.

In one degree of freedom

ω 2j = k / m, Lr = ΦP, M r = mΦ 2 , Φ ir = Φ

and the resonant amplitude from equation (4-22).

Lr Φ ir 1 Φ PΦ m P 1 1
u= ⋅ = = = u st
M r ω r 2Dr mΦ k 2D k 2D
2 2
2D

as found previously.

Before the modal analysis is started, the modes are sometimes normalized in such a
way that Φ nj = 1or M *j = 1 for each mode. In the latter case, the modes so
normalized are called orthonormal modes. Their coordinates are

Φ ij
Φ ij =
M *j

and the generalized mass

n
M j = ∑ m i Φ ij2 = 1
*

i =1

This normalization is used by some writers but it does not offer any particular
advantage.

The modal coordinates generally have the dimensions of the displacements, i.e.
translations or rotations; if only translations are involved, they can be taken as
dimensionless in which case the generalized coordinates assume the dimension of
length and the generalized masses are in kg (or slugs).

Page 4-9
CEE490b

Problem 4.1: Consider the five storey shear building analyzed in the tutorial.

The elevator drive produces a harmonic force acting on the 4th floor,

P4 (t ) = P4 cos ωt
P4 = 1.5 kN
The frequency ω of the drive is variable.

a) Find the resonant amplitudes of the top (5th) floor at natural frequencies:
ω = ω1 , ω = ω2 , ω = ω3
and also amplitudes at the operating speed of the elevator motor
ω = 21 (ω 3 + ω 4 )
The Damping ratio for all modes is D j = 0.01 and recall that, α j = Dω j

b) Evaluate also the physiological effects using Fig. 4.3.

Page 4-10
CEE490b

Fig. 4.3 Human Susceptibility to Vibration (after Reiher & Meister)

Page 4-11
CEE490b

Static Loading

Static loading can also be conveniently solved by means of modal analysis as a

special case of harmonic excitation. The pertinent formulae follow from the
preceding paragraph with the frequency of excitation ω ⇒ 0 `and thus cos ω t = 1.
With ω = 0 , loads Pi(t) = Pi and φ = 0 . From equation (4-19), the generalized
coordinate of mode j is:

Lj
η i ,st = (4-23)
M j ω 2j

and from equation (4-21), the static displacement of mass mi due to static loads

n Lj
u i ,st = ∑ Φ ij (4-24)
j =1 M j ω j
2

This is an exact approach, suitable to examine the effect of static loads in statically
indeterminate structures if the free vibration modes are known already from previous
analysis.

This approach can also be used to find the response to

a suddenly applied static load or a rectangular pulse
whose duration TP is sufficiently longer than the natural
period of the structure T1. In these cases, the maximum
(peak) response ≤ 2 x the static response.

Fig. 4.4
Examples:

4.2 The two storey shear building is exposed to

static wind load in the horizontal direction
given as P1 = 30 kN , P2 = 20 kN. Compute
the displacements u1, u2 and the stresses in the
columns .
Fig. 4.5

4.3 The five storey shear building is being designed for the Toronto area. Its width
is W = 24 m (tributary width for one frame = 6 m ). Calculate the wind loads
and deflections of the building, the stresses in the lowest columns and
maximum acceleration for return period 30 years, exposures A and C, and
damping ratio 1%.

Page 4-12
CEE490b

4.2 RESPONSE TO GROUND MOTION

Differential equations of motion depend on the nature of damping. The more

important case is that of relative damping.

Relative damping. - From Newton's second

law, forces acting on mass, mi are

n
mi U&&i = − ∑ k ir (U r − u g ) − c i (U& i − u& g )
r =1

where the absolute displacement

U i = u g + u i and the relative displacement
u i = U i − u g . With the damping force at
mass mi described as

− c i (U& i − u& g ) = −2mi α (U& i − u& g )

= −2mi αu& i

Fig. 4.6

The differential equations of the motion in terms of relative coordinates u become –

n
mi u&&i + 2mi α u& i + ∑ k ir u r = −mi u&&g , i = 1,2,3, …. n (4-25)
r =1

This is formally equal to equation (4-1) in which the equivalent exciting forces

Pi (t ) = −mi u&&g , i = 1,2,3, …. n

The damping satisfies equation (4-8a) and, therefore, modal analysis leads to
decoupled equations for generalized coordinates η i . The generalized force (the sign
can be omitted) is:

p j (t ) = ∑ Pi (t )φ ij
i
or

Page 4-13
CEE490b

n
p j (t ) = u&&g ∑ mi Φ ij = u&&g L j
i =1
n
where the earthquake participation factor L j = ∑ mi Φ ij
i =1

In matrix form, equations (4-25) can be written as

[m] {u&&} + [c ] {u&} + [k ]{u} = −[m]{1}u&&g (4-25a)

in which [c ] = 2α [m ] and {1} = [1 1 1 ... 1]

T

Equation (4-25a) can also be derived directly realizing that the inertia forces stem
from the absolute displacements Ui but the stiffness and damping forces are due to
the relative displacements ui. Then,

[m] {U } + [c ] {u&} + [k ]{u} = {0}

Eliminating U, equation (4-25) is obtained.

The relative displacement of mass i is, by equation (4-2),

n
u i (t ) = ∑ Φ ij η j (4-25b)
j =1

in which the generalized coordinate is given by equation (4-12),

p j (t ) Lj
η&&j + 2αη& j + ω 2j η j = = u&&g (t ) (4-26)
Mj Mj

This is analogous to one degree of freedom in which the solution depends on the
type of ground motion.

Transient ground motion produces response whose solution in 1 DOF is given by

the Duhamel (convolution) integral:

t
1
y (t ) = ∫
mω o 0
P (τ ) e −Dω o ( t −τ ) sin ω o (t − τ )dτ

With P (τ ) = L j u&&g (τ ) , substitution gives for the generalized coordinate:

Page 4-14
CEE490b

t
1 Lj −D ω ( t −τ )
η j (t ) = ∫
ωj Mj 0
u&&g (τ ) e j j sin ω j (t − τ )dτ

1 Lj
= V j (t )
ωj Mj

where

t
−D j ω j ( t −τ )
V j (t ) = ∫ u&&g (τ ) e sin ω j (t − τ )dτ (4-27)
0

The complete response is the sum of responses in individual modes, i.e.,

n
1 Lj
u i (t ) = ∑ Φ ijV j (t ) (4-28)
j =1 ω j M j

Integral V j (t ) = f (u&&g ,ω j , D j ) is exactly the same as in one degree of freedom and can
be obtained by numerical integration. V(t) has the dimension of velocity, (i.e. m/s).
equation (4-28) can be interpreted as Fig. 4.7 indicates.

4.3 SPECTRAL APPROACH TO TRANSIENT MOTION

It is usually not necessary to find the complete time history of the response. The
maximum response is decisive in most applications and this is obtained by
substituting the maximum value of integral Vj(t) into equation (4-28) using the same
notation as in one degree of freedom.

Vmax = Sv = spectral velocity (pseudo velocity)

Vmax Sv Sa
Sd = = = = spectral displacement
ωj ωj ω 2j
Sa = ω j Sv = spectral acceleration.

Spectral velocities can be computed for any typical earthquake. In Fig. 4.7 , the
spectral velocity of El Centro earthquake is given. Smoothed or averaged spectra
should be used possibly depending on site conditions (Figs. 4.8 and 4.9). In terms of
spectral displacement, the maximum (peak) displacement in the jth mode is:

1 Lj Lj
uˆ ij = Φ ij SV ( j ) = Φ ij Sd ( j )
ωj Mj Mj

Page 4-15
CEE490b

Fig. 4.7 Smoothed El Centro spectra reduced to a maximum

acceleration of 20%g

Page 4-16
CEE490b

Fig. 4.8 Average acceleration spectra for different site conditions

Fig. 4.9 84 percentile acceleration spectra for different site conditions

(after Seed, Ugos and Lysmer, 1974)

Page 4-17
CEE490b

The effective acceleration for mode j is

⎛ 1 Lj ⎞ Lj
u&&ij = ω 2j uˆ ij = ω 2j ⎜ Φ ij Sv ( j ) ⎟ = Φ ij Sa( j )
⎜ω M ⎟ M
⎝ j j ⎠ j

The stresses can be computed with a static load equal to the maximum effective
earthquake force acting on mj in mode j defined as:

Lj
q ij = m i x (accel. Amplitude) = mi ω 2j uˆ ij = m i Φ ij Sa( j )
Mj

The total maximum base shear which is a measure of earthquake loading is, for
mode j:

n Lj n L2j
Q j = ∑ q ij = Sa ( j ) ∑ m i Φ ij = Sa( j )
i =1 Mj i =1 Mj

The total response is a sum of responses in individual vibration modes (Fig. 4.10).

Fig. 4.10

The peak in individual modes does not appear at the same time. Its accurate value
could be obtained from the resultant time history. Approximately, the maximum
response of mass mi is:

Page 4-18
CEE490b

uˆ i ,max ≅ uˆ i2,1 + uˆ i2,2 + uˆ i2,3 + ..... = ∑ uˆj

2
i, j

One can write similar expressions for all types of responses:

e.g. Base Shear: Qmax = ∑Q j

2
j

Stress at location “i” σ i ,max = ∑σ j

2
i, j

With harmonic motion of the ground:

u g (t ) = u g cos ωt , u&&g (t ) = −u g ω 2 cos ωt

and loads:

n n
Pi (t ) = −mi u&&g (t ) = m i ω 2 u g cos ωt ; p j (t ) = ∑ Pi (t )Φ ij = ω 2u g cos ωt ∑ mi Φ ij
i =1 i =1

n
Steady state response is from equations (4-19) and (4-21), denoting L j = ∑ mi Φ ij ,
i =1

n L jω 2 1
u i (t ) = u g ∑ Φ ij cos(ωt + φ j )
j =1 M jω 2
⎡ ⎛ω 2
⎤
2 2
j
⎞ ⎛ ⎞
⎢1 − ⎜ ⎟ ⎥ + 4⎜ ω ⎟ D 2j
⎢ ⎜⎝ ω j ⎟
⎠ ⎥ ⎜ω ⎟
⎝ j⎠
⎣ ⎦

The resonant amplitude in mode j is;

Lj 1
ui (r ) = ug Φ ij
Mj 2D j

4.4 MODAL EQUIVALENT MODEL

In earthquake engineering, the "modal equivalent model" is sometimes used to

represent the structure. This model comprises n single degree of freedom systems
whose damping and natural frequencies are identical to those of individual vibration
modes and the masses and their heights are so determined that identical base
moments and base shears are obtained from both mode j and the equivalent model
(Fig.4.11).

Page 4-19
CEE490b

mn
m j S a(j)
φ ij
mn mj

m2
q ij
Hj
m1 hi

Mode j Equivalent 1 DOF Model

Fig. 4.11 Structural response in mode j and its equivalent 1 DOF model

Thus, for the base shear:

L2j
Qj = Sa( j ) = m j Sa( j )
Mj

and from here the equivalent mass is:

L2j
mj =
Mj

The equality of base moments (overturning moments) requires:

∑q h
i =1
ij i = m j Sa( j ) H j

which yields

Page 4-20
CEE490b

n Lj
∑m
n

∑ q ij hi i =1
i
Mj
Φ ij Sa( j ) hi
Hj = i =1
2
Mj = Mj
L Sa( j )
j L2j Sa ( j )

After abbreviation the equivalent height is:

n

∑m Φ i ij hi
Hj = i =1

Lj

All the modes can be represented as Fig.4.12 indicates.

mj

Hj

Fig. 4.12 Representation of an n-degree of freedom structure by n 1 DOF systems

Page 4-21
CEE490b

Problem 4.4

a) Calculate the earthquake response, i.e. displacements, earthquake forces,

base shear and stresses in the lowest columns of the five storey shear
building due to the 20% El Centro. Consider all modes and damping ratio 2%.

b) Calculate the response of the top floor and the base shear of the same five
storey building considering the four average acceleration spectra for different
site conditions (Fig.4.8), maximum ground acceleration 20% g, damping 5%
and the first mode only.

Problem 4.5

Calculate the resonant amplitudes of the five storey shear building due to harmonic
horizontal ground motion u g (t ) , whose amplitude is 0.01 in and frequency ranges
from 0 to 1.2 ω 5 . Assume D j = 0.01.

Problem 4.6

Adjust the Duhamel integral for numerical integration by taking all factors containing
t in front of the integration sign and prepare a flow chart for the computation of the
complete time history ui (t) from equation (4-28).

Answers to Problem 4.4a: (for l 1 = 4.00m )

Amplitudes: u51 = 4.8x10-3 m f1 = 5.44 ω 1 = 34.16

u52 = l.6xl0-4 m f2 = 13.90
u53 = 2.2x10-5 m

Base shear: Q1 = 424.82 kN

Column stress: σ 1 = 36.45 MPa for outside column

(first mode) σ 1′ = 42.53 MPa for inside column

Page 4-22
CEE490b

Chapter 5 - Response To Random Loads

5.1 General

A random process differs from the deterministic processes dealt with in the
preceding chapters in that it cannot be accurately predicted mathematically even if
the past time history is known. Such a process is most meaningfully described in
statistical terms. The basic statistical characteristics are reviewed first, presuming
that the random process x(t) may represent loading of the structure or its response.

A random process may be given in the form of one representative time history
(Fig. 5.1) or by a set of sample functions collected into an ensemble (Fig. 5.2).
When the statistical characteristics are extracted from the one long time history,
the procedure involved is called temporal averaging. The analysis of the
ensemble is known as ensemble averaging. Ideally, the time histories should
extend from t → −∞ to t → +∞

5.2 Basic Statistical Characteristics

Probability density function, p(x ) . This function defines the probability that x will
have a value in the range from x to x+dx. The typical bell-like shape of this function
is indicated in Fig. 5.3. Because all values may occur:

+∞
∫−∞
p( x )dx = 1 (5-1)

Probability distribution function, P(x). This function defines the probability of x

being smaller than or equal to a certain value α and thus:

a
P ( x ) = Pr( x ≤ a ) = ∫ p( x )dx (5-2)
−∞

This function is characterized by an S-like shape and is always bounded by the

limits 0 < P ( x ) < 1 (Fig. 5.4).

Mean value or expected value. This is the mean or average value of the
function and can be defined for the function x(t) as:

1 +T
x=
2T ∫
−T
x (t )dt (5-3)
T →∞

Page 5-1
CEE490b

Fig. 5.1 A random function

Fig. 5.2 Ensemble of samples of a random function

Fig. 5.3 Probability Density Fig. 5.4 Probability Distribution

Page 5-2
CEE490b

Fig. 5.2a Typical Signatures of wind-Induced Accelerations of a Tall Building

Page 5-3
CEE490b

The above components of acceleration can be combined

to estimate the peak resultant acceleration as follows:

aˆR ≤ aˆ x2 + aˆ y2 + r 2θ&&2

Summary of Fundamental Periods

Mode Analytical During Low Winds During High Winds
Estimate Avg. (sec) C.O.V. (%) Avg C.O.V. (%)
(sec) (sec)
1 N-S 9.0 6.26 5.6 7.16 1.4
2 E-W 7.25 6.32 2.4 6.89 1.9
3 Torsion 5.5 4.75 4.0 5.69 3.1

Fig. 5.2b) Observed Accelerations and Corresponding Power Spectra of a 43-

Storey Building as well as Fundamental Periods

Page 5-4
CEE490b

Realizing that all values have a total probability of occurring equal to 1, the
mean value can also be expressed using the probability density function as

+∞
x = ∫ xp( x )dx (5-4)
−∞

Thus, the value x can be viewed as the coordinate of the centroid of the area
under the curve p(x).

Finally, the mean value can be obtained from the ensemble of samples by
cutting through the ensemble at a certain time, e.g. t1, and calculating the
average value of the samples (Fig. 5.2). This gives

1 n
x = lim ∑ x i (t 1 )
n →∞ n i =1
(5-5)

Other notations used are x = E ( x ) =< x > .

Mean-square value is the average value of x2(t), is denoted, as

x 2 = E ( x 2 ) =< x 2 > and follows from the time history of x(t) as

1 +T
x2 =
2T ∫
−T
x 2 (t )dt (1st moment about the origin) (5-6)
T →∞

or from the ensemble as

1 n 2
x 2 = lim
n →∞ n
∑
i =1
x i (t 1 ) (5-7)

Variance is the mean-square value of the difference from the mean. Its notations
are σ x2 = ( x − x ) 2 and its value is:

1 +T
σ x2 =
2T ∫
−T
( x − x ) 2 dt (2nd moment about origin) (5-8)
T →∞
or from the probability density function:

+∞
σ x = ∫ ( x − x ) 2 p( x )dx (5-9)
−∞

Standard deviation is the square root of variance and is, therefore, also called
the root-mean-square value of x, or briefly r.m.s. Standard deviation is usually
denoted as:

Page 5-5
CEE490b

σ x = σ x2 = ( x − x ) 2 (5-10)

With regard to equations (5-1) and (5-9), the standard deviation can be viewed as
the radius of gyration of p(x) about x .

Standard deviation is an important magnitude and together with the mean value
x are the most important parameters which can characterize a probability
distribution function.

Gaussian or Normal probability density and distribution functions are defined as

( x − x )2
−
1 2σ x2
p( x ) = e (5-11)
2π σ x

and

( x − x )2
−
1 x
∫
2σ x2
P( x ) = e dx (5-12)
2π σ x −∞

This distribution is called “Normal” because it fits most natural phenomena. It

can be shown that a force described by a normal distribution produces response
of linear systems which also has a normal distribution.

- ∞ 84%

Fig. 5.5 Normal Distributions with Fig. 5.6 Probabilities associated with
x = a and different values of σ multiples of standard
deviation σ = sd

Page 5-6
CEE490b

The normal distribution is defined by the eman value, x , and standard deviation
σ x = σ . The magnitude of σ indicates the spread of the values of x about the
mean (Fig. 5.5). The dimensionless measure of this spread is the coefficient of
variation defined as the ratio σ / x .

The normal distribution is widely used and various probabilities following from it
are tabulated. For example, the probability of x being smaller than or equal to
the mean plus one standard deviation is 84%, i.e.

P ( x ≤ x + σ ) = 84%

This probability characterizes the pseudovelocity spectra plotted in Fig. 4.9. Other
probabilities are indicated in Fig. 5.6.

A given probability density of the process describes the percentage (proportion)

of time for which x takes on values in a certain range. However, it does not
provide any information on the rate of change in x(t) i.e., on the frequency
characteristics of the process. A more complete description of a random process
is contained in further statistical characteristics called correlation functions and
power spectral densities.

Autocorrelation function or more briefly correlation function is defined as the mean

value of the product of x(t) and x(t + τ ) where τ is a time lag, i.e.

1 +T
R x (τ ) = x (t )x (t + τ ) =
2T ∫−T
x (t )x (t + τ )dt (5-13a)
T →∞

From the ensemble (Fig. 5.2), the correlation function is obtained as

1 n
R x (τ ) = lim
n →∞ n
∑
i =1
x i (t ) x i ( t + τ ) (5-13b)

By evaluating the average products for different values of τ , the correlation

function of a process is established. The correlation function is even, droops
either in a smooth or oscillatory way and has the following properties:

d
R(0) = x 2 , R(∞ ) = 0, R(τ ) = R( −τ ), R (0 ) = 0
dτ

Since τ is real time in seconds, the correlation function indicates the speed with
which the correlation of the process diminishes and the time within which it vanishes
(Fig. 5.7).

Page 5-7
CEE490b

Fig. 5.7 Typical Autocorrelation Functions Fig. 5.8 Two-sided and one-sided
Power Spectral Densities

The above statistical characteristics allow for a few more definitions.

If the statistical characteristics are independent of the reference time t 1 (Fig. 5.2) the
process is called stationary: if they depend on time t 1 the process is nonstationary.
The discussion here is limited to stationary processes.

If the process is stationary and the temporal averages are equal to the ensemble
averages, the process is ergodic.

A centric process is a stationary process with x = 0. Covariance is one of the

measures of the extent to which two random variables x (t) and y (t) are related to
each other or correlated. Covariance is defined as

σ xy
2
= x(t )y (t ) =< x(t )y (t ) > (5-14)

When x(t) and y(t) are completely independent σ xy

2
=0;
σ xy
2
≠ 0 indicates correlation between the two variables, e.g. input force and
response.

In stationary processes the mean value x is a constant and it is, therefore, (in
favour of numerical accuracy to separate the mean value from the process and
analyze just the fluctuating random part of it), x ′(t ) = x(t ) − x . If x(t) is a load, x is
its static component which produces static deflection about which the structure
oscillates due to the effect of the fluctuating component x ′(t ) . Thus, the statistical
analysis can be limited to the fluctuating component because R x (τ ) = R x ′ (τ ) + x 2 .

Page 5-8
CEE490b

Power Spectral Density, S (f). This function describes the energy distribution of
the process with regard to frequency and is defined as the Fourier Transform of
the autocorrelation function R(τ ) , i.e.

∞
S(f ) = ∫ R (τ )e −i 2πfτ dτ (5-15)
−∞

This transformation yields an even, two-sided power spectrum indicated in Fig.

5.8. Because negative frequencies do not have a technical meaning, it is usually
preferable to define a one-sided power spectrum for positive frequencies only.
The area under both types of the spectra has to be the same and thus, the
magnitude of the one-sided spectrum is twice the magnitude of the two-sided
spectrum (Fig. 5.8). Splitting the integration interval into two, − ∞ to 0 and 0 to
+ ∞ and recalling that R(τ ) = R( −τ ) , the Fourier transform of the autocorrelation
function reduces to a cosine Fourier transform and the one-sided power spectrum
becomes

∞
S(f ) = 4 ∫ R (τ ) cos 2πfτdτ (5-16)
0

The inverse Fourier transform of the spectrum yields the

correlation function,

∞
R (τ ) = ∫ S(f ) cos 2πfτdτ (5-17)
0

The Fourier transform pair defined by equations (5-16) and (5-17) is also known
as the Wiener-Khintchin relationship.

The term power spectrum stems from electrical applications in which it has the
following physical meaning. Assume that x(t) is random voltage filtered through a
narrow band filter and that the power passed by the filter is measured. Then, this
power is proportional to the bandwidth of the filter and the spectral density of x(t)
at the centre frequency of the filter. The total power describes the variance of the
signal and thus,

∞ ∞
x 2 = ∫ S x (f )df = ∫ S x (ω )dω (5-18)
0 0

Expressing dω = 2πdf , equation (5-18) gives

S(f ) = 2πS(ω ) (5-19)

Equation (5-18) defines the most important property of the spectrum and also
suggests the dimension of a spectrum, because S(f )df must have a dimension of
x2. Consequently, SX(f) is in (dimension of x2)/frequency. Thus if x is

Page 5-9
CEE490b

displacement, Sx ( f ) is in m2/s-1 = m2s; the power spectrum of acceleration

is similarly m /s /s = m / s-3
2 4 -1 2

Other forms of power spectra used are the normalized spectrum and the
logarithmic spectrum.

Normalized spectrum S'(f) is defined as S ′(f ) = S(f ) / σ 2 and thus S(f ) = σ 2S ′(f )
and

∞
∫0
S ′(f )df = 1 (5-20)

Logarithmic spectrum is S l = fS(f ) / σ 2 , it is dimensionless and because

1
d log e f = df ,
f

∞ fS(f )
∫0 σ2
d log e f = 1 (5-21)

The relationship between S ′(f ) and S l (f ) is shown in Fig. 5.9.

Fig. 5.9 Relationship between (a) - normalized spectrum and (b) - logarithmic
spectrum

When the process x (t ) = const . x ′(t ) , the spectrum is, with regard to equations
(5-13) and (5-15)

S x (f) = (Const ) S 'x (f)

2
(5-21a)

SP (f) = m 2 S y&&g (f) (5-22)

where S y&&g (f ) is the spectrum of ground acceleration.

Page 5-10
CEE490b

Examples of Random Processes. - Examples of typical random processes

and their comparison with a deterministic harmonic process are shown in
Fig. 8.10. Mathematical expressions for some correlation functions and the
corresponding power spectra can be found in Ref. 10. Power spectra of a few
earthquake ground motions are plotted in Fig. 5.11.

x(t) p(x) Sx(f) Rx(τ)

Fig. 5.10 Typical random processes and their description in terms of probability
density, power spectral density and autocorrelation function.

L/ V 10
SV (f) = 4ΚV102
(2 + f 2 )5/6

Κ = surface roughness factor = 0.08 to 0.14,

L = scale length = 1200m,
V10 = mean wind velocity 10 m above ground
in m/s and f = fL / V 10 .

Fig. 5.10a Power spectral density of wind velocity fluctuations Sv(f) and schematic
of fluctuations in wake (Davenport, Harris)

Page 5-11
CEE490b

αg 2
S(ω ) = exp[−0.74(ω o / ω ) 4 ]
ω 5

where:
α = 8.1 x 10-3,
g = 9.81 m/s 2 and
ω o = g/U = frequency of spectrum peak

Fig. 5.10b Pierson-Moskowitz spectrum of sea surface elevation as function of

wind speed U.

Page 5-12
CEE490b

Fig. 5.11 . Power Spectra of Earthquake Ground Acceleration

Page 5-13
CEE490b

5.3 Response to Random Load in one Degree of Freedom

Relationship between Input and Output

A periodic force P(t) with period T = 1/ f1 can be represented by a complex Fourier

series as:

∞
1 T /2
P (t ) = ∑ c r e ir 2πf1t , c r = ∫ P (t )e −ir 2πf1t dt , r=1,2 ….. (5-23)
−∞ T −T / 2

The response of a SDF system to such a load can be obtained by means of

superposition of responses to individual components r in terms of the frequency
response function (admittance). The harmonic load

P (t ) = Po e iωt

yields response

Po
y (t ) = H (ω )e iωt
k

where

1
H (f ) = 2
(5-24)
⎛f ⎞ f
1 − ⎜⎜ ⎟⎟ + i 2D
⎝ fo ⎠ fo

However, the response to a series of harmonic loads (equation 5-23)

∞
1
y (t ) = ∑ c r H (f r )e ir 2πf1t , with f r = r f1 (5-25)
−∞ k

The mean square response can be expressed in terms of Parceval's

∞
y = ∑ cr
2 2
(5-26)
−∞

1
because (5-25) is again a Fourier series with amplitudes c r H (f r ) . Hence
k

∞
1
∑c
2 2
y2 = r H (f r ) (5-27)
k2 −∞

Page 5-14
CEE490b

realizing that

2 2 2
y 1y 2 = y1 y 2

A non-periodic (random) force can only be expressed in the above manner if

period T is extended to ∞ , thus from (5-27)

∞
1
∑ cr
2 2
y2 = lim 2 H (f r ) (5-28)
k 2 T →∞ 0

2 2
as c r and H (f r ) are even functions. Substitution for cr from 5-23 gives

2
∞ T /2
1 2
y = 2 lim ∑ 2 ∫ P (t )e
− ir 2πf1t 2
2
dt H (f r )
k T →∞ 0 T −T / 2

With period T → ∞

∞ ∞ T /2
1/ T → df , ∑ → ∫, ∫ P (t )e
− ir 2πf1t
dt → A(if ) as rf1 = f r → f and H (f r ) → H (f )
0 0 −T / 2

Also, the mean square response can be expresses by means of its power spectrum,
∞
y = ∫ S y (f )df . Hence:
2

∞ ∞
1 ⎡2 2⎤
∫0 S y (f )df = k 2 ∫ lim ⎢⎣T
2
A(if ) ⎥ H (f ) df
0
T →∞
⎦

As

⎡2 2⎤
lim ⎢ A(if ) ⎥ = S p (f )
T →∞ T
⎣ ⎦

i.e. the spectrum of the excitation P(t), the relation between the spectrum of the input
and the spectrum of the output is:

1
S p (f ) H (f ) = α (if ) S p (f )
2 2
S y (f ) = (5-29)
k 2

Page 5-15
CEE490b

2
in which H (f ) is the square of the modulus of the admittance function, which is
equal to the square of the dynamic magnification factor and is:

1
H (f ) = ε 2 =
2

[1− (f / f ) ]
o
2 2
+ 4D 2 (f / fo ) 2
(5-30)

The relationship between the input and the output described by equation 5-29 is
shown in Fig. 5.12.

Fig. 5.12 The relationship between spectrum of input and spectrum of output

Page 5-16
CEE490b

With response spectrum defined by equation 5-21, the mean square response y 2 is

∞ ∞
S (f ) H (f ) df
1
y 2 = ∫ S y (f )df = 2 ∫ p
2
(5-31)
0 k 0

or in terms of circular frequency

∞ ∞
dω
y 2 = ∫ 2πS y (ω )
2π ∫0
= S y (ω )dω
0

From equation 5-31 the rms displacement having the dimension of amplitudes, is
y 2 = σ y . The only complication is that the integral in equation 5-31 cannot be
generally evaluated in closed form. It can be evaluated approximately as:

f ∞
1 o
y ≅ 2 ∫ S p (f )df + 2 ∫ S p (fo ) H (f ) df
2 1 2

k o k 0
f
(5-32)
1 o 1 πf
= 2 ∫ S p (f )df + 2 S p (fo ) o
k 0 k 4D

2
This approximate evaluation is based on replacing H (f ) by unity for frequencies
from 0 to fo and on replacing the force power spectrum SP (f) by a constant (white)
spectrum SP (f o) whose magnitude is equal to the force spectrum for the natural
frequency of the system, fo (Fig. 5.13). The first part of equation 5-32 is called the
background effect and the second part the resonant effect.

If greater accuracy is needed, the integral in equation 5-31 can be evaluated using
the theory of residua or numerical integration.

Fig. 5.13 Approx. evaluation of response Fig. 5.14 Peak of random response

Page 5-17
CEE490b

When the damping is small and the spectrum flat, the second part of equation
5-32 yields sufficient accuracy and thus, the variance of the response is
approximately

1 πf 1 π ωo 1 π ωo
y2 ≅ SP (fo ) o = 2 2π SP (ω o ) = 2 SP (ω o ) (5-33)
k 2
4D k 4 2π D k 4 D

The input spectrum and the system natural frequency can be expressed in
terms of frequency f (Hz) or ω (rad/sec). In both cases the formulae are
formally the same.

From the variance y 2 the standard deviation (root-mean-square) of the

response follows as

σy = y2
The r.m.s. response depends on the square root of the damping ratio.

The standard deviation determines the distribution of all values of the response,
as can be seen from equation 5-11 and Fig. 5.6, but the peak, i.e. maximum
value of the response indicated in Fig. 5.14 is of primary importance for design.

Peak Value of Response

During each period of observation T, one largest (peak) value of the response can
be established. This largest value depends on the duration of the observation, T,
and the apparent frequency ν , which depends on the spectrum of the process and
is:

ν=
∫
0
f 2 S(f )df
(Hz) (5-34)
∞
∫
0
S(f )df

For a narrow band process such as the response of a lightly damped system, the
apparent frequency v is close to the natural frequency and thus, ν ≅ fo . The peak
values observed in individual observations may be assembled to yield a probability
density distribution (Fig. 5.15). The mean value of the peaks can be evaluated as:

yˆ ≅ gσ y (5-35)

in which the peak factor g = yˆ / σ y can be calculated using the formula:

Page 5-18
CEE490b

0.5772
g = 2 loge ν T + (5-36)
2 loge ν T

The peak factor ranges between about 2.5 and 4.5 (Fig. 5.16).
(see: Davenport, A.G., “The Distribution of the Largest Values of a Random
Function With Application to Gust Loading”, Proc. ICE, Vol. 28., No. 6739, June
1964, pp 187-196 ….. and
Rice, S.O. "Mathematical Analysis of Random Noise," Selected Papers on Noise
and Stochastic Processes, edited by N. Wax, Dover Publ., New York, 1954.)

Response to earthquakes. - With regard to equations 5-22 and 5-33, the variance
of earthquake response is

1 π ωo 2
y2 = m S y&&g (ω o ) (5-37)
k2 4 D

A more accurate analysis should consider nonstationarity but the assumption of

stationarity is conservative.

Fig. 5.15 Probability distributions of all values and peak values

Fig. 5.16

Peak Factor
vs. ν T

Page 5-19
CEE490b

Problem 5.1: Predict the seismic response of the one storey shear building
given in below to the El Centro 1940 earthquake in terms of random vibration.
The power spectrum of that earthquake is given in Fig. 5.llc. Assume damping
ratio D = 2% and strong motion duration T = 30 s. (In Fig. 5.llc, the power
spectrum corresponds to the original peak ground acceleration of 0.3 g.)

y
Each section of shear building is
supported by two columns having a
depth, d of 600 mm
h
I = 560 x 106 mm4
E = 2.0 x 105 Mpa
l h = 5.0 m

The participating mass of the

structure is:

m = 30,000 kg (for one bay)

Page 5-20
CEE490b

5.4 RESPONSE OF MULTI-DEGREE -OF-FREEDOM SYSTEMS TO RANDOM

LOADING

5.4.1 Fully Correlated Load

The motion of the ground or the forces acting directly upon masses mi are often
random. If the forces have the same time history (phase shift) at each mass but
different amplitudes they are fully correlated. This is the case with ground
excitation when the effective forces are:

Pi (t ) = ( −)mi u&&g (t )

or with direct excitation,

Pi (t ) = Pi f (t )

where u&&g (t ) or f (t ) are common for all masses. An example is a large wind gust
hitting a relatively small structure (Fig. 5.17). The response is again given by
equation 4.2, i.e.

u i (t ) = ∑ Φ ij η j (5-38) P (t)
j n
in which η j is given by equation (5-11),

M j η&& j + 2M j D j ω j η& j + K j η j = L j f (t )
= L j u&&g (t ) (5-39) P (t)
i

in which, for ground excitation

L j = ∑ mi Φ ij (5-40a) P (t)
i 1

or with external forces

L j = ∑ Pi Φ ij (5-40b)
i
Fig. 5.17
and

K j = ω 2j M j

Page 5-21
CEE490b

If f(t) or u&&g is random it can be described by its power spectrum Su&&g (ω ) or Sf (ω )

called generally S(ω ) . By equation 5-21a), the power spectrum of the right side of
equation.(5-39) is

S j (ω ) = L2j S(ω ) (5-41)

2
Equation (5-39) is an equation of SDF system H

and therefore the spectrum of coordinate η j is

1
by equation (5-29)
ω
1 ω
2
1 ⎛ω ⎞ S(ω)
Sη j (ω ) = L2j S(ω ) H ⎜ ⎟ (5-42)
K 2j ⎜ω j ⎟
⎝ ⎠
ω
where the square of the mode of the mechani-
S (ω)
cal admittance η

2
⎛ω ⎞ 1
H⎜ ⎟ = ω ω
⎜ω ⎟
1
2
⎝ j ⎠ ⎛ ω2 ⎞ ⎛ ⎞
⎜1 − ⎟ + 4D 2 ⎜ ω ⎟
⎜ ω2 ⎟ ⎜ω ⎟
⎝ j ⎠ ⎝ j ⎠
µi (t )
Variance of η j becomes:

2
L2j ⎛ω
∞
⎞
η = 2 ∫ S(ω ) H ⎜⎜
2 ⎟
⎟
(5-43)
⎝ω j
j
Kj 0 ⎠ -T 0 -T

The motion is

n
u i (t ) = ∑ Φ ij η j (t )
j =1

or with distributed systems

∞
u( x, t ) = ∑ Φ j ( x )η j (t )
j

The variance of this motion is obtained by squaring and averaging,

Page 5-22
CEE490b

2
1 ⎛
T
⎞ 1 ⎛
T
⎞
u = 2
i
2T
⎜ ∑
∫−T ⎜⎝ j Φ ij η j ( t ) ⎟
⎟ dt =
2 T ∫ ⎜
⎜ ∑ Φ 2 2
ij η j ( t ) + ∑ product terms ⎟dt
⎟
T →∞ ⎠ T →∞ −T ⎝
j ⎠
(5-44)
T
1 ⎛ n n ⎞ n n
=
2T −
∫T ⎜⎝ ∑∑
⎜
r =1 s =1
Φ ir Φ η η ⎟
is r s ⎟
⎠
dt = ∑∑
r =1 s =1
η r η s Φ ir Φ is
T →∞

The cross products between the generalized coordinates complicate the situation.
However, they can be neglected if

1) Natural frequencies are well separated

2) Damping is small or at least not very large.

Then

1 ⎛
T
⎞ 2 1
T
u = ∫−T ⎜⎝ ∑j Φ ijη j (t ) ⎟⎠dt =∑j Φ ij 2T −∫Tη j (t )dt
⎜ ⎟
2 2 2 2
i
2T
T →∞

T
1
As ∫
2T −T
η 2j (t )dt = η 2j = the variance of generalized coordinate as given by

Equation (5-43), the resonance of the displacement of mass m i is

2
n n ⎛ωL2j ∞
⎞
u (t ) = ∑ Φ η = ∑ Φ ⎜ ⎟ dω
K 2j ∫0
2
i
2
ij
2
j S (ω )2
H
ij ⎜ω ⎟
(5-45)
j =1 j =1 ⎝ j ⎠

The integral in equation (5-45) can be evaluated by means of the theory of

residua or numerically as already discussed. If damping is small and the spectrum
rather flat, an approximate solution indicated in equation (5-33) and Fig. 5.13 is
usually sufficiently accurate, i.e.

∞ ∞
π ωj
∫ S(ω ) H (ω ) dω ≅ S(ω )∫ H (ω ) dω = S(ω j )
2 2
(5-46)
0 0
4 D

With this approximation, equation (5-45) simplifies and the variance of the
displacement is

Page 5-23
CEE490b

π n L2j
u i2 (t ) =
4D
∑ Φ 2ij K 2j
S(ω j )ω j
j =1
S(ω j)
S (ω)

|H (ω)|
2

ω
Substituting for K j = ω 2j M j , it is j ω
also

π n L2j S(ω j ) 1 n L2j S(f j )

u (t ) =
2
i
4D
∑Φ
j =1
2
ij
M 2j ω 3j
=
64π 3 D
∑Φ
j =1
2
ij
M 2j f j3
(5-47)

The RMS displacement is σ ui = u i2 (t ) . Only one, two or three first modes usually
need to be considered. Very often, one mode is enough (the first or the second),
e.g. for buildings exposed to wind gusts or earthquake excitation. Max. (peak)
values follow from equation (5-35) and range from 3.5 to 4.5 RMS.

If the power spectrum is available as a function of frequency f, it is possible to

use either one as

1
S(ω j ) = S (f j )
2π

If the damping ratio, D, is assumed to be different for each vibration mode, it

remains as Dj behind the summation sign ∑ in equation (5-47).

Problem 5.2: Analyze the response of the five storey shear building to earthquake
excitation defined by the power spectrum shown in Fig. 5.11c (El Centro, 1940).
Calculate:
(a) Peak response in individual modes uij assuming the duration of the strong
motion T = 30 s.
(b) Equivalent seismic forces q ij = mi ω 2j uˆ ij
(c) Compare the results with those obtained by means of the pseudovelocity
spectrum.

Page 5-24
CEE490b

5.4.2 Partially Correlated Loads

When the loads Pi acting at individual stations of a structure (Fig. 5.17) are not
fully correlated, their total effect on the response in the fundamental mode is
reduced. This reduction is very significant in the case of wind loading as is
discussed in Chapter 6. The analysis requires a greater amount of input infor-
mation and is more difficult but can lead to useful observations (see: Novak, M.
"Random Vibration of Structures", Proc. 4th Intern. Conference on Application of
Statistics and Probability in Soil and Structural Engineering, Florence, 1983, pp.
539-550).

LITERATURE ON RANDOM LOADING

1. Bendat, J.S. and Piersol, A.G. "Measurement and Analysis of Random

Data", John Wiley & Sons, New York, 1968, p. 390.
2. Blackman, R.B. and Tukey, J.W. "The Measurement of Power Spectra"
Dover Publications Inc., New York, 1958, p. 190.
3. Bolotin, V.V. "Statistical Methods in Structural Mechanics", Holden-Day
Inc., 1969, p. 240.
4. Clough, R. W. and Penzien, J. "Dynamics of Structures", McGraw-Hill,
Inc., 1975, Ch. 4, pp. 389-517.
5. Crandal, S.H. and Mark, W.D. "Random Vibration in Mechanical
Systems", Academic Press, Inc., 1963, p. 166.
6. Robson, J.D. "An Introduction to Random Vibration", Edinborough
University Press, 1963, p. 150.
7. Vanmarcke, E. "Random Fields: Analysis and Synthesis", MIT Press,
Cambridge, Mass., 1983, p. 382.

Page 5-25
CEE490b

Chapter 6 - Response To Gusting Wind

6.1 General

Assume that the structure is smaller than the typical dimension of the atmospheric
vortex. Then the quasi-steady approach applies and the time variable drag force is

FD (t ) =
1
2
1
[
ρC D AV 2 (t ) = ρC D A V + v (t )
2
]
2
(6-1)

where ρ = air density = 0.0024 slugs/ft 3 = 1.23 kg/m3, CD = drag coefficient, A =

area exposed to wind, V (t ) = V + v (t ) = wind speed with V > v(t). Taking the mean
wind speed V in front of the bracket

2⎡ v (t ) ⎛ v (t ) ⎞ ⎤
2 2
⎡ v (t ) ⎤
2
V (t ) = V ⎢1 +
2
= V ⎢1 + 2 +⎜ ⎟ ⎥
⎣ V ⎥⎦ ⎣⎢ V ⎝ V ⎠ ⎦⎥

( )2
Because v (t ) / V < 1 and thus v (t ) / V << 1 , the square of the ratio v (t ) / V can be
omitted and the total drag force split into two parts:

1 2
1. the mean (static drag) FD = ρCD AV (6-2)
2
and

2. the fluctuating (dynamic) drag FD (t ) = ρC D AVv (t ) (6-3)

in which v(t) = fluctuating component of the wind speed. This fluctuating wind speed
has a spectrum Sv(f) and, therefore, the spectrum of the drag is, according to
equation 5.21a,

(
SF = ρC D AV Sv (f ) )2
(6-4)
or

2
4F D
SF = 2
Sv (f ) (6-5)
V

The relation between the size of the structure and the size of the vortex
(disturbance) can be introduced through an "aerodynamic admittance function"
analogous to the mechanical admittance,

Page 6-1
CEE490b

2
⎛ fL ⎞
χ = χ⎜ ⎟
2
(6-6)
⎝V ⎠

where L = characteristic dimension for buildings or structures equal to L = A . The

ratio fL / V is a dimensionless frequency. Only with "point" (small) structures does
the Aerodynamic Admittance, χ ≡ 1 . Otherwise the values of χ range between these
limits:

(fL /V ) → 0 , χ2 →1
(fL /V ) → ∞ , χ2 → 0

χ 2 is a drooping function indicated in Fig. 6.1.

Fig. 6.1 Aerodynamic admittance function

The aerodynamic admittance function is usually established analytically using

experimental data. With the aerodynamic admittance introduced into equation (6-5),
the fluctuating force spectrum is

2 2
4F D ⎛ fL ⎞
SF = χ ⎜ ⎟ Sv (f ) (6-7)
V ⎝V ⎠

With the spectrum given by equation (6-7), the response of a single degree of
freedom structure can be readily predicted. However, for large structures, the
approach must be somewhat extended to include the variation of wind speed with
height and the shape of the vibration mode.

These aspects were included in an approach adopted in the National Building Code
of Canada. This Code assumes that the first vibration mode is linear and sufficient
for the analysis and that the width of the face of the building is constant.

Page 6-2
CEE490b

The approach is called the Gust Factor Approach. The gust factor Cg is defined as
the ratio of the total peak displacement Û (or load) in a period T to the mean
displacement u (or load). Thus,

Uˆ
Cg = or Uˆ = C g u
u

and the peak load pˆ = C g p . The peak value is obtained from σ u by means of the
peak factor g; adding the static (average) displacement u , the total peak
displacement is (Fig. 6.2)

⎛ σ ⎞
Uˆ = u + gσ u = u ⎜1 + g u ⎟ (6-8)
⎝ u ⎠

Hence, the gust factor actually becomes

σu
Cg = 1 + g (6-9)
u

The gust factor C g can be evaluated using the method given for the evaluation of
wind loading in the NBC of Canada. This method incorporates all the ingredients of
the random vibration approach.

Fig. 6.2 Total structural response to gusting wind

Page 6-3
CEE490b

6.2 Wind Pressure By NBC

Detailed Procedure by the National Building Code of Canada, 1995

This procedure is facilitated by the charts given in Fig. 6.3.

The design external pressure p = p(Z ) varies with height Z on the windward wall, is
constant for other walls and is calculated as

p = q Ce Cg Cp (6-10)

in which: 2
q = q10 = reference mean velocity pressure in kN/m2 = 1 / 2 ρ V . It is given in
Chapter 1 of the Supplement to the NBC for different localities and return
periods of 10, 30 and 100 years; it corresponds to open terrain and the height
of 10m, hence V = V 10

Ce = exposure factor. It accounts for the variation of the wind pressure (wind speed)
with height and therefore, varies with height and exposure (surface
roughness). It can be read from Fig. 6.3b or calculated for three exposures:

A - open terrain: Ce = (Z/10)0.28, Ce > 1.0 (6-11a)

B - suburban areas: Ce = 0.5(Z/12.7)0.50, Ce > 0.5 (6-11b)
C - centres of large cities: Ce = 0.4(Z/30)0.72 Ce > 0.4 (6-11c)

In equation (6-10), Ce varies with height for the windward wall, i. e. C = Ce(Z) ,
for the roof CeH= Ce(H) and for the leeward wall Ce = Ce (H/2).

C g = gust effect factor. It accounts for dynamic effect of gusting wind and is defined
as the ratio: total peak response (load)/mean response(load). C g has one
value for the whole building and is evaluated using the mean wind velocity at
the top of the building, VH. This velocity is obtained from the reference velocity
V = V 10 as:

VH = V C eH (m/s) (6-12)

where CeH = Ce(H) and V = V 10 is:

V = q / C (m/s) (6-13)

1
with C = ρ = 650 x 10-6 for V in m/s and q in kN/m2.
2

Page 6-4
CEE490b

The gust effect factor follows from the formula:

C g = 1 + g p (σ / µ ) (6-14)

in which the coefficient of variation:

K ⎛ sF ⎞
σ /µ = ⎜⎜ B + ⎟ (6-15)
Ce ⎝ β ⎟⎠

Here, K = surface roughness factor equal to:

0.08 for exposure A,
0.10 for exposure B,
0.14 for exposure C;

CeH = Ce(H) = the exposure factor for height H;

B = background turbulence factor from Fig. 6.3c;

s = size reduction factor from Fig. 6.3d in which no = natural

frequency of the building in Hz;

F = gust energy ratio from Fig. 6.3e; and

β = damping ratio of the building.

In equation (6-14),
gp = peak factor = uˆ / σ u follows from Fig. 6.3f depending on the
average fluctuation rate (equation 5.34)

sF
ν = no
sF + β B

With these factors C g is obtained from equation (6-14). Finally in equation

(6-10),

C p = pressure coefficient given in Fig. 6.3a.

Page 6-5
CEE490b

b) Exposure Factor, Ce

a) Pressure Coefficients

c) Background Turbulence Factor, B

Fig. 6.3a,b,c Charts for evaluation of gust effect factor by NBC, 1995

Page 6-6
CEE490b

d) Size Reduction Factor, S

e) Gust Energy Ratio, F

f) Peak Factor, g
P

Fig. 6.3d,e,f Charts for evaluation of gust effect factor by NBC, 1995

Page 6-7
CEE490b

Acceleration - The gust factor can also be used to estimate the peak acceleration at
the top of the building, which is an important design criterion of human comfort.
Assume that the peak acceleration, â , derives only from the peak fluctuating
displacement û (Fig. 6.2) and that it can be approximately evaluated as in harmonic
motion; assume further that background turbulence does not contribute to it because
the associated frequencies are very low and thus the background turbulence factor
B = 0. Then, the peak acceleration is equal to:

σu
aˆ = ω o2uˆ = 4π 2 no2 g p µ
µ

With σ / µ from equation (6-15) and µ = u = Uˆ / C g , where Û is the total peak dis-
placement due to loading described by equation (6-10), the peak acceleration
becomes:

gp KsF ˆ
aˆ = 4π 2 n o2 ⋅U (6-17)
Cg Ce β

PROBLEM 6.1:

Evaluate the gust factor, design pressure and acceleration for a tall building to be
built on the waterfront in Toronto's downtown area. Consider wind blowing from the
south (lake) and from the north (downtown), i.e. exposures A,C, respectively.

Height H = 100 m
Width W = 24 m
Depth D = 20 m
Natural Frequency no = 0.2 cps
Damping ratio β = 0.01
Exposure A,C

For a 30 year return period, the reference hourly wind pressure in Toronto is from
Chapter 1 of the Supplement, q = 0.48 kN/m2. Evaluate the maximum acceleration
assuming that Û was found to be 0.02 m for both exposures. Explain the difference
between the two exposures.

Page 6-8
CEE490b

SOLUTION TO PROBLEM 6.1:

Given:
Height H = 100 m
Width W = 24 m
Depth D = 20 m
Natural Frequency no = 0.2 cps
Damping ratio β = 0.01
Exposure A,C
Reference Wind Pressure q = 0.48 kN/m2 for a Return Period of 30 years
Peak Displacement Û = 0.02 m

a) Evaluate response under NBCC Exposure “A” (Open Terrain Exposure)

• The gust factor is determined through the evaluation of:

⎛σ ⎞
Cg = 1 + g p ⎜ ⎟ (6-14)
⎝u⎠
σ K ⎛ sF ⎞
= ⎜⎜ B + ⎟ (6-15)
u Ce ⎝ β ⎟⎠

• K = 0.08 for Exposure A

• The exposure factor is evaluated for a height of 100 m,

C e = C e (100 ) = (Z / 10) 0.28 = (100 / 10) 0.28 = 1.91 , C e ≥ 1.0

• The Background Turbulence Factor is found from Figure B-2 as

B = 0.78 since W/H =24/100 = 0.24 and H = 100 m

• The reference wind speed is determined from

V = 0.48 = 27.2 m/s and the wind speed at the top of the
650 x10 −6
building, VH = 27.2 1.91 = 37.5 m/s.

n o H 0.2 x100
• The Reduced Frequency is: = = 0.53 and with the width
VH 37.5
to height ratio, W/H = 0.24, from Fig. B-3, the Size Reduction Factor,
s = 0.18

• The Gust Energy Ratio is obtained from Figure B-4, for a Wave
n 0. 2
Number, o = = 0.0053 , then F = 0.28,
VH 37.5

Page 6-9
CEE490b

• The Coefficient of Variation, for a structural damping ratio, β = 0.01

from equation (5) is now:

σ K ⎛ sF ⎞ 0.08 ⎛ 0.18 x 0.28 ⎞

= ⎜⎜ B + ⎟⎟ = ⎜ 0.78 + ⎟ = 0.49
u Ce ⎝ β ⎠ 1.91 ⎝ 0.01 ⎠

• The Average Fluctuation Rate ν , is needed to evaluate the Peak

Factor, g p , and is determined from:
sF
ν = no (6-16)
sF + βB
0.18 x 0.28
ν = 0 .2 = 0.19 Hz
0.18 x 0.28 + 0.01x 0.78

• The Peak Factor, is found from Figure B-5 as g p = 3.8 for a time
duration of 1 hour (T = 3600 seconds), or from the familiar equation:
0.5772
g p = 2 lnνT +
2 lnνT
0.5772
= 2 ln(0.19 x 3600 ) +
2 ln(0.19 x 3600 )
= 3.77

• The Gust Effect Factor is then, from equation (6-14):

⎛σ ⎞
Cg = 1 + g p ⎜ ⎟
⎝u⎠
= 1 + 3.77 x 0.49
= 2.86

• The pressure coefficient at the Windward Face at roof level is C p = 0.8

according to NBCC, so then has a design pressure of:

p = qC eC g C p (6-10)
= 0.48 x 1.91 x 2.86 x 0.8 = 2.10 kPa

• The pressure coefficient on the Leeward Face at roof level is

C p = −0.5 according to NBCC, so then has a design pressure of:
p = 0.48 x 1.91 x 2.86 x (-0.5) = -1.31 kPa

Page 6-10
CEE490b

• The maximum acceleration is determined from equation (6-17)

g p KsF
aˆ = 4π 2 n o2 ⋅ Uˆ (6-17)
Cg Ce β
3.77 0.08 x 0.18 x 0.28
aˆ = 4π 2 0.2 2 ⋅ 0.02
2.86 1.91x 0.01
= 0.019 m/s 2

Or, about 2 milli-g’s

b) Evaluate response under NBCC Exposure “C” (City Exposure)

• The Gust Effect Factor is then, from equation (6-14):

C g = 3.71

• The pressure at the Windward Face at roof level is

p = 0.48 x 0.95 x 3.71 x 0.8 = 1.35 kPa

• The pressure on the Leeward Face at roof level is:

p = -0.85 kPa
• The maximum acceleration is determined from equation (6-17)
aˆ = 0.021 m/s 2

Or, again, about 2 milli-g’s

Notice that :
• In Exposure A, (Open Terrain) we have higher mean wind speeds
at the top of the building, but lower gust factors, which results in
higher mean pressures, but lower gust pressures
• In Exposure C, (City Terrain) we have lower mean wind speeds at
the top of the building, but higher gust factors, which results in
lower mean pressures, but higher gust pressures

Page 6-11
CEE490b

6.3 Distributed Random Loading Due to Wind

Multiple Discrete Loads

Consider a randomly varying quantity z(t) made up of

two components, x(t) and y(t),

Fig. 6.4 Multiple loads

z(t)=x(t)+y(t)

The autocorrelation function of the combined input, R z (τ ) , is:

R z (τ ) =< z(t )z(t + τ ) >

=< [ x (t ) + y (t )][ x (t + τ ) + y (t + τ )] >
=< x (t )x (t + τ ) + x (t )y (t + τ ) + y (t )x (t + τ ) + y (t )y (t + τ ) >

Thus, R z (τ ) = R xx (τ ) + R xy (τ ) + R yx (τ ) + R yy (τ )

In which R xx (τ ) and R yy (τ ) are the autocorrelation functions of signals x and y and:

R xy (τ ) =< x (t )y (t + τ ) >
R yx (τ ) =< y (t )x (t + τ ) >

the cross-correlation or cross-covariance functions.

The correlation function of a combined signal is determined by the autocorrelations

of the components and by their cross-correlations. The cross-correlation will be zero
only if the components x(t) and y(t) are completely uncorrelated or unrelated.

When a signal consists of more components, so that:

z(t ) = x1 (t ) + x 2 (t ) + x 3 (t ) + ... + x n (t )

The autocorrelation R z (τ ) is:

R z (τ ) = R11 (τ ) + R12 (τ ) + R13 (τ ) + ... + R1n (τ ) +

R 21 (τ ) + R 22 (τ ) + R 23 (τ ) + ... + R 2n (τ ) +
R 31 (τ ) + R 32 (τ ) + R13 (τ ) + ... + R1n (τ )

which can be written as a double sum: R z (τ ) = ∑∑ R rs (τ )

r s

Page 6-12
CEE490b

or as a correlation matrix: [R z ] = [R ij ]

⎡R11 R12 L R1n ⎤

⎢R R L R 2 n ⎥⎥
[R z ] = ⎢ 21 22
⎢ M M O M ⎥
⎢ ⎥
⎣R n1 R n 2 L R nn ⎦

Properties of cross-correlations of stationary processes:

R xy ( −τ ) =< x(t )y (t − τ ) >=< y (t )x(t + τ ) >= R yx (τ )

R yx ( −τ ) =< y (t )x(t − τ ) >=< x(t )y (t + τ ) >= R xy (τ )
R xx ( −τ ) =< x (t )x(t − τ ) >=< x(t )x (t + τ ) >= R xx (τ )

Fig. 6.5 Time history of two signals, x(t) and y(t)

There is no relation between R xy (τ ) and R xy ( −τ ) ; R xy (τ ) and R yx (τ ) are in general,

unrelated.

R xy (τ ) and R yx (τ ) do not necessarily have their maximum values at τ =0.

Page 6-13
CEE490b

Fig. 6.6 Auto and Cross Correlation Functions

The distance between two peaks determines the average time delay between two
processes, e.g. wind speed and wind pressure.

Covariance

When τ =0, R xy (0) =< x (t )y (t ) >= σ xy

2
, which is covariance. This is the measure of
the extent to which two random variables, x,y are correlated. In they are completely
independent, σ xy
2
= 0 and also R xy = R yx = 0 .

Correlation Coefficient

A dimensionless form of covariance

σ xy
2

R= ≤ 1 .0 ,
σ xσ y
or, when σ x = σ y = σ , R = σ xy
2
/σ 2

When R = 1, the process is fully correlated

and if R = 0 , the process is completely
uncorrelated.

Fig. 6.7 Correlation Coefficient

The correlation length, L, can measure the span-wise correlation and is defined as:
∞
L
= ∫ R (θ )dθ in numbers of diameters, d.
d 0

Page 6-14
CEE490b

Further useful relationships are:

2
R xy (τ ) ≤ R x (0)R y (0)
R xy (τ ) ≤ 1
2
[R x (0) + R y (0)]

The correlation function coefficient, (normalized cross-covariance or cross-

correlation function), is, with zero means,

R xy (τ ) R xy (τ )
ρ xy (τ ) = =
R x (0)R y (0) σ xσ y

The area beneath the curve is a measure of the time over which the processes are
correlated,

∞
T = ∫ ρ xy (τ )dτ and is called the time scale. In the case of wind , T is the time
0

scale of turbulence, determined from ρ xx , ρ yy respectively.

Fig. 6.8 Effective time scale

A more complete picture of the process with respect to its frequency content can be
obtained from ”cross-correlations” and “cross-spectral densities”. In most cases, we
can describe the process in a simplified way through the spectral density and the
correlation coefficient, which can be replaced with the correlation length.

i.e. Local Spectra x Correlation Length

Page 6-15
CEE490b

6.4 The Along-Wind Response of Line-Like Structures (after Davenport)

We have examined the response of small, point-like structures to wind. These

structures are those that are sufficiently small, so that the bulk of the energy of the
turbulence of gusting is at wavelengths much greater than the typical dimension of
the structure. A line-like structure, on the other hand is one where that has
significant dimension transverse to the wind, however remains small in the sense
that the smallest wavelengths likely to be of significance must be large compared to
the breadth of the structure. The approach is applicable to slender towers,
transmission lines and some long span bridges.

The load per unit length on a slender structure of width B and length L may be
expressed as:

F (t ) = F + F ′(t )
2 (6-18)
= 1
2
ρU CDB + ρ UCDBu(t )

The response of the structure to the fluctuating component of the force, ρUCD Bu(t ) ,
may be computed mode-by-mode using modal analysis. If the mode shape for the
ith mode is µ i (z ) , then the modal force, Fi (t ) , is as follows;

L
Fi (t ) = ∫ C D Bρ Uu( z, t )µ i ( z )dz (6-19)
0

The spectral density function SF (f ) then becomes;

( )
L L
SF (f ) = C D Bρ U Sv (f )∫ ∫ R( z, z ′, f )µ i ( z )µ i ( z ′)dzdz ′
2
(6-20)
0 0

in which;

Sv (f ) is the spectrum of turbulence

R( z, z ′, f ) is the narrow band correlation function or normalized co-
spectrum of turbulence

The mean square value of the modal coefficient ai2 is then obtained from the
relationship;

∞
1
∫S F (f ) ⋅
(1 − (f / f i ) ) + 4(f / f i ) 2 ζ 2
2 2
df
ai2 = 0
(6-21)
(2πf i ) 4 M i2

Page 6-16
CEE490b

in which;

ζ is the damping as a fraction of critical

fi is the natural frequency of the ith mode
Mi is the Generalized Mass of the ith mode
L
M i = ∫ m( z )µ i ( z )dz
0

The displacement of the structure can now be computed by the superposition of

modes as follows;

y ( z, t ) = ∑ ai (t )µ i ( z ) (6-22)
i
or, in terms of the mean-square response;

y 2 ( z ) = ∑ ai2 µ i2 ( z ) (6-23)
i

The mean-square bending moment at position, z is given by:

BM 2 ( z ) = ∑ ai2 ⋅ BM i2 ( z ) (6-24)
i

in which BMi (z) is the bending moment at position z, when the deformed shape is
µ i (z ) . If the spanwise distribution of force varies due to spanwise changes of
diameter, the drag coefficient, the mean wind velocity and / or velocity spectrum,
Equation 6-20 can be written:

( )
2
S F ( f ) = C D Bρ U S v ( f ) ⋅ J ( f )
2
(6-25)

where

L L
J (f ) = ∫ ∫ R(ξ ,ξ ′; f ) γ (ξ )γ (ξ ′) µ i (ξ )µ i (ξ ′) dξdξ ′
2
(6-26)
0 0

ξ is a normalized coordinate, ξ = z / L and γ (ξ ) is a function incorporating the span

-wise changes in diameter, the drag coefficient, the mean wind velocity and velocity
2
spectrum. The function J (f ) is called the “Joint Acceptance Function” and defines
the sensitivity between the turbulence and the structural modes of vibration, which is
critical in defining the response of the structure.

Page 6-17
CEE490b

The Joint Acceptance Function

Equations (6-25) and (6-21) define the link between the gust fluctuations (which are
described by the velocity spectrum Sv (f ) ) and the modal force fluctuations and
displacements (provided by the Joint Acceptance Function). This function depends
on the mode shape and the velocity field, which can vary from structure to structure
as indicated in Figure 6.9.

The lamp standard in (a) can oscillate in both a fore and aft as well as in a twisting
mode, however, the wind excitation is concentrated at essentially one elevation,
namely that of the lamps. In the vertical structures in (b-d), the wind speed varies
with height; the fundamental mode of the building in (b) is nearly a straight line. The
diameter of the chimney in (c) varies with height and has a fundamental bending
type of mode shape. The guyed mast in (d) will have mode shapes consisting of
several half-waves. The horizontal bridge structures in (e) and (f) will likely have
near constant mean wind velocity along the span. The suspension bridge will have
near-sinusoidal mode shapes, while the cantilever bridge will have a twisting mode
about the axis of the pier as well as one in the fore and aft direction.

For a slender structure it is common practice to assume that the correlation of the
forces is the same as that for the transverse correlation of the longitudinal
component of the wind. That is;

⎡ f z − z′ ⎤
R( z, z ′, f ) ≈ exp ⎢− c ⎥ (6-27)
⎢⎣ V ⎥⎦

or

⎡ f z − z′ ⎤
R(ξ ,ξ ′, f ) = exp⎢− φ ⎥ (6-28)
⎣⎢ L ⎦⎥

cfL
where φ = , a dimensionless frequency, and c is the correlation coefficient,
V
commonly found to be between 8-10.

The JAF can now be written:

1 1
−φ ξ −ξ ′
J (f ) = ∫ ∫ e γ (ξ )γ (ξ ′) µ i (ξ )µ i (ξ ′) dξdξ ′
2
(6-29)
0 0

Page 6-18
CEE490b

Fig. 6.9 Oscillations of structures in turbulent wind (after Davenport)

Page 6-19
CEE490b

Table 6.1 indicates the form of the Joint Acceptance Function for a number of
common mode shapes, assuming that γ (ξ ) = 1.0 , or there is no variation in the force
per unit length. The functions are also plotted in Figures 6.10 and 6.11.

Several features of these functions are worth noting:

a) The mode shapes composed of deflections of the same sign decrease

monotonically for higher frequencies and can be well represented by
approximations of the form
2 1
J (f ) =
A + Bφ

b) The antisymmetric shapes have no response for small values of φ (i.e. for
large gust wavelengths). The gust then envelops the entire structure and the
antisymmetric mode shape neutralizes its effect. These mode shapes have a
peaked JAF, indicating that there will be a maximum response at a specific
wavelength.

c) Mode shapes having deflections of opposite sign, but not necessarily

antisymmetric have a finite lower asymptote as φ → 0 , reach a peak for
intermediate values of φ and fall off again as 1/ φ at larger values of φ . It is
important to note that for sinusoidal mode shapes, the peak of the JAF occurs
at φ ≈ 3n − 2 , where n is the number of half waves.

The qualification of “line-like” in the use of the Joint Acceptance Function implies that
all significant wavelengths influencing the structural response are greater than the
diameter of the structure (i.e. fD / V << 1). The flow is then “quasi-steady” For
structures such as transmission lines this assumptions is accurate, however as the
structure becomes larger and the slenderness decreases, (from chimneys to bridge
decks to buildings, for example) this assumption will exaggerate the response at
higher frequencies and provide overly conservative estimates of the wind loads.
2
Introducing the Aerodynamic Admittance Function, A(f ) , allows for the reduction in
force for the less slender structures.

In most cases in the estimation of the response of a structure to wind it is rarely

necessary to extend the summation of equation (6-20) beyond more than a few
modes, and in many cases the first term (especially in the case of buildings of
moderate height) is quite adequate. This is the basis for the wind load provisions in
the National Building Code of Canada. Long span bridges may require several
modes to adequately represent the loading with sufficient detail in order to
adequately predict the force effects. To date, satisfactory wind codes for long span
bridges have not been developed and wind tunnel testing must be undertaken to
define loads for design.

Page 6-20
 Table 6.1 Joint Acceptance Functions for line-like structures
CEE490b

Page 6-21
 CEE490b

Page 6-22
Fig. 6.10 Joint Acceptance Functions for line-like mode shapes – linear modes (after Davenport)
 CEE490b

Page 6-23
Fig. 6.11 Joint Acceptance Functions for line-like mode shapes – lsinusoidal modes (after Davenport)
CEE490b

Chapter 7 – Fatigue
7.1 General

Repetitive loading of a structure during vibration may bring about its failure at a
stress much lower than the static strength. Such a failure can often be attributed to a
phenomenon called fatigue. This type of failure has occurred in steel railroad
bridges, towers, chimneys and even in highway bridges. Thus it is important to
evaluate the stresses due to vibration with regard to fatigue.

Fatigue has much in common with plastic flow and fracture and is essentially a
process of slip within the crystals along the direction of the greatest shear. In static
fracture, the slip formation is spread throughout a large volume of the metal and is
accompanied by readily recognizable distortions such as contraction of the cross-
section; in fatigue, the slip formation is confined to a small localized volume, the
fracture plane is characteristically smooth and the metal suffers no distortion (Fig.
7.1).

Fig. 7.1 Modes of failure: (a) - static failure in tension test, (b) - fatigue failure
due to repetitive loading

The fatigue process, from the first loading cycle to rupture, has three distinct
stages. In the first stage, slip formation occurs within the interior of the grains if
the frequency of load application is higher than about 400 cycles per minute; at
lower frequencies, the slip formation extends from one grain boundary to the next
as in static tests.

The intermediate stage occupies the major part of fatigue life. The slip formation
becomes localized. The original small number of slip lines, formed in the first
stage, begin to thicken into bands and a microscopic crack appears. This crack
starts growing in a zig-zag form and the average direction at right. angles to the
tensile stress.

Page 7-1
CEE490b

In the final stage, the microscopic crack grows into a macroscopic crack which
rapidly spreads into the bulk of the metal rendering it unable to carry the load; an
abrupt failure follows.

The tendency to fatigue failure can be described in terms of fatigue limit or fatigue
strength.

Fatigue or endurance limit is the alternating stress which can be applied

indefinitely without failure occurring; 106 loading cycles is usually considered to
be sufficient to define the fatigue limit.

Fatigue strength is the nominal alternating stress which produces failure for a
specified number of cycles (Fig. 7.2). Alternatively, fatigue strength may be given
in terms of fatigue ratio which is the ratio of the fatigue strength to the ultimate
strength.

Fatigue limit and fatigue strength depend on a number of factors. The first one is
the type of loading (Fig. 7.3). As can be seen, torsional loading reduces the
available strength most but even bending reduces-the fatigue strength to about
one half of static strength. Surface roughness has a profound effect; fatigue limit
increases with increasing smoothness of the surface but increases only
marginally with static strength (grade of the steel) if the surface is not machined
(Fig. 7.4). Thus the choice-of higher grade steel need not eliminate the danger of
fatigue.

Fig. 7.2 Schematic of fatigue limit and fatigue strength

Page 7-2
CEE490b

Fig. 7.3 Fatigue strength for basic types of loading

Fig. 7.4 Variation of fatigue limit with surface roughness and tensile strength

Page 7-3
CEE490b

Fig. 7.5 Typical notches reducing fatigue strength

Fatigue strength is further reduced by stress concentration, which may be caused

by various notches (Fig. 7.5). The sensitivity of different materials to notches is
not the same and is characterized by notch sensitivity index.

Stress range applied in cyclic loading is the most important factor for any
specimen; the mean value of the stress seems fax less important than had been
thought until recently (Fig. 7.6).

The effect of weldments on the fatigue strength of steel structures is a very

unfavourable factor. This is so because welding provides the initial flaws and
defects which alleviate the need for the first stage of the fatigue process. Sharp
defects exist at the weld periphery or in the weldment and occur regularly even in
the common fillet and groove (V-butt) welds. At these defects, the crack growth
starts. The initial flaws are sharp intrusions of slag or porosity (gas pockets) and
may also result from weld repairs with incomplete fusion. These defects occur at
the flame-cut edges of plates as well.

For these reasons cover plates, stiffeners and various attachments, such as
those shown in Fig. 7.7, can reduce the fatigue strength to 1/2, 1/3 or even more
of the fatigue strength of the plain welded beam (Fig. 7.6 from: J.W. Fischer et
al. Effect of Weldments on the Fatigue Strength of Steel Beams," Nat. Coop.
Highway Research Program Report 102, Highway Research Board, Nat.
Academy of Engineering, 1970, p. 114). Considerable improvement can be
achieved by sealing the root of the weld and machining the surface (Fig. 7.8).

Experimental observations depend on test conditions and details which lead to

large scatter of results and a high degree of uncertainty. This is why statistics is
increasingly used to evaluate the results of fatigue tests and to predict fatigue life
of a structure. Further information on fatigue can be found in Refs. 11 and 12.

Page 7-4
CEE490b

Fig. 7.6 Mean fatigue strength and 95 percent confidence limits in terms of stress
range for rolled, welded and cover plated beams (J.W. Fisher et al., 1970)

Fig. 7.7 Typical details that reduce fatigue strength of beams (J.W. Fisher et al.,
1970)

Page 7-5
CEE490b

Fig. 7.8 Effect of weld contour and defects on stress flow lines and fatigue
behaviour of welded joints

Brittle Fracture. In a normal case of static overloading, the imminent failure is

fairly well predictable, because visible yielding, sag or cracks precede the
collapse. The brittle fracture, somewhat like a fatigue failure, occurs suddenly
and without visible signs of any imminent failure. Brittle failure, meaning a
fracture without yielding, can occur when low temperatures and peak stresses
exist. The peak stresses are produced by stress raisers such as weld cracks,
fatigue cracks, flaws or notches.

The material capability to resist brittle fracture depends on toughness.

Toughness represents the capability to take the load even in the presence of
flaws, notches or cracks and can be established by special standardized tests.

Page 7-6
CEE490b

7.2 Fatigue Analysis

The most common method used to examine the effect of cyclic stress on a structural
component is by examining the “S-N Curve”. This is a plot of the stress level
required to cause a fatigue failure of a component for a given number of repetitive
cycles.

N cycles to failure

S-N Curve
Stress level, S

1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08 1.E+09 1.E+10

Number of Cycles to Failure, N

Figure 7.9: Typical S-N Curve

However, most loading of structures does not consist of a single, cyclic stress level,
but a number of stress conditions, all contributing to the fatigue damage to the
structure. An empirical method was developed to evaluate the cumulative damage
done to a structural component due to different cyclic stress levels, each with their
own contribution to fatigue. This is the so-called “Miner’s Damage Law”.

Page 7-7
 CEE490b

Miner’s Damage Law

Miner postulated that fatigue failure would occur when:

n( y )
∑ N( y ) = 1
where:
N(y) is the number of cycles required to produce failure at amplitude, y
n(y) is the number of cycles realized at amplitude, y

Stress Distribution

σ1 σ vs N curve
MAXIMUM STRESS, σ

σ2

σ3 etc.

n1 n2 n3 N1 N2 N3
NUMBER OF CYCLES TO FAILURE, N

Figure 7.10 Miner’s Damage Law

Page 7-8
CEE490b

If we have a normally distributed process, the number of cycles in a given time T

becomes:

(y o − y ) (
− yo −y )
n( y ) = νT e 2σ 2
σ 2

n( y )
Hence: ∫ N( y ) dy = 1 is the damage criteria.

S-N Curve
Amplitude, y

knee
n

1.E+01 1.E+03 1.E+05 1.E+07 1.E+09

Number of Cycles

Figure 7.11 Damage Criteria

Page 7-9
CEE490b

Problem 7.1

Check the possible fatigue damage at the base of the tower shown here due to
resonant vortex shedding from the observation gondola.

Data:
Effective mass, m = 68,000 kg
Young’s Modulus, E = 2 x 105 Mpa
Height, h = 152.4 m
Equivalent uniform inertia of the cross section, Ieq = 1.116 m4
Inertia at Base Ibase = 3 x Ieq
Diameter at base, dbase = 15.24 m
Damping Ratio, D = 0.4%
Lift Coefficient, CL = 0.2
Strouhal Number, S = 0.2

The lateral force due to vortex shedding at resonance is P (t ) = Po cos ω o t , where

Po = 21 ρVcrit
2
CL ld and l and d are the height and diameter of the observation
gondola, respectively. The critical velocity for vortex shedding is V crit = f o d / S

Assume that:
a) Resonant Vortex Shedding occurs when 0.85 ≤ V / V crit ≤ 1.15 , where V
is the hourly mean wind speed.
b) The hourly mean wind speed has a “Weibull” probability distribution of the
form:
( C)
−V
K

P(> V ) = e where: C = 8 m/s and K = 1.8

Page 7-10
CEE490b

The natural frequency of the tower is equal to:

1
fo = k
2π m
3EI
1 h3
=
2π m
3 ⋅ 2 x1011 ⋅ 1.116
= 0.1592
152.4 3 ⋅ 68000
= 0.265 Hz

The critical wind speed for vortex shedding is equal to:

Vcrit = fo d / S
= 0.265 ⋅ 7.5 / 0.2
= 9.94 m/s

The stress at the base due to the vortex shedding forces at resonance with the
natural frequency of the tower is determined from:

P (t ) = Po cos ω o t

Po = 21 ρVcrit
2
C L ld
= 21 ⋅ 1.23 ⋅ 9.94 2 ⋅ 0.2 ⋅ 7.5 ⋅ 13.0
= 1.185 kN

1
at resonance, the maximum dynamic force is Po ⋅
2D

1 d
Po ⋅ ⋅ h ⋅ base
2D 2
σ base =
I base
1185 ⋅ 152.4 ⋅ 15.24
= 2
3 ⋅ 1.116
= 51.1 x10 6 N/m 2 = 51.1 MPa

The dynamic stress range is:

Fsr = σ max − σ min = 51.1 − ( −51.1) = 102.2 MPa

Page 7-11
CEE490b

From Figure K1 or Table 6A of CAN3-S16.1-M8, Fsr < Endurance Limit or Fatigue

strength for Categories A and B, therefore no further action is needed. However,
Fsr > Endurance Limit for Categories C to F, therefore it would be necessary to
estimate the number of cycles that the tower would likely be subjected to over its
lifetime.

To address this problem fully, one should know the likelihood of obtaining this
9.94 m/s wind speed for different directions. Then one could assess the fatigue at a
specific location around the circumference of the tower base, since:

⎛ M base ⋅ d cos α ⎞
⎜ ⎟
(Fsr )A =⎜ 2
⎟⎟
⎜ I base
⎝ ⎠
= 102.2 ⋅ cos α MPa
(Fsr )A = 102.2 MPa when α = 0
= 0 when α = 90

A)

The most conservative assumption that can be made is to assume that the wind
blows from the same direction all the time, so (Fsr )A is always 102.2 Mpa, whenever
the wind speed is in the critical range.

The number of cycles in the lifetime is equal to n = (0.265 x 60 x 60 ) x M

or 0.265 Hz x 60 seconds x 60 minutes x the number of hours during the lifetime, L
when V is within +/- 15% of Vcr or 8.4 m/s ≤ Vcr ≤ 11.4 m/s.

Assuming a lifetime of 50 years (from NBCC, for example), then

M = 365 x 24 x 50 x (P(V > 8.4) - P(V > 11.4))

= 438,000 x (e -(8.4/8) − e −(11.4 / 8 ) )
1.8 1.8

= 438,000 x 0.1848
= 80,946 hours per 50 years
n = 0.265 x 60 x 60 x 80,946
= 7.72 x 10 7 cycles per 50 years

From Figure K1, it can be seen that there would be fatigue damage unless the
connection used was designed to fall into Allowable Stress Category B.

Page 7-12
CEE490b

B) Assuming that the wind is equally likely to come from any direction,

α (Fsr )A = 102.2 x cosα n1=n/7 Cycle Life “N”

Category C, Fig K1
0° 102.2 1.1 x 107 1.2 x 106
15° 98.7 1.1 x 107 1.6 x 106
30° 88.5 1.1 x 107 2.2 x 106
45° 72.3 1.1 x 107 3.8 x 106
60° 51.1 1.1 x 107 ∞
F sr far below
75° 26.5 1.1 x 107 ∞
1.1 x 107 ∞ endurance limit
90° 0

Miners Damage Law assumes that:

nj
∑N
j
= Cumulative Damage j is the stress level
j

There would be fatigue damage over a broad range of wind directions, 0° < α < 45°,
should the detailing of the connection fall into Category C.

Page 7-13
CEE490b

7.3 The Estimation of Fatigue Life

The analysis of cracking in steel bridges can be accomplished through using a

linear-elastic fracture mechanics. With this method, the stresses very close to the
crack front, which cause crack extension, are treated as proportional to the stress
intensity factor, K. The size, shape and orientation of the crack play a major role in
determining the applicable K value.

Figure 7.12: Idealized crack conditions. (a), Surface crack; (b), through crack; and
(c), edge crack.

The stress intensity factor, K, for a surface crack depth, a, shown in Figure 7.12 for
three types of cracks can be related as follows:

K = Fe ⋅ Fs ⋅ Fw ⋅ Fg ⋅ σ πa

These correction factors modify σ πa , which is for the idealized case, to account
for the following:

Fs - the free surface

Fw - the finite width
Fg - the non-uniform stresses acting on the crack
Fe - the crack shape

Page 7-14
CEE490b

Numerous solutions for these correction factors can be found in the literature. A few
of the more important types are:

a
Fs = 1.211 − 0.186 for a semicircular crack in a semi-infinite plate subjected to
c
uniform stress

πa
Fw = sec for a central crack in a plate of uniform width
2b

for a three-dimensional elliptical crack shape, where E (k ) is an elliptical

1
Fe =
E (k )
integral:

π /2
[
E (k ) = ∫ 1 − k 2 sin 2 φ ]
1/ 2
dφ
0
c 2 − a2
and k2 = , where a / c is the minor to major axis ratio
c2

There are a number of approximate expressions for the gradient correction factor,
depending on the structural detail – gussets, stiffeners, cover plates, toe welds etc.
[1].

Fatigue Crack Growth Model

In order to assess the fatigue behaviour, the crack propagation relationship between
the crack growth rate and the range of the stress intensity factor, ∆K = K max − K min .
Since the crack size at the upper and lower limits of the load cycle are the same, the
stress intensity range is a function of the stress range. The “Paris Power Law” has
the form:

da
= C∆K n
dN

Figure 7.13 is a schematic representation of the crack growth relationship. A crack

growth exponent of n = 3 has been observed to be applicable to basic crack growth
rate data for structural steels as well as test data on welded members. The
corresponding average crack growth constant, C , was found to be between 1.2 to
2.18 x 10-13 if one uses mm for crack size and MPa m for ∆K . In summary an
upper bound for the crack growth relationship with the number of cycles can be
taken as:

Page 7-15
CEE490b

da
= 2.18 × 10 −13 ∆K 3
dN

Figure 7.13 General crack propagation relationship.

Since randomly variable loading is usually involved in every case of fatigue crack
propagation, an effective stress intensity range can be used, based on Miner’s Law
and a corresponding Miner’s Effective Stress Range, SrMiner . Hence, ∆K can now
be defined as:

∆K e = SrMiner Fi πa
where

[
SrMiner = ∑ α i Sri3 ]
1/ 3

This is the same relationship between cumulative damage and stress range,
namely, damage from variable loading is given by:

ni
∑ =1
Ni

where the ratio, ni / N i is the incremental damage done from a block of stress range
cycles Sri that occurs n i times. Failure is defined when the sum of the increments
equals unity.

Page 7-16
CEE490b

Fatigue Failure Estimates of the Aguasabon River Bridge

This bridge, located on the Trans-Canada Highway near Terrace Bay Ontario, is a
three span continuous plate girder bridge, constructed in 1948. The bridge has a
composite steel beam – reinforced concrete slab construction, with spans of 18.3-
24.4-18.3 m as shown in Figure 7.14. The longitudinal structural members consist of
four WF33 x 141 (84 cm deep) girders, haunched over the piers and abutments to
an overall depth of 1.3m. The haunch was fabricated by cutting the bottom flange
from the web fillet and welding a 16mm parabolic insert plate at the desired
locations. The main girders were field spliced at two points in the center span 6.7 m
from each pier as shown in Figure 7.15. The splice plates were located at the point
of dead load inflection.

In 1963, cracks were discovered at the vertical butt weld detail in three of the six
haunch inserts of the north interior main girder as shown in Figure 7.16. One of
these cracks extended 1.12 m into the girder web along a diagonal line starting from
this weld detail. These cracks were repaired by welding cover plates on each side of
the web and welding an insert in the bottom flange.

In 1973, the structure was subjected to a detailed investigation and dynamic testing.
As a result, four additional cracks were found. During the repairs, it was found that
the cracks stemmed from initial weld imperfections or inclusions in the short
transverse groove welds at the ends of the parabolic haunches. All cracks were
discovered before the flanges fractured because the details were located near the
point of inflection where the dead load stresses were small. Hence, large fatigue
cracks were able to develop from repetitive live loads without brittle fracture of the
remaining section.

During the 1973 tests, a test vehicle loaded to a gross weight of 405 kN traversed
the bridge and dynamic strains were recorded. Based on the dynamic strain
measurements and traffic conditions, a representative stress range histogram (Sr )
was determined as shown in Figure 7.17. The stress range histogram was used to
estimate the effective RMS stress range SrRMS and SrMiner . If all stress range
conditions above 6 Mpa are considered, the SrRMS = 13.2 MPa. The effective stress
range using Miner’s Law was equal to SrMiner = 13.2 Mpa for all stress cycles above
3.1 Mpa.

A sample of vehicles crossing the bridge indicated that 6 to 15 stress cycles above
43.1 Mpa would result on each passage. The smoothed histogram shown in Figure
7.17, indicated 25 million stress cycles exceeded 3.1 Mpa between 1948 and 1963
and an additional 16 million cycles between 1963 and 1973.

Page 7-17
CEE490b

Figure 7.14 Plan and Elevation of Aguasabon Bridge (crack locations are indicated
with the mark X).

Figure 7.15 Main girder haunch detail, field shear splice and connector detail of
Aguasabon Bridge.

Page 7-18
CEE490b

Figure 7.16 Location of weld crack in main girder (see Figure 5.3b for a photograph
of circled area).

Figure 7.17 Stress range histogram (smoothed) of Aguasabon Bridge.

Page 7-19
CEE490b

Figure 7.18 Assumed stages of crack growth (outer edge of weld inclusion is
shown as dark line).

The fatigue propagation stages were modeled by approximating the weld defect by
an ellipse as shown in Figure 7.18. The stress intensity factor for Stage 1 was
modeled as:

K = Fe Fw σ πa

where:

Fe =
1
E (k )
[
π /2
]2 1/ 2
and E (k ) = ∫ 1 − k sin φ
2
dφ and k =
2
2 c −a
c2
2

πa
Fw = sec
2b

Page 7-20
CEE490b

The minor to major diameters of the ellipse a / c = 0.11, b = 61 mm

The “penny” shape of the crack assumed in the second stage of growth resulted in:

π πa
K= sec σ πa
2 2b

where a is the crack radius and ainitial = c final for Stage 1.

The crack growth propagation equation used to estimate the fatigue life for each
stage of crack propagation was:

61mm da
N= ∫ −13
a − initial 2.18 × 10 ∆K 3

Several different crack sizes were assumed to assess the sensitivity of the initial
inclusion to the fatigue life. The results are shown in Table 1.

For the crack shown in Figure 7.19, it was estimated that the effective stress range
was 13.2 Mpa, occurring at a rate of 1.5 million cycles per year. At this rate
approximately, two to three years would be required before the crack penetrated into
the bottom flange during the first stage of crack growth. An additional 15 or 20 years
would be required before the crack penetrated the full depth of the bottom flange.
These estimates are in good agreement with that observed for the Aguasabon River
Bridge.

Table 1 Estimated Fatigue Life: SrMiner = 1.92 ksi (13.2 MPa)

Stage 1: Stage 2:
Crack Growth through Weld Crack Growth through Flange
Initial Cycles Years Initial Cycles Years
Crack/Size of to Crack/Radius of to
ai (in.) Stress Achieve (in.) Stress Achieve
0.20 14,050,000 9.4 1.40 25,500,000 17
(5 mm) (35.6 mm)
0.25 2,664,000 1.8 1.42 24,150,000 16
(6.4 mm) (36 mm)
0.30 40,500 0.3 1.45 21,960,000 14.6
(7.6 mm) (36.8 mm)
1.50 18,600,000 12.4
(38 mm)

References

1. Fisher, J.W., “Fatigue and Fracture in Steel Bridges”, John Wiley & Sons,
New York, 1984.

Page 7-21