University of Bath Department of Architecture & Civil Engineering Page 1 of 28

Structures 4 lecture notes

Buckling

Buckling calculations are very difficult except for a few special cases,

and so numerical methods on a computer are almost invariably used in

practice for buckling modes involving plates, shells and assemblies of

beams and columns. Single columns and beams aren’t too bad.

The compression flange of a beam acts more or less like a column

except that the torsional and warping resistance are involved.

However the simple special cases are still worth examining because

they tell us what to look for in a numerical analysis.

The elastica

dψ
The sagging curvature is . This is the definition of curvature. The
ds

sagging moment is equal to EI times the curvature. Sagging moment

is also equal to −Py and thus

University of Bath Department of Architecture & Civil Engineering Page 2 of 28

dψ
s

M = EI = −Py = −P ∫ sinψ ds
ds s=0

d 2ψ
EI = −P sinψ
ds 2
dψ d 2ψ dψ
EI 2
= −P sinψ
ds ds ds
1 ⎛ dψ ⎞
2

EI ⎜ ⎟ = P ( cosψ − cosψ 0 )
2 ⎝ ds ⎠

where ψ 0 is the value of ψ at s = 0 .

Hence we have

1 ⎛ dψ ⎞
2

EI ⎜ ⎟ = P ( cosψ − cosψ 0 )
2 ⎝ ds ⎠
ψ .
EI dψ
s= ∫
2P ψ =ψ 0 cosψ − cosψ 0

If we write ψ = 2θ , cosψ = cos2θ = 1 − 2sin 2 θ and

ψ ψ
EI 2dθ EI dθ
s= ∫
2P ψ =ψ 0 2 sin 2 θ 0 − sin 2 θ
= ∫
P sin θ 0 ψ =ψ 0 1 − k 2 sin 2 θ
2

.
1
k=
sin θ 0

This is an incomplete elliptic integral of the first kind, - see

http://en.wikipedia.org/wiki/Elliptic_integral

This means that it cannot be evaluated using elementary functions –

trigonometric functions, hyperbolic functions etc.

Alternative derivation:
University of Bath Department of Architecture & Civil Engineering Page 3 of 28

dy
tanψ =
dx
dψ d 2 y dx d 2 y
sec 2 ψ = 2 = cosψ
ds dx ds dx 2
d2y d2y d2y
dψ 2
dx 2 dx 2
= dx3 = =
sec ψ 3 3
ds
(
1 + tan 2 ψ ) 2 ⎛ ⎛ dy ⎞ 2 ⎞ 2
⎜ 1 + ⎜⎝ dx ⎟⎠ ⎟
⎝ ⎠

Thus

d2y
EI
dx 2 + Py = 0
3
⎛ ⎛ dy ⎞ ⎞ 2 2

⎜ 1 + ⎜⎝ dx ⎟⎠ ⎟
⎝ ⎠
dy d 2 y
EI
dx dx 2 + Py dy = 0
3
dx
⎛ ⎛ dy ⎞ 2 ⎞ 2
⎜ 1 + ⎜⎝ dx ⎟⎠ ⎟
⎝ ⎠
EI EI 1
− + Py 2 = 0
⎛ dy ⎞
2
⎛ dy ⎞ 2
2

1+ ⎜ ⎟ 1+ ⎜ ⎟
⎝ dx ⎠ 0 ⎝ dx ⎠

We can carry on, but we will get stuck again with an elliptic integral.

dy
However, if we assume that is small,
dx

1 1 1 ⎛ dy ⎞
2
⎛ ⎛ dy ⎞ 2 ⎛ dy ⎞ 2 ⎞
≈ 2 ≈1− ⎜ ⎟ so that EI ⎜ ⎜ ⎟ − ⎜ ⎟ ⎟ + Py 2 = 0 .
⎛ dy ⎞
2
1 ⎛ dy ⎞ 2 ⎝ dx ⎠ ⎝ ⎝ dx ⎠ ⎝ dx ⎠ 0 ⎠
1+ ⎜ ⎟ 1+ ⎜ ⎟
⎝ dx ⎠ 2 ⎝ dx ⎠

This is satisfied by
University of Bath Department of Architecture & Civil Engineering Page 4 of 28

⎛ P ⎞
y = Bsin ⎜ x
⎝ EI ⎟⎠
dy P ⎛ P ⎞
=B cos ⎜ x .
dx EI ⎝ EI ⎟⎠
⎛ dy ⎞ P
⎜⎝ ⎟⎠ = B
dx 0 EI

This gives the Euler column formula,

π 2 EI π 2 EA
P= = 2
L2 ⎛ L⎞
⎜⎝ ⎟⎠
r
A = cross-sectional area
I = Ar 2 .
r = radius of gyration
L
= slenderness ratio
r

For columns other than pin-ended columns, L is the effective

length. For cantilever columns the effective length is more than twice

the actual length, depending on how stiff the moment connection at

the base is.

Analysis of the elastica shows that if a column remains elastic the load

continues to increase as the column buckles.

University of Bath Department of Architecture & Civil Engineering Page 5 of 28

The above figure shows a buckled pin ended column of length L and

bending stiffness EI . The column is initially perfectly straight. The

relationship between the buckling load and the shortening due to

P δ π 2 EI
bending is = 1+ where PEuler = 2 is the Euler buckling load.
PEuler 2L L

This formula is obtained using complete elliptic integrals as described

in §2.7 of Timoshenko and Gere, Theory of Elastic Stability and applies

δ
for small values of .
L

However, it can be shown that for the truss column below that the load

F decreases with deflection (see question 3 in the 2013-14 Structures

4 exam). This is because as the truss deflects the buckled member

attracts more than its fair share of the load.

University of Bath Department of Architecture & Civil Engineering Page 6 of 28

Perry Robinson formula

See the A history of the safety factors by Alasdair N. Beal, The

Structural Engineer 89 (20) 18 October 2011,

http://anbeal.co.uk/TSE2011HistoryofSafetyFactors.pdf

for a fascinating discussion of safety factors including the Perry

Robertson formula.

⎛ πx⎞
Assume column has an initial bend, y = ζ L sin ⎜ ⎟ (note that ζ is
⎝ L⎠

dy
dimensionless) and that is SMALL. Then the sagging moment,
dx

⎛ d2y d2 ⎛ ⎛ π x⎞⎞⎞
M = EI × change of curvature = EI ⎜ 2 − 2 ⎜ ζ L sin ⎜ ⎟ ⎟ ⎟ = −Py
⎝ dx dx ⎝ ⎝ L ⎠⎠⎠
d2y π2 ⎛ πx⎞
EI + Py = −EIζ L sin ⎜ ⎟
dx 2 L2 ⎝ L⎠

⎛ πx⎞
Try solution y = Bsin ⎜ ⎟ then
⎝ L⎠

π2 ⎛ πx⎞ ⎛ πx⎞ π2 ⎛ πx⎞

−EIB 2 sin ⎜ ⎟ + PBsin ⎜ ⎟ = −EIζ L 2 sin ⎜ ⎟
L ⎝ L⎠ ⎝ L⎠ L ⎝ L⎠
π2
EIζ L .
B= L2 = ζ L
π2 P
EI 2 − P 1 − 2
L π EI
L2

The maximum stress is equal to

University of Bath Department of Architecture & Civil Engineering Page 7 of 28

PLζ P LAζ
P M P PB P Z P
σ max = + = + = + = + A Z .
A Z A Z A 1− P A 1− P
π EI
2
PEuler
2
L

I
Z= is the section modulus.
c

I Ar 2 P
If we set I = Ar , Z = =
2
σ = , σ max = σ y and
c c A

PEuler π 2 EI π 2E
σ Euler = = = , then we have
A AL2 ⎛ L ⎞ 2
⎜⎝ ⎟⎠
r

θσ
σy =σ + .
σ
1−
σ Euler

LAζ Lcζ
in which θ = = 2 .
Z r

Therefore

(σ Euler − σ )(σ y − σ ) + θσ Eulerσ = 0

.
( )
σ 2 − σ y + (1+ θ )σ Euler σ + σ yσ Euler = 0

1
(
σ y + (1+ θ )σ Euler −)1
(σ + (1+ θ )σ Euler )
2
σ= y − 4σ yσ Euler
2 2
1
(
= σ y + (1+ θ )σ Euler − )1
(σ + (1+ θ )σ Euler ) − 4 (1+ θ )σ yσ Euler + 4θσ yσ Euler .
2
y
2 2
1
(
= σ y + (1+ θ )σ Euler − )1
(σ − (1+ θ )σ Euler )
2
y + 4θσ yσ Euler
2 2

When θ = 0 ,

σ=
1
2
( 1
)
σ y + σ Euler − σ y − σ Euler
2 .
= σ y or σ Euler
University of Bath Department of Architecture & Civil Engineering Page 8 of 28

π 2E L E
Note that σ Euler = σ y when 2 = σ y so that =π .
⎛ L⎞ r σy
⎜⎝ ⎟⎠
r

When σ Euler is very small,

σ!
1
2
( )
σ y + (1+ θ )σ Euler −
1
2
σ y2 − ( 2 − 2θ )σ yσ Euler

σ
1
( ) 1
= σ y + (1+ θ )σ Euler − σ y 1− 2 (1− θ ) Euler
2 2 σy

1 ⎡ σ ⎤
!
1
2
( )
σ y + (1+ θ )σ Euler − σ y ⎢1− (1− θ ) Euler ⎥
2 ⎣ σy ⎦
! σ Euler

and when σ Euler is very large,

σ!
1
2
(σ y + (1+ θ )σ Euler −)1
2
(1+ θ )2 σ Euler
2
− ( 2 − 2θ )σ yσ Euler

(1− θ ) σ y
=
1
2
( )
1
σ y + (1+ θ )σ Euler − (1+ θ )σ Euler 1− 2
2 (1+ θ )2 σ Euler
⎛ (1− θ ) σ y ⎞
!
1
2
( )
1
σ y + (1+ θ )σ Euler − (1+ θ )σ Euler ⎜ 1−
2 2 ⎟
⎝ (1+ θ ) σ Euler ⎠
1 ⎛ (1− θ ) ⎞ σ y
= σ y ⎜ 1+ =
2 ⎝ (1+ θ ) ⎟⎠ 1+ θ
University of Bath Department of Architecture & Civil Engineering Page 9 of 28

E
Perry-Robertson graphs with = 1000 and θ = 0, 0.01, 0.1 and 1 .
σy

See http://en.wikipedia.org/wiki/Perry_Robertson_formula

This is the basis for column design.

Timoshenko or Cosserat column

University of Bath Department of Architecture & Civil Engineering Page 10 of 28

The figure on the left a shows the buckling of a battened or Vierendeel

column. The deformation has been exaggerated so that the deflected

shape can be seen.

The figure on the right shows a detail of two bays in which the circles

show the assumed points of contraflexure half way along the members.

If the column is treated as a Timoshenko or Cosserat beam, the

differential equations describing deformation of the column are

d ⎛ dy ⎞
M = Py = −EI composite ⎜⎝ − θ ⎟⎠
dx dx
dy
kθ = F = P
dx

in which x is the vertical coordinate along the column, y is the lateral

displacement of the column, M is the overall bending moment, P is

the overall axial load and F is the overall shear force. I composite is the

fully composite second moment of area and k is the shear stiffness of

the column.

Hence

d ⎛ dy P dy ⎞
EI composite ⎜ − ⎟ + Py = 0
dx ⎝ dx k dx ⎠
.
⎛ P⎞ d y
2
EI composite ⎜ 1− ⎟ 2 + Py = 0
⎝ k ⎠ dx

If the column is pin-ended and its length is length L , the differential

equation is satisfied by

πx
y = Bsin
L

if
University of Bath Department of Architecture & Civil Engineering Page 11 of 28

π 2 EI composite ⎛ P ⎞
P= ⎜⎝ 1− ⎟⎠
L2 k
⎛ P⎞
= PEuler ⎜ 1− ⎟
⎝ k⎠
.
⎛ P ⎞
P ⎜ 1+ Euler ⎟ = PEuler
⎝ k ⎠
kPEuler
P=
k + PEuler

The shear stiffness k depends upon the bending stiffness and length

of the individual members. In the above column all the horizontal

members all have length 2a so that the length CF is a . The vertical

members all have length 2b so that the lengths BC and CD are both

b . The horizontal members all have second moment of area I horizontal

and the vertical members all have second moment of area I vertical and

cross-sectional area Avertical . The members are all made from a material

with Young’s modulus E .

Shear deformation means that an angle such as DĈF is deformed from

π
a right angle to + θ due to bending of the members. The
2

connections between the members are assumed to be rigid.

If we ignore I vertical , then the parallel axis theorem gives

I composite = 2Averticala 2 .

To calculate the shear stiffness we first note that that the tip deflection

of a cantilever of bending stiffness EI and span S loaded with a point

wS 3
load W at its tip is . The shear force in each of the vertical
3EI
University of Bath Department of Architecture & Civil Engineering Page 12 of 28

F
members is and the shear force in the horizontal members is
2
F
2 b
2 = Fb . Thus the shear deformation is
a a

⎛ F 3 ⎞ ⎛ Fb 3 ⎞
b a
⎜ 2 ⎟ ⎜ a ⎟
⎜ 3EI vertical ⎟ ⎜ 3EI horizontal ⎟
⎜⎝ ⎟⎠ ⎜⎝ ⎟⎠ Fb 2 Fab
θ= + = + .
b a 6EI vertical 3EI horizontal

1
Hence k = 2 .
b ab
+
6EI vertical 3EI horizontal

Buckling of plates and shells

See, for example, Don O. Brush and Bo O. Almroth, Buckling of Bars,

Plates and Shells, McGraw-Hill, New York 1975.

Plates behave fairly well when they buckle, the load usually does not

drop off dramatically and may increase after buckling. On the other

hand shells, including axially compressed cylinders, can be very

imperfection sensitive so they collapse at a much smaller load than the

eigenvalue buckling load - see

Hunt, G. W., 2011. Reflections and symmetries in space and time. IMA

Journal of Applied Mathematics, 76 (1), pp. 2-26.

http://imamat.oxfordjournals.org/content/76/1/2
University of Bath Department of Architecture & Civil Engineering Page 13 of 28

Simply supported flat plate

Image from Brush and Almroth, Buckling of Bars, Plates and Shells

For a straight beam we have

d 2v
EI + Pv = 0
dx 2
d 4v d 2v
EI 4 + P 2 = 0
dx dx

where v is the displacement in the y direction. The corresponding

equation for a flat plate is

⎛ ∂4 w ∂4 w ∂4 w ⎞ ∂2 w
D⎜ 4 + 2 2 2 + 4 ⎟ +σ x 2 = 0
⎝ ∂x ∂x ∂y ∂y ⎠ ∂x
Et 3
D=
12 (1− υ 2 )

in which it is assumed that that there is only a membrane stress σ x in

the x direction. Membrane stress has units force per unit width. w is

the displacement in the z direction, t is the plate thickness and υ is

Poisson’s ratio. D is the bending stiffness per unit width and it

bd 3
replaces the E for a rectangular beam.
12
University of Bath Department of Architecture & Civil Engineering Page 14 of 28

Let us suppose that the plate is simply supported along y = 0 and y = b

and that it is long in the x direction. The differential equation is

satisfied and the y = 0 and y = b boundary conditions are satisfied by

2π x π y
w = Asin sin
λ b

if

2
⎛ 2π ⎞
2
⎛ ⎛ 2π ⎞ 4 ⎛ 2π ⎞ ⎛ π ⎞ ⎛ π ⎞ ⎞
2 2 4
⎛ ⎛ 2π ⎞ 2 ⎛ π ⎞ 2 ⎞
σ x ⎜ ⎟ = D⎜ ⎜ ⎟ + 2⎜ ⎟ ⎜ ⎟ + ⎜ ⎟ ⎟ = D⎜ ⎜ ⎟ + ⎜ ⎟ ⎟
⎝ λ ⎠ ⎝⎝ λ ⎠ ⎝ λ ⎠ ⎝ b⎠ ⎝ b⎠ ⎠ ⎝⎝ λ ⎠ ⎝ b ⎠ ⎠

so that

2
⎛ ⎛π⎞ ⎞
2

⎜ ⎛ 2π ⎞ ⎜⎝ b ⎟⎠ ⎟
σ x = D⎜⎜ ⎟ + ⎟
⎜ ⎝ λ ⎠ ⎛⎜ 2π ⎞⎟ ⎟
⎜⎝ ⎝ λ ⎠ ⎟⎠

which is minimum when

⎛ ⎛π⎞ ⎞⎛ ⎛π⎞ ⎞
2 2

dσ x ⎜ ⎛ 2π ⎞ ⎜⎝ b ⎟⎠ ⎟ ⎜ ⎜⎝ ⎟⎠ ⎟ 2π
b
= −2D ⎜ ⎜ ⎟ + ⎟ ⎜ 1− 2⎟ =0
dλ ⎜ λ⎝ ⎠ ⎛ 2π ⎞ ⎟ ⎜ ⎛ 2π ⎞ ⎟ λ 2 .
⎜⎝ ⎜ ⎟
⎝ λ ⎠ ⎟⎠ ⎜⎝ ⎜⎝ λ ⎟⎠ ⎟⎠
λ = 2b

Thus the buckling membrane stress is

⎛π⎞
2

σ x = 4D ⎜ ⎟ .
⎝ b⎠

Simple single degree of freedom models

These will be introduced in lectures, with particular reference to post-

buckled stability and imperfection sensitivity. In particular the

University of Bath Department of Architecture & Civil Engineering Page 15 of 28

difference between non-linear and linear (eigenvalue) buckling will be

emphasised.

Linear or eigenvalue buckling

The above structure is loaded by a system of loads that are all

multiplied by the same load factor λ . Imagine that it is analysed as a

linear elastic structure with stiffness matrix K and that the axial

forces in the members are found. These axial forces will all be

proportional to λ and from these forces we can find the geometric

stiffness matrix − λ G . The minus is put there because compressive

forces produce a negative stiffness.

Linear buckling occurs when

[K − λG]δ = 0
K −1 [ K − λ G ] δ = 0
⎡⎣ K −1K − λ K −1G ⎤⎦ δ = 0 .
⎡⎣ I − λ K −1G ⎤⎦ δ = 0
⎡ −1 1 ⎤
⎢⎣ K G − I δ=0
λ ⎥⎦
University of Bath Department of Architecture & Civil Engineering Page 16 of 28

We are only interested in the lowest buckling load for which the load

factor λ is equal to one over the highest eigenvalue of K −1G . The

corresponding eigenvector gives the mode shape.

Note the similarity to natural frequencies and mode shapes. Buckling

corresponds to the natural frequency becoming zero.

Note that linear eigenvalue buckling analysis gives no information

about imperfection sensitivity.

Therefore non-linear buckling analysis should always be done if there

is any question of imperfection sensitivity.

The P − Δ effect refers to the moment caused by side-sway of a

column. It is sometimes not clear what is meant by P − Δ analysis in a

piece of software regarding linear or non-linear. Note that rotation is

not a vector unless it is small and this assumption is often made even

in so called non-linear analysis.

University of Bath Department of Architecture & Civil Engineering Page 17 of 28

Lagrange's Equations of Motion

1 n n 1
Kinetic energy = T = ∑ ∑ M ijδiδ j = δ T Mδ where n is the number of
2 i=1 j=1 2

degrees of freedom. The M ij are functions of the δ ’s only.

dT 1 n n ⎛ n
∂ M ij ! ! ! ⎞
= ∑ ∑ ⎜ M ijδ!!iδ! j + M ijδ!iδ!!j + ∑ δ iδ jδ k ⎟
dt 2 i=1 j=1 ⎝ k=1 ∂δ k ⎠
⎡ ⎛ ⎛ ∂ M jk ∂ M kj ⎞ ⎞⎤
+
n
⎢
= ∑ δi ∑! ⎜
(
⎢ n ⎜ M +M
ij )
δj + ∑
ji !!
n ⎜
1 ⎝ ∂δ i ⎟
∂δ i ⎠ ! ! ⎟ ⎥
δ jδ k ⎟
⎥

i=1
⎢ j=1 ⎜ 2 2 k=1 2 ⎟⎥
⎢ ⎜⎝ ⎟⎠ ⎥
⎣ ⎦
⎡ ⎛ ⎛ ∂ M jk ∂ M kj ⎞ ⎛ ∂ M jk ∂ M kj ⎞ ⎞⎤
+ +
n ⎢
⎢
= ∑ δi ∑ ⎜
! (
n ⎜ M + M
ij
δj +∑
ji !! )
n ⎜
⎝ ∂δ i ∂δ i ⎟⎠ ! ! 1 n ⎜⎝ ∂δ i
δ jδ k − ∑
∂δ i ⎟⎠ ! ! ⎟ ⎥
δ jδ k ⎟
⎥

i=1
⎢ j=1 ⎜ 2 k=1 2 2 k=1 2 ⎟⎥
⎢ ⎜⎝ ⎟⎠ ⎥
⎣ ⎦
n ⎡
⎛ d ⎛ ∂T ⎞ ∂T ⎞ ⎤
= ∑ ⎢δ!i ⎜ ⎜ ! ⎟ − ⎟⎥
⎣ ⎝ dt ⎝ ∂δ i ⎠ ∂δ i ⎠ ⎥⎦
i=1 ⎢

If U is the strain energy and G is the gravitational potential energy, by

conservation of energy,

n ⎡ ⎛ d ⎛ ∂ T ⎞ ∂ T ∂U ∂ G ⎞ ⎤
d
0= (T + U + G ) = ∑ ⎢δi ⎜ ⎜  ⎟ − + + ⎟ ⎥
dt ⎣ ⎝ dt ⎝ ∂δ i ⎠ ∂δ i ∂δ i ∂δ i ⎠ ⎦⎥
i=1 ⎢

This equation applies for arbitrary δi and therefore

d ⎛ ∂ T ⎞ ∂ T ∂U ∂ G
− + + =0 .
dt ⎜⎝ ∂δi ⎟⎠ ∂δ i ∂δ i ∂δ i

These are Lagrange’s equations of motion and apply for i = 1 to n .

University of Bath Department of Architecture & Civil Engineering Page 18 of 28

Verlet integration

If the equations of motion can be rearranged so that:

( )
δ1 = q1 δ1, δ 2 ,..., δ n , δ1, δ2 ,..., δn , p1, p2 ,..., pn

δ = q (δ , δ ,..., δ , δ , δ ,..., δ , p , p ,..., p )

2 2 1 2 n 1 2 n 1 2 n

.
.
.
.
(
δn = qn δ1, δ 2 ,..., δ n , δ1, δ2 ,..., δn , p1, p2 ,..., pn )
in which the loads p1, p2 ,..., pn are known values of time and if we also

know the initial values of δ1, δ 2 ,..., δ n , δ1, δ2 ,..., δn , then we can step

through time updating values as follows:

δ1 = δ1 + δ1Δt δ1 = δ1 + δ1Δt

δ = δ + δ Δt
2 2 2 δ = δ + δ Δt
2 2 2

. .
. and . .
. .
. .
δn = δn + δn Δt δ n = δ n + δn Δt

There are a number of essentially similar ‘explicit’ methods like this

with names such as Verlet Integration, Störmer's method, Gauss–Seidel

and dynamic relaxation.

University of Bath Department of Architecture & Civil Engineering Page 19 of 28

Definitions of some terms used for non-

aeroelastic vibrations

Note that single degree of freedom systems are very important

because multidegree of freedom systems can be reduced to uncoupled

single degree of freedom systems using modal analysis and the

orthogonality conditions discussed in lectures – this is ignoring the

coupling due to damping.

Summary of results

In order to solve the equations

•• •
M δ + Dδ + Kδ = p

n
for a system with n degrees of freedom, we write δ = ∑ fr (t ) Δ r where
r=1

Δ r are the eigenvectors of K −1M .

If we ignore coupling due to damping,

•• •
mr fr + λr fr + kr fr = pr (t )
mr = Δ rT MΔ r
kr = Δ rT KΔ r
λr = 2c kr mr
pr (t ) = Δ rT p
c = non-dimensional damping ratio

which is the equation for a single degree of freedom system.

University of Bath Department of Architecture & Civil Engineering Page 20 of 28

Single degree of freedom system

Figure 1
Figure 1 shows a typical ‘random’ load, p(t ) .

The mean value of p(t ) is µ p =

2
1
T ∫ p (t ) dt as T → ∞ .
T
−
2

T
2
1
The auto-correlation function is Rpp (τ ) = ∫ p (t ) p (t + τ ) dt as T → ∞ .
T T
−
2

The auto-correlation function of p(t ) minus its mean is

T
2
1
T ∫ ⎡⎣ p (t ) − µ
T
p
⎤⎦ ⎡⎣ p (t + τ ) − µ p ⎤⎦ dt as T → ∞
−
2

= Rpp (τ ) − µ p2 .

The mean-square spectral density is

1 ∞ 1 ∞
φ pp (ω ) = ∫ ⎡⎣ Rpp (τ ) − µ p2 ⎤⎦ e −iωτ dτ = ∫ ⎡⎣ Rpp (τ ) − µ p2 ⎤⎦ cos (ωτ ) dτ . Here
2π −∞ 2π −∞

ω = 2π × frequency .

From the theory of Fourier transforms,

∞ ∞
Rpp (τ ) = µ p2 + ∫ φ pp (ω ) eiωτ dω = µ p2 + ∫ φ pp (ω ) cos (ωτ ) dω .
−∞ −∞
University of Bath Department of Architecture & Civil Engineering Page 21 of 28

Note that µ p , R pp (τ ) and φ pp (ω ) are all real, and R pp (−τ ) = R pp (τ ) and

φ pp (−ω ) = φ pp (ω ) .

σ p is the standard deviation of p(t ) which is defined as

T
2
1
σ p2 =
T ∫ ⎡⎣ p (t ) − µ
T
p
⎤⎦ ⎡⎣ p (t ) − µ p ⎤⎦ dt as T → ∞
−
2
∞
= Rpp ( 0 ) − µ p2 = ∫ φ pp (ω ) dω
−∞

The mean-square spectral density tells us about the amount of

‘energy’ at different frequencies, but gives no information about the

relative phase.

ψp is the root mean square (rms) value of p(t ) and

ψ 2p = µ2p + σ 2p = R pp (0) .

p(t)
x(t)

s λ

Figure 2
Figure 2 shows a mass - spring- damper. When the load p(t ) is

applied,

d2x dx
m 2 + λ + sx = p (t )
dt dt
so that

1 d 2 x 2c dx p (t )
+ +x=
Ω dt
2 2
Ω dt s
University of Bath Department of Architecture & Civil Engineering Page 22 of 28

s
where Ω = = 2π × natural frequency and the viscous damping factor,
m

λ
c= .
2 sm

µp
The mean value of x (t ) is µ x = .
s

⎛ φ pp (ω ) ⎞
⎜⎝ s 2 ⎟⎠
The mean-square spectral density of x (t ) is φ xx (ω ) = 2 .
⎛ ω 2 ⎞ ⎛ 2cω ⎞
2

⎜⎝ 1 − Ω 2 ⎟⎠ + ⎜⎝ Ω ⎟⎠

1
Figure 3 shows plots of 2
.
⎛ ω ⎞ ⎛ 2cω ⎞
2 2

⎜⎝ 1 − Ω 2 ⎟⎠ + ⎜⎝ Ω ⎟⎠
University of Bath Department of Architecture & Civil Engineering Page 23 of 28

Figure 3

The standard deviation of x (t ) is σ x =

∞
∫ φ xx (ω ) dω = Rxx ( 0 ) − µ x2 and if
−∞

c is small,

1 ∞
πΩ ∫ ⎡⎣ Rpp (τ ) − µ p2 ⎤⎦ cos (Ωτ ) dτ
1 πΩφ ( Ω ) 1 2π −∞
σx = pp
=
s 2c s 2c
∞
1 Ω ∫−∞ ⎡⎣ Rpp (τ ) − µ p ⎤⎦ cos (Ωτ ) dτ
2

=
s 4c
where 2πΩ is the natural frequency.
University of Bath Department of Architecture & Civil Engineering Page 24 of 28

πΩφ pp (Ω )
The ‘dynamic magnification factor’ is . Note that this
2cσ p2

dynamic magnification factor applies only to the dynamic

component of the load.

Figure 4 shows the response to the load.

Figure 4
Note the dynamic magnification which can be seen by comparing the mean and standard
deviation of the load and the response.

Multi-degree of freedom systems

Let us suppose the ‘load’ exciting a particular mode of vibration is

N
p (t ) = ∑ Ai qi (t ) where the qi (t ) are the pressures and the Ai are the
i=1

associated areas times displacement in the mode (which may be

positive or negative).

N
The mean of the load exciting the mode is µ p = ∑ Ai µqi and the auto-
i=1

correlation function is
University of Bath Department of Architecture & Civil Engineering Page 25 of 28

N N
Rpp (τ ) = ∑ ∑ Ai A j Rqiq j (τ )
i=1 j=1
N N
= µ p2 + ∑ ∑ Ai A j ⎡⎣ Rqiq j (τ ) − µqi µq j ⎤⎦
i=1 j=1

where the cross-correlation,

T
2
1
Rqiq j (τ ) = Rq j qi ( −τ ) = ∫ q (t ) q (t + τ ) dt as T → ∞ .
i j
T T
−
2

1 ∞
Again φ pp (ω ) = ∫ ⎡⎣ Rpp (τ ) − µ p2 ⎤⎦ e −iωτ dτ .
2π −∞

Note that in general Rqiq j ( −τ ) ≠ Rqiq j (τ ) so that cross mean-square

spectral density,

1 ∞
φqiq j (ω ) = φqiq j ( −ω ) = φq j qi ( −ω ) = ∫ ⎡ Rq q (τ ) − µq µq ⎤ e −iωτ dτ
⎣ ij j ⎦
will be a
2π −∞ i

complex function.

N N
However in doing the summation φ pp (ω ) = ∑ ∑ Ai A jφqiq j (ω ) , the
i=1 j=1

imaginary parts will cancel out.

In terms of correlation functions, the response is given by

⎛ N N ∞ ⎞
∫ ⎜ ∑ ∑
1 −∞ ⎝ i=1 j=1
A A ⎡ R
i j ⎣ qi q j (τ ) − µ qi q j ⎦ ⎟ cos ( Ωτ ) d τ
µ ⎤
⎠
σx = .
m 4c

A note on Fourier series

For our purposes a stochastic random load can be approximated by

the Fourier series,

( ))
∞
p (t ) = µ p + ∑ (
2φ pp (ω n ) Δω 2 cos ω n t + β p (ω n )
n=1
University of Bath Department of Architecture & Civil Engineering Page 26 of 28

2πn 2π
where ωn = and Δω = if the period, T , is sufficiently large. Stochastic means
T T
‘governed by the laws of probability’.
The autocorrelation function,
⎛⎡
( ⎤
))
⎞
∞
T
⎜⎢ µ p + ∑ (
2φ pp (ω n ) Δω 2 cos ω n t + β p (ω n ) ⎥ ⎟
1 ⎣
2
⎦
Rpp (τ ) = ∫ ⎜ ⎟ dt as T → ∞
n=1

T T ⎜⎡
( )) ⎤⎟
∞
− µ
2⎜⎢ p
⎝⎣
+ ∑ (
2φ pp (ω n ) Δω 2 cos ω n (t + τ ) + β p (ω n ) ⎥⎟
n=1 ⎦⎠
T

∫ ∑ (2φ (ω ) Δω 2 cos (ω (t + τ ) + β (ω )) cos (ω t + β (ω ))) dt

2 ∞
1
= µ p2 + pp n n p n n p n
T T n=1
−
2

( )
T

1 2 ⎛ ∞ ⎛ cos (ω nτ ) cos2 ω n t + β p (ω n ) ⎞⎞
=µ +2
∫T ∑ ⎜ 2φ pp (ω n ) Δω 2 ⎜ ⎟ ⎟ dt
( ) (
⎜⎝ −sin (ω nτ ) sin ω n t + β p (ω n ) cos ω n t + β p (ω n ) ) ⎟⎠ ⎟⎠
p
T n=1 ⎜
− ⎝
2
∞

(
= µ p2 + ∑ 2φ pp (ω n ) Δω cos (ω nτ ) )
n=1
∞

(
= µ p2 + 2 ∑ φ pp (ω n ) cos (ω nτ ) Δω )
n=1

You can also use the Fourier transform, but this seems to cause problems with spectral
T T
densities unless you limit the time to − ≤t≤ .
2 2

Duhamel's integral

Duhamel's integral is a bit like Verlet integration, except that it only

applies to linear systems.

The unloaded single degree of freedom system:

d2x dx 1 d 2 x 2c dx p (t )
m + λ + sx = 0 or + +x= where
dt 2
dt Ω dt
2 2
Ω dt s

s
Ω= = 2π × natural frequency and the viscous damping factor,
m

λ
c=
2 sm

is satisfied by
University of Bath Department of Architecture & Civil Engineering Page 27 of 28

( ((
x = e −cΩ t Asin ) )
1 − c 2 Ω t + B cos (( 1 − c ) Ω t )) .
2

If the mass is stationary and receives an impulse I at t = 0 , then when

t>0,

x=
I e
−cΩ t
sin (( 1 − c ) Ω t ) .
2

m
( 1 − c )Ω 2

Duhamel's integral treats the load as lots of little impulses so that

(( 1 − c )Ω (t − τ )) p(τ ) dτ .
t
1 −cΩ (t−τ )
x= ∫e
2

( 1− c )
sin
m 2
Ω 0

Seismic excitation

The ground motion in the above is y (note that horizontal ground

motion is more of a problem than vertical). The equation of motion is

x + λ ( x − y ) + s ( x − y ) = 0 .
m

We can rewrite this as

x + λ x + sx = λ y + sy
m

so that the ‘load’ is λ y + sy . However it is more usual to write

•• •
m ( x − y )+ λ ( x − y )+ s ( x − y ) = −m
y
University of Bath Department of Architecture & Civil Engineering Page 28 of 28

in which ( x − y ) is the motion relative to the ground and it is this

motion which causes the stresses in the structure. The ‘load’ is now

simply −m
y . This is easy to implement in matrix notation for multi-

degree of freedom systems.

Aeroelasticity

This is discussed in lectures using the example of the Fokker E.V (later

the D-VIII) monoplane (divergence) and the Tacoma Narrows bridge

(single degree of freedom non-classical flutter) - see

http://books.google.co.uk/books?id=DnQOzYDJsm8C&dq=stall+flutt

er+tacoma&source=gbs_navlinks_s

Chris Williams