Random Vibration of Structures

Lecture 2. Random Process

1. The Definition of Random Process

2. Stationary Random Process

3. Weakly Stationary Random Process

4. Autocorrelation Functions

5. Cross-correlation Functions

6. Fourier Transformation

7. Power Spectral Density

8. Properties of Power Spectral Density

1
1. The Definition of Random Process

Random process in random vibration is a random

function of time t.

The random process is reduced to a random variable

if time t is given.

2
2. Stationary Random Process

The Stationary Random Process is a random process

of which the PDF and joint PDF at arbitrary time in-
stants is independent of time t, i.e.,

p[x(t1)] = p[x(t1 + τ )] (1)

p[x(t1), x(t2)] = p[x(t1 + τ ), x(t2 + τ )] (2)

p[x(t1), x(t2), x(t3)] = p[x(t1 + τ ), x(t2 + τ ) + x(t3 + τ )]

(3)
......

3
3. Weakly Stationary Random Process

Weakly Stationary Random Process is a random pro-

cess of which the first few moments are independent of
time t.

The first few moments are usually the first and second
moments, i.e.,

mean: m = E[X(t1)] = E[X(t2)] (4)

second moment: E[X 2(t1)] = E[X 2(t2)] (5)

and consequently

variance: σ 2 = E{[X(t1) − m]2}

= E[X 2(t1)] − m2
= E{[X(t2) − m]2}
= E[X 2(t2)] − m2 (6)

4
4. Autocorrelation Functions

• For stationary process X(t), the autocorrelation

function Rx(τ ) is defined as:

Rx(τ ) = E[X(t)X(t + τ )] (7)

which is an important quantity in random vibration

analysis of structures.

• The properties of autocorrelation function of a sta-

tionary process X(t):

1. Rx(τ ) = Rx(−τ ) Rx(τ ) is symmetrical.

Proof: Because Rx(−τ ) = E[X(t)X(t − τ )],

Set t − τ = t0. Then t = t0 + τ and

Rx(−τ ) = E[X(t0+τ )X(t0)] = E[X(t)X(t+τ )] = Rx(τ )

5
 2. Rx(0) = E(X 2) ≥ 0

3. |Rx(τ )| ≤ Rx(0)

Proof: According to Schwarz’s inequality,

Rx2 (τ ) = {E[X(t)X(t+τ )]}2 ≤ E[X 2(t)]E[X 2(t+τ )] = Rx2 (0)

(8)

Hence

|Rx(τ )| ≤ Rx(0) (9)

4. If X(t) = X(t + τ ), then

Rx(τ ) = E[X 2] (10)

5. If X(t) is independent of X(t + τ ), then

6
 Rx(τ ) = E[X(t)]E[X(t + τ )] = (E[X])2 (11)

6. limτ →∞ Rx(τ ) = (E[X])2 = m2, because X(t) and

X(t + τ ) become independent as τ → ∞.

If m = 0, then limτ →∞ Rx(τ ) = 0.

7
5. Cross-correlation Functions

• The cross-correlation function Rxy (τ ) of two sta-

tionary processes X(t) and Y (t) is defined as:

Rxy (τ ) = E[X(t)Y (t + τ )] (12)

which is another important quantity in random vibra-

tion analysis of structures.

• The properties of cross-correlation function of sta-

tionary processes X(t) and Y (t):

1. Rxy (τ ) = Ryx(−τ )

Proof: Rxy (τ ) = E[X(t)Y (t + τ )]

= E[Y (t0)X(t0 − τ )] = Ryx(−τ ).

where t0 = t + τ .

• The properties of cross-correlation function of sta-

tionary processes X(t) and Ẋ(t):

8
 dRx(τ )
1. Rxẋ(τ ) = dτ
Proof:

Rxẋ(τ ) = E[X(t)Ẋ(t + τ )]
dE[X(t)X(t + τ )]
=
dτ
dRx(τ )
=
dτ

dRx(τ )
2. Rẋx(τ ) = − dτ

Proof:

Rẋx(τ ) = E[Ẋ(t)X(t + τ )]
dE[X(t0 − τ )X(t0)]
= (t0 = t + τ )
dτ
dRx(−τ )
=
dτ

9
 dRx(−τ )
= −
d(−τ )
dRx(τ )
= −
dτ

3. Rxẋ(τ ) = −Rẋx(τ )

4. Rxẋ(−τ ) = −Rxẋ(τ ). Hence

E[X(t)Ẋ(t)] = Rxẋ(0) = 0.

5. Rxẋ(0) = Rẋx(0) = 0

• The relation of the autocorrelation functions of X(t)

and Ẋ(t):

d2Rx(τ )
Rẋ(τ ) = −
dτ 2
Proof:

Rẋ(τ ) = Rẋẋ(τ ) = E[Ẋ(t)Ẋ(t + τ )]

10
 dE[Ẋ(t)X(t + τ )]
=
dτ
dRẋx(τ ) d2Rx(τ )
= =−
dτ dτ 2

6. Fourier Transformation

• Fourier transformation of a periodic function

If f (t) is a periodic function of period T , is at least

R +T /2
piece-wise continuous and the integral −T /2 |f (t)|dt ex-
ists, then

n=+∞ 2π
Cn expinω0t; ω0 =
X
f (t) = (13)
n=−∞ T
1 R T /2 −inω0 t
where Cn = T −T /2 f (t) exp dt.

Proof: Multiplying both sides of above equation with

exp−imω0t and integrating over a period, we obtain

11
Z
+T /2 −imω0 t +∞
X T /2Z
i(n−m)ω0 t
−T /2 f (t) exp dt = Cn exp dt
n=−∞ −T /2
(14)

Because

Z
T /2 i(n−m)ω0 t


 T ; (m = n)
−T /2 exp dt =  (15)
0; (m =
6 n)
hence
1 Z
T /2 −inω0 t
Cn = −T /2 f (t) exp dt (16)
T

• Fourier transformation of non-periodic function:

Non-periodic function can be considered as a special

periodic function with the period approaches infinity.

If f (t) is a non-periodic function, at least piece-wise

R +∞
continuous and the integral −∞ |f (t)|dt exists, then

Z
+∞ −iωt
F (ω) = −∞ f (t) exp dt (17)

12
and
1 Z +∞ iωt
f (t) = F (ω) exp dω (18)
2π −∞

Proof: Because

n=+∞ 2π
Cn expinω0t; ω0 =
X
f (t) = (19)
n=−∞ T

and
1 Z
T /2 −inω0 t
Cn = −T /2 f (t) exp dt (20)
T

Denoting ω0 or 2π/T by dω, nω0 by ω; as T ap-

proaches infinity, yields

Z
T /2 −iωt
lim T Cn = lim f (t) exp dt
T →∞ T →∞ −T /2
Z
+∞ −iωt
= −∞ f (t) exp dt = F (ω) (21)

13
and
n=+∞
Cn expinω0t
X
f (t) =
n=−∞
+∞
X 2π 1
iωt
= lim (CnT ) exp ( )( )
T →∞ −∞ T 2π
1 Z +∞ iωt
= −∞ F (ω) exp dω (22)
2π

• If limt→±∞ X(t) = 0, the Fourier transformation of

Ẋ(t) is:

Z
+∞
Ẋ(ω) = −∞ Ẋ(t) exp−iωt dt
Z
−iωt +∞ +∞
= X(t) exp |−∞ + iω −∞ X(t) exp−iωt dt
Z
+∞
= iω −∞ X(t) exp−iωt dt
= iωX(ω) (23)

• If limt→±∞ X(t) = 0 and limt→±∞ Ẋ(t) = 0, the

Fourier transformation of Ẍ(t) is:

Ẍ(ω) = −ω 2X(ω) (24)

14
15
7. Power Spectral Density

• Power spectral density is defined to be the Fourier

1
transformation of 2π R(τ ) of a stochastic process X(t),
i.e.,

1 Z +∞ −iωτ
Sx(ω) = −∞ R x (τ ) exp dτ (25)
2π

Hence,
1 1 Z +∞ iωτ
Rx(τ ) = −∞ Sx (ω) exp dω (26)
2π 2π

or Z
+∞ iωτ
Rx(τ ) = −∞ Sx (ω) exp dω (27)

• The cross power spectral density is defined to be

1
the Fourier transformation of 2π Rxy (τ ) of two station-
ary stochastic processes X(t), Y (t), i.e.,

16
 1 Z +∞ −iωτ
Sxy (ω) = R xy (τ ) exp dτ (28)
2π −∞

and Z
+∞ iωτ
Rxy (τ ) = −∞ Sxy (ω) exp dω (29)

17
8. Properties of Power Spectral Density

• Sx(ω) is real and even function of ω because Rx(τ )

is real and even.

• Sx(ω) is positive.

Proof: Because
Z
2 T
ε =| 0 X(t) exp−iωt dt|2 ≥ 0 (30)

and
Z Z Z
T −iωt 2 T −iωt T
| 0 X(t) exp dt| = 0 X(t) exp dt 0 X(s) exp+iωs ds
(31)

hence
Z Z
2 T T
E(ε ) = 0 dt 0 exp−iω(t−s) R(t − s)ds

(t − s = τ )
Z
T ZT −iωτ
Z
0 Z T +τ
= 0 [τ exp R(τ )dt]dτ + −T [ 0 exp−iωτ R(τ )dt]dτ
Z Z
T −iωτ 0
= 0 (T − τ ) exp R(τ )dτ + −T (T + τ ) exp−iωτ R(τ )dτ

18
 Z
+T
= −T (T − |τ |)R(τ ) exp−iωτ dτ (32)

hence

E(ε2) 1 Z +T (T − |τ |) −iωτ
lim = lim R(τ ) exp dτ
T →∞ 2πT T →∞ 2π −T T
1 Z +∞ −iωτ
= −∞ R(τ ) exp dτ = S(ω) ≥ 0 (33)
2π

• Sx(ω) is defined in (−∞, +∞). The negative fre-

quency is only for the convenience of mathematical de-
scription.

• E(X 2) = Rx(0) = −∞ +∞ R
S(ω)dω, which means that
S(ω) is the ensemble mean square of X(t) per unit fre-
quency.

• Relation between Rẋ(τ ) and Sx(ω).

Z
+∞ 2 iωτ
Rẋ(τ ) = −∞ ω Sx (ω) exp dω (34)

19
Proof:

d2Rx(τ )
Rẋ(τ ) = −
dτ 2
d2 Z +∞
= − 2 ( −∞ Sx(ω) expiωτ dω)
dτ
Z
+∞ 2 iωτ
= −∞ ω Sx (ω) exp dω (35)

• Relation between Rxẋ(τ ) and Sx(ω).

Z
+∞ iωτ
Rxẋ(τ ) = −∞ iωS x (ω) exp dω (36)

20
Assignments: Problems 2.1, 2.2

21