4 Frequency Based Fatigue

7/30/2019 4 Frequency Based Fatigue

http://slidepdf.com/reader/full/4-frequency-based-fatigue 1/57

Professor Darrell F. Socie

Department of Mechanical Science and Engineering

University of Illinois at Urbana-Champaign

© 2001-2010 Darrell Socie, All Rights Reserved

Frequency Based Fatigue



Frequency Based Fatigue © 2001-2011 Darrell Socie, All Rights Reserved 1 of 57

Deterministic versus Random

Deterministic – from past measurements the future positionof a satellite can be predicted with reasonable accuracy

Random – from past measurements the future position of a car can only be described in terms of probability andstatistical averages




Time Domain

Time (Secs)

S t r a i n ( u s t r

a i n )

Bracket.sif-Strain_c56

-750

750




Frequency Domain

Frequency (Hz)

0 20 40 60 80 100 120 140

102

S t r a i n ( u s t r a

i n )

ap_000.sif-Strain_b43

101

100

10-1




Histogram Domain

1500

-750

750

0

750

C o u n t s

rf_000.sif-Strain_b43




Outline

Statistics of Time Histories

Time Domain Nomenclature

Frequency Analysis

PSD Based Fatigue Analysis


http://slidepdf.com/reader/full/4-frequency-based-fatigue 7/57Frequency Based Fatigue © 2001-2011 Darrell Socie, All Rights Reserved 6 of 57

Statistics of Time Histories

Mean or Expected Value

Variance / Standard Deviation

Root Mean Square

Kurtosis

Skewness

Crest Factor

Irregularity Factor



Mean or Expected Value

Central tendency of the data

( )N

xXExMean

N

1ii

x∑====µ=



Variance / Standard Deviation

Dispersion of the data

( )N

)xx(

XVar

N

1i

2

i∑=

−

=

)X(Varx

=σ

Standard deviation



Root Mean Square

N

x

RMS

N

1i

2i∑

==

The rms is equal to the standard deviationwhen the mean is 0



Skewness

Skewness is a measure of the asymmetry of the dataaround the sample mean. If skewness is negative, thedata are spread out more to the left of the mean thanto the right. If skewness is positive, the data are spread

out more to the right. The skewness of the normaldistribution (or any perfectly symmetric distribution) is zero.

( )3

N

1i

3

i

N

)xx(

XSkewnessσ

−

= ∑=



Kurtosis

Kurtosis is a measure of how outlier-prone a distribution is. The kurtosis of the normal distribution is 3. Distributions thatare more outlier-prone than the normal distribution have

kurtosis greater than 3; distributions that are lessoutlier-prone have kurtosis less than 3.

( ) 4

N

1i

4

i

N

)xx(

XKurtosis σ

−

=

∑=



Crest Factor

The crest factor is the ration of the peak (maximum) valueto the root-mean-square (RMS) value. A sine wave has acrest factor of 1.414.



Irregularity Factor

Positive zero crossing

Peak 7

4

)P(E

)0(EIF ==

+

IF → 1 is narrow band signal

IF→

0 is wide band signal



Time Domain Nomenclature

Random

Stochastic

Stationary

Non-stationary

Gaussian

Narrow-band

Wide-band



Random

The instantaneous value can not be predicted at anyfuture time.



Stochastic

Stochastic processes provide suitable models forphysical systems where the phenomena is governedby probabilities.



Stationary

The properties computed over short time intervals, t +∆t,do not significantly vary from each other

∆t



Non-stationary



Gaussian

Normally distributed around the mean



Narrow-band



Wide-band




Frequency Domain Analysis

Fourier

FFT

Inverse FFT

Autospectral density Transfer Function




Fourier 1768 - 1830

Fourier studied themathematical theory of heatconduction.He established the partialdifferential equationgoverning heat diffusion andsolved it by using infiniteseries of trigonometric

functions.




Fourier Series

∑∞

=

ω+ω+=1k

okoko )tkcos(b)tksin(aa)t(X

∑∞

=

θ+ω+=1k

koko )tkcos(cc)t(X

∑∞

∞−=

ω=k

t jk

Soe]k[X)t(X

ak, bk, ck, XS[k] are Fourier coefficients

frequency, magnitude, and phase are all describedby the coefficients




Fourier Transform

Time

M a

g n i t u d e

F F T M a g n i t u d e

Frequency

The area under each spike represents the magnitudeof the sine wave at that frequency.

∆f

The magnitude of the FFT depends on the frequency window ∆f.




Fast Fourier Transform - FFT

Frequency (Hz)

0 20 40 60 80 100 120 140

16

14

12

108

6

4

2

0

F F T

S t r a

i n ( u s t r a i n )

∑=

θ+ω+=2/N

1kkoko )tkcos(cc)t(X




Fast Fourier Transform - FFT

Frequency (Hz)

0 20 40 60 80 100 120 140

150

100

50

0

-50

-100

-150

P h a s e ( D e g r e e s )

∑=

θ+ω+=2/N





Inverse FFT

∑=

θ+ω+=2/N


ck, θk, and kωo are all known

Time (Secs)

S t r a i n ( u s t r a i n )

-750

750




Autospectral Density

Function Units

Time History EU X(n)

Linear Spectrum EU S(n) = DFT(X(n))

AutoPower EU 2̂ AP(n) = S(n) · S(n)

PSD (EU 2̂)/Hz PSD(n) = AP(n) / ( Wf · ∆ f )

ESD (EU 2̂*sec)/Hz ESD(n) = AP(n) · T / ( Wf · ∆ f )




Calculating Autospectral Density

T 2T 3T nD TND

( ) ( )∑=∆

=Dn

1i

2

k

DD

k f XtNn

1f S

ND Block size

Magnitude only no phase information




Power Spectral Density - PSD

Frequency (Hz)

S t r a i n ( u s t r a i n ^ 2 ) / H z

Log Magnitude-Power-Strain_c56

100 101 102

105

104

103

102

101

100

Average power associated with a 1 Hz frequency windowcentered at each frequency, f .Phase information is lost.




Linear Spectrum

Frequency (Hz)

102


Log Magnitude-Linear-Strain_c56

100

101

100 101 102

Sometimes called Amplitude Spectral Density




Comparison

Frequency (Hz)

S t r a i n ( u s t r a i n ^ 2 )

100 101 102

105

104

103

102

101

100

10-1

PSD

Linear Spectrum




Frequency Domain Limitations

Time (Secs)

800

600

400

200

0

-200

-400


EASE1.EDT-Strain_c56

Non-stationary signals




Linear Spectrum

Frequency (Hz)

0 20 40 60 80 100 120 140

103

102

101

100

10-1

S t r a i n ( u s

t r a i n )




Transfer Function

m

input

output

Frequency (Hz)

100 101 102

104

103

10

2

input

output





Stationary LoadingWind

Sea State

Vibration




Assumptions

RandomGaussianStationary




Fatigue Analysis

Structuralmodel

Stresstime

history

RainflowMinerlinear

damageDurability

Loadtime

history

AccelerationPSD

Transferfunction

StressPSD

?

Time Domain

Frequency Domain




Dynamics Model

y(t)

m

k cx(t)

F(t)

2

2

2

2

dt

xdm)t(Fzk

dt

dzc

dt

zdm −=++

z(t) = y(t) – x(t)

2

22

nn2

2

dt

xdz

dt

dz2

dt

zd−=ω+ζω+

m

F(t)

y(t)

)t(Fdt

ydm

2

2

=




Nondimensional Amplitude

2

n

22

n

o

21

1

F

Xk

ωω

ζ+

ωω

−

=

0

1

2

3

4

5

0 1 2 3 4

oF

Xk

nω

ω

ζ = 1

ζ = 0.5

ζ = 0.25

ζ = 0.15




PSD Moments

frequency

M a g

n i t u d e 2 / H z

Gk( f )

f k

δf

∑=

δ=N

1kkk

n

kn f )f (Gf m




Expected Values

0

2

m

m)0(E =

2

4m

m)P(E =

40

2

2

mm

m

)P(E

)0(EIF ==

zero crossings

peaks

Irregularity factor




Probability Density Function

p( ∆Si )

∆Si

δS

P r o b a

b i l i t y d e n s i t y

Stress range

The probability P( ∆Si ) of a stress range occurring between

S)S(p)S(Pis

2

SSand

2

SS iiii δ∆=∆

δ+∆

δ−∆




Fatigue Damage

Cycles at level i ni = p( ∆Si ) δS N T

Total cycles N T = E( P ) T

Total time

b

1

'f

iif

S2

S)S(N

∆=∆Fatigue life

Fatigue Damage

( )∑∑ == ∆∆δ=∆=

N

1i b

1

i

ib

1

'f

N

1i if

i

S)S(p)S2(S T)P(E

)S(NnD

Fatigue damage is determined by p( ∆Si )




Narrow Band Solution

∆−

∆=∆

0

2

0 m8

Sexp

m4

S)S(p

Rayleigh distribution

This wide band signal hasthe same peak distribution

as this narrow band signal

IF → 1 is narrow band signal

IF → 0 is wide band signal




Dirlik Solution

0

2

32

2

2

21

m2

2

ZexpZD

R

Zexp

R

D

Q

Zexp

Q

D

)S(p

−+

−+

−

=∆

0m2

SZ

∆=

40

2

mm

mIF=

4

2

0

1m

m

m

m

mX =

2

2

m1

IF1

IFX2D

+

−=

R1

DDIF1D

2

112 −

+−−=

2

11

2

1m

DDIF1

DXIFR

+−−

−−=

( )

1

21

D4

RDDIF5Q

−−=

213 DD1D −−=

p( ∆S ) =f ( m0, m1, m2, m4 )




Loading History

Frequency (Hz)

0 20 40 60 80 100 120 140

102

S t r a i n

( u s t r a i n )

101

100

10-1

Frequency (Hz)

0 20 40 60 80 100 120 140

102

S t r a i n

( u s t r a i n )

101

100

10-1

Time (Secs)

S t r a i n

( u s t r a i n )

-750

750

Time (Secs)

S t r a i n

( u s t r a i n )

-750

750




Rainflow Ranges

0.01

0.1

1

10

100

1000

10000

0 250 500 750 1000

PSD

Time History

Stress Range

N u m b e r o f C

y c l e s




Relative Fatigue Estimates

SN Slope TH

PSD

N

N

10 371

5 6.6

3 1.8

TH

PSD

N

N

109

1.9

0.5

Dirlik Rayleigh




Fatigue Data

1

10

100

A m p l i t u

d e

100

Cycles101 102 103 104 105 106 107

1000

n = 10

n = 5

n = 3nSDamage ∆∝




Loading History

Time (Secs)

50 100 150 200 250 300

750

500

250

0

-250

-500

-750

S t r a i n G a g e ( u s t r a i n )

Bracket.sif-Strain_b43

0

1500

-750

750

0

750

C o u n t s

rf_000.sif-Strain_b43




Slope = 3

0

1500

-750

750

3.15

%

d a m a g e

Damage




Slope = 5

0

1500

-750

750

5.14

%

d a m a g

e

Damage




Slope = 10

0

1500

-750

750

20.78

%

d a m a g e

Damage




Documents

4 Frequency Based Fatigue