Вы находитесь на странице: 1из 33

Weibull Analysis

for
Life Data

Jang JuSu
Historical Background
1. Waloddi Weibull (1887-1979) invented the Weibull distribution in
1937.

2. His 1951 paper represents the culmination of his work in reliability


analysis.

3. The U.S.Air Force recognized the merit of Weibulls methods and


funded his research to 1975.

4. Leonard Johnson at General Motors, improved Weibulls methods.


(Weibull used mean rank values for plotting but Johnson suggested the use of
median rank values)

5. E.J.Gumbel proved that the Weibull distribution and the smallest


extreme value distributions(Type III) are same.

6. The engineers at Pratt & Whitney found that the Weibull method
worked well with extremely small samples, even 2 or 3 failures.
MOASOFT Training Course for Reliability 2
Advantages of Weibull Analysis
1. The primary advantage of Weibull analysis is the ability to provide
reasonably accurate failure analysis and failure forecast with
extremely small samples.

2. Another advantage is a simple and useful graphical methods.

MOASOFT Training Course for Reliability 3


Application of Weibull Analysis

Life Data or Accelerated Life Data

Weibull Analysis

Failure Analysis and Forecasting

Schedule Maintenance
Application to

Project Management for a System or Plant

Warranty and Cost Analysis

MOASOFT Training Course for Reliability 4


Weibull Distribution
1. Probability Density Function
1 t
t


f (t ) e
,t 0

2. Cumulative Distribution Function


t



F (t ) 1 e
, t 0

: Shape parameter

: Scale parameter

: Location parameter

MOASOFT Training Course for Reliability 5


Weibull Distribution
Shape parameter variations with 10 .
0.4 0.4

0.3 0.3
0 .5 1
0.2 0.2

0.1 0.1

5 10 15 20 25 30 5 10 15 20 25 30

0.4 0.4

0.3 0.3

2 2.5
0.2 0.2

0.1 0.1

5 10 15 20 25 30 5 10 15 20 25 30

MOASOFT Training Course for Reliability 6


Weibull Distribution
Scale parameter variations with 2 .

0.8 0.08

0.6 1 0.06
10

0.4 0.04

0.2 0.02

0.5 1 1.5 2 2.5 3 5 10 15 20 25 30

0.008 0.0008

0.006 100 0.0006 1000


0.004 0.0004

0.002 0.0002

50 100 150 200 250 300 500 1000 1500 2000 2500 3000

MOASOFT Training Course for Reliability 7


Weibull Distribution
Location parameter shift the curve along x-axis.

CDF : shape parameter variation

MOASOFT Training Course for Reliability 8


Weibull Distribution

1
1

MOASOFT Training Course for Reliability 9


Weibull Distribution

t




F (t ) 1 e

From the above equation,


If we set the value of time to t then F ( ) 0.632 .
So we can guess that is defined as the age at which
63.2% of the units will fail.





F ( ) 1 e 1 e 1 0.632 .

MOASOFT Training Course for Reliability 10


Interpretation

Assume that we have 2-types data, failure and suspended,

1. For scale parameter ,

In general the more we have suspended data the shape


parameter , , hardly change but the scale parameter , ,
will be increased.

2. For shape parameter ,

1 : Implies Infant mortality and we can suspect

MOASOFT Training Course for Reliability 11


Interpretation

Inadequate burn-in test or screening,

Production problems, misassembled, quality control,

Overhaul problems,

Solid state electronic failure.

1 : Implies random failures

Maintenance errors, human errors,

Failures due to nature,

MOASOFT Training Course for Reliability 12


Interpretation

Failures due to nature,

Mixtures of data from 3 or more failure modes or different s.

1 4 : Implies early wear out

Low cycle fatigue,

Corrosion or erosion.

4 : Implies ageing effects

MOASOFT Training Course for Reliability 13


Types of Data

0 Time 0 Time 0 Time


1 Complete 1 Type II 1 Type I
Data
2 2 2
3 3 3
Unit

Unit

Unit
4 4 4
5 5 5
6 6 6

0 Time 0 Time 0 Time


1 Multiply Failure 1 Random 1 Interval
Censored Censored (Grouped)
2 2 2 Data
3 3 3
Unit

Unit

Unit
4 4 4
5 5 5
6 6 6

MOASOFT Training Course for Reliability 14


Median Rank Regression

Failure time Status Median Rank Y X

1500 F 0.0946 7.3132 -2.3088


1750 S
2250 F 0.2523 7.7187 -1.2353
4000 F 0.4099 8.2940 -0.6397
4300 F 0.5676 8.3664 -0.1763
5000 S
7000 F 0.8041 8.8537 0.4887

MOASOFT Training Course for Reliability 15


Median Rank Regression

1. Adjusted Rank formula

reverse rank Rn 1 ( N 1)
Rn
reverse rank 1
R0 0 , n 1 , 2 , 3 , . . .

2. Benards Approximation formula

( Rn 0.3)
BRn
N 0.4
n 1, 2 , 3,...
MOASOFT Training Course for Reliability 16
Median Rank Regression


t

t


F (t ) 1 e 1 F (t ) e


t
t
1 /(1 F (t )) e ln[ 1 /(1 F (t ))]

ln ln[ 1 /(1 F (t ))] ln( t ) ln( )

y Bx A
MOASOFT Training Course for Reliability 17
Median Rank Regression

y ln( failure time)

x ln ln( 1 /(1 Median Rank of Y ))

1

e A
B

MOASOFT Training Course for Reliability 18


Median Rank Regression

Y
e

XC
CX t (Y XC ) 0 y1 x1 ,1

y2 x2 ,1 B
X Y X XC
t t
Y X C
.. .. ,1 A

y x ,1
C ( X t X ) 1 X tY n n

MOASOFT Training Course for Reliability 19


Median Rank Regression

MOASOFT Training Course for Reliability 20


Maximum Likelihood Estimation

t
1
f (t ) t e , t 0

k
L f (ti ) f (t1 ) f (t 2 )... f (t k )
i 1

ti
k

ti e
1

i 1

MOASOFT Training Course for Reliability 21


Maximum Likelihood Estimation

1
k
1 k
ln L k ln ln ti ti
i 1 i 1
Maximizing Logarithm Likelihood Function

(ln L) k 1
ln ti ti ln ti 0

(ln L) k
2 0
i
t


MOASOFT Training Course for Reliability 22
Maximum Likelihood Estimation

i ln ti
t
1 r 1
G( ) i 1
ln ti 0
k

i r i 1
t
i 1

1
k

t i
i 1
r

MOASOFT Training Course for Reliability 23


Maximum Likelihood Estimation

G( n )
n1 n
G`( n )

We so as to use Newton method have to find the derivative of G( )

MOASOFT Training Course for Reliability 24


Maximum Likelihood Estimation

k k

t i

(ln ti ) 2
(t i

ln t ) i
2
1
G`( ) i 1
i 1
2
k k

t
i 1
i

( t )
i 1
i
2

MOASOFT Training Course for Reliability 25


Maximum Likelihood Estimation

(ln L)
2 1 t t 2 t t 2
i (ln i ) i (ln i )
2

n
r

(ln L)
2 t t


i
( 1) i
( 1)
2
2 2 2

n
r


(ln L)
2
1 ti 1 ti
ln 1
n
t 1 t
i ln i 1


r

MOASOFT Training Course for Reliability 26
Maximum Likelihood Estimation


Z / 2 var( )
exp

Z / 2 var( )
exp



(ln t ln )

Z / 2 var( )
exp

Z / 2 var( )
exp

MOASOFT Training Course for Reliability 27


Maximum Likelihood Estimation

MOASOFT Training Course for Reliability 28


Conclusion

In case of small sample -> Monte Carlo Simulation!


this will make an analysis enhanced with small sample.

Topics MLE Rank Regression

Graphical None OK
Small sample Bad( more bias) more accurate
Forecast Bad more accurate
Risk Analysis ? ?
Confidence Rigorous None
Convergence ? Always
Mixed mode Only one Available

But large sample case, MLE preferred!

MOASOFT Training Course for Reliability 29


Arrhenius Model
Arrhenius Law

E
A e kT

where
E : activation energy : electron-volts.
k : Boltzmanns constant : 8.6171x10-5 per C.
T : Kelvin temperature
A : a constant that depends on product geometry,
specimen size, and other factors.

MOASOFT Training Course for Reliability 30


Arrhenius Model
Linearized relationship

log 0 ( 1 / T )

where

1 log( e) ( E / k ) 0.4343 E / k

E 2.303 k 1

MOASOFT Training Course for Reliability 31


Arrhenius Model
Arrhenius acceleration factor K

1 1
K / ' exp{( E / k )[ ]}
T T'

Example : E = 0.65 eV, T = 453 K(180 C), T = 533 K(260 C).

K = exp{(0.65/ 8.6171x10-5 )[1/453 1/533]} = 12

MOASOFT Training Course for Reliability 32


Arrhenius Model

MOASOFT Training Course for Reliability 33

Вам также может понравиться