Scribd
Upload
195 views

Reliability Distributions

Sildes presents reliability analysis using common distributions such as Weibull distribution [email protected]

Uploaded by

AhmedBassyouni
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
195 views

Reliability Distributions

Sildes presents reliability analysis using common distributions such as Weibull distribution [email protected]

Uploaded by

AhmedBassyouni
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 21

Reliability Distributions

Dr. Ahmed Bassyouni


[email protected]
[email protected]
315 307-8194
1

Reliability Distributions

Topics
I.

Weibull Distribution

II. Lognormal Distribution


III. Beta Distribution

IV. Probability Plots

I. Weibull Model
Weibull is a good general purpose distribution -- it may be used to
represent a variety of distributions.
Used mostly because of its ability to fit data rather than its
theoretical justification.
Very common tool in reliability prediction

CDF 1 - e-(x/b)a
b scale parameter
a shape parameter
shape parameter as it relates to reliability prediction.
a <1; decreasing failure rate with usage
a =1; constant failure rate EXPONENTIAL!
a >1; increasing failure rate
3 < a < 4 ; approaches symmetrical - normal (a=3.5)

Weibull Distribution Shapes


Turns out the Exponential Distribution is also a
special case of the Weibull.
Weibull - based on a shape parameter~ a; Scale
parameter ~ b

a=1
a = 3.5

prob density function, pdf

Weibull for Different Shape Parameters


a

= 0.5
a

= 3.5
a=.5
a=1
a =2
a =3.5

time

Weibull Properties
PDF

CDF

a a 1 ( x / b )a
x0
a x e
b
0
x 0

1 e

( x / b )a

x0
x0

E(X)

b * 1
a

V(X)

b 2 { 1

2
1 2

[
1

] }
a
a

Weibull Example
Let T represent time to failure (hours) of a
bearing in a mechanical shaft. T may be
modeled using Weibull with a= 0.5, b = 5000
hrs/failures.
Find E(T)
Determine the probability that a bearing will last
at least 6000 hours (reliability: R(t) = P(T > t)).

Mean-time-to-failure (MTTF)
In reliability, the expected value of a variable
following a Weibull (or any other failure
distribution, such as the exponential) is often
referred to as the mean time to failure (MTTF).
For example, if T is the time to failure for a
certain component and it follows a Weibull
distribution with a=1, what is the probability that
a component will last more than the MTTF?

Practical Uses of Weibull


Also used to model probabilities of building
excess (excess material, excess capacity).
Let X be the value that represents excess.
Example: suppose you have material thickness
requirement of 7 mm. From historical data
you fit X to a Weibull with parameters a= 2,
b = 4.
What is the likelihood of producing > 10 mm?
(3 in excess of the 7mm)

II. Lognormal Distribution


A nonnegative rv is lognormal if the rv Y =
ln(X) has a normal distribution so that
ln(X)~N(,)

PDF

CDF

[ln( x ) ]2 /( 2 2 )
e
x0

f x; , = 2 x
0
x 0

F x; , = P X x = Pln X ln x
ln( x)
ln( x)
= P( Z
) = (
)

Expected Value and Variance


of the Lognormal
E(X)

V(X)

/ 2
e
e

2 2

* e
1

Note: and are the mean and standard deviation of Y or ln(X),


not of X!

Lognormal - Key Characteristics


X may not be normal but ln(x) needs to be
normal.
Lognormal curves - positive skew
Normal f(ln(x);1,1)

0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

f(ln(x);1,1)

f(x;1,1)

Lognormal f(x;1,1)

10

15
x

20

25

0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
-3

-2

-1

1
ln (x)

Lognormal Example
Suppose a mechanical component has a
wear-out effect that follows a
lognormal distribution where =8.5; =
0.20
What is the probability that the unit
will last at least 3000 hours?

III. The Beta Distribution


Similar to uniform, if X follows a beta
distribution, is defined in a finite
interval [A,B]
PDF:
a 1
b 1
1
a b x A B x

A x B
f x;a , b , A, B = B A a b B A B A

otherwise

The case of A=0 and B=1, gives the standard beta distribution

Expected Value and Variance


of the Beta Distribution
E(X):

A (B A)

a
ab

V(X):

(B A) ab
2

a b a b 1

Common Applications of the Beta


PDF:
Project Management
A --> optimistic completion time for a task
B --> pessimistic completion time for task

Common Use of Standard Beta


A = 0 and B=1
X is proportion between 0 and 1.
Examples: % of time, % mixture.

IV. Probability Plots


Used to check a distribution assumption
In real life, it is hard to figure out a
distribution, typically we need sample data
to check and fit a distribution

If a distribution actually follows the


assumed distribution, the points in the
plot will follow a straight line (45)

How to?
By Hand:

Order n observations from smallest to largest


Plot the following pairs:
([100(i - 0.5)/n]th percentile, ith smallest observation in sample)
For i = 1 .. n

Example: suppose you have 10 observations.


1st sample value Vs. [100(1-0.5)/10]=5th percentile (1st obs vs.
5th percentile)
Plot pairs for each actual value in sample.
If all points fall on 45o line, then distribution under choice is
perfect fit.
The further points are scattered about line, the less likely the
data may be fit by the selected distribution (used to generate
the percentile values)

But mostly with software (Minitab!)

Problem
Data Vs. Normal

Data Vs. Exponential

Normal Probability Plot for w


ML Estimates - 95% CI

Exponential Probability Plot for w


ML Estimates - 95% CI

99

ML Estimates

95

Mean

218.21

StDev

126.794

99

Mean

AD*

Goodness of Fit
95

1.05

Percent

Percent

97

Goodness of Fit

70
60
50
40
30

AD*

90

80

20

70
60
50

10
5

30
10

1
0

-200

-100

100

200

Data

300

400

500

600

218.210

98

90
80

ML Estimates

500

1000

Data

1500

2.119

Analyzing Probability Plots


Look for straight lines.
One challenge - lots of distributions may fit
data. May want to search for best fit.
Use Minitab to compare distribution fits with
probability plots. (in Minitab, select lowest
AD score)
Use data from problem 90 and compare fits of
Normal, Exponential, Weibull, Lognormal.

Solutions
Slide 6-Weibull
E(X) = 5000 1 1/.5 = 5000 2! = 10k
P(X > 6000) = 1 P(X <= 6000)
W(6000; 5000,0.5)
Exp(-6000/5000)^.5 = 0.334

Lognormal
Z = (ln(3000) - 8.5 / 0.2) = 1 - (z=-2.55 ) =
.99461

You might also like