You are on page 1of 30

Reliability and Maintenance

(MANE 4015 )
Instructor: Dr. Sayyed Ali Hosseini
Winter 2015
Lecture #6

Parametric & Non-Parametric Data Analysis


Todays Lecture:
Rank statistics median rank, mean rank
Exponential plotting, 2-parameter Weibull plotting,
squares fit

MANE4015 Reliability and Maintenance

Lecture #6

least

Parametric & Non-Parametric Data Analysis


Given a set of failure data, can we tell if it fits some known probability
distribution? If not maybe we can construct a probability distribution
for the data.
Two approaches can be used to analyze data :
Non parametric
Parametric

MANE4015 Reliability and Maintenance

Lecture #6

Non-Parametric Data Analysis


Non-parametric data analysis normally refers to those techniques
that do not rely on data belonging to any particular distribution. It
assumes that the data are not drawn from a given probability
distribution.
No assumptions about underlying probability distribution.
Construct histograms, sample mean, sample variance, and rank
statistics.
Large # of data points use histogram to get failure distribution
function ( )
Small # of data points use rank statistics to get
definition

MANE4015 Reliability and Maintenance

Lecture #6

( ) or via

Parametric Data Analysis


Despite non-parametric data analysis, parametric data analysis
assumes that the data have come from a type of probability
distribution and makes inferences about the parameters of the
distribution.
By assuming that data comes from a probability distribution, this
method tries to fit data to that distribution and then estimate
distribution parameters (e.g. in exponential distribution)

MANE4015 Reliability and Maintenance

Lecture #6

Non-Parametric Data Analysis - Histogram


Data set contains

points (e.g.

failure times)

Find range of data set ( =

Choose interval width to divide data set into groups and plot data.

MANE4015 Reliability and Maintenance

==

1 + 3.3 log#$

Lecture #6

Example #1
70 components were tested and their time to fail was noted. The first
one failed in 21 hours and the last one failed in 84 hrs. Plot the failure
distribution function ( ). What is the probability that the component
will survive more than 55hrs?
39, 54, 21, 42, 66, 50, 56, 62, 59, 40, 41, 75, 63, 58
32, 43, 51, 60, 65, 48, 61, 27, 46, 60, 73, 36, 38, 54
60, 36, 35, 76, 54, 55, 45, 71, 54, 46, 47, 42, 52, 47
62, 55, 49, 39, 40, 69, 58, 52, 78, 56, 55, 62, 32, 57
45, 84, 36, 58, 64, 67, 62, 51, 36, 73, 37, 42, 53, 49

MANE4015 Reliability and Maintenance

Lecture #6

Example #1 Solution

, ** -

==

20 29
30 39
40 49
50 59
60 69
70 79
80 89

1 + 3.3 log

84 21
=
= 8.8
1 + 3.3 log#$ 70

/ - 0 12

MANE4015 Reliability and Maintenance

2
11
16
20
14
6
1

3/ = 4

Lecture #6

* 10

( ) = 4/

2/70 = .03
13/70 = .185
29/70 = .41
49/70 = .7
63/70 = .9
69/70 = .98
70/70 = 1

Example #1 Solution
: 3-;< *2
= 1 0 .7 = 0.3

> 55

* =:

> 55 = 1 :

MANE4015 Reliability and Maintenance

Lecture #6

55 = 1

55

Non-Parametric Data Analysis Rank Statistics


Small # of data points, use rank statistics to get Q(t).
Why do you have small # of data points?
Testing is destructive (part is expensive).
Not enough time to test.

MANE4015 Reliability and Maintenance

Lecture #6

10

Non-Parametric Data Analysis Rank Statistics


Steps:
Rank the failure times
number of components

in ascending order ( ), where

Approximate ( @ ) at each value


A
A

C ( @ ) =
B

C ( @ ) =

is the

by:

+1

0.3
+ 0.4

MANE4015 Reliability and Maintenance

Lecture #6

11

Example #2
The following are the times to fail (min) for 14 components, 97, 82,
72, 126, 113, 117, 103, 127, 127, 207, 154, 159, 199, 139. Plot the
failure distribution function Q(t) for the component. Use mean rank.

MANE4015 Reliability and Maintenance

Lecture #6

12

Example #2 Solution
B

1
2
3
4
5
6
7
8
9
10
11
12
13
14

. ; - 0

72
82
97
103
113
117
126
127
127
139
154
159
199
207

MANE4015 Reliability and Maintenance

( )

( ) = /( + 1)
1/15 = 0.0667
2/15 = 0.1333
3/15 = 0.2
0.266
0.333
0.4
0.466
0.533
0.6
0.667
0.733
0.8
0.866
0.933

Lecture #6

13

Example #2 Solution
From ( ) vs @ plot, you can determine for example the probability
that the component will last longer than 150 min.
1
0.9
0.8
0.7

Q(t)

0.6
0.5
0.4
0.3
0.2
0.1
0
72

82

97

103

113

117

126

127

127

139

154

159

199

207

MANE4015 Reliability and Maintenance

Lecture #6

14

Parametric Data Analysis Exponential Plotting


The objective is to fit failure distribution ( ) to some known
probability distribution, check if it fits, and then estimate parameters
from the plot.
For exponential distribution:
=1

DEF

DEF

ln

The above equation has a form of straight line / = I + J with yintercept J = 0.

MANE4015 Reliability and Maintenance

Lecture #6

15

Parametric Data Analysis Exponential Plotting

/= I+J =
ln

MANE4015 Reliability and Maintenance

K -< =

Lecture #6

16

Parametric Data Analysis Exponential Plotting


Steps
Compute ( ) using non-parametric methods.
Plot ln

#
#DL F

on Y-axis, and on X-axis.

If the result is a straight line passing through origin (y-intercept = 0),


then the failure data is exponentially distributed.
Estimate of

= * -< -0

If M = 1 then = 1/ M . Therefore, alternate way to find is to


lookup value of 1 on y-axis, and find corresponding value of M to
compute = 1/ M .
MANE4015 Reliability and Maintenance

Lecture #6

17

Parametric Data Analysis Exponential Plotting


/= I+J =
ln

1
K -< =

MANE4015 Reliability and Maintenance

Lecture #6

18

Example #3
The failure times for 10 components were recorded to be 11.2, 6.8,
5.2, 16.8, 17.8, 32.0, 23.4, 25.4, 19.6, 44.8. Determine whether this
components failure times can be model as exponential distribution.
And if it can, estimate the parameter .

MANE4015 Reliability and Maintenance

Lecture #6

19

Example #3 Solution
B

1
2
3
4
5
6
7
8
9
10

.;

5.2
6.8
11.2
16.8
17.8
19.6
23.4
25.4
32.0
44.8

MANE4015 Reliability and Maintenance

( ) = /( + 1)

1/11 = .0909
2/11 = 0.1818
3/11 = 0.2727
4/11 = 0.3636
5/11 = 0.4545
6/11 = 0.5454
7/11 = 0.6363
8/11 = 0.7272
9/11 = 0.8181
10/11 = .0909
Lecture #6

1/(1 ( ))
1.1
1.22
1.37
1.57
1.83
2.2
2.75
3.66
5.5
11

20

Example #3 Solution
Plot ln

#
#DL F

on Y-axis, and on X-axis. Slope of line = .


= 0.056 min.

From the plot you will find that


3
2.5
2
1.5
1
0.5
0
5.2

6.8

11.2

16.8

17.8

19.6

23.4

25.4

32

44.8

/
0.18 0.09
0.09
K -< =
K -< =
K -< =
= 0.056
I
6.8 5.2
1.6
If its a curve then its not exponential.
MANE4015 Reliability and Maintenance

Lecture #6

21

Least Squares Fitting


Consider exponential plotting. Rather than draw line by eyeball, we
want a more precise estimate of the line (y-intercept, slope). In this
case, Least squares fitting can be used.
Given data points (I@ , /@ ), find the line / = I + J that minimize
the mean square deviation of the vertical distance between the
points (I@ , /@ ) and the corresponding values on the line.
(IR , /R )

/
(I# , /# )

(IS , /S )

/ = I+J
/ /S

(IQ , /Q )

MANE4015 Reliability and Maintenance

I
Lecture #6

22

Least Squares Fitting


min K =

T /@ /
@V#

T /@ I@ J

@V#

The objective is to find , J to minimize K.


Solve

WK
= I/ I Q JI = 0
W
=

Answer is:
I/ =

I/ I . /Y
IQ

T I@ /@
@V#

I Q

J = /Y I

and
IQ =

WK
= /Y I J = 0
W

and

MANE4015 Reliability and Maintenance

T I@Q
@V#

Lecture #6

23

Parametric Data Analysis 2 Parameter Weibull Plotting


items tested and
function is:

fail by time , an estimate of the reliability


B

Weibull reliability distribution function:


B

MANE4015 Reliability and Maintenance

F [
D Z

Lecture #6

24

Parametric Data Analysis 2 Parameter Weibull Plotting


=1

DEF

F [
D Z

ln ln B

DEF

ln B( ) =

ln
_

ln ln B

= ` ln

= ` ln ` ln ^

The above equation has a form of straight line / = I + J. We can


plot ln ln B on Y-axis and ln [ ] on X-axis. If the result is a straight
line, then data fits a weibull distribution.

MANE4015 Reliability and Maintenance

Lecture #6

25

Parametric Data Analysis 2 Parameter Weibull Plotting

ln ln B( )

K -< = `

` ln ^
ln[ ]

MANE4015 Reliability and Maintenance

Lecture #6

26

Example #4
100 items are tested and the # of failures and times were observed.
Does the data fit a Weibull distribution?
32;2

# -0 0

; ( -2 *)

2 ( )

MANE4015 Reliability and Maintenance

17

12

Lecture #6

35

18

54

24

73

30

85

36

27

Example #4 Solution

B( )

I I * = ln ( )
/ I * = ln [ B( )]
1

= 6,
1

= 100,

=B 6 =

12

18

24

30

36

0.96

0.83

0.65

0.46

0.27

0.15

1.79

2.48

2.89

3.18

3.4

3.58

0.27

0.64

3.2 1.68 0.84 0.25

=4

100 4
= 0.96
100

ln ( 1) = ln (6) = 1.79
ln ln B

= ln ln B 6

= 3.2

MANE4015 Reliability and Maintenance

Lecture #6

28

Example #4 Solution
1
0.5
0
0

0.5

1.5

2.5

3.5

-0.5
-1
-1.5
-2
-2.5
-3
-3.5

MANE4015 Reliability and Maintenance

Lecture #6

29

Example #4 Solution
K -< =

/
1.68 (3.2))
1.52
K -< =
K -< =
= 2.203
I
2.48 1.79
0.69

As a result, ` 2.20

^ =?

c I *

3 < = ` ln ^ 7 = 2.20 ln ^ ln ^ = 3.18 ^ =

MANE4015 Reliability and Maintenance

Lecture #6

R.Qe

= 25.79

30

You might also like