You are on page 1of 31

ENGR 4015U Reliability and

Maintenance

Reliability and Maintenance


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

Reliability and Maintenance


Todays Lecture:
3 parameter Weilbull distribution
Weilbull plot ( =0), Nonlinear plots ( > 0),
Estimating
Jardine Appendix2
WeibullSoft demo software
3

Weibull Distribution
Introduction
2 parameter Weibull distribution:

= shape parameter, = scale parameter


4

Weibull Distribution
Introduction
3 parameter Weibull distribution (FigureA2.1):
= shape parameter
= scale parameter (characteristic life)
= location parameter (failure free period)

Weibull Distribution

Weibull Distribution
Similar to the 2-parameter Weibull:
If = 1, the Weibull reduces to the
exponential distribution with constant hazard
rate function (t) = 1/ and MTTF =
In general, < 1 represents a decreasing
hazard rate, > 1 represents increasing hazard
rate. See Figure A2.2
7

Weibull Distribution

Weibull Distribution
3-parameter Weilbull Plot
Can use Weibull paper see figure A2.5 and
handout.
If you get a straight line, then = 0.
Run WeibullSoft tutorial
9

Weibull
Distribution

10

Weibull Distribution
Note:
If you know in advance = 0, then 3 parameter
Weibull reduces to 2 parameter Weibull, we use
method cover in last lecture:

Has form of straight line Y = a - bx


Can plot ln[-lnR(t)] on Y-axis and ln[t] on X-axis.
11

Example 1
100 lamps were tested and the time for the failures
(hrs) were recorded. The time is divided into
intervals and the # of failures in the interval (and
corresponding % of failures) is calculated.
a) Plot the lamp failure data and estimate the
parameters.
b) What is the expected (mean) life of the lamp or
equivalently the MTTF?
c) What is the variance of the life of the lamp?
12

Example 1
Interval

F = # of Fails

%F = % of fails = F/N*100

Q(t) =%F

0-4
4-8
8-12
12-16
16-20
20-24
24-28
28-32
32-36
36-40
40-44
44-48
48-52
52-56
56-60
60-64
64-68
68-72
72-76
76-80

5
9
6
5
7
6
8
2
6
6
4
2
12
8

5
9
6
5
7
6
8
2
6
6
4
2
12
8

5
14
20
25
32
38
46
48
54
60
64
66
78
86

13

Example 1 Solution
(a) Number of components N = 100. Large # of data points, so we
can use histogram approach for finding failure distribution
function Q(t).
Plot data on Weibull paper, Q(t) on Y-axis, and t on X-axis.
Note:
For time t, use upper value of time interval. Hence, first point to
plot is
(t=4, Q(t) = 5)
Do not plot intervals where no fails occur.
From Weibull plot Fig A2.6 and handout, = 1.2, = 43, data fits
straight line so = 0.
14

Example 1
Solution

15

Example 1 Solution
(b) To obtain expected life, draw line from
estimator point on top left corner
perpendicular to plot. Read P value = 60%.
Therefore MTTF = = 40hrs from plot.

16

Example 1 Solution
c) The variance

or can use Nomograph (see handout):


From = 1.2, Factor for estimating from = 0.8
from graph.
= .8(43hrs)=34.4hrs
so
2 = 1176.5hrs
17

Nonlinear plots (nonzero )


What happens when you get a plot that is not
a straight line? This means the parameter
0.
How to handle this case?

18

Example 2
Suppose we put 20 motors on test for 2500hrs.
During that time interval, we had 12 failures at
the following times: 550, 1180, 1750, 720,
1330, 1920, 880, 1490, 2150, 1020, 1610, 2325.
Fit these failure times to a 3 parameter weibull
and estimate the values of the parameters.

19

Example 2 Solution
This is an example of censored data. i.e.,
didnt wait for all of the components to fail.
Small # of data points, so use Mean or Median
ranks to obtain failure distribution function
Q(t). Assume we use Median ranks.

20

Example 2 Solution
N= 20 (note we use 20 items and not 12 items)
For t = 550 hrs, Q(t) = (1 0.3)/(20+0.4) = .0343
or 3.43%
Note that the %Q(ti) values come from table in
Appendix 8 of the book and it has some rounding,
therefore it is a little different than the values eg.
3.43% obtained from rank i=1.
21

Example 2 Solution
Rank i
1
2
3
4
5
6
7
8
9
10
11
12
13

20

Time to fail (ti)


550hrs
720
880
1020
1180
1330
1490
1610
1750
1920
2150
2325

% Q(ti)
3.406
8.251
13.147
18.055
22.967
27.88
32.975
37.7
42.6
47.5
52.45
57.3
Censored data
22

Example 2 Solution
Plot % Q(ti) on Y-axis and time ti on X-axis see
Figure A2.7
Since we have a curve, this indicates > 0.
If we take (ti - 375) and plot again, we get a
straight line Figure A2.8 Therefore, = 375.
From new plot, = 1.32, = 2120hrs
note that true = 2120 + 375 =2495hrs and true
MTTF =mean from plot + 375 = ?
23

Example 2
Solution
Plot % Q(ti) on
Y-axis and
time ti on Xaxis see
Figure A2.7

24

Example 2 Solution
Since we have a curve, this indicates > 0.
If we take (ti - 375) and plot again, we get a
straight line Figure A2.8
Therefore, = 375.
From new plot, = 1.32, = 2120hrs
Note that true = 2120 + 375 =2495hrs and true
MTTF =mean from plot + 375 = ?
25

Example 2
Solution

26

Estimating the parameter


Rather than trial and error to obtain the parameter , can
use following approach:
Select 2 endpoints of the Weilbull plot that cover the entire
set of failure data. Let a and c be the projections of these
endpoints on the Y-axis (figure 2.10)
Bisect the distance between a and c. Let b be the midpoint.
Let the projections of a, b, and c on the X-axis be t1, t2, t3
Then
Run demo
27

Estimating the parameter

28

Example 3
Estimate for motor failure data in example 2.

29

Example 3 Solution
From Figure A2.11 - t1 = 500 hrs, t2 = 933 hours,
t3 = 2500hrs
So

30

Example 3
Solution

31

You might also like