Solution Manual

Problem 3
Using the data and the reliability equation of the Kaplan-Meier estimator, the following table can be
constructed:

State (ni-
Number of Number of Available
End fI(ni-ri)/ni
Failures, ri Suspensions, si Units, ni ri)/ni
Time
9 3 1, 20 0.8s0 0,850
t\ L 0 16 0,938 0,797
12 0 L 15 1,000 0,197
13 t 1 L4 o,929 0,740
15 0 1 12 1,000 0,740
L7 1 0 1.1 0,909 o,673
27 1 ,^, 0 10 0,900 0,605
22 0 9 1,000 0,505
i
,t
1.

8 1,000 0,605
24 0 1

26 o- 1 7 1,000 0,605
28 1 0 6 0,833 0,505
30 1 0 5 0,800 0,404
32 0 1. 4 l-,000 0,404
35 0 1 3 1,000 0,404
39 0 1 2 1,000 0,404
4! 0 1 1. 1,000 o,404

As can be determined from the preceding table, the reliability estimates for the failure times are:

Failure Reliability
Time Estimate
9 85,00%
1.1 79,70%
13 74,00%
17 67,30%
21 60,so%
28 5050%
30 40,40%

Page I 9
SOLUTION MANUAL

In this example, we have complete data only. The partial derivative of the log-likelihood
function, l. is given by:

, ll
#-! + rr; .)l
#,\ + Ai
tf
I I,i li

*:
Recall that when using the MLE method for the exponential distribution, the value of is
*i : i
equal to that of the first failure time. The first failure occurred at 5 hours, thus
-'t
hours . Substituting the values for ?'and we get:

1J
*.' - .',, i-:fi{}
,l
or:

I - tt i|li f.qilrrrm.i hr,ur

Page | 2
SOLUTION MANUAT

Construct the following table, as shown next.

Table Least Squares AnalYsis

-)
N ti F(ti) yr ti2 yr- ti.yi
1 5 0,0483 -0,0495 25 0,00245 -0,24753

2 10 o,1.L7O -0,7244 100 o,oL548 -t,2443

3 L5 0,1855 -o,2064 225 0,04260 -3,0961.4

4 20 o,2561 -o,2958 400 0,08753 -5,91697

5 25 0,3258 -o,3942 625 o,15542 -9,85571
6 30 0,3954 -o,5032 900 0,25320 -L5,0956

7 35 o,465L -o,6257 1725 o,39L47 -2t,8986

8 40 0,5349 -o,7655 1600 0,s8599 -30,620L
9 50 0,6046 -0,9279 2500 0,96092 -46,3929

10 60 a,6742 -L,12L5 3600 L,25770 -67,2883

tl 70 o,7439 -L,3622 4900 1,85s55 -95,3531

L2 80 0,8135 -L,6793 6400 2,82013 -L34,346

13 90 0,8830 -2,1456 8100 4,60352 -L93,702
L4 100 0,95L7 -3,0303 10000 9,18286 -303,032

530 L3,23L5 40500 22,1148 927,4899

l.i I I i.l
I rn,Lr,
- tI l,llI y,i:'l-i

l: ,---,
ti?L,t$91l lti:liiji 1f,.'j.i15i,i1 I

-t{-}. r{:{i * t{:i3{-i}r// 1; ?; {i"ll:?11

,1 tt ,. ai"i)-?? tii *.*3?1 I iirilrrri:s.:tir:ttr

.1 .1

I:,, 1- r, i:i.j:{i.r ri.rjl

tt
rj r i i)r.'i it.(l:;lll {}.1;ls
Jt 't!
} ,Y '.1
:-t 1l

* {'i.=7-:t
i il.r,l;l I '-: ;*.13fii ltr:trrs
. r'r.1:iiti r ttr i'16;
.fir: - ii-t.{}:?lil

Page | 3
Solution Manual
Problems 2
The problem dealing with censored / suspended data is that we do not know exactly when
the suspended life data will be failed. The plotted data in probability paper is only for
complete / failed data. Hence, we do not know exactly the order of the failed data. The best
(MON)
thing to find the correct order of the failed data is to calculate mean order number
of the failed data.

Item number 1 is complete and the least data, therefore item number 1 (t = 5L00 hr)will
pose number 1 in the order of the failed / complete data.

The suspended life time of item number 2 (S1) is 9500 hr which is less than complete life
time of item number 3 (F2). ltem number 2 (S1) might be failed after 9500 hr but less that
j.5O0O hr or it migh also be more than 15000 hours. Therefore item number 2 (F2) might
pose either number 2 or 3 in the order of complete data.

F2 can occur in the second position six ways and in the third position two ways as it is

summarized in the following table.

f2 hPocition? fl, k Pcsitiott 3

h 2
1 7 4 5 1

F; n q ii !': ri f, r
r1

f,l {.t & F-3 F: f+ OR 5: c.

sl {- F3 S1
(- fir
JI !-., f.1

S1 S1 r, F, .5; 5t r}
r',, i-:; 5': 5; sr 5r r; S,:

Mean Order Number (MoN) for F2 can be calculated as follows.

MlN2-(6x2)+?x3) :2'25
6+2

Using the same logic on the third faiture, it can be located in position numbers 3, 4 and
5.

fl, iaPo*iti**3 {, in ?ositio* tt fi irPasitiet 5

1 2
',-t- lz I 3

?-1 i;i i1 lr,1 rr fi .F1 f't

& & OR 5r n lr, *B 5r F, L'
1.1

nl {t & $ iu, fl, 5l 5;

51 "! Fi r: ir, st 5r 51
,i! 5n s, ls, tr- f,J {r

Mean Order Number (MON)for F3 can be calculated as follows.

Page | 7
 MoNj=#
= 4.125

The new rank order after considering the suspended data will be as follow.

F1=5L00 1 'L2:,;96o/o, ,97-t9!%

i F2=15ooo 36,17yo
F3=40000 , 4,;325 ' 7Q,83o/, ,' 29U%

From the plotted data in the graph it is known that the curve crossed
the ul" line at
t = 31500 hr

It means that

MTTF =11)"= 31500

)u = L/31500 = 3,17 46E-O5

Page I B
 N- 6

TTF m n Fo.5o,.,n (N-i+1)/i MR = CDF R=1-MR

1 7 L2 2 0,8220 6,0000 L5,857% 83,L43%
2 12 10 4 0,8295 2,5000 32,534% 67,466%
3 19 8 5 0,8208 1,3333 47,746yo 52,254%
4 29 6 8 0,7930 0,7500 62,704% 37,296%
5 41, 4 10 0,7368 0,4000 77,237% 22,763%
6 57 2 12 0,6186 0,1667 90,653% 9,347%

Comparison

F(t)
BETAAND F
Cumulative Benard's
TTF DISTRIBUTIONS
Binomial Approximation
APPROACH

7 \0,910% 10,938% L6,857%

12 26,444% 26,563% 32,534%
19 4L,t41% 42,L88% 47,746yo
29 57,859% 57,8L3yo 62,704%
41. 73,556% 73,438% 77,237%
67 95,378% 89,063% 90,653%

Page | 6
Solution Manual
Problem 1

Cumulative binomial

ValueofZwhenP=0.50
Order
N=1 N=2 N=3 N=4 N=5 N=5 N=7
No.
1 50,000% 29,289% 20,630% 75,9!0% 12,945% 10,9L0% 94,280%

2 70,7110/o 50,001% 38,5730/o 31.,380% 26..444% 22,849%

3 79,370% 67,427% 50,001-% 41.,L4L% 36,41L%

4 94,090% 68,620% 57,859% 49,999%

5 97,055% 73,556% 63,589%

6 95,3L8% 77,15t%
7
90,572%

TTF MR = CDF R=1-MR

L 7 LO,910% 89,090%
2 12 26,444% 73,556%
3 t9 4L,1-4Lyo 58,859%
4 29 57,859% 42,L41%
5 41, 73,556% 26,444%
6 67 95,318% 4,682%

Benard's Approximation for Median Ranks

MR = i -o3
// + 0.4

N- 6
) TTF MR = CDF R=1-MR
1. 7 L0,938% 89,063%
2 t2 26,563% 73,438%
3 19 42,188% 57,813%
4 29 57,813% 42,188%
5 41, 73,438% 26,563%
6 67 89,063% 1.0,938%

BETA AND F DTSTRTBUTIONS APPROACH

l.
j'rrt-.,
atn _
N-i , +I. t
r-T-- O,io,rn,r:
)]
tu:ji:!-lfIl
i: : ,]l

Page I 5
PROBABILIry PLOTTING

Steps in determining parameter(s) of probability distribution using probability plotting.

. Linearize the unreliability function
' Construct the probability plotting paper
. Determine the X and Y positions of the plot points
. Using the plot, read any particular time or reliability / unreliability value of interest'

1,. Linearize the unreliability function

Case : Two-Parameter Weibull Distribution.

The unreliability function (CDF) of two parameter weibull distribution is given by following
equation.

_.tt\p
F(t)-Q(f) -t-e 'it
This functio" ..::"r linearized in the form of y: m' X * b by doing the following.
0(r) : t- e-(il
1- Q(t) = e-G)P
, ,,9
ln(t - A(t)) = lnl e-t'tt I

tn(r-0(0) : -(+\'
tn(-tn(t- 0(r)) ) : tn( (;)' )

tn(tn(r- Q(t))-'\ : Pt"(+\

tnltn(#) l=frlnt- Frnrt

Then by setting:

!:tn{ h(*o.,))
and:
y: Blnt
the equation can then be rewritten as:
y:px-frlnrt
which is now a linear equation with a slope of:
m= fi
and an intercePt of:
6 : _plnr1

Page | 1
2. Construct the probability plotting paper

The next task is to construct the Weibull probability plotting paper with the appropriate y
and x axes.

6#
qq{
tF*n*
*
&
qs..
r.i n
* & re- }.
-: {3 * *
q
* *i3
*1e
**+
*,#
&.#
€r:$ ,t
*4&
*-*
4*.S*
/i
@:*
&.*a
fi*I
t-"t
I tl
--f=,i}
1 qif 1'; I
!*&

1€*

3**
E.#

:*
*s
+3*
sr ?s

,*.* *.. "*=r.L* ii. " iu:;,

f, !il

parameter (scale parameter, characteristic of life) can be read on x axis (t) by firstly set the
11

e(t) = 63.2%. This value is obtained by setting 1= 11. Following is the calculation.
i * ,.:

tJ\t : \i -i - r: i', I

'lt -f.
n.rrrr=
.,t!1.{...1?

Page | 2
3. Determine the X and Y positions of the plot points

a. Determine the X positions of plot point.

. The points on the plot represent our data or, more specifically, our times-to-failure data.

b. Determine the Y positions of plot point.

. To determine the y plotting positions, we must first determine a value indicating the
corresponding unreliability for that fail ure.
. ln other words, we need to obtain the cumulative percent failed for each time-to-failure'
For example, the cumulative percent failed by 10 hours may be 25%, by 20 hours 50%,
and so forth

The most widely used method of determining this value is the method of obtaining the
median ronk for each failure

MEDIAN RANK
ihe gVl*tjia* *;t;.:ic* ffi.*th{is is used t* *i=ii*ix *,q *s'i!:x*tt l:f th* **reiixhiilis- for *;:*h
i*!iu:re. The median rank is the value that the true probability of failure, Q(t), should have
at the ith failure out of a sample of Nunits at the 50% confidence level.

The rank can be found for any percentage point, P, greater than zero and less than one, by
solving the cumulative binomial equation for Z.V\Jzx repr*s*n'i* th* ranlt. or unreliability
estimate, for the lth failure in the following equation for the cumulative binomial:

'
\- I \ i\ , 1
'
.r ! '
\I
"-
Where.
N = the sample size
j = the order number.

i** r::*rii+rr r*y:lc is obtained by **iui*€ tr.1!* aqr:*ti*r' l*:- i at-# = *"5#.
i r'.
:-,, \-[').i !
| ;
l.t

The following table shows the value of Z forvarious sample size (N) when P= 0'50'

Order ValueofZwhenP=0.50
No. N=1 N=2 N=3 N=4 N=5 N=5 N=7
1 so% 29.289% 20.630% 15.910% 12.945% LO.910% 9.428%
2 70.7L1% 50.o01% 38.573% 3L.380% 26.444% 22.849%
3 79.370% 6L.427% s0.001% 41.L41% 36.4L1%
4 84.090% 68.620% 57.859% 49.999%
5 87.05s% 73.556% 53.589%
6 9s.3L8% 77.L51%
7 90.572%

Page | 3
 BETA AND F DISTRIBUTIONS APPROACH
A more straightforward and easier method of estimating median ranks is by applying two
transformations to the cumulative binomial equation, first to the beta distribution and then
to the F distribution, resulting in:

MR:+
- L+N-Jj+1'Fo.so;^;n
m=2(N-j+L)
^-1.j
TL_I

Where Fo.so;rn,n denotes the F distribution at the 0.50 point, with m and n degrees of
freedom, for failure j out of N units.

Benard's Approximation for Median Ranks

Another quick, and less accurate, approximation of the median ranks is also given by:

t,i i 1,,

This approximation of the median ranks is also known as Benard's opproximotion.

4. Using the plot, read any particular time or reliability / unreliability value of interest.

CASE EXAMPLE

j.. Sixunitsareputonalifetestandtestedtofailure.ThefailuretimesareT,t2,!9,29,4L,
and 67 hours.
a. Estimate the failure rate for a 1-parameter exponential distribution using the
probability plotting method.
Find the MTTF and the reliability equation of the unit.
b. ,Estimate the scale parameter (n) and shape parameter / slope (S) if the life data set
is assumed to follow weibull distribution-

2- The following data sets represents life test of five units.

esoo n

3 F2 1s000
4 S2 22000

Find the MTTF and the reliability equation of the unit if the distribution is assumed to follow
1--pa rameter exponential distribution.

Page | 4
 100.0
xm
E
80.0 o
70.0 o
5
o)
60.0
!
50.0
0)
g
40.0

30.0
{rrt ll
d
(o
='
T
o)
E
o
20.o

10.0

8.0
7.0
e
x.
o.o
5.0
>
#
_: 4.0
-o
N
6
t
3.0

2_O €
q.
o
g
i)
o
3
@'
Oj
1.0 t,
o
3.
0.8 o
o
0.7
0.6 7
g-
o)
0.5 a)

0.4 o
o
a,

0.3 *o
n,

-f,
{{
0_2
n
=
g
O)
@
o
a)
o
3
0.1
99.90

99.00

95.00
90.00

80.00
70.00
60.00
50.00

40-00

30.00
N
o
o
o-
20.00 d
,
o
d
o
o
q
l
10.00
o
I
odq

o
6
5.00 I
o
E
@

lq.
3.00 lo
lq
2.00
lr
lr
t.
l:
{:
.tq
lo
le
tii
l3
t -
-t ,;
IB
0.50 l&
IE
l5
l^
t:
0.30 lo
la
lo
tb
0.20 ls.
l6
to
l3
lo
-l
Rank Regression on Y

performing a rank regression on Y requires that a straight line be fitted to the set of available data
from the points to the line is
points ,uah thrt the sum of the squares of the vertical deviations
minimized.

Y (RRY) were derived'

The following equations for rank regression on

-\,, n

,
I 'u' I't'
i '.
0 t! - n..- -T---" h!
I

\
and:
1' \': \-r.
_ - - \ ; " _..

)", !t, -
lt
:l -\-
; ..
Il \":. *, 1
I
r- .l i. : !
\
12
r'*
*l
.1
*1
'$t
for and r:r are:
For Exponential distribution, the equations

F(tr) = L - e-L(t-Y)

1-F(t,) -e-7(t-Y)
ln(r - r(ri)) - ln e-t(t-Y)

ln(t--r(ti)):-7(t-T)
ln(r-F(ti)):-Lt+)ry
g;' 'trllit *'ir'i-
and:
-i't It

ranks- Once o and l' are obtained' then A

and the f-'if ' )is estimated from the median
For the one-parameter exponential'
,nd "3- can easily be obtained from above equations.
equations for estimating o and b become:

tt , I1,
,\'
I'tt"q'
t-
'tl, : --
\

Iri
e.. 1

Page | 1
MAXIMUM LIKELIHOOD ESTIMATE
likelihood estimation method is, with
. From a statistical point of view, the method of maximum
of the parameter estimation techniques'
Some exceptions, considered to be the most robust
o values of the parameters, for a given
The basic idea behind MLE is to obtain the most likely
distribution, that will best describe the data

lf .' is a continuous random variable wilh pdf:

- ti t,, t' ,

li, . I i* . ... . Iir, are h' unknown parameters which need to be estimated, with R
where
independentobservatiofls, ":
'i '' "'i''ii,whichcorrespondinthecaseof lifedataanalysisto
failure times.

The likelihood function is given by:

R

L(Ir, 0 2,..., 0 r,lx y, x 2,..., x n) = L = ll f @; er, 0 r, "', 0 1, )

i=l

The logarithmic likelihood function is given by:

A : lnI =frn! 1*;0r,02,...,0t )

The maximum likelihood estimators (or parameter values)

of 0r, 02, "' , 0o are obtained by
maximizing L or A

oL=o
" ?1 =o
ao,- ao,

CASE EXAMPLE

life test data were obtained' Assuming that

14 units were being reliability tested and the following
estimate the parameters using the MLE
the data follow a 2-parameter exponential distribution,
method.

6 30
Life Test Data
7 35
Data point Time-to- 8 40
index failure
9 50
1-0 60
1 5
7L 70
2 10
12 80
3 15
13 90
4 20
1-4 100
5 25

Page | 1
Non-Parametric Life Data Analysis

Kaplan-Meier
product limit estimafor) is used as an alternative
The Kaplan-Meier estimator (also known as the
estimates of the unreliability for probability
to the median ranks method for calculating the
given by:
pto,,inS purposes. The equation of the estimator is

fir,: - ;'-lf -;!"] ' I," ':Ir

1 i
where:
:lr , l'*t;rl nuxrbr'r *l d*ls p{}}nfs
It -'lt:r" lrll*l *ug:l-l:i' t-rf *lrits

1" =' rluillLlrt ;-.rfiieiluitu lll l.!rr ,1r* rl:ltrr gri-rupl' rtrtri
rrf rlirvivi*g unlls in th*: 3t& rir'ltrr
grfil:I}
s, -nrtrilh{.,r

TheKaplan-Meierestimator,alsoknownastheproductlimitestimator'canbeusedtocalculate
values for non-parametric reliability for data sets
with multiple failures and suspensions' The
equation of the estimator is given by:
t,

riir,] -11 i.-1. -rrt

;r--l
f
where:
tit -, llrt'ft.rl;rl 'tu:llilt'r *l'ri:rl;l g'l:-rirt1*

?t ,.. lhe :,*tt-l} ril1l::bfr r-:f *ttits

The variable
*c is defined bY:
i: a- j
r, :- r? - \*...
l--- 'i - \-,'
1--- r
r 1. '.'.,;r
J:* J:*
where:
r-, .' lLe $Li:Illi{:r *l fueil*r;:= lrl !tir' 3tt' t-}.+l* gr{rtrll
Iltt'litl111trtl' *{ *iispt::5:t}ri*$ i::- t}itl .;tt' rl*'t:r XrilLl!}
rl
ir
for times at which one or more failures
Note that the reliability estimate is only calculated it;
of at time values that have failures and
occurred. For the sake of calculating the value
occur slightly after the failures' so that the
suspensions, it is assumed that the suspensions
and included in the count of '1i
suspended units are considered to be operating

Page | 1
CASE EXAMPLE
A group of units are put on a life test with the
following results'

Number State state End

in State (F or S) Time

3 F 9

1, S 9

1, F 11

I S t2
1 F 13

L S 13

t S 15

t F 17

1 F 71.

1, S 22

1 S 24
t S 26

1 F 28

L F 30

1 S 32

1 S 35

1 S 39

L S 41

failure time'
Use the Kaplan-Meier estimator
to determine the reliability estimates for each

Page | 2