1

Fundamentals of Reliability Engineering

and Applications


Dr. E. A. Elsayed
Department of Industrial and Systems Engineering
Rutgers University
([email protected])



Systems Engineering Department
King Fahd University of Petroleum and Minerals
KFUPM, Dhahran, Saudi Arabia
April 20, 2009


2
Reliability Engineering
Outline
Reliability definition
Reliability estimation
System reliability calculations
2
3
Reliability Importance
One of the most important characteristics of a product, it
is a measure of its performance with time (Transatlantic
and Transpacific cables)

Products recalls are common (only after time elapses). In
October 2006, the Sony Corporation recalled up to 9.6
million of its personal computer batteries

Products are discontinued because of fatal accidents
(Pinto, Concord)

Medical devices and organs (reliability of artificial organs)
3
4
Reliability Importance
Business data
4
Warranty costs measured in million dollars for several large
American manufacturers in 2006 and 2005.
(www.warrantyweek.com)
Maximum Reliability level
R
e
l
i
a
b
i
l
i
t
y
W
ith
R
e
p
a
irs
Time
N
o
R
e
p
a
irs
Some Initial Thoughts
Repairable and Non-Repairable
Another measure of reliability is availability (probability
that the system provides its functions when needed).
5
Some Initial Thoughts
Warranty
Will you buy additional warranty?
Burn in and removal of early failures.
(Lemon Law).

Time
F
a
i
l
u
r
e

R
a
t
e
Early Failures
Constant
Failure Rate
Increasing
Failure
Rate
6
7
Reliability Definitions
Reliability is a time dependent characteristic.

It can only be determined after an elapsed time but
can be predicted at any time.

It is the probability that a product or service will
operate properly for a specified period of time (design
life) under the design operating conditions without
failure.
7
8
Other Measures of Reliability
Availability is used for repairable systems

It is the probability that the system is operational
at any random time t.

It can also be specified as a proportion of time
that the system is available for use in a given
interval (0,T).
8
9
Other Measures of Reliability
Mean Time To Failure (MTTF): It is the average
time that elapses until a failure occurs.
It does not provide information about the distribution
of the TTF, hence we need to estimate the variance
of the TTF.

Mean Time Between Failure (MTBF): It is the
average time between successive failures.
It is used for repairable systems.
9
10
Mean Time to Failure: MTTF
1
1
n
i
i
MTTF t
n
=
=

0 0
( ) ( ) MTTF tf t dt R t dt

= =
} }
Time t
R
(
t
)

1
0
1
2
2 is better than 1?
10
11
Mean Time Between Failure: MTBF
11
12
Other Measures of Reliability
Mean Residual Life (MRL): It is the expected remaining
life, T-t, given that the product, component, or a system
has survived to time t.


Failure Rate (FITs failures in 10
9
hours): The failure rate in
a time interval [ ] is the probability that a failure per
unit time occurs in the interval given that no failure has
occurred prior to the beginning of the interval.

Hazard Function: It is the limit of the failure rate as the
length of the interval approaches zero.
1
2
t t
1
( ) [ | ] ( )
( )
t
L t E T t T t f d t
R t
t t t

= > =
}
12
13
Basic Calculations
0
1
0 0
0
( )

, ( )
( )
( )

( ) , ( ) ( )
( )
n
i
f
i
f
s
r
s
t
n t
MTTF f t
n n t
n t
n t
t R t P T t
n t t n

=
= =
A
= = > =
A

Suppose n
0
identical units are subjected to a
test. During the interval (t, t+t), we observed
n
f
(t) failed components. Let n
s
(t) be the
surviving components at time t, then the MTTF,
failure density, hazard rate, and reliability at
time t are:
13
14
Basic Definitions Contd
The unreliability F(t) is

( ) 1 ( ) F t R t =


Example: 200 light bulbs were tested and the failures in
1000-hour intervals are
Time Interval (Hours) Failures in the
interval
0-1000
1001-2000
2001-3000
3001-4000
4001-5000
5001-6000
6001-7000
100
40
20
15
10
8
7
Total 200
14
15
Calculations
Time
Interval
Failure Density
( ) f t x
4
10

Hazard rate
( ) h t x
4
10

0-1000


1001-2000

2001-3000



6001-7000
3
100
5.0
200 10
=

3
40
2.0
200 10
=

3
20
1.0
200 10
=

..

3
7
0.35
200 10
=

3
100
5.0
200 10
=

3
40
4.0
100 10
=

3
20
3.33
60 10
=

3
7
10
7 10
=

Time Interval
(Hours)
Failures
in the
interval
0-1000
1001-2000
2001-3000
3001-4000
4001-5000
5001-6000
6001-7000
100
40
20
15
10
8
7
Total 200
15
16
Failure Density vs. Time
1 2 3 4 5 6 7 x 10
3

Time in hours
16

1
0
-
4

17
Hazard Rate vs. Time
1 2 3 4 5 6 7 10
3

Time in Hours
17

1
0
-
4

18
Calculations
Time Interval
Reliability ( ) R t
0-1000

1001-2000

2001-3000



6001-7000
200/200=1.0

100/200=0.5

60/200=0.33



0.35/10=.035

Time Interval
(Hours)
Failures
in the
interval
0-1000
1001-2000
2001-3000
3001-4000
4001-5000
5001-6000
6001-7000
100
40
20
15
10
8
7
Total 200
18
19
Reliability vs. Time
1 2 3 4 5 6 7 x 10
3

Time in hours
19
20
Exponential Distribution
Definition
( ) exp( ) f t t =
( ) exp( ) 1 ( ) R t t F t = =
( ) 0, 0 t t = > >
(t)
Time
20
21
Exponential Model Contd
1
MTTF

=
2
1
Variance

=
1
2 Median life (ln )

=
Statistical Properties
21
6
Failures/hr 5 10

=
MTTF=200,000 hrs or 20 years
Median life =138,626 hrs or 14
years
22
Empirical Estimate of F(t) and R(t)
When the exact failure times of units is known, we
use an empirical approach to estimate the reliability
metrics. The most common approach is the Rank
Estimator. Order the failure time observations (failure
times) in an ascending order:

1 2 1 1 1
... ...
i i i n n
t t t t t t t
+
s s s s s s s s
23
Empirical Estimate of F(t) and R(t)
is obtained by several methods

1. Uniform naive estimator

2. Mean rank estimator

3. Median rank estimator (Bernard)

4. Median rank estimator (Blom)
( )
i
F t
i
n
1
i
n +
0 3
0 4
.
.
i
n

+
3 8
1 4
/
/
i
n

+
24
Empirical Estimate of F(t) and R(t)
Assume that we use the mean rank estimator

24
1

( )
1
1

( ) 0,1, 2,...,
1
i
i i i
i
F t
n
n i
R t t t t i n
n
+
=
+
+
= s s =
+
Since f(t) is the derivative of F(t), then
1
1

( ) ( )

( )
.( 1)
1

( )
.( 1)
i i
i i i i
i
i
i
F t F t
f t t t t
t n
f t
t n
+
+

= A =
A +
=
A +
25
Empirical Estimate of F(t) and R(t)
25
1

( )
.( 1 )

( ) ln ( ( )
i
i
i i
t
t n i
H t R t
=
A +
=
Example:

Recorded failure times for a sample of 9 units are
observed at t=70, 150, 250, 360, 485, 650, 855,
1130, 1540. Determine F(t), R(t), f(t), ,H(t) ( )t
26
Calculations
26
i t (i) t(i+1) F=i/10 R=(10-i)/10 f=0.1/At =1/(At.(10-i)) H(t)
0 0 70 0 1 0.001429 0.001429 0
1 70 150 0.1 0.9 0.001250 0.001389 0.10536052
2 150 250 0.2 0.8 0.001000 0.001250 0.22314355
3 250 360 0.3 0.7 0.000909 0.001299 0.35667494
4 360 485 0.4 0.6 0.000800 0.001333 0.51082562
5 485 650 0.5 0.5 0.000606 0.001212 0.69314718
6 650 855 0.6 0.4 0.000488 0.001220 0.91629073
7 855 1130 0.7 0.3 0.000364 0.001212 1.2039728
8 1130 1540 0.8 0.2 0.000244 0.001220 1.60943791
9 1540 - 0.9 0.1 2.30258509
27
Reliability Function
27
0
0.2
0.4
0.6
0.8
1
1.2
0 200 400 600 800 1000 1200
Reliability
Time
28
Probability Density Function
28
0.000000
0.000200
0.000400
0.000600
0.000800
0.001000
0.001200
0.001400
0.001600
0 200 400 600 800 1000 1200
Density
Function
Time
29
Failure Rate
Constant
29
0.000100
0.000300
0.000500
0.000700
0.000900
0.001100
0.001300
0.001500
0.001700
0.001900
0 200 400 600 800 1000 1200
Failure Rate
Time
30
Exponential Distribution: Another
Example
Given failure data:

Plot the hazard rate, if constant then use the
exponential distribution with f(t), R(t) and h(t) as
defined before.

We use a software to demonstrate these steps.
30
31



Input Data
31
32
Plot of the Data
32
33
Exponential Fit
33
Exponential Analysis
35



Go Beyond Constant Failure Rate
- Weibull Distribution (Model) and
Others
35
36
The General Failure Curve
Time t
1
Early Life
Region
2
Constant Failure Rate
Region
3
Wear-Out
Region
F
a
i
l
u
r
e

R
a
t
e

0
ABC
Module
36
37
Related Topics (1)
Time t
1
Early Life
Region
F
a
i
l
u
r
e

R
a
t
e

0
Burn-in:
According to MIL-STD-883C,
burn-in is a test performed to
screen or eliminate marginal
components with inherent
defects or defects resulting
from manufacturing process.
37
38
2 1
Motivation Simple Example
Suppose the life times (in hours) of several
units are:
1 2 3 5 10 15 22 28
1 2 3 5 10 15 22 28
10.75 hours
8
MTTF
+ + + + + + +
= =
3-2=1 5-2=3 10-2=8 15-2=13 22-2=20 28-2=26
1 3 8 13 20 26
(after 2 hours) 11.83 hours >
6
MRL MTTF
+ + + + +
= =
After 2 hours of burn-in
39
Motivation - Use of Burn-in
Improve reliability using cull eliminator
1
2
MTTF=5000 hours
Company
Company
After burn-in
Before burn-in
39
40
Related Topics (2)
Time t
3
Wear-Out
Region
H
a
z
a
r
d

R
a
t
e

0
Maintenance:
An important assumption for
effective maintenance is that
component has an
increasing failure rate.

Why?


40
41
Weibull Model
Definition
1
( ) exp 0, 0, 0
t t
f t t
| |
|
| q
q q q

(
| | | |
= > > > (
| |
\ . \ . (

( ) exp 1 ( )
t
R t F t
|
q
(
| |
= = (
|
\ . (

1
( ) ( ) / ( )
t
t f t R t
|
|

q q

| |
= =
|
\ .
41
42
Weibull Model Cont.
1/
0
1
(1 )
t
MTTF t e dt
|
q q
|

= = I +
}
2
2
2 1
(1 ) (1 ) Var q
| |
(
| |
(
|
|
(
\ .
(

= I + I +
1/
Median life ((ln2) )
|
q =
Statistical properties
42
43
Weibull Model
43
44
Weibull Analysis: Shape Parameter
44
45
Weibull Analysis: Shape Parameter
45
46
Weibull Analysis: Shape Parameter
46
47
Normal Distribution
47
Weibull Model
1
( ) ( ) .
t
h t

u u

=


( )
1
( ) ( )
t
t
f t e

u u

=


0
( )
1
( ) ( )
t
F t e d

,
u

,
,
u u

=
}


( )
( ) 1
t
F t e

=



( )
( )
t
R t e

=




Input Data
Plots of the Data
Weibull Fit
Test for Weibull Fit
Parameters for Weibull
Weibull Analysis



Example 2: Input Data
Example 2: Plots of the Data
Example 2: Weibull Fit
Example 2:Test for Weibull Fit
Example 2: Parameters for Weibull
Weibull Analysis
61
Versatility of Weibull Model
Hazard rate:
Time t
1 | =
Constant Failure Rate
Region
H
a
z
a
r
d

R
a
t
e

0
Early Life
Region
0 1 | < <
Wear-Out
Region
1
| >
1
( ) ( ) / ( )
t
t f t R t
|
|

q q

| |
= =
|
\ .
61
62
( ) 1 ( ) 1 exp
1
lnln ln ln
1 ( )
t
F t R t
t
F t
|
q
| | q
(
| |
= = (
|
\ . (

=

Graphical Model Validation

Weibull Plot
is linear function of ln(time).
Estimate at t
i
using Bernards Formula

( )
i
F t
0.3

( )
0.4
i
i
F t
n

=
+
For n observed failure time data
1 2
( , ,..., ,... )
i n
t t t t
62
63
Example - Weibull Plot
T~Weibull(1, 4000) Generate 50 data
10
-5
10
0
10
5
0.01
0.02
0.05
0.10
0.25
0.50
0.75
0.90
0.96
0.99
Data
P
r
o
b
a
b
i
l
i
t
y
Weibull Probability Plot
0.632
q
|
If the straight line fits
the data, Weibull
distribution is a good
model for the data
63