Vous êtes sur la page 1sur 71

I n this chapter we shall consider a very important and growing area of applica-

I n this chapter we shall consider a very important and growing area of applica- tion of some of the probability concepts discussed in the previous chapters,

namely, Reliability Engineering. Reliability considerations are playing an in- creasing role in almost all engineering disciplines. Generally the term ‘reliabil- ity’ of a component (or a system, that is, a set of components assembled to per- form a certain function) is understood as its capability to function without break- down. The better the component performs its intended function, the more reliable it is. In the broader sense, reliability is associated with dependability, with suc- cessful performance and with the absence of breakdown or failures. However, from the engineering analysis point of view, it is necessary to define reliability quantitatively as a probability. Technically, reliability may be defined as the probability that a component (or a system) will perform properly for a specified period of time ‘t’ under a given set of operating conditions. In other words, it is the probability that the component does not fail during the interval (0, t). We give below the formal mathematical definition of reliability and discuss the related concepts.

The definitions given below with respect to a component hold good for a system

or a device also. If a component is put into operation at some specified time, say t = 0, and if T is the time until it fails or ceases to function properly, T is called the life length or time to failure of the component. Obviously T (0) is a continuous random variable with some probability density function f(t). Then the reliability or reliability function of the component at time ‘t’, denoted by R(t), is defined as R(t) = P(T > t) or 1 – P(T t)

= 1 – F(t) where F(t) is the cumulative distribution function of T, given by

F(t) =

t

0

f

( t ) dt

(1)

2

Probability, Statistics and Random Processes

 
 

t

Thus

R(t) = 1 –

f

( t ) dt

=

f

( t ) dt

(2)

 

0

t

Since F(0) = 0 and F() = 1 by the property of cdf, R(0) = 1 and R() = 0

 
 

d

i.e., 0 R(t) ≤ 1. Also since

dt F(t) = f(t),

 

we get

f (t) = – dR(t)

 

(3)

dt

Now the conditional probability of failure in the interval (t, t + t), given that the component has survived up to time t, viz.,

P{t T t + t/T t} =

ft

( )

t

1–

()

F t

, by the definitions of pdf and cdf

= f(t) t/R(t)

The conditional probability of failure per unit time is given by

f

(

t

)

R

(

t

)

and is

called the instantaneous failure rate or hazard function of the component, de- noted by l(t).

Thus

Now, using (3) in (4), we have

(5)

Integrating both sides of (5) with respect to t between 0 and t, we have

(4)

f

(

t

)

l(t) =

R

(

t

)

R

(

t

)

R

(

t

)

= l(t)

i.e.,

i.e.,

i.e.,

t

0

R

(

t

)

R

(

t

)

dt

t

0

{log R(t)}

=

=

log R(t) – log R(0) = –

log e [R(t)] = –

R(t) =

e

t

0

t

0

t

0

t

0

t

0

Using (6) in (4), we have

(

t ) dt

( t ) dt

(

t ) dt

(

t ) dt

( )

t

dt

[ R(0) = 1]

(6)

f(t) = l(t)

e

t

0

( t )

dt

(7)

The expected value of the time the failure T, denoted by E(T) and variance of T,

denoted by

2 are two important parameters frequently used to characterise

T

reliability. E(T) is called mean time to failure and denoted by MTTF.

Reliability Engineering

3

Now

 

tf (t)dt

0

tR ()t dt,

MTTF = E(T) =

= by (3)

0

on integration by parts

= – [tR(t)]

0

+

0

R(t)dt

Now

[tR(t)] t =

te

t

0

( t )

dt

t

=

t

e

t

0

( t )

dt

t

= 0

and

Using (9) and (10) in (8), we get

[tR(t)] t = 0 = 0 × R(0) = 0 × 1 = 0

MTTF =

0

R(t)dt

(8)

(9)

(10)

(11)

Var(T ) s T 2 = E{T E(T )} 2 or E(T 2 ) – {E(t)} 2

=

0

t

2

f (t)dt

– (MTTF) 2

(12)

Conditional reliability is another concept useful to describe the reliability of a component or system following a wear-in period (burn-in period) or after a warranty period. It is defined as

R(t/T 0 ) = P{T > T 0 + t/T > T 0 }

=

=

=

PT {

T

0

t

}

RT (

0

t

)

=

PT {

T

0

}

R ( T

0

)

e

T

0

0

t

( )

t

dt

e

T

0

0

( )

t

dt

=

e

T

0

0

t

() t

dt

e

T

0

T

0

t

( )

t

dt

T

0

0

() t

dt

(13)

(14)

Some of the probability distributions describe the failure process and reliability of a component or a system more satisfactorily than others. They are the expo- nential, Weibull, normal and lognormal distributions. We shall now derive the

4 Probability, Statistics and Random Processes

reliability characteristics relating to these failure distributions using the formu- las derived above.

If the time to failure T follows an exponential distribution with parameter λ, then its pdf is given by

(15)

f(t) = le lt , t 0

Then, from (2),

 

R(t) =

From (4),

l(t) =

t

e

t

dt

=

ft

(

)

e

t

(

R t

)

e

t

e

t

= l

t

e

t

(16)

(17)

This means that when the failure distribution is an exponential distribution with parameter l, the failure rate at any time is a constant, equal to l. Conversely when l(t) = a constant l, we get from (7),

f(t) = l .

e

t

t

dt

= le lt , t 0

Due to this property, the exponential distribution is often referred to as constant failure rate distribution in reliability contexts. We have already derived in the earlier chapters, that

MTTF = E(T ) =

1

(18)

and

Also

1

2

t

)

Var(T ) = s 2 T =

R(t/T 0 ) =

RT (

0

e

(

T

0

t

)

R ( T

0

)

e

T

0

by (16)

(19)

(20)

This means that the time to failure of a component is not dependent on how long the component has been functioning. In other words the reliability of the compo- nent for the next 1000 hours, say, is the same regardless of whether the compo- nent is brand new or has been operating for several hours. This property is known as the memoryless property of the constant failure rate distribution.

The pdf of the Weibull distribution was defined as f(t) = abt b1 e atb , t 0 An alternative form of Weibull’s pdf is

(21)

= e lt

f(t) =

t

1

t

exp

(22) is obtained from (21) by putting a =

1

,

q > 0, b > 0, t 0

(22)

. b is called the shape parameter

Reliability Engineering

5

and q is called the characteristic life or scale parameter of the Weibull’s distri- bution (22). If T follows Weibull’s distribution (22),

then

R(t) =

=

=

l(t) =

t

t

t

1

t

exp

.

e

x

dx ,

on putting

t

= exp

)

e x

f

(

t

t

t

R

(

t

)

.

=

1

dt

= x

(23)

(24)

As derived in the chapter ‘some special probability distributions’, we have

and

Now

1 MTTF = E(T ) = 1+ 2 1 Var(T ) = s 2 T
1
MTTF = E(T ) =
1+
2
1
Var(T ) = s 2 T = q 2
1
1
Rt (
T
)
0
R(t/T 0 ) =
R ( T
)
0
t
T
0
exp
=
T
0
exp
tT
T
0
0
= exp

2

(25)

(26)

(27)

If the time to failure T follows a normal distribution N(m, s ) its pdf is given by

In this case,

R(t) =

t

f

( t )

1 () t 2 f(t) = exp 2 2 2 MTTF = E(T ) =
1
()
t
2
f(t) =
exp
2
2
2
MTTF = E(T ) = m and
Var (T ) = s T = s 2 .
2

, |–< t < |

dt is found out by expressing the integral in terms of standard

normal integral and using the normal tables. l(t) is then found out by using (4).

6 Probability, Statistics and Random Processes

If X = log T follows a normal distribution N(m, s), then T follows a lognormal distribution whose pdf is given by

f(t) =

1 1 t exp log 2 st 2 2 s t m
1
1
t
exp
log
2
st
2
2 s
t
m

2

, t 0

(28)

where s = s is a shape parameter and t m , the median time to failure is the location parameter, given by log t m = m. It can be proved that

(29)

exp

s

2

2

MTTF = E(T ) = t m

and In this case,

Var(T ) = s T = t m exp(s 2 ) [exp (s 2 ) – 1]

2

2

(30)

F(t) = P(T t) = P{log T log t}

log

T

log t

 
 

,

 

 

log

T

log

t

m

1

t

Z

s

1

t

log

st

m

 

s

log

t

m

,

P

= P

= P

=

since log T follows a N(m, s)

where Z follows the standard

normal distribution N(0, 1). Then R(t) and l(t) are computed using (1) and (4).

Then R ( t ) and l ( t ) are computed using (1) and (4).

on widgets) manufactured by a certain company is given by f(t) =

The density function of the time to failure in years of the gizmos (for use

200

(t 10)

3

, t 0.

(a)

Derive the reliability function and determine the reliability for the first year of operation.

(b)

Compute the MTTF.

(c)

What is the design life for a reliability 0.95?

(d)

Will a one-year burn-in period improve the reliability in part (a)? If so, what is the new reliability?

(a)

f (t) =

R(t) =

R(1) =

200

(t 10)

3

f

( t ) dt

t

100

(1 10)

2

, t 0

=

100

100

(

t

10)

2

t

(

t

10)

2

= 0.8264.

Reliability Engineering

7

(b) MTTF =

=

0

R(t)dt

100

t

10

=

0

0

100

(

t

10)

2

dt

= 10 years.

(c) Design life is the time to failure (t D ) that corresponds to a specified reliability. Now it is requeired to find t D corresponding to R = 0.95

i.e.,

100

(t 10)

2

= 0.95

(t D + 10) 2 = 100.2632

t D = 0.2598 year or 95 days

(

t

t

(

R t

1)

=

100

100

=

121

R (1)

 

(t

11)

2

11

2

(t

11)

2

 

121

100

 

11)

2

(t

> 100

121

(t

10)

2

(d) R(t/1) =

Now

i.e., if

i.e., if

t

R(t/1) > R(t), if

(

10)

2

11)

2

10

> 10

11

t

11

i.e., 11t > 10t, which is true, as t 0 One year burn-in period will improve the relaibility.

Now

R(1/1) =

121

(1 11)

2

= 0.8403 > 0.8264.

The time to failure in operating hours of a critical solid-state

power unit has the hazard rate function ll(t) = 0.003

t

500

0.5

, for t 0.

(a)

What is the reliability if the power unit must operate continuously for 50 hours?

(b)

Determine the design ofe if a reliability of 0.90 is desired.

(c)

Compute the MTTF.

(d)

Given that the unit has operated for 50 hours, what is the probability that it will survive a second 50 hours of operation?

(a)

R(t) =

exp

t

0

( t )

dt

R(50) =

exp

= exp

50 t 0.003 500 0 0.003 2 3 2 . t 500 3
50
t
0.003
500
0
0.003
2
3 2
.
t
500
3

0.5

50

0

dt

8 Probability, Statistics and Random Processes

0.003 2 = exp 50 500 3 = exp[–0.03162] = 0.9689 (b) R(t D )
0.003
2
=
exp
50
500
3
= exp[–0.03162]
= 0.9689
(b) R(t D ) = 0.90
t
0.5
D
t
i.e.,
exp
0.003
dt
= 0.90
500
0
t
D
0.003
1 2
i.e.,
t
dt
= – 0.10536
500
0
0.003
2
3 2
i.e.,
t
= 0.10536
D
500 ×
3
2 3
3
500
∴ t D =
0.10536
= 111.54 hours.
2
0.003

(c) MTTF =

R(t)dt 0 0.003 2 3 2 500 3 = t e dt
R(t)dt
0
0.003
2
3 2
500
3
=
t
e
dt

0

3 2 2 at = e dt , where a = 0.003 3 500 0
3 2
2
at
=
e
dt ,
where a = 0.003
3
500
0
2
1 3
x
=
e
.
x
dx, on putting x = at 3/2
2 3
3
a
0
2
2
3
=
(2 3)
=
2 3
2 3
3a
3a
(5 3)
2

=

0.9033

2 3 a
2
3
a

, from the table of values of Gamma function.

= 45.65 hours.

(d) P(T 100/T 50) =

P T

(

100)

R

(100)

=

P T

(

50)

R

(50)

100 exp (t)dt = 50 0.002 = exp 100 500
100
exp
(t)dt
=
50
0.002
= exp
100
500

32

0.002 500
0.002
500

50

32

= exp[{– 0.08944} – {– 0.03162}]

= 0.9438

Reliability Engineering

9

t 0 , where t 0 is the maximum life of the blade.

The reliability of a turbine blade is given by R(t) =

1

t

t

0

2

, 0 t

(a)

Show that the blades are experiencing wear out.

(b)

Compute MTTF as a function of the maximum life.

(c)

If the maximum life is 2000 operating hours, what is the design life for a reliability of 0.90?

(a)

Now

R(t) =

1

t

t

0

2

l(t)= –

= –

, 0 t t 0

R

(

t

)

R

(

t

)

2

2

t

1

tt

00

t

1

t

0

=

2

t

0

t

l(t) =

2

(t

0

t)

2

> 0 and so l(t) is an increasing function of t.

When the failure rate increases with time, it indicates that the blades are experiencing wear out.

(b)

MTTF =

=

0

R(t)dt

t

0

3

1

=

t

0

0

t

t

0

3

1

t

0

0

t

t

0

=

2

t

0

3

dt

(c) When t 0 = 2000, R(t D ) = 0.90

i.e.,

1

t

D

2000

2

= 0.90

1

t

D

2000

t D = 102.63 hours.

= 0.9487

Given that R(t) =

0.001t e , t ≥ 0
0.001t
e , t ≥ 0

(a)

Compute the reliability for a 50 hours mission.

(b)

Show that the hazard rate is decreasing.

(c)

Given a 10-hour wear-in period, compute the reliability for a 50-hour mission.

(d)

What is the design life for a reliability of 0.95, given a 10-hour wear-in period?

10 Probability, Statistics and Random Processes

(a)

R(t) =

e

0.001t
0.001t

t 0

R(50) =

e

0.001 50
0.001 50

= 0.9512

R ( ) t 1 0.001 t 0.001 t (b) l(t) = = – e
R
( )
t
1
0.001
t
0.001
t
(b)
l(t) =
= –
e
e
0.001
R t
( )
t
0.001 ,
=
which is a decreasing function of t.
t
Rt (
T
)
0
(c)
R(t/T 0 ) =
R ( T
)
0
0.001
60
R
(60)
e
R(50/10) =
=
= 0.8651
R
(10)
0.001
10
e
(d)
R(t D /10) = 0.95
R ( t
10)
D
i.e.,
= 0.95
R (10)
0.001
(
t
10)
0.001 10
i.e.,
e
D
= 0.95 ×
e
0.001
(
t
10)
= 0.15129
D
t D = 12.89 hours.
A one-year guarantee is given based on the assumption that no

more than 10% of the items will be returned. Assuming an exponential distribution, what is the maximum failure rate that can be tolerated? If T is the time to failure of the item,

then

i.e.,

( no more than 10% will be returned)

P(T 1) 0.9 R(1) 0.9

i.e.,

i.e.,

i.e.,

t

e

1

dt

0.9

(–e lt ) 0.9 e l 0.9

1

l – 0.1054

l 0.1054/year

( T follows an exponential distribution)

A manufacturer determines that, on the average, a television set is used 1.8 hours per day. A one-year warranty is offered on the picture tube having a MTTF of 2000 hours. If the distribution is exponential, what percentage of the tubes will fail during the warranty period?

Reliability Engineering

11

Since the distribution of the time to failure of the picture tube is exponential, R(t) = e lt where l is the failure rate

Given that

MTTF = 2000 hours

0

e

t

dt

1

i.e.,

=

2000

= 2000 or l = 0.0005/hour

i.e.,

P(T 1 year) = P(T 365 × 1.8 hours) [ the T.V. is operated for 1.8 hours/day]

= 1 – P{T > 657}

= 1 – R(657)

= 1 e –0.0005 × 657

= 0.28

i.e., 28% of the tubes will fail during the warranty period.

Night watchmen carry an industrial flashlight 8 hours per night, 7 nights per week. It is estimated that on the average the flashlight is turned on about 20 minutes per 8-hour shift. The flashlight is assumed to have a constant failure rate of 0.08/hour while it is turned on and of 0.005/hour when it is turned off but being carried.

(a)

Estimate the MTTF of the light in working hours.

(b)

What is the probability of the light’s failing during one 8-hour shift?

(c)

What is the probability of its failing during one month (30 days) of 8-hour shifts?

In the earlier problems, we have not taken into account the possibility that an item may fail while it is not operating. Often such failure rates are small enough to be neglected. However, for items that are operated only a small fraction of time, failure during non-operation may be quite significant and so should be taken into account. In this situation, the composite failure rate λ c is computed as l c = fl 0 + (1 – f )l N , where f is the fraction of time the unit is operating and l 0 and l N are the failure rates during operating and non-operating times]

Since the flashlight is used only a small fraction of time (20 minutes out of 8 hours), we compute the composite failure rate l c and use it for other calculations. l c = fl 0 + (1 – f )l N

(a)

(b)

(c)

=

1

24

× 0.08 +

23

24 × 0.005 = 0.008125/hour

MTTF = 1

1

=

c

0.008125

= 123 hours

P(T 8) =

8

0

c e

c t

dt = 1–e 8lc = 0.0629

p = probability of failure in one day (night) = 0.0629

12 Probability, Statistics and Random Processes

q = 1 – p = 0.9371

n = 30

Probability of failure in 30 days

= 1 – Probability of no failure in 30 days

= 1 – nC 0 p 0 q n , by binomial law

= 1 – (0.9371) 30

= 0.8576

A relay circuit has an MTBF of 0.8 year. Assuming random failures,

(a)

Calculate the probability that the circuit will survive 1 year without failure.

(b)

What is the probability that there will be more than 2 failures in the first year?

(c)

What is the expected number of failures per year?

Since the failures are random events, the number of failures in an interval of

length t follows a Poisson process, given by P{N(t) = n} =

… ∞, where l, = failure rate.

e

t

(

t

)

n

n !

, n = 0, 1, 2,

Then the time between failures follows an exponential distribution with mean

Now

MTBF = Mean time between failures = 1

= 0.8 year

1

.

(a)

(b)

(c)

l =

1

0.8

per year = 1.25 per year

e

0

0!

P{N (1) = 0} =

= e 1.25 = 0.2865

012

1!

2!

0!

P{N(1) > 2} = 1 – e l

= 1 – 0.2865

1

1.25

(1.25)

2

2

= 0.1315

E{N(t)} = lt E{Number of failures per year} = l = 1.25

For a system having a Weibull failure distribution with a shape parameter of 1.4 and a scale parameter of 550 days, find

(a) R(100 days); (b) the B1 life; (c) MTTF; (d) the standard deviation, (e) the

design life for a reliability of 0.90.

The pdf of the Weibull distribution is given by f(t) =

t 0. Now b = 1.4 and q = 550 days

t

1

t

.

exp

,

Reliability Engineering

13

(a) R(t) = exp

R(100) =

(b)

exp

t

100

550

1.4

= 0.9122

The life time corresponding to a reliability of 0.99 is called the B1 life.

Similarly that corresponding to R = 0.999 is called the B.1 life.

Let t R be the B1 life of the system

Then

R(t R ) = 0.99

i.e.,

exp

t

R

550

1.4

=

0.99

t

R

550

1.4

= 0.01005

t R = 550 ×

(0.01005)

1

1.4

= 20.6 days

(c)

MTTF =

= 550 × = 550 ×

1 1 1.4
1
1
1.4

= 550 ×

(1.71)
(1.71)

= 550 × 0.91057, from the Gamma tables

= 500.8 days

(d)

(S.D.) 2

2 2 1 = q 2 1 1 2 2 1 2 1 1 550
2
2
1
=
q 2
1
1
2
2
1
2
1
1
550
=
1.4
1.4
2
=
550
(2.43)
2
(1.71)
2
2
=
550
1.43
(1.43)
(1.71)
= 550 2 [1.43 × 0.88604 × (0.91057) 2 ]

S.D = 550 × 0.66174 = 363.96 days

14 Probability, Statistics and Random Processes

(e)

Let t D be the required design life for R = 0.90

R(t D ) =

exp

t

D

550

1.4

t D = 550 ×

t

D

550

550

1.4

(0.10536)

1

1.4

=

0.90

= 0.10536

= 110.2 days

A device has a decreasing failure rate characterised by a two- parameter Wibull distribution with q = 180 years and b = 0.5. The device is required to have a design life reliability of 0.90.

(a)

What is the design life, if there is no wear-in period?

(b)

What is the design life, if there is a wear-in period of 1 month in the beginning?

(a) Let t D be the required design life for R = 0.90

R(t D ) =

exp

t

D

180

0.5

= 0.10536

t

D

180

0.5

=

0.90

t D = 180 × (0.10536) 2 = 1.998 2 years.

(b) If T 0 is the wear-in period, then

R(t/T 0 ) =

exp

tT

0

T

0

Let t W be the required design life with wear-in.

Then

exp

t

W

1

12

180

0.5

1

12

180

0.5

= 0.90

i.e.,

i.e.,

12

t

W

1

0.5

+

12

t

W

12

1

180

0.5

= 0.12688

12

180

1

12

180

0.5

= – 0.10536

12t W + 1 = 12 × 180 × 0.01610 = 34776 t W = 2.18 years

i.e.,

Thus a wear-in period of 1 month increases the design life by nearly 0.81 year or

10 months.

Reliability Engineering

15

Two