LECTURE NOTE - BASIC TOOLS OF RELIABILITY

Reliability has been defined as “the ability of a system or component to perform its
required functions under stated conditions for a specified period of time” by the Institute
of Electrical and Electronics Engineering (IEEE).

Let X be a non-negative random variable representing the lifetime of an item with life
distribution F (x), satisfying F (0) = 0.

Definition 1.1: The life distribution denoted by F(x) is the probability that the item
fails before time ‘x’, and is given by
F ( x ) = P  X  x  , F ( 0 ) = 0. (1.1)

Definition 1.2: The reliability of an item is defined as the probability that it will
perform satisfactorily for a specified period of time without a major failure, and is given
by
F ( x ) = P X  x  = 1 − F ( x ). (1.2)

It is also known as survival function, survival distribution, and survival probability.

Interpretation: If F ( 300 ) = 0.90 

If 100 identical items are operated for 300 units of time (say, hr.), 90 items are likely to
operate without failure for 300 hours and the rest fail some time in [0, 300].

Definition 1.3: The pdf or failure density is given as:

dF(x)
f(x) = (1.3)
dx

Also

F(x + x) - F(x)

f(x) = lim
x →0 x

P  X  x + x  - P  X  x 
= lim
x →0 x

1
 P X  x + x - X  x
= lim
x →0 x

P  x  X  x + x 
= lim (1.4)
x →0 x

 f(x) x P  x  X  x + x  (1.5)

If a new item is put to test at time x = 0, then (1.5) answers the question “What is the
probability that this item will fail in the interval (x, x + x ] ?” In other words, this
probability is approximately equal to the failure density function, f(x), at time x
multiplied by the length of the interval, x .

Definition 1.4: The failure rate function denoted by h ( x ) is defined as

P  x  X  x + x | X > x 
h(x) = lim (1.6)
x →0 x
This is the limit of the probability that a unit fail for the first and only time in a small
interval given that it survived to the beginning of the interval.
h ( x ) is a measure of the rate at which failures in the total population occur at time ‘x’.

It is also known as conditional failure rate function, or the hazard rate function.
It is the limit of a conditional probability but is not a conditional pdf. The function need
not integrate to one.
From (1.6) it follows that
h(x) x P  x  X  x + x | X > x  (1.7)

Consider an item that has survived until time ‘x’ , then (1.7) answers the question “What
is the probability that this item will fail in the next interval (x, x + x ] ?” In other words
this probability is approximately equal to the failure rate function, h(x), at time x
multiplied by the length of the interval, x .

Knowing any one of f(x), F(x), F ( x ) , or h(x) is enough to find all the others.

Theorem 1 : For a continuous random variable, the hazard rate function is

2
 f (x)
h(x) = , (1.8)
F( x)

for ‘x’ such that F ( x )  0.

Proof :
P  x  X  x + x | X > x 
h(x) = lim
x →0 x
P  x  X  x + x, X > x 
= lim
x →0 xP  X > x 

1 P  x  X  x + x 
= lim
F(x) x →0 x
f (x)
= .
F(x)

Theorem 2: For a continuous random variable with hazard rate function h(x) , the life
distribution and the failure density are given by:
x
−  h(u) du
F(x) = 1 - e 0 (1.9)
and
x
−  h(u) du
f(x) = h(x) e 0 , (1.10)
respectively.
x
Proof: Consider  h(u) du
0
x x f (u)
 h(u) du =  du
0 0 F(u)
x dF(u)
= 
0 F(u)

3
 x
= −  dF(u)
0 F(u)

Let F(u) = z  dF(u) = dz

x F(x) dz
  h(u) du = − 
0 1 z
1 dz
= 
F(x) z

= − log e (F(x))

x
−  h(u) du
 F(x) = e 0 ,0  x  , (1.11)

which is an alternative form of reliability and signifies that one can find F ( x ) from

h ( x ).

It follows from (1.11) that

x
−  h(u) du
F(x) = 1- e 0

dF(x)
Also f(x) =
dx

x
 d x  −  h(u) du
=   h(u) du  e 0
 dx 
 0 

x
−  h(u) du
= h(x) e 0 .

Definition 1.5: The 100pth percentile (or pth quantile) of a distribution F() is the age
p by which a proportion ‘p’ of the population fails. It is the solution of

4
 ( )
p = F p . (1.12)

It is also called the “100p percent point” or “p fractile”. The 50% point is called the
median, and is usually used as a “typical life”. The characteristic life is the time at
which 63.2% of the units will fail.

Definition 1.6: The cumulative hazard function is given by:

x
H(x) =  h(u) du
0

= − log e F ( x ) (1.13)

Definition 1.7:The conditional survival function of unit of age ‘x’ is the conditional
probability that a unit of age x will survive for an additional ‘t’ units of time and is given
by:

Fx ( t ) = P  X − x  t | X>x  (1.14)

(1.14) can be re-written as:

P X − x  t , X > x
Fx ( t ) =
P X > x

P X  t + x 
=
P X > x 

F(t + x)
= x such that F(x) > 0. (1.15)
F(x)

Mean Lifetime (Mean Time to Failure (MTTF)) of a unit

Theorem 3: The mean life time of a unit is


E ( X ) =  F(x)dx. (1.16)
0

Proof: By definition

5
 
E[X] =  xf (x)dx
0

T
= lim
x →0
 xf (x)dx
0

T
= lim  x dF(x)
T→
0

 T 
= lim TF(T) -  F(x)dx 
T→  
 0

 T 
= lim  −T(1 − F(T)) + T-  F(x)dx 
T→  
 0

 T T 
= lim  −T(1 − F(T)) +  dx-  F(x)dx 
T→ 
 0 0 
 T 
= lim  −T(1 − F(T)) +
T→ 
 (1 − F(x)) dx 

 0


=  F(x)dx
0

since lim T(1 − F(T)) = lim T F(T)

T → T →

= lim TP[X > T]

T →

= lim T
T →  dF(x)
T


 lim
T →  x dF(x) = 0 (as E[X] exists) 
T →
lim T(1 − F(T)) =0.
T

6
Mean Residual Lifetime of a unit of age x
Mean Residual Lifetime of a unit of age x is given by:

eF (x) = E (X − x) | X > x  (1.17)

(1.17) can be re-written as


eF (x) =  Fx (t) dt
0


F(t + x)
=  F(x)
dt
0


F(u)
=  F(x) du which is obtained by putting t+x = u  dt = du
x


1
F(x) x
Thus, eF (x) = F(u)du . (1.18)

While the failure rate function at x provides information about the immediate future
after time x, the mean residual life function provides information about the whole
interval after x.

It follows from (1.18) that e F (0) = E[X] .

CLASSES OF LIFE DISTRIBUTIONS BASED ON THE NOTIONS OF AGING

‘Aging’ is categorized into positive aging, negative aging and no-aging. Positive aging
is the adverse effect of age on the residual lifetime of the unit, the dual concept of
negative aging meaning a beneficial effect of age on the unit’s residual lifetime. ‘No-
aging’ is interpreted as the phenomenon that age has no effect on the residual lifetime
of the unit. The memoryless property of exponential distribution is viewed as no-aging
property. Further since exponential distribution is the only distribution with constant

7
failure rate function and constant mean residual life, hence no-aging also corresponds
to constant failure rate and constant mean residual life. The notion of aging plays an
important role in reliability and maintenance theory. Classes of life distributions based
on aging properties help a reliability practitioner to consider a problem of a character
somewhat different from the usual ones. Instead of assuming that nothing is known
about the underlying life distribution, it is assumed that some type of aging property
(say, failure rate is increasing or decreasing) holds for the life distribution. Various
classes of life distributions have been devised to describe positive (negative) aging
property of a unit throughout its life span.

Definition 1.8 : A life distribution F has Increasing Failure Rate (IFR) if the failure fate
function h F (x)  h(x) is increasing 0  x <  .

Definition 1.9 : A life distribution F has Increasing Failure Rate Average (IFRA) if the
x

 h(u) du H(x)
average failure fate function A F (x) = 0 = is increasing 0 < x <  .
x x

Definition 1.10 : A life distribution F has New Better than Used (NBU) if

Fx ( t )  F ( t ) ; x,t  0 .

This is equivalent to saying that the survival probability of a unit of age x is smaller
than the survival probability of a new unit.

Definition 1.11 : A life distribution F has Decreasing Mean Residual Life (DMRL) if

e F ( x ) is decreasing on [0, ) .

Definition 1.12 : A life distribution F has New Better than Used in Expectation (NBUE)
if

e F ( x )  e F ( 0 ) =  F x  0.

The NBUE property states that mean residual lifetime of a unit of age x is smaller than
the mean lifetime of a new unit.

8
These classes of life distributions are related through the following implications:

IFR  IFRA  NBU

 
DMRL  NBUE

The duals of the classes depicting negative aging can be obtained by replacing
Decreasing (D) by Increasing (I) , Better (B) by Worse (W) and by reversing the
direction of inequality.