5.

0 Reliability

When reliability is mentioned, we would often consider issues

like:

 How long will this battery last?

 Will my cellphone always have a clear signal
anywhere in Metro Manila?
 How many components should I place as
redundant components to make my system
better?
 What should be the warrantee period of our
product so that we will minimize costs of full
replacements?

Reliability is the probability that a given product will successfully

perform a required function without failure, under specified
environmental conditions, for a specified period of time.

Failure must be clearly specified in the first place:

 Let’s take an electric battery, for example: if it is being
recharged, a battery is said to fail the first time it does not reach
100 Ampere-hours, its minimum required recharging limit.
Thereby, the battery that should have lasted 100 hours at 1A
constant power drain, only reaches 60 hours.
 A cooling tower in a chemical process is said to fail if it floods.
This may be due to a lower fluid drainage rate because the
cooling elements are bunched up.
 The same cooling tower can also fail if channeling occurs—air
and water flow in channels within the cooling tower and no
contact between the hot water and the cooling air occurs.
 A bridge may fail when the level between the bridge and the
cantilever edge deviates by more than 3 inches.
 A clock goes slower by more than 3 seconds every 24 hour
period.
 A car fails when the ride becomes rough.
 A tire fails when it cannot sustain the normal airpressure put in
for more than 1 hour.

5.1 Lifetime Distribution

Lecture Notes: Reliability page 83

 The lifetime distribution of a product is the basic information
from which all measures of reliability are evaluated. It consists of
the lifetimes of all items in the population of a product. Lifetime
may be expressed in hours, days, number of use, number of
kilometers, etc.

The life distribution of a product is known if its shape is known

along with the estimated input parameter values of the distribution.
Shapes may be normal bell-shaped curves, exponential slide-shape,
or others as may apply. When the input parameters are known for
the lifetime distribution, we may then use this information to predict
the reliability of the product across its lifetime.

5.2 Failure Rate or Hazard Rate h(t)

Failure rate (or hazard rate) h(t) is the likelihood (probability)

that a product will fail as a function of its age. It is measured as the
proportion that will fail in the next unit of time, out of those units
htat have survived up to that time. The following table should
demonstrate this concept.

No. of
Interval No. of Percent survivors Reliability Failure
(years) failures fail in at end of at end of rate at
interval interval interval end of
interval
0-0 0 0 1000 1.000 0.450
0< t  1 450 45% 550 55% 0.145
1< t  2 80 8% 470 47% 0.181
2< t  3 85 8.5% 385 38.5% 1.000
3< t  4 385 38.5% 0 0% --.--
Total 1,000 100%

This means that 45% of the survivors after time 0 should be

expected to fail in the next one-year interval. Which is 450 out of
1000 survivors.

The next failure rate means that 80 of the remaining 550

survivors at the end of year 1 is expected to fail in between the 1st and
2nd birthday of the product.

This table’s figures can be graphically shown in the column and

line graphs on the next page.

Lecture Notes: Reliability page 84

 50% 1
Percent Failure 40% 0.8

Reliability
30% 0.6

20% 0.4

10% 0.2

0%
0
1 2 3 4 1 2 3 4 5

Years Years

1
Hazard or failure rate

0.8

0.6

0.4

0.2

0
1 2 3 4
years

Normally, failure rate may be constant through its lifetime but

increases dramatically when a certain age is reached. The failure rates
for common applications of reliability is shown in the next page. (page
84).

Mathematical Expression for Failure/Hazard rate h(t)

For any population that has life distribution represented by a

probability distribution function of lifetime t, call this f(t), and
Reliability R(t) = the probability that lifetime will exceed t  1-P(t<t) =
1-F(t), the hazard rate can be expressed as:
f (t )
h(t ) 
R (t )
Most electronic devices have an exponentially distributed
lifetime, we should derive an expression for its failure/hazard rate.

For an exponential distribution with expected time , we can

express reliability function R(t) as

R(t) = 1- F(t)

Lecture Notes: Reliability page 85

 = 1-(1-e –t/
 or e–t/
The failure rate is therefore a constant:
t
1 
e
f (t )  1
h(t )   
R(t ) 
t


e

Meaning, there is 1 failure every  period of time.

Infant
Mortality Wearout

Failure
Rate Failure Chance failures
Rate

Essential y constant failure rate

like in most electronic hardware, Bathtub curve: failure rate curve;
mature electromechanical
Complex mechanical equipment
equipment

Failure
Rate Failure
Rate

Decreasing failure rate:

Computer software, Increasing failure rate:
Structural members Parts subject to degradation
like rubber parts

Lecture Notes: Reliability page 86

5.4 Mean Time between Failure (MTBF) or Mean time to
Failure (MTTF)

Mean time to failure (MTTF) is the mean of the life distribution.

It is the average of life length for all the units in the population. MTTF
is usually used to refer to products that fail and are replaced. (ex.
Disposable dry-cell batteries) Mean time between failures (MTBF), on
the other hand, is used to refer to components that are repairable.
When lifetime follows an exponential distribution, knowledge of MTBF
or MTTF is adequate to evaluate and predict all measures relating to
reliability.

For an exponential distribution:

1 1
  
MTTF (or MTBF) failure rate (1 )


In summary, if a product’s life distribution can be approximated by the

exponential distribution:
t
1 
Life distribution (pdf) : f (t )  e 

t

Reliability at time t : R(t )  e 

1
Failure Rate : h(t ) 


Mean Lifetime : MTTF (or MTBF) = 

Illustrative Example:
(Adapted from JM Juran and FM Gryna. Quality Planning and Analysis, (1993),
McGraw-Hill, p. 291)

A washing machine requires 30 minutes to clean a load of

clothes. The mean time between failures of the machine is 50 hours.
Assuming a constant failure rate, what is the chance of the machine
completing a cycle without failure? If the reliability of the machine has
to be at least 95%, what should be the recommended MTBF?

Lecture Notes: Reliability page 87

Ans: Since there is a constant failure rate h(t), we can therefore say
that the time to failure follows an exponential distribution. with
parameter MTBF = = 50 hours.

To complete a cycle without failure means failure occurs after 30

minutes. So the probability of no failures within 30 minutes is the
reliability of the machine.

R(t) = e–t/
R(t) = e–(0.5 hr/50 hrs)
= 0.99005

There is a 99 % chance of completing the wash cycle.

To find MTBF= such that R(t) = 0.95, we equate this value with
the reliability formula with t=0.5 hour.

R(t) = e–(0.5 / ) = 0.950 and solve for 

We find the natural logarithm of both sides:

ln [e–(0.5 / )] = ln(0.950)

-(0.5/) = -0.05129329439

=9.7479 hours

We need the washing machine to have a mean time between failure

of as low as 9.75 hours to have a 95% chance of finishing the
laundry in 30 minutes.

5.5 Estimating exponential distribution parameterand its

Failure rate 1/

We may use the following formula to estimate the mean lifetime:

t i
average time to failure t for n samples of failures.
 i 1

n
Where ti= number of hours/cycles, etc at which

Lecture Notes: Reliability page 88

 the ith unit in the sample fails
For i=1,2,3 …n total units in the sample.

Example:

Eight Christmas light bulbs were put on test and their lifetimes
until failures were recorded as follows:
Failure no. 1 2 3 4 5 6 7 8
Failure time 35 53 78 130 135 167 203 253
(hours) 8 5 4 5 2 4 5 9

Estimate the MTTF and failure rate for the next hour of the lights, what
is the probability that another such light would last all 19 days of
continuous use of the holiday season?
n

MTTF= t i
= average of all failure hours = 1322.75 hrs
 i 1

Failure rate = (MTTF)-1 P(last 19 days) = R (19x24=456

hours)
= 1/1322.75 = exp(-456/1322.75)
= 0.000756 failures per hour = 0.7084

5.6Reliability of Complex systems:

D
R(t)=80%

G
R(t)=60%

A B C E
R(t)=95% R(t)=90% R(t)=99% R(t)=75%

H I
R(t)=92% R(t)=97%
Components in series

F
R(t)=85%
H and I in series and
Lecture Notes: Reliability parallel wipage
th G 89

Components in parallel
 A complex system may consist of a large number of components
each with its own reliability value. We may model such a complex
system by a system reliability diagram:

5.6.1 Reliability of Components in Series:

When components are in series, the system works (or is said to be

reliable) only if each one of the components work. When one or more
component fails, then the system fails.

This can be thought of as a pipeline system such that water coming

from the left needs to get to to the right through the components.
When pipes are in series, there is only one passageway for the water,
and any pipe that fails, would stop the flow from left to right (or right to
left, either way.)

Like in components A, B and C in the system diagram on the

previous page: if any one of A, B or C would fail, then this is akin to
the water junction being broken, and hence, water could not flow from
left to right.

The formula for the reliability of n components in series is

Rsys in series =  R = R1xR2x……x Rn

i 1
i

= P(ABC…..Z)
Logical “AND” = “A works and B works and C works and all
others work”

As in the reliability diagram again, the reliability of ABC in series

(and not considering the other components D to I):

RABCsubsystem = RAxRBxRC = (0.95)(0.90)(0.99) =

0.84645

5.6.2 Reliability of Components in Parallel:

When components are in parallel, the system will be reliable if at

least one component works. If all components fail, then and only then
will the system fail.

Lecture Notes: Reliability page 90

 Using the water pipe analogy again, if pipes are in parallel, then
water can flow from left to right if at least one pipeline is working.
Water will not get from source to sink only if all pipes are clogged or
not working.

In the system diagram shown on page 87, we can see that D, E

and F are in parallel. We therefore say that D, E and F are redundant
components, and therefore actually increase the reliability of the
subsystem containing them.

To compute for the Reliability of systems with components in

parallel, we can use the following logical transformations:

Rsystem in parallel = P(At least one of {D or E or F} works)

= 1-P(all fail)
= 1- P(D’E’F’) where P(D’)=1-P(D),
P(E’)=1-P(E), etc.
but since P(D) = RD, P(E)=RE, and P(F)=RF
= 1-[(1-RD)x(1-RE)x(1-RF)]

The formula for the reliability of n components in parallel is

Rsys in parallel = 1   (1  Ri ) =
i 1

1-[(1-R1)(1-R2)….(1-Rn)]
= P(ABC…….Z )
Logical “OR” = “A works or B works or C works or at least
others work”

As in the reliability diagram again, the reliability of DEF in parallel

(and not considering the other components A,B,C nor G,H, I):

RDEFsubsystem = 1-(1-RD)(1-RE)(1-RF) =
= 1-(1-0.80)(1-0.75)(1-0.85) =
0.9925.

Lecture Notes: Reliability page 91

 It should be remarked that when high reliability components are
placed in series, (as in A,B and C whose reliabilities are all greater
than 90%), their serial reliability drops to a number smaller than the
single least reliable component’s reliability value.

But when components that are not as reliable are placed in

parallel, (as in D, E and F whose reliabilities are all no greater than
85%), we end up with a reliability that is higher than the single
highest among the components’ reliabilities.

This tells us that redundant systems (i.e. parallel components)

actually enhance the reliability of a system, while serial systems
actually are less reliable than their individual parts.

If one works out the reliability of the remaining G,H and I

subsystem, one should arrive at a reliability value of 0.95696. (from 1-
{(0.4)[1-(0.92)(0.97)]}. Furthermore, the system reliability of the
whole system of three subsystems in series is 0.8039.

Practice Problems:

1. What is the MTTF of a component whose tested population showed

a constant failure rate and 95% of which failed by 2000 hours?
2. Determine the reliability of a cellphone battery on its 500 th
recharging when it is known that the battery lifetime is normally
distributed lifetime has a mean of 1,000 rechargings and a standard
deviation of 300 recharges.
3. Draw a system reliability diagram of components A to G that follows
the following logic expressions:
a. (ABC)  [(DE)(FG)]
b. (AB)C{D(EF) G}
c. [(AB)C]  (DEF)  G

4. Compute for the system reliability for 3a to 3c given the following

information on Reliability probability: A=98%, B= 95%, C=92%,
D=90%, E=85%, F=80%, G=70%

5. Let’s suppose that there are only three areas that a bicycle could
fail: if the pedal gear couldn’t bite on the chain, if the chain
breaks, and if both brakes (front wheel and rear wheel) fail.
Suppose further that the pedal gear has a lifetime that follows an
exponential distribution with a MTTF of 1500 hours of continuous
use, the chain’s lifetime is normally distributed with mean lifetime

Lecture Notes: Reliability page 92

 until failure of 1400 hours of continuous use and standard deviation
of 500 hours, and each of the two brakes are said to be 90%
reliable at the next trip, but one needs only one brake to function to
have the bicycle work. Model this bicycle’s reliability using a
system reliability diagram and label the reliability of the 4
components: (1) pedal gear, (2) chain and (3) front wheel brake
and (4) rear wheel brake. What is the reliability of the bicycle at the
point where it has been working for a total of 1300 hours?

6. All machines on a shopfloor uses toolbits whose lifetime follows an

exponential distribution with mean life of 400 hours. Your
maintenance policy goes like this: if a toolbit fails, it is promptly
replaced. When the preventive maintenance time period lapses (in
t hours from the time of the last maintenance), then all toolbits are
replaced unilaterally—that is, the toolbits are replaced whether they
are old or new. If you want to replace toolbits before they fail and
are willing to tolerate only a maximum of 20% of failures for all
toolbits, what should be your preventive maintenance time period
t?

7. Your car’s radiator usually uses up all its water (full content at the
beginning) in about 5 hours of continuous motor run about 90% of
the time. From your experience, this water is used up following a
normal distribution with mean time of 4 hours.. When should you
replenish radiator water back to full level such that you will not run
out of water 95% of the time?

Lecture Notes: Reliability page 93