46

CHAPTER – 3

SYSTEM RELIABILITY MODELS AND REDUNDANCY

TECHNIQUES IN SYSTEM DESIGN
Table of Contents

S. No Description Page No.

3.1 Failures and Failure Modes 48

3.2 Causes of failures and unreliability 51

3.3 Reliability of a product from test data 52

3.4 Mean Time To Failure (MTTF) 54

3.5 Time Dependent Hazard Models 56

3.5.1 Field –data curves 56

3.5.2 Constant Hazard Model 57

3.5.3 Linear Hazard Model 58

3.5.4 Non-linear Hazard Model 58

3.5.5 Gamma Model 59

3.5.6 Other Models 60

3.6 Stress-Dependent Hazard Models 61

3.7 Computation of Reliability function using Markov model 62

3.8 System Reliability Models 64

3.8.1 Series Systems 65

3.8.2 Systems with Parallel Components 66

3.8.3 k-out-Of-m Systems 69

3.8.4 Non series – Parallel Systems 72

3.8.5 Systems with mixed-mode failures 77

 47

3.8.6 Fault-Tree Technique 78

3.9 Redundancy Techniques in System Design 81

3.9.1 Component Versus Unit Redundancy 82

3.9.2 Weakest-Link Technique 84

3.9.3 Mixed Redundancy 87

3.9.4 Standby Redundancy 87

3.9.5 Redundancy Optimization 88

3.9.6 Problem Formulation 90

3.9.7 Computational Procedure 91

3.10 Conclusions 92
 48

CHAPTER - 3

“A system is a collection of components, subsystems

and/assemblies arranged to a specific design in order to achieve the

desired functions with acceptable performance and reliability”. The

types of components, their qualities, their quantities and the way in

which they are set inside the system have direct effect on system

reliability. Malfunction of a component or part may lead to the

breakdown of the whole system in few cases and may not in others,

depending upon the functional relationship among the components.

This necessitates a cautious study of component failures and failure

modes and their functions and also their failure models.

3.1 FAILURES AND FAILURE MODES

“Failure is defined as Non-conformance to some defined

performance criterion”. Some products have well defined failures while

others do not. For example, electric bulbs and switches have well

distinct failures. Either they are working or nonworking (failed). Such

products are known as two state products. Some products like

voltage – stabilizers, resistors etc. work in a range. For example, the

output voltage of a stabilizer may lie inside the limits of V - and V +.

When the output voltage crosses these limits only, the device is

supposed to have failed. For evaluating the quantitative reliability of a

device, the concept of failure and their details is to be used.

 49

Several years of knowledge of failure data of different devices

has revealed that based upon the nature of failure, the failures can be

grouped into different kinds. When a large collection of elements are

put into operation, it is possible that there are a huge number of

failures initially which are called initial failures or infant mortality.

These initial failures are due to production defects, such as weak

parts, poor fit, poor insulation, bad assembly etc. This period is

known as the burn-in or debugging period, as the malfunctioning units

are eliminated in the initial failure periods.

“The failures which occur after initial failures, due to the sharp

change in parameters determining the performance of the units, either

as a result of the change in the working stresses or environment

conditions are called random failures or catastrophic failure”. Random

failures are few in long period of operation during which it is difficult

to predict the time of stress occurrence and their amplitude.

With the passage of time, the units get worn out and begin to

weaken. A gradual decrease in the values of the parameters affects

the performance of the product and when the parameters exceed the

limits of tolerance, the product fails. This area is called the wear-out

region where the failure rate increases and the prediction of wear-out

failure is very complicated.

The characteristic curve depicting the above modes of behavior

which if often known as bath-tub curve is shown in the Fig.3.1.

Manufacturers who manufacture high-reliability products subject

their products to an initial burn-in-period in order to reduce

 50

manufacturing defects. The useful life time of the product is the

period t2 – t1. After time ‘t2’, the product is changed with a pre-tested

product.

Infant
Mortality Wear out
Failure Rate

Useful life

T1 Time T2

Fig. 3.1 Bath-tub curve

A curve is plotted with the failure rate on the y-axis and with

the product life on the x-axis. The life can be in cycles, actuations,

years, hours, minutes, or any other units of time or use, which are

quantifiable. Failures among surviving units per time unit is known as

failure rate. From this plot, it can be shown that many products begin

their lives with a higher failure rate due to poor workmanship, poor

quality control of incoming parts, manufacturing defects, etc. and

exhibit a declining failure rate. The failure rate then generally

stabilizes to more or less constant rate where the failures observed are

a chance failure which is known as useful life region. With usage, the
 51

products experience more wear and tear, the failure rate begins to

grow and the products failure begins to occur related to wear-out. The

mortality rate is more during the first year or so of life, then drops to

a low constant level during teenage and early adult life and then rises

as we progress in years in case of human mortality. Infant mortality

occurs in early age is characterized by a decreasing failure rate.

Occurrence of failures during this period is not random in time but

rather the result of a few components with gross defects and the lack

of adequate quality assurance controls in the manufacturing process.

Causes for increase in failure rate /wear out failures are:

 Owing to oxidation, friction wear, shrinkage, corrosion, atomic

migration, breakdown of insulation, fatigue, etc.

 Populations of substandard items owing to microscopic flaws where

the components fail when random fluctuations (transients) of

stress go beyond the item strength.

 Usually related to quality assurance and manufacture, e.g. joints,

connections, welds, cracks, wrap, impurities, dirt, insulation or

coating flaws, incorrect positioning or adjustment.

3.2 CAUSES OF FAILURES AND UNRELIABILITY

The causes for failures of component and equipments in a

system can be many. Some are known and others are unknown due

to the complexity of the system and its environment. A few of them

are listed below:

1. Poor component design or system design

 52

2. Wrong method of production-technique

3. Lack of complete knowledge and skill

4. High complexity of equipment

5. Very poor maintenance policies,

6. Organizational inflexibility and complexity

7. Human errors

8. Failure resulting from environmental factors ,software elements

and human factors

9. Common mode failure in a redundant system where all

replicated units fail by a common factor.

Component hazard data can be obtained in the following two ways:

a) It is not reasonable to ascertain failure rates of products for all

working conditions from part-failure data obtained from either

failure reports of consumers or from life-tests. It is not

impossible through interpolation or extrapolation to forecast

reliability under prescribed stress conditions from the data

available.

b) From the basic failure rates and working stresses applying

stress models.

3.3 RELIABILITY OF A PRODUCT FROM TEST DATA

Let us consider a population comprising of a group of N items

subjected to operation while time t = 0. As time grows, the products

fail. Let us assume that Ns(t) denote the number of surviving products
 53

and Nf(t ) indicate the products that have failed after time t . Each

item independently fails with the probability of failure.

Ft  1  Rt  ….. (3.1)

Where R(t) represents the probability of survival.

Then

N S (t ) N (t )
Rt    1 f ….. (3.2)
N N

“The hazard rate regarded as a measure of instantaneous speed of

failure is defined as

N s (t)  N s (t  t) 1 d N s (t)

Z(t)  lim 
t 0 N s (t) t N s (t) dt ….(3.3)

Differentiating R(t) with respect t,

dR(t ) 1 dN S (t )

dt N dt

substituting in terms of the hazard rate,

dR (t ) N (t ) z (t )
 S   R(t ) z (t )
dt N

Then,

1 dR (t )
z (t )   …(3.4)
R (t ) dt

Integrating on both the sides,

t
 z ( x)dx   logR t 
0

or

t
R(t )  exp    z ( x)dx  ….. (3.5)
 0 

then,
 54

t
F (t )  1  exp    z ( x) dx 
 0 

dF (t )
f (t ) 
dt
t
= z (t ) exp   z ( x)dx 
 0 

 z (t ) R (t )

Therefore,

f (t ) f (t )
z (t )   …. (3.6)
R(t ) 1  F (t )

3.4 MEAN TIME TO FAILURE (MTTF)

The probable time to failure for a non-repairable system is

described as Mean time to failure. “MTTF is an estimate of the

average, or mean time until component’s first failure or disruption in

the operation of the product, process, procedure, or design occurs”.

MTTF assumes that the product cannot resume any of its normal

operations as it cannot be repaired. “It is frequently used to describe

the system or equipment reliability, of use when estimating

maintenance costs and is significant even if there is no constant

failure rate”. MTTF is properly used only for components that can be

repaired and returned to service.

The knowledge about the mean time to failure of a product is

important rather than the complete failure details. MTTF for all the
 55

products which are identical in their design and operate under

identical conditions is assumed to be same. If we have life-tests data

on the population of N items with failure times t1, t2,……..tn then the

MTTF can be mathematically expressed as

1 n
MTTF   ti
N i 1
……. (3.7)

When a component is described by hazard model and its

reliability function, then the MTTF is a function of the random

variable T relating the time to failure of the component. MTTF can be

defined as


MTTF  E[T ]   tf (t )dt .......(3.8)
0

dF (t ) dR (t )
f (t )  
dt dt

Hence, MTTF    tdT (t )
0


 tR (t ) 0   R ( x)dt
0


  R(t )dt .........(3.9)
0

The MTTF can also be evaluated as the Laplace - transform of R(t) i.e.,

 t
MTTF   t ( R)dt  lim  R( x) dx
0 t 0 0

t
However, lim  R( x) dx  lim R( s )
t  0 s 0

Where R(s) is the Laplace transform of R(t)

Thus, MTTF  limS0 R(s) ……... (3.10)

 56

3.5 TIME DEPENDENT HAZARD MODELS

3.5.1 Field –Data Curves

From the failure analysis data, hazard rate of a component can

be computed by using the formula,

N S (t  t )
z (t )  ….. (3.11)
N S (t )t

for various time intervals. A general hazard-rate curve is shown in

Fig.3.2. It is assumed that the time intervals are identical. The time

intervals need not be equal.

()

z(t)

0 20 40 60 80 100
TIME (→)

Fig. 3.2 Hazard-rate curve

When the data happens to be large and the time interval

approaches zero, the piecewise hazard-rate function will tend to be the

continuous hazard-rate function. Using such failure curves, the

failure rate of other identical components operating in identical

conditions can be predicted. But this requires a few mathematical

models which approximately describe the failure behavior. A few such

 57

models which depict various shapes of failure curves will be discussed

in the subsequent sections.

3.5.2 Constant-Hazard Model

This model assumes that hazard rate is constant which does

not significantly increase with component age. When a product is

functioning during its useful lifetime constant-hazard model is

precisely suitable.

Reliability of a constant-hazard model:

R(t) = 1 - F(t) = exp (-λt) ..... (3.12)

Let us consider constant failure rate and an exponential probability

distribution;

Failure rate; f (t) = λ e–λt .. (3.13)

Reliability function, R (t)= e–λt

Then, hazard function,

h (t)= f(t)/R(t) = λe–λt / e–λt …. (3.14)

When the hazard rate is a constant and is equal to the failure

rate, the ‘constant-hazard model’ takes the form

Z (t)= …….(3.15)

Where  is a constant. This attribute is being shown by many

components, particularly, electronic products. For many years, this

model has been used in reliability studies.

An item with constant hazard rate will have the following

reliability and unreliability functions.

f (t) = e - t …..... (3.16)

R (t) = e – t …….. (3.17)

 58

F (t) = 1-e – t …….. (3.18)

The mean time to failure of the item is:

 1
MTTF   e  t dt  ... (3.19)
0 
3.5.3 Linear – Hazard Model

Several components which are under mechanical stresses will

fail due to deterioration or wear-out. The hazard rate increases with

time for such components. The linear-hazard model which is the

simplest time-dependent model has the form

Z (t) = bt, t > 0 ….. (3.20)

where b is a constant


R(t )  exp    bt dt  ….. (3.21)
 0 

 exp(bt 2 / 2)

f (t )  bt exp(bt 2 / 2) ..... (3.22)

where f (t) is a Rayleigh-density function. The portion beyond the

useful period in the bath-tub curve, might follow this model for few

cases. The mean time to failure is given by

 2
MTTF   e  bt /2
dt
0

T (1/ 2) 
  ….. (3.23)
2 b/ 2 2b

3.5.4 Non-Linear Hazard Model

The hazard rate which is not always a linearly increasing

function of time has a more general form of the hazard model:

 59

Z(t) = atb ….. (3.24)

Where a and b are constant

This gives us


R(t )  exp  at b 1 /(b  1)  ….. (3.25)


f (t )  at b exp  at b 1 / b  1)  .….. (3.26)

The above general form is called the Weibull model which

generates a wide range of curves for various values of a and b. This

model represents the constant-hazard model which includes both the

previously discussed models when b=0 and represents linearly

increasing-hazard model when b=1. The parameter b the shape of z(t)

and a affects the amplitude and so they are called shape and scale

parameters respectively.

For the above model,

 1 
 
MTTF   b  1 ….. (3.27)
1 /( b 1)

b  1 a 
 b  1 

dT (d )

(ad ) a

1
Where d 
b 1

3.5.5 Gamma Model

As the gamma is a variable life distribution model, it may offer a

good fit to a few sets of failure data. However, it is not widely used as

a life distribution model for the common failure mechanisms.

The hazard model and it’s associated functions are

 60

t
 a 1 e
z (t ) 
(a  1)!
 t
R(t )
…………. (3.28)

R (t )  
n 1
 t  j

e  t …....... (3.29)
j 0 j!

 a 1
f (t ) 
 a  1!
  t  e  t …..…….. (3.30)

Where  is a positive integer and  is a positive constant

MTTF = a / 

For a > 1, z(t) increases and for a=1, it becomes a constant-hazard

rate model. The above functions are applicable to a component which

is replaced (a-1) times by identical components with practically zero

replacement time. In effect a total of ‘a’ components were used in

sequence to accomplish the task. The case a = 1 represents that there

is only one component.

3.5.6 Other Models

In statistics and probability theory, the exponential distribution

is a set of continuous probability distribution which describes the

time elapsed between trials in a Poisson process, i.e. a process in

which trials occur independently and continuously at a steady average

rate. It may be noted that the exponential distribution is not the

identical to the class of exponential sets of distributions, which is a

large class of probability distributions that includes the exponential

distribution as one of its members, but also includes the Poisson

distribution, binomial-distribution, normal distribution, gamma

distribution, and many others.

 61

There are few more models that are applied sometimes to

describe the failure curves which do not fit in with the earlier

discussed models. The shifter model takes the form

Z (t) =a (t-t0) b where t > t ….. (3.31)

This is used when the initial hazard rate is almost zero for some time.

This model is called the three-parameter Weibull model. This model

roughly combines the constant and linear models to describe a failure

curve that is initially constant and then increases.

The following model also can serve this purpose to a limited extent.

Z (t) = aect ….. (3.32)

Where a and c are constants

For this,

 
R(t )  exp  a / c  e a  1  ….. (3.33)

This model is generally known as the exponential-hazard model.

3.6 STRESS-DEPENDENT HAZARD MODELS

Mostly, the reliability of a component is defined under stated

operating and environmental conditions which implies that any

variation in these situations can affect the failure rate of the

component and thus its reliability. For almost all components the

failure rate is stress-dependent. A component reliability which is

influenced by more than one kind of stress, a power function model of

the form below is used.

h(t )  z(t ) 1a1  2a2 ….. (3.34)

 62

1 and 2 are stress conditions, a1, a2 are positive constants. For

instance, for an electrical item, the stress may be

Operating Voltage
1 
rated voltage

Operating current
2 
rated current

The above models are applied for accelerated testing of products.

Statistics shows that the variable ‘a’ varies from 1 to 8 depending

upon the variety of component. The other factors like quality factors,

application factor, complexity factor, etc will also be used for the

accurate estimation of failure rates.

3.7 COMPUTATION OF RELIABILITY FUNCTION USING MARKOV

MODEL

This method of estimating the reliability of a part can be used if

the hazard rate of the part is known. The part is assumed to exist in

only two states:

State 0: the part is good

State 1: the part is failed

1-z(t) (t) 1

z(t) (t)

State 0 State 1

Fig. 3.3 Markov Graph

 63

If a part is in state 0 at time t, then the probability of

component failure and reach state 1 in the time interval t is z (t) t,

where z(t) = the hazard rate of the component

If P0 (t) = the reliability that the component will not fail in time t;

P1 (t) = the probability that the component will fail in time t;

Then, the probability that the component will stay in state 0 at time t

+ t is given by

P1 (t  t )  z (t )tP0 (t )  P1 (t ) ….. (3.35)

When the component fails, the probability of its remaining in state 1

(failed state) is unity because the component is not repaired. The

probability of the component being in state 1 at time t + t is,

P0 t  t   P0 t 
  z t P0 t  ….. (3.36)
t

The probability equations can be written as

P1 t  t   P1 t 
 z t P0 t  ….. (3.37)
t

As t  0, We have

dP0 t 
  z t P0 t 
dt

The solution of this equation yields

t
log P0 t     z x dx  c1
0

t
P0 t   exp   z x dx  c1 
 0 

t
 c2 exp    z  x dx  ….. (3.38)
 0 
 64

Where c1 and c2 are constants of ingtegration Initially when t=0 the

component is in state 0 and therefore,

P0 0  1  c2

Thus the reliability of the component is

t
Rt   P0 t   exp    z  x dx  ….. (3.39)
 0 

3.8 SYSTEM RELIABILITY MODELS

The main difficult task of a systems engineer is to evaluate

various reliability parameters of the systems he deals with. The

system configuration may differ from simple (consisting of one or two

elements) to complex (involving thousands of elements). One method

for analyzing such systems is to disintegrate them into subsystems of

suitable size, each representing a precise function. Reliabilities of all

the subsystems are evaluated and then combined to find the reliability

of the complete system using certain probability laws. However, this

approach requires total information about the physical structure of

the system and the kind of its functions to evaluatefind the behavior

of the system whenever a subsystem fails.

The task of a system engineer is to evaluate reliability of various

systems. The system configurations vary from a simple one consisting

of one or two components to a complex system consisting of plenty of

components. Such systems can be analyzed by decomposing the

system into subsystems of suitable size, each performing a specific

function. After decomposition of the system, the reliabilities of

 65

subsystems are evaluated and combined to found out the reliability of

the whole system using certain probability laws. This technique

requires a comprehensive knowledge of the physical structure and to

evaluate adequately well to the behavior of the entire system. The

subsystem may consist of one or more components whose reliabilities

are known.

The reliability models for various kinds of subsystems (or

systems) are developed in this section. For all models the assumption

is that each component fails independently of other, i.e. the failure of

any component does not alter the failure of rest of the components.

3.8.1Series Systems

In a series system where components are connected serially, the

failure of one of its components leads to system failure.

Assume a system is having a total of ‘n’ components. When the

functional diagram suggests that for the successful operation of the

system, the accurate operation of all the ‘n' components, then we call

that the system arrangement is a series type. Such systems are

represented as shown in Fig.3.4 for the purpose of reliability

estimation. The information from the IN end will reach the OUT end

only if all the ‘n’ components function properly. Many complex

systems can be reduced to a simple structure.

IN 1 2 n OUT

Fig. 3.4 Series System

 66

Let ‘Ei’ represent the event that component ‘i’ is working state

and‘ ’ be the event that component ‘i’ is in failure state. The

intersection of E1, E2, E3,……, En represents the event of success of

the system.

λ1, λ2, λ3, …… λi,…… λn are the components hazard rates of 1, 2, 3,

…..i,…..n respectively.

Pi(t) be the probability that component ‘i’ is working at time ‘t’

Reliability of a series system,

Rs = the probability that all the units are in good condition

Rs  Pr(E1  E 2  E3   E n )

 Pr(E1 ). Pr(E2 / E1 ). Pr(E3 / E 2 .E1 ) …...(3.40)

 P1 (t) . P2 (t) . P3 (t)  Pn(t)

 e  1t .e  2t .e 3t .....e  n t ......(3.41)

= Product of reliabilities of the components

i.e Reliability of a series system is the product of its component

reliabilities.

3.8.2Systems with Parallel Components

“Parallel system is a system in which components are connected

in parallel and the system does not fail, even one component is in

good working condition i.e the system fails only when all components

have failed”.

Let, the event that component ‘i’ is in good working state be ‘ Ei’.

The event that component ‘i’ is in failure state be ‘ Ei’ the number of
 67

components in the system be ‘ n’ the hazard rates of components 1,

2,3,..i,…..n are λ1, λ2, λ3,… λn respectively.

2
IN OUT

Fig. 3.5 Parallel System

Let the probability that component ‘i’ is functioning at time ‘t’ be

Pi(t) System unreliability,

= Pr( E1 ∩ E2 ∩ E3……∩ En ) ….. (3.42)

= Pr(E1).Pr(E2 /E1).Pr(E3/E2.E1)..Pr(Em)

= P1(t) . P2(t) . P3(t) …… Pn(t)

    
 1  e  1t . 1  e   2t . 1  e  3t ..... 1  e   n t  ….. (3.43)

= Product of unreliability of components.

System reliability,

RP = 1- Unreliability of the system

i.e Reliability of a parallel system =1-the product of its component

unreliability.

By adding one more unit to the existing system which is having

already m units, the system reliability will increases to

Rm 1 (t )  1  [1  p(t )]m 1 ….. (3.44)

and the increase in reliability is given by

 68

Rm 1 (t )  1  [1  p (t )]m 1 ….. (3.45)

 1  p (t )   [1  p (t )] m 1
m

 p(t )[1  p(t )]m

Hence the following recursive formula is available for estimating the

reliability

R1 t   P t 

Rm t   Rm1 t   Rm1 t ,  2 .. (3.46)

If the failure rates are constant


R t   1  1  e    m

The mean time to fail (MTTF) of the system is


MTTF   1  1  e  
0
   dt
m

Putting 1  e   x, we get


1 1 1  x m 
MTTF  dx
 0  1  x 

1 1
 1  x  x 
 .......x m 1 dx
2

 0

1 1 1 1
 1    .............  
 2 3 m

1 m 1
 
 i 1 i
….. (3.47)

For higher values of ‘m’,

1
MTTF  [ In( m)  0.577  1 / 2m] ….. (3.48)



  [exp(1t )  exp(2t )  ........exp(mt )
0
 69

It is observed that the enhancement in the mean life with

redundancy is logarithmic. If the component reliabilities are not equal,

the problem becomes complex.

3.8.3 k-out-Of-m Systems

One of the important practical systems is where more than one

of its parallel components is required to meet the demand. For

example, in a power generating station where there are four

generators, two generators are sufficient to provide the necessary

power to the consumers. The other two generators are added to

improve the supply reliability. One more instance is a four-engine

aircraft where two engines are required for successful operation and

two are kept as standby units. Many such systems are present in

industries and other applications. The models which are developed

for a simple parallel system cannot be applied to these systems.

The binomial distribution can be used to estimate the reliability

for systems which have identical and statistically independent

components. If the probability of survival of each component is p,

then the probability that precisely x out of m components surviving is

given by

P( M , x)  B(m, x) p x (1  p)m  x ….. (3.49)

Where

m
B(m, x )   
x
 70

is the binomial coefficient. If the least number of components

required for operation of the system without any failure is k , then the

system will function if k, k+1, k+2,…. Or m components are

functioning. The system reliability is the total of binomial

probabilities, x varying from k to m, i.e.,

m
R   B(m, i ) p i (1  p ) m i ….. (3.50)
i k

When k=m the system becomes a series-system and when k=1 the

system becomes a parallel system.

In a k-out-of-m system, the reliability of the system is improved

by the addition of m-k units. Such systems sometimes are called

partial-redundant systems to represent that k>1. The k units are

called basic units whose functioning is essential for the success of the

system. If the demand on the system increases, the number of basic

units required will increase and hence the redundant units decrease.

F1 F3

G1 G2

F2 F4

Fig. 3.6 Power distribution system for a chemical

Processing industry

For example, a 4-unit power generating system where each

unit with a capacity of 65 MW and can supply a demand of up to 260

 71

MW. If the capacity is higher than the demand, the system can meet

the demand even if there is a failure of any unit .When there is

demand for less than 130 MW, the system will function as a 2-out-of-

4 system. There is a reserve of another 130 MW. We can safely

increase the demand up to a further 130 MW without installing any

additional unit but increase is done at the cost of reliability. A total

demand of, say 195 M.W would mean that the system works as a 3-

out-of-4 system with a reduced reliability which is given by

R  6 p 2 (1  p)2

There is a corresponding reduction in the mean life of the

1
system. The reduction in case of constant failure rates is T 
2

Using the binomial failure probabilities, the unreliability of a k-out-of-

m system can be stated as

m
 m
Q ( k , m)     (1  p) i p m i ….. (3.51)
i  m  k 1 i 

In case of high reliable components with constant hazard rates

Q(k , m)  B(m, k 1) (t ) mK 1 ….. (3.52)

If we substitute k=1 in the above equation, it reduces to

Q(k , m)  (t ) m ….. (3.53)

If the values of m and k are large, the approximate formula for mean

life is

 1
 m 
1 2
MTTF  ln   ….. (3.54)
  1
k 
 2 
 72

3.8.4 Non Series – Parallel Systems

A complex system when simplified can produce a non – series

parallel configuration. Bridge configuration comes under this type.

The reliability of this type of configuration can be estimated using

appropriate probability rules and logic diagram. A simple non series-

parallel structure which is a bridge configuration is shown in Fig.

3.7. S1, S2 …… are the subsystems and direction of arrows show the

flow process. Such cases can be analyzed by using another approach

called the logic diagram technique.

S3
S1

S5
IN OUT

S2 S4

Fig. 3.7 Bridge Network

The system diagram is converted into a logic diagram that

consists of many simple parallel paths between IN and OUT terminals

by using the logic-diagram approach. The successful functioning of the

system depends on the successful operation of the elements in

different paths. If there is at least one continuous way between IN and

OUT terminals, the system will function successfully. The logic

diagram for the system shown in Fig3.7 is represented in Fig.3.8. The

subsystem S5 is unidirectional like any other subsystems.

 73

S1 S3

S1 S5 S4
IN OUT

S2 S4

Fig. 3.8 Logic Diagram for Fig 3.7

The logic diagram is a plain series-parallel system whose

reliability can be estimated by combining the models developed in

sections 2 and 3 but care has to be exercised to recognize

interdependency of paths. The failure of each subsystem is assumed

to be independent of the failure of the paths is not independent since

some subsystems find place in more than one path. The use of the

following rule takes care of this problem.

If Pr(E1) = p1p2 and pr(E2) = p2p3 then

Pr(E1) * Pr(E2)= p1p2p3 ….. (3.55)

That is pi * pi = pi for any value of i.

Let the reliability of the subsystem, Si= Pi and the path i

reliability be Ri.

Let the reliability of the path i = Ri and the subsystem Si

reliability be Pi and.

R1 = p1p3

R2 = p2p5p4

R3 = p2p4

Then the entire structure reliability is

 74

R=1-(1-R1) (1-R2) (1-R3)

= R1 + R2 + R3 - R1R2 -R2R3 - R1R3 + R1R2R3

On simplification, we get

R=p1p3+p1p5p4+p2p4-p1p3p4p5-p1p2p4p5-p1p2p3p4+p1p2p3p4p5 …….. (3.56)

For p1=p2= ….. =p5=p

R=2p2 +p3 - 3p4 + p5 ………… (3.57)

There are some cases where the subsystem S5 can be a

bidirectional element. For instance, in the systems shown in Fig. 3.9,

the subsystem S5 transmits signals in either direction. The logic

diagram showing the paths for these systems is illustrated in fig. 3.10.

S1 S2 S1 S2

IN OUT
S5
IN S5 OUT

S3 S4 S3 S4

(a) Power distribution system (b) Relay network

Fig. 3.9 Bridge networks

 75

S1 S2

S1 S5 S4
INPUT OUTPUT

S3 S5 S2

S3 S4

Fig. 3.10 Logic diagram of systems in Fig.3.9

The reliability of bridge networks shown in Fig. 3.9 is given by

R=1-(1-p1p2) (1-p3p4) (1-p1p5p4) (1-p3p5p2)

= p1p2 +p3p4+p1p5p4+p3p5p2-p1p2p3p4 - p2p3p4p5 - p3p4p5p1 -p4p5p1p2 -

p5p1p2p3+2p1p2p3p4p5 …....(3.58)

For the case p1 = p for all i

R = 2p2 + 2p3 - 5p4 + 2p5 .…….(3.59)

For a value p=0.8 Eqs. (3.57) yields R=0.890 and Eq. (3.59) yields

R=0.91

If we put p5 = o in Eqs. (3.57) and (3.58) the results would be

R = p1p3+p2p4 - p1p2p3p4 (from Eq. (3.56)

R = p1p2 + p3p4 - p1p2p3p4 .…….(3.60)

The slight difference in the positions of p2 and p3 is due to the

difference in the physical location of S2 and S3.

The condition ps=0 means that the subsystem S5 is no more in

operation and the system configuration is as shown in Fig. 3.11. In

this case the reliability is given by Eq. (3.61)

 76

R=1-(1-p1p2) (1-p3p4) ……..(3.61)

It may be noted that Eqs. (3.36) and (3.37) are identical. Similarly, if

we assume p5=1, Eq. (3.34) yields

R= p1p2 + p3p4 + p1p4 + p3p2 - p2p3p4 - p3p4p1 – p1p2p4 -

p1p2p3 +p1p2p3p4p5 ……..(3.62)

Section 1 Section 2
S1 S2

IN OUT

S3 S4

Fig. 3.11 Bridge network with open link

The assumption of p5=1 implies that the subsystem is replaced

by a direct link of unit reliability and the system can be represented

as shown in Fig.3.12. The system reliability is the product of the

reliabilities of sections 1 and 2.

Further, we know that

R(Section 1) = 1-(1-p1) (1-p3)

=p1+p3 -p1p3

R(Section 2) =1-(1-p2) (1-p4) = p2+p4-p2p4

Section 1 Section 2
S1 S2

IN OUT

S3 S4

Fig. 3.12 Bridge network with a link of unit reliability

 77

then, R=(p1+p3-p1p3) (p2+p4-p2p4)

=p1p2+p2p3+p3p4+p4p1-p1p2p3 - p2p3p4-p3p4p1- p4p1p2 +

p1p2p3p4 ....(3.63)

It may be noted that Eq. (3.63) is the same as that of Eq.(3.62)

for equal Ps, Eq. (3.63) becomes

R =4p2 - 4p3 +p4

= 0.9216 for p 0.8

3.8.5 Systems With Mixed-Mode Failures

Until now it is assumed that when there is an open circuit

relating the OUT and IN terminals of a system, it is considered to have

failed. Such type of failure is known as “open-mode failure”. But

there are some exceptions. For example, an electric appliance will fail

to function when there is an electric short-circuit between OUT and IN

terminals. A relay is declared to be failed if it fails to open when it is

required to open. Such failures are called “short-mode failures”.

When a system possess these type of components, both type of

failures with their failure probabilities should be considered in

estimation of system-reliability.

The reliability of a component subject to both failures, is given

by

p=1-(q0+qs) -----(3.64)

Where q0 and qs are the failure probabilities due to open-mode

and short-mode respectively. Consider a system having two capacitors

which are connected in parallel as shown in Fig. 3.13a. If there is an

open-circuit in one of the capacitors, it does not cause a system to

 78

fail. If there is a short-circuit across any one of them, it will cause the

system to fail. The capacitors are considered to be in series for the

short-circuit type in the reliability sense.

For evaluating probabilistic failures the logic diagrams are

shown in Fig. 3.13(b)

C
C C
C2

(i) Open mode (ii) short mode

(a) System mode (b) logic mode

Fig. 3.13 Systems with mixed mode failures

3.8.6 Fault-Tree Technique

Fault tree analysis technique was developed in the early

1960’s. Since then they have been adopted for a broad variety of

engineering disciplines for performing reliability and safety analysis.

They graphically signify the relationship of failures and other events

within a system. Basic events at the base of the fault tree are

connected via logic symbols (known as gates) to one or more top

events. These peak events represent known hazards or system failure

modes for which reliability is predicted. Crucial events at the base of

the fault tree usually represent human and component faults for

which statistical failures and repair data is available.

Common basic events are

Switch fails closed

 79

Pump failure

Operator does not respond

Temperature controller failure

Fault trees can be used to investigate large and complex

systems. Fault trees are particularly capable of representing and

analyzing redundancy arrangements. Further common cause events

are easily handled.

The fault-tree approach to reliability analysis of complex

systems is based on the events that will cause the system to fail. “A

fault tree (FT) is a diagrammatic representation of all possible fault

events, their logical combinations, and their relationship to the system

failure”. The faults which are at the lowest level of the system are

usually represented at the base of the tree and the system fault at the

top. The events which are at the lowest level are known as “basic

events”. “The events resulting from combinations of basic events

which may or may not be of interest and such events are represented

as intermediate events”. The system failure is found out by combining

the failure probabilities of basic events to obtain the failure

probabilities of intermediate events and lastly the top event.

For simplicity sake, a 3-unit system as shown in Fig. 3.14 is

considered. This system will fail when the component C1 fails, or by

the failure of components C2 and C3 combined, or by the failure of all

the three components. The fault tree diagram of this system is shown

in Fig. 3.15. For combining two or more events the logical OR and

AND gates are used.

 80

C2

C1

C3

Fig. 3.14 A three unit system

An AND gate is a multi-input mechanism which produces an

output signal only when all its inputs function. If Y represents the

output and X1, X2,……. are the input events, then

Y = X1  X2  X3  …….. Xn

An OR gate, on the other hand, produces an output even when one of

its inputs is success. In this case,

Y = X1  X2  X3  ……. Xn

System fails

OR
gate

E1 E1

AND
gate

E2 E3

Fig. 3.15 Fault-tree diagram for the system in Fig. 3.14

 81

3.9 REDUNDANCY TECHNIQUES IN SYSTEM DESIGN

“Redundancy is the provision of alternative means or parallel

paths in a system for accomplishing a given task such that all means

must fail before causing a system failure”. System reliability and

mean life can be increased by additional means by applying

redundancy at various levels. The different approaches of introducing

redundancy in the system are:

1. A duplicate path is provided for the entire system itself which is

known as system or unit redundancy.

2. A redundant path is provided for each component individually

which is called component redundancy.

3. In the third approach, the weak components should be identified

and strengthened for reliability.

4. In the last approach, a mix of the above techniques is used

depending upon the reliability requirements and configuration of

the system which is known as mixed redundancy.

The application of a particular technique depends upon many

factors, for example the weight, size, initial cost and operating

characteristics of components or systems. Particularly in electrical

and electronic systems redundancy use at the component level

introduces certain deviation in operating characteristics of the main

systems. Particular attention should be given to such systems.

Redundancy can either be ‘active’ in which case all redundant

elements operate simultaneously in performing the same function or

 82

‘standby’ in which the duplicate element is switched into service when

a primary element falls.

3.9.1 Component Versus Unit Redundancy

Duplication at the unit level is easy rather than at the

component level. But higher reliability is achieved through component

redundancy than unit redundancy.

Let us consider a two-component series system. Redundancy

can be applied in two ways as shown in Fig. 3.15. Assuming that the

units are statistically independent and identical at each level, the

reliability of the system with unit-redundancy is

Ru=1-(1-p1p2)(1- p1p2)

=2p1p2- p12 p22 ………(3.65)

C1 C2 C1 C2

C2 C1 C2
C1
C

(a) Unit redundancy (b) Component redundancy

Fig. 3.16 Two-element active redundant systems:

Where the reliabilities of components C1 and C2 are p1 and p2. In case

of component redundancy, the reliability is

Rc  [1  (1  p1 ) 2 ] [1  (1  p2 ) 2 ]

= 4 p1 p2  p12 p22  2 p12 p22 ……..(3.66)

If we assume p1 = p2 for simplicity sake , we obtain

Ru  2 p 2  p 4
 83

Rc  p 2 ( 2  p ) 2 ……..(3.67)

Then,

Rc  Ru  p 2 [( 2  p ) 2  ( 2  p 2 )]

= p2(2-4p+2p2)

=2p2(1-P)2 …….. (3.68)

It is obvious from Equation (6.1) that Rc - Ru > 0 for 0 < p < 1 and Rc

- Ru = 0 for p =1 which proves that the redundancy at the component

level is more enhanced than redundancy at the unit level to the level

as far as reliability is concerned. This is also correct even if the

primary components of the system are nonidentical.

This analysis can be extended to a more wider case where the

unit consists of n components in series. Suppose that m-1

components are arranged in parallel at each stage, the system

reliability would be

Rc  [1  (1  p ) m ]n …… (3.70)

In unit redundancy, suppose m-1 units are added across the

primary unit, the reliability is

Ru  1  [1  p n ]m …….. (3.71)

It shows that a series system which has five redundant

components has a greater reliability than a series system which has

three unit redundant components.

Another important redundancy technique is to use partial

redundancy popularly known as k-out-of-m system becomes a series

structure when k=m and a parallel structure when k=1. Consider a

 84

simple 2-out-of-3 system. This system has, physically three

components in parallel and the system is successful as long any two

of them are working. In this case, component redundancy means

duplication of each component.

The reliability of system with unit redundancy or no component

is

R=p2(3-2p)

=0.648 for p=0.6

with unit redundancy this would become

Ru  1  (1  0.648) 2  0.876

It may be noted that component redundancy change the system into a

“2-out-of-6-system” and the net reliability would be

6
6
Rc     p i (1  p) 6i
i 2  i 

 15 p 2 (1  p) 4  20 p 3 (1  p) 3  15 p 4 (1  p) 2  6 p 5 (1  p)  p 6

=0.959 for p=0.6

Again component redundancy is higher than unit redundancy.

3.9.2 Weakest-Link Technique

For a series structure, reliability is at the most the same as the

reliability of the weakest component in the structure. Assume a simple

system having two components A and B in series. Their reliabilities

are 0.8 and 0.4 respectively. This system will have a reliability of 0.32

which is much less than 0.4, the reliability of the weakest component.

The system reliability can be enhanced by one of the following

methods:-
 85

1. Apply redundancy across B only

2. Apply redundancy across A only.

3. Apply redundancy across both A and B individually (component

redundancy).

Various system configurations and their resultant reliabilities

are presented in table 3.1. Types (b) and (c) shows the use of one

additional component to improve the reliability. Their reliabilities

show that the application of redundancy across weaker equipment

results in higher reliability when compared to the redundancy across

stronger equipment. When an attempt is made to increase the

reliabilities of various configurations of a series structure, it is seen

the overall structure reliability is less than the weakest section

reliability. So the pay off will not be much when investment is done

on a section except the weakest one. If the costs of equipments A and

B are same, then the reliability-cost ratio of type (b) is lower than that

of type (c).

Table 3.1 Redundancy Techniques

Type System Configuration System Reliability

 pa  0.8 and pb  0.4 
A B
0.4
Ra  pa pb  0.32
0.8
A

A B 2
B Rb  1  1  pa   pb  0.384
 
A
A B 2
C Rc  pa 1  1  pb    0.512
 
B
B
A
Rd  2 pa pb  pa2 pb2  0.5376
D
A B
A B
Re  pa pb  2  pa  2  pb   0.6282
E
A B
 86

A B 3
R f  pa 1  1  pb    0.6272
 
B
F
B

A B 4
Rg  pa 1  1  pb    0.6963
 
B

G B

A B
Rh  1  1  pa pb  2  pb   1  pa pb 

H
B  0.6674
A B

A B 2 3
Ri  1  1  pa   1  1  pb  
  
B
I A  0.7526
B

A B 3
R j  1  1  pa pb   0.6855
A B
J
A B

A B 2
Rk  1  1  pa pb  2  pb  
B
 0.7618
K A B

A B 3 3
Rl  1  1  pa   1  1  pb  
  
A B
L  0.7777
A B

Now the configurations (c), (f) and (g) are considered in which

the redundancy is provided across the component B only. In (c) the

reliability of B is increased to 0.512 which is still less than the

reliability of A. Therefore, it is advisable to improve the reliability of B

further as shown in (f). It may be noted that the reliability of type (e) is
 87

higher than the reliability of type (f) even though both of them employ

the same number of equipments. In (f) the reliability of B is improved

to 0.904 which is now higher than 0.8. Any further improvement

should be done in A. Compare the reliabilities of types (g) and (i) both

having the same number of equipments. Type (h) is superior to

type (g).

3.9.3 Mixed Redundancy

Component and unit redundancies are easy to design and

simple to implement. But they may not be the best configurations

and there might be scope for further enhancement in their reliability-

cost ratios. A comparison of types (j), (k) and (l) will demonstrate this.

In configuration (k), weak component reliability is enhanced first and

then the unit redundancy is applied. Its reliability is more than the

reliability of the component redundancy shown in (I). The reliability of

(k) can be further increased by providing a link in between two

sections.

3.9.4 Stand by Redundancy

All the components and equipments are not suitable for "active

redundancy”. For example, resistors and capacitors generate design

problems when they are placed actively in parallel. If one of two such

components which are operating in parallel fails, there will be a

change in the circuit constants. Similarly, two electric generators

which are operating at different frequencies cannot be placed together

in parallel. Standby redundancy is used to increase the reliability of

such systems.
 88

The failed component or equipment is replaced automatically or

physically by its "equivalent" In standby redundancy. The reliability of

the operator or sensing and switching mechanism must be considered

for such cases.

Suppose in a system, a switch (contact) activated by a

feedback sensing and control device is used for that replacement of

factory Unit-1 by a standby Unit-2, then the system reliability is given

by

R  p1 p2 pc  (1  p1 ) p2 pc ps  p1 (1  p2 ) p1 pc ……… (3.72)

Where

p1 = Unit-1 reliability

p2 = Unit-2 reliability

pc = switch reliability

ps = sensing and control device reliability

pt = probability that a chance and premature

switching may not occur.

3.9.5 REDUNDANCY OPTIMIZATION

Using redundancy, adequately reliable systems can be

constructed using less reliable elements. By providing parallel

elements at each stage the reliability of a system can be enhanced but

the complexity, weight, size, cost etc. of the system increase

substantially. So it is essential to maintain the cost and weight of the

system to a least and reliability to a highest level when designing a

 89

redundant system. For simple parallel systems, the cost and reliability

of can be expressed as

C = F(c, m, method of redundancy)

R = G (p, m, method of redundancy).

Where ‘p’ and ‘c’ are the reliability and cost and of each element and

‘m’ is the number of parallel elements.

Thus, the optimal redundancy problem is reduced to the

determining the values of m for maximum redundancy for a given ‘C’

or minimum ‘C’ for a given ‘R’.

The redundancy optimization problem can be solved by using a

simple method with cost and volume constraints. The assumptions

made in solving the problem are as follows.

(i) The number of redundant components in each stage is a

continuous variable and the final solution is obtained by

rounding off the solution to integer values.

(ii) The cost, volume/weight of any stage raise linearly with the

number of components in that stage.

(iii) Active redundancy is considered.

(iv) The failure of elements is independent of others

The probability of failure of elements because of short circuit is zero.

 90

3.9.6 Problem Formulation

At least one element in each stage must function for the

successful operation of a series system. If ‘mi’ elements each with

unreliability ‘qi’ are used at stage ‘i’ then the reliability of the entire

system having ‘n’ stages is :

R= =

R= …….(3.73)

and the unreliability is

Q= ……(3.74)

When qimi is small, their product may be neglected.

Then

R=1- ……. (3.75)

and Q= …….(3.76)

The total reliability in a multi-constraint problem is

limited by the following constraints.

j=1,2,……..K .…………….(3.77)

Where ‘Cij’ is the jth resource used at ith stage. The constraints

might represent total cost (C1), total volume (C2), etc. Equation (3.75)

and (3.76) represent the objective functions and the equation (3.77)

represent the constraint function.

Now the problem can be stated as

Find Vector M = (m1, m2…….. mn) which will

Maximize R=

Subject to mi > 1 i = 1, 2, ………. n

Cij > 0 j = 1, 2, ……….. k

 91

 cij mi < Cj

3.9.7 Computational Procedure

Since mi is assumed to be a continuous variable, the maximum

R can be found by differentiating equation (1) with respect to mi and

equality to zero.

From (1) log R =

log R=

Differentiating, we get = (qimi log qi)/(1 - qimi)

= (- qimilog(qi))/ (1 - qimi ) *

(3.78)

Neglecting the products of qm, equation (3.78) can be written as

= - qimi log (qi) i = 1, 2 … n .…. (3.79)

The optimal value of mi = mi* i s the solution of

= - qimi log (qi) = 0

qimi * log (qi) = q2mi *log (qi) = qn mn* log (qi) .….(3.80)

Assuming that m1 * is known, other values m2 *, m3 *…..mn * can be

determined by equation (3.80)

mI*= i=2,3,……..n ….(3.81)

Where E = q1mi*

H = log (q1)

G = log (qi)

The detailed procedure for computing the vector ‘M’ is given below:
 92

Step 1: Choose the initial value m1*

Step 2: Find m1* i = 2, 3, ……….n

Step 3: Check cost and volume constraints. If they are not violated

go to otherwise go to step 5.

Step 4: Give a small positive increment ‘dm’ to m1*, go to step 2

Step 5: Round of mi* such that the constraints are not violated.

Step 6: Calculate the system reliability, cost and volume of this

systems.

3.10 Conclusions

The basic analytical functions in reliability engineering along

with construction of reliability Bathtub curve are presented in this

chapter. Various types of system configuration and their reliability

functions are provided. The problem formulation and computational

procedure for redundancy optimization of series- parallel system with

cost and volume/weight constraints is presented.