You are on page 1of 18

DK598X_book.

fm copy Page 27 Thursday, March 3, 2005 10:47 AM

3
Reliability of System Structures

The next logical step in our construction of mathematical


models of reliability is to enhance the system structure models
by the addition of probability — the second of the attributes
of the definition of reliability. As we do this, we will refer to
the probabilities as reliabilities despite the fact that we have
not yet included all four attributes in our models. Naturally,
we expect that the reliability of a system will be represented
as a function of the reliabilities of its constituent components.

3.1 PROBABILITY ELEMENTS


Keeping in mind the fact that we represent system state as
a binary variable, φ, define the system reliability, Rs, to be the
probability that the system is functioning:

Rs = Pr[φ = 1] (3.1)
Observe that an artifact of the binary definition of the
system state is that the system reliability is also the expected
value of the system state variable:

E[φ] = 1·Pr[φ = 1] + 0 ·P[φ = 0] = Pr[φ = 1] (3.2)


A similar pair of definitions applies to the component
status variables. That is, we let

27
DK598X_book.fm copy Page 28 Thursday, March 3, 2005 10:47 AM

28 Nachlas

ri = Pr[ xi = 1] (3.3)
where, because of the fact that the xi are binary, it is again
the case that the reliability and expected value correspond.
For a system comprised of n components, we take

r = {r1 , r2 , ......, rn }
to be the vector of component reliability values. Given the
defined notation, it is reasonable to expect that the system
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

reliability can be expressed as a function of the component


reliabilities. In terms of general notation, we represent this as

Rs ( r ) = Pr[φ( x) = 1] (3.4)
and we devote the next section to the realizations of this
expression.

3.2 RELIABILITY OF SYSTEM STRUCTURES


The formulation of the system reliability function is often
relatively straightforward, but the general forms must be
constructed carefully. The key issue to consider is whether or
not the components are mutually independent. In this con-
struction, we follow the same four cases that we examined in
the previous chapter.

3.2.1 Series Systems


Based on the form of the system structure function, the gen-
eral statement of the reliability function for a series system is
n

Rs = Pr[φ( x) = 1] = Pr[ ∏ x = 1]
i =1
i (3.5)

Now, in general
n n

Pr[ ∏ x = 1] ≥ ∏ Pr[ x = 1]
i=1
i
i=1
i
DK598X_book.fm copy Page 29 Thursday, March 3, 2005 10:47 AM

Reliability of System Structures 29

and equality holds only when the components are mutually


independent. Thus, for a series system comprised of indepen-
dent components, the system reliability may be stated as
n

Rs = ∏r
i=1
i (3.6)

We should note that, regardless of whether or not the


components are independent, the system reliability function
is an increasing function of the component reliability values
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

and is a decreasing function of the number of components.


In most series systems, the components are independent
with respect to their probabilities of proper function. One
noteworthy class of structures for which the components are
not independent is the set of systems for which the compo-
nents share loads. This is discussed further in Chapter 14.

3.2.2 Parallel Systems


The expression corresponding to Equation 3.5 for a parallel
system is
n n

Rs = Pr[φ( x) = 1] = Pr[ i =1
xi = 1] = Pr[ ∏ (1 − x ) = 0]
i =1
i (3.7)

and it is again the case that


n n

Pr[  x = 1] ≤  Pr[ x = 1]
i=1
i
i=1
i

so a parallel system of independent components has reliability


function
n

Rs = r
i=1
i (3.8)

Examination of this function indicates that the system


reliability function for a parallel system is increasing in both
DK598X_book.fm copy Page 30 Thursday, March 3, 2005 10:47 AM

30 Nachlas

the component reliability values and in the number of com-


ponents.
For system structures other than series and parallel, the
computation of the system reliability from the component
reliabilities is not as straightforward.

3.2.3 k-out-of-n Systems


For the generic k-out-of-n system, there is no compact state-
ment of the reliability function. We can only state that
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

Rs = Pr[ ∑ x ≥ k]
i=1
i (3.9)

and even when the components are independent, there is no


convenient form for this function. The single exception occurs
when the n components are independent and identical (have
the same reliability). In that case, the system reliability is
given by the sum of binomial probabilities for k or more
functioning components:
n
 n
Rs = ∑  j  r (1 − r)
j=k
j n− j
(3.10)

For most other k-out-of-n systems, the use of the mini-


mum-path- and minimum-cut-based methods of the next sec-
tion provide the most effective approach to evaluating system
reliability.

3.2.4 Equivalent Structures


The fact that every system structure has an equivalent rep-
resentation as a combination of series and parallel forms
suggests that we can use the equivalent structures to evaluate
system reliability. While this is true, the fact that the mini-
mum paths and minimum cuts that comprise the equivalent
forms are usually not independent implies that we will be
able to obtain bounds on system reliability rather than exact
system reliability values. The process of using the minimum
DK598X_book.fm copy Page 31 Thursday, March 3, 2005 10:47 AM

Reliability of System Structures 31

paths and minimum cuts is again illustrated by the Whet-


stone bridge shown in Figure 2.3
Before developing the minimum-path- and minimum-
cut-based bounds, let us note that, using the same approach
as we did for the system status, we can determine the system
reliability by complete enumeration. Specifically, we can enu-
merate all possible system states, obtain the probability that
each state occurs, and sum those probabilities that correspond
to system function. For the case in which the components are
independent, the result of the enumeration is shown in Table
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

3.1.
Based on the expressions in Table 3.1, the system reli-
ability is
Rs = r1r2r3r4r5 + r1r2r3r4(1 – r5) + r1r2r3(1 – r4)r5
+ r1r2r(1 – r3)r4r5 + r1r2(1 – r3)r4(1 – r5)
+ r1r2(1 – r3)(1 – r4)r5 + r1(1 – r2)r3r4r5
+ r1(1 – r2)r3r4(1 – r5) + r1(1 – r2)r3(1 – r4)r5
+ r1(1 – r2)(1 – r3)r4r5 + r1(1 – r2)(1 – r3)r4(1 – r5)
+ (1 – r1)r2r3r4r5 + (1 – r1)r2r3r4(1 – r5)
+ (1 – r1)r2r3(1 – r4)r5 + (1 – r1)r2(1 – r3)r4r5
+ (1 – r1)r2(1 – r3)(1 – r4)r5(3.11)
With considerable algebraic effort, this reduces to
Rs = r1r4 + r2r5 + r1r3r5 + r2r3r4 − r1r2r3r4 − r1r2r3r5 − r1r2r4r5
− r1r3r4r5 − r2r3r4r5 + 2r1r2r3r4r5 (3.12)
In addition, for a structure such as the bridge, it would
be reasonable for all five components to be identical and to
have the same reliability. Then,

ri = r ∀i
and the system reliability function reduces to the polynomial
Rs = 2 r2 + 3r3 − 5r4 + 2r5 (3.13)
DK598X_book.fm copy Page 32 Thursday, March 3, 2005 10:47 AM
32
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

Table 3.1 Reliability Values for Paths of the Whetstone Bridge


x φ(x) Pr[φ(x) = 1] x φ(x) Pr[φ(x) = 1]
{1, 1, 1, 1, 1} 1 r1r2r3r4r5 {0, 1, 1, 1, 1} 1 (1 − r1)r2r3r4r5
{1, 1, 1, 1, 0} 1 r1r2r3r4(1 − r5) {0, 1, 1, 1, 0} 1 (1 − r1)r2r3r4(1 − r5)
{1, 1, 1, 0, 1} 1 r1r2r3(1 − r4)r5 {0, 1, 1, 0, 1} 1 (1 − r1)r2r3(1 − r4)r5
{1, 1, 1, 0, 0} 0 r1r2r3r(1 − r4)(1 − r5) {0, 1, 1, 0, 0} 0 (1 − r1)r2r3r(1 − r4)(1 − r5)
{1, 1, 0, 1, 1} 1 r1r2(1 − r3)r4r5 {0, 1, 0, 1, 1} 1 (1 − r1)r2(1 − r3)r4r5
{1, 1, 0, 1, 0} 1 r1r2(1 − r3)r4(1 − r5) {0, 1, 0, 1, 0} 0 (1 − r1)r2(1 − r3)r4(1 − r5)
{1, 1, 0, 0, 1} 1 r1r2(1 − r3)(1 − r4)r5 {0, 1, 0, 0, 1} 1 (1 − r1)r2(1 − r3)(1 − r4)r5
{1, 1, 0, 0, 0} 0 r1r2(1 − r3)(1 − r4)(1 − r5) {0, 1, 0, 0, 0} 0 (1 − r1)r2(1 − r3)(1 − r4)(1 − r5)
{1, 0, 1, 1, 1} 1 r1(1 − r2)r3r4r5 {0, 0, 1, 1, 1} 0 (1 − r1)(1 − r2)r3r4r5
{1, 0, 1, 1, 0} 1 r1(1 − r2)r3r4(1 − r5) {0, 0, 1, 1, 0} 0 (1 − r1)(1 − r2)r3r4(1 − r5)
{1, 0, 1, 0, 1} 1 r1(1 − r2)r3(1 − r4)r5 {0, 0, 1, 0, 1} 0 (1 − r1)(1 − r2)r3(1 − r4)r5
{1, 0, 1, 0, 0} 0 r1(1 − r2)r3(1 − r4)(1 − r5) {0, 0, 1, 0, 0} 0 (1 − r1)(1 − r2)r3(1 − r4) (1 − r5)
{1, 0, 0, 1, 1} 1 r1(1 − r2)(1 − r3)r4r5 {0, 0, 0, 1, 1} 0 (1 − r1)(1 − r2)(1 − r3)r4r5
{1, 0, 0, 1, 0} 1 r1(1 − r2) (1 − r3)r4(1 − r5) {0, 0, 0, 1, 0} 0 (1 − r1)(1 − r2)(1 − r3)r4(1 − r5)
{1, 0, 0, 0, 1} 0 r1(1 − r2) (1 − r3) (1 − r4)r5 {0, 0, 0, 0, 1} 0 (1 − r1)(1 − r2)(1 − r3)(1 − r4)r5
{1, 0, 0, 0, 0} 0 r1(1 − r2)(1 − r3)(1 − r4)(1 − r5) {0, 0, 0, 0, 0} 0 (1 − r1)(1 − r2)(1 − r3)(1 − r4)(1 − r5)

Nachlas
DK598X_book.fm copy Page 33 Thursday, March 3, 2005 10:47 AM

Reliability of System Structures 33

Clearly, system reliability evaluation by this method is rather


demanding and offers plenty of possibility for error.
An alternate approach is to construct bounds on system
reliability. Three sets of bounds on system reliability have
been defined. It is appropriate to compute all three sets and
to combine the information they provide to obtain the nar-
rowest possible interval within which the system reliability
will lie.
The first set of bounds is reasonably obvious and is gen-
erally not very tight. These are the series and parallel bounds.
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

To compute these bounds, we simply treat the system compo-


nents as if the system configuration were a series structure
of independent components and calculate a lower bound, and
we then calculate an upper bound assuming a parallel con-
figuration. Thus,
n n

bs = ∏r ≤ R ≤ r = b
i=1
i s
i=1
i p (3.14)

For a system such as the bridge structure, assuming all


of the components are identical implies that these bounds
reduce to

bs = r n ≤ Rs ≤ 1 − (1 − r) n = bp

Example calculations of these bounds are presented in Table


3.2.
A second set of reliability bounds may be constructed
using minimum paths and minimum cuts. Remember that
the minimum-cut equivalent structure has the minimum cuts
arranged in series, and recall that for a series structure in
general,

 n  n

Pr 


i =1
xi = 1 ≥

∏ Pr  x = 1
i =1
i

Applying this inequality to the minimum-path structures


yields the minimum-cut lower bound on system reliability:
DK598X_book.fm copy Page 34 Thursday, March 3, 2005 10:47 AM

34 Nachlas

bmcl = ∏ Pr κ (x) = 1 ≤ Pr ∏ κ (x) = 1


k k
k k (3.15)
= Pr φ( x) = 1 = Rs

The same reasoning can be applied to the minimum


paths. The minimum-path equivalent structure has the min-
imum paths arranged in a parallel configuration. Therefore,
applying the inequality
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

 n  n

Pr 

i =1
xi = 1 ≤

 Pr  x = 1
i =1
i

to the minimum paths yields the minimum path upper bound


on system reliability:

bmpu =  Pr ρ (x) = 1 ≥ Pr  ρ (x) = 1


j
j
j
j

(3.16)
= Pr φ( x) = 1 = Rs

For the example bridge structure having five identical


components, these bounds are computed as follows:

κ 1 ( x) =  x = 1 − (1 − x )(1 − x ) ,
i∈C1
i 1 2

so Pr κ 1 ( x) = 1 =  r = 1 − (1 − r)
i∈C1
i
2

κ 2 ( x) =  x = 1 − (1 − x )(1 − x ) ,
i∈C2
i 4 5

so Pr[κ 2 ( x) = 1] =  r = 1 − (1 − r)
i∈C2
i
2

κ 3 ( x) =  x = 1 − (1 − x )(1 − x )(1 − x ) ,
i∈C3
i 1 3 5
DK598X_book.fm copy Page 35 Thursday, March 3, 2005 10:47 AM

Reliability of System Structures 35

so Pr[κ 3 ( x) = 1] =  x = 1 − (1 − r)
i∈C3
i
3

κ 4 ( x) =  x = 1 − (1 − x )(1 − x )(1 − x ) ,
i∈C4
i 2 3 4

so Pr[κ 4 ( x) = 1] =  x = 1 − (1 − r)
i∈C4
i
3
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

ρ1 ( x) = ∏x = x x
i∈P1
i 1 4 ,

so Pr[ρ1 ( x) = 1] = ∏r = r
i∈P1
i
2

ρ2 ( x) = ∏x = x x
i∈P2
i 2 5 ,

so Pr[ρ2 ( x) = 1] = ∏r = r
i∈P2
i
2

ρ3 ( x) = ∏x = x x x ,
i∈P3
i 1 3 5

so Pr[ρ3 ( x) = 1] = ∏r = r
i∈P3
i
3

ρ4 ( x) = ∏x = x x x
i∈P4
i 2 3 4 ,

so Pr[ρ4 ( x) = 1] = ∏r = r
i∈P4
i
3

Then,
DK598X_book.fm copy Page 36 Thursday, March 3, 2005 10:47 AM

36 Nachlas

( ) (1 − (1 − r) ) ( )
2 2
bmcl = 1 − (1 − r)2 3
≤ Rs ≤ 1 − (1 − r 2 )2 (1 − r 3 )2

= bmpu

Example calculations of these bounds are also included


in Table 3.2.
The minimum paths and minimum cuts may be used to
define a third set of bounds. These are known as the minimax
bounds. Starting with the minimum paths, recall that the
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

structure function for a parallel system may also be expressed


in terms of a maximum. That is, as stated in Equation 2.5,

φ( x) = max xi
i
{ }
so

Pr[φ( x) = 1] = Pr[max xi = 1]
i
{ }
and for a parallel arrangement of minimum paths,

Pr[φ( x) = 1] = Pr[max ρ j ( x) = 1]
j
{ }
In general, for any set of probabilities

{ }
max Pr[ρ j ( x) = 1] ≤ Pr[max ρ j ( x) = 1]]
j j
{ }
so the minimax lower bound on system reliability is

{
bmml = max Pr[ρ j ( x) = 1] ≤ Rs
j
} (3.17)

Applying the same logic to the minimum cuts, we have


from Equation 2.3

φ( x) = min xi
i
{ }
so
DK598X_book.fm copy Page 37 Thursday, March 3, 2005 10:47 AM

Reliability of System Structures 37

Pr[φ( x) = 1] = Pr[min xi = 1]
i
{ }
and for the parallel configuration of the minimum cuts

Pr[φ( x) = 1] = Pr[min κ k = 1]
k
{ }
In general,

{ }
Pr[min κ k ( x) = 1] ≤ min Pr[κ k ( x) = 1
k k
{ }
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

And the resulting minimax upper bound on system reliability


is

{
Rs ≤ min Pr[κ k ( x) = 1 = bmmu
k
} (3.18)

Therefore, the minimax reliability bounds are

{ }
bmml = max Pr[ρ j ( x) = 1] ≤ Rs ≤ min Pr[κ k ( x) = 1 = bmmu (3.19)
j k
{ }
For the bridge with identical components, the computa-
tion of the minimax bounds proceeds as follows:

{ }
bmml = max Pr[ρ j ( x) = 1] = max r 2 , r 2 , r 3 , r 3 = r 2
j
{ }
and

{
bmmu = min Pr[κ k ( x) = 1
k
}
{
= min 1 − (1 − r)2 ,1 − (1 − r)2 ,1 − (1 − r)3 ,1 − (1 − r)3 }
= 1 − (1 − r)2
So

r 2 ≤ Rs ≤ 1 − (1 − r)2
To illustrate the computation and behavior of the bounds,
we have calculated the values of each of the bounds for the
bridge with identical components for several values of the
DK598X_book.fm copy Page 38 Thursday, March 3, 2005 10:47 AM

38 Nachlas

Table 3.2 Computed Values of the System Reliability


and the Bounds on System Reliability
r bs bmcl bmml Rs bmmu bmpu bp
0.99 0.951 0.999 0.980 0.999 1.000 1.000 1.000
0.95 0.774 0.995 0.903 0.995 0.998 1.000 1.000
0.90 0.591 0.978 0.810 0.979 0.990 0.997 1.000
0.75 0.273 0.852 0.563 0.861 0.938 0.936 0.999
0.60 0.078 0.618 0.360 0.660 0.840 0.748 0.990
0.50 0.031 0.431 0.250 0.500 0.750 0.569 0.969
0.25 0.001 0.064 0.063 0.139 0.438 0.148 0.763
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

0.10 0.000 0.003 0.010 0.022 0.190 0.022 0.410

component reliability. These values as well as the actual system


reliability value (from Equation 3.13) are shown in Table 3.2.
Note that one may not in general assume that one of the
sets of bounds will be tighter than the others. A reasonable
approach to using the bounds is to take the greatest of the
lower bounds and the smallest of the upper bounds to obtain
the tightest possible bounds on system reliability. Doing this
for the entries of Table 3.2 indicates that very satisfactory
bounds are obtained.

3.3 MODULES
The idea that the components that comprise a system may
sometimes be partitioned into modules may be extended to
the calculation of system reliability bounds. There are three
key ways in which this may be pursued. Recall that the
algebraic representation of system state using modules is

φ( x) = φ(ψ ( x))

where

ψ ( x) = {ψ 1 ( x), ψ 2 ( x), ........, ψ m ( x)}

One possible approach to the use of the partition is to


calculate the reliability of each module and to use the result-
ing values as component values in the bounds defined in the
DK598X_book.fm copy Page 39 Thursday, March 3, 2005 10:47 AM

Reliability of System Structures 39

previous section. If the modules contain relatively few com-


ponents or are configured in either a series of parallel struc-
ture, this method will be fairly straightforward.
If, on the other hand, some of the modules are themselves
rather complicated, but the system is designed with the mod-
ules in series, then the series computation applied to each of
the upper and the lower bounds provides a pair of bounds on
system reliability. In this case, the bounds on module reliabil-
ity are obtained using the methods of the previous section.
The third possibility is that one or more of the modules
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

are complicated, and the system configuration of the modules


is not a simple one. In this case, a lower bound on system
reliability can be computed by applying the minimum-cut
lower-bound calculation at the system level to the minimum-
cut lower bounds for the modules.

3.4 RELIABILITY IMPORTANCE


In view of the influence of the reliabilities of the components,
it is reasonable to ask which components have the greatest
(or least) impact on system reliability. One possible reason for
examining this question is to help decide which of the com-
ponents should be improved first. The idea that the effect of
each component is worth considering has led several authors
to define various measures of “reliability importance.” Among
the several forms that have been suggested, the one based on
derivatives seems to us to be the most logical. The formal
definition is
Defn. 3.1: The reliability importance of component i, IR(i),
of a coherent system is the derivative of the system reli-
ability function with respect to the component i reliability.
That is,
d
I R (i) = Rs (3.20)
dri
An appropriate interpretation of this definition is that
the component for which the component reliability imposes
the greatest gradient on the system reliability function is the
most important. Consider some specific example cases.
DK598X_book.fm copy Page 40 Thursday, March 3, 2005 10:47 AM

40 Nachlas

For a series structure comprised of n independent com-


ponents, the realization of expression is
n n

∏r = ∏r
d d
I R ( j) = Rs = i i
drj drj i=1 i=1
i≠ j

Thus, for a series system of independent components, the


reliability importance of any component is equal to product
of the reliabilities of the other components. Since component
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

indices are usually arbitrary, assume the components have


been numbered so that

r1 ≤ r2 ≤ r3 ≤ ........ ≤ rn

Under this indexing, we can see that

I R (1) ≥ I R (2) ≥ .......... ≥ I R ( n)

so that the weakest component has the greatest importance.


For a series system, this seems intuitively reasonable.
For a parallel system comprised of independent compo-
nents, the corresponding analysis is

d  
n n n

I R ( j) =
d
drj
Rs =
d
drj i=1
ri = 1 −
drj  ∏i=1
(1 − ri ) =
 ∏ (1 − r )
i=1
i

i≠ j

Assuming the same ordering of the component indices as


indicated above, the most important component in a parallel
system is the strongest. That is

I R (1) ≤ I R (2) ≤ .......... ≤ I R ( n)

Intuitively, it seems reasonable that the strongest com-


ponent is most important to reliability. The corresponding
point that investments in component reliability improvement
should begin with the most reliable component is less appar-
ent but equally accurate.
DK598X_book.fm copy Page 41 Thursday, March 3, 2005 10:47 AM

Reliability of System Structures 41

Resolving questions of how to enhance system reliability


has been one of the areas in which reliability analysts have
contributed to system design efforts. Reliability importance
measures provide a basis for evaluating the cost effectiveness
of investments in component redesign or other improvement
strategies.

3.5 RELIABILITY ALLOCATION


Another approach to enhancing system reliability is by intro-
ducing redundancy at selected component locations. That is,
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

the system configuration is altered by replacing a single com-


ponent with two or more copies of the component in parallel.
The problem of selecting the components for which this is
done is known as the reliability allocation problem. It is
assumed that each copy of a component included in the system
has a cost. The cost might actually represent the price of the
component or may represent weight or any other consequence
of allocating the component to the design. Then, the problem
of designating the locations and magnitudes of component
redundancy can be stated as an integer mathematical pro-
gram. In fact, there are two plausible algebraic forms for the
optimization problem. We may minimize system cost subject
to a reliability constraint, or we may maximize system reli-
ability subject to a budget constraint.
Regardless of the system configuration, we assume that
we have an expression for the system reliability, Rs. Let mi
represent the number of copies of component i placed in par-
allel at the component i location in the system configuration
and let ci represent the unit cost for component i. Then, one
integer programming representation of the system design
problem is
n

Minimize ∑c mi=1
i i

subject to
Rs ≥ Rtarget
mi ≥ 1 ∀i
mi integer
DK598X_book.fm copy Page 42 Thursday, March 3, 2005 10:47 AM

42 Nachlas

Here, we minimize the cost to obtain a target system


reliability level. The alternate problem is to
Maximize Rs
subject to
n

∑c m ≤ C
i=1
i i budget

mi integer
Naturally, for both of the optimization problems, the alge-
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

braic statement of the system reliability function has terms


of the form
1 − (1 − ri )mi
for the contributions of the reliabilities of the component
positions. Equally clear should be the fact that it is the alge-
braic form of the system reliability function that determines
how difficult it is to solve either of the optimization problems.
A recent paper by Majety, Dawande, and Rajgopal [14] pro-
vides an efficient algorithm for solving the integer programs
in general, and recent work by Rice, Cassady, and Wise [15]
suggests that, because of the relatively small number of fea-
sible solutions, enumeration strategies will often yield solu-
tions efficiently.
A distinction should be made here between reliability
allocation and reliability apportionment. Reliability appor-
tionment is the process of assigning reliability targets to sub-
systems during system design. There are algorithms for doing
this, the most popular of which is called the ARINC method,
which is described by Lloyd and Lipow [16]. Usually, the
algorithms are based upon a balance between the reliability
importance of the subsystems and the cost of enhancing the
reliability of existing subsystems designs. The key difficulty
with the apportionment task is that the assignment process
is driven by subjective criteria, in that enhancement cost is
usually represented by an effort function. The origin of the
effort function is often unclear. For this reason, the algorithm
is not developed here. In contrast, the allocation models
defined above treat the question directly.
DK598X_book.fm copy Page 43 Thursday, March 3, 2005 10:47 AM

Reliability of System Structures 43

3.6 CONCLUSION
The analyses considered to this point provide a means for
relating system reliability to component reliability for many
types of equipment designs. Several exceptions have been
noted, and some of these will be addressed later in the text.
For the system configurations that are based on binary com-
ponent states and independent components, the models and
analyses treated so far are sufficient to permit a reductionist
approach to reliability analysis. That is, for these simplest of
systems, reliability may be studied at the component level
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

because the dependence of system reliability on component


reliability is well defined. For very many system designs, the
ability to focus independently on individual component reli-
ability performance is essential to achieving the high levels
of reliability we now enjoy.

3.7 EXERCISES
1. For the system of Problem 5 of Chapter 2, assume
that r = {0.96, 0.96, 0.96, 0.90, 0.90, 0.94, 0.94). Com-
pute the three types of reliability bounds as well as
actual system reliability.
2. For the system of Problem 6 of Chapter 2, assume
that r1 = r2 = 0.96, r3 = 0.95, r4 = 0.98, r5 = r6 = 0.92,
r7 = 0.93, and r8 = r9 = 0.90. Compute the three types
of reliability bounds as well as actual system reliabil-
ity. Also, compute the lower bound under the modular
decomposition you constructed for the system.
3. Compute the reliability of a 3-out-of-4 system for
which all components have a reliability of 0.85.
4. Compute the reliability of a three-component parallel
system in which all components have a reliability of
0.75.
5. Suggest an alternative reliability importance mea-
sure to the one of Definition 3.1.
6. For the system of Problem 7 of Chapter 2, assume
that r1 = 0.85, r2 = 0.80, r3 = 0.95, r4 = 0.75, and r5 =
0.90. Compute the three types of reliability bounds
as well as actual system reliability.
DK598X_book.fm copy Page 44 Thursday, March 3, 2005 10:47 AM

44 Nachlas

7. For the system of Problem 8 of Chapter 2, assume


that r1 = r2 = 0.90, and r3 = r4 = 0.80. Compute the
three types of reliability bounds as well as actual
system reliability.
Downloaded by [Engineers Australia ] at 01:25 15 May 2014

You might also like