Anda di halaman 1dari 13

Cybernetics and Systems Analysis, Vol. 47, No.

6, November, 2011
FUZZY-ALGORITHMIC RELIABILITY ANALYSIS
OF COMPLEX SYSTEMS
A. P. Rotshtein UDC 681.5.015:007
Abstract. A new approach to system reliability analysis is proposed. This approach combines the
descriptive tools of Glushkovs algorithmic algebra and the quantitative tools of L. Zadehs fuzzy logic.
The rules for transition from operations in an algorithmic algebra to operations with membership
functions of fuzzy sets are obtained. These rules allow evaluating the correctness distribution of the
algorithm execution depending on the values of the measurable parameters.
Keywords: fuzzy reliability of a system, algorithmic algebra, fuzzy logic, correctness distribution,
operator structures, logic structures.
INTRODUCTION
Logic-probabilistic models are widely used in the theory of reliability of complex systems [1, 2]. They allow
calculating a system failure probability based on a Boolean function that relates the event system failure with events
failures of elements.
Complex multifunction systems (a man, a computer, an industrial enterprise, etc.) are sets of objects and processes [3].
In modeling the reliability of operation of such systems, it is necessary to consider two essentially different types of events:
(i) failure onset and (ii) failure detection and emergency maintenance. Boolean algebra, which underlies logic-probabilistic
methods [1, 2] is well suited to describe events of the first type. To model events of the second type, i.e., correction and control
procedures used to enhance the reliability, it is more natural to employ system operation algorithms [46].
V. M. Glushkovs algebra of regular algorithms [7, 8] is a convenient formal tool to describe events related to the
onset, detection, and elimination of failures in a system. It contains operations that generate logical conditions (logical
structures) and operations that generate operators (algorithmic structures). The rules of transition from operations of
algorithmic algebra to their probabilistic analogs, which allow estimating system reliability, are proposed in [9].
Linguistic estimates of uncertainty based on fuzzy logic [10] and theory of possibilities [11] may be an alternative to
statistical methods, which are traditionally used in reliability theory. Failure probability can be formalized with the use of
membership function [10], which depends on the values of parameters that influence the reliability of an element (Fig. 1).
In the paper, we propose an approach to system reliability analysis that combines the descriptive means of Glushkov
algorithmic algebra [7] and the estimation means of Zadeh fuzzy logic [10].
The initial idea underlying the proposed theory is a quantitative estimate of the elements of algorithmic description
(operators and conditions) with the use of membership functions of fuzzy sets [10]. These functions are determined by
experts and specify possibility distributions [11] for correct execution of operators and conditions depending on the
measurable parameters. Each operator and logic structure is associated with a fuzzy knowledge base, which determines the
conditions for the correct execution of this structure. Due to this, the problem of system reliability estimation reduces to a
problem of identification (or a problem of diagnostics) using fuzzy knowledge bases [12, 13], where the object of
identification is the initial logical-algorithmic description; the output of the object is the degree of correctness of the
execution of the algorithm; inputs of the object are measurable variables that determine the correctness of the execution of
operators and conditions with the use of membership functions; the inputsoutput interrelation is a hierarchical system of
fuzzy knowledge bases (a derivation tree), which corresponds to the structure of the regular algorithm.
919
1060-0396/11/4706-0919

2011 Springer Science+Business Media, Inc.


Jerusalem College of Technology, Machon Lev, Israel, rot@jct.ac.il. Translated from Kibernetika i Sistemnyi Analiz,
No. 6, pp. 102115, NovemberDecember 2011. Original article submitted September 28, 2009.
920
1. ALGORITHMIC ALGEBRA
An algebra of regular algorithms [7, 8] is specified by a quadruple of sets
A, B, , W W
1 2
, (1)
where A = ( , , ,... ) A B C is a set of operators; B= ( , , ,... ) a b g is a set of conditions; W
1
is a set of operations that
generate logical conditions (logical structures); and W
2
is a set of operations that generate operators (operator
structures).
Operations from the set W
1
include Boolean operations of conjunction (a a b
1 2
= ), disjunction (a a g
1 2
= ),
negation ( a h = ), and left multiplication of a condition by an operator (Aa e = ). It is assumed that condition can take only
two values: 1 for true and 0 for false.
Operations from the set W
2
include the operation of multiplication of operators A A B
1 2
= (linear structure);
a-disjunction ( )
a
A A C
1 2
= (alternative structure for if-then-else); and a-iteration {
a
A D }= (direct while-do iteration).
According to the regularization theorem [7, 8], an arbitrary operator (algorithm) can be represented in terms of other
operators and conditions with the use of the above-mentioned operations. As the auxiliary operation from the set W
2
, inverse
a-iteration (do-while) { } A G
a
= related to the main a-iteration by the relation { } } A A A
a a
= { can be used.
Applying algebra (1) combines, on a unified logical-algorithmic basis, two essentially different approaches to the
description of events necessary for the system to reliably (correctly) execute the task.
Elemental S-approach, which provides reliability modeling based on the structure of system elements:
series-parallel schemes, fault trees [1, 2]. Subalgebra A, W
1
is used in this case.
Functional F-approach, which provides reliability modeling based on the structure of system functions [46].
Subalgebra A, W
2
is used.
2. FUZZY RELIABILITY OF OPERATORS
This section proposes the concepts of correct (incorrect) execution of operator elements of the reliability model. Two
types of operators are considered, which correspond to the stages of onset and elimination of failures during system
operation: working operator and correction operator.
2.1. A working operator is an element of the process of system operation such that without its correct execution the
problem cannot be solved. During the execution of a working operator, disturbances (errors, defects) are only brought into
the system but are not detected and eliminated.
Depending on the number of measurable parameters, i.e., input variables influencing the correctness of the execution
of a working operator, we will consider two cases: one-parameter and multiparameter.


(
)

()

Fig. 1. Interrelation between probability theory and fuzzy logic
in reliability estimation.
FUZZY LOGIC
Membership
function
Parameter
(failure
cause)
Failure
possibility
PROBABILITY THEORY
Data
processing
Failure
probability
(failure
consequence)
The values of measurable parameters are assumed to depend on time, i.e., in the one-parameter case, x x t = ( ), and in
the multiparameter case x x t
i i
= ( ), i n =1 2 , ,... , . In each of these cases, the output of the operator is a linguistic variable L,
which is regarded as the correctness of the execution of an operator A and is estimated by two fuzzy terms: A
1
for correct
execution of the operator A and A
0
for incorrect execution of the operator A.
One-Parameter Case. Following [10], fuzzy terms A
1
and A
0
are defined as follows:
A x x
A
V
1 1
=

m ( ) / , A x x
A
V
0 0
=

m ( ) / ,
where m
A
x
1
( ) and m
A
x
0
( ) are membership functions whereby variable x belongs to the terms A
1
correct and
A
0
incorrect, m m
A A
x x
0 1
1 ( ) ( ) = - , and V is a universal set on which the measurable parameter x is defined.
Following [11], the membership functions used here can be regarded as the distribution of the possibility of correct
and incorrect executions of the operator A.
Example 1. Table 1 presents the membership functions that correspond to the distributions of the correct (incorrect)
execution of four operations described by a working operator. These functions are set up under the following assumptions.
For the first two operations (transmitting a message and passing an exam), there exist some threshold values of the parameter
(a
1
and a
2
), below and above of which the possibility of correct execution of the operations is zero. For two other operations
(driving a car and etching a printed circuit board), there exists a value of the parameter (b
0
), for which the correctness of the
execution of the operation is equal to unity, and values of the parameter (b
1
and b
2
) that determine zero possibility of the
correct execution of the operation.
Multiparameter Case. In this case, the correctness of the execution of the operator A depends on the vector of
variables X = ( , ,... , ) x x x
n 1 2
. This dependence is assumed to be defined by an expert system of fuzzy rules:
IF x a x a x a
n n 1 11 2 21 1
= = = AND AND
OR x a
1 12
= AND x a
2 22
= AND x a
n n
=
2

OR x a
m 1 1
= AND x a
m 2 2
= AND x a
n nm
=
THEN L A =
1
,
921
TABLE 1. Examples of Membership Functions
Operation Parameter
Membership function of the correct (A
1
) and
incorrect (A
0
) execution of operations
Message transmission
Number of symbols
Passing an exam Points
Driving a car
on a road
Distance to sidewalk
Etching a printed
circuit board
Time of etching
x 0
1
0 1
m
a
2
a
1
x 0
1
0 1
m
b
2
b
1 b
0
where a
ij
is a fuzzy term (low, below average, average, ) for the estimate of the variable x
i
(i n =1 2 , ,... , ) in the jth
conjunction row ( j m =1 2 , ,... , ).
Then the multidimensional membership function of vector X x x x
n
= ( , ,... , )
1 2
in the terms A
1
correct and
A
0
incorrect is determined using the fuzzy logic equations [12, 13]:
m m
A
n
j
m
i
n
ij i
x x x x
1
1 2
1 1
( , ,... , ) ( ) =
= =
, m m
A
n
A
n
x x x x x x
0
1 2
1
1 2
1 ( , ,... , ) ( , ,... , ) = - ,
where m
ij i
x ( ) is the membership function of the variable x
i
in the fuzzy term a
ij
(i n =1 2 , ,... , ; j m =1 2 , ,... , ).
2.2. Correction Operator R is introduced into the algorithm to eliminate the errors in the execution of the working
operator A.
Let m
A
x
1
( ) and m
R
x
1
( ) be the distributions of the correctness of the execution of the operators A and R dependent on
the parameter x. Let m m
R A
x x
1 1
( ) ( ) " x V, which corresponds to the extension of the correctness distribution function of
the working operator. Then the function m
R
x
1
( ) can be defined by
m m
R A
r
x x
A
1 1
1 1 ( ) [ ( )] = - - ,
(2)
where r
A
is the coefficient of extension of the function m
A
x
1
( ) corresponding to the correctness quality.
For practical calculations, we will use the following values: if the correctness quality is low, then r
A
1; if the
correctness quality is below average or average, then r
A
3 and r
A
5, respectively; if the correctness quality is above
average or high, then r
A
7 and r
A
9, respectively.
If, for example, the correctness of the working operator is m
A
x
1
0 5 ( ) . = , then the correctness of the execution of the
correctness operator m
R
x
1
( ) for different levels r
A
is determined as follows: if r
A
=1, then m
R
x
1
0 5 ( ) . = ; if r
A
= 3 , then
m
R
x
1
0 875 ( ) . = ; if r
A
= 5 , then m
R
x
1
0 967 ( ) . = ; if r
A
= 7 , then m
R
x
1
0 992 ( ) . = ; and if r
A
= 9 , then m
R
x
1
0 9998 ( ) . = .
3. FUZZY RELIABILITY OF LOGICAL CONDITIONS
In modeling the reliability of a system as a collection of objects and processes, two types of logic functions should be
distinguished: external and internal. External conditions are related to performance check of objects, for example hardware.
Internal conditions correspond to the correctness check of the execution of elements of the process of system operation, i.e.,
operators of the algorithm.
Whether an external condition is true depends on the logic function a a a
a
= f ( , ,... )
1 2
, where a
i
is serviceability
condition for the ith object included in the system. The truth of an internal condition depends on the correct execution of the
section of the algorithm that is controlled by this condition.
A logic function a is formally defined as a mapping X
a
1 0 ( ), where 1 is true; 0 is false; X
a
is the vector of
parameters that are checked when condition a is checked. The following outcomes are possible:
a
1
(a
0
) : condition a is objectively true (false);
a
11
(a
10
): an objectively true condition a is recognized true (false) after a check;
a
00
(a
01
): an objectively false condition a is recognized false (true) after a check.
Cases 10 and 01 pertain to errors of the first and second kinds, which are associated with false alarm and
missed rejects, respectively.
Since the boundary between true (1) and false (0) is fuzzy, these outcomes should be considered fuzzy terms, which
can be formalized by membership functions. If the truth of the condition a depends on a set of parameters X
a
, then the
multidimensional distribution of the truth can be obtained with the use of an expert knowledge base similarly to the case of
multipleparameter operator.
Let us consider the one-parameter case, where the truth of the condition a is determined by a unique parameter, i.e.,
X
a
= x V. Then the fuzzy outcomes of the condition a can be presented as
922
a m
a
1 1
=

( ) / x x
V
, a m
a
0 0
=

( ) / x x
V
,
a m a m
a a
11 11 10 10
= =

( ) / , ( ) / x x x x
V V
,
a m a m
a a
00 00 01 01
= =

( ) / , ( ) / x x x x
V V
,
where m
a
1
( ) x [ ( )] m
a
0
x is the distribution of the truth (falsity) of the condition a; m
a
11
( ) x [ ( )] m
a
10
x is the distribution of
the possibility of executing the condition a without an error of the first kind (with an error of the first kind); m
a
00
( ) x
[ ( )] m
a
01
x is the distribution of the possibility of executing the condition a without an error of the second kind (with an
error of the second kind), and
m m
a a
1 0
1 ( ) ( ) x x = - , m m
a a
10 11
1 ( ) ( ) x x = - , m m
a a
01 00
1 ( ) ( ) x x = - " x V.
Errors of the first and second kind occur since the real values of the truth and falsity of the conditions a are
underestimated, i.e.,
m m
a a
11 1
( ) ( ) x x , m m
a a
00 0
( ) ( ) x x " x V.
If checking the condition a does not involve errors of the first and second kind, there is the equality sign in these
relations. As the possibility of errors of the first and second kind increases, the inequalities amplify. Therefore, it is
possible to use the operation of the concentration of fuzzy sets [10] and to present functions m
a
11
( ) x and m
a
00
( ) x as
m m
a a
a
11 1
1
( ) [ ( )] x x
k
= , m m
a a
a
00 0
0
( ) [ ( )] x x
k
= ,
(3)
where k
a
1
(k
a
0
) is the coefficient of contraction of the truth (falsity) distribution function of the condition a.
If condition a is regarded as the equipment serviceability control (external condition) or control of the correct
fulfillment of operators of the algorithm (internal condition), then coefficients k
a
1
and k
a
0
reflect the predisposition of the
control to errors of the first and second kinds, respectively.
For modeling based on linguistic estimate, we may say that if the possibility of an error of the first (second) kind is
absent, then k
a
1
( ) k
a
0
1 ; if the possibility of an error of the first (second) kind is below average or average, then k
a
1
( ) k
a
0
2 or k
a
1
( ) k
a
0
3 ; if the possibility of an error of the first (second) kind is above average or high, then k
a
1
( ) k
a
0
4
or k
a
1
( ) k
a
0
5 .
If control is carried out by a human, errors of the first kind (false alarm or rejection of a good result) may be related to
the degree of objectivitybias of the checker, and errors of the second kind (missed rejects) to the degree of
vigilancenegligence of the checker. In this case:
k
a
1
1 = if the checker is very objective;
k
a
1
5 = if the checker is very biased;
k
a
0
1 = if the checker is very vigilant;
k
a
0
5 = if the checker is very negligent.
Example 2. Let us consider a learning and knowledge control process, where is a working operator corresponding
to the learning stage; a is a logic condition corresponding to the control stage.
We assume that m m
a
1 1
( ) ( ) x x
A
= , where x is a parameter that can be regarded as the percentage of correctly executed
tasks. Figures 24 show the distributions of the possibility of correct and incorrect execution of the operator and condition a.
Let x = 85. Then from Fig. 2 we obtain the possibility of correct and incorrect execution of the learning stage:
m
A
1
85 0 5 ( ) . = , m
A
0
85 0 5 ( ) . = .
923
The possibility of knowledge estimation with an error of the first kind (recognizing the estimate x = 85 to be
insufficient to pass the test) is
m m
a a
10 11
85 1 85 ( ) ( ) = -
=
= 0 5 1
0 75
1
. (
.
for the checker is very objective),
instea
k
a
d of for somewhat biased examiner),
i
05 2
0875
1
. (
.
k
a
=
nstead of for considerably biased examine 05 3
1
. ( k
a
= r),

i.e., as the bias of the examiner increases (or the objectivity decreases), the possibility of an error of the first kind
increases.
The possibility of knowledge estimation with an error of the second kind (recognizing the estimate x = 85 to be
sufficient to pass the test) is
m m
a a
01 00
85 1 85 ( ) ( ) = -
924
Fig. 2. The distribution of the possibility of correct (a) and incorrect (b) learning.
80 85 90
x
0.0
0.5
1.0
80 85 90
x
0.0
0.5
1.0
a b
m
A
1
m
A
0
80 85 90
x 0.0
0.5
1.0
80 85 90
x
0.0
0.5
1.0
Fig. 3. The distribution of the possibility of knowledge control without errors of
the first (a) and second (b) kind
a b
m
A
11
m
A
00
k
a
1
1 =
2
k
a
0
1 =
3
3
2
Fig. 4. The distribution of the possibility of knowledge control with errors of the
first (a) and second (b) kind
80 85 90
x
0.0
0.5
1.0
80 85 90
x
0.0
0.5
1.0
a
m
A
10
m
A
01
b
3
2
k
a
1
1 = k
a
0
1 =
2
3
=
= 05 1
075
0
. ( ),
.
for vigilant and strict examiner
istea
k
a
d of for lenient examiner),
istead of f
05 2
0 875 05
0
. (
. .
k
a
=
or very lenient examiner), k
a
0
3 =

(
i.e., as the vigilance of the examiner decreases (the lenience increases), the possibility of referring the estimate x = 85 to
the knowledge level sufficient to pass the test increases.
4. FUZZY RELIABILITY OF OPERATOR STRUCTURES
Operator structures are formed by operations of algorithmic algebra that generate operators. To set up models of
fuzzy reliability, we will use the concept of the graph of possibility. This graph reflects the logic of the events that lead to
correct (incorrect) execution of the structure and is an analog of fuzzy knowledge base (set of IF-THEN rules [12, 13]).
Nodes of the graph correspond to the events of the beginning and the end of elements of the structure (operators and
conditions), and arcs are weighted by the distributions of correct (incorrect) execution of these elements. In each graph, there
are a unique initial node and two absorbing nodes, which reflect the end of the execution of the structure with positive and
negative outcomes. The distribution of the possibility of correct (incorrect) execution of the structure is determined by all the
paths on the graph that lead from the initial node to a node with positive (negative) result.
4.1. Linear F
B
-Structure. This structure (Fig. 5a) is associated with a record
B A A =
1 2
, (4)
where A
1
and A
2
are working operators depending on parameters x
1
and x
2
; B is an equivalent operator.
The events occurring when S
B
-structure is executed are determined by the graph of possibilities (Fig. 5a), where 3 is
the beginning of the execution of the operator A
1
; 4 (5) is the end of correct (incorrect) execution of the operator A
1
and
beginning of the execution of the operator A
2
; 1 (2) is the end of correct (incorrect) execution of the structure.
The arcs of the graph are weighted by the distributions of correct and incorrect execution of the operators A
1
and A
2
dependent on the parameters x
1
and x
2
.
The graph in Fig. 5 is associated with fuzzy logic equations that relate the possibilities of correct and incorrect
execution of the operator B with similar possibilities for the operators A
1
and A
2
:
m m m
m m m
B
B
x x x x
x x x x
1
1 2
1
1
1
2
0
1 2
1
1
0
2
( , ) ( ) ( ),
( , ) [ ( ) ( )
=
= ] ( ).

m
0
1
x
(5)
In the case of a sequence of n operators B A A A
n
=
1 2
... , the models the correctness (incorrectness) estimate have the
form
m m m m
B
n n
x x x x x x
1
1 2
1
1
1
2
1
( , ,... , ) ( ) ( ) ( ) = K ,
m m
B
n
B
n
x x x x x x
0
1 2
1
1 2
1 ( , ,... , ) ( , ,... , ) = - .
4.2. Alternative F
C
-Structure. In the general case, this structure (Fig. 6, a) is specified by the formula
C A A
1 1 2
= ( )
a
,
where C A
1 1
= if a =1 and C A
1 2
= if a = 0; A
1
and A
2
are arbitrary operators.
In reliability modeling, of greatest interest is a special case that corresponds to the algorithm controlcorrectness
without a feedback (Fig. 6b):
C E U = ( )
w
, (6)
where w is a condition checked during the control and depended on the parameter x; E is an identical operator
corresponding to fixing the results of the control; U is correction operator; C is an equivalent operator.
925
Events that occur during the execution of an alternative structure are determined on the graph of possibilities
(Fig. 6c), where 3 is a dummy input node, 5 (6) is the beginning of the execution of the structure provided that w =1(w = 0),
4 is the beginning of the execution of the operator U, 1 (2) is the end of the correct (incorrect) execution of the structure.
The graph in Fig. 6 yields fuzzy logic equations for estimating the possibility of correct and incorrect execution of the
operator C :
m m m m m m m
w w w w w w
C
x x x x x x
1 1 11 1 10 0
( ) ( ) ( ) [ ( ) ( )] [ ( ) = {
00 1
0 1
1
( )] ( ),
( ) ( ).
x x
x x
U
C C
}
= -

m
m m
(7)
4.3. Iterative F
D
-Structure. This structure (Fig. 7a) can be written as
D R = { }
n
, (8)
where n is a logic function dependent on the parameter x; R is correction operator; and D is an equivalent operator.
Iterative structure can be regarded as an algorithm equipment serviceability check (n) repair (R) with a feedback.
The operator R is assumed to be executed until the condition n becomes true.
926
Fig. 5. Linear structure: F
B
-structure (a);
S
B
-structure (b).
B
A
1
A
2
5
3
4
2 1
1
m
1
1
( ) x
a b
m
1
2
( ) x
m
0
1
( ) x
m
0
2
( ) x
0
w
U
1
C
a
0
1
C
1
A
1
A
2
4
3
5
2 1
6
Fig. 6. Alternative F
C
-structure.
a
b c
m
w
1
( ) x
m
w
0
( ) x
m
w
10
( ) x
m
w
00
( ) x
m
w
11
( ) x
m
U
x
1
( ) m
U
x
0
( )
m
w
01
( ) x
0
R
4
3
5
2 1
1
n
D
6
3
1 2
4 a
1
c
1
b
1
b
a
c
Fig. 7. Iterative F
D
-structure.

m
n
1
( ) x
m
n
0
( ) x
m
n
10
( ) x m
n
00
( ) x
m
n
11
( ) x
m
R
x
1
( ) m
R
x
0
( )
m
n
01
( ) x
The contents of the nodes and weights of arcs on the graph of possibilities (Fig. 7b) completely coincides with the
alternative structure (see Fig. 6c) provided that R U = and n w = . However, the iterative structure generates cycles 454
and 464 on the graph of possibilities (Fig. 7b).
Figure 7 shows an equivalent graph of possibilities obtained from the original graph (Fig. 7b) by deleting arcs 5
and 6. The weights of arcs of the equivalent graph are as follows:
a x x = m m
n n
1 11
( ) ( ), c x x = m m
n n
0 01
( ) ( ),
b x x x x = [ ( ) ( )] [ ( ) ( )] m m m m
n n n n
1 10 0 00
,
a x x
R
1
1 11
= m m
n
( ) ( ), c x x
R
1
0 01
= m m
n
( ) ( ),
b x x x x
R R
1
1 10 0 00
= [ ( ) ( )] [ ( ) ( )] m m m m
n n
.
The tree into which the graph in Fig. 7 is transformed yields fuzzy logic equations for the estimate of the possibilities
of correct and incorrect execution of the iterative structure
m
m
D
D
x a b a b b a b b b a
x c
1
1 1 1 1 1 1
0
( ) [ ] [ ] [ ]
( ) [
=
=
K
b c b b c b b b c

1 1 1 1 1 1
] [ ] [ ] K
. (9)
5. FUZZY RELIABILITY OF LOGIC STRUCTURES
Logic structures are formed by operations of algorithmic algebra that generate logic conditions. The following logic
structures are of greatest interest in reliability modeling:
b a a =
1 2
(S

-structure),
g a a =
1 2
(S

-structure),
which correspond to series ( ) S

and parallel ( ) S

connection of elements.
Let y
i
be a measurable parameter that determines the reliability of the ith element. Then fuzzy logic equations for the
estimate of the possibilities of correct and incorrect execution of logic structures have the form
m m m
b a a
1
1 2
1
1
1
2
1 2
( , ) ( ) ( ) y y y y = ; m m
b b
0
1 2
1
1 2
1 ( , ) ( , ) y y y y = - ,
m m m
g
a a
1
1 2
1
1
1
2
1 2
( , ) ( ) ( ) y y y y = ; m m
g g
0
1 2
1
1 2
1 ( , ) ( , ) y y y y = - . (10)
The passage from an arbitrary Boolean function to the distribution of its correct execution can involve the
above-mentioned relations and orthogonalization methods outlined in [1].
6. CALCULATIONS
Applying the fuzzy logic equations obtained in Secs. 5 and 6 provides the replacement of operations and with the
respective arithmetic operations. The following arithmetization rules are most popular in fuzzy set theory [10, 11]:
m m
m m
m m
1 2
1 2
1 2
=

min ( , ) , (standard intersection)


(algebraic product);

m m
m m
m m m m
1 2
1 2
1 2 1 2
=
+ -
max ( , ) (standard union) ,
(algebraic sum).

If standard intersection ( min) = is applied in iterative structures controlcorrection with a feedback we have
b b b b b b b
1 1 1 1 1 1 1
= = K min ( , ,... , ) .
927
This does not allow modeling the growth of the possibility of correct execution of the iterative F
D
-structure with the increase
in the number of control cycles; therefore, it is expedient to use algebraic products and sum in fuzzy logic equations.
Since only exclusive AND ( ) is used in graphs of possibilities (Figs. 57) and in the fuzzy logic equations
corresponding to these graphs, the replacement m m m m
1 2 1 2
= + holds for the algebraic sum.
Relevant transformations yield the following transition rules to pass from operations in algorithmic algebra [7] to
respective operations over membership functions [10], which characterize the fuzzy correctness of the execution of operators
and conditions:
B A A x x x x
B A A
= =
1 2
1
1 2
1
1
1
2
1 2
m m m ( , ) ( ) ( );
C E U x x x x x
C
= = + + ( ) ( ) ( ) ( ) [ ( ) ( )] [
w
w w w w
m m m m m
1 1 11 1 10
{ m m m
w w
0 00 1
( ) ( )] ( ) x x x
U
} ;
D R
x a ba
b
a x x a x
D
R =
= +
-
= =
{ }
( ) ,
( ) ( ), (
n
n n
m
m m m
1
1
1
1 11
1
1
1
1
) ( ),
( ) ( ) ( ) ( ),
( )
m
m m m m
m
n
n n n n
11
1 10 0 00
1
1
x
b x x x x
b x
R
= +
= m m m
n n
10 0 00
( ) ( ) ( ), x x x
R
+

b a a m m m
b a a
= =
1 2
1
1 2
1
1
1
2
1 2
( , ) ( ) ( ) y y y y ;
g a a m m
g
a a
= = - - -
1 2
1
1 2
1
1
1
2
1 1 1
1 2
( , ) [ ( )][ ( )] y y y m y .
The formulas for iterative structure are obtained on the assumption that there are no constraints on the feasible
number of cycles.
The successive applications of these relations allow replacing the initial algorithm with a unique operator whose
membership function characterizes the fuzzy reliability (correctness) of the task execution depending on the values of
parameters of the operators and conditions.
7. EXAMPLE OF RELIABILITY CALCULATION
Let us consider a logic-algorithmic description of events related to the onset, detection, and elimination of failures
during the execution of some function (operation, work) (Fig. 8). To this description there corresponds an algebraic
expression
F R A E U = { }
n w
0
( ), (11)
where n
0
1 0 = ( ) if the necessary equipment is serviceable (nonserviceable); R is repair of the equipment; A is
execution of the function; w =1 0 ( ) if the function is executed correctly (incorrectly); E is an identical operator
corresponding to fixing the results of control w and U is error correction.
The equipment necessary to execute the function is assumed to have parallel-series structure
n n n n n
0 1 2 3 4
= ( ) ( ) , (12)
where n
i
is serviceability condition for the ith element of the equipment, i =1 4 ,... , .
Formulas (11) and (12) are associated with the tree in Fig. 9, which determines the order of applying operator ( F-)
and logical (S-) structures in reliability calculation.
To carry out the calculation, it is necessary to specify the following membership functions: m
A
x
1
( ) is the correctness
distribution of the execution of the operator A depending on the parameter x and m
n
i
y
i
1
( ) is the distribution of nonfailure
operation of the ith element of the equipment depending on the parameter y
i
, i =1 4 ,... , .
For simplicity, we assume that the distribution of nonfailure operation is the same for all the elements of the
equipment, i.e., m m
n
n
i
y y
i
1 1
( ) ( ) = , i =1 4 ,... , .
928
The necessary distributions are represented in Figs. 10 and 11. The calculation was carried out for three variants of the
initial data specified in Table 2. Variant 1 corresponds to the best case where errors of the first and second kind are absent, and
repair and correction are performed with high quality. Variants 2 and 3 reflect the average and worst cases, respectively.
929
A
U w
0
1
1
0
R
n
0
1
2
3
4
n
1
n
2
n
3
n
4
Fig. 8. Logic-algorithmic model.
A U
w
R
n
1
n
2
n
3 n
4
F
C
F
B
F
D
S

Fig. 9. Reliability tree.


50 75 100 x
0
1
Fig. 10. Correctness distribution of the
execution of the operator A.
80 90 y
0
1
Fig. 11. The distribution of nonfailure
operation of an element
of the equipment.
m
A
x
1
( )
m
n
1
( ) y
TABLE 2. Initial Data for the Numerical Example
Element of
the algorithm
Parameter
Variants of calculations
1
2 3
n
m y
n
1
( )
Type I errors
Absent,
k
n
1
1 =
Average level, k
n
1
3 = High level, k
n
1
5 =
Type II errors
Absent,
k
n
0
1 =
Average level, k
n
0
5 = High level, k
n
0
9 =
R
Repair quality High, r
n
0
9 =
Average, r
n
0
5 = Low, r
n
0
1 =
A m
A
x
1
( )
w
Type I errors
Absent,
k
A
1
1 =
Average level, k
A
1
3 =
High level, k
A
1
5 =
Type II errors
Absent,
k
A
0
1 =
Average level, k
A
0
5 = High level, k
A
0
9 =
U
Correction
quality
High, r
A
= 9 Average, r
A
= 5 Low, r
A
= 1
For each of the variants, three-dimensional function m
F
x y
1
( , ) is calculated, which characterizes the distribution of
the correctness of the execution of algorithm (11) depending on the parameters x and y. The calculation results are presented
in Fig. 12. They show that as the quality of check (n and w) and correction (R and U) decreases, the membership function
m
F
x y
1
( , ) is concentrated (compressed). This means that the domain of feasible values of parameters (x, y), for which the
required level of correctness of the algorithm execution is achieved, becomes narrower.
The example illustrates how the proposed method can model the reliability of the system based on its
logic-algorithmic description and available expert-linguistic information about the correct execution of operators and
conditions.
Note that the well-known methods of reliability modeling based on probability models [1, 2, 4, 6] do not allow
tracing the influence of the parameters of elements of the system on its reliability. Thus, the proposed method provides a
new possibility of optimal system design based on an operation reliability (correctness) criterion and constraints on the
domain of feasible values of parameters.
CONCLUSIONS
The paper has proposed a method to model system reliability based on the algebra of regular algorithms and fuzzy set
theory. In contrast to the well-known approaches [14], which use Boolean algebra as descriptive means, the method allows
modeling (on a unified logic-algorithmic basis) the events related to the onset, detection, and elimination of failures in a system.
A fuzzy-algorithmic algebra has been developed to implement the method. It represents formal transition rules to pass
from operations in an algebra of regular algorithms to the corresponding operations over membership functions, which
specify the distributions of correct execution of operators and conditions. Applying these rules allows presenting the
reliability model as a hierarchical system of fuzzy knowledge bases that correspond to operator and logical structures.
Unlike the classical probabilistic approach, the proposed method does not need statistical data and allows restricting
the consideration to expert knowledge. Moreover, a radically new possibility of modeling the reliability with regard for
constraints on the domain of feasible values of parameters of the elements is provided.
REFERENCES
1. I. A. Ryabinin, Reliability and Safety of Structurally Complex Systems [in Russian], Izd. S.-Peterb. Univ.,
St.-Petersburg (2007).
2. E. J. Henley and H. Kumamoto, Reliability Engineering and Risk Assessment, Prentice-Hall Inc., Englewood Cliffs
(1981).
3. V. M. Glushkov, Introduction to Automatic Control Systems [in Russian], Tekhnika, Kyiv (1974).
4. A. I. Gubinskii, Reliability and Performance of Ergatic Systems [in Russian], Nauka, Leningrad (1982).
930
Fig. 12. The distribution of the correctness of the execution of algorithm (11): variant 1 (a);
variant 2 (b); variant 3 (c)
100
1.0
0.8
0.6
0.4
0.2
m
F
1
0.0
50
a
x
y
90
80
m
F
1
50
100
80
y
x
0.0
0.2
1.0
0.8
0.6
0.4
c
x
50
100
80
y
90
m
F
1
0.4
0.6
0.8
1.0
0.2
0.0
b
90
5. I. V. Safonov, Formalized reliability analysis of algorithmic processes, Upravl. Mashiny i Sistemy, No. 3, 9295
(1973).
6. A. P. Rotshtein and P. D. Kuznetsov, Design of Faultless Man-Machine Technologies [in Russian], Tekhnika, Kyiv
(1992).
7. V. M. Glushkov, Automata theory and formal transformations of microprograms, Kibernetika, No. 5, 110 (1965).
8. V. M. Glushkov, G. E. Tseitlin, and E. L. Yushchenko, Algebra. Languages. Programming [in Russian], Naukova
Dumka, Kyiv (1989).
9. A. P. Rotshtein, Probabilistic-algorithmic models of man-machine systems, Avtomatika, No. 5, 8187 (1987).
10. L. A. Zadeh, The concept of a linguistic variable and its application to approximate reasoning, Information
Sciences, 8, No. 3, 199249 (Part I); 8, No. 4, 301357 (Part II); 9, No. 1, 4380 (Part III) (1975).
11. L. A. Zadeh, Fuzzy sets as a basic for a theory of possibility, Fuzzy Sets and Systems, 1, 328 (1978).
12. A. P. Rotshtein and D. I. Katelnikov, Identification of nonlinear objects by fuzzy knowledge bases, Cybern. Syst.
Analysis, 34, No. 5, 676683 (1998).
13. A. Rotshtein, Design and tuning of fuzzy rule-based systems for medical diagnosis, in: N.-H. Teodorescu,
A. Kandel, and G. Lain (eds.), Fuzzy and Neuro-Fuzzy Systems in Medicine, CRC Press, Boca-Raton (1998),
pp. 243289.
14. K. Y. Cai, Introduction on Fuzzy Reliability, Kluwer Academ. Publ., Boston (1996).
931

Anda mungkin juga menyukai