Sie sind auf Seite 1von 5

Reliability Analysis of a Repairable Parallel System

with Repair Time Omission


Yu Pang, Hong-Zhong Huang*, Yu Liu, Qiang Miao, Zhonglai Wang
School of Mechatronics Engineering
University of Electronic Science and Technology of China
Chengdu, Sichuan, 610054, P.R. China
*Corresponding author. E-mail: hzhuang@uestc.edu.cn
AbstractBased on some practical problems in industrial
engineering, a species of two-unit parallel repairable system
model is introduced in this paper. The prominent feature of the
model is described as follows: if a repair time is sufficiently short
(less than some threshold value) that does not affect the system
operation, i.e. the effect could be ignored, the system can be
considered as being operating during this repair time. Two
scenarios are discussed where the threshold value is regarded as
a constant and non-negative random variable, respectively.
Reliability indices such as instantaneous availability are
formulated.
Keywords- availability; Markov model; parallel system; repair
time omission

I.

section and the formulations of the indices are proofed in detail.


A numerical example is presented in Section V, as well as the
steady-state availability of the system. Conclusion is provided
in Section VI.
II.

BASIC ASSUMPTIONS OF THE ORIGINAL PARALLEL


SYSTEM

Before formulating the new model, the original two-unit


parallel repairable system is defined at first [5][6]:
The system is consisting of two different units as
shown in Fig.1. The system fails if and only if the units
are both in failure state, and there is only one repair
facility available.

INTRODUCTION

A great deal of research has been devoted to system


reliability for its wide applications in power plants,
manufacturing systems, and military equipment [1][2].
Reliability analysis of repairable system is a momentous
branch of reliability theory, maintaining a high level reliability
is often an essential requisite for repairable system [3].
In the former literatures, two basic assumptions are
involved in most of research on repairable system: 1) when the
operating unit fails, repair begins immediately; 2) the system
instantaneously falls into the failure state when it is out of work.
However, in some practical situations, for example, if the
demands on the system by customers are not too frequent,
they more likely miss a small repair time, or are at most
delayed by a negligibly short time in receiving service. In
generally, if the system is temporarily under repair which has
no effect on system operating, the system can be regarded as
being operating during such a repair interval [4]. In other words,
if a short repair time does not result in the system being out of
work; such a repair interval can be omitted from the downtime
period. Taking this into consideration, Zheng et al.[4] calculate
system availability for a single-unit Markov repairable system.
In this paper, the reliability of a two-unit parallel system is
analyzed with considering the repair time omission introduced
in [4]. Reliability indices such as system availability are
formulated. The remainder of this paper is organized as follows:
In Section II, some basic assumptions of the original parallel
system model are given. Difference between the original model
and the new model are analyzed in Section III. A
comprehensive reliability analysis is discussed in the next
Sponsored by the National Science Foundation of China (50775026) and
the National Defense Key Foundation of China (9140A27020308JB34).

978-1-4244-4905-7/09/$25.002009 IEEE
41

Assume that the failure time and the repair time of


each unit follow exponential distribution Fi (t) and Gi (t)
with the parameter of i and i , respectively. All the
random variables involved are mutually independent.
Thus, one has:
F i ( t ) = 1 e i t , i > 0

G i ( t ) = 1 e i t , i > 0

i = 1, 2 ; t 0

The two units are new at the initial time ( t =0), and
the repaired unit is restored into as good as new
condition.

According to stochastic behavior of the entire system, there


exist five system states:
State 1: unit 1 is up (operating), unit 2 is down (failed). In
other words, unit 2 is under repair while unit 1 is working; the
system is still operating;

State 2: unit 2 is up, unit 1 is down; the system is in the


operating state;
State 3: unit 1 is under repair, unit 2 is waiting for repair;
the system in its failure state;

The new system model is still operating if the original


system model is failed but the repair time of the system
is less than the threshold value ;

The new system model is failed only when the original


system model is failed and repair time is greater than
the threshold value ;

The threshold value can be either a constant or a


non-negative random variable. The distribution is
H ( ) if is considered as a non-negative random
variable.

State 4: unit 2 is under repair, unit 1 is waiting for repair;


the system is also in failure state.
The state space is defined as E = {0,1,2,3,4} , the working
state space is W = {0,1,2} while the failure state space is F = {3,4} .
Let X (t ) representing the stochastic process of the system
state at time instant t , thus one has:
0,
1 ,

2,

X (t ) = 3,

4,

both two units are operating


unit 1 is up, unit 2 is down
unit 2 is up, unit 1 is down
unit 1 is under repair,
unit 2 is waiting for repair
unit 2 is under repair,
unit 1 is waiting for repair

at time t ;
at time t ;
at time t ;
at time

t;

at time

t;

In particular, when = 0 , the new system model changes


into the original system model; when = , the new system is
never down. A possible sequence of system state changes of
the original system model and the new system model is shown
in Fig.2.

Apparently, {X (t ), t 0} forms a homogeneous continuous


time Markov process in state space E .

III.

BASIC ASSUMPTIONS OF THE NEW MODEL

The main difference between the original system model and


the new system model (parallel repairable system with
considering repair time omission) is that: when the original
system fails, the repair facility is still repairing the failed unit
which fails earlier. Given a threshold value ( 0) , if any
repair time involved is shorter than , the system can be
regarded as in its operating state during the repair interval. That
is to say the new system model is considered to be still in up
state during this repair interval while the original system model
is down state during the same period.
In contrast with the original system model, the new parallel
repairable system model is described as follows:

The new system model consists of two units and only


one repair facility is available. The two units are new at
the initial time ( t =0); all the random variables are
mutually independent; the failed units can be regarded
as good as new after repair;

Whether the failed unit is waiting for or is receiving


repair depends on the state of the repair facility. If the
repair facility is available (no occupied), the failed unit
will receive repair immediately after failure. The
waiting period is called repair delay time. The failure
time and repair time of these two units follow the
exponential distribution Fi (t ) and Gi (t ) as in the
original model. Qi (t ) denotes the distribution of the
repair delay time of the unit i which is formulated in
the Section IV;

The new system model is operating if the original


system model is operating;

42

Let the stochastic process X(t) denoting the state of the new
system model at any time instant t , and one has:
0 , the new system is up at time t
~
X (t ) =
1, the new system is down at time t

It is obvious that the Markov property is not held in the new


system model.

IV.

RELIABILITY INDICES ASSESSMENT

Considering the new system model, the instantaneous


system availability is discussed in this section. It is the
derivation of the reliability indices that would be carried on in
this section.
~

A(t ) and A(t) denote the instantaneous availabilities of the

original system model and the new system model, respectively.


The instantaneous availability of a single repairable unit i
( i =1,2) is denoted by Ai (t ) , and it is formulated as [5]:
Ai (t ) =

i + i

i + i

e ( i + i )t

For a single repairable unit i ( i =1,2) with considering the


repair delay, the Laplace form of the instantaneous availability
Ai' (t ) is well defined as [9]:
Ai* ( s ) =

1 / s (1 Fi ( s ))
,
1 F ( s )G ( s )Q ( s )
i

where Fi ( s) , Gi (s) and Q i (s) represent the Laplace-Stieltjes


transforms of Fi (t) , Gi (t) and Qi (t ) , respectively.
Proposition 1: Assuming that the threshold value i ( i = 1,2 )
are constants, thus the instantaneous availability of the new
model is formulated as follows:
min(t ,1 ) s1
~
A(t ) = A(t ) + 12 A1 (t s1 ) A2' (t s2 ) P(s1 < Y1 < 1 )ds2 ds1

0
0
min(t , 2 ) s2

'
12 A1 (t s1 ) A2 (t s2 ) P(s2 < Y2 < 2 )ds1ds2

(1)
Proof: In conformity with the assumption of the new
system model, the instantaneous availability A~(t ) can be
derived from the definition as:

~
~
A ( t ) = P (the new system is up at time t ) = P X ( t ) = 0
~
~
= P X (t ) = 0 , X (t ) = 0 + P X (t ) = 0 , X (t ) = 1
~
~
+ P X (t ) = 0 , X (t ) = 2 + P X (t ) = 0, X (t ) = 3
~
+ P X (t ) = 0 , X (t ) = 4

) (
) (
)

(
(

) (
)

) (

(2)

denotes the probability that unit 1 is receiving repair while unit


2 is waiting for repair, and the repair time is less than 1 at
time instant t, along with the independence of the variables,
one has the formulation of P (X~ (t ) = 0, X (t ) = 3) :

~
P X (t ) = 0, X (t ) = 3

s1

A1 (t s1 )1 P(s1 < Y1 < 1 )

A2' (t

s 2 )2 ds2 ds1

Proposition 2: If i is a non-negative random variable with


distribution function denoted as H i ( ) , ( i = 1,2 ), the
~
instantaneous availability A(t) of the new system model is
formulated as follows:
t 1 s1
~
A(t) = A(t ) + 12 A1 (t s1 ) A2' (t s2 )P(s1 < Y1 < 1 )ds2 ds1dH1 ( )
0 0 0
t s1

+ 12 A1 (t s1 ) A2' (t s2 )P(s1 < Y1 < 1 )ds2 ds1dH1 ( )


t 0 0
t 2 s2

0 0 0
t s2

+ 12 A1' (t s1 ) A2 (t s2 )P(s2 < Y2 < 2 )ds1ds2 dH2 ( )

= P(unit 1 is under repair at timet and repair time is no longer than 1 )


min(t ,1 )

+ 12 A1' (t s1 ) A2 (t s2 )P(s2 < Y2 < 2 )ds1ds2 dH2 ( )

= P(unit 1 is under repair,unit 2 is waitingfor repair at timet


and the repair time is no longer than 1 )
P(unit 2 is waitingfor repair at timet )

Therefore, the instantaneous availability of the new model


can be assessed via substituting Eqs.(2)(4)(5) into Eq. (1).

~
and
Thus, only two terms
P X (t ) = 0, X (t ) = 3
~
are
needed
to
figure
out.
The
first
term
P X (t ) = 0, X (t ) = 4

min(t,2 )s2
~
P X(t) = 0, X(t) = 4 = 12A1' (t s1)A2(t s2)P(s2 < Y2 <2)ds1ds2

(5)

~
~
~
A(t) = A(t) + P X (t) = 0, X (t) = 3 + P X (t ) = 0, X (t ) = 4 .

Furthermore, A1 (t s1 ) represents the probability that unit 1


is up at time t s1 . For the failure time follows exponential
distribution, 1 denotes the constant failure rate of unit 1 while
A1 (t s1 )1 P(s1 < Y1 < 1 )ds1 stand for the probability of the
state of unit 1 switching from working state to failure state in
(t s1 , t s1 + ds1 ) and the repair time Y1 is within the interval
(s1,1) . In other words, it denotes that unit 1 is being repaired at
time t and the repair time is less than the threshold value 1 .
Eq.(4) is hold because of the independence between s 2
and A1 (t s1 )1 P( s1 < Y1 < 1 ) .

Thus, A~(t ) can be simplified as follows:

The repair time of the unit 1 is equal to the summation of


the repair delay time of unit 2 and time deviation between the
failure time of units 1 and 2. The repair delay time of the unit 2
follows the distribution denoted as Q2 (t ) = G1 [t ( s1 s 2 )] .
With the same manner, if unit 2 fails before unit 1, one can
assume s1 < s 2 and the repair delay time of the unit 1 follows
the distribution denoted as Q1 (t ) = G 2 [t ( s 2 s1 )] . In generally,
the expression of the repair delay time of the unit i is known
as Qi (t ) = G j [t ( s j s i )] , i j , i, j = 1,2 , s j > s i .

Likewise, the following formula can be derived:

Similarly, the availability of the original system model can


be expressed as:
~
~
A (t ) = P X (t ) = 0 , X (t ) = 0 + P X (t ) = 0 , X (t ) = 1
~
+ P X (t ) = 0, X (t ) = 2

within the time interval (t s1 , t s1 + ds1 ) , unit 2 transits from


its working state to the state of waiting for repair in time
interval (t s 2 , t s 2 + ds 2 ) after the failure of unit 1.
Therefore, s 2 < s1 can be obtained from t s1 < t s 2 .

t 0 0

(3)
(4)

Considering the scenario that unit 1 fails before unit 2, Eq.(3)


denotes that unit 1 switches from working state to failure state

43

Proof: According to the analysis in Proposition 1, one has:

~
P X (t ) = 0, X (t ) = 3
min(t ,1 ) s1

t 1

'

12 A1 (t s1 ) A2 (t s 2 ) P( s1 < Y1 < 1 )ds2 ds1dH1 (t )

For the new system model, let = i = 1 ( i = 1,2 ) and < t ,


using Eq.(1), one can get the instantaneous availability of the
new model as follows:

0
s1

min(t ,1 ) s1
~
A(t ) = A(t ) + 12 A1 (t s1 ) A2' (t s 2 ) P(s1 < Y1 < 1 )ds2 ds1

= 12 A1 (t s1 ) A2' (t s2 ) P( s1 < Y1 < 1 )ds 2 ds1dH1 ( )


0 0 0

t s1

+ 12 A1 (t s1 ) A2' (t s2 ) P( s1 < Y1 < 1 )ds2 ds1dH1 ( )

0
min(t , 2 ) s2
0

(6)

t 2 s2

+ 12 A1' (t s1 ) A2 (t s2 )P(s2 < Y2 < 2 )ds1ds2 dH2 ( )

1
(1 F ( s ))
s
A2 ( s ) =
1 F ( s )G ( s )Q 2 ( s )

t 0 0

(7)
Proposition 2 is proofed via substituting Eqs.(2)(6)(7) into
Eq. (1). Single helical gear reducer is widely used in the
engineering. The model of the single gear reducer is given in
Fig. 2.

2 + 2

+ 2 + 2

(s + 0 .5 )2

(s + 0 .5 )3 0 .5 3 e 0.5 (s s
1

A1*(s) = A2*(s)

1
= 3+
s

= 0.75

(s + 0.5)2 =
(s + 0.5)3 0.53

(s + 0.5)2
2

3
2

s (s + 0.75) +
4

2 1
s+
3 2
2
2
3 3
s + +
4 4

Using the inverse Laplace transform [10], the instantaneous


availability of the single repairable unit i with repair delay
involved is equal to:

F (t ) = F1 (t ) = F2 (t ) = 1 e 0.5t

G (t ) = G1 (t ) = G2 (t ) = 1 e 0.5t

Q1 (t ) = G2 [t ( s 2 s1 )] = 1 e

To reduce the mathematical complexity, suppose that


s1 = s 2 , and one has:

NUMERICAL EXAMPLE

In this section, a two-unit parallel system is studied to


verify the effectiveness of the proposed formulation. The two
units are assumed to be with identical failure and repair rate.
Given = 1 = 2 = 0.5 , = 1 = 2 = 0.5 , the steady-state
system availability of the original system and some relevant
distributions are calculated as follows [5]:
2

s2
11
'
A2 (t s 2 )P ( s 2 < Y < 1) A1 (t s1 )ds1ds 2
40
0

Laplace transform of A2 ' (t ) is known as:

0 0 0
t s2

A=

s1
11
'
A1 (t s1 )P(s1 < Y < 1) A2 (t s 2 )ds2 ds1
40
0

The expression of Ai ' (t ) should be worked out at first. The

= 12 A1' (t s1 ) A2 (t s2 )P(s2 < Y2 < 2 )ds1ds2 dH2 ( )

V.

= A(t ) +

Similarly, one has:


~
P X (t ) = 0, X (t ) = 4

'
12 A1 (t s1 ) A2 (t s 2 ) P( s 2 < Y2 < 2 )ds1ds2

t 0 0

0.5t

A ' (t ) = Ai' (t ) =

0.5( s2 s1 )

1 2 4t
3
+ e
cos
t
3 3
4

( i = 1,2 )

Next, one can get

Q2 (t ) = G1[t ( s1 s 2 )] = 1 e 0.5t e 0.5( s1 s2 )

The Laplace-Stieltjes transform of Fi (t ) , Gi (t) and Qi (t ) ,


( i = 1,2 ), are given by:

0.5
Fi ( s) = e st dFi (t ) =
s
+
0.5
0

s1

A (t s )ds
'

1
2 3t
3
1
3
sin
s1 + e 4 cos
t
t
3
3
4
4
3

3 ( t s1 ) 1
3 ( t s1 )
2 3 ( t s1 )
+ e 4
+
sin
cos

3
4
4
3

Now, the formulation of the instantaneous availability of


the new system model can be simplified as:

0.5
G i ( s) = e st dGi (t ) =
s + 0.5
0

s1
~
11
A(t ) = A(t ) + A(t s1 )P(s1 < Y < 1) A' (t s 2 )ds2 ds1
20
0

0.5
Q 1 ( s) = e st dQ1 (t ) =
e 0.5( s2 s1 )
s + 0.5
0

The explicitly expression of A~ (t ) is complicated to express,


but easy to evaluate when time instant t is given. Furthermore,

0.5
Q 2 ( s) = e st dQ2 (t ) =
e0.5( s1 s2 )
s + 0.5
0

44

with time instant approaching infinity, the steady-state


availability can be calculated as in (8).
s1
1
~
1
A = A + lim A(t s1)P(s1 < Y < 1) A' (t s2 )ds2ds1
2 t 0
0

s1
1

= A + lim
+
e( + )(t s1 ) es1 e A' (t s2 )ds2ds1
2 t 0 + +
0

1 s1
1
1
= A + lim 1 + es1 t e 2

8 t 0

= 0.75 +

REFERENCES

3
1

2 t
3
1
3
s1 + e 4 cos t sin t +

3
4
4

3
3

e 2
ds1
3

(
)

2 e 4 1 cos 3(t s1 ) + 1 sin 3(t s1 )

4
4
3

[1]

1
(7 * 0.6065 4)
24

0.76

(8)
It is obvious that the availability of the new system model
is higher than that of the original system model because the last
two terms in Eq.(1) are positive, and it has been verified in this
numerical example. It would be more useful if is given
objectively according the industrial problem.

VI.

ACKNOWLEDGMENT
This research was partially supported by the National
Science Foundation of China under contract number 50775026
and the National Defense Key Foundation of China under
contract number 9140A27020308JB34.

CONCLUSION

In this paper, a general reliability analysis for a two-unit


parallel Markov repairable system model with repair time
omission is presented. The character of the studied model is
that not every repair action would affect the systems operation.
The instantaneous system availabilities are formulated. The
analysis approach is general and could be applied in a number
of management and engineering situations. More complicated
system model and relevant reliability indices will be discussed
in the future research.

45

L. A. Sanabria, T. S. Dill, Power system reliability assessment suitable


for a deregulated system via the method of cumulants, Electr. Power
Energy System, vol. 20, 1998, pp. 203211.
[2] D. F. Percy, K. A. H. Kobbacy, H. E. Ascher, Using proportional
intensities models to schedule preventive maintenance intervals, IMAJ.
Math. Appl. Bus. Industry, vol. 9, 1998, pp. 289302.
[3] Ying Liu, Wansheng Tang, Ruiqing Zhao, Reliability and mean time to
failure of unrepairable systems with fuzzy random lifetimes, IEEE
Transactions on Fuzzy Systems, vol. 15, no. 5, Oct. 2007, pp. 1009-1026.
[4] Zhihua Zheng, Lirong Cui, Alan G. Hawkes, A study on a single-unit
Markov repairable system with repair time omission, IEEE
Transactions on Reliability, vol. 55, no. 2, June 2006, pp. 182-188.
[5] Jinhua Cao, Kan Cheng, Introduction to Reliability Mathematics,
Science Press, Beijing, 1986.
[6] JinHua Cao, YanHong Wu, Reliability analysis of a multi state
repairable system with a replaceable repair facility Acta Mathematicae
Applicate Sinica, vol. 4, no. 2, May 1988, pp. 113-121.
[7] D. Towsley, C.G. Rommel, J.A. Stankovic, Analysis of fork-join
program response times on multiprocessors, IEEE Transactions on
Parallel and Distributed Systems, vol. 1, no. 3, July 1990, pp. 286 303.
[8] Seidman Thomas I., `First come, first served' can be unstable, IEEE
Transactions on Automatic Control, vol. 39, no. 10, Oct. 1994, pp. 21662171.
[9] Yong Mao, Cailiang Li, Yinghui Tang, Reliability analysis of one unit
repairable system with delay Repair, Journal of UEST of China, vol. 29,
no. 5, 2000, pp. 545-548.
[10] C. J. Savant, Fundamentals of the Laplace Transformation, McGrawHill Book Co., New York, 1962.

Das könnte Ihnen auch gefallen