Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

ARTICLE IN PRESS

Reliability Engineering and System Safety 92 (2007) 1601–1607


www.elsevier.com/locate/ress

Extended block diagram method for a multi-state system


reliability assessment
Anatoly Lisnianski
Israel Electric Corporation Ltd, Reliability Department, New Office Building, P.O.B. 10, Haifa, Israel
Accepted 22 September 2006
Available online 17 November 2006

Abstract

The presented method extends the classical reliability block diagram method to a repairable multi-state system. It is very suitable for
engineering applications since the procedure is well formalized and based on the natural decomposition of the entire multi-state system
(the system is represented as a collection of its elements). Until now, the classical block diagram method did not provide the reliability
assessment for the repairable multi-state system. The straightforward stochastic process methods are very difficult for engineering
application in such cases due to the ‘‘dimension damnation’’—huge number of system states. The suggested method is based on the
combined random processes and the universal generating function technique and drastically reduces the number of states in the multi-
state model.
r 2006 Elsevier Ltd. All rights reserved.

Keywords: Block diagram; Reliability; Multi-state system; Universal generating function; Random process

1. Introduction damnation’’. At first, the state-space diagram building or the


model construction for MSS is not a simple job. It is a
Traditional binary reliability models allow only two difficult, non-formalized process that may cause numerous
functional states, for a system and for each of its components: mistakes even for a relatively small MSS. The problem of
perfect functionality (UP) and complete failure (DOWN). identifying all of the states and transitions correctly is a very
Multi-state system (MSS) reliability analysis relates to difficult assignment. Secondly, solving models with hundreds
systems for which one cannot formulate an ‘‘all or nothing’’ of states can challenge the available computer resources.
type of failure criterion [1]. Many real-world systems are For an MSS consisting of n different repairable elements
composed of multi-state components, which have different where every element j has Q kj different performance levels
performance levels and several failure modes with various one has a model with K ¼ nj¼1 kj states. This number can
effects on the system’s entire performance. Reliability models be very large even for a relatively small MSS.
were extended from the binary state (UP or DOWN) to the If the stochastic process is identified as the Markov
finite number of performance levels (MSS model). Random process, then a system of K differential equations must be
processes methods are suggested for repairable MSS in order solved in order to find the state probabilities of MSS. If
to assess system reliability [2,3]. According to these methods, sojourn times in some states are non-exponentially
a state-space diagram of MSS should be built and the distributed, then the semi-Markov process application
transitions between all the states should be defined. A system often provides a good opportunity to obtain the solution
evolution is represented by a continuous-time discrete state [4]. The number of integral equations in the system that
stochastic process. should be solved using the semi-Markov approach is equal
The main disadvantages of using random processes to the square of the total number of MSS states. Therefore,
models for MSS reliability evaluation are the ‘‘dimension the total number of integral equations (that should be
solved in order to find the states probabilities for the MSS
E-mail addresses: anatoly-l@iec.co.il, lisnians@zahav.net.il. by using straightforward semi-Markov method) will be K2.

0951-8320/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ress.2006.09.013
ARTICLE IN PRESS
1602 A. Lisnianski / Reliability Engineering and System Safety 92 (2007) 1601–1607

In a general case simulation may be performed in order defined based on previously determined states probabilities
to assess MSS reliability. The simulation technique is also for all elements and the system structure function. At this
very sensitive to the states number of the model. It has the stage, the UGF technique provides a simple procedure that
same difficulties during the model construction stage and is based solely on algebraic operation.
often requires enormous computational resources during Without the loss of generality, we consider a multi-state
the solution stage. element with minor failures and repairs (as defined in [7]).
Therefore, the method’s development, which is based on With every state i there is an associated performance gji of
simplified procedures and can reduce the problem’s the element j. The states are ordered so that gji+1Xgji for
dimension, may be extremely beneficial for reliability any i. Minor failures and repairs cause element transitions
engineers. This paper presents such a method and uses a from state i, where 1pipkj, only to the adjacent states i1
special mathematical technique—the universal generating and i+1 respectively. It will be in transition to the state i1
function (UGF) and is termed the combined UGF and the from the state i if failure occurs in the state i, and it will be
random processes method. The UGF technique was in transition to the state i+1 if the repair will be completed.
introduced in [5]. More details about the UGF can be In the state kj it may be only the failure and transition to
found in [6,7], with its comprehensive, up-to-date descrip- the state kj1 and in the state 1 it may be only the repair
tion of the UGF technique. and transition to the state 2.

2. Performance stochastic process for a multi-state element 2.1. Markov model for multi-state element

Generally, any element j in MSS can have kj different If all failures and repair times are exponentially
states corresponding to the different performances, repre- distributed then the performance stochastic process will
sented by the set gj ¼ fgj1 ; . . . ; gjkj g, where gji is the have a Markov property and can be represented by a
performance rate of element j in the state i, Markov model [3]. Here for simplicity we omit index j and
i 2 f1; 2; . . . ; kj g. The generic MSS model [1] consists of assume that the element has k different states as presented
the performance stochastic processes Gj ðtÞ 2 gj , j ¼ 1; . . . ; n in Fig. 1. For a Markov process, each transition from the
for each system element j, and the system structure state s to any state m (s; m ¼ 1; . . . ; k) has its own
function that produces the stochastic process correspond- associated transition intensity designated as asm. In our
ing to the output performance of the entire MSS: case, any transition is caused by the element’s failure or
GðtÞ ¼ fðG 1 ðtÞ; . . . ; Gn ðtÞÞ. repair. If mos, then asm ¼ lsm , where lsm is a failure rate
In traditional binary-state reliability interpretation, [8] a for the failures that cause the element transition from state
reliability block diagram shows the interdependencies s to state m. If m4s, then asm ¼ msm , where msm is a
among all of the elements. The purpose is to show by
concise visual shorthand, the various block combinations
(paths) that result in the system’s success. Each block of the
reliability block diagram represents one element of the k
function contained in the system. All blocks are configured
in series, parallel, standby, or combinations thereof as μk-1,k λk,k-1
appropriate. The diagram blocks follow a logical order,
which relates the sequence of events during the prescribed
system operation. The reliability model consists of a k-1
reliability block diagram and an associated mathematical
λk-1,k-2
or simulation model. μk-2,k-1
In a multi-state interpretation, each block of the
reliability block diagram represents one multi-state element ... ... ...
of the system. A logical order of the blocks in the diagram
is defined by the system structure function
fðG 1 ðtÞ; . . . ; Gn ðtÞÞ and each block’s j behavior is defined μ2,3 λ2,3
by the corresponding performance stochastic process Gj(t). 2
According to the suggested method, during the first stage
a model of stochastic process should be built for every
multi-state element. Based on this model, a state prob- μ1,2 λ2,1
abilities
pji ðtÞ ¼ PrfG j ðtÞ ¼ gji g; i 2 f1; . . . ; kj g 1

for every MSS’s element j 2 f1; . . . ; ng can be obtained.


At the second stage the output performance distribution Fig. 1. State-space diagram for Markov model of repairable multi-state
for the entire MSS at each time of instant t should be element.
ARTICLE IN PRESS
A. Lisnianski / Reliability Engineering and System Safety 92 (2007) 1601–1607 1603

corresponding repair rate. The corresponding performance


gs is associated with each state s.
Let ps(t), s ¼ 1; . . . ; k is the state probabilities of the
element’s performance process G(t) at time t:
ps ðtÞ ¼ PrfGðtÞ ¼ gs g; s ¼ 1; . . . ; k; tX0.
The following system of differential equations for finding
the state probabilities ps(t), s ¼ 1; . . . ; k for the homo-
geneous Markov process can be written (Trivedi, 2002) as
2 3
dps ðtÞ 4X k Xk
¼ pi ðtÞais 5  ps ðtÞ asi . (1)
dt i¼1 i¼1
ias ias

In our case, all transitions are caused by the element’s Fig. 2. State-space diagram for semi-Markov model of repairable multi-
failures and repairs corresponding to the transition state element.
intensities ais and are expressed by the element’s failure
and repair rates. Therefore, the corresponding system of where
differential equations may be written as Q12 ðtÞ ¼ F 1;2 ðtÞ, (4)
dp1 ðtÞ Z t
¼  m12 p1 ðtÞ þ l21 p2 ðtÞ,
dt Qlþ1l ðtÞ ¼ ½1  F lþ1;lþ2 ðtÞ dF lþ1;l ðtÞ for 1plpk  2,
0
dp2 ðtÞ
¼ m12 p1 ðtÞ  ðl21 þ m23 Þp2 ðtÞ þ l32 p3 ðtÞ, (5)
dt
... ¼ ... Z t

dpk ðtÞ Qllþ1 ðtÞ ¼ ½1  F l;l1 ðtÞ dF l;lþ1 ðtÞ for 2plpk  1,
¼ mk1;k pk1 ðtÞ  lk;k1 pk ðtÞ. ð2Þ 0
dt (6)
We assume that the initial state is the state k with the
best performance. Therefore, by solving system (2) of Qkk1 ðtÞ ¼ F k;k1 ðtÞ. (7)
differential equations under the initial condition pk ð0Þ ¼ 1, The kernel matrix (3) and the initial state k (with the best
pk1 ð0Þ ¼    ¼ p2 ð0Þ ¼ p1 ð0Þ ¼ 0, the states probabilities performance) completely define the semi-Markov process,
ps(t), s ¼ 1; . . . ; k can be obtained. which describes the stochastic behavior of a multi-state
element.
For every element we designate ylm(t) the probability
2.2. Semi-Markov model for multi-state element that the semi-Markov stochastic process which starts from
initial state l at instant t ¼ 0, will be in state m at instant t.
Now we assume that for every element j, 1pjpn time to Probabilities ylm(t), l; m ¼ 1; 2; . . . ; k can be found from the
failure is distributed according to the arbitrary cumulative solution in the following system of integral equations:
distribution function (c.d.f.) F ðjÞ i;i1 ðtÞ for any state i,
" #
k Z t
Xk X
1pipkj. Analogously, for any state i, 1pipkj time to ylm ðtÞ ¼ dlm 1  Qlm ðtÞ þ qls ðtÞysm ðt  tÞ dt,
repair is assumed to be distributed according to c.d.f. m¼1 s¼1 0
ðjÞ
F i;iþ1 ðtÞ. As was done in the previous section, we also omit
l; m ¼ 1; . . . ; k, ð8Þ
index j in further computations. The state-space diagram
and corresponding transitions are presented in Fig. 2. where
In order to define the semi-Markov process that defines dQis ðtÞ
the element’s functioning, the corresponding kernel matrix qis ðtÞ ¼ and dlm ¼ 1 if l ¼ m; dlm ¼ 0 if lam.
dt
jQlm ðtÞj, l; m ¼ 1; 2; . . . ; k should be obtained. Each element
Qlm(t) of this matrix determines the probability that We assume that the process always starts from the state
transition from state l to state m occurs during the time k (the best state). Hence the states probabilities of multi-
interval [0, t]. According to [4] we obtain state element, which should be defined based on the
  solution of the system of integral Eq. (8), are the following:
 0 Q12 ðtÞ 0 0 ... 0 0 

  pk ðtÞ ¼ ykk ðtÞ; pk1 ðtÞ ¼ ykk1 ðtÞ; . . . ; p1 ðtÞ ¼ yk1 ðtÞ. (9)
   Q21 ðtÞ 0 Q23 ðtÞ 0 ... 0 0 
Q ðtÞ ¼  ,
lm  ... ... ... ... ... ... . . .  At the first stage ‘‘small’’ Markov or semi-Markov

  models should be constructed for each element of the entire
 0 0 0 0 ... Qkk1 ðtÞ 0 
MSS. If the performance stochastic process for multi-state
(3) element is the Markov process, then the Markov model of
ARTICLE IN PRESS
1604 A. Lisnianski / Reliability Engineering and System Safety 92 (2007) 1601–1607

this element consists of k differential equations. If the (b) MSS expected output performance at instant t40
performance stochastic process for a multi-state element is !
X
K X
K
a semi-Markov process, then the semi-Markov stochastic EðtÞ ¼ dE ðUðz; tÞÞ ¼ dE pi ðtÞzgi ¼ pi ðtÞgi :
model for this element consists of k2 integral equations. By i¼1 i¼1
solving these equations, we obtain the performance (13)
probability distribution pi ðtÞ ¼ PrfGðtÞ ¼ gi g, i ¼ 1; . . . ; k
for every multi-state element at each time instant t.
(c) MSS expected performance deficiency at t40 for
arbitrary constant demand w !
3. Multi-state system reliability evaluation X
K
Dðt; wÞ ¼ dD ðUðzÞ; wÞ ¼ dD pi ðtÞzgi ; w
At the second stage based on determined state prob- i¼1
abilities for all elements, the individual UGF for each X
K

element should be defined. Then by using composition ¼ pi ðtÞ maxðw  gi ; 0Þ: ð14Þ
i¼1
operators over UGF of individual elements and their
combinations in the entire MSS structure, one can obtain
the resulting UGF for the entire MSS by using simple
4. Numerical example
algebraic operations. This UGF defines the output
performance distribution for the entire MSS at each time
Consider a flow transmission system that is presented in
instant t. MSS reliability measures can be easily derived
Fig. 3 consisting of three elements (pipes). The oil flow is
from this output performance distribution.
transmitted from left to right. The performance of the pipes is
The following steps should be executed during the
measured by their transmission capacity (tons per minute).
second stage:
Times to failures and times to repairs are distributed
exponentially for all elements. Elements 1 and 2 are repairable
1. Having performances gji and corresponding probabil- and each one has two possible states. A state of total failure
ities pji(t) for each element j, j ¼ 1; . . . ; n, i ¼ 1; . . . ; kj for both elements corresponds to a transmission capacity of 0
one can define the UGF for this element in the and the operational state corresponds to the capacities of 1.5
following form : and 2 t/min, respectively so that
gjkj
uj ðzÞ ¼ pj1 ðtÞzgj1 þ pj2 ðtÞzgj2 þ    þ pjkj ðtÞz . (10) G1 ðtÞ 2 fg11 ; g12 g ¼ f0; 1:5g and G 2 ðtÞ 2 fg21 ; g22 g ¼ f0; 2g.
The failure rates and repair rates corresponding to these
2. The composition operators Ofs (for elements connected two elements are:
in a series), Ofp (for elements connected in parallel) and
OfB (for elements connected in a bridge structure) lð1Þ 1 ð1Þ
2;1 ¼ 7 year ; m1;2 ¼ 100 year
1
for element 1;
should be applied over the UGF of individual elements
lð2Þ
2;1 ¼ 10 year1 ; mð2Þ
1;2 ¼ 80 year1 for element 2:
and their combinations. These operators were defined
in [7], where corresponding recursive procedures for Element 3 is a multi-state element with minor failures
their computation were introduced for different types and minor repairs. It can be in one of three states: a state
of systems. Based on these procedures, the resulting of total failure corresponding to a capacity of 0, a state
UGF for the entire MSS can be obtained: of partial failure corresponding to a capacity of 1.8 t/min,
X
K and a fully operational state with a capacity of 4 t/min.
Uðz; tÞ ¼ pi ðtÞzgi , (11) Therefore,
i¼1
G3 ðtÞ 2 fg31 ; g32 ; g33 g ¼ f0; 1:8; 4g.
where K is the number of the entire system states and gi
is the entire system performance in the corresponding The failure rates and repair rates corresponding to element
state i, i ¼ 1; . . . ; K. 3 are
3. Applying the operators dA, dE, dD introduced in [7] over
the resulting UGF of the entire MSS one can obtain the lð3Þ 1 ð3Þ ð3Þ 1
3;2 ¼ 10 year ; l3;1 ¼ 0; l2;1 ¼ 7 year ,
following MSS reliability indices: mð3Þ ð3Þ 1 ð3Þ 1
1;3 ¼ 0; m1;2 ¼ 120 year ; m2;3 ¼ 110 year .
(a) MSS availability A(t, w) at instant t40 for
arbitrary constant demand w The system output performance rate is defined as the
! maximum flow that can be transmitted. The MSS structure
X
K
function is
gi
Aðt; wÞ ¼ dA ðUðz; tÞ; wÞ ¼ dA pi ðtÞz ; w
i¼1 Gs ðtÞ ¼ fðG 1 ðtÞ; G 2 ðtÞ; G3 ðtÞÞ
X
K
¼ minfG 1 ðtÞ þ G 2 ðtÞ; G 3 ðtÞg.
¼ pi ðtÞ1ðgi  wX0Þ. ð12Þ
i¼1 The demand is constant: w ¼ 2:0 t=min.
ARTICLE IN PRESS
A. Lisnianski / Reliability Engineering and System Safety 92 (2007) 1601–1607 1605

Fig. 3. Multi-state system structure and state-space diagrams for the system’s elements.

Applying the described above two-stage procedure, we For element 3:


proceed as follows: g3 ¼ fg31 ; g32 ; g33 g ¼ f0; 1:8; 4:0g; p3 ðtÞ ¼ fp31 ðtÞ; p32 ðtÞ; p33 ðtÞg.

1. According to the Markov method we build the


following systems of differential equations for each 2. Having the sets gj, pj(t) for j ¼ 1; 2; 3 obtained in the
element separately (using the state-space diagrams first step, we can define the u-functions of the individual
presented in Fig. 3 inside the corresponding elements). elements as:
For element 1:
u1 ðzÞ ¼ p11 ðtÞzg11 þ p12 ðtÞzg12 ¼ p11 ðtÞz0 þ p12 ðtÞz1:5 ,
dp11 ðtÞ=dt ¼  mð1Þ ð1Þ
1;2 p11 ðtÞ þ l2;1 p12 ðtÞ, u2 ðzÞ ¼ p21 ðtÞzg21 þ p22 ðtÞzg22 ¼ p21 ðtÞz0 þ p22 ðtÞz2 ,
dp12 ðtÞ=dt ¼  lð1Þ
2;1 p12 ðtÞ þ mð1Þ
1;2 p11 ðtÞ. u3 ðzÞ ¼ p31 ðtÞzg31 þ p32 ðtÞzg32 þ p33 ðtÞzg33
Initial conditions are: p12 ð0Þ ¼ 1, p11 ð0Þ ¼ 0. ¼ p31 ðtÞz0 þ p32 ðtÞz1:8 þ p33 ðtÞz4 .
For element 2:
dp21 ðtÞ=dt ¼  mð2Þ ð2Þ
1;2 p21 ðtÞ þ l2;1 p22 ðtÞ,
These functions are also presented in Fig. 3 under the
corresponding elements:
dp22 ðtÞ=dt ¼  lð2Þ
2;1 p22 ðtÞ þ mð2Þ
1;2 p21 ðtÞ. 3. Using the composition operators Ofs and Ofp for flow
Initial conditions are: p21 ð0Þ ¼ 1, p22 ð0Þ ¼ 0. transmission, MSS with flow dispersion [7] we obtain
For element 3: the resulting UGF for the entire series-parallel MSS

dp31 ðtÞ=dt ¼  mð3Þ ð3Þ UðzÞ ¼ Ofs ðOfp ðu1 ðzÞ; u2 ðzÞÞ; u3 ðzÞÞ.
1;2 p31 ðtÞ þ l2;1 p32 ðtÞ,

dp32 ðtÞ=dt ¼ lð3Þ ð3Þ ð3Þ ð3Þ


3;2 p33 ðtÞ  ðl2;1 þ m2;3 Þp32 ðtÞ þ m1;2 p31 ðtÞ, In order to find the resulting UGF for elements 1 and 2
dp33 ðtÞ=dt ¼  lð3Þ ð3Þ
3;2 p33 ðtÞ þ m2;3 p32 ðtÞ. connected in parallel, the operator Ofp applied to
individual UGF u1(z) and u2(z):
Initial conditions are: p31 ð0Þ ¼ p32 ð0Þ ¼ 0, p33 ð0Þ ¼ 1.
A closed-form solution was obtained for each of these
three systems of differential equations. Corresponding Ofp ðu1 ðzÞ; u2 ðzÞÞ ¼ Ofp ðp11 ðtÞz0 þ p12 ðtÞz1:5 ; p21 ðtÞz0
expressions for the states probabilities are presented in þ p22 ðtÞz2 Þ
the Appendix.
¼ p11 ðtÞp21 ðtÞz0 þ p12 ðtÞp21 ðtÞz1:5
Therefore, one obtains the following performance
distributions: þ p11 ðtÞp22 ðtÞz2 þ p12 ðtÞp22 ðtÞz3:5 .
For element 1:
g1 ¼ fg11 ; g12 g ¼ f0; 1:5g; p1 ðtÞ ¼ fp11 ðtÞ; p12 ðtÞg. As can be observed the resulting UGF is a simple
For element 2: product of u1(z) and u2(z).
In order to find the UGF for the entire MSS, where
g2 ¼ fg21 ; g22 g ¼ f0; 2:0g; p2 ðtÞ ¼ fp21 ðtÞ; p22 ðtÞg. element 3 is connected in a series with two elements, 1 and
ARTICLE IN PRESS
1606 A. Lisnianski / Reliability Engineering and System Safety 92 (2007) 1601–1607

2 that connected in parallel, the operator Ofs should be 1


applied.
UðzÞ ¼ Ofs ðu3 ðzÞ; Ofp ðu1 ðzÞ; u2 ðzÞÞÞ ¼ Ofs ðp31 ðtÞz0 p5(t)
0.75
þ p32 ðtÞz1:8 þ p33 ðtÞz4 ; p11 ðtÞp21 ðtÞz0
þ p12 ðtÞp21 ðtÞz1:5 þ p11 ðtÞp22 ðtÞz2 þ p12 ðtÞp22 ðtÞz3:5 Þ
¼ p31 ðtÞp11 ðtÞp21 ðtÞz0 þ p31 ðtÞp12 ðtÞp21 ðtÞz0 0.5

þ p31 ðtÞp11 ðtÞp22 ðtÞz0 þ p31 ðtÞp12 ðtÞp22 ðtÞz0


p2(t)
þ p32 ðtÞp11 ðtÞp21 ðtÞz0 þ p32 ðtÞp12 ðtÞp21 ðtÞz1:5 0.25 p4(t) p3(t)
p1(t)
þ p32 ðtÞp11 ðtÞp22 ðtÞz1:8 þ p32 ðtÞp12 ðtÞp22 ðtÞz1:8
þ p33 ðtÞp11 ðtÞp21 ðtÞz0 þ p33 ðtÞp12 ðtÞp21 ðtÞz1:5
0
þ p33 ðtÞp11 ðtÞp22 ðtÞz2 þ p33 ðtÞp12 ðtÞp22 ðtÞz3:5 . 0 0.04 0.08 0.12 0.16 0.2
The resulting UGF U(z) is also a product of poly- time (years)
nomials, but the powers of z are found as the minimum of
Fig. 4. Probabilities of different performance levels.
powers of the corresponding terms.
Taking into account that
p31 ðtÞ þ p32 ðtÞ þ p33 ðtÞ ¼ 1;
p21 ðtÞ þ p22 ðtÞ ¼ 1 and p11 ðtÞ þ p12 ðtÞ ¼ 1, The instantaneous performance deficiency Dt at any time
t for the constant demand w ¼ 2 t/min:
we obtain the UGF that determines the output perfor-
mance distribution g, p(t) of the entire MSS in the X
5

following form: Dt ¼ dD ðUðzÞ; wÞ ¼ pi ðtÞ maxð2  gi ; 0Þ


i¼1
X
5
¼ p1 ðtÞð2  0Þ þ p2 ðtÞð2  1:5Þ þ p3 ðtÞð2  1:8Þ
UðzÞ ¼ pi ðtÞzgi ,
i¼1 ¼ 2p1 ðtÞ þ 0:5p2 ðtÞ þ 0:2p3 ðtÞ.
where Note: Instead of solving the system of K ¼ 2  2  3 ¼
12 differential equations (as should be performed in the
g1 ¼ 0; p1 ðtÞ ¼ p11 ðtÞp21 ðtÞ þ p31 ðtÞp12 ðtÞ þ p31 ðtÞp11 ðtÞp22 ðtÞ,
straightforward Markov method) here we solve only three
g2 ¼ 1:5 t=min; p2 ðtÞ ¼ p12 ðtÞp21 ðtÞ½p32 ðtÞ þ p33 ðtÞ, systems: one of the third-order and two of the second-
g3 ¼ 1:8 t=min; p3 ðtÞ ¼ p32 ðtÞp22 ðtÞ, order. The further derivation of the entire system states
g4 ¼ 2:0 t=min; p4 ðtÞ ¼ p33 ðtÞp11 ðtÞp22 ðtÞ, probabilities and reliability indices is based on using simple
algebraic equations.
g5 ¼ 3:5 t=min; p5 ðtÞ ¼ p33 ðtÞp12 ðtÞp22 ðtÞ.
These probabilities (output performance distribution of 5. Conclusions
the entire MSS) are presented in Fig. 4 as functions of time.
Based on the UGF U(z) of the entire MSS, one can This paper suggests a new method for complex multi-
obtain the MSS reliability indices by using expressions state system reliability assessment. The presented method
(12)–(14), respectively. extends the classical reliability block diagram method to a
The instantaneous MSS availability for the constant repairable multi-state system. It is suitable for engineering
demand level w ¼ 2.0 t/min applications because the procedure is well formalized and
! based on the natural decomposition of the entire multi-
X5
AðtÞ ¼ dA ðUðzÞ; wÞ ¼ dA gi
pi ðtÞz ; 2 state system. Until now, the classical block diagram
i¼1 method did not provide reliability assessment for a
X
5 repairable multi-state system. Mathematical models that
¼ pi ðtÞ1ðF ðgi ; 2ÞX0Þ ¼ p4 ðtÞ þ p5 ðtÞ. are based on straightforward stochastic processes are very
i¼1 difficult for engineering application due to the ‘‘dimension
The instantaneous mean output performance at any damnation’’—huge number of system states.
instant t40, The advantages of the proposed approach are:

X
5  Simplification of the multi-state system model building.
E t ¼ dE ðUðzÞÞ ¼ pi ðtÞgi ¼ 1:5p2 ðtÞ þ 1:8p3 ðtÞ Instead of building a complex model for the entire multi-
i¼1
state system, one should built n separate relatively
þ 2p4 ðtÞ þ 3:5p5 ðtÞ. simple models for system elements.
ARTICLE IN PRESS
A. Lisnianski / Reliability Engineering and System Safety 92 (2007) 1601–1607 1607

 Simplification of the process for solving a system of [3] Trivedi K. Probability and statistics with reliability, queuing and
equations. Instead of solving one high-order system of computer science applications. New York: Wiley; 2002.
[4] Limnios N, Oprisan G. Semi-Markov processes and reliability.
differential (for Markov process) or integral (for semi-
Boston/Berlin: Birkhauser; 2001.
Markov process) equations, one must solve n low-order [5] Ushakov I. A universal generating function. Sov J Comput Syst
systems for each system element. Science 1986;24:37–49.
[6] Gnedenko B, Ushakov I. Probabilistic reliability engineering. NY/
Chichester/Toronto: Wiley; 1995.
References [7] Lisnianski A, Levitin G. Multi-state system reliability. Assessment,
optimization, applications. New Jersey/London/Singapore: World
[1] Natvig B. Multi-state coherent systems. In: Jonson N, Kotz S, editors. Scientific; 2003.
Encyclopedia of statistical sciences, vol. 5. New York: Wiley; 1984. [8] Modarres M, Krivtsov V, Kaminskiy M. Reliability engineering and
[2] Aven T, Jensen U. Stochastic models in reliability. New York, Berlin: risk analysis. A practical guide. New York, Basel: Marcel Dekker;
Springer; 1999. 1999.

You might also like