Sie sind auf Seite 1von 11

Practical implications

A stochastic maintenance The system availability reference table derived


management model with from the stochastic model of this study is practical
for maintenance managers allocating maintenance
recovery factor personnel as well as formulating the maintenance
cycle time. To meet the satisfactory system
Cheng-Hua Wang and availability, the optimal combination of the
number of maintenance personnel and
Sheue-Ling Hwang maintenance cycle can be found to determine
effective and efficient maintenance schedules and
plans. As to the practical implication of the model
solution, the stochastic model of this study that has
integrated quantitative and qualitative models,
provides a simple and practical solution to the
The authors maintenance problem. As well as the maintenance
cycle and maintenance personnel, other
Cheng-Hua Wang is Lecturer, Department of Business
Administration, Chang-Jung Christian University, Tainan, Taiwan.
parameters such as human error rate and system’s
Sheue-Ling Hwang is Professor, Department of Industrial tolerance time were obtained by the simulated
Engineering and Engineering Management, National Tsing-Hua experiment. In such way, this maintenance
University, Hsinchu, Taiwan. management model could reduce perplexity of the
previous mathematical models. Moreover, a
Keywords similar strategy of model development can be
Maintenance, Human failure, System monitoring applied to various maintenance management fields
as long as the relevant parameters are obtained.
Abstract
In this study, recovery factor was considered in the maintenance
management model that integrates the quantitative method and
qualitative concept. The model is developed to solve the 1. Introduction
practical parameters in a maintenance management task, such
as the number of maintenance personnel and maintenance cycle Most industrial production systems are subject to
time. The stochastic model is applied to construct the deterioration with usage and age. Such system
relationships among maintenance cycle, maintenance personnel deterioration may lead to higher production costs
allocation, human recovery factor, and a system’s tolerance time. and lower product quality and the increased
In addition, a simulated experiment was conducted to find out
possibility of breakdown. Therefore, a well-
the supplementary parameters such as individual latent human
error, individual critical human error, recovery rate, and a planned maintenance management program is
system’s tolerance time. Since system availability is the criterion important to reduce costly breakdowns (Dohi et al.,
of this maintenance management model, the final solution of 2001). In some cases, system failure leads to
this model provides a system availability reference table of the uncountable costs, for example, the failure of an
combination of the number of maintenance personnel and the airplane engine during flight may bring property
maintenance cycle time. This system availability reference table damage and loss of life. Therefore, extensive
is a practical tool for maintenance managers. maintenance is necessary to improve the system
availability (Murthy and Hwang, 1996).
Electronic access
Maintenance management plays a critical role
The Emerald Research Register for this journal is in preventing deterioration and failure in systems.
available at Usually, maintenance management can be defined
www.emeraldinsight.com/researchregister as the combination of technical maintenance and
The current issue and full text archive of this journal is associated administrative actions intended to
available at retain an item in a system or restore the system to a
www.emeraldinsight.com/1355-2511.htm normal state. The maintenance management
objectives can be summarized as follows:

This study was supported by a research grant from


the National Science Council under the project
Journal of Quality in Maintenance Engineering number NSC 90-2218-E-007-003. The authors also
Volume 10 · Number 2 · 2004 · pp. 154-164 acknowledge the editor, Professor Kaio and the
q Emerald Group Publishing Limited · ISSN 1355-2511 anonymous reviewers for their helpful comments on
DOI 10.1108/13552510410539222 an earlier draft of this manuscript.
154
A stochastic maintenance management model Journal of Quality in Maintenance Engineering
Cheng-Hua Wang and Sheue-Ling Hwang Volume 10 · Number 2 · 2004 · 154-164

.
ensuring system function (availability, (Dey and Jaisingh, 1988; Berg, 1984). The
efficiency and product quality); available literature on discrete time maintenance
.
ensuring system life (asset management), models predominantly treats an equipment
ensuring safety; and deterioration process as a Markov chain, and
.
ensuring human well-being (British Standards similar approaches have also been discussed on the
Institution, 1984). development of maintenance system for road
surface maintenance (Pat-Cornell et al., 1987).
To accomplish the goals of maintenance
Most of these stochastic approaches predict the
management, maintenance management models
future state of equipment based only on the
are derived which cover four characteristics:
current condition without any regard to past
(1) A description of a technical system.
break-down, usage or maintenance activities.
(2) Modeling of deterioration of the system.
As to the qualitative models, Nowlan and Heap
(3) A description of the available information
(1978) first addressed the general description of
about the system and the actions in
reliability-centered maintenance (RCM). RCM
management.
was founded in the 1960s and initially oriented
(4) An objective function and an optimization
towards airplane maintenance. It is only now, more
technique which helps find the best balance
than 20 years later, that RCM has started to break
(Dekker, 1996).
through in many industries. It directs maintenance
In general, maintenance management models can effort to those parts and units where reliability is
be classified into quantitative models and critical. It can be regarded as the more qualitative
qualitative models as shown in Figure 1. approach to maintenance whereas mathematical
The quantitative models includes deterministic models are the quantitative approach (Anderson
models and stochastic models, such as linear and and Neri, 1990).
nonlinear programming, dynamic programming, Another important qualitative approach should
Markov decision methods, decision analysis be mentioned is total productive maintenance
techniques, search techniques, and heuristic (TPM), originating from Japan, which solves
approaches (Sherif, 1982). maintenance problems using a quality circles
The deterministic model was developed for method (Nakajima, 1986). This approach has the
minimizing the total cost of maintenance potential of providing almost a seamless
management and planning optimal maintenance integration of production and maintenance
schedule (Jardine, 1973; Gupta, 1981; Jayabalan through development of a strong partnership
and Chaudhuri, 1992). Stochastic models were (Maggard and Rhyne, 1992).
developed to address continuous operating As automated systems are getting sophisticated,
systems, as well as intermittent operating systems to fulfill systems’ goals during operation, these

Figure 1 Classification of maintenance management models

155
A stochastic maintenance management model Journal of Quality in Maintenance Engineering
Cheng-Hua Wang and Sheue-Ling Hwang Volume 10 · Number 2 · 2004 · 154-164

systems require extensive maintenance, and most maintenance time cycle and the number of
maintenance activities have to be operated by maintenance personnel on system availability
humans. Human factors in maintenance under the consideration of recovery factor. It is
management are considered in the following expected to find out the appropriate maintenance
functions: cycle and maintenance personnel allocation in
.
training maintenance personnel; order to improve the efficiency of system operation.
.
diagnosing and error recovering;
.
recording maintenance data; and
.
handling equipment (Dhillon, 1986).
2. Concepts of a stochastic maintenance
Figure 2 shows the human roles in the
maintenance management system.
management model
Quantitative models can simplify a realistic
The human errors can be classified into two types:
maintenance management situation so that the (1) Critical human error which will cause system
tasks can be managed easily, and system breakdown immediately.
performance can be predicted precisely. However, (2) Latent human error which does not lead to
there exists the gap between theory and practice. immediate system breakdown, and may be
The reasons are as follows: recovered by maintenance personnel in the
.
Maintenance mathematical models are system tolerance time period.
difficult to understand and to interpret. For
example, a Markov decision model is far more If the latent error is recovered, then the system can
difficult to understand and to interpret than a be operated normally. Hence, the concept of a
routing problem. human error recovery function that has been taken
.
Many papers have been written for into consideration in the system design may
mathematical purposes only. Mathematical improve equipment efficiency, reduces operation
results and structures of optimal equations, and maintenance cost, and increases the system
however, may not be easy to practice for users. availability (Dhillon, 1986; Dirickx and Kistner,
1979).
On the other hand, qualitative models may For improving the system availability, one can
describe the important concepts that are not easy to either decrease the human error rate in the
quantify. For example, the TPM emphasizes maintenance task, or positively improve the
smooth partner relationships for production and equipment to tolerate the slight latent human
maintenance performance, and human factors error. If a latent human error occurs during the
stress on human training, communication, and operating process and has been recovered in time,
error diagnosis tasks. However, the qualitative then the system can be operated continuously,
models lack objective numeric analysis to estimate otherwise, these human errors will lead to system
maintenance management performance precisely. breakdown. The system can be continuously
Therefore, in the present study, a model which operated if no error occurs during the maintained
integrates the stochastic process (quantitative operating process. However, either a critical
method) and human factors (qualitative method) is human error or equipment failure will lead to
proposed to predict the influence of the system breakdown.

Figure 2 Human in the maintenance management system

156
A stochastic maintenance management model Journal of Quality in Maintenance Engineering
Cheng-Hua Wang and Sheue-Ling Hwang Volume 10 · Number 2 · 2004 · 154-164

Since human error is not avoidable, one can either (3) State 2 – the system is being maintained.
make a greater effort in training the operators, (4) State 3 – the system has broken down due to
selecting appropriate operators to perform the equipment failure or critical human error or
task, or improving the mechanical equipment to latent human error not recovered in time.
reduce the human error, for example, increasing Once the system has broken down, it will be
the tolerance time of the machine so that the latent repaired at the system repair rate r.
human error in the system can be recovered in
In this model, a latent human error, a critical
time. In addition, since prevention is better than
human error, or equipment failure may occur
cure, the system availability can be improved by during the operation or maintenance period.
keeping up appropriate maintenance Either a critical human error or equipment failure
management. will lead to immediate system breakdown.
In the real world, the maintenance time is Nevertheless, as long as the latent human error has
usually considered to be continuous, while system been recovered in time, the system could be
state is discrete. Therefore, in this study, a operated normally. When the system has broken
continuous-time stochastic model and a down, it will be repaired eventually.
supplementary variable technique (Cox and In state 0, the system is operated normally. If at
Milles, 1968) are applied for finding the least one critical human error occurs or equipment
appropriate maintenance time cycle and number fails (Rc þ rm), then state 0 will change to state 3.
of maintenance personnel as well as the influence On the other hand, if at least one latent human
of the tolerance time on error recovery. error occurs (Rl), then state 0 will change to state
1. In addition, operation state (state 0) may change
to maintenance state (state 2) with maintenance
cycle rate l. In state 1, if a latent human error
3. Modeling process occurs and it has been found and recovered by the
operators, then state 1 will become state 0 with
Figure 3 is the transition diagram of the stochastic recovery rate G(a). When recovery time exceeds
maintenance management model, and the states the system tolerance time a, state 1 will change to
are defined as follows: state 3 with transient rate 1 2 GðaÞ. Furthermore,
(1) State 0 – the system is in normal operation. state 1 may change to state 2 with maintenance
(2) State 1 – a latent human error exists in the cycle rate l. In state 2, if the maintenance
system and the system is still in normal personnel do not make a latent human error or a
operation. critical human error, this state will change to

Figure 3 Transition chart of system operation

157
A stochastic maintenance management model Journal of Quality in Maintenance Engineering
Cheng-Hua Wang and Sheue-Ling Hwang Volume 10 · Number 2 · 2004 · 154-164

state 0. If at least one latent human error occurs, (4) In this system, the equipment’s failure rate
state 2 will change to state 1. If at least one critical rm and the critical human error rate rc are
human error occurs, state 2 will become state 3. In constant.
state 3, the system has broken down. It will be (5) Equipment breakdown and critical human
repaired with repair rate r, and become state 0. errors are mutually independent.
(6) Latent human error can be recovered at rate
G(a) in the time interval [0, a ], and the
3.1 Notation
system will be as new when it is repaired.
The symbols of this model are described as follows:
(7) System repair rate r is constant.
Pi(t): the probability of the system in state i at
(8) There are N persons who operate and
time t.
maintain this system.
P1(t,y): the probability of the system in state 1 when
(9) The maintenance personnel is parameter N
the latent human error occurred at time t
in this model.
and the recovery time has been used for
(10) Maintenance personnel are identical and
time y.
independent.
rl: individual latent human error rate.
(11) The system will be stopped when it is being
rc: individual critical human error rate.
maintained.
rm: the equipment failure rate.
(12) If latent human error exists in the system, it
r: the system repair rate.
will be recovered in the tolerant time
N: number of maintenance personnel who also
operate the system. interval.
Rl: the rate when at least one latent human (13) The maintenance task is parallel.
error occurs, and Rl ¼ 1 2 ð1 2 rlÞN .
Rc: the rate when at least one critical human 3.3 Analysis of model
error occurs, and Rc ¼ 1 2 ð1 2 rcÞN . In this model, the latent human error has to be
R: the rate when neither latent human error recovered and the supplementary variable
nor critical human error occurs, and technique is applied to solve this model. As the
R ¼ ½1 2 ðrl þ rcÞN . supplementary variable technique is applied to this
a: the upper limit of the system’s tolerance model, the model becomes a stochastic process.
time. The supplementary variable can help this model to
l: maintenance cycle rate which represents record the time period when the latent human
the maintenance frequency. error occurs as well as the state of the system.
S: Laplace transformation. The properties of this process are easily
g(y): the probability density function of the obtained directly, denoting by Pi(t), the probability
recovery time y. that state i is occupied at time t, and:
G(y): the cumulative distribution of g(y).  
pi ðt þ Dt Þ ¼ prob: state i at t þ Dtjstate i at t
ð1Þ
3.2 Assumptions
Thus we can specify the system as being in state 0,
The assumptions of this model are summarized as
in state (1, y), states 2, and 3, where state (1, y) is
follows:
occupied when the individual is in state 1 and has
(1) The human error can be separated into two
been there for time y. The coefficient of P1(t, y) is
forms:
.
Latent human error. The latent human error now the conditional probability of not broken-
may occur during state 0 or state 2 with the down given y in operating form:
same rate rl. The system may not break prob: ½state 1 occupied at t having
down after the latent human error occurs,  
been entered in t  y  Dy; t  y 
but the system will have broken down
p1 ðt;yÞ ¼ lim
when the recovery time exceeds the Dy!0 Dy
tolerance time of the system. ð2Þ
.
Critical human error. The critical human
error may occur during state 0 or state 2  
with the same rate rc. If the critical human pn;m t þ Dt; y þ Dtjt; y ¼ limþ
Dy!0
error exists in the system, it will cause
immediate system breakdown. prob:ðy þ Dt , Y , y þ Dt þ Dy; in state n at
(2) The failed system will not operate until it has t þ Dtjin state m at time t and Y ¼ yÞ
been repaired.
(3) The system will be fully repaired whenever it Dy
fails. ð3Þ
158
A stochastic maintenance management model Journal of Quality in Maintenance Engineering
Cheng-Hua Wang and Sheue-Ling Hwang Volume 10 · Number 2 · 2004 · 154-164

   
To perform mathematical calculations as follows: p xtþDt ¼ 3 ¼ p xtþDt ¼ 3; xt ¼ 3
   
þ p xtþDt ¼ 3; xt ¼ 0 þ p xtþDt ¼ 3; xt ¼ 1
   
PðxtþDt ¼ 0Þ ¼ PðxtþDt ¼ 0; xt ¼ 0Þ þ PðxtþDt ¼ 0; þ p xtþDt ¼ 3; xt ¼ 2 ¼ p xtþDt ¼ 3jxt ¼ 3
 
xt ¼ 1Þ þ PðxtþDt ¼ 0; xt ¼ 2Þ þ PðxtþDt ¼ 0; xt ¼ 3Þ £ pðxt ¼ 3Þ þ p xtþDt ¼ 3jxt ¼ 0 £ pðxt ¼ 0Þ
¼ PðxtþDt ¼ 0jxt ¼ 0Þ £ Pðxt ¼ 0Þ þ PðxtþDt ¼ Z aþDt
 
0jxt ¼ 2Þ £ Pðxt ¼ 2Þ þ PðxtþDt þ 0jxt ¼ 3Þ £ P þ p xtþDt ¼ 3jxt ¼ 1 £ g Y ðyÞdy
Z Dt a 
þ p xtþDt ¼ 3jxt ¼ 2 £ pðxt ¼ 2Þ:
ðxt ¼ 3Þ þ pðxtþDt ¼ 0; xt ¼ 1jY ¼ yÞ £ g Y ðyÞdy Z aþDt
0  
Z Dt Note p xtþDt ¼ 3jxt ¼ 1 £ g Y ðyÞdy
Note : PðxtþDt ¼ 0; xt ¼ 1jY ¼ yÞ £ g Y ðyÞdy  a 
0 ¼ p xtþDt ¼ 3jxt ¼ 1 £ p1 ðt; yÞ
¼ pðxtþDt ¼ 0jxt Þ ¼ 0Y ¼ yÞ £ P 1 ðt; yÞ: p3 ðt þ Dt Þ ¼ p3 ðtÞ £ ð1  r £ Dt Þ þ p0 ðtÞðrm þ RcÞ
Z aþDt
£ Dt þ p2 ðtÞ £ Rc £ Dt þ gðyÞdy þ 0ðDt Þ:
a
Equation (4) shows the probability of the system in
ð7Þ
state 0 and at time t þ Dt:
The boundary conditions are as follows:
p0 ðt þ DtÞ ¼ p0 ðtÞ £ ð1  rl £ DtÞ £ ½1  ðrm þ rcÞ
Z Dt p1 ðt; 0Þ ¼ rl £ p0 ðtÞ;
  
£ Dt þ gðyÞdy þ p2 ðtÞ £ 1  ½1  ðrc þ rmÞN } as t ¼ 0; then p0 ð0Þ ¼ 1; p1 o; y ¼ 0; p2 ð0Þ ¼ 0;
0
p3 ð0Þ ¼ 0:
£ Dt þ p3 ðtÞ £ Dt þ oðDtÞ:
ð4Þ
By equation (3), we have that:

Equation (5) shows the probability of the system in


›p1 ðt; yÞ ›p1 ðt; yÞ  
þ þ GðyÞ þ l
state 1, and recovery time has lasted Dt: ›t ›y ð8Þ
£ p1 ðt; yÞ ¼ 0;
 
p1 ðt þ Dt; y þ DtÞ ¼ p1;1 t þ Dt; y þ Dtjt; y
 
£ p1 ðt; yÞ þ p1;0 t þ Dt; y þ Dtjt; y £ p0 ðtÞ þ p1;2 p02 ðtÞ þ {R þ Rl þ 1  ½1  ðrl þ rcÞN }
  ð9Þ
ðt þ Dt; y þ DtÞ £ p2 ðtÞ; p1 ðt þ Dt; y þ DtÞ £ p2 ðtÞ ¼ P 1 t; y £ l þ P 0 ðtÞ £ l;
 
¼ p1 ðt; yÞ £ 1  gðyÞ £ Dt £ ½1  GðaÞ £ Dt 
£ ð1  l £ Dt Þ þ oðDt Þ: p03 ðtÞ þ r £ p3 ðtÞ ¼ p0 ðtÞ £ ðrm þ rcÞ
Note : p1 ð0; yÞ ¼ 0: Z
  ð10Þ
þ p2 ðtÞ £ r þ 1  gðyÞ £ P 2 ðt; yÞdy:
ð5Þ

Use the Laplace transform method to solve


Equation (6) shows the probability of the system in equations (8) to (10):
state 2, at time t þ Dt:  
sL p0 ðtÞ  p0 ð0Þ þ ðrl þ l þ rm þ rcÞ £ sL
         
P ¼ 2 ¼ P xtþDt ¼ 2; xt ¼ 2 p0 ðtÞ ¼ sL p3 ðtÞ £ 1  ð1  ðrl þ rcÞÞN

xtþDt
   Z a ð11Þ
þ P xtþDt ¼ 2; xt ¼ 0 þ P xtþDt ¼ 2; xt ¼ 1
  þ gðyÞ £ sLðp1 ðt; yÞÞdy;
¼ P xtþDt ¼ 2jxt ¼ 2 £ P ðxt ¼ 2Þ 0
 
þ P xtþDt ¼ 2jxt ¼ 0 £ P ðxt þ 0Þ
    ›p1 ðs; yÞ  
þ P xtþDt ¼ 2jxt ¼ 1 £ P ðxt ¼ 1Þ; sL p1 ðt; yÞ þ  p1 ð0; yÞ þ gðyÞ þ GðaÞ þ l
p2 ðt þ Dt Þ ¼ p2 ðtÞ £ ð1  Rl £ Dt Þ £ ð1  R £ Dt Þ ›y

  £ sL p1 ðs; yÞ ¼ 0;
£ , 1  1  ½1  ðrm þ rcÞN £ Dt . þp1 ðtÞ
ð12Þ
£ l £ Dt þ p0 ðtÞ £ l £ oðDt Þ:
ð6Þ  
sL p2 ðtÞ  p2 ð0Þ þ Rc þ Rl
 
þ f1  ½1  ðrl þ rcÞN g £ sL p2 ðtÞ ð13Þ
Equation (7) shows the probability of the system in     
state 3, at time t þ Dt: ¼ l sL p0 ðtÞ þ sL P 1 ðt; yÞ ;

159
A stochastic maintenance management model Journal of Quality in Maintenance Engineering
Cheng-Hua Wang and Sheue-Ling Hwang Volume 10 · Number 2 · 2004 · 154-164

    maintenance personnel and maintenance cycle on


sL p3 ðtÞ  p3 ð0Þ þ r £ sL p3 ðtÞ
  the recovery time can be predicted.
¼ sL p0 ðtÞ £ ðrm þ rcÞ þ sL½p2 ðtÞ £ Rc
Z 1 ð14Þ A total of 40 students from the Industrial
    Engineering Department in National Tsing Hua
þ 1  gðyÞ £ sL p1 ðt; yÞ dy:
a University participated in this experiment. Among
these 40 subjects, 37 subjects were male. In this
Summary: simulated system (Figure 4), there are two paths
p0 ðsÞ ¼ fðs þ rl þ rm þ rc þ lÞ connecting every tank (pump 1 carries water to
tank 1 and tank 2, pump 2 carries water to tank 1
M3 r l
 £ l £ Q3ðsÞ þ £ ðrm þ rcÞ þ and tank 3, and pump 3 carries water to tank 2 and
M1 sþr M1 tank 3). Every pipeline has two control valves, one
 
£ Q3ðsÞ þ rl £ M2 þ Q1ðsÞ þ GðaÞg1 ; automatic and the other manual.
The experiment comprises two parts. The first
 part was designed to estimate the human error
1
p1 ðsÞ ¼ rl £ ½1  expðsa  la  a rate. There were 15 tasks randomly appearing in
s þ l þ GðaÞ
 the monitor for the subject to operate. Four tasks
£ GðaÞ  N1ðsÞ £ p0 ðsÞ; are infusing water into one tank, six tasks are
infusing water into two tanks and five tasks are
maintaining the pump or valve. The subject
 Z Dy
 controls the pumps and valves depending on the
l situation portrayed in the monitor.
p2 ðsÞ ¼ ð1 þ rl Þ þ Q1ðsÞ £ exp  gðyÞdy ;
M1 0 The second part was designed to obtain the
detection and recovery time when the latent
human error occurred in the system. In the
l beginning, some pumps and valves were
p3 ðsÞ ¼ £ ðrc þ rmÞ þ p2 ðsÞ þ rl
sþr improperly opened, and the subject had to close

those being opened incorrectly.
£ M2 £ Q1ðsÞ :

Note:
5. Result
M1 ¼ {s þ Rc þ Rlz; þ 1  ½1  ðrl þ rcÞN };
M2 ¼ expðla  aGðaÞ þ aÞ; Data (168 units) collected from the reference feed-
water system experiment, were used to calculate
M3 ¼ 1  ½1  ðrl þ rcÞN ; the parameters of the mathematical equations.
Z a
    Then, from a simplified mathematical model,
N1ðsÞ ¼ exp sy  ly  GðaÞ £ yÞ £ GðyÞ dy;
different combinations of the maintenance
Z 01
    personnel and the maintenance cycle in each
N2ðsÞ ¼ exp sy  ly  GðaÞ £ yÞ £ gðyÞ dy; degree of system availability can be presented.
a
Z 1 Z Dt
From the result of the data analysis by using the
 
Q1ðsÞ ¼ 1  gðyÞ exp  gðyÞdy statistical package STATISTIC, the recovery time
a 0 (g(y)) fits the gamma distribution where
£ expðy  syÞdy; a ¼ 2:81753, b ¼ 0:138002.To avoid the effect of
 
Q2ðsÞ ¼ exp sy  ly  GðaÞy ; the outlier value, the system tolerance time is
Z Dt
decided by the median of the experimental data. In
Q3ðsÞ ¼ ð1 þ rl Þ þ Q2ðsÞ £ exp  gðyÞdy : the reference feed-water system, the tanks and the
0 channels are fixed, and the subjects operated the
pumps and the valves, hence it was assumed that
4. Experiment the equipment failure rate came from the pump or
the valve. Referring to the data of the WASH-1400
An experiment is conducted by simulating the (US NRC, 1964), the equipment failure rate is set
situation where an operator controls a reference at 0.002. If the system has broken down, it can be
feed-water system of a nuclear power plant. The repaired immediately, hence let the repair rate be
purpose of this experiment is to obtain the raw data 1, rc ¼ 0:0033 (the rate of the critical human
of the probability density function g(y), and some error), rl ¼ 0:16 (the rate of the latent human
parameters in the model, which contains rl and rc. error), rm ¼ 0:002 (the rate of the equipment
These estimated parameters simplify the breakdown), r ¼ 1 (the repair rate of the system),
mathematical equations of the stochastic model, a ¼ 19s (the upper limit of the system tolerance
hence the influence of different numbers of time).
160
A stochastic maintenance management model Journal of Quality in Maintenance Engineering
Cheng-Hua Wang and Sheue-Ling Hwang Volume 10 · Number 2 · 2004 · 154-164

Figure 4 A reference feed-water system

The system states probabilities are as follows: As the parameters obtained from the
experiment are placed in the analytic solution, we
P o ðt ! 1Þ ¼ t!1
lim P 0 ðt Þ ¼ lim sP o ðsÞ found a significant relationship between the
s!0
number of maintenance personnel and the system
availability, and between maintenance cycle and
P 1 ðt ! 1Þ ¼ t!1
lim P 1 ðt Þ ¼ lim sP 1 ðsÞ system availability. When the number of
s!0
maintenance personnel is under three, the system
availability increases by the increase of the
P 2 ðt ! 1Þ ¼ t!1
lim P 2 ðt Þ ¼ lim sP 2 ðsÞ maintenance cycle. However, when the number of
s!0
maintenance personnel is over seven, the system
availability decreases by the increase of
maintenance cycle. When the maintenance
In the steady state, the system availability is:
personnel is between four and six, the availability
AV ¼ t!1
lim ½P 0 ðt Þ þ P 1 ðt ÞP 2 ðt Þ: curve slightly shows a “U” pattern (Figure 5).
Using the data of the reference feed-water
experiment, the optimal solution for the model is
The analytic solution of this model is complex, to dispose two maintenance personnel in the
therefore, some software packages (MATLAB, system under frequent maintenance cycle, so that
EXCEL, STATISTICAL) are used to solve the the system availability will be over 0.95 (Figure 6).
model. By means of software computation, the Limitations have to be considered when the
system availability can be predicted by the different frequent maintenance cycle is selected. One is that
combinations of the number of maintenance the frequent maintenance cycle means the system
personnel and maintenance cycle rate. Table I, efficiency is low. The other is that the frequent
column 1 represents the maintenance cycle (from maintenance cycle results in higher maintenance
0.05 to 1), increasing the value of l means that the costs.
maintenance cycle becomes more frequent. Row 1
represents the different number of maintenance
personnel (from 1 to 10). Various combinations of 6. Discussion
maintenance cycle rate and maintenance personnel In general, the aim of the maintenance
number could create predictive system availability. management mathematical models is to find the
161
A stochastic maintenance management model Journal of Quality in Maintenance Engineering
Cheng-Hua Wang and Sheue-Ling Hwang Volume 10 · Number 2 · 2004 · 154-164

Table I System availability on various maintenance cycle rate (l) and number of maintenance personnel (n)
N
l n51 n52 n53 n54 n55 n56 n57 n58 n59 n 5 10
0.05 0.840602 0.839029 0.83312 0.827104 0.824906 0.823446 0.822433 0.821694 0.821138 0.820707
0.1 0.846787 0.845839 0.834169 0.822306 0.817924 0.815016 0.812996 0.811524 0.810415 0.809557
0.15 0.852512 0.851878 0.834563 0.817046 0.810487 0.806146 0.803127 0.800926 0.799268 0.797986
0.2 0.857846 0.857481 0.834614 0.811644 0.802929 0.797157 0.793143 0.790218 0.788013 0.786309
0.25 0.862845 0.862918 0.834569 0.80635 0.795484 0.788294 0.783292 0.779646 0.776899 0.774775
0.3 0.867555 0.868401 0.834626 0.801352 0.78834 0.779745 0.77376 0.769398 0.76611 0.763569
0.35 0.872014 0.874093 0.834933 0.796793 0.781656 0.771652 0.764687 0.75961 0.755785 0.752828
0.4 0.876254 0.88012 0.835606 0.792775 0.775519 0.76411 0.75617 0.750382 0.746021 0.742649
0.45 0.880301 0.886574 0.836728 0.789364 0.76999 0.757188 0.748275 0.741779 0.736883 0.733098
0.5 0.884176 0.893518 0.838356 0.786606 0.765116 0.750927 0.741045 0.733841 0.728413 0.724216
0.55 0.884774 0.900999 0.840532 0.784526 0.760931 0.745351 0.734501 0.726591 0.720631 0.716023
0.6 0.89148 0.909043 0.843276 0.783132 0.757437 0.740467 0.72865 0.720037 0.713545 0.708527
0.65 0.894938 0.917664 0.846602 0.782422 0.754578 0.736272 0.72349 0.714173 0.707151 0.701723
0.7 0.898279 0.926867 0.850509 0.782389 0.752485 0.732755 0.719008 0.708988 0.701436 0.695598
0.75 0.901516 0.936648 0.854993 0.783013 0.751017 0.729897 0.715187 0.704464 0.696383 0.690135
0.8 0.904655 0.946999 0.860045 0.784276 0.750173 0.727679 0.712006 0.70058 0.691969 0.685312
0.85 0.907704 0.957904 0.865648 0.786152 0.750171 0.726076 0.709438 0.69731 0.68817 0.681103
0.9 0.910668 0.969347 0.871785 0.788961 0.750317 0.725061 0.70746 0.694629 0.684959 0.677483
0.95 0.913551 0.98131 0.878437 0.791648 0.751254 0.724609 0.706044 0.69251 0.68231 0.674425
1 0.91636 0.993771 0.885583 0.795215 0.752712 0.724692 0.705163 0.690926 0.680197 0.671902

Figure 5 The relationship between system availability and number of maintenance personnel

optimum balance between the costs and benefits of and efficient schedules and plans. Maintenance
maintenance. Quantitative models focus on actions may be performed either at discrete times
numeric analysis, and they can lead to the or continuously. Many researchers use a stochastic
following four outcomes. First of all, maintenance process to formulate effective maintenance
policies can be evaluated and compared management actions to reduce equipment failures
considering cost effectiveness and system and derive optimal maintenance policies (Murthy
availability. Second, maintenance performance and Hwang, 1996; Dedopoulos and Shah, 1995).
can be obtained from the structure of optimal In the simulated reference feed-water system
policies. Third, models can assist in the timing experiment, the latent human error rate reaches
aspect: how often to inspect or maintain. Finally, 0.16. If the latent error has not been recovered in
models can also be of help in determining effective time, it may seriously diminish system maintenance
162
A stochastic maintenance management model Journal of Quality in Maintenance Engineering
Cheng-Hua Wang and Sheue-Ling Hwang Volume 10 · Number 2 · 2004 · 154-164

Figure 6 The relationship between system availability and maintenance cycle

performance. The recovery action can be viewed as As to the advantages of qualitative models, the
an immediate maintenance action, and recovering current qualitative models deal with maintenance
the human latent errors is valuable for maintenance quality and system availability, especially those not
management. In addition, from this experiment, easily quantified by strict mathematical models.
the human error types and human latent error This paper dealt with a model that incorporates
recovery time are obtained. To classify the human equipment failure rate and human error rate by
errors and to fit the distribution function of recovery using a stochastic process and the supplementary
time are very helpful for designing a real variables estimated from a simulated experiment
maintenance management system. For preventing to solve the optimal maintenance personnel
critical human error in the maintenance allocation and maintenance cycle. This integral
management task, the recovery factor should be quantitative and qualitative model provides a
considered especially with respect to maintenance simple and practical solution to the maintenance
personnel training. problem. With the exception of the maintenance
Generally, the recovery function is designed in cycle and maintenance personnel, other
the operation system for the operator to improve parameters were obtained by the simulated
the operation performance (Dhillon, 1986). In this experiment. Thus, this maintenance management
paper, the recovery factor has been taken into model could reduce the complexity of the previous
account in the model of maintenance management mathematical models, and it can be applied to
so that the system availability can be estimated various maintenance management fields as long as
more precisely. the related parameters are obtained.
Most results of research on maintenance In this model, system availability is used as a
personnel allocation, maintenance cycle time performance evaluation criterion and cost is not
formulation, policies are subject to cost (Jayabalan considered although it is an important aspect of
maintenance management. Hence, for the further
and Chaudhuri, 1992; Dohi et al., 1998;
study, this model can be modified to estimate the
Gopalakrishnan et al., 1997). In this study, the
maintenance management total cost.
system availability is a criterion for allocating
maintenance personnel and formulating the
maintenance cycle time. Based on the criterion, the
optimal combination of the number of maintenance 7. Conclusion
personnel and maintenance cycle can be found for
improving maintenance management performance. The analytic solution of the stochastic model in
In other words, too many maintenance personnel or this study is as complex as the general
too frequent maintenance would not increase the mathematical maintenance management model.
system availability. Nevertheless, some parameters in the model such as
163
A stochastic maintenance management model Journal of Quality in Maintenance Engineering
Cheng-Hua Wang and Sheue-Ling Hwang Volume 10 · Number 2 · 2004 · 154-164

human error rate and a system’s tolerance time were Dedopoulos, I.T. and Shah, N. (1995), “Preventive maintenance
obtained from a simulated reference feed-water policy optimization for multipurpose plant equipment”,
system experiment, and that significantly reduced Computers & Chemical Engineering, Vol. 19, pp. 693-8.
Dekker, R. (1996), “Application of maintenance optimization
the complexity of this analytic solution model.
model: a review and analysis”, Reliability Engineering and
Furthermore, a recovery factor was incorporated System Safety, Vol. 51, pp. 229-40.
into the maintenance management task that more Dey, D.K. and Jaisingh, L.R. (1988), “Estimation of system
precisely estimated system availability. availability for independent series components with
Mathematical maintenance management Weibull life distributions”, IEEE Transactions on Reliability,
models deal objectively with maintenance Vol. 37 No. 4, pp. 401-5.
schedule, cost, and policy, whereas qualitative Dhillon, B.S. (1986), Human Reliability with Human Factors,
Pergamon Press, New York, NY.
maintenance management models focus on Dirickx, Y.M.I. and Kistner, K.P. (1979), “Reliability of a repairable
personnel training, partner relationships, and system with redundant units and preventive maintenance”,
maintenance data collection which are not easily IEEE Transactions on Reliability, Vol. 28 No. 2, pp. 170-1.
quantified by a strict mathematical model. Dohi, T., Kaio, N. and Osaki, S. (1998), “Minimal repair policies for
Combining the advantages of both approaches, an economic manufacturing process”, Journal of Quality in
this study derives an integrated objective Maintenance Engineering, Vol. 4 No. 4, pp. 248-62.
quantitative analysis method and a reasonable Dohi, T., Ashoko, A., Kaio, N. and Osaki, S. (2001), “Optimizing
the repair-time limit replacement schedule with
qualitative maintenance management model to discounting and imperfect repair”, Journal of Quality in
evaluate maintenance management performance, Maintenance Engineering, Vol. 7 No. 1, pp. 71-84.
and provides a more active concept of the recovery Gopalakrishnan, M., Ahire, S.L. and Miller, D.M. (1997),
factor to increase system availability. The “Maximizing the effectiveness of a preventive
contributions of this paper are as follows: maintenance system: an adaptive modeling approach”,
.
integrates the advantages of a quantitative and Management Science, Vol. 43 No. 6, pp. 827-40.
qualitative maintenance management model Gupta, A. (1981), “Reduced tree search for optimal preventive
maintenance of generating facilities”, IEEE Transactions
for evaluating the maintenance performance on Reliability, Vol. R-30, pp. 476-7.
precisely; Jardine, A.K.S. (1973), Maintenance Replacement and Reliability,
.
provides a system availability table of the Pitman Publishing, London.
various numbers of maintenance personnel Jayabalan, V. and Chaudhuri, D. (1992), “Cost optimization of
and maintenance cycle time which can be a maintenance scheduling for a system with assured
practical reference for the maintenance reliability”, IEEE Transactions on Reliability, Vol. 41 No. 1,
manager; and pp. 21-5.
Maggard, B.N. and Rhyne, D.M. (1992), “Total productive
.
incorporation of a recovery function into
maintenance: a timely integration of production and
maintenance management task, which can maintenance”, Production and Inventory Management
prevent system breakdown and improve the Journal, pp. 6-10.
system availability. Murthy, D.N.P. and Hwang, M.C. (1996), “Optimal discrete and
continuous maintenance policy for a complex unreliable
Further study can be investigated in the following machine”, International Journal of Systems Science,
directions: Vol. 27 No. 5, pp. 483-95.
.
applying the technology of multiple criteria Nakajima, S. (1986), “TPM challenge to the improvement of
decision making so that system availability, productivity by small group activities”, Maintenance
maintenance cost and maintenance Management International, Vol. 6, pp. 52-6.
performance can be traded-off more Nowlan, F.S. and Heap, H.F. (1978), Reliability-centered
appropriately; and Maintenance, Technical Report AD/A066-579, Dolby
Access Press, San Francisco, CA.
.
based on the present model, maintenance Pat-Cornell, M.E., Lee, H.L. and Tagaras, G. (1987), “Warning of
management decision support systems can be malfunctions: the decision to inspect and maintain
developed for on-line maintenance processes on schedule or on demand”, Management
management. Science, Vol. 33 No. 10, pp. 1277-90.
Sherif, Y.S. (1982), “Reliability analysis: optimal inspection and
maintenance schedules of failing systems”, Micro-
References Reliability, Vol. 22, pp. 59-115.
US Nuclear Regulatory Commission (NRC) (1964), Reactor Safety
Anderson, R.T. and Neri, L. (1990), Reliability-Centred Study: An Assessment of Accident Risks in US Commercial
Maintenance – Management and Method, Elsevier Nuclear Power Plants, WASH-1400 (NUREG-75/014), US
Science Publishing, New York, NY. NRC, Washington, DC.
Berg, M. (1984), “A preventive replacement policy for units
subject to intermittent demand”, Operations Research,
Vol. 32, pp. 584-95.
British Standards Institution (1984), BS 3811 Glossary of Further reading
Maintenance Terms in Tero-technology, British Standards
Institution, London. Canfild, R.V. (1986), “Cost optimization of periodic preventive
Cox, D.R. and Milles, H.D. (1968), The Theory of Stochastic maintenance”, IEEE Transactions on Reliability, Vol. 35,
Process, John Wiley & Sons, New York, NY. pp. 78-81.
164

Das könnte Ihnen auch gefallen