Professional Documents
Culture Documents
Linderman
Linderman
www.elsevier.com/locate/dsw
Abstract
Organizational leaders increasingly recognize process management as an essential element in organizational per-
formance. Two key tools for process management––Statistical Process Control and Maintenance Management––can
create profound economic benefits, particularly when they are coordinated. This paper demonstrates the value of
integrating Statistical Process Control and maintenance by jointly optimizing their policies to minimize the total costs
associated with quality, maintenance, and inspection. While maintenance is often scheduled periodically, this analysis
encourages ‘‘adaptive’’ maintenance where the maintenance schedule accelerates when the process becomes unstable.
This paper presents a number of models to demonstrate the economic behavior and value of coordinating process
control and maintenance. Finally, a sensitivity analysis is conducted to develop insights into the economic and process
variables that influence the integration efforts.
Ó 2004 Published by Elsevier B.V.
1. Introduction
Organizational leaders continually confront the challenge of improving the performance of their orga-
nizationsÕ assets. General Electric, Motorola, Miliken, Procter and Gamble, Ford, and Dupont, among
others, incorporate process management into their business strategies. In 1999, General Electric invested
half a billion dollars in process improvement initiatives and expects more than $2 billion in benefits for the
fiscal year (Pande et al., 2000). The strategic nature of process management forces organizations to develop
integrated management systems, which requires an increased level of coordination among management
practices (Lee et al., 1999).
*
Corresponding author. Tel.: +1-612-626-8632; fax: +1-612-624-8804.
E-mail addresses: klinderman@csom.umn.edu (K. Linderman), kmckone@babson.edu (K.E. McKone-Sweet).
1
Tel.: +1-781-239-4245.
Nomenclature
While practitioners and academics recognize the relationship between product quality, process quality,
and equipment maintenance (Tsuchiya, 1992; McKone et al., 1998), research in integrating these efforts has
been limited. Academics focus mostly on quality models (see reviews in Porteus and Angelus, 1997; Ho and
Case, 1994; Kolesar, 1993; Montgomery, 1980) or maintenance models (see reviews in McKone and Weiss,
1998; Valdez-Flores and Feldman, 1989; Pierskalla and Voelker, 1976; McCall, 1965) in isolation. Cassady
et al. (2000) performed a preliminary investigation on the integration of maintenance and Statistical Process
Control. They developed a Simulation-Optimization Model to minimize total-quality costs by simulta-
neously optimizing maintenance and process-control policies. Cassady et al. (2000) further suggested that
more research needs to be conducted on integrating these practices.
326 K. Linderman et al. / European Journal of Operational Research 164 (2005) 324–340
In this research, we develop a generalized analytic model to determine the optimal policy to coordinate
Statistical Process Control and Planned Maintenance to minimize total expected cost. We propose inte-
grating these two management practices into a coordinated Monitoring-Maintenance Model. In our model,
Statistical Process Control monitors the equipment and provides signals indicating equipment deteriora-
tion, while Planned Maintenance is scheduled at regular intervals to preempt equipment failure. The
determination of an unstable process, via Statistical Process Control, results in an early Reactive Main-
tenance to restore the equipment. Otherwise a Planned Maintenance occurs after a specified period of
operation. In this sense, we propose an ‘‘adaptive’’ maintenance policy, where the maintenance schedule
adapts to the stability of the process. The completion of maintenance returns the process to its original
operating condition and results in a process renewal. The model derives and optimal policy to minimize the
cost per unit time. The derivation of the coordinated cost function follows a renewal-reward process (Ross,
1996, p. 132).
The proposed Monitoring-Maintenance Model follows the framework in Fig. 1. The framework requires
the joint determination of the Planned Maintenance and Statistical Process Control parameters and reveals
the assertion that adaptive maintenance ought to be done as an informed event, given the state of process
variability and more specifically process stability.
The process begins in an in-control state with a Process Failure Mechanism 2 that follows a Weibull
m
distribution (f ðtÞ ¼ km mtm1 eðktÞ where k; m; t P 0). (Dodson (1994) and Hopp and Spearman (1996) describe
the importance of the Weibull distribution in characterizing equipment failures, Banerjee and Rahim (1988)
also utilize a Weibull process failure mechanism, McWilliams (1989) further illustrates the applicability of
the Weibull distribution to a wide range of industrial settings.) Assume that process inspections occur after
h hours of production to determine whether the process has shifted from an in-control to an out-of-control
state. The quality characteristic is measured and plotted on a control chart to assess the state of the process.
If the control chart does not signal an out-of-control condition after k inspection intervals, then scheduled
or Planned Maintenance occurs at the (k þ 1)th sampling interval. However, if the control chart signals an
out-of-control condition at any of the j inspections, a search for an assignable cause takes place to validate
2
Process Failure Mechanism is the distribution function that characterizes the length of time that the process remains in the in-
control state.
K. Linderman et al. / European Journal of Operational Research 164 (2005) 324–340 327
the signal. A valid control-chart signal then results in Reactive Maintenance. We assume that the com-
pletion of Planned or Reactive maintenance restores the equipment to a ‘‘good-as-new’’ condition (a re-
newal). Fig. 1 illustrates all possible scenarios that can occur when cycling between process monitoring and
maintenance.
Fig. 2a illustrates the timeline of events that describes Scenario 1 from Fig. 1. Assume that the process
begins in-control and inspections occur after h hours of production to determine whether the process has
shifted from an in-control to an out-of-control state. Sometime between the jth and (j þ 1)th sampling
interval an assignable cause occurs and the process shifts to an out-of-control state. The process continues
to operate; however, the control chart does not detect an out-of-control condition until the (j þ i)th sample.
A time lag is associated with collecting the data and plotting the results on the chart. The control chart then
signals an out-of-control condition and a search for an assignable cause takes place to validate the signal. A
valid control-chart signal then results in Reactive Maintenance that restores the equipment to a ‘‘good-as-
new’’ condition (a renewal).
In Scenario 2 in Fig. 1, the process shifts to an out-of-control state, but the control chart does not
signal an out-of-control condition before the Planned Maintenance. Fig. 2b illustrates the timeline of
events that describes this scenario. As in Scenario 1, the process begins in an in-control state and
sometime between the jth and (j þ 1)th sampling interval, the process shifts to an out-of-control state.
However, the process continues to operate because the control chart does not detect an out-of-control
condition. At the (k þ 1)th sampling interval, maintenance begins, and the out-of-control state is iden-
tified. We consider this to be Reactive Maintenance because the out-of-control condition occurred before
the scheduled maintenance and additional time and expense will be incurred to identify and resolve the
equipment problem. Completing the maintenance causes the process to renew and return to the in-control
state.
In Scenario 3 in Fig. 1, the process still remains in an in-control state at the time of the Planned
Maintenance. Fig. 2c illustrates the timeline of events that describes this scenario. In this scenario, the
process continues in an in-control state the entire time. The Planned Maintenance takes place at the
(k þ 1)th sampling interval to preempt a process failure. Typically, the activities associated with planned
maintenance are less costly than those associated with Reactive Maintenance because preparation activities
can be conducted off-line before the Planned Maintenance. Finally, after maintenance completion, the
process renews itself.
This general approach assumes that the process monitoring and maintenance schedule follows a rolling
schedule. When the process undergoes either Reactive or Planned Maintenance, the next planned main-
tenance is scheduled to occur after ðk þ 1Þh hours of operation. The Planned Maintenance, of course, can
be preempted by Reactive Maintenance if an out-of-control condition is determined. Both Reactive and
Planned Maintenance restore the process and the cycle begins again.
We base the coordinated policy for the Monitoring-Maintenance Model on the total relevant costs
associated with inspection, maintenance, and quality. Model development follows a renewal-reward process
in which a renewal occurs at the completion of maintenance. Therefore, the long-run average cost per unit
time can be expressed as the expected cost per cycle divided by the expected time of cycle (Ross, 1996,
p. 132). The expected cycle time and cycle cost for each of the three scenarios follows.
328 K. Linderman et al. / European Journal of Operational Research 164 (2005) 324–340
Control Chart
th
j (j+1)th (j+2)th (j+i)th
Sample Sample Sample Sample
Control Chart
In control
(b) Out of control Period
Period
Control Chart
Maintenance
Cycle Scheduled
Begins . . . Maintenance
Completed
In control Period
(c)
Fig. 2. Diagrams of three monitoring-maintenance scenarios: (a) Scenario 1––out-of-control state detected, (b) Scenario 2––out-
of-control state not detected, (c) Scenario 3––in-control state, planned maintenance performance.
K. Linderman et al. / European Journal of Operational Research 164 (2005) 324–340 329
state and the control chart detects the out-of-control state during the first k inspections. The expected in-
control time consists of the mean time to failure and the expected amount of time investigating false alarms,
Z kh m
km mtm eðktÞ sT0
E½In-Control TimejS1 ¼ ðkðkþ1ÞhÞ m dt þ ð1 c1 Þ :
0 1e ARL0
truncated Weibull distribution.) Therefore, the expected in-control time resembles the expected in-control
time for Scenario 1 except for the indices of integration. The equation follows:
Z ðkþ1Þh m
km mtm eðktÞ sT0
E½In-Control TimejS2 ¼ ðkðkþ1ÞhÞ m dt þ ð1 c1 Þ :
0 1e ARL0
The derivation of the expected out-of-control time consists of the difference between the scheduled
maintenance time and the expected time until the process shifts out of control, plus the time to perform a
Reactive Maintenance. The mathematical description follows:
Z ðkþ1Þh m
km mtm eðktÞ
E½Out-of-Control TimejS2 ¼ ðk þ 1Þh m dt þ T2 :
0 1 eðkðkþ1ÞhÞ
Now, the expected cycle time, E½Cycle TimejS2, is the sum of the expected in-control and out-of-control
times.
(Recall, c2 is an indicator variable which indicates if production occurs during a Reactive Maintenance.)
Because the control chart does not give an out-of-control signal, the cost of sampling cost consists of the
product of the sampling costs and the number of samples taken, E½Cost of SamplingjS2 ¼ ðCF þ CV nÞk.
Finally, the cost of false alarms and performing a Reactive Maintenance is given as
sCY
E½Cost of AlarmsjS2 ¼ ARL 0
þ CR . Now, the sum of the above costs (the costs of quality loss, sampling,
false alarms, and maintenance) provides the expected cost per cycle for Scenario 2, E½Cycle CostjS2.
Because we assume it is a renewal-reward process, the following equation provides the expected cost per
hour for the Monitoring-Maintenance Model:
K. Linderman et al. / European Journal of Operational Research 164 (2005) 324–340 331
where
ESPC&PM ½Cycle Cost ¼ E½Cycle CostjS1P ½S1 þ E½Cycle CostjS2P ½S2 þ E½Cycle CostjS3P ½S3;
ESPC&PM ½Cycle Time ¼ E½Cycle TimejS1P ½S1 þ E½Cycle TimejS2P ½S2 þ E½Cycle TimejS3P ½S3
and
P ½S1 ¼ F ðkhÞP ðSignalingjout-of-control stateÞ;
P ½S2 ¼ F ððk þ 1ÞhÞ F ðkhÞP ðSignalingjout-of-control stateÞ;
P ½S3 ¼ 1 F ððk þ 1ÞhÞ:
P ðSignalingjout-of-control stateÞ is the probability that the control-chart signals given that the process has
shifted to an out-of-control state. The actual computation of this probability depends on the type of control
chart and out-of-control criteria used. For example, the probability that an X chart signals when seven
consecutive sample means fall above the center line will be different from the probability that an X chart
signals when a single point falls outside the control limits. F ðtÞ is the cumulative distribution function for
the Process Failure Mechanism. As a result, F ððk þ 1ÞhÞ gives the probability that the process shifts out of
control before the Planned Maintenance.
In the Planned Maintenance Model, the decision variable PM defines the Planned Maintenance interval,
which is similar to ðk þ 1Þh in the coordinated model. In this model, no inspections occur. When the process
shifts to an out-of-control state, it continues to operate in an out-of-control state until the scheduled
maintenance time. The length and cost of the maintenance is, however, dependent on the condition of the
process at the time of the scheduled maintenance. The derivation of the cycle time and cycle cost follows:
EPM ½Cycle Time ¼ PM þ T2 F ðPMÞ þ T3 f1 F ðPMÞg;
Z PM Z PM
m m
EPM ½Cycle Cost ¼ CI km mtm eðktÞ dt þ CO PM km mtm eðktÞ dt þ c2 T2 þ CR F ðPMÞ
0 0
The derivation of the Statistical Process Control Model follows from Scenario 1 in the Monitoring-
Maintenance Model. When k approaches infinity in the Monitoring-Maintenance Model, it reduces to the
Statistical Process Control Model. The derivation of the cycle time and cycle cost follows. Mathematically,
the derivation of this model follows from Scenario 1 by substituting s0 with s and s0 with s (see Appendix C).
(Note, the above formulation also assumes that T0 < h when c1 ¼ 0.)
Z 1
m s0 T0
ESPC ½Cycle Time ¼ km mtm eðktÞ dt þ ð1 c1 Þ s0 þ nE þ hðARL1 Þ þ T1 þ T2 ;
0 ARL 0
Z 1
m s0 C Y
ESPC ½Cycle Cost ¼ CI km mtm eðktÞ dt þ CO ½hðARL1 Þ s þ nE þ c1 T1 þ c2 T2 þ þ CR
0 ARL0
R 1 m m ðktÞm
k mt e dt þ hðARL1 Þ s þ nE þ c1 T1 þ c2 T2
þ ðCF þ CV nÞ 0 :
h
In this Statistical Process Control Model, minimizing (3) determines the optimal policy variables (the
sampling interval, sample size, and width of control-chart limits).
ESPC ½Cycle Cost
ESPC ½Hourly Cost ¼ : ð3Þ
ESPC ½Cycle Time
The general Monitoring-Maintenance Model applies to a variety of control charts and out-of-control
criteria. See Ryan (2000) and Montgomery (2001) for details about different control charting schemes and
associated average run length performance. To demonstrate the use of our coordinated model, we apply the
model to an X chart that signals an out-of-control condition when a sample mean falls outside the control
limits. This time-honored chart has received a lot of attention by researchers and practitioners and serves as
the foundation to modern Statistical Process Control. For this reason, we use the X chart to illustrate
coordinating process control and maintenance. The X chart formulation follows.
Assume that when the process operates in-control, the quality characteristic follows a normal distri-
bution with an average measurement of l0 and standard deviation r. An out-of-control state results in a
shift in the process mean by d standard deviations from l0 to l1 ¼ l0 þ dr. Further assume that the process
can only be returned to the in-control state through a maintenance intervention. The X chart monitors the
process where a sample size n is collected after h hours of operation. The upper and lower control limits are
set at L standard deviations from the mean. The control chart signals an out-of-control condition when the
sample mean exceeds the control limit. The general model applies to an X chart by making the following
substitutions: ARL0 ¼ 1=a, ARL1 ¼ 1=ð1 bÞ, P ðSignalingjout-of-control stateÞ ¼ 1:0 Pkl j¼1 b, where
& R hk ’
m m ðktÞm
0
k mt e dt
l¼
h
(d e is the next largest integer.)
6. Sensitivity analysis
A sensitivity analysis promotes our understanding of the Monitoring-Maintenance model and illustrates
the importance of making coordinated decisions. In addition, comparisons are made between the Moni-
K. Linderman et al. / European Journal of Operational Research 164 (2005) 324–340 333
toring-Maintenance Model and using Process Control or Planned Maintenance in isolation. This investi-
gation develops insights into the impact of the model parameters and their effects on coordination.
We conduct a Design of Experiments (DOE) study to build model intuition and explore the sensitivity of
the model parameters with optimal policy variables (n; h; L; k) and hourly cost. The economic Monitoring-
Maintenance Model contains 15 input parameters and three indicator variables that determine the cost per
hour for a given set of policy variables. This analysis assumes the following settings for the indicator
variables, c1 ¼ 1, c2 ¼ 0, c3 ¼ 0. (Production continues during an assignable cause search and stops during
Reactive and Planned Maintenance.) These policies commonly occur in practice. Table 1 presents the high
and low levels of the 15 input parameters in the DOE, which are adapted from Prabhu et al.Õs (1997) study
of adaptive Statistical Process Control. As a result, this restricts the analysis to these ranges of values.
The sensitivity analysis employs a fractional factorial design to study the effects of the model parameters
on the optimal hourly cost and associated policy variables (n ; h ; L ; k ). Studying these response variables
1510
using a 2IV design requires 32 experimental runs. This resolution IV design separates main effects but
some two-factor interactions are aliased. As a result, the study limits conclusions to main effects. The design
generators are F ¼ ABC, G ¼ ABD, H ¼ ABE, J ¼ ACD, K ¼ ACE, L ¼ ADE, M ¼ BCD, N ¼ BCE,
O ¼ BDE, and P ¼ CDE.
Typically, replicated experimental runs in a DOE provide an estimated experimental error, and an F -test
determines the significance of the factors. In this setting, no experimental error occurs with replication. The
contribution of each factor to the total sum of squares is compared to determine the relative significance. As
a result, the DOE, and more specifically the relative factors, is determined using normal probability plots as
suggested by Neter et al. (1990).
To determine the optimal policy (n ; h ; L ; k ), we constrain the feasible region of the decision variables
to ensure the selection of a reasonable solution. We constrain the sample size to an integer from 1 to 20. A
constraint on ARL0 to values greater than 100 avoids excessive false alarms, and a constraint on ARL1 to
values less than 10 promotes early detection of an out-of-control condition. A constraint on the Planned
Maintenance time, kh, to values greater than 40 hours allows for sufficient production between Planned
Maintenance intervals. Clearly, these constraints will vary according to the needs of the production
environment. Finally, the design parameters h and L must be positive. A formal statement of the optimi-
zation problem follows:
Minimize fEq: ð1Þg
subject to ARL0 > 100; ARL1 < 10; kh > 40; n 6 20; h; L > 0 and n; k 2 N þ :
The Hooke and Jeeves pattern search algorithm (Bazaraa and Shetty, 1993) is used to search for an
optimal solution. We also use the above mathematical programming problem to identify optimal solutions
for the Planned Maintenance Model and the Statistical Process Control Model, replacing the objective
function with Eqs. (2) and (3) respectively.
Table 1
Experimental data
Factor A B C D E F G H J K L M N O P
Variable Shape-m Mean-h E T0 T1 T2 T3 CI CO CY CR CP CF CV d
Low 1 100 0.01 1 1 4 4 100 300 200 5000 2400 1 0.2 0.5
High 4 500 0.1 4 4 8 8 200 600 500 10000 4800 5 1 2
334 K. Linderman et al. / European Journal of Operational Research 164 (2005) 324–340
The analysis of the DOE allows for the determination of the significant main effects on the policy
parameters––n , h , L , and k ––and on hourly cost, Cost . Fig. 3 presents the normal plot of the effect
estimates with Cost as the response variable. This plot shows the normalized parameter estimates against
the normal quintile error. The negligible effects lie close to the line, and the large effects deviate from the
line. For example, Fig. 3 illustrates that increasing CI (in-control costs) and CR (reactive-maintenance costs)
results in a sizeable increase in the optimal cost. On the other hand, an increase in h (the mean failure time)
or m (the shape parameter) results in a sizable decrease in optimal costs.
Table 2 summarizes the results from the DOEs normal probability plots for cost and each of the optimal
policy variables (n , h , L , k , and kh ). The factors reside in the columns and the responses in the rows. The
sign in each column indicates the effect of the factor on the response. A positive sign indicates that varying
the factor from the low to high level increases the response; a negative sign indicates a decrease in the
response. A blank entry means that the response is insensitive to the factor. Table 2 suggests that T0 , T1 , T2 ,
T3 , CY , and CP have little impact on the optimal cost and associated design parameters.
Interestingly, the Process Failure Mechanism plays an important role in determining the optimal policy
parameters and the overall cost. For example, when the shape parameter (m) increases, the costs, sample
size, and inspection interval decrease. When the shape parameter increases, the variance in the time to
failure decreases and it becomes easier to predict the time of failure. In addition, when the time between
failure increases (h), the inspection and maintenance intervals increase and, logically, the overall costs
decrease.
CI
1 CR
Normal Score
-1
ν
θ
-5 0 5 10 15
Standardized Effect
Table 2
Summary of results for monitoring-maintenance model
Factor Shape-m Mean-h E T0 T1 T2 T3 CI CO CY CR CP CF CV d
n
) )
h ) + ) + ) +
L ) ) +
k )
ðkhÞ ) +
Cost ) ) + +
K. Linderman et al. / European Journal of Operational Research 164 (2005) 324–340 335
PM Cost
Coordinated Cost
SPC Cost
Cost
0 1 2 3 4 5 6 7
Shape Parameter
To further illustrate the sensitivity of the coordinated Monitoring-Maintenance Model to the Process
Failure Mechanism, a comparison is made between the Statistical Process Control Model, the Planned
Maintenance Model, and the Monitoring-Maintenance Model. Fig. 4 compares the optimal cost for each of
the models for various levels of the shape parameter.
This figure confirms McWilliams (1989) argument that Statistical Process Control charts are insensitive
to the assumed Process Failure Mechanism: Costs are the same for various levels of the shape parameter
when the mean is held constant. On the other hand, the graph indicates that the Planned Maintenance
Model is sensitive to the shape parameter, where the optimal hourly cost increases as the shape parameter
decreases. This is intuitively appealing because the Weibull distribution has a decreasing hazard rate when
the shape parameter is less than one. Planned Maintenance is not effective with a decreasing failure rate.
(Note that decreasing hazard rates often do not occur in real-life equipment maintenance settings; however,
Hopp and Spearman (1996) describe some industrial settings where it may occur.) The Monitoring-
Maintenance Model is sensitive to changes in the shape parameter but not nearly as sensitive as the Planned
Maintenance Model. This suggests that supplementing Planned Maintenance with Statistical Process
Control dampens the sensitivity to the Process Failure Mechanism.
Table 3
Summary for sensitivity of cost
Factor Shape-m Mean-h E T0 T1 T2 T3 CI CO CY CR CP CF CV d
Cost ) ) + + +
PM
Cost ) + + + )
SPC
Cost ) ) + +
SPC/
PM
Reactive Maintenance costs have a large impact on overall cost. Without Statistical Process Control to
detect problems, corrective action does not occur. Only shorter Planned Maintenance intervals can mitigate
the possibility of out-of-control production.
The DOE suggests that the Statistical Process Control Model is sensitive to the process mean (d). A small
shift in the process mean makes it difficult for a control chart to detect an out-of-control condition. The
coordinated Monitoring-Maintenance Model helps mitigate this problem by supplementing Statistical
Process Control with Planned Maintenance.
Fig. 4 also offers insight into when the coordinated policy is more cost effective that the non-coordinated
policies. When the shape parameter is less than 1.0 (a decreasing failure rate), there is no benefit in using
the coordinated model over using just Statistical Process Control, because Planned Maintenance is not
effective for decreasing failure rates. In addition, when the shape parameter is high (in this case, greater
than 4.0), there is little benefit in using the coordinated model over the Planned Maintenance Model. As
the shape parameter increases, the process failure time becomes more predictable. Planned Maintenance
can be used to renew the process before it goes out of control, making maintenance an effective control
tool.
A comparison between the coordinated and uncoordinated process-control and maintenance policies
also provides insight into the value of integrating these practices. Often practitioners give little thought to
how maintenance and inspection decisions affect one another. We investigate the impact of making
uncoordinated decisions using the DOE. To evaluate the cost of an uncoordinated decision, we first
determine the optimal Statistical Process Control policies from Eq. (3) and the optimal maintenance pol-
icies from Eq. (2). Using these parameters to evaluate Eq. (1) produces the cost of making uncoordinated
decisions and provides a basis for comparison to coordinated decisions. The percent increase in hourly cost
of making uncoordinated decisions over coordinated decisions ranged from 0.1% to 54%. Table 4 provides
a summary of the DOE results for the percent increase in making uncoordinated versus coordinated
decisions.
These results suggest that increases in Planned Maintenance costs (CP ) and hourly out-of-control costs
(CO ) increase the penalty of not coordinating, and decreases in the shape parameter (m) and the in-control
Table 4
Summary for sensitivity for penalty of not coordinating
Factor Shape-m Mean-h E T0 T1 T2 T3 CI CO CY CR CP CF CV d
% Inc. ) ) + +
cost
K. Linderman et al. / European Journal of Operational Research 164 (2005) 324–340 337
cost (CI ) increase the penalty of not coordinating. Coordinated decision-making works well with high
Planned Maintenance costs, making frequent maintenance very costly, or high out-of-control costs, making
process-control maintenance more important.
7. Conclusions
This research offers a promising conceptual contribution that suggests that maintenance and process-
control decisions, often considered independent in practice, need to be coordinated. In our model, Sta-
tistical Process Control monitors the equipment and provides signals indicating equipment deterioration,
while Planned Maintenance is scheduled at regular intervals to preempt equipment failure. Our research
demonstrates the benefits from conducting maintenance as an adaptive maintenance policy, where the
maintenance schedule adapts to the stability of the process. While the modelÕs effectiveness is demonstrated
using an X chart, it is possible to use a variety of other control charts, such as Exponentially Weighted
Moving Average, thereby expanding the modelÕs applicability to diverse environments with unique process-
control approaches.
The analysis demonstrates considerable economic benefit in coordinating process-control and mainte-
nance policies and indicates conditions for which the benefits are most pronounced. In addition, this re-
search provides expanded insight into the implications of Process Failure Mechanism when determining
process-control and maintenance policies. These results suggest that careful consideration should be given
to the Process Failure Mechanism when integrating Planned Maintenance and Statistical Process Control.
As a final point, our study illustrates the economic benefits that can result from using economic criteria to
determine control chart design. However, if practitioners want to use Rules of Thumb rather than eco-
nomic criteria, we demonstrate that there is still economic benefit in making coordinated Statistical Process
Control and maintenance decisions.
338 K. Linderman et al. / European Journal of Operational Research 164 (2005) 324–340
Finally, this paper emphasizes the importance of integrated management systems. We demonstrate that
good maintenance and quality practices developed in isolation are not as effective as coordinated policies.
We expect that the benefits of coordination shown in this paper are actually fewer than might be experi-
enced in practice. After all, if the maintenance, production, and quality personnel work together to develop
control methods, they are likely to arrive at other process improvements that may not have been addressed
without collaboration.
Appendix A
Z ðkþ1Þh
TA
s ¼ E TA h jTA < ðk þ 1Þh ¼ ðt hbt=hcÞfTA ðtÞ dt
h 0
Z ðkþ1Þh Z ðkþ1Þh Z ðkþ1Þh k Z
X ðjþ1Þh
¼ tfTA ðtÞ dt hbt=hcfTA ðtÞ dt ¼ tfTA ðtÞ dt h bt=hcfTA ðtÞ dt;
0 0 0 j¼0 jh
Appendix B
X
k
¼ j½FTA ððj þ 1ÞhÞ FTA ðjhÞ
j¼1
X
k
m
¼ eðkjhÞ keðkðkþ1ÞhÞ ;
j¼1
Appendix C
Because a Planned Maintenance is never performed, s0 and s0 are the limiting case of s and s respectively.
As a result,
" #
X
k X
1
0 ðkjhÞm ðkðkþ1ÞhÞ m
s ¼ lim s ¼ lim e ke ¼ eðkjhÞ and
k!1 k!1
j¼1 j¼1
"Z " ##
hðkþ1Þ
m X
k
m X
1
m
s0 ¼ lim s ¼ lim mkm1 tm eðktÞ dt h eðkjhÞ keðkðkþ1ÞhÞ ¼hh eðkjhÞ ;
k!1 k!1 0 j¼1 j¼1
References
Banerjee, P.K., Rahim, M.A., 1988. Economic design of x-bar control charts under Weibull shock models. Technometrics 30, 407–414.
Bazaraa, M., Shetty, C., 1993. Nonlinear Programming: Theory and Algorithms. Wiley, New York.
Cassady, R.C., Bowden, R.O., Pohl, E.A., 2000. Combining preventive maintenance and statistical process control: A preliminary
investigation. IIE Transactions 32, 471–478.
Dodson, B., 1994. Weibull Analysis. Quality Press, Milwaukee, WI.
Girshick, M.A., Rubin, H., 1952. A BayesÕ approach to a quality control model. Annals of Mathematical Statistics 23, 114–125.
Feigenbaum, A.V., 1961. Total quality control: Engineering and management; the technical and managerial field for improving
product quality, including its reliability, and for reducing operating costs and losses. McGraw-Hill, New York.
Ho, C., Case, K.E., 1994. Economic Design of Control Charts: A literature review for 1981–1991. Journal of Quality Technology 26,
39–53.
Hopp, W., Spearman, M.L., 1996. Factory Physics: Foundations of Manufacturing Management. Irwin, Chicago, IL.
Ishikawa, K., 1976. Guide to Quality Control. Asian Productivity Organization, Tokyo.
Kolesar, P.J., 1993. Scientific quality management and management science. In: Graves, S.C. (Ed.), Handbooks in OR & MS, vol. 4.
Elsevier Science Publishers, Amsterdam.
Lee, T.E., Shiba, S., Wood, R.C., 1999. Integrated Management Systems: A Practical Approach to Transforming Organizations. John
Wiley, New York.
McCall, J.J., 1965. Maintenance policies for stochastically failing equipment: A survey. Management Science 11 (5), 493–524.
McKone, K.E., Schroeder, R.G., Cua, K.O., 2001. The impact of total productive maintenance practices on manufacturing
performance. Journal of Operations Management 19 (1), 39–58.
McKone, K., Weiss, E.N., 1998. Total productive maintenance: Bridging the gap between practice and research. Production
Operations Management 7 (4), 335–351.
McWilliams, T.P., 1989. Economic control chart designs and the in-control time distribution: A sensitivity study. Journal of Quality
Technology 21, 103–151.
Montgomery, D.C., 1980. Economic design of control charts: A review of literature survey. Journal of Quality Technology 12, 40–43.
Montgomery, D.C., 2001. Introduction to Statistical Quality Control, fourth ed. John Wiley & Sons, New York.
Moskowitz, H., Plante, R., Chun, Y.H., 1994. Effect of process failure mechanisms on economic x-bar control charts. IIE Transactions
26, 12–21.
Neter, J., Kutner, M.H., Nachtsheim, C.J., Wasserman, W., 1990. Applied Linear Statistical Models. Irwin, Chicago, IL.
Pande, P.S., Neuman, R.P., Cavanagh, R.R., 2000. The Six Sigma way: How GE, Motorola, and other top companies are honing their
performance. McGraw-Hill, New York.
Pierskalla, W.P., Voelker, J.A., 1976. A survey of maintenance models: The control and surveillance of deteriorating systems. Naval
Research Logistics Quarterly 23, 353–388.
Porteus, E.L., Angelus, A., 1997. Opportunities for improved statistical process control. Management Science 43 (9), 1214–1228.
Prabhu, S.S., Montgomery, D.C., Runger, G.C., 1997. Economic-statistical design of an adaptive X chart. International Journal of
Production Economics 49, 1–15.
Ross, S., 1996. Stochastic Processes, second ed. John Wiley & Sons, New York.
340 K. Linderman et al. / European Journal of Operational Research 164 (2005) 324–340
Ryan, T.P., 2000. Statistical Methods for Quality Improvement, second ed. John Wiley & Sons, New York.
Tsuchiya, S., 1992. Quality Maintenance: Zero Defects Through Equipment Management. Productivity Press, Cambridge, MA.
Valdez-Flores, C., Feldman, R.M., 1989. A survey of preventative maintenance models for stochastically deteriorating single-unit
systems. Naval Research Logistics Quarterly 36, 419–446.