1

Prediction of Passive
Xi Gu
Department of Mechanical Engineering,
University of Michigan,
Maintenance Opportunity
2350 Hayward Street,
Ann Arbor, MI 48109
e-mail: [email protected]
Windows on Bottleneck
Xiaoning Jin
Machines in Complex
Department of Mechanical Engineering,
University of Michigan,
2350 Hayward Street,
Manufacturing Systems
Ann Arbor, MI 48109 In this paper, we investigate hidden opportunities for performing proper maintenance
e-mail: [email protected] tasks during production time without causing production losses. One of the maintenance
opportunities on a machine is when the machine is starved or blocked due to the occur-
Jun Ni rence of random failures on its upstream or downstream machines. Such failure-induced
Fellow ASME starvation or blockage time is defined as a passive maintenance opportunity window
Department of Mechanical Engineering, (PMOW), and is predicted on the bottleneck machines in manufacturing systems with
University of Michigan, different configurations. The effectiveness of the PMOW prediction algorithm is validated
2350 Hayward Street, through case studies in both simulations and an automotive assembly plant.
Ann Arbor, MI 48109 [DOI: 10.1115/1.4029906]
e-mail: [email protected]
Keywords: manufacturing system, real-time maintenance policy, maintenance opportunity
window

1 Introduction between production and maintenance [4]. Chang et al. [5] defined
MOWs as the hidden opportunities for maintenance during pro-
Maintenance is one of the most important operations in manu-
duction time while the production continuity is still guaranteed.
facturing systems, and it is the number one system operational
Such time window is provided by the material handling devices in
cost factor [1]. There are different types of maintenance tasks per-
manufacturing systems, such as bins, shelves, and conveyors.
formed in manufacturing systems, including (1) repair when
These devices function as buffers to store work-in-process (WIP)
machines break down; (2) regularly scheduled preventive mainte-
jobs and decouple machine downtimes [6].
nance tasks; and (3) incidental maintenance tasks that are urgent
We further classify MOWs into two different types. For the first
but take short time, such as adding coolant, changing tools, and
type, one machine can be strategically shut down for maintenance
inspections [2]. Development of effective maintenance policies
during its production time if the short-term system production
can help manufacturing enterprises to reduce maintenance cost by
requirement can be satisfied by utilizing the inventories in the
making better utilization of their maintenance resources.
machine’s downstream buffers. Such kind of maintenance oppor-
Although maintenance can keep the machines and equipment
tunities is defined as active MOWs. The second type of MOWs
operating in good condition, arbitrarily stopping machines for
comes from a machine’s blockage and starvation time that is
maintenance could interrupt regular production and affect system
induced by the propagation of the downtime of other machines in
throughput. Therefore, a conflict arises between the production
the system. We define such MOWs as PMOWs.
manager and the maintenance manager: the former wants to keep
Maintenance policies in multicomponent systems have been
machines operating to satisfy daily production target but with lit-
studied for decades. Cho and Parlar [7] reviewed papers investi-
tle concern about the machine health condition, while the latter
gating the policies in the systems where components may or may
wants sufficient stoppage time to perform adequate maintenance
not depend on each other, economically or stochastically, and
tasks [2]. A traditional way to resolve the conflict is to schedule
Nicolai and Dekker [8] further considered the structural depend-
maintenance tasks during nonproduction shifts or weekends [3].
ence. Some group maintenance and opportunistic maintenance
Such policy is sometimes difficult to meet the required system
policies were reviewed by Wang [9]. However, most of the litera-
performance in terms of cost-effectiveness and production effi-
ture focuses on the reliability of individual components and makes
ciency. First, it introduces extra overtime labor cost. Usually the
assumptions about their interdependency. There is a lack of
salaries during weekends and holidays are one and half to two
study on the analytical relationship among these components. For
times of regular salaries. Second, as manufacturing systems
manufacturing systems, Ambani et al. [10] used a continuous time
become more and more complex, there are so many maintenance
Markov chain model to develop condition-based maintenance
tasks that not all of them can be completed in nonproduction time.
policies in serial production lines. Lee et al. [11] investigated the
Third, this policy is usually static and does not respond to the
optimal inspection policy for a manufacturing system with two
real-time system conditions. In order to reduce overtime costs and
machines in serial or parallel. The systems considered in
make more efficient use of maintenance resources, mathematical
Refs. [10,11] contain no buffers, which makes their application
models should be developed to capture the interdependence
limited. The real-time buffer levels directly affect how the down-
time of one machine could propagate to its surrounding machines
1
Corresponding author. [12], and hence the existence of buffers plays an important role in
Contributed by the Manufacturing Engineering Division of ASME for publication
in the JOURNAL OF MANUFACTURING SCIENCE AND ENGINEERING. Manuscript received
developing short-term maintenance policies [13]. Chang et al.
September 30, 2013; final manuscript received February 19, 2015; published online used a continuous flow model to investigate the maintenance
March 12, 2015. Assoc. Editor: Jaime Camelio. opportunities in serial lines in Ref. [5], and further accounted for

Journal of Manufacturing Science and Engineering JUNE 2015, Vol. 137 / 031017-1
C 2015 by ASME
Copyright V

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 the location of the slowest station in Ref. [14]. Liu et al. [15] In Sec. 2.1, the critical downtime in serial lines is analyzed.
investigated the cost of downtime incidents in serial lines and Then in Sec. 2.2, an EPSL method is developed. Section 2.3
found that once the downtime of the machine exceeds a threshold, calculates the critical downtime in a complex system.
the cost will increase linearly as the downtime increases. Li et al.
[16] studied how system throughput is affected by the propagation
of disruptions in production subsystems. These papers provide 2.1 Calculation of the Critical Downtime in a Serial Line.
insights on how the line dynamics will change when machine We consider an N-machine-(N  1)-buffer serial line in Fig. 1,
failures occur in serial lines. Investigating the problem in where machine Mi is located either on the upstream or the down-
manufacturing systems with more complex configurations is more stream of the bottleneck machine Mb . Backward and forward
interesting but challenging. aggregation methods have been developed in Ref. [19] to analyti-
In this paper, we predict PMOWs on the bottleneck machine in cally derive DTi , the critical downtime for machine Mi , as
a complex manufacturing system. A complex manufacturing sys- 8
>
> X
b1 X
b1
tem is defined as a system where the downtime of one machine >
> Nj ð0Þ  tb  tj if 1  i < b
may propagate to another machine through different lines in the >
>
>
> j¼i j¼i
system. The bottleneck machine is the one whose stoppage could <
interrupt the system throughput in the strongest manner [17]. DTi ¼ 0 if i ¼ b (1)
>
>
Therefore, in practice, it is always desired not to actively shut >
> Xi1 
>
>
down the bottleneck machine, and hence to predict its PMOWs is >
> Cj  Nj ð0Þ  tb if b < i  N
:
important. Short-term bottleneck machines can be detected by j¼b
“turning point” method [18]. Chang et al. [14] showed that the
slowest machine in the system satisfies the concept of a turning where Ci and Ni ð0Þ are the capacity and real-time levels of buffer
point and thus can be regarded as the bottleneck machine. More- Bi, respectively.
over, Liu et al. [15] showed that the stoppage of the slowest From Eq. (1), although the expressions of DTi ’s are different
machine will result in permanent system production losses, dem- based on the relative locations of Mi and MbP , their structures are
onstrating that the slowest machine is the least desired to be similar. If Mi is on the upstream of Mb , then b1j¼i Nj ð0Þ  tb is the
stopped. Therefore, in this paper, we treat the bottleneck machine time for Mb to process all the parts between machine Mi and
Pb1
and slowest machine as equivalent. machine Mb , until Mb becomes starved; and j¼i tj is the time it
PMOWs on the bottleneck machine come from the occurrence will take to make Mb resume working after Mi is restored from the
of random failures on other machines in the system. To determine downtime. Similarly, if Mi is on the downstream of Mb ,
whether a failure on a nonbottleneck machine will make the
P i1  
j¼b Cj  Nj ð0Þ  tb is the time for Mb to feed the empty space
bottleneck machine idle or not, we first investigate the critical in the buffers between machine Mi and Mb before Mb is blocked,
downtime. If the actual downtime of one machine is shorter than which is the same as the critical downtime since resuming Mi will
its critical downtime, it will not induce idle time on the bottleneck immediately make Mb unblocked. These similarities lead to the
machine. Gu et al. [19] calculated the critical downtime analyti- construction of an EPSL, which will be discussed next.
cally in discrete production systems with serial and nonserial
structures.
Furthermore, as an event-driven decision support tool, the 2.2 Construction of an EPSL. In a serial line, one machine
prediction of PMOW should respond rapidly if there is updated is either on the upstream or downstream of another machine,
failure information. In this paper, we not only consider PMOWs while in a complex system, the relative locations of two machines
under a single failure, but also investigate how to update PMOWs may not be so straightforward. Therefore, a unified form of the
when multiple failures occur. With the prediction and updating of calculation of DTi ’s is needed, regardless of whether Mi is on the
PMOWs, maintenance crews can get prepared in advance, so that upstream or the downstream of Mb .
proper preventive maintenance tasks can be carried out on the bot- In this section, we develop an EPSL method to obtain DTi ’s. If
tleneck machines more effectively. Mi is the machine whose critical downtime needs to be calculated,
The rest of the paper is organized as follows. In Sec. 2, we then an EPSL is constructed in a way such that machine Mi is the
investigate the critical downtime (DT*) for machines in a complex first machine and the bottleneck machine Mb is located at the end
manufacturing system by using the equivalent pseudo serial line of the line. Based on the result in Ref. [5], the critical downtime
(EPSL) method. Based on the critical downtime, a mathematical for machine Mi can be calculated by
model is developed in Sec. 3 to predict the PMOW on the bottle-
neck machine. The PMOW is analyzed under a single failure in a DTi ¼ Ticons  Tires (2)
complex system and under multiple failures in a serial line.
Finally, these results are combined to predict PMOWs under mul- where Ticons (time to consume) is the time it takes Mb to process
tiple failures in a complex system. Section 4 presents two numeri- all the buffer contents between Mi and Mb , and Tires (time to
cal case studies where the PMOW prediction model is illustrated resume) is the time it takes for the first part processed by the
and its effectiveness is validated through simulation and real plant repaired Mi to traverse to Mb .
data. Section 5 contains the concluding remarks and the future In order to construct an EPSL where the bottleneck machine is
works. the last machine, the machines which are located on the down-
stream of the bottleneck machine need to be reversed to
its upstream in the EPSL. For example, in order to analyze the
2 Calculation of the Critical Downtime effect of the downtime of MN in the serial line in Fig. 1,
The critical downtime (DT ) is defined as the maximum time Mb  Bb      BN1  MN are reversed in the pseudo line
that one machine can be down without making the bottleneck MNr  BrN1      Brb  Mbr , as shown in Fig. 2.
machine idle. In the figures throughout the paper, M and B stand If a failure occurs on machine Mi (i > b) in the real line, and the
for machine and buffer, respectively. It is assumed that machine failure is long enough, then machines Mi1 ; Mi2 ; :::; Mb will be
Mi has a constant cycle time ti , while traveling time at buffers is blocked successively; and if the failure occurs on Mir , then
negligible. In a system, one machine is starved if one of its
upstream buffers is empty; and it is blocked if one of its down-
stream buffers is full and the next machine is down or blocked. In
addition, the first machine is never starved and the last machine is
never blocked. Fig. 1 An N-machine-(N 2 1)-buffer serial line

031017-2 / Vol. 137, JUNE 2015 Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 3 Prediction of PMOWs
The above method of critical downtime calculation enables us
to predict the PMOW on the bottleneck machine when failures
Fig. 2 An EPSL for the downstream machine MN occur in the system. In this section, the PMOW is first predicted
in a single failure scenario, and then updated when sequential fail-
machines Mi1 r r
; Mi2 ; :::; Mbr will be starved successively. By an ures occur.
appropriate selection of the parameters of the pseudo line, the star-
vation of Mbr will be equivalent to the blockage of Mb . 3.1 Prediction of PMOWs Under Single Failure in a
The real line and its EPSL are equivalent in terms of the time to Complex System. Consider that at time 0, a failure occurs on
consume and the time to resume. More specifically, the time to machine Mi and will last for DTi . Based on Sec. 2.3, the critical
consume the empty space in buffers Bi1 ; :::; Bb equals the time to downtime of Mi can be calculated as DTi ¼ mink¼1;2;:::;Ki
cons res
consume the parts in the reversed buffers Bri1 ; :::; Brb ; and the ðTi;k  Ti;k Þ. If the actual downtime is shorter than (or equal to)
time to make Mb unblocked after the downtime of Mi ends is the the critical downtime, i.e., DTi  DTi , the bottleneck machine
same as the time to make Mbr unstarved after Mir resumes running. Mb will not be starved or blocked. However, if DTi > DTi;k
Consequently, the critical downtimes in these two lines are also (k ¼ 1; :::; Ki ), Mb may have an idle interval resulted from line li;k .
equivalent. To predict the PMOW on Mb requires the joint consideration of
This result is stated in the following lemma: the idle intervals in all of these lines.
LEMMA 1. The effect of the failure occurring on the downstream First, we consider these idle durations independently. In line
cons res
of the bottleneck machine can be equivalently analyzed by that li;k , Mb will be idle from time Ti;k to DTi þ Ti;k , i.e.,
on the corresponding upstream machine in the EPSL, through the
h 
following transformation: cons res
PMOWk ¼ Ti;k ; DTi þ Ti;k ðk ¼ 1; :::; Ki Þ (5)
Nir ð0Þ ¼ Ci  Ni ð0Þ for i ¼ b; …; N  1
(3)
tri ¼ 0 for i ¼ b þ 1; …; N Also, the length of PMOWk is
 
res cons
where Nir ð0Þ is the initial level of the reversed buffer Bri , and tri is jPMOWk j ¼ max 0; DTi þ Ti;k  Ti;k
the cycle time of the reversed machine Mir in the EPSL.  
Proof. See the Appendix. ¼ max 0; DTi  DTi;k (6)
A numerical example will be given in Sec. 4.1 to illustrate the
construction of an EPSL. The calculation of the critical downtime which demonstrates that if DTi is greater than the critical down-
in such a line serves as a basis for the calculation of the critical time DTi;k , Mb will have an idle duration of (DTi  DTi;k ) in line
downtime in a complex system. li;k . This agrees with the result claimed in Ref. [15].
Next, we consider the PMOW resulted from lines li;1 ; :::; li;Ki
2.3 Calculation of the Critical Downtime in a Complex jointly. First, these lines are arranged in the ascending order of
System. In a complex manufacturing system, two machines Mi cons
Ti;k cons
such that Ti;ð1Þ cons
     Ti;ðK . Then, the earliest possible
iÞ
and Mb may be connected through multiple equivalent serial lines idle duration on Mb may occur in line li;ð1Þ , i.e.,
(which may be either real or pseudo), and the downtime of Mi can
propagate to Mb through these different lines. h 
Figure 3 illustrates an example of such a system, where cons
PMOWð1Þ ¼ Ti;ð1Þ res
; DTi þ Ti;ð1Þ (7)
machine M1 is a cutting machine, machines M2 –M5 are stamping
machines, and the bottleneck machine Mb is a welding machine  
which joins the parts from both lines together. In such a system, Since Mb has an idle duration of PMOWð1Þ , which starts
cons
the two stamping lines share the same cutting machine M1 . If a before time Ti;ð2Þ , the additional idle time that is caused by the
failure occurs on M1 , the impact of the failure can propagate to  
downtime in li;ð2Þ will start at Ti;ð2Þcons
þ PMOWð1Þ  while it still
Mb through both lines l1;1 and l1;2 . res
ends at DTi þ Ti;ð2Þ . Therefore, when the joint effect in lines li;ð1Þ
Generally, if there are Ki equivalent serial lines connecting Mi
and Mb , then in line li;k ðk ¼ 1; :::; Ki Þ, the critical downtime DTi;k and li;ð2Þ is considered, the PMOW on Mb is
can be calculated as DTi;k ¼ Ti;k cons
 Ti;kres
. The critical downtime
for the system is the minimum of all DTi;k ’s, i.e.,
 
DTi ¼ min DTi;k ¼ min Ti;k cons res
 Ti;k (4)
k¼1;2;:::;Ki k¼1;2;:::;Ki

Therefore, if the actual downtime is shorter than DTi , then the

bottleneck machine will not be idle in any of the equivalent serial
lines.

Fig. 4 The state of the bottleneck machine in lines li;ð1Þ and li;ð2Þ
when these lines are considered independently (a) and
Fig. 3 An example of a complex manufacturing system jointly (b)

Journal of Manufacturing Science and Engineering JUNE 2015, Vol. 137 / 031017-3

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 [h   
PMOWð2Þ ¼ PMOWð1Þ cons
Ti;ð2Þ þ PMOWð1Þ ; DTi þ Ti;ð2Þ
res investigate how to update PMOWs in response to multiple failures
in a serial line.
(8) Assume that the jth failure Fj occurs on machine Mij at time Tj
and lasts for DTij . The corresponding time to consume and time to
Figure 4 illustrates the state of the bottleneck machine Mb in resume are denoted as Ticonsj
and Tires
j
, respectively. The updated
lines li;ð1Þ and li;ð2Þ , when these two lines are considered independ- PMOW on the bottleneck machine Mb at time Tj (after Fj occurs)
ently (Fig. 4(a)) and jointly (Fig. 4(b)). In this figure, “original” is denoted as PMOWðTj Þ. Then, after the first failure F1 ,
means the part that Mb is working on is in the original system PMOWðT1 Þ can be predicted by Eq. (5), as
before the failure occurs; and “new” means the part that is proc-
essed by Mi after it resumes running. Figure 4 shows that the idle h 
duration on Mb caused in line li;ð1Þ delays the propagation of the PMOWðT1 Þ ¼ T1 þ Ticons1
; T1 þ DTi1 þ Tires
1
(10)
downtime in line li;ð2Þ . The idea of calculating such “delay of
propagation” will be frequently used in the following analysis. When the second failure F2 occurs, PMOWðT2 Þ can be updated
This procedure can be applied recursively to predict the according to the following two cases:
PMOW on Mb , which is caused jointly in lines li;ð1Þ ,…,li;ðkþ1Þ , as Case 1: T2 þ Ticons  T1 þ Ticons
2 1
In this case, the impact of failure F1 will propagate to Mb
PMOWðkþ1Þ before that of F2 , so the occurrence of F2 will not affect
[h   
PMOWðT1 Þ. Moreover, the existence of PMOWðT1 Þ on Mb may
¼ PMOWðkÞ cons
Ti;ðkþ1Þ þ PMOWðkÞ ; DTi þ Ti;ðkþ1Þ
res
(9)
delay the time when the impact of F2 propagates to Mb . We
denote DðF2 Þ as this delay, which can be calculated as
Finally, since there are Ki lines connecting Mi and Mb , h \
PMOWðKi Þ is the PMOW caused by the single failure on machine DðF2 Þ ¼ T1 þ Ticons ; T 1 þ DT i þ T res
½T2 ; þ1Þ (11)
1 1 i1
Mi .
As discussed above, in a serial line, the total idle time on the
bottleneck machine increases linearly in DTi as long as DTi Then, PMOWðT2 Þ can be updated as
exceeds its threshold DTi . The following lemma demonstrates h 
that such property still holds in a complex system. PMOWðT2 Þ ¼ T1 þ Ticons
1
; T 1 þ DTi1
þ T res
i1
LEMMA 2. In a complex system, if machine Mi is down for time [h 
DTi , and its critical downtime is DTi , then the total idle time on T2 þ Ti2 þ jDðF2 Þj; T2 þ DTi2 þ Tires
cons
2
the bottleneck machine is jPMOWi j ¼ maxð0; DTi  DTi Þ.
Proof. See the Appendix. (12)
However, unlike a serial line, where the bottleneck machine
could have at most one idle duration, in a complex system, a fail- Case 2: T2 þ Ticons
2
< T1 þ Ticons
1

ure event may cause multiple idle durations on the bottleneck The second case occurs when the first idle duration on machine
machine. Naturally, one may consider grouping these separate Mb is caused by F2 . Then, this idle duration will delay the propa-
idle durations together and strategically shutting the bottleneck gation for F1 , and this delay is calculated as
machine down for jPMOWi j. The following proposition shows h 
that such grouping will not bring additional idle time to the bottle- DðF1 Þ ¼ T2 þ Ticons2
; T2 þ DTi2 þ Tires
2
(13)
neck machine, as long as it is shut down before it is affected by
the failure event.
PMOWðT2 Þ is then updated as
PROPOSITION 1. Let j1 ¼ mink¼1;:::;Ki fk : DTi > DTi;ðkÞ g, then for
cons
any T0  Ti;ðj , if the bottleneck machine is shut down during h 
1Þ
time ðT0 ; T0 þ jPMOWi jÞ, it will have no additional idle time PMOWðT2 Þ ¼ T2 þ Ticons
2
; T2 þ DTi2 þ Tires
2
afterward if no future failure occurs. [h 
Proof. See the Appendix. T1 þ Ticons
1
þ jDðF1 Þj; T1 þ DTi1 þ Tires
1

Note that the maintenance crews can only move the entire or (14)
partial idle durations on the bottleneck machine earlier; otherwise
the total length of PMOW will be reduced. For example, if a
To conclude the two cases, the PMOW after the second failure
PMOW starts 5 min later but the maintenance crews cannot get
F2 can be updated as
prepared in 7 min, at least 2 min of the maintenance opportunities
will be wasted. [ h 
PMOWðT2 Þ ¼ Tk þ Ticons
k
þ j DðFk Þj; T k þ DT ik
þ Ti
res
k
k¼1;2
3.2 Prediction of PMOWs Under Multiple Failures in a
Serial Line. In Sec. 3.1, PMOWs have been predicted under a (15)
single failure. However, it is possible that during the downtime of
one machine, another random failure occurs. In this section, we where

8h 
< T þ T cons ; T þ DT þ T res T2 þ Ticons < T1 þ Ticons
2 i2 2 i 2 i2
DðF1 Þ ¼ 2 1

: [ T2 þ Ticons  T1 þ Ticons
2 1

8
< [ T2 þ Ticons
2
< T1 þ Ticons
1
DðF2 Þ ¼ h T
: T1 þ Ti ; T1 þ DTi1 þ Tires
cons
½T2 ; þ1Þ T2 þ Ticons  T1 þ Ticons
1 1 2 1

031017-4 / Vol. 137, JUNE 2015 Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Next, we use mathematical induction to generalize the
PMOW updating procedure under multiple failures in a serial line.
Define function Hs ðFk Þ ¼ ½Tk þ Ticons
k
þ jDðFk Þj; Tk þ DTik þ Tires
k
Þ,
which is the additional PMOW on the bottleneck machine that is
caused by failure Fk , and DðFk Þ is the delay of propagation of the
impact of failure Fk for the bottleneck machine.
Based on Eq. (15), it is assumed that the PMOW after failures
Fn can be updated as
[ [
PMOWðTn Þ ¼ Hs ðFk Þ ¼ Hs ðFðkÞ Þ (16)
k¼1;:::;n k¼1;:::;n

 
where the order of FðkÞ ’s satisfies TðkÞ þ Ticons þ DðFðkÞ Þ
  ðkÞ
Fig. 5 Prediction of PMOW under multiple failures in a com-
 Tðkþ1Þ þ Ticons þ DðFðkþ1Þ Þ, such that the impact of FðkÞ
ðkþ1Þ plex system
propagates to Mb before that of Fðkþ1Þ .
Then, when a new failure Fnþ1 occurs at time Tnþ1 ,
PMOWðTnþ1 Þ can be updated through the following steps.
the latter updates the PMOWs when sequential failures occur in a
(1) Arrange the n þ 1 failures in an ascending order of the serial line. The integration of the two scenarios leads to the pre-
time when their impacts propagate to the bottleneck diction of PMOW in a more general yet more complicated sce-
machine Mb . nario, i.e., PMOWs under multiple failures in a complex system.
(a) Initially, set the delay of propagation for the new fail- Figure 5 illustrates the idea of such integration: once a new
ure Fnþ1 as DðFnþ1 Þ ¼ [ and its order index as j ¼ 1; machine failure occurs in a complex system, it can be regarded as
(b) Determine whether the impact of failure Fnþ1 propa- multiple equivalent failures that occur simultaneously in equiva-
gates to Mb before that of failure FðjÞ :   lent serial lines; and then, for each of the equivalent failures,
If Tnþ1 þ Ticons
nþ1
þ jDðFnþ1 Þj < TðjÞ þ Ticons ðjÞ
þ DðFðjÞ Þ, PMOWs can be updated.
Fnþ1 impacts Mb before FðjÞ . Go to step (c).   Assume at time T1 , the first failure F1 occurs on machine Mi1 ,
If Tnþ1 þ Ticons
nþ1
þ jDðFnþ1 Þj  TðjÞ þ Ticons ðjÞ
þ DðFðjÞ Þ, which is connected to the bottleneck machine through Ki1 equiva-
Fnþ1 impacts Mb after FðjÞ . Update the delay of propa- lent serial lines. Therefore, it can be regarded as there are totally
S Ki1 equivalent failures occurring simultaneously at time T1 , and
gation for Fnþ1 as DðFnþ1 Þ ¼ DðFnþ1 Þ Hs ðFðjÞ Þ
T the PMOW can be predicted based on the analysis in Sec. 3.1. Let
½Tnþ1 ; þ1Þ and set j ¼ j þ 1. If j ¼ n þ 1, Fnþ1 is F1;k denote an equivalent failure of F1 in the kth line, and order
the last failure that impacts Mb and go to step (c). cons
them by FðkÞ ¼ F1;ðkÞ (k ¼ 1; :::; Ki1 ) such that T1;ðkÞ cons
 T1;ðkþ1Þ .
Otherwise repeat step (b) to compare Fnþ1 with the From Eq. (9), PMOW at time T1 can be calculated as
updated FðjÞ .
(c) Arrange the n þ 1 failures in the ascending order of the [
PMOWðT1 Þ ¼ Hc ðFðkÞ Þ (18)
time when
8 their impacts propagate to Mb , as k¼1;:::;Ki1
< FðkÞ k ¼ 1; :::; j  1
F0ðkÞ ¼ Fnþ1 k¼j . The “¼” here h  
:
Fðk1Þ k ¼ j þ 1; :::; n þ 1 where Hc ðFðkÞ Þ ¼ TgðkÞ þ Ticons gðkÞ ;hðkÞ
þ DðFðkÞ Þ; TgðkÞ þ DTigðkÞ ;hðkÞ
means that the parameters of the updated failure F0 ’s, þTires Þ is the PMOW caused by failure FðkÞ ; and
gðkÞ ;hðkÞ
such as T 0 , DT0 , T cons0 , and T res0 , are the same as those S T 
of the corresponding failure F’s. DðFðkÞ Þ ¼ l¼1;:::;k1 Hc ðFðlÞ Þ TðkÞ ; þ1 is the delay of propa-
gation for failure FðkÞ . gðkÞ and hðkÞ are the machine and line indi-
(2) Update the delay of propagation for the Sfailures that impact T ces, such that the equivalent failure FðkÞ comes from the effect of
0 0
 b after Fnþ1 , as DðFðkÞ Þ ¼ j¼1;:::;k1 Hs ðFðjÞ Þ
M the actual failure FgðkÞ in line hðkÞ, i.e., FðkÞ ¼ FgðkÞ;hðkÞ . Note that
TðkÞ ; þ1 for k ¼ j þ 1; :::; n þ 1. gðkÞ ¼ 1, for all k ¼ 1; :::; Ki1 .
(3) Update
S the new PMOW as PMOWðTnþ1 Þ Then, we assume that at time T2 , a second failure F2 occurs on
¼ k¼1;:::;nþ1 Hs ðF0ðkÞ Þ. Set FðkÞ ¼ F0ðkÞ ðk ¼ 1; :::; n þ 1Þ so machine Mi2 , which is connected to the bottleneck machine
that it can be written as through Ki2 lines. The procedure to update PMOW is outlined as
[ follows:
PMOWðTnþ1 Þ ¼ Hs ðFðkÞ Þ (17)
k¼1;:::;nþ1
(1) Decompose failure F2 into its equivalent failure F2;ðkÞ ’s
(k ¼ 1; :::; Ki2 ) such that Ticons
2 ;ð1Þ
     Ticons
2 ;ðKi2 Þ
.
(2) For each F2;ðkÞ (k ¼ 1; :::; Ki2 ), apply steps 1–3 in Sec. 3.2
Equation (17) has the same format as Eq. (16), which completes to update PMOWðT2;ðkÞ Þ, which is the PMOW caused
the mathematical induction. Therefore, these steps can be used jointly by F1;ð1Þ ; F1;ð2Þ ; :::; F1;ðKi1 Þ and F2;ð1Þ ; F2;ð2Þ ; :::; F2;ðkÞ .
recursively to update thePMOWwhen a new failure occurs. More- (3) Finally, the PMOW at time T2 can be updated as
over, among these three steps, step 1 is the key. Once the
sequence of the failures (in terms of when their impacts propagate [
to the bottleneck machine) is obtained, it is not difficult to update PMOWðT2 Þ ¼ PMOWðT2;ðKi2 Þ Þ ¼ Hc ðFðkÞ Þ
the delay of propagation and the PMOW. k¼1;:::;Ki1 þKi2

(19)

3.3 Prediction of PMOWs Under Multiple Failures in a

Complex System. In Secs. 3.1 and 3.2, PMOWs have been Furthermore, if a future failure occurs, the PMOW can be
predicted under a single failure in a complex system and under updated again by following this procedure. Generally, when a
multiple failures in a serial line, respectively. The former predic- failure Fnþ1 occurs at time Tnþ1 , the flowchart to update
tion is built upon the construction of equivalent serial lines and PMOWðTnþ1 Þ from PMOWðTn Þ is shown in Fig. 6.

Journal of Manufacturing Science and Engineering JUNE 2015, Vol. 137 / 031017-5

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 8 Decomposition of system into equivalent serial lines l2;1
(a) and l2;1 (b)

Table 2 Parameters of the reversed machines and buffers

Machine M1r M2r M4r Buffer Br0 Br1

Cycle time (s) 0 0 0 Initial level 1 1

Table 3 PMOW prediction for case study 1

jPMOW2 j(s)

2% 5% 10%

DT2 (s) Analytical Mean SD Mean SD Mean SD

0 0 0 0 0 0 0 0
50 0 0 0 0 0 0.06 0.95
100 0 0 0 0 0 1.44 5.91
150 0 1.73 2.62 5.91 7.32 23.34 19.96
200 50 50.40 4.04 54.23 10.37 70.87 23.55
250 100 100.31 3.87 104.10 10.13 119.65 23.55
300 150 150.19 3.93 153.95 10.11 170.59 24.83
350 200 200.16 3.96 203.84 10.19 220.80 22.66
400 250 250.22 3.96 254.03 10.18 269.82 25.56
450 300 300.21 3.97 304.01 10.20 320.23 23.00
500 350 350.24 3.99 353.97 10.16 369.90 23.55

Fig. 6 Flowchart for PMOW prediction under multiple failures

in a complex system
Table 1. In closed-loop systems, such as automotive production
paint shops, some machines are not only starved by parts, but also
by pallets which are used for holding the parts [20]. In this case
study, machines M1 , M2 , M3 , and M4 are where the parts are
processed on the pallets. After machine M4 completes its work,
the finished part and the pallet will be sent to buffers B4 and B0 ,
Fig. 7 System layout for case study 1 respectively. Therefore, M4 will be blocked if either buffer B4 or
B0 is full. Moreover, although it is assumed that machine M1 is
Table 1 Parameters of the system for case study 1 not starved by parts, it can still be starved by pallets when B0 is
empty.
Machine M1 M2 M3 M4 M5 M6 We investigate the PMOW for different cases of the downtime
of machine M2 (i.e., DT2 ). l2;1 and l2;2 are the two equivalent
Cycle time (s) 62 60 59 61 60 65 serial lines connecting from machine M2 to M6 , as shown in
Buffer B0 B1 B2 B3 B4 B5 Figs. 8(a) and 8(b), respectively. The cycle times of the reversed
machines, as well as the initial levels of the reversed buffers are
Capacity 5 3 2 4 3 5 shown in Table 2. Based on the analysis in Secs. 2 and 3, it can
cons res cons
Initial level 4 2 1 2 1 2 be calculated that T2;1 ¼ 390, T2;1 ¼ 240 and T2;2 ¼ 325,
res  
T2;2 ¼ 60. Therefore,
S DT 2;1 ¼ 150, DT 2;2 ¼ 265, and PMOW 2
¼ ½325; DT2 þ 60Þ ½390 þ j½325; DT2 þ 60Þj; DT2 þ 240Þ.
4 Case Studies To validate the effectiveness of the prediction algorithm, we
have built a simulation model of the closed-loop system by using
In this section, we present two case studies to illustrate the a commercial simulation software, SIMUL8 [21]. Moreover, in
PMOW prediction algorithm. The first case study predicts the order to make the case more practical, the machine cycle times in
PMOW on the bottleneck machine in a closed-loop system, and the simulations are assumed to follow Gaussian distributions.
the analytical result is validated by simulation. The second case First, we study the jPMOW2 j under different DT2 ’s. The stand-
study focuses on PMOW prediction under multiple failures using ard deviation of the cycle time on each machine is set to be 2%,
data from a real automotive plant. 5%, and 10% of its mean. The total idle time of the bottleneck
machine M6 before it completes 20 parts is logged, and its mean
4.1 Case Study 1: PMOW Prediction in a Closed-Loop and standard deviation (SD) are shown in Table 3 and Fig. 9.
System. In the first case study, we investigate the PMOWs in a These results shows that when DT2 > 150, jPMOW2 j will
closed-loop system as shown in Fig. 7, where the machine cycle increase approximately linearly as DT2 increases. This agrees
times, buffer capacities, and initial buffer levels are provided in with the analytical result in Lemma 2. Therefore, the PMOW

031017-6 / Vol. 137, JUNE 2015 Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 9 PMOW prediction for case study 1

Fig. 10 WIP of the bottleneck machine M6 over time

prediction algorithm is applicable to the systems where the varia- plant to validate its effectiveness in practice. The system in this
tions of the machine cycle times are small. case study consists of three groups of machines and conveyors
Then, we study the WIP [22] on the bottleneck machine M6 between successive groups, as shown in Fig. 11(a). The bottleneck
under different DT2 ’s. The results when DT2 ¼ 150, 250, 350, and machine Mb is the first machine in group 3. Moreover, if a failure
450 s are shown in Fig. 10 (the standard deviation on the process- occurs, maintenance will be carried out on that machine and the
ing time is assumed to be 2% of its mean). These results demon- whole group of machines will be stopped, until the maintenance is
strate that if DT2 is within 150 s, there is almost no PMOW on completed. In this case study, we only consider the PMOWs
M6 . If DT2 increases to 250 s, which is between DT2;1 and DT2;2 , caused by failures in groups 1 and 2. This system can be modeled
there will be one idle duration on M6 , which resulted from line as a three-machine–two-buffer (3M2B) system, as shown in
l2;1 . However, if DT2 is greater than DT2;2 , such as 350 and 450 s, Fig. 11(b), with the system parameters in Table 4. The cycle time
M6 will have two separate idle durations (the third one is negligi- of machine Mi (i ¼ 1, 2) in Fig. 11(b) is equal to the summation of
ble), where the first one is caused in l2;2 and the second is in l2;1 . the cycle times of all the machines in group i. Moreover, the cycle
When DT2 continues to increase, the first idle duration will keep time of the buffers will be included in the calculation of T res .
growing while the length of the second idle duration keeps In order to show how the PMOW is updated under multiple fail-
constant. ures, machine M16 is shut down for 15 min, starting at 1:00 p.m.,
Moreover, this case study indicates that the PMOW prediction and machine M14 is shut down for 10 min, starting at 1:20 p.m.
tool is also applicable to the case when the downtime DTi is not These two events are regarded as failures F1 and F2 , respectively.
deterministic. If the probability distribution of DTi is known, The factory information system (FIS) updates the WIP of all buf-
then the probability distribution of PMOW can be predicted fers and machines every 5 min, as shown in Fig. 12. Based on the
accordingly. The optimal maintenance policy under such proba- real-time information at 1:00 p.m., the PMOW under F1 can be
bilistic PMOWs will be investigated in the future. predicted as ½1 : 25 : 00; 1 : 26 : 13Þ, indicating a 73-s idle dura-
tion on the bottleneck machine Mb , starting from 1:25 p.m. When
the second failure F2 occurs at time 1:20 S p.m., the PMOW can
4.2 Case Study 2: PMOW Prediction Under Multiple be updated as ½1 : 25 : 00; 1 : 26 : 13Þ ½1 : 32 : 13; 1 : 41 : 13Þ,
Failures in a Real Manufacturing Plant. The PMOW prediction indicating that another 9-min idle duration on Mb will start at
tool has also been implemented to a real automotive assembly 1:32 p.m. The real-time plant floor data after 1:20 pm (also shown

Journal of Manufacturing Science and Engineering JUNE 2015, Vol. 137 / 031017-7

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 11 System layout (a) and its model (b) for case study 2

Table 4 Parameters of the 3M2B system delivered to Mb shortly. Similarly, the WIP in B2 is 2 at 1:30 p.m.,
indicating that the second idle duration will start at around
Machine/buffer M1 B1 M2 B2 Mb 1:32 p.m.; while B2 has a large WIP at 1:40 p.m. so that the second
idle duration on Mb will end soon. These results validate the effec-
Capacity 6 11 7 24 1
tiveness of PMOW prediction algorithm under multiple failures.
Cycle time (min) 4.44 0.71 6.23 3.53 1

5 Conclusions
In this paper, an analytical model has been developed to predict
the PMOWs on the bottleneck machine under multiple machine
failures in a complex manufacturing system. First, through the
proposed technique of generating EPSLs, the critical downtime
for each machine in a complex system is calculated. Then, a
PMOW prediction algorithm for a complex system is developed
when a single failure occurs on one machine. It is found that, in a
complex system, the total idle time of the bottleneck machine
equals to the difference between the machine’s actual downtime
caused by the failure and its critical downtime. Moreover, the
PMOW prediction algorithm under multiple failures has also been
investigated. Case studies in simulations and a real automotive
plant have been conducted to demonstrate the methodology and
insights.
The effectiveness of the PMOW prediction algorithm has also
been validated when it is applied to systems with small variations
in processing time. The analytical model also provides a sound
basis for research on systems with more uncertainties and varia-
tions, such as large variation on machine processing time and
probabilistic downtime caused by the failure. Such randomness
will be taken into consideration in our future work, to make
PMOW prediction algorithm more robust for real plant
Fig. 12 PMOW prediction and validation using real-time data implementation.
from FIS The future work related to the maintenance opportunity window
technique includes: (1) estimating PMOWs analytically when the
cycle times of machines are not constant; (2) developing methods
to correct PMOW based on the feedback information; and (3)
integrating maintenance opportunities into the design of manufac-
in Fig. 12) are utilized to validate the prediction. From the WIP turing systems.
records in FIS, there are indeed two idle durations on Mb , as
predicted. Moreover, with a detailed observation of the WIP in
buffer B2 , one can obtain a more accurate estimation of the Acknowledgment
starting/completion time of the two idle durations. For the first This work was supported by National Science Foundation
idle duration, it shows that at 1:20 p.m. buffer B2 has four parts, (Grant No. 0825789) and the Industry/University Cooperative
which will allow Mb to process for additional 4 min before it gets Research Center for Intelligent Maintenance Systems (NSF Grant
starved; and at 1:25 p.m. B2 has one new part, which will be No. 1134676).

031017-8 / Vol. 137, JUNE 2015 Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 cons res
Appendix These results prove that PMOWðjmþ1 Þ  ½Ti;ðj 1Þ
;DTi þ Ti;ðj ÞÞ
 mþ1
and jPMOWðjmþ1 Þ j ¼ DTi  DTi;ðj Þ .
Proof of Lemma 1. We consider a machine Mi on the down-  
mþ1
Based on steps (a) and (b), PMOWðjÞ  ¼ DTi  DTi;ðj Þ 8j.
stream of the bottleneck machine in Fig. 1, i.e., i ¼ b þ 1,…, N, Let jM ¼ maxk¼1;:::;Ki fk : DTi > DTi;ðkÞ g and jM ¼ argmink¼1;2;:::;jM
and its corresponding machine Mir in the EPSL (Fig. 2). Based on DTi;ðkÞ ¼ argmink¼1;2;:::;Ki DTi;ðkÞ . Then, DTi;ðj Þ ¼ DTi and
the above analysis, in the ESPL, the time to consume is M
jPMOWi j ¼ jPMOWðjM Þ j ¼ DTi  DTi;ðj Þ ¼ DTi  DTi;ðK Þ ¼ DTi
Xi X
i1   M i
Ticons;r ¼ r
Nj1 ð0Þ  tb ¼ Cj  Nj ð0Þ  tb DTi .
j¼bþ1 j¼b Summarizing
 these
 cases leads to jPMOWi j
¼ max 0; DTi  DTi . ⵧ
which is also the time to consume all the empty spaces between Proof of Proposition 1. If jPMOWi j ¼ 0, there is no idle time
machines Mi and Mb in the original line. on Mb , and the proposition is true.
Moreover, the time to resume in the EPSL can be calculated as Otherwise, jPMOWi j ¼ DTi  DTi > 0. We consider a virtual
X
i cons
line li;ð0Þ , where Ti;ð0Þ res
¼ T0 and Ti;ð0Þ ¼ T0  DTi , such that
Tires;r ¼ trj ¼ 0  
DTi;ð0Þ ¼ DTi and PMOWð0Þ ¼ ½T0 ; T0 þ jPMOWi jÞ.
j¼bþ1
Then based on Lemma 2, PMOWðkþ1Þ ¼ PMOWðkÞ if
which is equal to that in the original line, since when the traveling DTi;ðkþ1Þ  DTi;ðkÞ . Since DTi;ð0Þ ¼ DTi ¼ mink¼1;2;:::;Ki DTi;ðkÞ ,
time between buffers is negligible, Mb will run immediately after
PMOWðkÞ ¼ PMOWð0Þ for all k 2 f1; 2; :::; Ki g.
Mi resumes running.
For the critical downtime of machine Mi in Fig. 2, we have Therefore, there are no additional idle durations on Mb in lines
li;ð1Þ ,…, li;ðkþ1Þ . ⵧ
DT;r cons;r
i ¼ Ti  Tires;r
X
i X
i
r
¼ Nj1 ð0Þ  tb  trj References
j¼bþ1 j¼bþ1
[1] Mobley, R. K., 2002, An Introduction to Predictive Maintenance, Butterworth-
X
i1   Heinemann, Wobrun, MA.
¼ Cj  Nj ð0Þ  tb ¼ DTi [2] Koren, Y., 2010, The Global Manufacturing Revolution: Product-
Process-Business Integration and Reconfigurable Systems, Wiley, Hoboken,
j¼b
NJ.
which agrees with the result in Eq. (1) when machine Mi is on the [3] Li, J., Blumenfeld, D. E., Huang, N., and Alden, J. M., 2009, “Throughput
downstream of Mb . Therefore, the effect of a failure on machine Analysis of Production Systems: Recent Advances and Future Topics,” Int. J.
Prod. Res., 47(14), pp. 3823–3851.
Mi which lies on the downstream of the bottleneck machine can [4] Jin, X., and Ni, J., 2013, “Joint Production and Preventive Maintenance Strat-
be equivalently evaluated in the EPSL. egy for Manufacturing Systems With Stochastic Demand,” ASME J. Manuf.
Proof of Lemma 2. If DTi  DTi , PMOWi ¼ ;, and Sci. Eng., 135(3), p. 031016.
jPMOWi j ¼ 0, the result is simply true. [5] Chang, Q., Ni, J., Bandyopadhyay, P., Biller, S., and Xiao, G., 2007,
“Maintenance Opportunity Planning System,” ASME J. Manuf. Sci. Eng.,
Otherwise, we prove it by mathematical induction. 129(3), pp. 661–668.
[6] Dallery, Y., and Gershwin, S. B., 1992, “Manufacturing Flow Line Systems: A
(a) Induction basis: Review of Models and Analytical Results,” Queueing Syst., 12(1), pp. 3–94.
Define j :¼ arg mink¼1;2;:::;j DTi;ðkÞ , and j1 ¼ mink¼1;:::;Ki [7] Cho, D. I., and Parlar, M., 1991, “A Survey of Maintenance Models for Multi-
fk : DTi > DTi;ðkÞ g. Then, j1 ¼ arg mink¼1;2;:::;j1 DTi;ðkÞ Unit Systems,” Eur. J. Oper. Res., 51(1), pp. 1–23.
cons res [8] Nicolai, R. P., and Dekker, R., 2008, “Optimal Maintenance of Multi-
¼ j1 , PMOWðj1 Þ ¼ ½Ti;ðj 1Þ
; DTi þ Ti;ðj  Þ Þ, and PMOWðj1 Þ j Component Systems: A Review,” Complex System Maintenance Handbook,
1

¼ DTi  DTi;ðj Þ > 0. Springer, London, UK.
1 [9] Wang, H., 2002, “A Survey of Maintenance Policies of Deteriorating Systems,”
(b) Inductive step: Eur. J. Oper. Res., 139(3), pp. 469–489.
Assume that there exist jm 2 f1; :::; Ki g and jm [10] Ambani, S., Li, L., and Ni, J., 2009, “Condition-Based Maintenance Decision-
:¼ arg mink¼1;2;:::;jm DTi;ðkÞ , such that PMOWðjm Þ Making for Multiple Machine Systems,” ASME J. Manuf. Sci. Eng., 131(3),
  p. 031009.
cons res  
 ½Ti;ðj1 Þ ; DTi þ Ti;ðj Þ Þ and PMOWðjm Þ ¼ DTi  DTi;ðjm Þ . [11] Lee, S., Li, L., and Ni, J., 2013, “Markov-Based Maintenance Planning Consid-
m
ering Repair Time and Periodic Inspection,” ASME J. Manuf. Sci. Eng.,
There are two possible cases. 135(3), p. 031013.
Case 1: DTi  DTi;ðkÞ for all k ¼ jmþ1 ; :::; Ki . [12] Armbruster, D., G€ ottlich, S., and Herty, M., 2011, “A Scalar Conservation Law
With Discontinuous Flux for Supply Chains With Finite Buffers,” SIAM J.
In this case, Ki ¼ jm and PMOWi ¼ PMOWðjm Þ proved. Appl. Math., 71(4), pp. 1070–1087.
Case 2: 9k 2 fjmþ1 ; :::; Ki g, such that DTi > DTi;ðkÞ . [13] Ni, J., and Jin, X., 2012, “Decision Support Systems for Effective Maintenance
Operations,” CIRP Ann. Manuf. Technol., 61(1), pp. 411–414.
In this case, we let jmþ1 ¼ mink¼jm þ1;:::;Ki fk : DTi > DTi;ðkÞ g [14] Chang, Q., Biller, S., and Xiao, G., 2010, “Transient Analysis of Downtimes
and jmþ1 ¼ arg mink¼1;2;:::;jmþ1 DTi;ðkÞ . From Eq. (9), and Bottleneck Dynamics in Serial Manufacturing Systems,” ASME J. Manuf.
Sh res cons cons Sci. Eng., 132(5), p. 051015.
PMOWðjmþ1 Þ ¼ PMOWðjm Þ DTi þ Ti;ðj  Þ  Ti;ðj Þ þ Ti;ðj
mþ1 Þ
; [15] Liu, J., Chang, Q., Xiao, G., and Biller, S., 2012, “The Costs of Downtime Inci-
m m
res
DTi þ Ti;ðjmþ1 Þ Þ. dents in Serial Multistage Manufacturing Systems,” ASME J. Manuf. Sci. Eng.,
134(2), p. 021016.
Case 2 can be further divided into the following two [16] Li, Y., Chang, Q., Brundage, M. P., Xiao, G., and Biller, S., 2013, “Standalone
subcases. Throughput Analysis on the Wave Propagation of Disturbances in Production
Sub-Systems,” ASME J. Manuf. Sci. Eng., 135(5), p. 051001.
Subcase 2.1: DTi;ðjmþ1 Þ  DTi;ðjm Þ . Then, jmþ1 ¼ jm and [17] Kuo, C. T., Lim, J. T., and Meerkov, S., 1996, “Bottlenecks in Serial Production
res cons cons res
½DTi þ Ti;ðj  Þ  Ti;ðj Þ þTi;ðj
mþ1 Þ
; DTi þ Ti;ðj mþ1 Þ
Þ ¼ [. There- Lines: A System-Theoretic Approach,” Math. Probl. Eng., 2(3), pp. 233–276.
m m   [18] Li, L., Chang, Q., and Ni, J., 2009, “Data Driven Bottleneck Detection of Man-
fore, PMOWðj Þ ¼ PMOWðj Þ , and PMOWðj Þ 
mþ1 mþ1 mþ1 ufacturing Systems,” Int. J. Prod. Res., 47(18), pp. 5019–5036.
¼ DTi  DTi;ðj Þ . [19] Gu, X., Lee, S., Liang, X., Garcellano, M., Diederichs, M., and Ni, J., 2013,
mþ1 “Hidden Maintenance Opportunities in Discrete and Complex Production
Subcase 2.2: DTi;ðjmþ1 Þ < DTi;ðjm Þ . Then, jmþ1 ¼ jmþ1 and Lines,” Expert Syst. Appl., 40(11), pp. 4353–4361.
res
Ti;ðj res
> Ti;ðj cons
 Þ þ Ti;ðj
cons
 Ti;ðjÞ and PMOWðj Þ  [20] Li, J., and Meerkov, S. M., 2008, Production System Engineering, WingSpan
 mþ1 Þ m mþ1 Þ m mþ1 Press, Livermore, CA.
¼PMOWðj Þ þj½DTi þT res T cons þT cons ;DTi þT res Þj
m i;ðjm Þ i;ðjm Þ i;ðjmþ1 Þ i;ðjmþ1 Þ
[21] http://www.simul8.com
[22] Hopp, W. J., and Spearman, M. L., 2008, Factory Physics, McGraw-Hill,
¼DTi DTi;ðj Þ. New York.
mþ1

Journal of Manufacturing Science and Engineering JUNE 2015, Vol. 137 / 031017-9

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use