You are on page 1of 6

The International

International
Proceedings
The Federation
of the 20th World
Federation of Congress
of Automatic Control
Automatic Control
Toulouse,
The France,
Proceedings of the
International
Toulouse, France, July
20th9-14,
World
Federation
July 9-14, 2017
of Congress
Automatic Control
2017
The International
Available online at www.sciencedirect.com
Toulouse, France,Federation of Automatic Control
July 9-14, 2017
Toulouse, France, July 9-14, 2017
ScienceDirect
IFAC PapersOnLine 50-1 (2017) 1087–1092
Performance Evaluation of Bernoulli Serial
Performance
Lines with Evaluation
Waiting Time of Bernoulli Serial
Constraints
Performance Evaluation of Bernoulli Serial
Lines with Waiting Time Constraints
Lines with Waiting
Jun-Ho Time Li
Lee ∗∗ Jingshan Constraints
∗∗
∗∗
Jun-Ho Lee Jingshan Li
Jun-Ho Lee ∗∗ Jingshan Li ∗∗

∗ School of Business, Jun-Ho LeeUniversity, Jingshan Li ∗∗05029, Republic of
School of Business, Konkuk Konkuk University, Seoul Seoul 05029, Republic of
∗ Korea. (e-mail:
(e-mail: junholee@konkuk.ac.kr).

School of Business,
Korea. Konkukjunholee@konkuk.ac.kr).
University, Seoul 05029, Republic of
∗∗
∗∗ School of Business, Konkuk University, Seoul 05029,University
Republic ofof
Department
Department of
of Industrial
Korea. (e-mail:
Industrial and
and Systems
Systems Engineering,
junholee@konkuk.ac.kr).
Engineering, University of
∗∗
Wisconsin,
Department
Wisconsin, Korea.
Madison,
of
Madison, (e-mail:
WI
Industrial
WI 53706,
53706,andjunholee@konkuk.ac.kr).
USA.
Systems
USA. (e-mail: jingshan.li@wisc.edu)
Engineering,
(e-mail: University
jingshan.li@wisc.edu) of
∗∗
Department
Wisconsin, of Industrial
Madison, WI 53706, and USA.
Systems Engineering,
(e-mail: University of
jingshan.li@wisc.edu)
Wisconsin, Madison, WI 53706, USA. (e-mail: jingshan.li@wisc.edu)
Abstract:
Abstract: Manufacturing
Manufacturing systems systems with with waiting
waiting time
time constraints
constraints are are widely
widely observed
observed in in battery,
battery,
automotive,
Abstract: semiconductor,
Manufacturing food
systems and
with many
waiting other
time
automotive, semiconductor, food and many other industries. Under such requirements, industries.
constraints Under
are such
widely requirements,
observed in the
battery,
the
Abstract:
residence time
automotive,
residence timeManufacturing
of aa part
part in
semiconductor,
of in systems
a buffer
a buffer
food and with
is waiting
is constrained
constrained
many time
other by constraints
byindustries.
aa given
given time
time are widely
limit
Under
limit andobserved
such
and the part in
requirements,
the part battery,
exceeding
the
exceeding
automotive,
the
the limit
residence
limit timeis semiconductor,
is subject
of a part
subject to
to quality food
in a buffer
quality and
inspection many andother
and
is constrained
inspection byindustries.
could
could abegiven
be scrapped
time
scrappedUnder withsuch
with
limit requirements,
aa certain
and the
certainpartprobability. the
exceeding
probability.
residence
To
the evaluate
limit time
is the of
subject a part
performance
to in a
quality buffer
of suchis
inspectionconstrained
systems,
and in by
could a
this given
be
To evaluate the performance of such systems, in this paper, we study Bernoulli serial lines paper, time
we
scrapped limit
study
with and
a the
Bernoulli
certainpart exceeding
serial lines
probability.
the
To
with limit
withevaluate
waiting
waiting is subject
the
time to quality
timeperformance
limits.
limits. For ofinspection
For two-machine
two-machine
such systems, and
lines,could
lines, in thisbepaper,
analytical
analytical scrapped
formulas with
we study
formulas aBernoulli
are derived
are certain
derived to probability.
to predict
predict the
serial lines
the
To
system
with
systemevaluate
waiting the
performance
performance performance
in
time limits. the
in theFor steady
steady of such
state.
two-machine systems,
In
state. In casecase of
lines, in this
longer
of analytical paper,
lines,
longer lines, we
direct
formulas study
analysis Bernoulli
are derived
direct analysis is serial lines
intractable,
to predict
is intractable, thus
the
thus
with
an
system waiting
an iterativeperformance
iterative time limits.
aggregation
aggregation in the For
procedure
steady
procedure two-machine
is
is presented.
state. lines,
In case
presented. analytical
ofBased
longeron
Based on formulas
extensive
lines,
extensive are derived
direct numerical
analysis
numerical to predictthus
is experiments,
intractable,
experiments, the
the
system
convergent
an performance
iterative results of
aggregation in
thethe steady
procedure state.
can
is beIn case
obtained
presented. of longer
and
Based lines,
high
on direct
precision
extensive
convergent results of the procedure can be obtained and high precision in performance estimation analysis
in is
performance
numerical intractable, thus
estimation
experiments, the
an iterativeresults
is achieved.
is achieved.
convergent aggregation procedure can
of the procedure is presented.
be obtained Based
and on high extensive
precisionnumerical
in performance experiments,
estimationthe
convergent
is achieved. results of the procedure can be obtained and high precision in performance estimation
© achieved.
is 2017, IFAC (International
Keywords: Federation of Automatic Control) Hosting machine,
by Elsevier Ltd. All rights reserved.
Keywords: Time Time constraint,
constraint, performance
performance evaluation,
evaluation, Bernoulli
Bernoulli machine, yield yield analysis
analysis
Keywords: Time constraint, performance evaluation, Bernoulli machine, yield analysis
Keywords: Time constraint, performance evaluation,
1. INTRODUCTION
INTRODUCTION Bernoulli machine,
Manufacturing systemsyield haveanalysis
attracted substantial
substantial re- re-
1. Manufacturing systems have attracted
1. INTRODUCTION search attentions
Manufacturing
search attentionssystems (see monographs
have attracted
(see monographs by
by Viswanadham
substantial and
Viswanadham re-
and
1. INTRODUCTION Manufacturing
Narahari
search
Narahari (1992);
attentions
(1992);systems
Buzacott
(see haveand
monographs
Buzacott and attracted
Shanthikumarsubstantial
by Viswanadham
Shanthikumar re-
(1993);
and
(1993);
search
Papadopoulos
Narahari attentions et
(1992); (see
al. monographs
(1993);
Buzacott andGershwin
Papadopoulos et al. (1993); Gershwin (1994); Li and by Viswanadham
Shanthikumar(1994); Li and
(1993);
Time Narahari
Meerkov (2009) (1992);
(2009) Buzacott
and reviewsand by Shanthikumar
Dallery (1994); (1993);
and Gershwin
Gershwin
Time constraints
constraints are are widely
widely observed
observed in in many
many manufac-
manufac- Papadopoulos
Meerkov et al.
and (1993);
reviews Gershwin
by Dallery and Li and
turing constraints
Time systems. For Forare instance,
widely in in battery in
observed cellmany
manufacturing
manufac- Meerkov Papadopoulos
(1992); et
Papadopoulos al.
(2009) and and
(1992); Papadopoulos (1993);
and Heavey
reviews
Heavey Gershwin
(1996);
by (1996); (1994);
Li
DalleryLiand et al. Li and
(2009);
Gershwin
et al. (2009);
turing systems. instance, battery cell manufacturing
Time
process,
turing constraints
constraints
systems. areon widely
For instance,waiting observed
times
in times
battery in
(i.e., many
the
cellthe timemanufac-
manufacturing a part Meerkov
Hudson
(1992); et (2009)
al.
Papadopoulos and
(2015)). reviews
Among
and Heavey by
them, Dallery
Bernoulli
(1996); Liand
et Gershwin
production
al. (2009);
process, constraints on waiting (i.e., time a part Hudson et al. (2015)). Among them, Bernoulli production
turing
can systems. For instance, in times
battery cellthemanufacturing (1992);
lines
lines have Papadopoulos
been studied and Heavey
extensively. (1996);
The Li
The et production
al. analysis
initial (2009);
can stay
process,
stay in
in aa machine
constraints
machine onor a
a buffer)
orwaiting
buffer) are enforced
are (i.e.,
enforced in
in tube
time tube fir-
fir- Hudson
a part have et al.
been(2015)).
studied Among them,
extensively. Bernoulli
initial analysis
process,
ing,
can stay constraints
material
in a filling
machine on
and waiting
orheating,
a buffer)times
andare (i.e.,
granular
enforcedthe time a
preparation.
in tube part
fir- Hudson
has
lines
has been
have
been et carried
al.
been(2015)).
carried out
studied
out Among
by
by Lim them,
et
extensively.
Lim et al.
al.Bernoulli
The(1990) production
initial
(1990) for
for highly
analysis
highly
ing, material filling and heating, and granular preparation.
can
During
ing,
Duringstay in afillings,
cell
material
cell machine
fillings,
filling theorcells
and
the a buffer)
cells
heating,needand
need arebe
to
to be enforced toinreach
tube the
baked preparation.
granular
baked to reach fir- lines
the reliable
has
reliable have
been been studied
production
carried
production lines,
out
lines, extensively.
byand Lim
and an
an et al. The
aggregation
aggregation initial
(1990) foranalysis
approach
approach has
highly
has
ing, material
designed
During filling
temperature and heating,
range and
after granular
adding eachpreparation.
material, has
been
reliablebeen carried
introduced
production byout by
Jacobs
lines, Lim
and and
an et al.
Meerkov
aggregation (1990)
(1995). for
approach highly
Follow-
has
designedcell fillings, the range
temperature cells needaftertoadding
be baked eachto reach
material,the been introduced by Jacobs and Meerkov (1995). Follow-
During
and
and thethecell
designed fillings,
next
next the range
material
temperature
material cells
mustneed
must be to
be
after be baked
added
adding
added within
each
within to aareach the reliable
certain
material,
certain ing
ing them,
been production
introduced
them, bylines,
performance
performance Jacobs andandan Meerkov
analysis
analysis aggregation
with
with diverse approach
(1995).
diverse has
scheduling
Follow-
scheduling
designed
time
and limit.
the temperature
Otherwise,
next material range
the
must after
cells
be haveadding
added to beeach
within material,
scrapped.
a certain In been
ing introduced
requirements,
them,
requirements, suchby as
performance
such Jacobs
as rework andand
analysis
rework Meerkov
and with (1995).
reentrance,
diverse
reentrance, hasFollow-
been
scheduling
has been
time limit. Otherwise, the cells have to be scrapped. In
and
time the
automotive
limit.
automotive next materialplants,
stamping
Otherwise,
stamping must
plants,
the be have
oils
cells
oils added
will
will to within
be
be sprayed aonto
be scrapped.
sprayed certain
onto the
theIn ing carriedthem,
requirements,
carried outperformance
out by Biller
Billeraset
bysuch analysis
al.
al. (2009);
etrework and
(2009);with diverse
Liu and
and Li
reentrance,
Liu Lischeduling
has(2009);
been
(2009);
time
top andlimit.
automotive Otherwise,
bottom parts
stamping of the
the cells
blank have
batch to be
after scrapped.
the washing In requirements,
Wang
carried and
out Liby such
(2010).
Biller asIn
et rework
addition,
al. and
(2009); the reentrance,
system
Liu and has
performance
Li been
(2009);
top and bottom parts plants,
of the blank oils will
batch be after
sprayed the onto
washingthe Wang and Li (2010). In addition, the system performance
automotive
operation
top and to stamping
bottom protect
parts plants,
from
of the oils
dirt.
blank will
However,
batch be sprayed
such
after the onto
protection
washingthe carried
with
Wang out
quality
and Liby Biller
control
(2010). et
device
In al. (2009);
has
addition, been
the Liu and
analyzed
system Liby (2009);
Chiang
performance
operation to protect from dirt. However, such protection with quality control device has been analyzed by Chiang
top
will and
operation
will be bottom
be fading
fading parts
after
to protect
after four
four of to
from thesix
to blank
six
dirt.hours
hours batch
if they
However,
if theyafter
arethe
such
are not
not washing
sent to
protection
sent to Wang(2006);
with
(2006); and
quality Li control
Meerkov
Meerkov (2010).
and In addition,
and device
Zhang
Zhang (2010,
has
(2010, the
been system
2011), andperformance
analyzed
2011), and variabilities
by Chiang
variabilities
operation
stamping
will to
press,
be fading protect
which
after fourfrom
will dirt.
lead
to lead However,
to dimension
six hours dimension
if they aresuch protection
quality
not sentissues. with
have
to (2006);
have beenquality
beenMeerkovcontrol
addressed device
by
and Zhang
addressed Li
by Li and has
and
(2010, been
Meerkov
Meerkov analyzed
(2000).
2011), (2000). byWhereas
Chiang
and variabilities
Whereas
stamping press, which will to quality issues.
will
It is be
also
stamping fading after
observed
press, which four
that willto
long six hours
waiting
leadwaiting if they
times
to dimension are
in not sent
automotive
quality to (2006);
most
have of
beenMeerkov
studies and
focus
addressed Zhang
by onLi (2010,
steady
and
issues. most of studies focus on steady state analysis, Meerkov 2011),
state
Meerkov and
analysis,
(2000). variabilities
Meerkov
Whereas
It is also observed that long times in automotive
stamping
paint
It shops
is also
paint shopspress,
willwhich
will
observed lead
that
lead will
to
tolong leadwaiting
degraded
degradedto dimension
paint quality
timesquality.
paint in issues. have
Similar
automotive
quality. Similar and Zhang
most
and been
Zhang addressed
(2008);
of studies focus
(2008); byonListeady
Zhang
Zhang and
et al.Meerkov
et al. (2013); (2000).
Ju et
state analysis,
(2013); Ju al.Whereas
et al. (2017)
Meerkov
(2017)
It is also
problems
paint shopsobserved
canwillbe leadthat
foundtoin long waiting
indegraded
semiconductor times in automotive
manufacturing,
paint quality. Similar and most
haveZhang
have of studies
analyzed
analyzed focus
transient
(2008); Zhang
transient on steady
properties. state
et al. (2013);
properties. analysis,
Extensive al.Meerkov
application
Ju etapplication
Extensive (2017)
problems can be found semiconductor manufacturing,
paint
food shops canwill
food processing,
problems
processing, lead
foundtoindegraded
be etc.
etc. paint quality.
semiconductor Similar and
manufacturing, studies
have
studies Zhanghave (2008);
have
analyzed been
been Zhang
carried
transient
carried outetusing
al. (2013);
properties.
out using the
the Ju etapplication
Bernoulli
Extensive
Bernoulli al. (2017)
reliability
reliability
problems
food can be etc.
processing, found in semiconductor manufacturing, have model,
studies analyzed
which
have beentransient
has shown
carried properties.
broad
out using
model, which has shown broad practicability in manufac-theExtensive
practicability
Bernoulli inapplication
manufac-
reliability
In such
such systems, etc. after finishing
finishing a a process
process at at the
the preceding
preceding studies have
In
food systems,
processing, after model,
turing whichbeen
turing systems
systems has(Licarried
(Li and Meerkov
shown
and out
Meerkov
broad using the Bernoulli
(2009)).
practicability
(2009)). in reliability
manufac-
machine,
In
machine, a
such systems, part needs
a part needs to
after finishing be moved
to be moved to
a process the
to the next
at next machine
the preceding
machine model, which has shown broad practicability in manufac-
In suchthe systems, after finishing a process turing systems (Li and Meerkov (2009)).
within
machine,
within the apre-determined
part needs
pre-determined to time limits.
be moved
time limits. toIf itat
it
If the the at
stays
next
stays preceding
at aa buffer
machine
buffer turingHowever,
However, only limited
only (Li
systems limited work
work is
and Meerkov is related
(2009)).to
related to systems
systems with with
machine,
longer
within than
the a part
the needs
limit,
pre-determined itto isbe
time moved
subject
limits.to
longer than the limit, it is subject to be scrapped with ato be
If the
it next
scrapped
stays at machine
a with
buffer a time
However,
time constraints.
only
constraints. For
limited
For example,
work
example, isLiberopoulos
related
Liberopoulos to et al.
al. (2007)
systems
et (2007)
with
within
certainthe
longer
certain than pre-determined
probability
the limit,
probability it istime
according
according tolimits.
subject
to to If
quality
quality itscrapped
inspection
beinspection buffera However,
stays at aresults.
results.
with consider
time
consider anonly
constraints.
an limited
automatic work isLiberopoulos
transfer
For example,
automatic transfer related
line
line withtomaterial
with systems
material with
et al. deteri-
(2007)
deteri-
longer
The
certain than
yield of the
the
probability limit, it
non-defective is subject
according toparts parts to
qualityis be scrapped
determined
inspection results.with
by the a time
orating
consider constraints.
and
an For
possible
automatic example,
scrapping
transfer Liberopoulos
during
line withlong et al.
failure
material (2007)
times.
deteri-
The yield of the non-defective is determined by the orating and possible scrapping during long failure times.
certain
one
one that
The probability
yield
that does
of the
does not according
exceed
notnon-defective
exceed toparts
waiting
waiting quality
time
time inspection
limits,
limits, or
is determined or the the consider
results.
by one
the one Naebulharam
orating
Naebulharam an automatic
and and
and Zhang
possibleZhang transfer
(2014)
scrapping line
(2014)during with
introduce
introduce longmaterial
an
anfailure deteri-
aggregation
times.
aggregation
The
that
one yield
that of
satisfies
does thenotnon-defective
quality standard
exceed waitingparts
even is determined
though
time it
limits, exceeds
or by
the the
one orating
method
Naebulharam
method and
for
for possible
performance
and Zhang
performance scrapping
(2014) during
evaluation
evaluation and
introduce
andlong anfailure
present
present times.
system-
aggregation
system-
that satisfies quality standard even though it exceeds the
one that
time satisfies
that does
limits. TheThenot
quality exceed
system
standard waiting
production time limits,
rate, defined
even though or
defined
it exceedsthe one
by the
the methodNaebulharam
theoretic
theoreticfor and
properties Zhang
performance
properties of (2014)
Bernoulli introduce
production
evaluationproduction
of Bernoulli an
and present aggregation
lines with
system-
lines with
time limits. system production rate, by
that satisfies
expected
time number
limits. quality of
The system standard
non-defective even
production though
parts it
produced
rate,produced exceeds by the method
quality
theoretic for performance
deterioration
properties of evaluation
functions.
Bernoulli
defined by the quality deterioration functions. The transient behavior The and present
transient
production system-
behavior
lines with
expected number of non-defective parts
time limits. The theoretic properties offunctions.
Bernoulli production
transientlines with
last
last machine
expected
machine numberin
in a asystem
production
of production
non-defective
production line
line duringrate,produced
parts
during adefined
a cycle,
cycle, willby the
by the
will be of
of two-machine
be quality Bernoulli
deterioration
two-machine Bernoulli lines
lines with The
with residence
residence time con-
behavior
time con-
expected
dependent
last machine number
on the
in a of
yieldnon-defective
of
production such line parts
processes
during produced
witha waiting
cycle, by the
time
will be quality
straints
of deterioration
has
two-machine
straints has been
been studied
Bernoulli
studied functions.
by
by Ju
lines
Ju et
et The(2017)
al.
with
al. transient
(2017)
residenceand
and abehavior
Markov
time
a con-
Markov
dependent on the yield of such processes with waiting time
last machine
constraints.
dependent
constraints. on inthea yield
production
of such line duringwith
processes a cycle,
waiting will
timebe of two-machine
decision
straints
decision has beenBernoulli
process
process model
studiedis
model lines
isbyused with
Ju etfor
used residence
al.production
for (2017)
production time
andcontrol con-
a Markov
control to
to
dependent
constraints. on the yield of such processes with waiting time straints
decision has been
process studied
model isby Ju
used et al.
for (2017)
production and a Markov
control to
constraints. decision process model is used for production control to
Copyright
Copyright ©© 2017
2017 IFAC
IFAC 1110
1110
2405-8963 ©
Copyright © 2017, IFAC (International Federation of Automatic Control)
2017 IFAC 1110Hosting by Elsevier Ltd. All rights reserved.
Peer review under responsibility
Copyright © 2017 IFAC of International Federation of Automatic
1110Control.
10.1016/j.ifacol.2017.08.387
Proceedings of the 20th IFAC World Congress
1088
Toulouse, France, July 9-14, 2017 Jun-Ho Lee et al. / IFAC PapersOnLine 50-1 (2017) 1087–1092

reduce scrap rate. In addition, there have been extensive the scenario when a part is expensive and the qual-
efforts on inventory control and scheduling of production- ity deterioration can be prevented by repair or re-
inventory systems with time constraints (see reviews by processing with a certain probability βi . Semicon-
Nahmias (1982) and Karaesmen et al. (2011)). The ductor manufacturing is a typical example.
schedulability analysis and optimal scheduling methods (vi) Machine mi , i = 1, . . . , M − 1, is blocked during a
of automated manufacturing machines in time-constrained time slot if it is up, buffer bi is full at the beginning
processes can also be found in Kim et al. (2016) and Wu of the time slot, and machine mi+1 does not take a
et al. (2008). part from bi . Machine mM is never blocked.
In spite of these efforts, serial production lines with waiting (vii) Machine mi , i = 2, . . . , M , is starved during a time
time constraints still need in-depth study and efficient slot if it is up and buffer bi−1 is empty at the
methods for performance evaluation of such systems are beginning of the time slot. Machine m1 is never
still required. Therefore, this paper is devoted to develop- starved.
ing such methods. With above assumptions, the following measures are under
The rest of this article is organized as follows: Section consideration:
2 introduces the assumptions and formulates the prob- • Consumption rate at mi , i = 2, . . . , M , (CRi ): the
lem. Based on such assumptions, a closed formula and expected number of parts consumed by mi during one
an efficient procedure for performance evaluation of two- cycle in the steady state.
machine and longer lines are provided in Sections 3 and • Yield at mi , i = 2, . . . , M , (Yi ): the expected number
4, respectively. The properties of the proposed procedure, of parts produced at mi during one cycle with no time
such as convergence and accuracy, are also studied. Fi- constraint violation at bi−1 in the steady state.
nally, Section 5 summarizes the work and discusses future • Scrap rate at mi , i = 2, . . . , M , (SRi ): the expected
research. Due to space limitation, all proofs are omitted number of parts scrapped at mi during one cycle in
and can be found in Lee et al. (2016). the steady state; that is, (CRi − Yi ) × (1 − βi ).
• Production rate at mi , i = 2, . . . , M , (P Ri ): the
2. SYSTEM DESCRIPTION expected number of non-defective parts produced at
mi during one cycle in the steady state; thiat is,
In this paper, we consider an M -machine serial line shown CRi − SRi .
in Figure 1, where the circles represent the machines and
the rectangles are the buffers. The following assumptions The problem to be studied in this paper is: Under assump-
and descriptions define the machines, the buffers, the tions (i)-(vii), develop an efficient method to evaluate the
waiting time limits and their behaviors. production rate of the system (i.e., P RM ).
(i) The production line consists of M machines, m1 , . . . , Solutions to the problem are presented in Sections 3 and
mM , and M −1 buffers, b1 , . . . , bM −1 , separating each 4 below.
pair of consecutive machines.
(ii) Machines have a constant and identical processing 3. TWO-MACHINE LINES
time which is slotted with cycle time. Hereafter, the
cycle time is assumed to be 1 time unit without loss For two-machine lines with a waiting time constraint,
of generality. direct analysis of production rate and other performance
(iii) The machines follow the Bernoulli reliability model; measures is possible. As all parts will go through machine
that is, in each cycle, machine mi , i = 1, . . . , M , is m2 no matter the waiting time constraint is violated or
able to produce a part with probability pi and fails not, the consumption rate of machine m2 will be equal to
to do so with probability (1−pi ). The machine status the production rate of all parts (including both defective
changes at the beginning of a cycle. and non-defective ones), which can be obtained as (see Li
(iv) Buffer bi , i = 1, . . . , M − 1, has a finite capacity and Meerkov (2009)):
Ni (1 ≤ Ni < ∞). First-in-first out (FIFO) is
assumed for the buffer outflow process and the buffer CR2 = p2 [1 − Q(p1 , p2 , N1 )] = p1 [1 − Q(p2 , p1 , N1 )], (1)
occupancy changes at the end of the cycle.
(v) From the time when a part is stored into buffer bi , where Q(p1 , p2 , N1 ) represents the probability that buffer
i = 1, . . . , M − 1, after finishing its process at mi , b1 is empty, and is defined as:
the maximum allowable waiting cycle before starting 
 1−p
the next process at mi+1 is constrained by Ti . When 
 , if p1 = p2 = p,
a part stays at bi longer than Ti , it will be scrapped N +1−p
Q(p1 , p2 , N ) := (1 − p1 )(1 − α) (2)
with probability (1 − βi+1 ) after finishing its next 
 , if p1 = p2 ,
 1− αp1 N
process at mi+1 ; whereas a part goes to the next p2
buffer with probability βi+1 without any quality and
degradation (see Figure 1). p1 (1 − p2 )
α= . (3)
Remark 1. In this model, a part keeps process- p2 (1 − p1 )
ing at the downstream machine even though its
waiting cycle exceeds the time limit in the buffer. From Naebulharam and Zhang (2014), it is shown that
Whereas some studies assume direct scrap of expired the probability that buffer b1 has i − 1 parts when m1
parts from the buffer (e.g., Naebulharam and Zhang produces a part into the buffer, denoted as Pi−1 , can be
(2014) and Ju et al. (2017)), this model reflects calculated as:

1111
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Jun-Ho Lee et al. / IFAC PapersOnLine 50-1 (2017) 1087–1092 1089

Scrapped Scrapped

Fig. 1. An M -machine production line with waiting time constraints.

Q(p1 , p2 , N1 ) 4.1 Aggregation Procedure


P0 = , (4)
(1 − p1 )(1 − Q(p2 , p1 , N1 ))
A machine-aggregation approach for performance evalua-
Pi = αi P0 , i = 1, . . . , N1 − 1. (5) tion of serial lines has been introduced in existing literature
and it is shown that the approach is efficient and leads
For a given buffer occupancy i ∈ {1, . . . , N1 }, let X1,i to accurate estimation (Li and Meerkov (2009); Naebul-
denote the waiting time of the top part in b1 to start its haram and Zhang (2014)). However, most of previous
next process at m2 . Then X1,i is the sum of variables studies do not consider waiting time constraints and hence
(the time to finish the process at m2 ) following the they cannot be applied to the proposed research directly.
identical geometric distribution Geo(p2 ), and X1,i follows Furthermore, in contrast to the traditional lines, lines with
the negative binomial distribution, i.e., X1,i ∼ N B(i, p2 ). waiting time constraints may not have the characteristics
Then that facilitate convergence of the aggregation procedure,
  such as monotonicity, which will impose significant effect
x−1 i
P [X1,i = x] = p (1 − p2 )x−i , i = 1, 2, . . . , N1 . on the study. Below, we present an iterative aggregation
i−1 2 procedure reflecting waiting time constraints and analyze
Using these probabilities, we can obtain the stationary its convergence and accuracy.
probability that a part exceeds the waiting time limit T1 , The key idea of the aggregation procedure is to contin-
for buffer capacity N1 : uously aggregate two consecutive machines into one ma-
chine, and use it to aggregate with the next consecutive
N1
 machine, until the whole line is represented by one ma-
Pi−1 (1 − P [X1,i ≤ T1 ]) chine. The proposed procedure is introduced as follows:
i=1
  Procedure 1. Iterative aggregation procedure
N1
 T1 
 
j−1   
= Pi−1 1 − pi2 (1 − p2 )j−i  . (6) pbi (s + 1) = pi 1 − Q pbi+1 (s + 1), pfi (s), Ni , (10)
i=1 j=i
i−1
i = 1, . . . , M − 1, s = 0, 1, 2, . . . ,
Then the yield Y2 can be calculated as: pfi (s + 1) = (1 − βi )Y (pfi−1 (s + 1), pbi (s + 1), Ni−1 , Ti−1 )
 
Y2 = Y (p1 , p2 , N1 , T1 ) +βi pi 1 − Q(pfi−1 (s + 1), pbi (s + 1), Ni−1 ) ,(11)
 N1  T1   
j−1 i i = 2, . . . , M, s = 0, 1, 2, . . . ,
= CR2 1 − 
Pi−1 1 − p (1 − p2 ) j−i
.(7)
i=1 j=i
i−1 2 with initial conditions

In addition, the scrap rate and line production rate can be pfi (0) = pi , i = 1, . . . , M, (12)
obtained: and boundary conditions

SR2 = (CR2 − Y2 ) · (1 − β2 ), (8) pf1 (s) = p1 , s = 0, 1, 2, . . . , (13)


P R2 = CR2 − SR2 . (9) pbM (s) = pM , s = 0, 1, 2, . . . . (14)

4. M > 2-MACHINE LINES In the procedure, equation (10) represents the backward
aggregation, which is similar to the one introduced in Li
For M > 2-machine lines, direct analysis is intractable and Meerkov (2009). However, the forward aggregation,
due to high complexity in the system. In fact, even for represented in equation (11), needs to accommodate the
lines without waiting time constraints, a closed formula is scenario introduced in assumption (v) of Section 2, i.e.,
not available for M > 2-machine lines (Li and Meerkov parts exceeding waiting time limits will be inspected and
(2009)). Thus an approximation method is pursued. Here some of them will be defective. Thus, its production rate
we present an iterative aggregation procedure to provide addresses both parts within waiting time constraints and
a precise estimation of the system production rate. parts exceeding the constraints but passing the inspection.

1112
Proceedings of the 20th IFAC World Congress
1090
Toulouse, France, July 9-14, 2017 Jun-Ho Lee et al. / IFAC PapersOnLine 50-1 (2017) 1087–1092

4.2 Convergence 0.95

For lines without waiting time constraints, the convergence 0.9


of pfi (s)’s and pbi (s)’s in terms of s can be mathematically
proved based on the monotonicity and boundness of pro- 0.85
duction rate (see Li and Meerkov (2009)). However, when
it comes to lines with waiting time constraints, it is all but
0.8
impossible to prove the convergence since production rate
has no monotonicity with respect to upstream machine’s
reliability. Therefore, the convergence of Procedure 1 is 0.75

examined numerically.
0.7
We first introduce the numerical results on convergence of 0 1 2 3 4 5 6 7 8 9 10 (s)
Procedure 1 with various line configurations. The numer-
ical experiments are carried out for M ∈ {3, 4, . . . , 15}.
For each M , 10,000 lines are randomly generated from the Fig. 2. A 5-machine line with pi = 0.9, i = 1, . . . , 5,
following sets: Ni = Ti = 3, i = 1, . . . , 4, and βi = 0.5, i = 2, . . . , 5.
Numerical Fact 1. pbi ’s and pfi ’s are convergent if pi ’s,
pi ∈ (0.7, 0.99), i = 1, . . . , M, i = 1, . . . , M , are identical or monotonically increasing in
Ni ∈ {3, 4, . . . , 7}, i = 1, . . . , M − 1, i, i = 1, . . . , M .
Ti ∈ {Ni − 1, Ni , . . . , Ni + 5}, i = 1, . . . , M − 1, (15) In addition, we investigate the convergence of the proce-
βi ∈ (0.3, 0.9), i = 2, . . . , M. dure with respect to line balancing. Numerical experiments
are carried out for 5-machine lines with 10,000 instances
With each randomly generated instance from (15), 300 following the same sets of (15). The balance of the lines is
iterations of Procedure 1 are carried out and the following measured by |pmax − pmin |, where
criterion is used to determine the convergence of the
procedure, pmax = max[p1 , p2 , . . . , p5 ],
pmin = min[p1 , p2 , . . . , p5 ].
|pfM (300) − pfM (299)| < 0.0001. (16)
As shown in Table 2, Procedure 1 always converges if
As shown in Table 1, 98% of lines, on average, have |pmax − pmin |  1, i.e., a line is well balanced. Since in
f
reached convergent PM ’s. It is notable that it typically practice many of real production lines are designed to be
converges within 5 to 10 iterations. An example of such well balanced, this implies that the procedure converges
cases is illustrated in Figure 2, which shows the execution and is applicable to these lines.
result of Procedure 1 for a 5-machine line where pi = 0.9,
i = 1, . . . , 5, Ni = Ti = 3, i = 1, . . . , 4, and βi = 0.5, Table 2. Convergence rate of Procedure 1 with
respect to line balancing
i = 2, . . . , 5. As one can see, in this case, all of pfi ’s and
pbi ’s converge to certain values and pf5 (= pfM ) indicates |pmax − pmin | Converging (%) Non-converging (%)
the line production rate (i.e., P R5 ). < 0.01 100.0 0.0
< 0.02 100.0 0.0
Table 1. Convergence rate of Procedure 1 < 0.03 100.0 0.0
< 0.05 100.0 0.0
Number of machines Converging (%) Non-converging (%) < 0.1 99.9 0.1
M =3 99.2 0.8 < 0.2 99.2 0.8
M =4 98.3 1.7
M =5 98.1 1.9
M =6 98.1 1.9
To further investigate the behavior of non-convergent
M =7 98.1 1.9 cases, we examine pfM (s) as iterations of Procedure 1 are
M =8 98.0 2.0 carried out. Figure 3 depicts a typical behavior of pfi ’s
M =9 98.2 1.8
and pbi ’s for a 5-machine line with {p1 , p2 , p3 , p4 , p5 } =
M = 10 97.7 2.3
M = 11 98.1 1.9
{0.94, 0.97, 0.99, 0.8, 0.86}, Ni = 5, i = 1, . . . , 4, Ti = 3,
M = 12 97.9 2.1 i = 1, . . . , 4, and βi = 0.5, i = 2, . . . , 5. As one can
M = 13 98.2 1.8 see, each pfi or pbi oscillates between two values as the
M = 14 97.8 2.2 iteration count increases. We define these two values as pjU i
M = 15 98.1 1.9
Average 98.1 1.9
and pjL
i , j ∈ {f, b}, which denote the larger and smaller
ones among the two oscillating values, respectively. From
extensive numerical experiments, we obtain the following
As there still exist non-converging cases that account for numerical fact on oscillating properties of pfi ’s and pbi ’s.
about 2%, we further investigate the factors that may
impact convergence. First, based on 10,000 production Numerical Fact 2. All non-converging pfi ’s and pbi ’s are
lines that are randomly generated with (15), we observe oscillating between two values defined by pfi U and pfi L ,
the following, summarized as a numerical fact below: and pbUi and pbL
i , respectively.

1113
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Jun-Ho Lee et al. / IFAC PapersOnLine 50-1 (2017) 1087–1092 1091

1 When the procedure results in oscillating results, under


0.9
Numerical Fact 2, pjU
i and pjL
i can formally be defined as
follows:
0.8  jU

 p = lim pji (s), pjL lim pji (s + 1),
i = s→∞
 i
 s→∞
0.7 

 if lim pji (s) > lim pji (s + 1),
s→∞ s→∞
0.6 (18)


 pjU = lim pj (s + 1), pjL = lim pj (s),

0.5 
 i s→∞ i i s→∞ i

otherwise.
0.4
0
3
6
9
12
15
18
21
24
27
30
33
36
39
42
45
48
51
54
57
60
63
66
69
72
75
78
81
84
87
90
93
96
99
p2f p3f p4f p5f p1b p2b p3b p4b Note that based on extensive numerical experiments, we
observe that such limits exist, i.e., the upper and lower
Fig. 3. A 5-machine line with oscillating pfi ’s and pbi ’s. bounds converges. Then, intuitively, three types of approx-
imations for line production rates can be provided using
4.3 Accuracy pfMU and pfML as follows:
When Procedure 1 converges, the line production rate is pfMU + pfML
easily defined: Average: ,
2
P R = P RM = pfM . fU
Upper: pM , (19)
To examine the accuracy of P R estimation, the production
Lower: pfML .
rate obtained from the procedure, P RP , is compared to
that from simulation experiments, P RS . 300 iterations Table 4 shows gaps between the production rates obtained
of Procedure 1 are conducted to obtain P RP ; that is, by Procedure 1 and simulation experiments using param-
P RP = pfM (300). We note that the execution time of eters generated in (15). ‘Average’, ‘Upper’, and ‘Lower’
the procedure with a program coded in Java is almost defined in (19) are used to compute the gaps in Table 4.
0 even for a large number of iterations. The simulation is The generation of instances and simulation settings are
performed for M -machine lines, M ∈ {3, 4, . . . , 15}, with similarly conducted with those for converging cases. From
a program coded in Java. For each M , 10,000 virtual lines these results, we can observe that the approximation of
are randomly generated with the same sets in (15). To production rate computed by average value of pfMU and
obtain more convincing results, the simulation is carried
out for 3 times repetitively for each instance with 400,000 pfML gives the most accurate one. Moreover, in such cases,
cycles after 100,000 warming-up cycles. The confidence the gap is about 2% on average. Therefore, Procedure 1
intervals are within 0.02% of the mean values. The gap can also be utilized for non-converging cases with high
between the results obtained from simulation and aggre- accuracy. Note that when the procedure converges, all
gation procedure is defined as three measures will be the same.
Table 4. Accuracy of non-converging cases
|P RP − P RS |
Gap(%): × 100. (17) Number Gap (%) Gap (%) Gap (%)
P RS of machines Average Upper Lower
The comparison results are presented in Table 3. As one M =3 2.20 3.71 4.61
can see, the gap is less than 2% on average. For large M , M =4 1.70 2.87 3.23
M =5 1.70 2.71 2.31
the gap is almost saturating at about 2%. Therefore, we
M =6 2.08 2.87 2.54
can conclude that Procedure 1 provides a highly accurate M =7 1.97 2.93 2.11
approximation of line production rate for most practical M =8 1.97 2.78 2.53
production lines. M =9 2.20 3.15 2.30
M = 10 2.08 3.10 2.26
Table 3. Accuracy of converging cases M = 11 2.34 3.14 2.30
M = 12 2.18 2.99 2.30
Number of machines Gap (%)
M = 13 2.20 3.11 2.16
M =3 0.48 M = 14 2.10 2.62 2.02
M =4 0.85 M = 15 2.21 2.96 2.04
M =5 1.16 Average 2.06 2.94 2.34
M =6 1.38
M =7 1.54
M =8 1.69 In addition to production rate, other measures introduced
M =9 1.78
M = 10 1.89
in Section 2 can be directly computed with pfi ’s and pbi ’s
M = 11 1.95 of the proposed procedure.
M = 12 2.00
M = 13 2.02 5. CONCLUSIONS
M = 14 2.05
M = 15 2.05 In this paper, a serial production line model with Bernoulli
Average 1.60 machines, finite buffers, and waiting time constraints is

1114
Proceedings of the 20th IFAC World Congress
1092
Toulouse, France, July 9-14, 2017 Jun-Ho Lee et al. / IFAC PapersOnLine 50-1 (2017) 1087–1092

presented to analyze the performance of such systems. A Li, J. and Meerkov, S.M. 2009. Production Systems Engi-
closed formula and an efficient procedure for performance neering, Springer.
evaluation of two-machine and longer lines are presented, Liberopoulos, G., Kozanidis, G. and Tsarouha, P. 2007.
respectively. Extensive numerical experiments are carried Performance evaluation of an automatic transfer line
out to examine the convergence and accuracy of the pro- with wip scrapping during long failures. Manufacturing
posed procedure and finally it is shown that the procedure & Service Operations Management, 9(1), 62-83.
provides an accurate approximation of production rate and Lim, J.-T., Meerkov, S. M. and Top, F. 1990. Homoge-
is applicable to most of practical problems. neous, asymptotically reliable serial production lines:
theory and a case study. IEEE Transactions on Auto-
In future work, this study can be extended to other relia-
matic Control, 35(5), 534-534.
bility models (e.g., exponential, Weibull, and general dis-
Liu, Y. and Li, J. 2009. Modelling and analysis of split
tributions). Another direction is on optimal buffer design
and merge production systems with Bernoulli reliability
and production control to maximize the line performance
machines. International Journal of Production Research,
since production rate has no monotonicity on buffer size
47(16), 4373-4397.
when waiting time limits are introduced. Finally, applying
Meerkov, S.M. and Zhang, L. 2008. Transient behavior
the results on the factory floor to validate the model and
of serial production lines with Bernoulli machines. IIE
method will be pursued.
Transactions, 40(3), 297-312.
Meerkov, S.M. and Zhang, L. 2010. Product quality in-
REFERENCES spection in Bernoulli lines: Analysis, bottlenecks, and
design. International Journal of Production Research,
Biller, S., Marin, S. P., Meerkov, S. M. and Zhang, L. 2009. 48(16), 4745-4766.
Closed Bernoulli production lines: analysis, continuous Meerkov, S.M. and Zhang, L. 2011. Bernoulli production
improvement, and leaness. IEEE Transactions on Au- lines with quality-quantity coupling machines: mono-
tomation Science and Engineering, 6(1), 168-180. tonicity properties and bottlenecks. Annals of Opera-
Buzacott, J.A. and Shanthikumar, J.G. 1993. Stochastic tions Research, 182(1), 119-131.
Models of Manufacturing Systems, Prentice Hall, En- Naebulharam, R. and Zhang, L. 2014. Bernoulli serial lines
glewood Cliffs, NJ. with deteriorating product quality: performance eval-
Chiang, S.-Y. 2006. Bernoulli serial production lines with uation and system-theoretic properties. International
quality control devices: Theory and application. Mathe- Journal of Production Research, 52(5), 1479-1494.
matical Problems in Engineering, 1-30. Nahmias, S. 1982. Perishable inventory theory: a review.
Dallery, Y. and Gershwin, S.B. 1992. Manufacturing flow Operations Research, 30(4), 680-708.
line systems: a review of models and analytical results. Papadopoulos, H.T., Browne, J. and Heavey, C. 1993.
Queuing Systems, 12(1), 3-94. Queueing Theory in Manufacturing Systems Analysis
Gershwin, S.B. 1994. Manufacturing Systems Engineering, and Design. Chapman & Hall.
Prentice Hall, Englewood Cliffs, NJ. Papadopoulos, H.T. and Heavey, C. 1996. Queueing theory
Hudson, S., McNamara, T. and Shaaban, S. 2015. Unbal- in manufacturing systems analysis and design: a clas-
anced lines: where are we now? International Journal of sification of models for production and transfer lines.
Production Research, 53(6), 1895-1911. European Journal of Operational Research, 92(1), 1-27.
Jacobs, D. and Meerkov, S.M. 1995. A system-theoretic Viswanadham, N. and Narahari, Y. 1992. Perfor-
property of serial production lines: improvability. Inter- mance Modeling of Automated Manufacturing Systems.
national Journal of Systems Science, 26(4), 755-785. Prentice-Hall. Englewood Cliffs, NJ.
Ju, F., Li, J. and Horst, J. A. 2017. Transient analy- Wang, C. and Li, J. 2010. Approximate analysis of reen-
sis of serial production lines with perishable products: trant lines with Bernoulli reliability model. IEEE Trans-
Bernoulli reliability model. IEEE Transactions on Au- actions on Automation Science and Engineering, 7(3),
tomatic Control, 62(2), 694-707. 708-715.
Karaesmen, I., Scheller-Wolf, A. and Deniz, B. 2011. Man- Wang, J., Hu, Y. and Li, J. 2010. Transient analysis
aging perishable and aging inventories: Review and fu- to design buffer capacity in dairy filling and packing
ture research directions. Planning Production and In- production lines. Journal of Food Engineering, 98(1), 1-
ventories in the Extended Enterprise, 393-436. 12.
Kim, H.-J., Lee, J.-H. and Lee, T.-E. 2016. Schedulabil- Wu, N., Chu, C., Chu, F. and Zhou, M. C. 2008. A Petri
ity analysis for noncyclic operation of time-constrained net method for schedulability and scheduling problems
cluster tools with time variation. IEEE Transactions on in single-arm cluster tools with wafer residency time
Automation Science and Engineering, 13(3), 1409-1414. constraints. IEEE Transactions on Semiconductor Man-
Lee, J.-H., Zhao, C. and Li, J. 2016. Bernoulli production ufacturing, 21(2), 224-237.
lines with waiting time constraints. Technical Report, Zhang, L., Wang, C., Arinez, J. and Biller, S. 2013.
Department of Industrial and Systems Engineering, Uni- Transient analysis of Bernoulli serial lines: Performance
versity of Wisconsin, Madison, WI. evaluation and system-theoretic properties. IIE Trans-
Li, J., Blumenfeld, D.E., Huang, N. and Alden, J.A. (2009) actions, 45(5), 528-543.
Throughput analysis of production systems: recent ad-
vances and future topics. International Journal of Pro-
duction Research, 47(14), 3823-3851.
Li, J. and Meerkov, S. M. 2000. Production variability
in manufacturing systems: Bernoulli reliability case.
Annals of Operations Research, 93(1), 299-324.

1115

You might also like