Professional Documents
Culture Documents
A Heu Problem Uristic A Mwithu Approac Unrelate Chtos Ed Parall Solve H Lel Mach Hybrid F Hines Flow-Sh Hop SCH Heduling
A Heu Problem Uristic A Mwithu Approac Unrelate Chtos Ed Parall Solve H Lel Mach Hybrid F Hines Flow-Sh Hop SCH Heduling
17)
March
h 2017, Volumee 28, Number 1
pp. 61- 774
Ebrahim
m Asadi Gangraj
G
Ebrahim Assadi Gangraj, Babol Noshirrvani Universsity of Technology
KEYWOR RDS A
ABSTRACT
T
Bottleneck--based IIn hybrid flow-shop
f s
scheduling problem (HFS)
( with unrelated
Heuristic, p
parallel maachines, a set s of n jobss is processsed on k machines.
m A
Lower bounnd, m
mixed integ ger linear programmi
p ing (MILP)) model forr the HFS
Hybrid floww shop, s
scheduling problems with
w unrela ated paralleel machiness has been
Makespan.
p
proposed to
o minimize the maximuum complettion time (m makespan).
S
Since the prroblem is shown
s to bee NP-compllete, it is neecessary to
u heuristtic methodss to tacklee the modeerate- to large-scale
use l
p
problems. T article presents
This p a new
n bottleneck-based heuristic
h to
s
solve the problem.
p Too improve the
t perform mance of thee heuristic
m
method, a local
l searchh approachh is embeddded in the structure off
t heuristicc method. To
the T evaluate the perform mance of thee proposed
h
heuristic meethod, a neew lower boound is deveeloped baseed on Kurz
a Askin [1]
and [ lower boound. For evaluation
e p
purposes, tw
wo series off
t
test problem ms, small- and largee-sized probblems, are generated
u
under differrent producction scenarrios. The empirical
e reesults show
t
that the avverage diffeerences betw ween lowerr bound annd optimal
s
solution as well as low wer bound and
a heuristicc method arre equal to
2
2.56% andd 5.23%, rrespectivelyy. For more investiggation, the
p
proposed h
heuristic meethod is co ompared with
w other well-known
w
h
heuristics inn the literature. The results
r verify
fy the efficieency of the
p
proposed h
heuristic m
method in terms of the numbeer of best
s
solutions.
© 2017 IUST
T Publication,, IJIEPR. Voll. 28, No. 1, All
A Rights Resserved
1. In
ntroduction
n1 machines, there exist one or moree unrelated
m
In the classical flow shhop problem
m, a set of joobs paarallel machiines in each stage [3]. Inn scheduling
must be pprocessed onn a number of sequential litterature, thiss problem is recognized asa a flexible
machines, processing routes of alll jobs are thet flow shop (FF FS), flexible flow line (F
FFL), hybrid
same [2]; but in the hybrid flow w-shop (HF FS) flow line (H HFL), or flo ow shop with multiple
scheduling problem with unrelated parallel prrocessors (FSMP). The hybridh flow shops exist
inn many real--world manuufacturing ennvironments
* suuch as seemiconductorr manufactturing [4],
Correspondin
ng author: Ebra
ahim Asadi Ganggraj auutomobile assembly plant p [5], packaging
Email: e.asadi@
@nit.ac.ir inndustry [6], steel manuufacturing [7], [ printed
Received 27 FFebruary 2017; revised
r 15 April 2017; accepted 11
June 2017 ciircuit-board assembly [8], and mettal forming
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1
A Heuristic Approoach to So olve Hybrid Flow-Shop p
62 Ebraahim Asadi Gangraj
G
Scheduliing Problem with
w Unrelateed Parallel Maachines
[9]. Somee authors addressed this classiccal onn a FFS witth k stages so s that theree are one or
problem in their studiess [10-18]. more identical machiness at any stage.
m s They
In the HFS S with unrelaated parallel machines, the t prroposed threee-step TOC--based heuristic for this
job processing times in each stagee are differeent prroblem. With h respect to the absolutee makespan
and dependdent on the machine
m typee. This may be vaalue, they co ompared the proposed heeuristic with
due to thee differencees between the machinnes shhifting bottlleneck heurristic, hybrrid shifting
themselvess, to the fact that each tyype of machiine boottleneck-loccal search heeuristic [26]. Chen and
is better suited
s for a particular job, whereeas C
Chen [27] considered the totall tardiness
others are not, or beecause the job j has som me m
minimization on the FFS problem witth unrelated
particular ccharacteristiccs and it muust be assignned paarallel mach hines in eaach stage and a with a
to machinees that are phhysically nearr to it [3]. boottleneck stage in the production line. Two
In the last decade, therre are some comprehensi
c ive boottleneck-bassed heuristiccs with threee machine
reviews onn hybrid floww-shop scheduuling problem m. seelection rulees were prop posed for thhe research
Vignier, Biillaut, and Proust
P [19] accomplished
a da prroblem.
review of tthis problem. After that, Kis and Pessch Because of NP P-completen ness of the HFS problem
[20] reviewwed the exacct methods for f the k-staage inn a strong sense, it is difficult
d to evaluate
e the
HFS probleem with iden ntical paralleel machines to peerformance of the suub-optimal procedures.
minimize m makespan or total flow tim me. Quadt annd Soome researchers derivedd lower bouunds for the
Kuhnt[16] also propossed taxonom my for k-staage H scheduling problem. Haouari and M’Hallah
HFS
HFS schedduling proced dures focusinng on heuristtic [228] presentedd a new low wer bound foor makespan
procedures. Finally, Ribas
R et al. [3] classifiied att two-stagee flexible flow shopp. Santos,
papers in the HFS environment
e from varioous H
Hunsucker annd Deal [29] developeed a lower
aspects, succh as the diff
fferent constrraints, machiine boound for mu ultiple identical FFS prooblem. The
features, solution apprroaches, andd optimization prrocedure for developing a global bouund involves
criteria. deetermining a lower bound for each e stage.
The bottlenneck phenom mena occur frequently in Sooewandi andd Elmaghrby [30] propposed some
some prodduction systeems. In TO OC (theory of loower boundss for the thrree-stage FF FS problem
constraints)) productionn philosophyy, Goldratt annd w
with identicaal parallel machines
m baased on an
Cox [21] stated the idea that the t bottleneeck auuxiliary problem from the originaal problem.
resources govern thee system’s performancce. K
Kurz and Ask kin [1] imprroved the loower bound
Drum-Bufffer-Rope (DB BR) proposeed by Goldratt prroposed by Leon
L and Raamamoorthy [31] for the
and Fox [222] is a popuular schedulinng approach in H problem with identical parallel machines.
HFS m
bottleneck--based envirronment. DB BR focuses on Inn this researcch, we consider the HFS S scheduling
scheduling bottleneck workstations
w that affect the
t prroblem with unrelated parallel machhines. In the
upstream and downnstream workstations by b laast decades, many researchers connsidered the
scheduling on the bottleneck. Therefore,T the
t H
HFS problemm with identtical machinnes in each
Bottleneck-based sch heduling approach
a h
has sttage and thhe HFS with w unrelateed parallel
attracted soome researchhers in the lasst decades. m
machines waas rarely sttudied. Therrefore, this
Acero-Dom minguez and d Paternina-A Arboleda [223] paaper addresses a new heuuristic algoriithm for the
proposed a bottleneckk-based algoorithm for the t H
HFS scheduliing problem with unrelaated parallel
HFS schedduling probllem with reelated parallel m
machines based on TOC production philosophy.
machines aand makespaan objective function.
f Thhey A
Also, a new lo ower bound was proposeed based on
proposed a three-step approach: 1) bottleneeck thhe lower bouund of Kurzz and Askin [1] for the
identification; 2) compputation of reelease and tail reesearch pro oblem to evaluate
e thee heuristic
times of eeach job for the bottlenneck stage annd allgorithm.
scheduling of jobs at the bottlenneck stage; 3) Thhe rest of th he paper is organized
o as follows: in
scheduling jobs on non-bottleneeck resourcces seection 2, a mathematicaal model foor the HFS
based on sccheduling onn the bottlenneck. Chen annd sccheduling problem
p wiith unrelateed parallel
Chen [24] pproposed a three-step
t boottleneck-bassed m
machines is described.. A bottleeneck-based
heuristic too solve a fleexible flow-liine schedulinng heeuristic algoorithm is proposed
p in section 3.
problem w with a botttleneck staage, with the t Seection 4 is dedicated
d to present a loower bound
objective oof minimizing g the makesppan. foor the reseaarch probleem. In secttion 5, the
Paternina-A Arboleda Montoya-To
M orres, Acerro- exxperimental study is preesented to evaluate
e the
Dominguezz, and Herrrera-Hernanddez [25] dealt loower bound anda the propo osed heuristiic according
with the prroblem of thhe makespann minimization too some expeerimental facctors. Finallyy, the main
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1
A Heuristic Approach to Solve Hybridd Flow-Shop Scheduling
S
Ebrahim Assadi Gangrajj 63
Problem witth Unrelated Parallel
P Machhines
findings off this paper as well as suggestions
s f
for Decision variiables:
D
the future rresearch areaas are addresssed in section : Completio on time of joob j in stage I;
I
6. : Start tim
me of job j in stage I;
2
2. Moddel Developm ment 1
This sectioon is devotedd to presentt a new mixxed 0
integer lineear programm ming (MILP)) model for the
t
HFS with unrelated parallel
p macchines at eaach 1
stage. At fifirst, we define the necesssary notatioons 0
to present tthe MILP moodel. Seet H is defi fined to omiit unnecessaary decision
Set and ind dex: vaariables and d constraintss in the mathematical m
, : Job inddex; m
model expresssion. If the decision vaariables are
, : Stage iindex; deefined regard ding the rannge of their indices, the
: Machinees index in each stage; nuumber of decision variabbles increasees to a huge
: Numberr of jobs; nuumber and the t efficienccy of the mathematical m
: Numberr of stages; m
model will decrease drastiically [32]. For F instance,
: Numbeer of machinees in stage I;; is equal to o 1, the valuue of is 0, and vice
: Set of joob indices neecessary to define
d decision veersa.
variables ( , | ). 2--1. Model foormulation
W respect to the notatiions summarrized in the
With
Input and parameterss:
peervious section, the HF FS mathematical model
: A large number;
w unrelatedd parallel maachines can be modeled
with
: Proccessing timee of job j in stage i on
ass MILP as foollows:
machine k;
(11)
subject to:
1,2, … , (22)
.
1,2, … , (44)
0
1,2, … , (88)
∈ 0,1
1
1,2, … , & 1,2, … ,
, 1,2, … , & , ∈ (99)
∈ 0,1
1,2, … ,
The objecttive functionn, as presentted in Eq. (1), shhows the relation betweeen starting tiime of each
minimizes makespan, and consttraint set ((2) joob in the laast stage annd makespann objective
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1
A Heuristic Approoach to So olve Hybrid Flow-Shop p
64 Ebraahim Asadi Gangraj
G
Scheduliing Problem with
w Unrelateed Parallel Maachines
function. CConstraint seet (3) shows that each job thhe overall sy ystem perfo ormance" (G Goldratt and
must be asssigned to onne machine in each stagge. Foox, 2004). Performancee enhancem ment of the
Constraint set (4) dem monstrates thhat each job is boottleneck stage
s can improve thhe overall
available at the beg ginning of the planning peerformance; in other words, fi finding the
horizon. Thhe relationshhip between starting tim mes boottleneck staage and expploiting it caan optimize
for job j in two consecuutive stages is stated in the
t thhe whole sy ystem perforrmance. Bassed on this
constraint set (5). Coonstraint sets (6) and (7) ( iddea, the heu uristic algorrithm consissts of four
show the rrelationship between tw wo consecutiive stteps. In the first step off the proposeed heuristic
jobs that aare processed d on one maachine in eaach appproach, the bottleneck stage
s is idenntified based
stage. If thhe two jobs, l and j, aree processed on onn the relativve workloadd in each sttage. In the
one machinne and job l is processedd before jobb j, seecond step, sequence of the joobs at the
constraint ((6) is activatted to prevennt interferennce boottleneck staage is deteermined based on two
between thhe processing g operationss of these tw wo prroposed paraameters: the first one caalculates the
jobs; on thhe other sid de, the roles of these tw wo neecessary timme for takingg the job too bottleneck
constraint sets are changed. Finally, booth sttage and the second one estimates thhe minimum
constraints (8), (9) forcce variables X and Z to timme to pass from
f the botttleneck stagee to the last
assume binnary values 0 or 1. sttage; the jobbs are arrannged by thee ascending
orrder of these parameters. At the thirdd step, it can
3. Th
he Proposeed Heuristic Algorithm
m tryy to sequen nce the jobss in the nonn-bottleneck
This sectioon is dedicateed to presennt the propossed sttages: upstreaam and dow wnstream stagges. For this
bottleneck--based heuristic for the HFS problem m. puurpose and ini the upstreeam stages, tthe jobs are
With respeect to unrelated parallel machines in arrranged baseed on the coompletion tiimes in the
each stagee, a machin ne selection rule will be prrevious stagge and start time in thee bottleneck
applied to ddetermine thhe assignmennt of the jobs in sttage. Also, for
f downstreeam stages, the t jobs are
each stagee. This algoorithm is baased on TO OC seequenced baased on an ascending order o of the
philosophyy and a local search is innserted into the
t coompletion times in thhe previous stage and
structure off the proposeed heuristic to
t improve the
t esstimated tim me to pass the produuction line.
overall perfformance. Fiinally, after determiningg the initial sequence, a
3-1. Assum
mption loocal search approach
a is applied to improve the
The hybridd flow shop problem,
p connsidered in thhis quuality of the generated solutions. Thhe algorithm
research, aassumes thaat there are d stages annd iss now describbed in more details
d as folllows:
includes a bottlenecck stage B. B There are a Sttep 1. Findin ng the bottleeneck stage
M unrelateed parallel machines
m in stage i. Theere For each
e stage i, compute flow ratio
are n jobs wwith the sam
me routing andd must visit all ∑ ∑
FR
R ,i 1,2, … , m. Thee stage with
the stages consecutivelly. The proccessing time of
a job in a stage is knnown and deepends on the t thhe largest FRR is defined as the bottleeneck stage
machine assigned to the t stage. A machine can c annd is denoted d by B.
process onlly one job att a time and a job must be Sttep 2. Sequeencing in bottleneck stage B
processed on one maachine in anny stage. Job For each job j, comppute R
preemptionn is not alllowed, andd there is no n ∑ min d , k 1,2, … , M , j 1,2, 1 …,n .
machine brreakdown annd no set-upp time requirred R denotes thee minimum necessary
n tim
me for job j
before the jobs
j are processed on anny machine. In too get to the bottleneck staage.
this researcch, the workkload is deterrmined in eaach For each job j, comppute D
stage to identify thee bottleneckk stage. The T
∑ FR R , k 1,2, … , M , j 1,2, … , n .
workload ofo stage i is computed
c by the sum of the
t
processing time at a particular
p staage divided by
b D denotes thee estimated time
t needed for job j to
its numberr of machiines, denoteed as FR paass through from bottleeneck stage to the last
∑ ∑ sttage.
ge with the largest FR is
. The stag Arrannge jobs by inncreasing ordder of R . If
defined as tthe bottlenecck stage. thhere is more than one job b with the same R , rank
3-2. Heurristic algoriithm thhem in increasing order of D . If thhere is more
This algorrithm is based
b on thhe theory of thhan one job with
w the sam me D , select jjob with the
constraints (TOC) man nufacturing philosophy
p th
hat laargest processing time in the bottlenecck stage.
state the iddea that "boottleneck ressources goveern Sttep 3. Sequeencing in non-bottleneck k stages
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1
A Heuristic Approach to Solve Hybridd Flow-Shop Scheduling
S
Ebrahim Assadi Gangrajj 65
Problem witth Unrelated Parallel
P Machhines
Uppstream stagees: V If the currrent partiall schedule inncludes the
V.
Forr the first stage,
s i 1 , compute for f entire n jo
obs, then stopp; otherwisee, go to Step
each job j, d min d , k 1,2, … , s . II.
Arrange joobs by increeasing the order o of d . If 3--3. Machinee selection ru ule
Inn order to co onsider unreelated paralleel machines
there is m more than on ne job with the same p ,
inn each stage, we must apply
a machinne selection
select a jobb with the larrgest processsing time at the t ruule to assig gn each job to machinnes. In this
bottleneck stage. reesearch, for a given job, the
t job is asssigned to all
Forr upstream m stage i, i 2, … , B , thhe machines (available anda unavailabble) in each
compute C , and ST S . Arraange jobs by b sttage and seleects a machiine that has the earliest
increasing the order of o C , . If there is moore coompletion tiime. This proposed
p sellection rule
than one joob with the saame C , , raank them in an m cause idle period of thhe job. How
may wever, due to
increasing order of ST . If there is more than one o unnrelated parallel machinnes, an unavvailable but
job with thhe same ST , select job with w the largeest m
more efficien
nt machine may
m producee an earlier
processing time in the bottleneck
b sttage. coompletion tim me for a givven job; in other
o words,
Doownstream sttages: thhis rule may prefer an unnavailable machine
m with
Forr downstream m stage h, compute
c C , a short processsing time to an availabble machine
w a long prrocessing tim
with me.
and D ∑ FR R , k 1,2, … , s , j
1,2, … , n . Arrange jobbs by increaasing order of 4
4. Low
wer Bound
C , . If there is moore than onee job with the t Due to the NP
D P-completeneess of the HF FS problem
same C , , rank them m in increasinng order of D . w unrelatedd parallel maachines in eaach stage, it
with
If there is more than one o job withh the same D , iss very hard and time-coonsuming too obtain the
select job w with the larggest processiing time at the t opptimum solu ution for meedium- and large-sized
bottleneck stage. prroblems. Onn the other haand, we needd criteria to
Step 4. Loccal search evvaluate the proposed heuristic. Theerefore, we
I. Sellect the first job in the innitial sequennce w develop a lower bounnd to comparee the results
will
and let iit be the currrent partial seequence. off the proposeed heuristic with a loweer bound, so
II. Sellect the nextt job in the innitial sequennce thhat we can assess
a the quality
q of thee solutions.
and insert the job into the positions beforre, Thhis lower bo ound is based on Kurz and Askin
betweenn, and after every
e two coonsecutive joobs (22001) lower bound which is developped for HFS
of the cuurrent partiall sequence. prroblem with identical parrallel machinnes. For this
III. Calculaate makespann for each paartial sequennce puurpose, two changes havve been impllemented in
producced in Step II while adjusting a jobbs' thhe structure of the loweer bound; att first, with
enterinng sequence at the bottleeneck stage to reespect to unnrelated paraallel machinnes at each
be the same
s as that at the first stage. sttage, we rewwrite the low
wer bound baased in this
IV. Select the partial sequence with w minimuum neew assumption. Subseqquently, a neew term is
makesppan and let the t partial seequence be the t appplied in the structure of the loweer bound to
currentt partial sequuence. immprove its quality.
q The lower boundd, proposed
byy Kurz and Askin (22001), is as a follows:
1
maax min min min min (10)
, ,…,
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1
A Heuristic Approoach to So olve Hybrid Flow-Shop p
66 Ebraahim Asadi Gangraj
G
Scheduliing Problem with
w Unrelateed Parallel Maachines
, ,…, ∑ ∑ ∑ ∑
∑ ∑ ∑ ∑ (12)
teest problemss in this secction are lim
mited to the
5. Comp
putational Study
S smmall-sized problems.
p Taable 1 summ marizes the
This secttion is deevoted to evaluate the t exxperimental factors used u to define
d the
performancce of the loower bound and proposed prroduction scenarios fo or the 20 small-sized
heuristic. FFor evaluatio
on purposes,, two series of prroblems.
test probleems are generated
g unnder differeent
production scenarios. The
T first seriees is dedicatted Tab. 1. Experimenntal factor for small
to small-sizzed problemms and the seecond series is size problems
devoted to medium- and large-sizedd problems. Experim
mental factor Feature
5-1. Comp parison of th
he proposed d lower boun nd Numbber of jobs U[3,5]
with the optimal solution
s forr small-sized Numbeer of stages U[2,4]
problems N
Number of machines
m in each
e U[1,3]
Firstly, too examine the perform mance of the t s
stage
proposed loower bound,, this sectionn compares the t Processsing time U
U[5,10]
lower bounnd with the optimal
o soluttion. We try to
find the opptimal valuee of the objeective function Regarding Taable 1, four experimental
R e l factors are
for test prooblems using the proposed MILP in thhis ussed to define small-sized problems: number of
research. L Lingo (versioon 16) is appplied to sollve joobs, number of stages, number
n of machines
m in
the probleems, optim mally. Due to the NP- N eaach stage, an nd processinng time. Thee results of
completeneess of the HFS
H schedulinng problem, it thhe comparisoon between loower bound and optimal
is very exppensive to achhieve the opttimal solutioons soolution are deescribed in Table
T 2:
to the mediium- and larrge-sized prooblems. Hencce,
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1
A Heuristic Approach to Solve Hybridd Flow-Shop Scheduling
S
Ebrahim Assadi Gangrajj 67
Problem witth Unrelated Parallel
P Machhines
14 4 4 3122 500 52 4.00%
15 5 2 31 533 53 0.00%
16 3 3 322 233 25 8.70%
17 4 2 22 21 22 4.76%
18 4 2 12 299 29 0.00%
19 3 3 233 25 26 4.00%
20 4 4 1122 500 50 0.00%
allgorithm and d lower boound, respecctively. A
According to Table 2, the proposedd lower bounnd seeries of comp putational exxperiments arre produced
can obtain the optimall solutions in i 53% of teest too evaluate the proposeed heuristic based on
problems aand solver cannot attain the optim mal diifferent prodduction scennarios. These scenarios
solution affter 10 minuutes in three--test problem
ms. coonsist of fivve experimenntal factors: number of
The averagge differencee between low wer bound annd joobs, number of stages, numbern of machines
m in
optimal soolution (optiimal gap) equals
e 2.56%%. eaach stage, variation
v off processing time, and
Furthermorre, in 14 (ouut of 17) testt problems, the
t boottleneck po osition. Tabble 3 summ marizes the
optimal gapp is lower thhan 5%. Theerefore, we can
c exxperimental factors usedd to define production
conclude thhat the lowerr bound has good
g quality to sccenarios. Thhe number of o jobs has three
t levels
achieve thee optimal solutions. w values seet at 10, 20,, and 50 (low
with w, medium,
5-2. Comp parison of the propoosed heuristtic annd high). Thhe number of o stages also has three
with lowerr bound for medium- an nd large-sized leevels with values
v set at
a 3, 6, andd 12 (low,
problems m
medium andd high), where
w the number
n of
In order tto evaluate the perform mance of thet m
machines in each
e stage is generated frrom discrete
proposed hheuristic meethod, relative percentaage unniform distriibution in thee range of [22, 4] and [4,
deviation (RPD) from thet lower boound is appliied 6]] (low and high).
h Processsing time haas one level
as follows: inn which the non-bottlene
n eck stages haave uniform
C C diistribution inn the range of
o [5, 10] andd bottleneck
RPD 100
0 (13) sttage has unifform distribuution in the raange of [11,
C
155]. Finally, the bottleneeck is locateed in three
In this eqquation, and are the
t pllaces, first 1//3, second 1//3 and third 1/3.
1
makespans obtained by
b the proposed heuristtic
Regarding the fivee experimeental facto ors boound. As mentioned above, thee objective
considered at Table 3, there are 54 production fuunction for all
a the test problems
p is to
t minimize
scenarios aand 10 test problems
p aree generated for
f thhe makespan n. The experrimental results of 540
each scenaario. These test probleems (540 teest teest problemss are providded in the appendix
a in
problems) are used to compare the performannce taable 6.
of the proposed heuristiic with that of
o the lower
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1
A Heuristic Approoach to So olve Hybrid Flow-Shop p
68 Ebraahim Asadi Gangraj
G
Scheduliing Problem with
w Unrelateed Parallel Maachines
Table 4 shhows the aveerage of relaative deviation taable 4, in all the in nstances, thee proposed
from the lower boun nd based oon some teest heeuristic, in average,
a hass 5.4% diffeerence from
problems. These results are pressented for the t thhe lower bounnd.
various expperimental faactors. As it can be seen in
Tab. 4. Average of
o relative deviation
d frrom lower bound
b (in percent)
p
Num
mber of stagee
L H
Jobs Bottlenneck positionn
3 6 12 3 6 12
First
F 1/3 3.81
3 2.33 4.15 13.33 12.17 8.66
10 Seecond 1/3 3.34
3 3.668 4.56 14.15 12.45 8.84
T
Third 1/3 3.91
3 4.15 4.66 16.43 12.94 10.21
First
F 1/3 2.32
2 3.003 3.32 6.04 5.39 4.55
20 Seecond 1/3 2.48
2 3.12 3.18 7.16 6.73 5.83
T
Third 1/3 5.25
5 3.223 4.02 6.92 7 6.98
First
F 1/3 1.54
1 1.66 1.97 3.69 4.18 4.14
50 Seecond 1/3 1.48
1 1.52 1.71 3.7 4.45 4.37
T
Third 1/3 1.64
1 1.89 1.87 3.32 4.23 5.11
With respeect to Table 4, 4 the averagge RPD for thet eaach stage, we cannot identify any a distinct
entire tesst problemss is equall to 5.23% %. paattern.
Furthermorre, the com mputational results sho ow Fuurthermore, we applied Figs.
F 1-3, whhich present
several notteworthy poiints. When the t number of thhe RPD trend ds for the number of jobss to confirm
jobs increaases, the RPD D significantlly decreases in thhe pervious conclusions.
c In all the fiigures, each
any experimental facto ors. Also, ass can be seeen, pllot name coonsists of th hree sectionns: the first
bottleneck position do oes not affecct the RPD in seection presennts the numbber of stages,, the second
entire scennarios. Wheen the num mber of stagges seection denotees the numb ber of machines in each
increases from L to H, the RPD R increasses sttage, and thee third term identifies
i thee number of
significantlly. But, in the number off machines inn joobs.
jobs=10
18
16
14
3L10
12
relative diviation
6L10
10
12L10
8
3H10
6 6H10
4 12H10
2
0
third 1/3
t second 1/3 first 1/3
Fig.
F 1. Relattive deviatiion from loower bound
d for jobs=110
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1
A Heuristic Approach to Solve Hybridd Flow-Shop Scheduling
S
Ebrahim Assadi Gangrajj 69
Problem witth Unrelated Parallel
P Machhines
jobs=20
8
6
3L20
relative diviation
5 6L20
4 12L20
3H20
3
6H20
2 12H20
0
third 1/3 second 1/3 first 1/3
Fig.
F 2. Relattive deviatiion from loower bound
d for jobs=220
jobs=50
6
3L50
4
relative deviation
6L50
3 12L50
3H50
2 6H50
12H50
1
0
third
d 1/3 second 1//3 first 1/3
Fig.
F 3. Relattive deviatiion from loower bound
d for jobs=550
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1
A Heuristic Approoach to So olve Hybrid Flow-Shop p
70 Ebraahim Asadi Gangraj
G
Scheduliing Problem with
w Unrelateed Parallel Maachines
6. Conclusion
C n
This reseaarch conduccted the prroposed low wer
bound which can achieeve the optim mal solution in
many instaances so thaat the solverr cannot attaain
the optimaal solution inn acceptable time in som me
instances, the
t lower boound gives a good solutioon.
The lower bound has 2.56%2 differrence from the
t
lower bounnd in small instances.
i Thherefore, it can
c
be used to evaluate oth her heuristicc performancce.
The heuristtic algorithm
m is proposedd in this articcle
for the HFFS schedulingg problems with unrelatted
parallel maachines, whiich is a goodd heuristic forf
C objecctive functioon, such thhat it has, on
average, 5.4% deviatio on from the lower bounnd.
The resultts show thaat the bottleeneck-heuristtic
algorithm outperformss the otherr well-know wn
heuristics ssuch as CDS, Slope, and NEH.
Considerinng other ch haracteristics of the HF FS
scheduling problemss, such as a sequencce-
dependent set up tiime, reentraant jobs annd
identical machines
m caan be openned for futuure
studies. IIn addition, new boottleneck-bassed
approachess can be stuudied for othher schedulinng
problems w with bottleneck stage.
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1
A Heurristic Approach too Solve Hybrid Flo
ow-Shop Schedulin
ng
Ebrahim Assadi Gangraj 7
71
Problem
m with Unrelated Parallel Machinees
Appendix
Tab. 6. Ex
xperimental ressults of the testt problems
N.S. 3 6 12
2
Job N..M. B.P. Firsst 1/3 Second 1/3 Third 1/3 First 1/3 Second 1/3 Third 1/3
T Firrst 1/3 Secon
nd 1/3 Third 1/3
1
T.P. Cmax x LB Cmax LB Cmax LB B Cmax LB Cmax LB Cm
max LB Cmax x LB Cmax LB Cmax LB
1 70 67 76 74 76 766 142 139 142 137 143 134 262 244 248 239 268 260
2 77 70 69 69 71 688 134 133 132 123 142 135 248 238 262 251 250 244
3 81 79 79 77 85 822 141 134 151 145 135 127 250 242 256 250 262 242
4 78 78 75 72 68 633 149 143 145 142 144 138 280 264 262 248 284 268
5 75 71 74 69 80 788 144 141 143 141 147 137 252 246 290 281 262 254
L
6 69 66 68 64 72 711 152 147 135 130 140 131 266 260 270 253 274 260
7 79 74 78 73 71 700 143 142 146 142 154 150 258 250 276 258 258 244
8 75 75 82 80 80 733 146 143 151 145 144 148 258 244 260 249 254 242
9 77 74 77 77 77 733 139 137 140 134 136 129 284 270 254 242 270 254
10 71 71 70 69 84 822 141 139 141 137 140 134 264 260 296 284 286 282
10
11 56 50 68 61 60 499 107 98 107 100 108 93 212 199 209 189 210 189
12 66 58 56 52 53 477 115 104 117 110 100 85 198 175 221 207 221 204
13 54 49 59 50 66 611 98 90 104 87 121 103 211 198 197 179 212 200
14 60 51 63 55 60 499 103 94 103 89 96
9 91 222 202 203 184 199 182
15 57 51 55 48 52 444 109 92 99 87 109 101 206 193 214 195 208 191
H
16 52 45 62 50 67 600 101 91 102 92 102 89 209 189 207 193 214 201
17 59 52 50 43 57 488 112 104 114 98 111 95 215 203 220 201 225 203
18 63 54 59 51 64 600 104 88 119 104 118 102 201 183 211 197 220 193
19 67 61 57 55 57 466 105 91 96 89 97
9 89 208 191 219 201 221 197
20 57 51 61 53 50 422 95 84 106 94 109 100 222 204 203 188 206 179
21 138 131 139 133 151 1444 289 281 310 300 322 315 399 390 408 392 399 383
22 125 122 148 146 141 1355 276 266 316 311 306 295 410 395 397 388 412 401
23 148 145 151 150 138 1300 287 280 307 299 318 311 414 403 401 390 415 395
24 139 137 139 139 125 1177 297 289 301 293 319 305 393 385 389 376 391 373
20 L 25 134 132 135 131 134 1266 281 276 299 288 301 289 399 388 411 399 412 403
26 125 123 125 122 136 1277 292 284 307 298 313 305 413 402 423 410 421 405
27 125 122 137 132 151 1444 280 266 314 301 304 296 425 404 413 395 400 380
28 137 135 125 120 148 1422 274 265 306 295 316 304 406 390 424 412 413 400
29 145 143 148 144 139 1322 276 269 301 292 311 300 394 381 407 396 395 378
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1
A Heuristic Approoach to So olve Hybrid Flow-Shop p
74 Ebraahim Asadi Gangraj
G
Scheduliing Problem with
w Unrelateed Parallel Maachines
[17] Nahavvandi, N. an nd Asadi Gangraj,
G E. “A
“ [225] Paterninaa-Arboleda, C., Montoyaa-Torres, J.,
New lower
l boundd for Flexibble Flow Shop Acero-Doominguez, M. and Herrera-
Probleem with Unrrelated Paralllel Machiness”, Hernandeez, M. “Sch heduling jobbs on a k -
Internaational Journal off Industrial stage flex
xible flow-shop” Annalss Operation
Engineeering & Production
P R
Research, Vool. Researchh, Vol. 164, No. 1, (20007), pp. 29–
25, Noo. 1, (2014), pp. 65-70. 40.
International Journal of In
ndustrial Engineeering & Producction Research, March
M 2017, Vo
ol. 28, No. 1