Proceedings

The of
of the
International
Proceedings 20th
20th World
Federation
the Worldof Congress
Automatic Control
Congress
The International
Proceedings
Toulouse,
The France,
International Federation
of the 20th9-14,
July World
Federation of Automatic
of Congress
2017
Automatic Control
Control
Toulouse, France, July
Available online at www.sciencedirect.com
The International
Toulouse, July 9-14,
9-14, 2017
France,Federation of Automatic Control
2017
Toulouse, France, July 9-14, 2017
ScienceDirect
IFAC PapersOnLine 50-1 (2017) 2255–2259
Some new ideas for assembly line balancing
Some new ideas for assembly line balancing
Some new ideas for research
assembly line balancing
research
research
Alexandre Dolgui ∗ Evgeny Gafarov ∗∗
∗
∗∗
Alexandre
Alexandre Dolgui Dolgui ∗∗ Evgeny
Evgeny GafarovGafarov ∗∗ ∗∗
∗ Alexandre Dolgui Evgeny Gafarov
∗
IMT Atlantique, LS2N - UMR CNRS 6004, La Chantrerie, 4, rue
∗ IMT Atlantique, LS2N - UMR CNRS 6004, La Chantrerie, 4, rue
∗ IMTKastler
Alfred Atlantique, - B.P. LS2N20722,- UMRF-44307CNRS 6004,cedex
Nantes La Chantrerie,
3, France (email:4, rue
IMTKastler
Alfred
Alfred Atlantique,
Kastler -- B.P.
B.P. LS2N20722,
20722,- UMRF-44307
F-44307CNRS 6004,cedex
Nantes
Nantes
alexandre.dolgui@imt-atlantique.fr) La Chantrerie,
cedex 3,
3, France 4, rue
France (email:
(email:
Alfred
∗∗ Kastler
V.A. Trapeznikov - B.P. 20722, F-44307 Nantes
alexandre.dolgui@imt-atlantique.fr)
alexandre.dolgui@imt-atlantique.fr) cedex 3,
Institute of Control Sciences of the Russian France (email:
∗∗
∗∗ V.A. Trapeznikov alexandre.dolgui@imt-atlantique.fr)
Institute
Institute of of Control
st. 65,Sciences
117997 of the
the Russian
∗∗ V.A. of
Academy Trapeznikov
Sciences, Profsoyuznaya Control Sciences of
Moscow, Russian
Russia
Academy
Academy V.A. of Trapeznikov
of Sciences,
Sciences, (email: Institute
Profsoyuznaya
Profsoyuznaya of Control
st. 65,Sciences
st. 65,
eugenemath@gmail.com) 117997 of
117997 the Russian
Moscow,
Moscow, Russia
Russia
Academy of Sciences, (email:
(email: Profsoyuznaya st. 65, 117997 Moscow, Russia
eugenemath@gmail.com)
eugenemath@gmail.com)
(email: eugenemath@gmail.com)
Abstract: The design of assembly lines is an important issue in manufacturing engineering,
Abstract:
Abstract: The
management anddesign
The design
control. of assembly
of The
assemblyidle time lines
linesis is
isthean important
anmost
important issue in
in manufacturing
issueperformance
interesting manufacturingindex for engineering,
engineering,
assembly
Abstract:
management
management
line design. TheThe
and design
control.
and classical of
control. The assembly
The
simple idle lines
time is
idleline-balancingis
thean
time is the most important
most issue
interesting
interesting
problem (SALBP) in manufacturing
performance
performance index
consists index engineering,
for
for
of assigning assembly
assembly
tasks,
management
line
line design.
design.
necessary and
forThe
The control.a The
classical
classical
processing simple
simple
product, idleline-balancing
time is the most
line-balancing
to workstations interesting
problem
problem
such (SALBP)
(SALBP)
that theperformance
idleconsists
consists index
of
time (number for
of assigningof assembly
assigning tasks,
tasks,
stations,
line design.
necessary
necessary
cycle time,for The
forcost) classical
processing
processing a simple
a
is minimized product,
product, line-balancing
while to workstations
toprecedence
workstations problem
such
such that
constraints (SALBP)
that the
the idle
idle
between consists
time of assigning
tasks(number
time (number
are of tasks,
of stations,
satisfied. stations,
From
necessary
cycle time,
cycleworst-case
the for processing
cost)
time, cost)analysisis a
minimized
is minimized product,
point while while to
of view, workstations
precedence
precedence
the SALBP such
constraints
constraints that
problems the idle
between
between time
tasks (number
are of
satisfied.
tasks areinsatisfied.
are NP-hard stations,
strong sense.From
From
cycle
the time,
worst-case cost)
the worst-casein analysis
Nevertheless, is minimized
analysis
practice,pointpoint
it is alsowhile
of view,
of view, precedence
the SALBP
the SALBP
important constraints
to be able problems
problems between
to compareare tasks
NP-hard
are NP-hard arein satisfied.
strong
in strong
real instances From
sense.
sense.
of SALBP.
the worst-case
Nevertheless,
paper,in
Nevertheless,
In this analysis
inthe simplepoint
practice,
practice, it
it is
assembly of view,
is also
also the SALBP
important
important
line balancing to
to be problems
able
beproblem
able to
to of are 1NP-hard
compare
compare
type real
real in strong
instances
instances
(SALBP-1) is of
of sense.
SALBP.
SALBP.
considered
Nevertheless,
In
In this
where paper,
this the
paper, inthe
cycle thepractice,
simple
simple
time it is and
assembly
assembly
is fixed also theimportant
line balancing
line objective
balancingto be able
isproblem
problem
to to of
compare
of
minimize type
type real
the11 number instances
(SALBP-1)
(SALBP-1) is of
is SALBP.
considered
considered
of stations. Two
In
wherethis paper,
the cycle
where the cycle ways
unconventional the simple
time
time are is assembly
fixed
is fixedproposedand
and the line
the balancing
objective
to objective isproblem
to of
minimize
is to minimize
help to estimate type the 1 (SALBP-1)
number
the number
the complexity of
of of
such is considered
stations.
stations.
a problem Two
Two
where the
unconventional
unconventional
instances: cycle
reduction time
ways
waysofare is
are fixed
proposed
theproposed and
graph of to theto objective
help to is to
estimate
help to constraints
precedence minimize
estimate the the the
complexity
to complexity number
planar one and of of
such
of such stations.
a problem
a problem
transformation Two
unconventional
instances:
instances:
of the problem reduction
reductionways
to ofare
of
a problemtheproposed
the graph
graph of to
of help to
precedence
precedence
of maximization. estimate
Weconstraints
constraints
show how the
to
to complexity
planar
planar
these one of can
one and
techniques and such bea employed
problem
transformation
transformation
instances:
of
of the
and thewhy reduction
problem
problem
they to a
to useful
are of the
a problem
problem graph
of
to analysis ofofprecedence
of maximization. assemblyWe
maximization. We constraints
lineshow
show how to
how
balancing planar
these
these one
techniques
techniques
problem andcan
can
instances. transformation
be
be employed
employed
of
and why they are useful to analysis of assembly line balancing problem instances. be employed
and thewhyproblem
they to
are a problem
useful to of
analysismaximization.
of assembly We lineshow how
balancing these techniques
problem can
instances.
© 2017,
and whyIFACthey(International
are useful toFederationanalysis of ofAutomatic
assembly Control) Hosting problem
line balancing by Elsevierinstances.
Ltd. All rights reserved.
Keywords: Modeling of assembly units, Modeling of manufacturing operations, Job and
Keywords:
Keywords:
activity Modeling
ModelingAssembly
scheduling, of assembly
of assembly units,
lineunits, Modeling
Modeling
balancing, of manufacturing
of manufacturing
Algorithms, operations,
Graphs operations, Job Job andand
Keywords:
activity Modeling
scheduling, of assembly
Assembly line units, Modeling
balancing,
activity scheduling, Assembly line balancing, Algorithms, Graphs of
Algorithms,manufacturing
Graphs operations, Job and
activity scheduling, Assembly line balancing, Algorithms, Graphs
1. INTRODUCTION Battaia and Dolgui (2013); Hazir et al. (2015). Different
1. INTRODUCTION
1. INTRODUCTION Battaia and
Battaia
application anddomains
Dolgui (2013);
Dolgui (2013); Hazir et
Hazir
are considered etforal.
al.example
(2015). in
(2015). Different
Different
Dolgui
1. INTRODUCTION Battaia
application
application and
and Ihnatsenka Dolgui
domains
domains
(2009),(2013);
are Hazir
considered
are Borisovsky
considered for et al.
for (2015).
example
example
et al. (2014), Different
in
in Dolgui
Dolgui
Battini
Modern production systems are characterized by short and application
and Ihnatsenka
et al.Ihnatsenka domains
(2016), Hazir(2009),
(2009), are considered
Borisovsky
andBorisovsky
Dolgui (2015), for
et example
al. (2014),
et al. (2014), in Dolgui
Battini
BentahaBattiniet al.
Modern life-cycles,
Modern
product productionhigh
production systems
systems
level of are
are characterized
characterized
automation, by short
by
emergence shortof et and al.Ihnatsenka
et al. (2016), (2009),
Hazir and Borisovsky
Dolgui (2015), et al. (2014),
(2015), BentahaBattiniet al.
al.
Modern production systems are characterized by short (2015), (2016),
and Hazir
Battaia etand
al. Dolgui
(2017). Bentaha et
product
product
new life-cycles, high
life-cycles,
manufacturing high level of
level
equipment ofand
automation,
automation,
technologies, emergence
emergence
and high of et
of al. (2016),
(2015), Haziret
and Battaia
Battaia etand Dolgui (2015), Bentaha et al.
al. (2017).
(2017).
product life-cycles, high level ofand automation, emergence (2015),
of Salveson and al.
new
new manufacturing
manufacturing
investment. equipment
equipment and technologies,
technologies, and high
and high (2015), and (1955) has suggested
Battaia et al. (2017).a linear program to describe
new manufacturing
investment. equipment and technologies, and high Salveson
Salveson
all possible(1955)
(1955) has
taskhas suggested for
suggested
assignments aa linear
linear program line.
program
an assembly to describe
to describe
There
investment. Salveson (1955)
Assembly
investment.is a key production activity, to put it in place, all all no
is possible
possible taskhas
task
constraint on suggested
assignments
assignments a linear
for
for
task splitting anand
an program
assembly
assembly to describe
line.
line.
so it may There
There
gener-
Assembly
Assembly
it is often is isnecessary
aa key
key production
production activity,
to developactivity,assembly to lines.
to put it
put itAssembly
in place,
in place, all is nopossible task
constraint assignments
on task for
splitting an assembly
and so it line.
may There
gener-
Assembly
it
it is often
is
lines often isnecessary
a key production
necessary
are production to develop
to develop
systems activity,
thatassembly
assembly
include to lines.
put
lines.
serially place, is
itAssembly
inlocated
Assembly
atenoinfeasible
is
ate no
constraint
constraint
infeasible
on task Bowman
solutions.
on
solutions.task
splitting (1960)
splitting
Bowman
and so added
and
(1960) so
it maya gener-
it
addedmay a
’non-
gener-
’non-
it is often necessary tooperations
develop assembly lines. Assembly ate infeasible
divisibility’ solutions.
constraint by Bowman
using a (1960)
zero-one added
integer a ’non-
program.
lines
lines are production
are
workstations production systems
in whichsystems that
that includeare
include
(tasks) serially
serially located ate
located
continuously infeasible
divisibility’ solutions.
constraint Bowman
by using
using (1960)
a zero-one
zero-one added
integer a ’non-
program.
lines are production divisibility’
An assembly constraint
line by
balancing a
problem can integer
also be program.
modeled
workstations
workstations
processed. Theyin are systems
in which
which operations
operations
employed that include
(tasks)
in(tasks)
variousare areserially located
continuously
continuously
industries like An divisibility’
with a tree constraint
An assembly
assembly line
where eachbyarc
line balancing
balancing using a zero-one
problem
problem
represents can
can integerbe program.
also be
also
a station modeled
modeled
and each
workstations
processed.
processed.
the automobile, Theyin which
are
They electronics,operations
employed
are employed in
etc. in (tasks)
various
various
where are continuously
industries
the industries
objective islike like
to with An
with assembly
a tree line
where balancing
each arc problem
represents can
a also
stationbe modeled
and each
processed. They areofemployed path a tree where
corresponds toeach
a arc
feasible represents
balancing a station
solution, and
see each
Jack-
the
the automobile,
automobile,
produce large electronics,
electronics,
series the same etc. in
etc. various
where
where
(or the industries
the
similar) objective
objective
products.is islike
to with
to path a tree where
corresponds toeach
a arc
feasible represents
balancing a station
solution, and
see each
Jack-
the automobile, electronics, etc. (or
where the objective path corresponds
son (1956). Dynamic to a programming
feasible balancing solution,
techniques aresee Jack-
used in
produce
produce large series
large series of the
of the same
same (or similar)
similar) products.is to path
products. corresponds
son (1956).
(1956). Dynamic to (1997).
a programming
feasibleThe
programming balancing solution,
techniques aresee Jack-
used in
The
producedesign
large of series
assemblyof thelines
same is an
(orimportant
similar) products. problem in son son
Agnetis and Dynamic
Arbib mosttechniques
recent are
Branch used in
and
The design of assembly lines is an important problem in Agnetis
Agnetis
Bound (1956).
andDynamic
and Arbib
algorithm programming
Arbibfor(1997).
(1997).
SALBP-1 The is
The most
mosttechniques
recent in
recent
presented are
Branch
Branch usedand
Morrison in
and
The design of assembly
manufacturing engineering, linesmanagement
is an important problemsee
and control, in BoundAgnetis and Arbib
algorithm for(1997).
SALBP-1 The most
is recent
presented Branch
in Morrisonand
The design
manufacturing
manufacturing
Askin of assembly
and Standridge engineering,
engineering, linesmanagement
(1993), isDolgui
an important
management and
andand problem
control,
control,
Proth (2010). in Bound
see
see algorithm for SALBP-1 is presented in Morrison
et al. (2014).
manufacturing
Askinbalancing
and Standridge engineering,
Standridge (1993), management
Dolgui and and control,
Proth (2010).see et Bound
et al. algorithm for SALBP-1 is presented in Morrison
al. (2014).
(2014).
Askin
The and of station(1993),
workloads Dolgui is theand most Proth (2010).
interesting
Askin
The and Standridge
The balancing
balancing
performance of station
of
index station (1993),
workloads
workloads
for assembly Dolgui
lineis and
isdesign.
the
the most
most Proth
The (2010). In Wee and
interesting
interesting
classical
In
et Wee
al. (2014). Magazine (1982), heuristics based on bin
and Magazine
Magazine (1982), heuristics based ondedi-
bin
The balancing of station workloads In Wee
packing and
algorithms are (1982),
suggested. heuristics
There are basedalsoon bin
performance
performance
simple index
index
line-balancing forproblem
for assembly
assembly lineisdesign.
line
(SALBP) theconsists
design. most The
The interesting
classical
ofclassical
assign- packingIn Wee
packing and Magazine
algorithms are (1982),
suggested. heuristics
There based
are also on bin
dedi-
performance index for assembly line design. The classical algorithms are suggested. There
cated heuristics, for example, the ranked positional weight are also dedi-
simple
simple
ing tasks line-balancing
line-balancing
necessary for problem
problem (SALBP)
(SALBP)
assembling consists
consists of
a product, of assign-
toassign-
work- cated packing
cated algorithms
heuristics, for are suggested.
example, the There
ranked are also
positional dedi-
weight
simple line-balancing problem (SALBP) consists (RPW) heuristics,
algorithm, forHelgeson
example,and theBirnie
ranked(1961):
positional
first weight
assign
ing tasks
ing tasks
stations necessary
necessary
such that the for
for assembling
assembling
idle time (or aanumber
product,
product, of of
to
to assign-
work- cated
work-
stations, (RPW)
(RPW)
the heuristics,
tasks algorithm,
algorithm,
which forHelgeson
have example,
Helgeson
long theBirnie
and
and
chains ranked
Birnie
of positional
(1961):
(1961):
succeeding first
first
tasks.weight
assign
assign
The
ing tasks
stations
stations
cycle necessary
such
such
time, that
that
cost) foridle
the
is the
minimizedassembling
idle time
time (or
and(or anumber
product,
number
precedence of
of to work- (RPW)
stations,
stations,
constraints the tasks algorithm,
which haveHelgeson
long and
chains Birnie
of (1961):
succeeding first
tasks. assign
The
stations such that the idle time the
lengthtasksof which
the have
chain is long chains
measured of succeeding
either by the tasks.
number Theof
cycle
cycle
between time,
time,taskscost)
cost) issatisfied.
areis minimized
minimized and(or
and number of
precedence
precedence stations, the
constraints
constraints lengthtasksof which
the chainhaveis
chain long
isor chains
measured ofofsucceeding
either the tasks.
byoperation
number Theof
cycle time, cost) length
successor of the
operations measured
by the sum either the by the number times.of
between
between tasks
tasks areissatisfied.
are minimized and precedence constraints length
satisfied. successor of operations
the chain
operations isor measured
or(1961)
by the either
thesuggested
sum of the
the by the number
operation times.of
A fundamental
between tasks areand comprehensive analysis of SALBP successor
satisfied. successor
Kilbridge and Wester by sum of a operation
method times.
based on
A fundamental
fundamental and (1986).
comprehensive analysisonof SALBP Kilbridge
ofassembly Kilbridge operations
and Wester or
and Westerof(1961)
graph presentation by
(1961) the sum
suggested
suggested
precedence of the
a operation
method
a methodTasks
constraints. times.
based
basedare on
on
A
was done in Baybars and comprehensive analysis
Recent surveys SALBP Kilbridge and Wester
graph presentation
presentation of(1961) suggested
precedence aaremethod
constraints. basedare
Tasks on
are
A
was
was fundamental
done
done in
in and
Baybars
Baybars comprehensive
(1986).
(1986). Recent
Recent
line balancing techniques are presented in Ghosh and graph analysis
surveys
surveys on
onof SALBP
assembly
assembly graph
assigned layer by of
layer, precedence
because constraints.
there no Tasks
precedence
was
line done in
balancing Baybars (1986).
techniques areRecent
presentedsurveys in on
Ghoshassembly
and assigned
assigned
constraintspresentation
layer
layer by
between oftasks
by layer,
layer, precedence
because
because
of the constraints.
there
there
same are
are no of
no
layer Tasks
precedence
precedence
a are
graph.
line
Gagnon balancing
(1989); techniques
Rekiek et al.are presented
(2002); Boysen inetGhosh and assigned
al. (2007); constraints layer by layer,
between tasksbecause the there
of the same are no of
layer precedence
of
line balancing
Gagnon
Gagnon (1989); techniques
(1989); Rekiek et
Rekiek al.are
et al. presented
(2002);
(2002); Boysen
Boysen inet
etGhosh and constraints
al. (2007);
al. (2007); between tasks of same layer aa graph.
graph.
Gagnon (1989); Rekiek et al. (2002); Boysen et al. (2007); constraints between tasks of the same layer of a graph.
Copyright © 2017 IFAC 2291
2405-8963 ©
Copyright © 2017,
2017 IFAC (International Federation of Automatic Control)
IFAC 2291 Hosting by Elsevier Ltd. All rights reserved.
Copyright
Peer review©under
2017 responsibility
IFAC of International Federation of 2291Control.
Automatic
Copyright © 2017 IFAC 2291
10.1016/j.ifacol.2017.08.189
Proceedings of the 20th IFAC World Congress
2256
Toulouse, France, July 9-14, 2017 Alexandre Dolgui et al. / IFAC PapersOnLine 50-1 (2017) 2255–2259
Arcus (1966) developed an heuristic called COMSOAL
(COmputer Method for Sequencing Operations for As-
sembly Lines). It randomly selects operations and after
a large number (e.g. 1000) of iterations keeps the best
solution. Evolutionary algorithms are also developed, see
Falkenauer and Delchambre (1992) for a grouping genetic
algorithm (GGA).
In this paper, the simple assembly line balancing problem
of type 1 (SALBP-1) is considered where the cycle time
is fixed and the objective is to minimize the number of
stations, see Baybars (1986) and Scholl (1999). Two un-
conventional ways are proposed to estimate the complexity
of such a problem instances and to improve algorithms to
solve the problem: modification of the graph of precedence
relations and investigation of a problem with an opposite
optimality criteria. We show how these techniques can be
employed and why they are useful to analysis of assem-
bly line balancing problem instances and to increase the
performance of existing methods.
The Simple Assembly Lines Balancing Problem of type
1 (SALBP-1), considered in this paper, is formulated as
follows:
A set N = {1, 2, . . . , n} of operations is given. For each
operation j ∈ N , a processing time tj ≥ 0 is known. A
cycle time c ≥ max{tj , j ∈ N } is also known and fixed.
Furthermore, the finish-start precedence relations i → j
are defined between the operations according to an acyclic
directed graph G.
The objective is to assign each operation j, j = 1, 2, . . . , n,
to a station in such a way that:
- number m of stations used is minimized;
-∑for each station k = 1, 2, . . . , m, its total load time
j∈Nk tj does not exceed c, where Nk is the set of
operations assigned to station k;
- precedence relations are fulfilled, i.e. if i → j, i ∈ Nk1
and j ∈ Nk2 , then k1 ≤ k2 .
It is known that from the worst case analysis point of
view, the SALBP-1 is NP-hard in the strong sense. The
literature on SALBP-1 is rich. Nevertheless, some specific
studies on comparison of SALBP-1 instances is missing.
Indeed, it will be useful to suggest techniques helping to
evaluate a relative complexity of real life instances and
academic benchmarks. Also, in real life applications, it is
important to know if it is possible to solve this problem
for a given number of tasks. What is why the possibility
to evaluate the complexity of a SALBP instance is an
important practical issue.
From this perspective, here two new ways for simple as-
sembly line balancing research are suggested: i) reduc-
tion of precedence graphs to planar ones to minimize
the number of precedence relations have to be considered
and ii) transformation of the SALBP-1 to a maximization
problem to estimate its solution space. These approaches
open completely new and promising ways for comparison
of SALBP instances. They can be also a departure point
to develop new optimization algorithms.
In Section 2, a new idea dealing with transformation of
precedence graphs to planar ones is discussed. Examples of
such transformations are given. This approach can reduce
2292
the complexity of specific examples of simple assembly line
balancing problems as well as can be used to compare
the known benchmark instances employed to test different
optimization algorithms.
In Section 3, a new simple assembly line balancing problem
where the number of stations is maximized, is suggested.
Only solutions with maximal station loads are considered.
Some examples of solving such a problem are reported.
This transformation of SALBP-1 to a maximization prob-
lem can also be used to evaluate the complexity of simple
assembly line balancing problem instances.
2. PLANAR GRAPH FOR PRECEDENCE
RELATIONS
In literature, order strength (OS) is considered as a key
characteristic of SALBP benchmarks, see for example
Scholl (1993). OS is estimated according to number n·(n−
1)/2 of precedence relations to measure SALBP instance
complexity (http://www.assembly-line-balancing.de). In
this section, it is demonstrated how to reduce graph of
precedence to planar one as well as benefits of such a
reduction in term of number of precedence relations.
Here it will be proved that any graph can be reduced
to planar one with a number of precedence relations no
greater than 3n − 6. If in the original graph, operation i
is a predecessors of operation j (immediate or not), than
this relation remains in the modified planar graph too.
Thus, we can suggest modifying a graph before solving
an instance because this modification reduce the order
strength of the problem and so its complexity.
In Lazarev and Gafarov (2009), authors present a trans-
formation of graph G of precedences to planar one for
the Resource-Constrained Project Scheduling Problem
(RCPSP). The same approach can be used for SALBP-
1.
Theorem 1. For any instance of SALBP-1 with n opera-
tions and v precedence relations, there exists an analogous
instance with a planar graph G′ with n′ operations and v ′
relations, where n + v ≥ n′ + v ′ .
An analogous instance can be obtained from the original
by adding ”dummy” operations (with tj = 0) and deleting
all unnecessary relations. The proof of the theorem follows
from Lemmas 6 and 3.
Lemma 2. If there is a subgraph G′ ⊂ G that is isomorphic
to the special graph K3,3 (complete bipartite graph on six
vertices, three of which connects to each of the other three,
also known as the utility graph), then it is possible to
transform it into a planar subgraph by adding ”dummy”
jobs (with tj = p = 0) and deleting all the unnecessary
relations. See the transformation rules shown in Figures 1,
2 and 3.
Lemma 3. If there is a subgraph G′ ⊂ G that is isomor-
phic to the special graph K5,2 (the complete graph on
five vertices), then it can be transformed into a planar
subgraph by deleting all the unnecessary relations. See the
transformation rule shown in Figure 4.
Thus, according to the Theorem of Pontryagin and Kura-
tovski Kuratowski (1930), the modified graph is planar.
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Alexandre Dolgui et al. / IFAC PapersOnLine 50-1 (2017) 2255–2259
Fig. 1. Transformation of K3,3
Fig. 2. Transformation of Kk,k
Fig. 3. Transformation of K4,4
Fig. 4. Transformation of K5,2
According to Euler’s Theorem, in such a planar graph
v ′ ≤ 3n′ − 6.
The number of precedence relations influences the running
time and theoretical complexity of algorithms. Usually,
the time complexity of solution algorithms for SALBP-
1 looks like O(f (n + v)), i.e. depends almost equally on n
and v. For example, in a list scheduling algorithm, after
assigning an operation, it is necessary to consider all arcs
directed from it, to check if its successors are available for
the next assignment. Since, in a planar graph the value
n+v is reduced, then running time will be reduced too. Of
course we add dummy jobs, but algorithms can be adapted
in a way that dummy jobs will not affect running time
substantially, because it is known which jobs are dummy.
2293
2257
Moreover, by transforming graphs of precedence relations
for benchmarks to planar ones, the benchmarks will be
normalized, and thus it will be possible to compare all
algorithms in the exactly same conditions.
3. MAXIMIZATION OF NUMBER OF STATIONS ON
SOLUTIONS WITH MAXIMAL STATION LOADS
In this section, instead of the standard SALBP-1, where
the number of stations is minimized, the simple assembly
line balancing with the opposite objective criterion is
considered: the number of stations is maximized. Similar
approaches are know in the scheduling theory, see for
example Aloulou et al. (2004).
To make this maximization problem non trivial, it is
assumed that all stations should have a maximal workload,
i.e. for two stations m1 and m2 (m1 < m2 ), there is no
operation j assigned to station m2 which can be moved
to station m1 without violating precedence or cycle time
constraints.
Let m be the number of stations for a feasible solution
with maximal station loads, mmin be the minimal number
and mmax be the maximal number of stations for solutions
with maximal station workloads.
Denote maximization simple assembly line balancing prob-
lem as max-SALBP-1. In the following, it will be proved
that the max-SALBP-1 is NP-hard in the strong sense.
3-Partition problem:
A set N = {b1 ,∑ b2 , . . . , bn } of n = 3m positive integers
n B B
is given, where i=1 bj = mB and 4 < bj < 2 , j =
1, 2, . . . , n. Does there exist a partition of N into m subsets
N 1 , N 2 , . . . , N m such that each subset consists exactly of
three numbers and the sum of the numbers in each subset
is the same, i.e.,
∑ ∑ ∑
bj = bj = . . . = bj = B?
bj ∈N 1 bj ∈N 2 bj ∈N m
Lemma 4. max-SALBP-1 is NP-hard in the strong sense.
Proof. Use a reduction from the 3-Partition problem.
Consider an instance of the 3-Partition problem with 3m
numbers. Let M = (mB)2 . Construct an instance of
max-SALBP-1 with 3m + 1 operations, where tj = M +
bj , j = 1, 2, . . . , 3m and t3m+1 = M . In addition, let
c = B + 4M − 1.
If for the instance of the 3-Partition problem the answer is
”YES”, then there exists an optimal solution with m + 1
stations. Operations which correspond to numbers from
the set N i are assigned to the station i, i = 1, 2, . . . , m.
The operation 3m + 1 is assigned to the station m + 1. If
the answer is ”NO” , then mmax = m, and on a station, 4
operations are assigned (including the operation 3m + 1).
In Queyranne (1985), it was proven that for any polyno-
mial time heuristic for SALBP-1, the worst-case ratio is at
least 32 , i.e. for a heuristic algorithm there is an instance
for which mm 3
min ≥ 2 .
As a consequence the following lemma holds.
Lemma 5. max-SALBP-1 cannot be approximated with a
ratio < 32 unlike P = N P .
Proceedings of the 20th IFAC World Congress
2258
Toulouse, France, July 9-14, 2017 Alexandre Dolgui et al. / IFAC PapersOnLine 50-1 (2017) 2255–2259

SALBP-1 is a generalization of the Bin-Packing Problem Some ∑ other Upper Bounds are used, e.g., U B2 =
(BP ), thus this approximation can be compared with 2· tj
⌈ c ⌉ − 1. Upper bounds are recomputed for each sub-
known results for the BP (e.g., see Queyranne (1985)),
problem. Unfortunately, the B&B proposed cannot solve
where there is a heuristic for which m ≤ 11/9mmin + 1.
the majority of benchmarks in 10 minutes (600 seconds)
For BP , it is known that for any polynomial time heuristic,
on CPU INTEL CORE 2 DUO 2.4 Hz (only one processor
the worst-case ratio is at least 32 , as well. But it holds only
is used). However, the results obtained, see Table 1, can be
for mmin = 2, m = 3, i.e. the absolute error is equal to used to estimate the difference between the minimal and
1. For mmin > 2, the worst case is m ≤ 11/9mmin + maximal numbers of stations for these benchmarks.
1. It is possible to conjecture, that the same relation
holds for SALBP-1 too, but in Queyranne (1985), authors The purpose of this numerical tests was not to check the
showed that the worst-case ratio 32 holds for any absolute efficiency of the algorithm proposed but to give a view
error, i.e., for any polynomial time heuristic and for any how big is difference between the maximal and minimal
mmin ≥ 2, there is an instance for which mm 3
min = 2 , where
numbers of stations for considered benchmarks.
m is computed by the heuristic. Table 1. Minimal and maximal numbers of
A similar result for max-SALBP-1 can be demonstrated. stations for known benchmarks
Lemma 6. For any polynomial time heuristic for max- Instance n c mmin mmax running time (sec)
SALBP-1 and for any given absolute error q ≥ 3, there Arcus1 83 3786 21 24 600.00
max
is an instance for which mm = 32 and mmax − m = q. Arcus2 111 5755 27 30 600.00
Barthol2 148 84 51 57 600.00
Proof. Consider the following reduction from the Parti- Barthold 148 403 14 16 600.00
tion Problem. Let an instance of the Partition Problem Bowman 8 20 5 5 <0.01
with n values Buxey 29 27 13 16 600.00
∑n be studied. Without lost of generality assume
A = 12 i=1 bi . Construct an instance ∪ ∪ of max-SALBP-1
∪
Gunther 35 41 14 16 600.00
with a set of operations N = N1 N2 . . . Nq . Assume Hahn 53 2004 8 9 600.00
Heskiaoff 28 138 8 10 600.00
c = 2n2 A · n2q . Each set Nl , l = 1, 2, . . . , q, contains Jackson 11 7 8 9 <0.01
2n + 3 operations, with processing times tj = Ml + bj , j = Jaeschke 9 6 8 8 <0.01
1, 2, . . . , n, and tj = Ml , j = n + 1, . . . , 2n + 1 and t2n+2 = Kilbridge 45 56 10 12 600.00
t2n+3 = Tl = c − ((n + 1)Ml + A − 1), where Mq = 2n2 A Lutz1 32 1414 11 13 600.00
and Ml = Ml+1 ·(n+2), l = q −1, q −2, . . . , 1. In addition, Lutz2 89 11 49 52 600.00
assume i → j, ∀i ∈ Nl , ∀j ∈ Nl+1 , l = 1, 2, . . . , q − 1 and Lutz3 89 75 23 25 600.00
2n + 2 → i, where 2n + 2, ∀i ∈ Nl , l = 1, 2, . . . , q. Mansoor 11 48 4 5 <0.01
Mertens 7 6 6 6 <0.01
Jobs from the different subsets Nl , l = 1, 2, . . . , q, cannot Mitchell 21 14 8 10 0.36
be assigned to the same station. Jobs 2n+2 and 2n+3 from Mukherje 94 176 25 27 20.90
the same subset are assigned to different stations too. If for Roszieg 25 14 10 11 247.32
the instance of the Partition problem the answer is ”YES”, Sawyer 30 25 14 17 600.00
then there exists an optimal solution with mmax = 3q Scholl 297 1394 50 54 600.00
stations, where jobs from the subset Nl , l = 1, 2, . . . , q, are Tonge 70 160 23 26 600.00
assigned to stations 3(q − 1) + 1, 3(q − 1) + 2, 3(q − 1) + 3 Warnecke 58 54 31 36 600.00
according to their assignment from the proof of Lemma 4. Wee-mag 75 28 63 63 14.66
For the instance in a line balance with mmin = 2q stations,
jobs from each subset Nl , l = 1, 2, . . . , q, occupy only 2 If the running time less than 600, then the value mmax
stations. is optimal. After running the algorithm for 60 seconds,
the same values mmax for all the instances was obtained,
The lemma is proven.
except for Arcus2 (after 60 sec, mmax = 29) and Lutz1
The calculation results of maximal number of stations (after 60 sec, mmax = 12). These results show that the
for several benchmarks from http://www.assembly-line- maximal deviation mmax − mmin found does not exceed
balancing.de are reported here. To maximize the number 20%.
of stations, a simple B&B algorithm with the deep-first
branching strategy and a simple Upper Bound based on 4. CONCLUSION
the following observations were developed.
Let us compute P redj which is a set of all predeces- In this paper, two new ideas for a future research on
sors (not immediate, as well) for each operation j, j = simple assembly line balancing of type 1 (SALBP-1) are
1, 2, . . . , n. Then, the earliest station Sj to which the oper- proposed: i) techniques to reduce graphs of precedence
ation to planar ones; ii) the maximization version of SALBP-
∑ j = 1, 2, . . . , n, can be assigned is computed as Sj = 1, called max-SALBP-1.
( ti )+tj
i∈P red
j
⌈ c ⌉. Let M Lk be the minimal load of the The technique to reduce graphs of precedence to planar
station k, k = 1, 2, . . . , n. Let tmin
k = minj {tj |Sj ≤ k} and ones deals with a more precise calculation of complexity
tmax
k = max {t |S
j j j ≤ k}. Then,∑ Lk∑=m max{tk , c −
M min
for benchmarks. All benchmarks can be normalized by
tmax
k + 1} and U B 1 = min{m| t j − l=1 M L l ≤ 0} is applying such a reduction, thus only planar graphs of
an Upper Bound for the maximal number of stations. benchmarks will be compared and not any graphs. The
perspectives for this research are in analyzing behavior of

2294
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Alexandre Dolgui et al. / IFAC PapersOnLine 50-1 (2017) 2255–2259
different known algorithms and differences in their perfor-
mances on all known normalized benchmarks. An interest-
ing topic for the future research is also the development of
new line balancing algorithms based on the properties of
planar graphs.
The maximization version of SALBP-1 provides new ele-
ments for understanding why some benchmarks harder to
solve than others with greedy heuristics.
In this paper, it was demonstrated that this new problem
max-SALBP-1, is NP hard in strong sense. A tentative
algorithm to solve it is also proposed. Numerical tests are
given for several known benchmarks. Further studies are
necessary on all available benchmarks to understand better
relations between the difference max − min number of
stations and problem complexity. Also, an open question
is how this characteristic of problem can be used for
the development of new and more efficient algorithms for
SALBP-1.
ACKNOWLEDGEMENTS
The work was partially supported by the Conseil Régional
des Pays de la Loire, France.
REFERENCES
Agnetis, A. and Arbib, C. (1997). Concurrent operations
assignment and sequencing of particular assembly prob-
lems in flow lines. Annals of Operations Research, 69, 1
– 31.
Aloulou, M., Kovalyov, M., and Portmann, M.C. (2004).
Maximization problems in single machine scheduling.
Annals of Operations Research, 129, 21–32.
Arcus, A. (1966). Comsoal: A computer method of
sequencing operations for assembly lines. International
Journal of Production Research, 4, 259 – 277.
Askin, G. and Standridge, C. (1993). Modeling and
Analysis of Manufacturing Systems. John Wiley & Sons.
Battaia, O. and Dolgui, A. (2013). A taxonomy of
line balancing problems and their solution approaches.
International Journal of Production Economics, 142(2),
259–277.
Battaia, O., Dolgui, A., and Guschinsky, N. (2017). Deci-
sion support for design of reconfigurable rotary machin-
ing systems for family part production. International
Journal of Production Research, 55(5), 1368–1385.
Battini, D., Delorme, X., Dolgui, A., Persona, A., and
Sgarbossa, F. (2016). Ergonomics in assembly line bal-
ancing based on energy expenditure: a multi-objective
model. International Journal of Production Research,
54(3), 824–845.
Baybars, I. (1986). A survey of exact algorithms for the
simple assembly line balancing. Management Science,
32, 909–932.
Bentaha, M.L., Battaia, O., and Dolgui, A. (2015). An
exact solution approach for disassembly line balancing
problem under uncertainty of the task processing times.
International Journal of Production Research, 53(6),
1807–1818.
Borisovsky, P.A., Delorme, X., and Dolgui, A. (2014).
Balancing reconfigurable machining lines via a set par-
titioning model. International Journal of Production
Research, 52(13), 4026–4036.
2295
2259
Bowman, E. (1960). Assembly line balancing by linear
programming. Operations Research, 8 (3), 385 – 389.
Boysen, N., Fliedner, M., and Scholl, A. (2007). A classi-
fication of assembly line balancing problems. European
Journal of Operational Research, 183, 674–693.
Dolgui, A. and Ihnatsenka, I. (2009). Balancing modular
transfer lines with serial-parallel activation of spindle
heads at stations. Discrete Applied Mathematics, 157(1),
68–89.
Dolgui, A. and Proth, J.M. (2010). Supply Chain Engi-
neering:Useful methods and techniques. Springer.
Falkenauer, E. and Delchambre, A. (1992). A genetic al-
gorithm for bin packing and line balancing. Proceedings
of the IEEE International Conference on Robotics and
Automation, 1186 – 1192.
Ghosh, S. and Gagnon, R.J. (1989). A comprehensive
literature review and analysis of the design, balancing
and scheduling of assembly systems. International
Journal of Production Research, 27, 637 – 670.
Hazir, O., Delorme, X., and Dolgui, A. (2015). A review
of cost and profit oriented line design and balancing
problems and solution approaches. Annual Reviews in
Control, 40, 14–24.
Hazir, O. and Dolgui, A. (2015). A decomposition based
solution algorithm for u-type assembly line balancing
with interval data. Computers Operations Research,
59, 126–131.
Helgeson, W.B. and Birnie, D.P. (1961). Assembly line
balancing using the ranked positional weight technique.
Journal of Industrial Engineering, 12 (6), 394 – 398.
Jackson, J. (1956). A computing procedure for the line
balancing problem. Management Science, 2, 261 – 271.
Kilbridge, M. and Wester, L. (1961). A heuristic method
for assembly line balancing. Journal of Industrial Engi-
neering, 12, 292 – 298.
Kuratowski, K. (1930). Sur le probleme des courbes
gauches en topologie. Fundamental Mathematics, 15,
271–283.
Lazarev, A. and Gafarov, E. (2009). Transformation of the
network graph of scheduling problems with precedence
constraints to a planar graph. Doklady Mathematics,
79/1, 1–3.
Morrison, D., Sewell, E., and Jacobson, S. (2014). An
application of the branch, bound, and remember algo-
rithm to a new simple assembly line balancing dataset.
European Journal of Operational Research, 236/2, 403–
409.
Queyranne, M. (1985). Bounds for assembly line balancing
heuristics. Operations Research, 33, 1353–1359.
Rekiek, B., Dolgui, A., Delchambre, A., and Bratcu, A.
(2002). State of art of assembly line design optimization.
Annual Reviews in Control, 26, 163–174.
Salveson, M. (1955). The assembly line balancing problem.
Journal of Industrial Engineering, 6 (3), 18 – 25.
Scholl, A. (1993). Data of assembly line balancing problem.
Technische Hochshcule Darmstadt, 16/93, 32 pp.
Scholl, A. (1999). Balancing and Sequencing of Assembly
Lines. Physica-Verlag, A Springer-Verlag Company,
Heidelberg.
Wee, T. and Magazine, M. (1982). Assembly line balancing
as generalized bin packing. Operations Research Letters,
1, 56 – 58.