Robotics and Computer Integrated Manufacturing xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Robotics and Computer Integrated Manufacturing

journal homepage: www.elsevier.com/locate/rcim

A problem design and constraint modelling approach for collaborative

assembly line planning
⁎
Daniel Neumann , Jan Keidel
Chemnitz University of Technology, Chemnitz, 09111, Germany

A R T I C LE I N FO A B S T R A C T

Keywords: Assembly line planning is the interface between agents of different professions, competencies, skills and ex-
Collaboration periences. Each agent is characterized by specific objectives and constraints, that must be considered when
Assembly line design different products and processes are merged across multiple levels of various decision making units. A systemic
Assembly line planning platform for the dependencies of goals, constraining relationships of elements to be assigned and the inter-
Collaborative planning
dependencies to properties of the elements is designed. Hereby, the objectives and the constraints are evaluated
Collaborative problem modelling
and mapped within the assembly line planning (ALP) process, resulting in a comprehensive graph with the
processes, structures, resources and their properties with its dependencies. The dependencies include time cri-
teria but also technical, processual, ergonomic, qualitative, part-dependent along with resource-dependent,
structural constraints. The dependency map is integrated within an existing ALP process to achieve a temporal
precedence on the incoming and outgoing information resulting in a dynamic process oriented presentation of
the objectives and constraints. The elements of the dependency map are categorized and analysed based on the
evaluations and a problem model is created for the presented ALP problem. The problem model is solved using a
genetic algorithm which is parameterized in accordance to the problem model complexity and characteristic.

1. Introduction costumer needs. Multi-purpose machines with automated tool swaps

This paper provides a multi-objective approach of assembly line
planning within the automotive industry. The approach differs from
other approaches in scientific literature by considering collaboration
aspects between planning participating actors of different professions,
competencies and practical skills. The paper is being divided into four
further sections. At first, a basic summary of assembly line planning is
given to introduce the considered problem, essential properties and
characteristics, as well as possible solution methods. Subsequent, the
concept of multi-objective precedence modelling is given. This concept,
including solution procedures, will be clarified and explained by a
practical example, which is illustrated based on a car model. Remarks
are closed by conclusions.
2. Assembly line planning
Assembly lines are flow-line production systems originally devel-
oped for mass production of standardized products. In recent times,
assembly lines are increasingly important in mass customization.
Especially, but not solely in automotive industry an individualization of
products must be taken into consideration to respond to different
allow varying models, yet respectively excluding considerable setup
costs [1].
In general, operations are performed by serially aligned production
units. Workpieces pass these stations successively as they are moved
along the line by using a transportation system, e.g. a conveyor me-
chanism [1]. These assembly systems are associated with high invest-
ment costs. Because of that, the configuration of an assembly line is
very important. In this planning process, all decisions regarding
equipment, alignment and capacity (number of stations, stroke cycle
time, number of workers in a stroke) of stations for a given product or
product group with a high number of operations and precedence rela-
tions must be made before the production starts. In scientific literature,
this configuration problem of assembly systems is called Assembly Line
Planning (ALP) or Assembly Line Balancing (ALB). The Simple As-
sembly Line Balancing Problem (SALBP) describes an assembly line as a
one-product-problem, a serial line, one worker per station and the ob-
jective of maximization of workload. This problem could be declared as
the academic standard problem formalization [1,2].
Boysen, Fliedner and Scholl introduced a classification scheme of
the General Assembly Line Balancing Problem (GALBP). It is divided
into three elements, which are noted as tuples [α, β, γ]. In this context α

⁎
Corresponding author
E-mail addresses: daniel.neumann@wirtschaft.tu-chemnitz.de (D. Neumann), jan.keidel@wirtschaft.tu-chemnitz.de (J. Keidel).

https://doi.org/10.1016/j.rcim.2018.05.002
Received 8 June 2017; Received in revised form 13 April 2018; Accepted 2 May 2018
0736-5845/ © 2018 Elsevier Ltd. All rights reserved.

Please cite this article as: Neumann, D., Robotics and Computer Integrated Manufacturing (2018), https://doi.org/10.1016/j.rcim.2018.05.002
D. Neumann, J. Keidel Robotics and Computer Integrated Manufacturing xxx (xxxx) xxx–xxx

describes the characteristics of the precedence relation graph, β the Table 1

characteristics of assembly lines and stations and γ the different plan- Solution approaches for assembly line planning.
ning objectives, e.g. minimal number of stations or minimal costs of Author Year Problem GA LS ACO ABC MA PSO Source
assembly line configuration. In this regard, SALB characteristics are type
chosen as common reference for classification, meaning that only de-
Rubinovitz 1995 ALBP x [6]
viations from that basic problem description must be provided by using
Tsujimura 1995 GALBP x [7]
the tuple notation. The presented classification scheme has been Plans 1999 SALBP-E x [8]
adopted from the machine scheduling scheme of Graham et al., which Pastor 1999 ALBP x [9]
was used and refined by Brucker et al. for project scheduling [1–3]. Ponnambalam 2000 GALBP x [10]
The Mixed-Model Assembly Line Balancing Problem (MMALBP), Bautista 2000 ALBP x x [11]
Rekiek 2001 ALBP x [12]
coupled with several variants of the same product is very important
Bautista 2005 TSALBP x X [13]
within the automobile industry. In general, this problem relates to the Rekiek 2006 MMALBP x [14]
operative task of determination of a variant sequence (Mixed-Model Vilarinho 2006 ALBP-I X [15]
Sequencing Problem) [4]. In addition, team organization, parallel tasks Zhang 2007 SALBP-1 X [16]
Su 2007 MALBP x [17]
and stations, processing alternatives, a very high number of parts and
Zhang 2008 ALBP-wa x [18]
tools as well as a division into production segments must be taken into Guo 2008 FALBP x x [19]
consideration. Cao 2008 MALBP x [20]
In this context, a wide gap between theoretical discussion and Bautista 2009 TSALBP x [21]
practical applications does exists. Most of the standard optimization Kao 2009 ALBP-1 x [22]
Khaw 2009 ALBP/ U- x X [23]
models are not able to consider all those practical aspects together with
Type ALBP
task-station assignment restrictions (combined tasks, assembly high, Chica 2009 TSALBP X [24]
ergonomic and qualification aspects) and multiple planning objectives. Jonnalagedda 2010 MALBP-2 x [25]
Furthermore, different roles and actors are essential for the planning Brudaru 2010 ALBP/ U- x [26]
process. Involved persons or groups can be divided into process and Shaped
Tang 2010 SALBP x [27]
product planners, product developers, logistics, plant and station Chica 2011 TSALBP X x [28]
planners and information as well as material supply planners [1–3]. Chica 2011 TSALBP x [29]
The depicted problem class can be assigned to Combinatorial Qian 2011 MALBP x x [30]
Optimization because of the finite solution space. The resulting high Sulaiman 2011 SALBP-1 X [31]
Hu 2011 Two-sided x [32]
complexity of these comprehensive planning problems, interconnected
ALBP
with the problem size, necessitates the use of capable solution methods. Zhuo 2011 U-Type x [33]
For practical occurring problems with more than a thousand operations, ALBP
the use of exact solution methods is impossible concerning the purpose Noushabadi 2011 SALBP-2 x [34]
of computation time. In this case, heuristic methods can be used to get Razali 2011 ALBP x [35]
Chica 2012 TSALBP x [36]
satisfying solutions combined with an acceptable computation time. Nilakantan 2012 Robotic x x [37]
Such methods can be divided into problem specific heuristics (con- ALBP
structive procedures) and metaheuristics (meta-strategies) [2,4,5]. Liao 2012 MALBP x [38]
Especially in the last two decades, a large variety of heuristic ap- Adham 2012 ALBP x [39]
Jaturanonda 2013 ALBP PL x [40]
proaches for different specifications of ALBP was developed.
Smoothness
Constructive procedures, like priority rule based approaches or in- Yu 2013 MMALB x [41]
complete enumeration approaches, as well as metaheuristics, like, Liu 2013 ALBP x x [42]
Genetic Algorithms (GA), Ant Colony Optimization (ACO) or local Saif 2014 SALBP x [43]
search strategies (LS), are used for searching feasible solutions to the Davidrajuh 2014 ALBP x [44]
Triki 2014 ALBP x [45]
assembly line balancing problem. As stated in Table 1 Genetic Algo- Chica 2015 r-TSALBP- x [46]
rithms are mainly used to solve the assembly line planning problem. m/a
Therefore, the problem formalization as well the modelling system is Sikora 2015 ALBP-2 x [47]
created with respect to use a genetic algorithm as solution search. 21 17 7 1 2 2

3. Collaborative precedence modelling

Those input parameters, especially the precedence relations and the
restrictions, are related or are extracted by decision makers, using a
According to the presented problem models, various objectives can
wide variety of information exemplified in Table 1. The evaluation of
be aimed for within the assembly line planning process. The most ob-
ergonomic aspects is carried out after the first assembly line is designed
vious objective is the minimization of costs. The question of optimizing
based on the assembly height and the assembly zone, as well as the
the costs hereby is, as always, limited to the direction it is aimed at. An
manufacturing equipment and possible product handling technology.
assembly line with minimal costs for logistics is most certainly not
In industrial practice, it is not common that all the precedence re-
equal to an assembly line with minimal production costs (i.e., labour
lations are either known when the assembly line design process is in-
costs, machinery, manufacturing site), and not equal to an assembly
itiated, nor will one planner state all the precedence relations.
line designed with minimal production planning costs. Besides the op-
Therefore, it only seems obvious to carry out investigations regarding
timization objective(s) following basic input parameters are used for
the potential of collaborations within the assembly line design process.
assembly line design [48]:
So far, merely the construction process takes the collaborative assembly
into consideration. Reviewing the process's perspective, no efforts have
▪ precedence relations between a set of elements (structural, proces-
been made to establish a collaborative planning process. In accordance
sual),
to the concept presented before (as presented in [49]), an approach for
▪ cycle time as time limit for structural element where process ele-
collaborative precedence modelling will be presented on behalf of au-
ments are assigned to,
tomotive assembly line planning.
▪ assignment constraints for tasks related to stations, competencies,
The pool of information is created by the decision makers and
logistics elements and technology.

2
D. Neumann, J. Keidel Robotics and Computer Integrated Manufacturing xxx (xxxx) xxx–xxx

Table 2 ▪ i before j
Input parameters for industrial assembly line design. ▪ i after j,
Input section Relevant planning data ▪ i parallel j,
▪ i assigned to j = j has assigned i
Assembly line section sub-assembly lines; cycle times ▪ i part of j = j consists of i,
materials/product handling technology
▪ i depends on j
Station machinery/manufacturing equipment offered
product height
Workplace assembly zone with different weights:
work force qualification
Task task description; task time; product/model ▪ must (not),
resources (machinery/manufacturing equipment)
▪ should (not),
container size per part/task
assembly location/assembly height/assembly zone ▪ independent,
Assembly process precedence relation
required parts and tasks and distance:
Logistics container size offered per part
equipment/container bundling
▪ i directly precede j: Position i = Position j − 1
▪ i precede by n elements j: Position i = Position j − n
planners in form of documents within the innovation process. These ▪ i assigned to j on nodeLevel k
planners give their restrictions and data to the subsequent innovation
process steps in form of documents. In case any changes must be made, The nodes represent planning elements, classes and properties. A
the planners will focus on the received documentation, regardless of planning element can be a task, a machine, a station or anything else
further possible changes to other areas of interest. All created docu- relevant to the assembly line planning process. A class represents a
ments are directly or indirectly linked to the individual objectives of the stereotype of the planning elements. These relations are used to model
participants within the innovation process, as shown in Table 2. the relation between elements and subsequently also the element
Due to the search methods for the underlying assembly line balan- classes as shown in Fig. 1. This process initiates the assembly line
cing problems (ALBP) for new products, it is common, that a pre- planning project and defines the content for the collaborative modelling
cedence relation is modelled. The precedence graph is an essential part and the solution evaluation system.
of the assembly line balancing, and many agents are directly or in- The assembly planning process in the traditional sense consists of
directly involved in the restriction model. Therefore, a collaborative following element classes, which are considered for the planning pro-
precedence modelling technique is presented in Fig. 1. The idea is that cess:
not only the precedence relations defined in the construction process
are used, but also preferences of the planner as well as restrictions are ▪ structure element,
being considered. The upfront modelling of this information accelerates ▪ process element,
the iterations allowing an earlier detection of conflicts. The search ▪ resource element,
quality benefits from this upfront information, enabling a conflict re- ▪ product element and
solution within the modelling process [49] (Table 3). ▪ properties.
To allow a search method to find different weighted interrelations
between the planning elements (elements, classes, properties), the re- Those classes build the basis for more detailed classes and for the
lations between a start node i and a target node j can be modelled as collaborative problem modelling system. As soon the interdependencies
sequence, assignment, dependency or mixed relation with following between the relations, the types of the connected nodes as well the level
directions: of the nodes are modelled, the interface for the solution search must be
defined. For the assembly line planning this is the highest level of

Fig. 1. Individual design for collaborative problem modelling system.

3
D. Neumann, J. Keidel
Table 3
Roles and documents required for assembly line design.
Planning Roles Document for assembly line design
Product Product plan, list of priority parts
Assembly equipment Work piece plan
Assembly layout Targeted layout
Assembly process Product parts list
Logistics process Logistics concept, logistics plan
Technology Technology set, container list
Industrial engineering Assembly line plan,
Location Material flow, parts list, plant layout, ergonomic and legal restrictions
process elements and the lowest level on structure elements, meanwhile
the elements of other classes are considered as additional components
to the solution optimization.
The dependencies between the relations, the classes and the ele-
ments are modelled by different. In contrast to the relations that con-
nect the elements and classes, the parameter relations of the properties
are considered as following:
IF StartNode has Property A with PropertyValue = 1
AND TargetNode has Property A with PropertyValue = 2
THEN StartNode is assigned to StructureNode i
AND TargetNode is assignet to StructureNode j
As it is not possible to merge precedence, assignment, dependencies
and mixed relations, three lists are being created based on the relation
type (sequence, assignment, dependency). To allow a satisfactory eva-
luation of the relationship, the dependencies are carried out in different
matrices for the type of nodes connected by the relation.
First, the parallel relationship is regarded building a cluster matrix,
showing the importance of clustering different tasks or operators for
parallel assembly tasks. Second, the precedence relation is build, re-
garding cluster the tasks which are marked as parallel. Using those
kinds of relations, all relations between tasks, tasks and operators, op-
erators and operators as well as between properties of tasks and op-
erators can be illustrated. Therefore, no zoning or separate priorities
between tasks are required. Only a precedence graph as well a solution
search is carried out. A task hereby can be every type process, not ne-
cessarily a value adding one. An operator can be a workplace, a station,
an assembly line section, a machine, a tool, but also a human worker. A
precedence relation can be modelled using different modelling ap-
proaches (i.e., adjacent matrix, precedence graph, and rulebook).
The parallel and precedence relations are tested for conflicts. This is
done by using in-depth first searches. If all precedence relations are
feasible, a solution search is initiated. In the more common case that the
dependencies are not feasible, single precedence relations are modelled
with highest emphasis (must direct relations and must relations), fea-
sibility is then checked once again. If there is no inconsistency, the
“should”-dependencies are added. In case of a conflicted dependency,
the nodes (tasks, operators) are identified and send back to the modeller
with a recommendation of the most important node due to page rank
and the degree of centrality. This approach helps the planner to un-
derstand the importance of a node for all modelled dependencies. The
feedback of the planner is always required due to the difference of
product and process knowledge between all the modellers. Therefore, a
Table 4
Example for a multi-agent modelled dependency map.
Start node type Start node id Target node type
Task F1 Task
Task F2 Workstation
Task K1 Task
Workplace 1 Workstation
Robotics and Computer Integrated Manufacturing xxx (xxxx) xxx–xxx
Individual objectives
Maximize product quality
Minimize equipment costs
Minimize areas used, minimize station count
Minimize production costs
Minimize logistics time and distance; minimize the buffer at the station
Minimize technological variety
Minimal assemblers per assembly line; minimal supply time
Use all existing personnel
further conglomeration of the planners can be required, considering
only the defined dependency flaws.
A backup approach would be the encapsulation of the inconsistent
dependencies within a placeholder that contains a summed-up proces-
sing time along with all related tasks and operators. The conflict can be
solved manually after possible solutions have been found. The ad-
vantage is that fast processing is possible without continuous feedback
required for any imperfection of the data sets.
4. Example
The method results in a relatively complex building up approach,
nevertheless allows a learning process for the assembly line planning
problem, knowledge management and data acquisition, as well as
mining approaches. The main advantage is that artificial learning and
data mining approaches can be used and other data can be linked to the
once existing precedence graph, which is based on a once created
model.
As an example of proof an extraction of a real-life assembly line is
carried out. A multi-model with 204 tasks and stochastic processing
times as well as the relative frequency is modelled in five iterations
identified by the modelled edges after a period (57 edges, 114 edges,
177 edges, 202 edges, 314 edges). The approach considers tasks,
workstations, workplaces and task clusters. Task clusters are a specific
group of tasks sharing a set of specific properties. The drift resulting in
this assumption is not considered in this solution approach, but can be
carried out using the objective of minimizing the drift. For the drift the
processing time for a frequency of 100% is used to determine the
model-based overlap time. The dependencies between the tasks are
modelled by eight agents, mainly in accordance with their professional
scope. An example of the tabular modelling is presented in Table 4.
In case a planner does model a set of tasks to be parallel, two pos-
sible ways are proposed. A minimal distance of a cycle time, so that one
of the tasks jumps to the next station. The next station suggests another
workforce on the identical physical station. For the precedence graph
an encapsulation of the parallel graphs is therefore useful. The parallel
dependencies are clustered together. For example, the process planner
models in a way that the task F1 precedes task F2, and task K1 is
parallel to F2. Those tasks can be encapsulated. Meaning that any re-
lation made to F1 or F2, (respectively F2 or K1) is also a precedence
relation to group F1F2 consisting of F1 and F2 (F2K1 for F2 and K1).
Especially in the assembly line planning process where agents with
various scopes are brought together within one precedence model, the
main problem is that conflicts will arise. The computational solution of
Target node id Edge weight id Edge direction id Edge distance Agent id
F2 1 1 (prec) −1 (any) 1
1 1 2 (assign) 1 2
F2 1 2 (assign) 1 4
1 2 1 (prec) 1 3
4
D. Neumann, J. Keidel Robotics and Computer Integrated Manufacturing xxx (xxxx) xxx–xxx

Therefore, two scenarios that always include manual feedback are

proposed. It is necessary that the involved agents review the incon-
sistencies and solve eventual conflicts with third parties (related to the
conflict).

Strategy 1: encapsulation of conflicted dependencies

The encapsulations of the conflicted dependencies allow a solution

search. The times for the encapsulated tasks are accumulated, which
allows a placeholder in case only on edge is modelled incorrectly. In
case of a fully planned station with a cycle time overlapping a place-
holder, a new solution search should be initiated. An example is shown
in Fig. 2.

Strategy 2: recommendation based on degree, network flow and centrality

measures

A precedence graph is analysed using network characteristics as

shown in Table 5. The degree of a node gives the number of relation it
has. The smaller the value degree of a node, the least connected nodes it
has. Therefore, if any changes are made, only small interdependencies
should occur.
Fig. 2. Predefined classes from relevant elements in assembly line planning. Based on the Degree, the Centrality and the Eccentricity, a proposal
for the involved planner would be to edit an edge connected to F1 for
Table 5 the first cycle as well as an edge connected to K4 or K5 for the second
Precedence structure analysis of the nodes involved in precedence conflicts. cycle involved. As there is no certainty if an edge is modelled in-
correctly, or there are restrictions not apparent within the analysed
InDegree OutDegree Degree Eccentricity Eigenvector centrality
situation, further action is required. From the input data alone, it is not
K2 3 1 4 3 0.5173 possible to infer to the correct edge for the assembly line planning.
F1 1 1 2 4 0.3988 The decision on how the conflicts in the precedence graph are re-
F2 2 3 5 2 0.4600 solved is carried out within meetings or simple dialogues of the in-
K3 2 2 4 4 0.1810
K4 1 1 2 6 0.1829
volved planners. In case there is no solution to the conflict, other per-
K5 1 1 2 5 0.1829 sonnel can be involved or risk management tools can be deployed. For
this example, it is assumed, that the edge from F1 to F2 and from K5 to
K3 are decided as negligible, as shown in Figs. 3–5.
a conflicted precedence graph is not presented yet. The main problem The solution search is initiated using a nearest neighbour heuristic.
of the collaboration is the conflict finding but also the conflict resolu- Hereby a list of all neighbour nodes incoming and outgoing to every
tion. This matters even more, because no search algorithm can find a node are created. The nodes are sorted based on their indegree. The first
valid assembly plan with an inconsistent graph/precedence relation. node is selected and the solution creation starts. The solution is

yes
Problem modelling

Collect all necessary

Model all dependencies Send conflicted Agents
elements (types: process, Model dependencies
Agent

necessary between edges to agents react?

structure, property, between element types
elements
resource)

no

yes
Problem Modeler

Send conflicted nodes no

Merge depedencies by all Conflicts Fix Dependency Map Conflicts?
Search for conflicts yes and edges to problem
planners exist?
analysis
Problem analyis

no
Analyse node content Analyse position in Evaluate probability
System

and edge weight problem model of failure of all edges

no
Solution
search

Search, evaluate and Satisfying

yes
review solutions solution?

Final Solution

Fig. 3. Collaborative precedence modelling.

5
D. Neumann, J. Keidel
Fix dependency map
Generate dependency map
using restrictive weights and
any direction level
Conflicts? yes Identify conflicted edges
no
Generate depedency map
Analyse conflicts and
using restrictive weights, any
recommend highest ranked
direction level and distances
Conflicts?
no
Generate dependency map Review and adapt
using all modelled relations dependency map
Conflicts?
no
Computable constraint model
Fig. 4. Dependency map fixture process.
generated iteratively based on the outgoing list for so long no other
connection to an outnode already used has an innode. Then the next
node on the first list is used, until all nodes are assigned to a structural
element, i.e. a workplace or a workstation. As soon the available time is
smaller than the processing time required for the task to be assigned to
the structural element, a new structure is generated.
The optimization is carried out using a Genetic Algorithm (GA) with
swap mutation. Genetic Algorithms are stochastic optimization proce-
dures from the field of metaheuristics. In this approach, a 1:1 assign-
ment between the set of genotypes (search space) and phenotype (so-
lution space) is chosen. This approach uses a permutation chromosome
coding.
The selected individuals are varied by the genetic operators
Crossover and Mutation. While Crossover mixed the genetic material of
parents randomly, Mutation varies genetic sequences of one individual.
The new individuals called children are evaluated according to their
fitness. After that, the iteration can be completed using a defined re-
placement scheme. This scheme is responsible to replace determined
individuals in the population by new generated one.
This process will be executed over a high number of iterations,
which causes in a perpetual development of the population. Therefore,
the individuals get a better fitness because of the selection pressure.
Conceptual transferred to the optimization problem better solutions can
be reached from iteration to iteration. Within this approach, several
setting and operators can be chosen. At first, it is possible to choose the
6
Robotics and Computer Integrated Manufacturing xxx (xxxx) xxx–xxx
yes
User Generate fixed dependency
yes no
reviewed? using encapsulations
yes
Conflicts? tEncapsulation
yes > tAvailable?
yes
Temporarily eliminate most
no probable edges
no
number of individual within a population and the number of executed
iterations. There are a Binary Crossover operator as well as a Swap
Mutation and an Inversion Mutation Operator. The best results are
reached by using a combination of Swap and Inversion operator, which
is used within this example, too. The fitness function can be defined
according to the underlying restrictions of the assembly line planning
problem.
In this example, the working height as well as the assembly line
section of each operation, which has to be carried out at the assembly
line, is integrated in the fitness function. By adding an additional
iteration where tasks without indegree are checked to fill gaps in
working stations before so that any task can be assigned to work-
stations, if available time of a workstation is bigger than the processing
time of the considered task.
A reduced structure of the solution method is shown in Fig. 6. Be-
sides the classes of a typical genetic algorithms, the initiation strategy,
the problem structure based on the interdependencies of the element
classes as well as the decode strategy are passed to the genetic algo-
rithm.
For the evaluation the configuration parameters working height,
assembly line section and sum up decode are tested. The evaluation in
Fig. 7 shows that an increasing number of constraints results in an in-
crease of computation time (left y-axis), decrease in the fitness of the
best found solution and an increase in the average acceptance rate of
the best found solution by the agents (right y-axis). The constraints
D. Neumann, J. Keidel Robotics and Computer Integrated Manufacturing xxx (xxxx) xxx–xxx

F1
K2 4
4 4 4 H C
6 F2
4 6 4 E
K1 1 4 1
S
4
3
K3 K5 V T
4
4 4

K4

F1
K2 4 4
4 4 H C
6 F2

K1 4 6 4 E
1 4 1
S 4
K3/K4/K5
3
K3 K5 V T
4
4 4

K4

Fig. 5. Encapsulation of conflicted edges.

Fig. 6. Class diagram of adaptable genetic algorithm scheme.

considered are based on the assignment of process elements to the
workstation, so the tasks are only assigned to a workstation with pro-
cess elements with exactly the same property items. Further, it is ob-
vious that the sum up heuristic will find further improvements of as-
signing non-related process elements.
As the raw numbers will not justify the extra work for modelling
additional edges the acceptance rate of a solution will be. As more
edges are modelled the higher the possibility of an acceptance of the
proposed solution by the algorithm.
However, a feasible assembly line can be found within a short
amount of time. In conclusion, it shows that an automated dependency
conflict resolution, which is feasible, will result in a worse average
station load. Instead, a rapid feedback can be given to the planning
personnel allowing a face-to-face resolution. This small example reveals
a conflict resolution approach in precedence graphs used in collabora-
tive assembly line planning.
5. Summary
This paper, which is based on a collaborative problem solving

7
D. Neumann, J. Keidel
16
14
12
10
8 0.6
6
4
2
0 0
0,0,0 0,0,1 0,1,0 0,1,1
Fitness Best Soluon
Fig. 7. Evaluation of configuration parameters (working height, assembly line section, sum up at decode) based on fitness of best solution, solution acceptance rate
and computing time.
concept, shows that the distributed precedence modelling using parallel
and precedence dependencies opens a new interdisciplinary scope of
identifying conflicts and resolving conflicts within the problem mod-
elling for assembly line planning. This approach adds graph algorithms,
data preparation for algorithmic solution searches and distributed
modelling in scope of assembly line planning. As a starting point, this
paper presents three of many possible approaches for resolving conflicts
within a precedence graph, created by distributed agents using the
above-mentioned collaborative problem-solving concept. Further re-
search of the graph resolution will be carried out over multiple graph
levels as well the automatic graph resolution of parallel and precedence
dependencies using artificial learning and data analysis. The solution
search is carried out using a combination of heuristic, genetic algo-
rithms and a problem-based decoding strategy to allow a solution
search with multiple constraints. As the multitude of constraints and
agents create conflicts on different levels a feasible solution can be
found. Regarding the problem modelling considerations have to be
made for further research to find an optimal measure for involving the
information from multiple parties.
Acknowledgements
We gratefully acknowledge the support by the Federal Ministry of
Education and Research, Germany for sponsoring the joint project
MontAss (FID: 01IS15029B). The project aims for designing an assis-
tance system for collaborative goal, precedence and restriction model-
ling with intelligent and autonomous agents. These input parameters
are the basis for automatic tasks assignment using a genetic algorithm
for assembly line balancing.
In addition, we thank Prof. Dr. J. Käschel for his guidance, support
and the possibility to work on this research project.
Robotics and Computer Integrated Manufacturing xxx (xxxx) xxx–xxx
1.2
1
0.8
0.4
0.2
1,0,0 1,0,1 1,1,0 1,1,1
Soluon Accepted Compung Time
References
[1] N. Boysen, A. Fliedner, A. Scholl, A classification of assembly line balancing pro-
blems, Eur. J. Oper. Res. 183 (2) (2007) 674–693.
[2] A. Scholl, C. Becker, State-of-the-art exact and heuristic solution procedures for
simple assembly line balancing, Eur. J. Oper. Res. 168 (3) (2006) 666–693.
[3] C. Becker, A. Scholl, A survey on problems and methods in generalized assembly
line balancing, Eur. J. Oper. Res. (168) (2006) 694–715.
[4] A. Scholl, Balancing and Sequencing of Assembly Lines, Physica Verlag, 1999.
[5] S. Bock, Modelle und verteilte Algorithmen zur Planung getakteter Fließlinien,
Gabler, Wiesbaden, 2000.
[6] J. Rubinovitz, G. Levitin, Genetic algorithm for assembly line balancing, Int. J. Prod.
Econ. 41 (1-3) (1995) 343–354.
[7] Y. Tsujimura, M. Gen, E. Kubota, Solving fuzzy assembly-line balancing problem
with genetic algorithms, Comput. Ind. Eng. 29 (1-4) (1995) 543–547.
[8] J. Plans, A. Corominas, Modelling and Solving the SALB-E Problem, IEEE,
Piscataway, NJ, 1999, pp. 356–360.
[9] R. Pastor, A. Corominas, Assembly Line Balancing with Work Zone and Station Load
Time Windows Constraints, IEEE, Piscataway, NJ, 1999, pp. 448–453.
[10] S.G. Ponnambalam, P. Aravindan, G. Mogileeswar Naidu, A multi-objective genetic
algorithm for solving assembly line balancing problem, Int. J. Adv. Manuf. Technol.
16 (5) (2000) 341–352.
[11] J. Bautista, R. Suarez, M. Mateo, R. Companys, Local Search Heuristics for the
Assembly Line Balancing Problem with Incompatibilities Between Tasks, Robotics
and Automation Society; IEEE Service Center, Piscataway, NJ, 2000, pp.
2404–2409.
[12] B. Rekiek, P. Lit, F. Pellichero, T. L'Eglise, P. Fouda, E. Falkenauer, A. Delchambre,
A multiple objective grouping genetic algorithm for assembly line design, J. Intell.
Manuf. 12 (5/6) (2001) 467–485.
[13] J. Bautista, J. Pereira, Ant algorithms for a time and space constrained assembly
line balancing problem, Eur. J. Oper. Res. 177 (3) (2005) 2016–2032.
[14] B. Rekiek, A. Delchambre, Assembly Line Design, Springer-Verlag London Limited,
London, 2006, p. 1 online resource.
[15] P.M. Vilarinho, A.S. Simaria, ANTBAL, Int. J. Prod. Res. 44 (2) (2006) 291–303.
[16] Z.-Q. Zhang, W. Cheng, L. Tang, Z. Bin, Ant Algorithm with Summation Rules for
Assembly Line Balancing Problem, first ed., Harbin Institute of Technology Press,
Harbin, P.R. China, 2007, pp. 369–374.
[17] P. Su, Y. Lu, Combining Genetic Algorithm and Simulation for the Mixed-Model
Assembly Line Balancing Problem, IEEE, Piscataway, NJ, 2007, pp. 314–318.
[18] W. Zhang, M. Gen and L. Lin, A Multiobjective Genetic Algorithm for Assembly Line
Balancing Problem with Worker Allocation, Piscataway, NJ, IEEE, 2008, p.
3026–3033.
[19] Z.X. Guo, W.K. Wong, S. Leung, J.T. Fan, S.F. Chan, A genetic-algorithm-based
optimization model for solving the flexible assembly line balancing problem with
work sharing and workstation revisiting, IEEE Trans. Syst. Man Cybern. Part C 38
(2) (2008) 218–228.
[20] Z. Cao, S. Ma, Balancing and Sequencing Optimization of the Mixed Model
Assembly Lines, IEEE, Piscataway, NJ, 2008, pp. 732–736.
[21] J. Bautista, J. Pereira, A dynamic programming based heuristic for the assembly
line balancing problem, Eur. J. Oper. Res. 194 (3) (2009) 787–794.
[22] H.H. Kao and D.H. Yeh, "An Algorithm for Type-I Assembly Line Balancing Problem
8
D. Neumann, J. Keidel
with Optimal Workload Balance," Piscataway, N.J., IEEE, 2009, p. 822–826.
[23] C.L. Khaw, S.G. Ponnambalam, Multi-rule Multi-Objective Ant Colony Optimization
for Straight and U-type Assembly Line Balancing Problem, IEEE, Piscataway, N.J.,
2009, pp. 177–182.
[24] M. Chica, O. Cordon, S. Damas, J. Bautista, Adding Diversity to a Multiobjective Ant
Colony algorithm For Time and Space Assembly Line Balancing (2009) 364–369.
[25] V.L.V. Jonnalagedda, B.M. Dabade, Heuristic Procedure for Mixed Model Assembly
Line Balancing Problem, IEEE, Piscataway, N.J., 2010, pp. 552–556.
[26] O. Brudaru, C. Rotaru, Dynamic Segregative Genetic Algorithm for Assembly Lines
Balancing, IEEE, 2010, pp. 229–236.
[27] Q. Tang, Z. Xiao, Y. Liang, M. Deng, Z. Xi, Novel Approach for Balancing Manual
Automobile Assembly Based on Genetic Algorithm, IEEE, Piscataway, N.J., 2010,
pp. 2028–2032.
[28] M. Chica, O. Cordon, S. Damas, J. Bautista, A Multiobjective Memetic Ant Colony
Optimization Algorithm for the 1/3 Variant of the Time and Space Assembly Line
Balancing Problem, IEEE, Piscataway, NJ, 2011, pp. 1–7.
[29] M. Chica, O. Cordon, S. Damas, Tackling the 1/3 Variant of the Time and Space
Assembly Line Balancing Problem by Means of a Multiobjective Genetic Algorithm,
IEEE, Piscataway, N.J., 2011, pp. 1367–1374.
[30] X. Qian, Q. Fan, Solving Multi-manned Assembly Line Balancing Problem by a
Heuristic-mixed Genetic Algorithm, IEEE, IEEE Computer Society, Piscataway, Los
Alamitos, California, 2011, pp. 320–323.
[31] M.N.I. Sulaiman, Y.-H. Choo, K.E. Chong, Ant Colony Optimization with Look
Forward Ant in Solving Assembly Line Balancing Problem, IEEE, Piscataway, NJ,
2011, pp. 115–121.
[32] X. Hu, Heuristic Algorithm for Two-Sided Assembly Line Balancing Problem with
Multi-Objectives, IEEE, Piscataway, NJ, 2011, pp. 1407–1410.
[33] C. Zhuo, Y. Xu, Y. Tan, A Research of the Stochastic U-type Assembly Line Balancing
Problem, IEEE, Piscataway, NJ, 2011, pp. 874–878.
[34] M.E. Noushabadi, U. Bahalke, K. Dolatkhahi, A.M. Yolmeh, A Simulation
Optimization Approach to Un-Paced Assembly Line Balancing Problem-II with
Additional Reworking Issue, IEEE, Piscataway, NJ, 2011, pp. 1–6.
[35] N.M. Razali, J. Geraghty, Biologically Inspired Genetic Algorithm to Minimize Idle
Time of the Assembly Line Balancing, IEEE, Piscataway, NJ, 2011, pp. 105–110.
[36] M. Chica, Ó. Cordón, S. Damas, J. Bautista, Multiobjective memetic algorithms for
time and space assembly line balancing, Eng. Appl. Artif. Intell. 25 (2) (2012)
254–273.
Robotics and Computer Integrated Manufacturing xxx (xxxx) xxx–xxx
[37] J.M. Nilakantan, S.G. Ponnambalam, An Efficient PSO For Type II Robotic Assembly
Line Balancing Problem, IEEE, Piscataway, NJ, 2012, pp. 600–605.
[38] L.M. Liao, C.J. Huang, A Multi-Agent Based Rescheduling Framework for Mixed-
Model Assembly Line Balancing, IEEE, Piscataway, NJ, 2012, pp. 474–478.
[39] A.A.J. Adham, R.B.M. Tahar, Enhancing Efficiency of Automobile Assembly Line
Using the Fuzzy Logical and Multi-Objective Genetic Algorithm, IEEE, Piscataway,
NJ, 2012, pp. 1–7.
[40] C. Jaturanonda, S. Nanthavanij, S.K. Das, Heuristic procedure for the assembly line
balancing problem with postural load smoothness, Int. J. Occup. Saf. Ergon. 19 (4)
(2013) 531–541.
[41] H. Yu, W. Shi, A Genetic Algorithm for Multi-Model Assembly Line Balancing
Problem, IEEE, Piscataway, NJ, 2013, pp. 369–371.
[42] C. Liu, H. Wen, Hybrid Particle Swarm Algorithm For Assembly Line Balancing
Problem in Complicated Products, IEEE Press, Piscataway, N.J., 2013, pp. 902–905.
[43] U. Saif, Z. Guan, W. Liu, C. Zhang, B. Wang, Pareto based artificial bee colony
algorithm for multi objective single model assembly line balancing with uncertain
task times, Comput. Ind. Eng. 76 (2014) 1–15.
[44] R. Davidrajuh, Solving Assembly Line Balancing Problems with Emphasis on Cost
Calculations: A Petrinets Based Approach, IEEE, Piscataway, NJ, 2014, pp. 99–104.
[45] H. Triki, A. Mellouli, W. Hachicha, F. Masmoudi, Assembly Line Balancing in an
Automotive Cables Manufacturer Using a Genetic Algorithm Approach, IEEE,
Piscataway, NJ, 2014, pp. 297–302.
[46] M. Chica, J. Bautista, S. Damas, O. Cordon, Adaptive IDEA for Robust
Multiobjective Optimization, Application to the r-TSALBP-m/A, Conference
Publishing Services, IEEE Computer Society, Los Alamitos, California; Washington;
Tokyo, 2015, pp. 1013–1020.
[47] C.G.S. Sikora, T.C. Lopes, H.S. Lopes, L. Magatao, Genetic Algorithm for type-2
Assembly Line Balancing, IEEE, Piscataway, NJ; Piscataway, NJ, 2015, pp. 1–6.
[48] P. Sivasankaran, P. Shahabudeen, Literature review of assembly line balancing
problems, Int. J. Adv. Manuf. Technol. 73 (9-12) (2014) 1665–1694.
[49] D. Neumann, J. Keidel, Collaborative formalization and modelling of multi-objec-
tive optimization problems, in: C.F. Oduoza (Ed.), Proceedings of the 25th
International Conference on Flexible Automation and Intelligent Manufacturing:
Designing for Advanced, High Value Manufacturing and Intelligent Systems for the
21st Century: 23-26 June, 2015, Wolverhampton, UK, vol. 2, The Choir Press,
Wolverhampton, UK, 2015, pp. 124–131.