Journal of Manufacturing Systems 40 (2016) 3753

Contents lists available at ScienceDirect

Journal of Manufacturing Systems

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

Method and software application to assist in the conceptual design of

aircraft nal assembly lines
A. Gmez a, , J. Ros a , F. Mas b , A. Vizn a
a
b

Department of Mechanical Engineering, Polytechnic University of Madrid, Jos Gutirrez Abascal, 2, 28006 Madrid, Spain
Department of PLM, Process and Tool Solutions, AIRBUS Group, 41020 Sevilla, Spain

a r t i c l e

i n f o

Article history:
Received 13 November 2015
Received in revised form 12 April 2016
Accepted 21 April 2016
Keywords:
Assembly process design
Assembly line denition
Process evaluation

a b s t r a c t
The decision-making process during the conceptual design of aircraft nal assembly lines requires of
software applications to assist industrialization engineers. This work proposes a method, a supporting
model and ad-hoc software development to address that issue. The proposed method divides the problem
of the assembly process denition into four parts: logistic plan denition, assembly line denition, layout
design and evaluation of complete solutions. The proposed approach is implemented in a prototype
software application that is integrated within a commercial software system widely used in the aerospace
sector. Feasible design solutions, which are created by the user interactively, are evaluated and ranked
with a Fuzzy Logic based multi criteria utility. A case study shows the feasibility of the proposed approach.
2016 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved.

1. Introduction
Literature shows the relevance of the assembly process within
the aircraft lifecycle. In aircraft manufacturing projects, the assembly has a big impact on the production volume and cost, where
the latter is estimated as nearly as the 30% of the total cost
[14]. Ultimately, the execution of assembly processes, manually
or automatically, requires work instructions and/or robot programs. To be able to generate them, designers work collaboratively
along the functional and industrial design process [510]. Collaborative working procedures are developed to integrate teams and
processes. Although, still there are issues related to project management and software tools capabilities and interoperability, the aim
is to create a common industrial Digital Mock Up (iDMU) [512].
The denition of the assembly process can be structured into
three main top-level tasks: create conceptual assembly process,
dene assembly process and develop detailed assembly process [4].
The latter task deals with the denition of the basic assembly tasks,
the documentation to be used by the assembly personnel and the
programs to be used in the automatic machines. The execution of
the two prior top-level tasks is necessary to dene the lowest level
assembly information. Precisely, the create conceptual assembly
process task denes the scope of this work.

Corresponding author. Tel.: +34 617268269.

E-mail addresses: alejandro.gomez.page@alumnos.upm.es (A. Gmez),
jose.rios@upm.es (J. Ros), fernando.mas@airbus.com (F. Mas),
antonio.vizan@upm.es (A. Vizn).

One key objective of the collaborative approach is to integrate

the functional design and the industrial design from the very beginning. This implies, starting the conceptual assembly process design
when the product is still at an early development state [4,7,8,13].
Industrial design decision-making is a challenge when implications
are difcult to be evaluated due to the product and processes low
denition state [13].
Scientic literature acknowledges the need to support industrial
engineers during the denition of the assembly process at any of
the previously mentioned three top-level tasks [1519]; additional
references can be found in [4,13]. In the aerospace sector, at the conceptual assembly process phase, the main objective of the industrial
decision-making process is to evaluate alternatives for the aircraft
nal assembly line (FAL) [4,10,13,20]. Taking these works as an
antecedent, there are four main contributions presented in this
paper. The rst one is the method to guide the designer by splitting
the conceptual assembly denition into: logistic plan, assembly
line, layout and evaluation. The second one is the extension of the
aircraft assembly line conceptual design model to support such a
guidance method. The third one is the enhancement of the cost
model by considering an activity-based approach to evaluate both
transport and assembly cost. The fourth contribution is the adoption of a multi-criteria evaluation method, which considers time,
cost and resource usage-efciency, to support the decision-making
process.
The proposed method aims to support the conceptual design of
aircraft FAL based on static workstations. The following features
characterize this kind of FAL. FALs are mostly designed as synchronous [14], i.e. all the workstations have the same cycle time

http://dx.doi.org/10.1016/j.jmsy.2016.04.002
0278-6125/ 2016 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved.

38

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

Fig. 1. Sequence of tasks of the proposed method.

[16,19]. Assembly operations are performed manually in a high percentage [24], so the number of assigned worker and their hourly
cost are key factors in manufacturing process costing [29,30]. In
addition, a FAL involves parts and subassemblies with large dimensions, which requires a careful distribution of spaces within the
assembly line facilities [10,11].
Fig. 1 shows the method proposed in this work. Four main
tasks are considered to create an assembly process conceptual
design solution: denition of the product assembly structure (or
as-planned structure), denition of the FAL, denition of the FAL
layout and denition of the logistic plan. The last three activities
must be completed following an iterative loop, as it is shown in
Fig. 1. This ensures the feasibility of the solutions, since the different
parts of the problem are coupled. A FAL conguration implies some
resources requirements that have an impact on the FAL layout. The
FAL layout determines space requirements has to be fullled by any
available plant to be part of the network that denes the logistic
plan. On the other hand, the logistic plan represents a big percentage of the total cost [3134]. Therefore, it can be optimized in rst
place, and distributing the industrial workload into a set of plants.
Then, a FAL conguration and layout can be dened considering the
available surface of such plants as space constraints.
A feasible conceptual solution for the FAL is dened each time
the tasks within the loop are completed at least once. Then, the
solution can be evaluated using a multi-criteria Fuzzy Logic-based
system. Several solutions, or variants of a single one, can be generated by executing the previously mentioned tasks. Then, they can
be compared by the evaluation system to nd the best one in terms
of costs, time and efciency in resources usage. Section 4 details
the proposed method and its implementation.
To support the proposed method, an object-oriented Product,
Process, Resources (PPR) meta-model was required. The created
model (see Fig. 2) takes as antecedent, a prior model created to support industrialization conceptual design [4]. This antecedent PPR
model was modied to fulll the requirements of the proposed
method and to be implemented in the commercial software system CATIA/DELMIA V5. Rules were dened to estimate costs and
times parameters. As input initial data, a library of resources and
processes is required for each assembly design specic case. Such a
library is called processes and resources dictionary (PRD). Section
2 details the created models.
There are four main topics involved in this work. The rst topic
is assembly planning. Assembly planning deals with the denition
of the product structure showing the main sub-assemblies to be
built and the execution sequence. The second topic is assembly line

denition. Assembly line denition comprises the denition of the

assembly line concept; this implies dening the workstations, the
joints to be executed in each workstation, allocating resources and
estimating space requirements. The third topic to consider is the
cost estimation for assembly line congurations, which is the purpose of cost models. The fourth topic is multi-criteria evaluation of
assembly line congurations. Multi-criteria evaluation deals with
the evaluation of alternatives when there are several parameters
to be considered. The literature review is included across the paper
within each section.
The paper is structured as follows: Section 2 presents the
developed models. Section 4 shows the proposed method and its
implementation in a prototype application. Section 5 shows a case
study. Section 6 shows the results and the paper ends with the
conclusions.
2. Developed models
This section describes the models created to support the proposed method. Such models are the basis for the development of
a prototype application integrated within a commercial software
system (CATIA/DELMIA V5).
2.1. Object oriented meta model
The information necessary for the dened approach is supported
by a PPR model (Products, Processes and Resources), which was
dened from the initial structure proposed by Mas et al. [4,13]. The
initial structure was modied to generalize the model structure, to
support the detailed properties of a FAL solution, mainly related
to operations and resources, and to enhance the integration with
DELMIA V5 native structure.
The original model was modied to work as a meta-model to
generalize its structure. This way, one unique PPR model is used to
support the information required by any particular case. It works as
a base to develop any particular model dealing with processes and
resources. The proposed method implements a specic model of
processes and resources in the form of a dictionary. Table 1 shows
an extract of the dictionary. Fig. 2 shows an UML class diagram
of the created meta-model. Tables 2 and 3 show an extract of the
specic information for a given case, which belongs to the case
study presented in Section 5.
The nal structure presents some variations from the initial proposal from Mas et al. [4,13]. Most of the properties are not directly
associated as class attributes, but they are represented as objects

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

39

Fig. 2. PPR meta-model with Products (up and left part), and Resources and Processes (right and down part).

of the class MfgAttribute. This approach is necessary for the metaclasses to support that properties may differ from one activity or
resource type to another. The unit of measurement of the attributes
(e.g. D /h, Tm) are also included in the meta-model. The following
two sections describe the main parts of the PPR information model.
2.1.1. Product
The part of the model related to Product supports information
about individual components, industrial nodes and the join of components. The class PhysicalProduct represents a component. The
class Joint has two components of the class PhysicalProduct and
an attribute related to the type of joint. The class SubAssembly
represents components set in groups. Each dened subassembly
is manufactured in a separated assembly line, which is represented
by an IndustrialNode object. The IndustrialNode class supports the
as-planned product structure.
The Joint class is modeled to support the as-prepared product structure, where the precedence in the execution of joints is
specied. Each Joint object is linked to some others, one Next,
one Previous and a number of Simultaneous. These associations

represent the assembly sequence, which denes the input, output

and intermediate products within an IndustrialNode object, and
that conforms the as-prepared product structure.
2.1.2. Processes and resources
Upper level classes model processes and resources. Any assembly operation belonging to an assembly process is represented by
an object of the class ActivityObject. An ActivityObject belongs to
a category represented by an ActivityType object and it has a nonxed number of properties, which are represented by MfgAttribute
objects. Assembly operations may have sub operations that also
modeled as ActivityObject. Resources are structured in a similar
way with the class ResourceObject the class ResourceType and the
class MfgAttribute.
2.2. Processes and resources dictionary (PRD): specic data
dictionary
Every case to be studied requires a specic low-level Processes
and Resources data dictionary, the described meta-model supports

40

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

Table 1
Specic simple and compound activities from a processes and resources dictionary (PRD).
ID

Type

Var.

Req. res.s

Skills

Att.s

Involved operations

Op. type

Op. var.

Subject

R1

Riveting

Manual

Worker

LS

M. oper.

Rivet applier

LS

Auto CNC
machine

Cost
Time
U cost
U time
Nb elem.s
Cost

Worker

Rob. oper.

Time
...
Cost

Drill to 1st D the

witness holes

Drilling

Open

Match holes mesh

Rivet rst wave
Drill to nal diameter
2nd wave
Counterbore 2nd wave
Rivet 2nd wave
Load and Position P1

N
Y

2
12
13

Generic
Riveting
Drilling

Manual
Open
Open

Y
Y
Y

14
15
1

Counter-boring
Riveting
Pos. at jig

Open
Auto
Base comp.

Load and Position P2

Pos. at jig

Added comp.

Assembling

CC

10

Pos. at jig

Added comp.

Auto

R2

MJ3

Assembling

CC

Time
U cost
U time

JT4

Join-up

FF

Specic jig

Cost
Time
U cost
U time

...
Assemble
...
Retire from jigs

Heaviest
(H)
Not H

Set

Table 2
Resources types from a PRD.
ID

Type

Assignment

Addition

Attributes (name-type-unit)

Lifter

Activity

D/A

Worker

Activity

D/A

Capacity
Cost rate
Assignment to activity
Name
Skills()
Speciality
Mean speed
Cost rate
Assignment to activity
Name

this approach. Those specic datasets are named as PRD. To apply

the proposed method, the dictionary must specify every kind of
joint that exists within the product model, as a compound activity
with several involved operations. The class Joint is the link between
the Product section and Processes section of the meta-model.
Table 1 shows an extract of a PRD instance. It shows an activity
of type riveting with two possible variants: manual and automatic.
It shows an activity of type assembly and variant composites, and
with a set of sub operations, two of them of type riveting. Finally,
it shows an activity of type join-up with its corresponding sub
operations, one of them of type assembly.
In the PRD showed in Table 1, the columns represent the properties of the activities. An activity has some properties, which are
explicitly modeled as attributes in the classes ActivityObject and
ActivityType of the meta-model (Fig. 2). For instance, Id (ID), Variant

Real
Real
Real
String
Enumeration
String
Real
Real
Real
String

Tm
D /h

step/s
D /h

(VAR), RequiredSkills (SKILLS) and Subject (SUBJECT) are properties of this kind. Other properties are dened as objects of the class
MfgAttribute. For instance, Addition (A), for controlling how adding
new resources affect to the activity: Linear at speed (LS) when
more resources nish the job faster, Linear at capacity (LC) when
they sum their capacity or Not (N) when only one resource of that
type can be assigned to the activity. The property is Separable (S)
denes if a child activity can be separated from its parent activity.
The property Priority (P) denes the precedence order of an activitys children. The Attributes (ATT.s) column contains properties
that are specic to each PRD.
Conceptually, Resources are modeled in a similar way to Processes. Table 2 shows two examples of resource types (lifter and
worker). The column ASSIGNMENT represents the attribute that
determines the kind of activities to which those resources can be

Table 3
A resource model in a PRD.
ID

Model

Type (s)

Attributes (name-value-type-unit)

TJ5

SinCrane5X11

Transporter jig; lifter

Name
Capacity
Mean speed
Cost
Life expectancy

Sinum Overhead bridge

crane large size
36
1
39 500
3650

CAD path
String
Real
Real
Real
Integer

C:\Doc. . .
Tm
m/s
D
days

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

41

Table 4
A human specialization in a PRD.
ID

Specialization

Skills

Attributes (name-value-type-unit)

SA

Systems assembler

Pneumatic and uids systems installer

Electronic systems installer
Pneumatic and uids systems verier
Electronic systems verier
Electric systems verier

Wage
Extra hours wage
Mean speed
Mean speed (electric systems verier)

assigned. Assignment to activity is the percentage of time that the

physical resource is used by activities. This value is determined
when an assembly line is evaluated, to calculate the efciency of
the resources.
The class ResourceModel models a specic commercial resource
of a ResourceType. A ResourceModel object points to a specic
CAD le containing the geometric representation of the commercial
resource. That geometric representation is needed to dene the FAL
layout. MfgAttribute objects specify the attributes of the resource.
Table 3 shows an example of resource included in the PRD.
The class HR Specialization models a specic kind of resource
of type worker. Table 4 shows the case of systems assembler.
The column SKILLS corresponds with the Skills attribute. This
attribute allows controlling that only those workers having the
skills required by an activity can be assigned as resources to such
activity. The column ATTRIBUTES includes attributes (i.e. wage)
dened as objects of the class MfgAttribute.
Manufacturing plants are considered as a resource type. A
specic plant is modeled as a ResourceModel object and it is characterized by its location, available surface and its cost per square
meter and per day. Those properties are linked to the object as
MfgAttributes objects.
The structure of the meta-model allows assigning several
resources to an activity and assigning resources that are shared
among different activities.
The class TransportNode models the nodes of the logistic network. They can represent either an industrial node of the process
or the factory from where a single component is transported. To
dene the logistic plan, two decisions must be taken for every transport node: selecting an industrial location and selecting a transport
mean to reach the location of the next node in the network. Each
decision is represented by an ACO Node object, which is linked to
the ResourceObject that represents the selected industrial plant or
to the ActivityObjects that represent the transport activities. The
ACO Node class was included to support the meta-heuristic algorithm developed to optimize the logistic plan.

3. Equations of the model

A system of equations is required to analyze any developed
assembly alternative according to the criteria considered in the
method. Equations are grouped into four categories: time, cost, efciency in the use of resources and uncertainty of the estimations.
Each group is essential to evaluate any feasible FAL conguration.
Table 5 shows the nomenclature used in the equations.
Estimating times in assembly lines can be performed by different ways. In deterministic estimations each task has a predened,
xed length, which ts well in highly automated lines. On the other
hand, stochastic approaches consider the variability in any given
activity duration according to a statistical distribution and the consequent cumulative time variation. Dynamic approaches include
reduction of times due to learning phenomena [26]. The present
model follows a stochastic approach to estimate times, variability
is included as uncertainty in times (Eqs. (16)(19)). The inclusion
of learning curves is considered for future developments.

14
20
1
0.7

Real
Real
Real
Real

D /h
D /h
1/s
1/s

Because of the synchronous line conguration hypothesis, the

process cycle time is calculated as the largest value of the workstations cycle time (Eq. (1)), and it sets the duration of all the stations.
Time of stations and operations is estimated using Eq. (2). For
simple activities, it is taken from a list of measured times included in

Table 5
Nomenclature of the equations.
General nomenclature:
TC is cycle time, dened as the time elapsed between two consecutive
assembling products being processed at a same step.
TT is total time, dened as the time to complete the whole considered
assembly process
TNS is the total number of stations within the process
Time equations nomenclature:
NCi is the number of children of station or composed activity i.
Ti is the time, or length, of process element i (e.g. industrial node i, station i,
operation i,. . .).
TMi is the time measured for the type of a given simple operation i under
controlled conditions (i.e. for one element using a mandatory resource with
known speed)
NEi is the number of elements (e.g. number of holes, rivets to perform) of a
given simple operation i.
STi is the speed of the mandatory resource employed to measure the type of
activity i time.
SR1 is the speed of the mandatory resource nally assigned to activity i.
NP1i is the number of activity is children with priority 1 (i.e. without any
precedence constraint with other child)
NHi is the number of shifts set for the station which includes activity i.
TEj is time to end, dened as the time from the beginning of a given child
activity j to its father activity is completed.
NASj is the number of activities included in a precedence sequence after
activity j.
BNSk is the number of stations in the main branch, or maximum number of
station and transport steps of the process branches until the completion of
the operations within industrial node k.
NPrk is the number of direct precedent nodes of industrial node k.
NSkl is the number of stations to be completed within node k after the
entrance of the output product of node l.
NoSl is the number of stops set for the transport of output products from
node l.
Costs equations nomenclature:
C is the total cost of the considered assembly process.
CnoT is the total cost excluding the transport of products between nodes.
Ci is the cost of industrial node i.
CoTi is the cost of the transport from industrial node i to its subsequent node.
NSi is the number of stations within industrial node i.
CAj is the cost of station or activity j.
NRk is the number of resources directly assigned to activity k (or to activities
within node k).
CHr is the hourly cost of resource r.
Wr is the wage of human resource r.
WEr is the wage for special shifts of human resource r.
CSr is the remaining amortization total value of resource r.
ELr is the remaining life expectancy of resource r.
Efciency equations nomenclature:
E is the overall efciency.
Er is the efciency associated to resource r.
Tar is the time of an activity a, which resource r is assigned to.
Uncertainty equations nomenclature:
UTT is the uncertainty in the total process time estimation.
UTC is the uncertainty in the cycle time estimation.
UTi is the uncertainty in an activity i time estimation.
UTEj is the uncertainty in an activity j time to end of its father estimation.
UC is the uncertainty in the process cost estimation.
UUCi is the uncertainty in a process element i cost estimation.

42

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

Fig. 3. Example of industrial nodes network.

the PRD, and it is adjusted with the mean speed of the key assigned
resource (i.e. the rst resource in the denition at the PRD); the
number of elements involved and the numbers of work shifts of the
workstation. Time of workstations or composed activities is given
by their critical path time, which is calculated using the variable
time to end, dened as the time from the beginning of an activity to
the end of its father activity (Eq. (3)). The time of a composed activity is the maximum value of time to end between its child activities
with Priority 1.
TC = MAXiTNS T (i)

Ti =

(1)

TMi NEi (STi /SR1i )/NHi ,

NP1i

MAXp


TEj =

(TEp ),

NASj = 0
NASj

Tj + MAXn

f =FAL

(TEn ),

(3)

NASj > 0

) TC

MAXlNPrk (NSkl

(2)

NCi > 0

Tj ,

TT = ( BNSf 
BNSk =

NCi = 0

+ BNSl + NoSl + 1)

(4)
(5)

Total process time is the time required to assemble one unit. It

is given by the product of cycle time and the maximum number
of stations in the main branch (BNS) of the nal node. BNS is the
maximum number of workstations and transports of the logistic
network branches that end in a node. This value is calculated using
Eq. (5). Extra stops in the transport count as additional transport
activities. Fig. 3 shows an example with four industrial nodes, a, b,
c and the nal one, FAL. BNS for the nal node would be given by
Eq. (6):
BNSFAL = MAX{(3 + BNSa + 0 + 1), (3 + BNSb + 0 + 1), (1 + BNSc + 1 + 1)}

(6)
Cost is another essential aspect in Assembly processes denition
and evaluation. A brief review of costs estimations approaches is
provided to facilitate a better understanding of the way followed
to include costing in the proposed model.
There are two main perspectives for cost modeling and estimation: based on production volume and based on activities (ABC)
[29]. While production based models separates costs in variable
costs, proportional to the production volume, xed costs and overhead cost; ABC assigns cost to activities, based on the cost driver of
each activity, which requires estimating the usage of the resources
assigned [29]. Different research works dealing with assembly
alternatives evaluation adopted the ABC approach.
Amen [30] considers the resources hourly cost and assigns
resources with similar hourly cost to the same workstations. The
objective is to minimize the assembly line cost by minimizing the
cost of inactive times. Baud et al. [31] propose an ABC model to
evaluate solutions for the logistic strategy. Cost of transport activities is calculated using the products physic volume as a resource
cost driver, and processing time is used as cost driver for assembly
operations. Kampker et al. [32] and Tang et al. [33] include the Net
Present Value concept in their models, to evaluate, respectively, the

assembly process and the logistic chain. They both use ABC models,
but they also take into consideration the investments required at
each solution.
Some other works also implement simplied ABC models,
where cost is calculated as the sum of the hourly cost of the
resources assigned to every workstation multiplied by the cycle
time and the resource assignment percentage [25,27,34].
The proposed cost model was designed following an activitybased approach. It considers activities executed for every product
unit produced. This approach allows calculating the total cost per
day and per assembled aircraft. It also allows future improvements,
such as implementing learning curves and including additional
activity types.
The cost of the whole assembly process is estimated as the sum
of all the industrial nodes costs (Eq. (7)). The cost of each industrial
node is calculated by two different ways. Eq. (8) includes costs of
the involved workstations and the transport to the next node. The
cost of a workstation or an activity is calculated by multiplying the
directly assigned resources hourly cost by the cycle time, plus the
costs of the children activities (Eq. (9)). A resource hourly cost is
given by Eq. (10), considering the expectancy of life, the number of
shifts that each resource is assigned to and the different wage for
special shifts in the case of humans.
That way to estimate an industrial node cost is used when the
denition of its involved assembly line is still in progress, so the
designer may have an idea of the implications of his decisions (e.g.,
assigning a resource instead of another one). However, the obtained
value is inaccurate, since the sharing of resources between activities
and workstations is not included in the calculations at this point in
time. The nal set of resources units is dened at later steps and
then the cost is calculated again.
The second way estimates an industrial node cost by using costs
of the actually assigned resources, as Eq. (11) sets. This way can be
followed only when the nal resources conguration is achieved.
C=

NI

i=1

Ci

Ci = CoTi +
CAk =

NRk


CHr =

r=1

j=1

CAj

(8)

TC CHr +

NCk
l=1

CAl

Wr + (NH 1) WEr ,

human resource

NH CSr /ELr ,

other case

Ci = CoTi +
CnoT =

(7)

NSi

NRi

r=1

TC CHr

NI NRi
i=1

r=1

(9)

(10)

(11)

TC CHr

(12)

Efciency in the use of assigned resources is considered in the

evaluation of alternatives. One resource usage efciency is its percentage of direct usage time. Total resources efciency is dened as
the sum of all resources usage efciency, balanced with their hourly
costs (Eq. (14)). Introducing Eq. (13) and considering the total cost
without transport costs, Efciency can be expressed with Eq. (15).

NAr

Er =
E=
E=

T
a=1 ra
TC

NR

r=1

(13)

NR

[Er CHr ]/

NR NAr

 Tar CHr
r=1 a=1

CnoT

r=1

CHr

(14)

(15)

Calculating uncertainty is necessary to give users an idea of the

estimations accuracy and to include stochastic variability in times

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

and costs computing. Eqs. (16)(19) are used to calculate uncertainty in times, according to the propagation rule.
A single activity time uncertainty requires values from the measured activities time list included in the PRD (Eq. (16) if the number
of children NCi is zero). For composed activities, a recursive procedure is used to calculate the cumulative uncertainty of the critical
path length (Eqs. (16) and (17)). Uncertainty in cycle time is considered as the maximum time uncertainty between the workstations
(Eq. (18)). Eq. (19) gives uncertainty in total time. Uncertainty
in transport times is not considered in the model, as the logistic
network is dened to t the cycle time resulting of dening the
assembly line processes.

UTi =

UTMi NEi (STi /SR1i )/NHi ,

NP1i

MAXp

UTEj =

UTj2 + UTEk2 ,

UTC = UTl ,
UTT =

(UTEp ),

l = s,

BNSf 

f =FAL

NCi = 0

(16)

NCi > 0
MAXsNAS (UTEs )

k = s,

(17)

MAXsNS (Ts )

(18)

UTC

(19)

Total cost uncertainty is calculated using the cost uncertainty of

industrial nodes (Eq. (20)). Eq. (21) gives the cost of an industrial
node introducing the term C, dened as the hourly costs whose
uncertainty is not considered. Replacing the value of C, given by Eq.
(21), and considering that uncertainty in transport costs is null, Eq.
(23) is obtained.
The cost uncertainty of any activity is calculated using Eq. (24).
It considers all the involved resources (i.e., directly assigned or
assigned to any child) and introduces the percentage that every
resource is shared the activity. This value is taken as 100% for stages
previous to dene the nal resources conguration.

NI

UC =

i=1

UCi2

(20)

Ci = CoTi + TC C

(21)

UCi =
UCi =

CoTi2 + C2 UTC 2

C CoT 
i
i
TC

(22)

UTC

(23)

UCj = UTC

NRj
r=1

(CHr PSrj )2

(24)

Eq. (15) is needed to obtain the equation to calculate the uncertainty in the total efciency. Applying the propagation rule, and
considering that a resource hourly cost uncertainty is null, then Eq.
(25) is obtained.

UE =

2
[E UC] +

r=1 a=1
NR

NAr

[UTar ]2 [CHr ]

CnoT

(25)

4. Proposed method and its implementation into a

prototype application
The proposed method can be described as a sequence of steps
needed to dene each feasible conceptual solution for the FAL.
Such steps are represented in the diagram showed in Fig. 1: dene
product as-planned structure, dene logistic plan, dene FAL conguration and layout. Feasible solutions can be evaluated using a
multi-criteria evaluation system. This section explains each step
more deeply, paying special attention to the denition of assembly
lines and the evaluation of solutions.

43

Taking the developed models as input, the proposed method is

implemented as a prototype application in the commercial software system CATIA/DELMIA V5. The application comprises a set of
VBA macros. Files of type: CATPart, CATProduct, CATProcess and
xls; are used to extract and store information. The processes and
resources dictionary (PRD) is stored in a Microsoft Excel le. The
application manages objects that conform to the PPR meta-model
(Fig. 2) and to the native CATIA/DELMIA data structure (i.e. part,
product, activity and attribute). Each feasible solution is stored in a
CATProcess le. The values that characterized each feasible solution
are located under the parameters node within the corresponding
element of the PPR tree. A macro exports such values into an Excel
le, where the Fuzzy Logic based utility for the nal evaluation is
implemented. A programmed function allows opening the document and executing the evaluation.
4.1. Dene aircraft as-planned structure
One rst step to initiate the conceptual design of an assembly system is to dene the relationship among components in
an assembly [16]. The starting point is the product functional
structure named as-designed. A common approach is to consider
that the following two steps can conceptually dene the assembly process: rstly, establishing a product structure with the main
sub-assemblies (i.e. the subgroups of components that must be
assembled together); and secondly, dening the sequence in which
the components are joined to conform the product [2,4,1324].
Usually, the concept of joint is used in the denition of an assembly sequence [1624]. Although, some authors approach is based
on the analysis of feasible product disassembling ways to nd the
assembly sequence [16]. From the very beginning, the joint concept
was adopted in this work and in its direct antecedents [4,10,13,20].
In the aerospace sector, at conceptual assembly process phase,
the as-designed product structure contains major structural components (e.g. center fuselage, rear fuselage, center wing box, right
wing, etc.). The prior two steps approach involves the creation of
two different product structures. The as-planned structure denes
the main sub-assemblies or industrial nodes. The as-prepared
structure incorporates an assembly sequence for the joints to be
executed [2,4,13,20]. Theoretically, several as-planned structures
can be derived from a single as-designed structure.
However, concurrency work between product functional design
and industrial design harmonizes aircraft design solutions and
industrialization manufacturing solutions. As a result a combined
as-designed as-planned product structure is agreed. Similarly,
depending on the sequence to execute the joints, several asprepared structures can be derived from a single as-planned
structure. At this phase, an industrialization engineer takes the
agreed as-planned product structure, denes feasible as-prepared
structures and evaluate them to make a decision on the alternative (i.e. the pair of as-planned and as-prepared structures) to be
adopted. Time and cost are the main criteria to make the decision,
and they depend heavily on the logistic plan, on the assembly lines
layout and on the resources use efciency.
Automating the generation of as-planned and as-prepared
structures and the selection of the best alternative is the focus of
extensive research [16]. For instance, Dmoly et al. [21,22] propose
a collaborative framework to generate all the possible assembly
sequences, considering the links between components and their
type, and to select one that ts with the maximum cycle time. The
approach adopted in this work, makes use of rules to dene for
each major component its feasible joints. The user denes the asplanned structure interactively and such rules limit the selection
to feasible components for each joint being dened.
According to the proposed method, in this step the major components of the aircraft, that conform the functional as-design

44

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

product structure, are grouped into subassemblies that have to be

made to build the complete aircraft. This grouping creates the asplanned structure, constituted by a number of industrial nodes,
each one of which involves the manufacturing of a dened sub
assembly. The total volume and weight of each subassembly are
estimated, since those data are key when selecting the means
needed for their transport between plants or within a plant. An initial implementation of this part of the proposed method, illustrated
with a case study, can be found in [20].
The input information to dene the as-planned structure is
the list of components and the joints between them. Mas et al.
[13] and Gmez et al. [20] show a rst version of the prototype
development, with a top-level description of the assistant tool. The
as-planned conguration represents the industrial breakdown and
indicates the grouping of components into assemblies (e.g. industrial nodes) that will be created in the manufacturing units. The user
executes the denition of joints and the as-planned structure interactively, supported by a graphical user interface (GUI), included in
the developed prototype, to create and modify industrial nodes and
functions that automatically refreshed the structure of the product
in CATIA [20].
4.2. Dene supply chain and logistic plan
Firstly, it comprises the allocation of a manufacturing plant to
each industrial node of the as-planned structure. Secondly, it comprises the selection of the transport means between plants. If the
transport mean fullls the requirements of load capacity and space
then more than one product unit can be transported at a time.
Based on Ant Colony Optimization [23], a meta-heuristic algorithm
is proposed to optimize the logistic plan.
The rst implementation of the logistic plan denition comprises the user manual selection of one plant for each industrial
node dened in the as-planned structure and then the selection of
the transport option between every pair of plants directly linked.
A selected transport mean must fulll requirements imposed by
weight and dimensions of the product, and be available as an option
for the origin and destination points. This rst version of the logistic
plan denition was implemented in the rst prototype application presented by Gmez et al. [20]. The implementation of the
meta-heuristic algorithm developed to optimize the logistic plan is
currently under development.
4.3. Dene nal assembly line (FAL)
In general, aircraft nal assembly lines can be implemented in
two main ways, as a set of static workstations or as a continuously
moving line. In the former case, components arrive to each workstation to be assembled by executing a set of basic operations;
each workstation generates one of the subassemblies dened in
the product as-planned structure. The resulting subassemblies are
moved from one workstation to the next one after completing all
the tasks during the workstation cycle time. When all the workstations have the same cycle time, as it is the case in the aeronautical
sector, the assembly line is named synchronous [14,16,19]. To
assign and order tasks among workstations, satisfying precedence
constraints and objective functions, is the purpose of Assembly
Line Balancing (ALB). Extensive research exists addressing different
ALB problems depending on the problem hypothesis and solving
method [14,16,2427].
The denition of a FAL is based on requirements related mainly
to delivery plan, budget, logistic plan, technologies availability and
space availability. The conceptual design contains the workstations,
their precedence constraints and tasks assigned to workstations
due to technology and resources constraints [4,10,13,14]. Additionally, each type of joint, (e.g. fuselagefuselage), has a set of

tasks to be executed [13]. When adopting a deterministic approach,

each tasks has a predened time. The production of major components is usually carried out in sites located apart, which needs also
considering transport requirements [13,19,20]. The traditional ALB
approach is not suitable to dene the concept of an aircraft FAL system, where key features, such as the number of workstations, have
to be dened attending to technological criteria rather than using a
mathematical function to minimize a specic parameter [14]. The
approach adopted in this work is the development of a guidance
tool, which requires knowledge from the designer as input and provides data to assist during the decision making process. The tool
guides and assists the designer to dene feasible static FAL congurations and it facilitates the evaluation of the feasible solutions.
In this sense, the proposed approach is different from a traditional
ALB approach. Similarly, this approach follows the trend of a higher
level of interaction between users and applications to allow more
exibility during the denition of the assembly process [15].
The as-planned structure is the input to assembly lines definition. As starting hypothesis, each industrial node requires an
assembly line, which requires at least a workstation. Each industrial node requires the denition of its corresponding assembly line.
The FAL is a specic type of assembly line that produces the nal
aircraft as output.
Fig. 4 synthesizes the procedure of dening assembly lines, as
it is proposed in the presented method. The procedure follows a
sequence depicted in the left side of the gure. The central area
shows the results of each step in the process structure and the left
side outlines the impacts of those results on the layout and the line
digital representation. Each step will be explained further in the
following subsections. In Fig. 4, Joints 9 and 11 were set in the rst
place of the sequence, as simultaneous; Joints 8, 10 and 15 were set
in a second place.
In a rst step, considering the as-planned structure, the industrialization engineer creates an as-prepared structure, by dening the
assembly sequence. This implies ordering the joints to be executed.
The as-prepared structure implies new intermediate subassemblies
to be built at each plant.
Then, the line operations conguration must be dened. A
default conguration is automatically created: each group of simultaneous joints is assigned to one workstation. Depending on the
type of joint, a predened set of tasks is created. Repetitive tasks
are automatically deleted. The industrialization engineer conducts
the nal conguration interactively, allocating operations in different workstations and creating new stations or activities if it is
necessary. In Fig. 4, three workstations were dened. The second
and third activities associated with Joint 9 were deleted and the last
ones of Joints 9 and 11 were reallocated in the third workstation.
Third, the user assigns resources to the activities to be executed
in each workstation, during the resources assignation process the
suitability of each resource is checked. It denes a complete set
of involved products and resources, so the layout can be dened,
which is actually considered part of the next stage within the
methodology. A nal step includes checking that space requirements of the line fulll the plant available surface. It is possible to
repeat previous steps to improve a solution or generate variants.
4.3.1. Assembly sequence denition
The rst step to design an assembly line is to dene the assembly
sequence, or when the components will be added to the product
being assembled. A typical way to do it is by dening a sequence
for the execution of the joints between pairs of components. Such a
sequence denes the real subassemblies that will be built. Fig. 4, at
its top, shows how a set of components are planned to be assembled
by sequencing their related joints (811 and 15). In this case, wings
are tted to the wing box, stabilizers to the fuselage, and nally
both sub-assemblies are joined together.

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

45

Fig. 4. Conceptual ow diagram, process structures created and Computer Aided views of the assembly line denition.

Fig. 5 shows the GUI form designed for this stage of the method.
The left list box shows the parts and sub-assemblies that have to
be joined. Joints can be moved up and down within the sequence.
The joint information box shows the intermediate products that

would be generated. They are calculated when clicking on the Set

Sequence button.
The adopted approach supports the simultaneous execution
of joints. In the depicted as-prepared example, rst, Joint 10 and

Fig. 5. Graphical user interface to dene the assembly sequence or as-prepared product structure.

46

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

Fig. 6. GUI to dene the assembly line workstations and results.

Joint 15 are dened as simultaneous, and then Joint 8, Joint 9 and

Joint 11. Initially, that implies a processes list with at least two
workstations, one for each simultaneous joint. Although this situation requires a further interactive conrmation by the designer,
since some simultaneous joints could be conducted in the same
workstation. The application automatically creates the as-prepared
product structure, after clicking on Apply. When clicking on
Create, the application creates a process structure, with the workstations according to the joint execution sequence and with their
corresponding product components associated.
4.3.2. Workstations denition
The starting point in the workstations denition is the asprepared product structure dened in the previous step. A process
structure is drawn from the as-prepared structure. Each joint corresponds with a basic assembly process of type structure (e.g.
ActivityType), which, in turn, has a set of tasks/operations to be executed depending on the attributes TypeOfAct and Variant (Fig. 2),
i.e. join up, assembling and composites (Table 1). The PRD contains
all the initial information about the basic assembly processes and
their operations.
Taking the initial process structure as input, the industrialization designer interactively continues with the denition of the
workstations. To do so, the designer uses the developed GUI and
make decisions, such as moving an activity to the next or the previous workstation or creating a new workstation. The developed GUI
shows the impact of each decision over the cycle time (Fig. 6).
The developed application creates a CATProcess le that stores
the initial process structure by assigning to every Joint (or simultaneous Joints) one single workstation. The initial workstations are
created with the assembly activities and operations that conform
to the PRD. If two joints are executed simultaneously, is normal
that activities with shared components are duplicated (e.g. activity
Load fuselage in jigs would appear twice in a workstation created for two joints involving the fuselage). When activities with
the same type and subject are detected within a workstation, the
user is asked if one of them must be deleted. Fig. 6 shows the GUI
with the dialog box.
New workstations and operations can be included by clicking
on the New after buttons. Operations can be assigned to different workstations by clicking on Up or Down arrow buttons. Only
operations whose property IsSeparable has the value Yes can be
changed. Precedence constraints are respected, when an operation

is assigned to the next workstation, all the activities that have to be

executed after it will be automatically reassigned too.
Fig. 6 shows that a third workstation was created as the last one.
Initially, a new workstation presents a time with a value equal to
zero. That is because initially, a newly created workstation does
not have any assigned activity or resource. The value of the Cycle
Time at plant is the maximum value for the dened workstations,
and the Cycle Time of Assembly Process is the maximum value for
the dened industrial nodes. At this stage, the former time can be
reduced by manually balancing the involved operations. The times
are calculated by using the reference time associated to each basic
task or operation in the PRD. In the following stage, once resources
are assigned to activities, the times are calculated using the speed
of the nally assigned resources. When the Apply button is
clicked, the initial CATProcess document is updated to store the
dened process structure. The initial manual balancing conducted
by the designer is relevant at this early assembly line conceptual
design phase. Later, discrete event simulations to tune the workstations and FAL balancing are conducted during the denition
phase [13].
4.3.3. Resources and Operations Management
After the denition of the workstations, there are two pending
aspects. First, no real existing resource is still assigned to any activity. Second, there could be activities without a dened variant in
the PDR, i.e. manual or automatic.
The PRD contains a number of resource models, and each one
belongs to one or more resource types. Every activity, depending on
its type, requires one or several resources of some specic type. For
each required resource type, at least one resource model has to be
assigned to each activity. Depending on the property Assignment,
more than one model could be assigned. That will increase either
the total resources speed or the total capacity. Human resources are
considered in a similar way: with specializations instead of models,
and with required skills instead of a required type. To carry out
an activity, it could happen that more than one skill is required
from a worker. For some activities, resources must fulll capacity
requirements (i.e. a lifter must have a loading capacity bigger than
the product weight).
Based on the information dened in the PRD, the application
makes an initial automatic allocation of resources. For all the activities, when the PRD contains a single model of a resource type
required by an activity, then the model is assigned to the activity.

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

47

Fig. 7. GUI to select variants of activities and assign resources.

After this initial automatic step, activities having resources pending of allocation are displayed with the tag NAY!! (for Not
Assigned Yet) and activities without dened variant display the tag
Gen!!.
Fig. 7 shows the GUI designed to assist the designer with the
variant selection and the resources assignment. The list in the left
side allows exploring the process structure. For every activity, the
variant can be selected, and the resource models or specializations
assigned from one list. Properties of activities and resources can be
checked to evaluate different scenarios.
The designer can automatically assign the best resource, the best
activity conguration (i.e. the set of assigned resources and the variant) or the best conguration of the whole process, by clicking on
the three respective buttons. The criteria that denes best can be
switched from time to cost with the buttons (D and t) located in
the left side below the activity list.
The dened conguration is checked when Apply is clicked. As
a result of the checking, when different assigned resource models
are found, for activities that could share such a resource (i.e. the
activities are in the same workstation and not overlapped in time),
the designer is asked to conrm if a single model can be selected
and shared. The dened rule is: choose the cheapest resource
model that fullls the capacity requirements. The FAL conguration created by the designer with the assistance of the developed
application is stored in a CATIA/DELMIA V5 le of type CATProcess.
The designer can generate several FAL conguration alternatives
and each one is stored in a CATProcess le.

4.4. Dene FAL layout

The industrialization engineer denes the FAL layout interactively by graphically positioning the different resources, jigs and
tools and industrial means assigned at the prior stage. Assigned
resources are automatically loaded into the active document of
the commercial software system (DELMIA CATProcess document).
The nal layout determines the space requirements needed for the
assigned plant. This interactive approach is valid for this kind of
xed-position assembly system, when considering the layout of a
dynamic job-shop assembly environment then a heuristic based
solution could be more appropriated [28].
With the assistance of the application, the 3D geometric representation of each resource can be loaded into the CATProcess
document and the designer can congure the assembly line layout interactively, using functionalities available in CATIA/DELMIA
to position elements within a geometric space.

4.5. Evaluate FAL solution

As it was previously mentioned, the selection of an aircraft FAL
concept depends on several parameters or decision criteria. In this
work, to support the decision-making process three parameters
are considered: time, cost and resource usage-efciency. There are
several methods to address the multi-criteria decision making situations [35]. Among them, Fuzzy Logic (FL) has also the advantage of
supporting uncertainty and inputs considered vague or ambiguous.
It accepts that an element can be member of a set with a membership degree lower than 100% (e.g., an assembly process solution
can be good at cost in a 66% and very good in a 34%). The application of FL to solve multi-criteria situations in manufacturing is
quite extensive, for instance in scheduling and planning [36], and
in supply chain management and supplier selection [37,38]. Fuzzy
Logic was the method selected for this work.
Once the industrialization designer creates feasible conceptual
FAL design solutions, this last step allows evaluating each created
solution and gives a score to each of them. Then, the set of solutions
can be compared and ranked. A FL-based utility was developed to
perform this evaluation, attending to parameters regarding time,
cost and resources efciency.
Fuzzy Logic (FL) is the approach adopted to develop the evaluation solution. This is an example of one of the implemented logic
rules: IF time is good AND cost is very good AND efciency is good,
THEN the solution gets a score of 8 out of 10. Table 6 shows an
extract of the implemented logic rules.
For each solution, three process parameters are evaluated: cycle
time, cost and efciency of the assigned resources. Each input is
evaluated using ve fuzzy sets that characterize its domain, from
A to E, being A the best qualier and E the worst one. The functions have a triangular shape. Eq. (26) denes the mathematical
membership function of one fuzzy set, of parameters a, b and c. The

Table 6
Rules table of the FL based system.
Time

Cost

Eff.

A
A

A
A

A
B

9
8

D
D

C
C

B
C

6
1

E
E

E
E

D
E

1
1

48

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

the interval [ai ,ei ] for the inputs of all the evaluated alternatives,
for a better differentiation between them.

Fig. 8. Membership functions of the input parameters fuzzy sets.

domain of each input parameter is dened by ve domain parameters (i.e. a, b, c, d and e), as there are ve sets of triangular geometry
(Fig. 8). Sets A and E are different, with a constant membership
value of 1 for input values smaller than a or greater than e. Since
the best efciency is the highest one, the distribution of the fuzzy
sets along the input axis changes for parameter efciency, following the sequence: E, D, C, B, A; instead of the opposite sequence used
for cost and time.

Sabc (x) =

(x a)/(b a), x [a, b)

(c x)/(c b),

x [b, c]

0,

any other case

et = Ncyc0 tcycobj

(27)

at = 40%et

(28)

ec = cos tMAX

(29)

ac = 30%et

(30)

eeff = 5

(31)

aeff = 50

(32)

Usually, single deterministic values are used as input to FL systems, and memberships are given by mathematical functions, as the
one expressed in Eq. (26). However, due to the statistical nature of
most of the parameters used in this work, where the parameter
has an expected nominal value and a variance, the membership to
a fuzzy set Sabc is calculated by using Eq. (33), where Sabc is the
conventional membership expression of Eq. (26). E and  are the
expected value and variance, and p is the probability density function, as Gaussian, given by Eq. (34). Using its deterministic value,
an input parameter could be member at the most of two fuzzy sets,
but considering its statistical distribution, it could belong to more
fuzzy sets (Fig. 9).

(26)

Every set has the same amplitude, so two parameters are enough
to dene a domain. Eqs. (27)(32) dene standard values for the
extreme parameters of the domains time, cost and efciency (i.e.
et , at , ec , ac , eeff and aeff , respectively). tcycobj and costMAX are initial
data, specic of each assembly case. The latter refers the maximum
allowed value of cost. tcycobj represent the objective cycle time for
an adequate production rate. Ncyc0 refers a reference value for the
total number of workstations within the process, set as if there was
one only workstation for each joint.
The value of domain parameters can be changed to improve the
rating accuracy. The extreme parameters et and ec will always adopt
the values given by Eqs. (27) and (29), but at and ac can be modied
for a better adjustment. Both parameters aeff and eeff may also vary.
It is highly recommended that every mean value Xi is included in

S

abc

(E, ) =
a

Sabc (x) p(E, , x) dx

1
p(E, , x) = ( 2) e(xE/)

(33)
(34)

The domain total length is not relevant for scoring an alternative,

as membership values of distributions affecting intervals (,ai ]
and [ei ,) are calculated considering a constant value of 1 for the
membership functions of sets A and E at them.
The input membership values activate rules that dene the
membership to output fuzzy sets, according to the Mamdami
approach. There are nine output fuzzy sets representing scores 1
to 9 over 10. Table 6 is an extract of the rule base. Each row associates one combination of input parameters rates with one score
e.g., second row means:
If time is rated as A And cost as A And efciency as B Then nal
score is 8.
The activation degree of a rule is given by the t-norma of the Minimum, which sets it as the smallest membership value of the input

Fig. 9. Membership functions for parameter time and Gaussian distribution (E,) = (40, 2.5) h, being a and e 19.2 h and 48 h.

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

49

Fig. 10. Output fuzzy sets and resulting joint membership function.

parameters to the sets involved in the rule. For the rule described
above, if time, cost and eff. had membership values of 0.7, 0.8 and
0.3 to their fuzzy sets A, A and B, the rule would be activated with
a degree of 0.3.
Since there are three input parameters with ve sets, 125 rules
are required, but some simplications can be done. For instance,
time belonging to the worst fuzzy set always implies the worst
output set, 1, as it represents the delivery time, which would be
exceeded. That denes 25 rules at one time. The nal score is
ranked between 0 and 10 (Fig. 10). The nal joint membership
function, usually called Defuzzy function, is built by using the
Sup-Min composition principle [39]. The nal score is calculated

as the position of the center of gravity of the area under such a

function.

5. Case study
A case study was developed to prove rstly the effectiveness of
the proposed method. Secondly, the execution of the case study
allows validating the developed prototype application. The prototype application aims assisting the industrialization designer in
the decision-making process to dene aircraft FAL conceptual solutions. The following subsections illustrate the case study.

As planned
Fuselage

LW
RW

V_Stab

LW
RW
WB

H__Stab

As prepare
Co
omplex Assembly n1

LW
RW
WB
RFBL
RFBU
RFF
CFL
CFU
FFL
FFU
FFB
VSU
VSR
VSF
VSM
HSRL
HSRR
HSRF
HSRM
HSLL
HSLR
HSLF
HSLM
HSCB

H_Stab

Compllex Assembly n2

Fuselage

RFBL
RFBU
RFF
CFL
CFU
FFL
FFU
FFB
VSU
VSR
VSF
VSM
Fig. 11. As Planned view and As Prepared view of the nal industrial node, dened for the case study.

WB

HSRL
HSRR
HSRF
HSRM
HSLL
HSLR
HSLF
HSLM
HSCB

V_Stab

50

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

Fig. 12. Gantt and 3D CATIA V5 representations of both studied alternatives.

5.1. Initial conditions: product, joints and PRD

The simplied product for the case study is an aircraft prototype
based on the AIRBUS A400M model, with 24 major components and
with parts belonging to the fuselage, empennage and wings. In total,
the prototype has 23 joints. A processes and resources dictionary
(PRD) was created to specify a base of activities and resources types.
Tables 13 show an extract of the PRD.
5.2. As-planned and as-prepared product structures
The as-planned product structure was dened with four subassemblies: fuselage, vertical stabilizer, horizontal stabilizer and
complete aircraft; as Fig. 11 shows. For the complete aircraft subassembly, the as-prepared product structure was set by sequencing
the joints in two steps. Joints between the complete fuselage, the
vertical stabilizers and the wing box (highlighted in Fig. 11) would
be executed simultaneously in a rst step, and the wings and horizontal stabilizer would be added in a second step.
5.3. Assembly line congurations
Considering the as-prepared structure, two alternatives were
studied for the aircraft FAL. The rst alternative was dened
with two workstations and the second one with three workstations. In the second case, the third workstation comprised testing
operations to be carried out after executing the main assembling
activities. The alternative with three workstations was dened to

reduce the workstation cycle time. The cheapest resources congurations were adopted for both alternatives. The layouts were
designed using the developed GUI. Additionally, standard DELMIA
V5 functions were used to generate Gantt diagrams. Fig. 12 shows
the Gantt diagrams and the 3D representations of the studied
alternatives.
A second study was executed to evaluate the impact of selecting
automated or manual variants for the drilling activities. Table 10
shows the properties of both compared alternatives; such data
(i.e. resource mean speed, resource cost and worker cost) are
included in the case PRD. The next section presents and discusses
the obtained results.
6. Results and discussion
As time related hypothesis, the FAL is synchronous, consequently, all the workstations have the same value as cycle time
target. The target for the workstation cycle time is equal to the
takt time. The FAL cycle time is the sum of the cycle times of all
the workstations. Every workstation that is precedent of another
workstation includes a transportation operation as last task.
In the rst part of the case study, from the results (Table 7), it
can be observed that including a third workstation causes a reduction on the workstation cycle time. Consequently, the takt time
would be also reduced. Fig. 12 shows both Gantt diagrams, with the
workstations highlighted for each alternative. Obviously, an extra
workstation has an impact in the resources needed, some resources
has to be duplicated, some new jigs might be required and the FAL

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

51

Table 7
Results for rst studied alternatives.
Alternative

Surf. (m2 )

Cost (D /cycle)

UCost

tcyc

tproc

2 Workstations
3 Workstations

6052
11 996

171 018.63
234 087.20

10 162D
13 453D

10.07 h
8.53 h

20.14
25.59

occupies a larger surface, and so requesting a larger plant, all of

which results in a rise in cost. The developed application calculated
the parameters according to the proposed models and input them
into the evaluation utility. Fig. 13 shows the activation degrees of
the fuzzy sets; Table 8 contains the activated rules, highlighting
the taken activation value for each output set and Fig. 14 shows the
nal defuzzication function, all for the rst alternative.
The above results depend on the input fuzzy sets denition
parameters: an alternative parameter will be better rated if set A

UTime

EffT

2.52 h
0.5 h

18.4%
28.6%

Table 8
Activated fuzzy rules function for the rst alternative of the rst study.
A
A
A
A
A
A
A
A
A
A
A
A
A
A
B
B
B
B
B
B
B
B

A
A
A
A
A
B
B
B
B
C
C
C
C
C
A
A
A
A
A
B
B
B

A
B
C
D
E
A
B
D
E
A
B
C
D
E
A
B
C
D
E
A
B
C

9
8
6
6
5
8
7
6
5
7
7
6
5
4
9
8
7
5
5
8
7
7

0.001
0.039
0.039
0.039
0.039
0.001
0.064
0.094
0.054
0.001
0.064
0.094
0.094
0.054
0.001
0.039
0.039
0.039
0.039
0.001
0.064
0.454

B
B
B
B
B
B
B
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C

B
B
C
C
C
C
C
A
A
A
A
A
B
B
B
B
B
C
C
C
C
C

D
E
A
B
C
D
E
A
B
C
D
E
A
B
C
D
E
A
B
C
D
E

5
5
7
7
5
5
4
7
6
5
5
4
6
5
5
4
3
5
5
4
4
3

0.427
0.054
0.001
0.064
0.172
0.172
0.054
0.001
0.039
0.039
0.039
0.039
0.001
0.064
0.108
0.108
0.054
0.001
0.064
0.108
0.108
0.054

is moved closer to its expected value. While parameters et and ec

(i.e., center values for fuzzy set E at time and cost domains) cannot vary, as they are given by the assembly process study case;
other denition parameters may be changed for a better adjustment of the system to each group of alternatives to study. As a
sensitivity analysis, the described alternatives were evaluated with
different parameters congurations to check their dependency to
the nal scores. Table 9 shows the obtained nal scores for seven
congurations.
Conguration 1 centers set A for time and cost (i.e., at and ac )
at the 20% of the objective values (i.e., et and ec ) and efciency
extremes (i.e., aeff and eeff ) at 50% and 5%, respectively. Congurations 24 increase the center parameters of set A (decrease in the
case of efciency) making the system less strict. That makes the

Table 9
Final scores obtained with different evaluation parameters congurations.
Conguration

Parameters

Alternative

Final score

et = 50 h; ec = 300 000D
at = 0.2et ; ac = 0.2ec
aeff = 0.5; eeff = 0.05
et = 50 h; ec = 300 000D
at = 0.3et ; ac = 0.3 ec
aeff = 0.4; eeff = 0.05
et = 50 h; ec = 300 000D
at = 0.4et ; ac = 0.4 ec
aeff = 0.3; eeff = 0.05
et = 50 h; ec = 300 000D
at = 0.5et ; ac = 0.5ec
aeff = 0.2; eeff = 0.05
et = 50 h; ec = 300 000D
at = 0.5et ; ac = 0.5ec
aeff = 0.5; eeff = 0.05
et = 50 h; ec = 300 000D
at = 0.5et ; ac = 0.7ec
aeff = 0.5; eeff = 0.05
et = 50 h; ec = 300 000D
at = 0.2et ; ac = 0.4ec
aeff = 0.4; eeff = 0.05

2 Workstations

5.1192

3 Workstations
2 Workstations

4.2108
5.8986

3 Workstations
2 Workstations

5.0679
6.2770

3 Workstations
2 Workstations

5.8938
7.9756

3 Workstations
2 Workstations

6.6820
5.8552

3 Workstations
2 Workstations

5.1599
5.8784

3 Workstations
2 Workstations

6.2545
5.7975

3 Workstations

4.993

7
Fig. 13. Membership to fuzzy set of process parameters for the rst alternative of
the study.

52

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

Table 10
Results for the second study.
Alternative

Res. model/Specialization

Attributes

Drillig Op.s within assemble activities

St1.7 in manual variant

Driller
HDr200

Drillig Op.s within assemble activities

St1.7 in automated variant

Str.assembler
Auto drilling machine
ADM80

Mean speed (el.s/s)

Cost (D )
Life expectancy (days)
Cost (D /h)
Mean speed (el.s/s)
Cost (D )
Life expectancy
Cost (D /h)

Auto machines operator

Fig. 14. Output joint membership function for the rst alternative of the rst case.

scores increase, but the rst alternative remains being considered

the best.
Conguration 5 uses the most permissive values for time and
cost domains, the worst elds for alternative 2, and the most
restrictive ones for efciency, its best eld, but the score keeps
larger for the rst one. Only in the extreme case of conguration 6,
where parameter ac is increased as much as each alternative cost
can hardly affect its nal score, the second alternative is rated as
the best one.
The nally adopted parameters conguration for the case study
is number 7, which ensures that the process parameters distributions are mostly included within the intervals [ai ,ei ], for all the
domains of the compared alternatives. Figs. 13, 14 and Table 8 were
obtained from evaluating the rst alternative with the conguration 7.
For the second part of the case study, the alternative with two
workstations, ranked as best, is considered. The drilling operations
were changed from automated to manual to evaluate the impact
on the cost. To avoid the impact on the workstation cycle time,
the change was only applied to activities that were not part of the
critical path, in particular, children tasks of the rst assemble operation (ninth operation in the lower Gantt diagram show in Fig. 12).
Table 10 shows the obtained results, choosing the manual variant
implies a small cost saving with the same workstation cycle time,
so the score for this case got a minor rise, from 5.7975 to 5.7923
over 10.
7. Conclusions
This paper proposed a method and a prototype application
to guide and assist an industrialization engineer in the creation
and evaluation of conceptual aircraft nal assembly lines (FAL).
The prototype application is integrated within a commercial software system widely used in the aerospace sector. It assists the

0.5
3200
280
12
2.5
8650
150
14.5

Cost (D /cycle)

Final score

171 018.63 ( = 521.64)

57 975

171 540.27

57 923

designers during the conceptual phase of the industrialization

decision-making process, when evaluating the impact of different
FAL congurations is difcult and time consuming due to the low
denition state of product and processes.
The material presented in [4,10,13,20] are antecedent works and
initial states of the work presented in this paper. In relation to them,
there are four contributions to point out; some of them were explicitly identied as future works in such prior papers. The rst one is
the enhancement of the initial method to guide the designer by
splitting the FAL conceptual denition into: logistic plan, assembly
line, layout and evaluation. The second one is the enhancement and
extension of a prior aircraft assembly line conceptual design model
[4], to support the guidance method. This enhancement led to the
denition of the processes and resources dictionary (PRD). The
third one is the enhancement of the cost model by considering an
activity-based approach to evaluate both transport and assembly
cost. This makes the approach more general and allows including
in the future additional activities that are not currently considered.
The fourth contribution is the adoption of a multi-criteria evaluation method, based on Fuzzy Logic, which considers time, cost and
resource usage-efciency, to support the decision-making process.
A case study with two parts was executed. The rst part objective was to assist in identifying the best alternative between two
feasible solutions. The second part objective was to evaluate the
impact of a change in a given alternative. The execution of the case
study and the obtained results show the viability of the proposed
method.
In addition to completing the implementation of the metaheuristic algorithm to optimize the logistic plan, future works aim
developing an algorithm to implement capabilities to generate the
as-prepared product structure automatically, and extending the
processes and resources dictionary to incorporate more types of
processes, such as electrical or furnishing.
Acknowledgements
The authors would like to thank their colleagues from AIRBUS
Group and Polytechnic University of Madrid that kindly helped and
contributed to this work.
References
[1] Raju JA. Conceptual design and cost optimization methodology. In: 44th
AIAA/ASME/AHS structures, structural dynamics and materials conference.
2003. AIAA 2003-1505.
[2] Buttereld J, Mawhinney P, Crosby S, Price M, Curran R, Armstrong CG, et al.
An integrated approach to the conceptual development of aircraft structures
focusing on manufacturing simulation and cost. In: AIAA 5th aviation, technology, integration and operations conference. 2005.
[3] Weber A. Assembly automation takes off in aerospace industry. Assembly
Magazine April; 2015. Available from: http://www.assemblymag.com/articles/
92790-assembly-automation-takes-off-in-aerospace-industry.
[4] Mas F, Ros J, Menndez JL, Gmez A. A process-oriented approach to modeling the conceptual design of aircraft assembly lines. Int J Adv Manuf Technol
2013;67:77184.
[5] Wang L, Shen W, Xie H, Neelamkavil J, Pardasani A. Collaborative conceptual
design state of the art and future trends. Comput Aided Des 2002;34:98196.

A. Gmez et al. / Journal of Manufacturing Systems 40 (2016) 3753

[6] Pardessus T. Concurrent engineering development and practices for aircraft
design at Airbus. In: 24th ICAS conference. 2004.
[7] Ming XG, Yan JQ, Wang XH, Li SN, Lu WF, Peng QJ, et al. Collaborative process
planning and manufacturing in product life cycle management. Comput Ind
2008;41:15466.
[8] Curran R, Zhao X, Verhagen WJ. Concurrent engineering and integrated aircraft
design. In: Concurrent engineering in the 21st century. Berlin: Springer Intl.
Pub.; 2015. p. 571605.
[9] Lu SC-Y, Elmaraghy W, Schuh G, Wilhelm R. A scientic foundation of collaborative engineering. Ann CIRP 2007;56:60534.
[10] Mas F, Gmez A, Menndez JL, Ros J. Proposal for the conceptual design of
aeronautical nal assembly lines based on the industrial Digital-Mock Up concept. In: Product lifecycle management for society. Berlin: Springer; 2013.
p. 109.
[11] Mas F, Menendez JL, Oliva M, Ros J, Gomez A, Olmos V. iDMU as the collaborative engineering engine: research experiences in airbus. In: Engineering,
technology and innovation, 20th intl. ICE conference. 2014.
[12] AeroSpace and Defence Industries Association of Europe. ASD SSG
through life cycle interoperability report; 2014. Available from: http://www.
asd-ssg.org/c/document library/get le?uuid=57ad4d3b-47be-4392-aa7dc6e449edfc8d&groupId=11317.
[13] Mas F, Ros J, Gmez A, Hernndez JC. Knowledge-based application to dene
aircraft nal assembly lines at the industrialisation conceptual design phase. Int
J Comp Integ M 2015 2015, http://dx.doi.org/10.1080/0951192X.2015.1068453
[ahead-of-print].
[14] Ros J, Mas F, Menndez JL. Aircraft nal assembly line balancing and workload
smoothing: a methodological analysis. Key Eng Mater 2012;502:1924.
[15] Wang L, Keshavarzmanesh S, Feng H-Y, Buchal RO. Assembly process planning
and its future in collaborative manufacturing: a review. Int J Adv Manuf Technol
2009;41:13244.
[16] Hu SJ, Ko J, Weyand L, ElMaraghy HA, Lien TK, Koren Y, et al. Assembly
system design and operations for product variety. CIRP Ann Manuf Technol
2011;60:71533.
[17] Boerl CR, Pedrazzoli P, Saccol M, Rinaldi R, De Pascale G, Avai A. Integrated computer aided design for assembly system. CIRP Ann Manuf Technol
2001;50:1720.
[18] Colledani M, Franchini F, Micchetti F, Ratti A, Taurisano A. A software platform
for the multi-objective early-stage design of automotive assembly lines. IFACPap Online 2015;48:228792.
[19] Anselmetti B, Fricero B. Aid tool for the design of process and aircraft assembly
lines. Aerosp Sci Technol 2012;23:38798.
[20] Gmez A, Mas F, Menndez JL, Ros J. A knowledge-based applifor
industrialization
design.
Proc
Eng
2013;63:31826,
cation
http://dx.doi.org/10.1016/j.proeng.2013.08.178.
[21] Dmoly F, Yan X-T, Eynard B, Rivest L, Gomes S. An assembly oriented design
framework for product structure engineering and assembly sequence planning.
Robot CIM-Int Manuf 2011;27:3346.

53

[22] Dmoly F, Dutartre O, Yan X-T, Eynard B, Kiritsis D, Gomes S. Product relationships management enabler for concurrent engineering and product lifecycle
management. Comput Ind 2013;64:83348.
[23] Wanga H, Ronga Y, Xiang D. Mechanical assembly planning using ant colony
optimization. Comput Aided Des 2014;47:5971.
[24] Morato C, Kaipa KN, Gupta SK. Improving assembly precedence constraint generation by utilizing motion planning and part interaction clusters. Comput
Aided Des 2013;45:134964.
[25] Battaa O, Dolgui A. A taxonomy of Line Balancing problems and their solution
approaches. Int J Prod Econ 2013;142:25977.
[26] Rekiek B, Dolgui A, Delchambre A, Bratcu A. State of art of optimization methods
for assembly line design. Annu Rev Control 2002;26:16374.
[27] Manavizadeh N, Tavakoli L, Rabbani M, Jolai F. A multi-objective mixedmodel assembly line sequencing problem in order to minimize total costs
in a Make-To-Order environment, considering order priority. J Manuf Syst
2013;32:12437.
[28] Keshavarzmanesh S, Wang L, Feng H-Y. A hybrid approach for dynamic
routing planning in an automated assembly shop. Robot CIM-Int Manuf
2010;26:76877.
[29] Shuford Jr RH. Activity-based costing and traditional cost allocation structures.
In: Cost estimators reference manual. New York: Johan Wiley & Sons Inc; 1995.
p. 4194.
[30] Amen M. Cost-oriented assembly line balancing: model formulations, solution
difculty, upper and lower bounds. Eur J Oper Res 2006;168:74770.
[31] Baud-Lavigne B, Agard B, Penz B. Mutual impacts of product standardization
and supply chain design. Int J Prod Econ 2012;135:5060.
[32] Kampker A, Burggra P, Wesch-Potente C, Petersohn G, Krunke M. Life cycle oriented evaluation of exibility in investment decisions for automated assembly
systems. CIRP-JMST 2013;6:27480.
[33] Tang ZE, Goetschalck M, McGinnisb L. Modeling-based design of strategic supply chain networks for aircraft manufacturing. Proc Comput Sci
2013;16:61120.
[34] Moncayo-Martnez LA, Recio GG. Bi-criterion optimisation for conguring an
assembly supply chain using Pareto ant colony meta-heuristic. J Manuf Syst
2014;33:18895.
[35] Velasquez M, Hester PT. An analysis of multi-criteria decision making methods.
IJOR 2013;10:5666.
[36] Azadegan A, Porobic L, Ghazinoory S, Samouei P, Kheirkhah AS. Fuzzy logic in
manufacturing: a review of literature and a specialized application. Int J Prod
Econ 2011;132:25870.
[37] Devs GM, Ribeiro LC. A fuzzy logic approach to supply chain performance
management. Int J Prod Econ 2011;134:17787.
[38] Kumara D, Singh J, Singh OP. A fuzzy logic based decision support system for
evaluation of suppliers in supply chain management practices. Math Comput
Model 2013;58:167995.
[39] Reyero R, Nicols CF. Sistemas de control basados en lgica borrosa: fuzzy
control. Madrid: Omron Electronics; 1995.