Robotics and Computer-Integrated Manufacturing 27 (2011) 33–46

Contents lists available at ScienceDirect

Robotics and Computer-Integrated Manufacturing

0736-58

doi:10.1

n Corr

E-m

x.yan@s

louis.riv1 Te2 Te3 Te4 Te

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

An assembly oriented design framework for product structure engineeringand assembly sequence planning

Frederic Demoly a,n, Xiu-Tian Yan b,1, Benoıt Eynard c,2, Louis Rivest d,3, Samuel Gomes a,4

a Mechatronics, Methods, Models and Skills Laboratory (M3M), Universite de Technologie de Belfort-Montbeliard (UTBM), F 90010 Belfort Cedex, Franceb Department of Design, Manufacture and Engineering Management, University of Strachclyde, James Weir Building, 75 Montrose Street, Glasgow G1 1XJ, United Kingdomc Department of Mechanical Systems Engineering, CNRS 6253 Roberval, Universite de Technologie de Compi�egne (UTC), BP 60319, Rue du Docteur Schweitzer,

F 60203 Compi�egne Cedex, Franced Laboratoire d’Ingenierie des Produits Procedes et Syst�emes (LIPPS), Ecole de Technologie Superieure de Montreal (ETS), 1100 rue Notre-Dame Ouest,

Montreal (Quebec), Canada H3C 1K3

a r t i c l e i n f o

Article history:

Received 27 May 2009

Received in revised form

4 May 2010

Accepted 28 May 2010

Keywords:

Assembly-oriented design

Assembly sequence planning

Design for assembly

Product structure

45/$ - see front matter & 2010 Elsevier Ltd. A

016/j.rcim.2010.05.010

esponding author. Tel.: +33 0 3 84 58 37 50;

ail addresses: frederic.demoly@utbm.fr, mfdem

trath.ac.uk (X.-T. Yan), benoit.eynard@utc.fr

est@etsmtl.ca (L. Rivest), samuel.gomes@utb

l.: +44 141 548 2852; fax: +44 141 552 7986

l.: +33 0 3 44 23 79 67.

l.: +1 514 396 8984; fax: +1 514 396 8595.

l.: +33 0 3 84 58 30 06; fax: +33 0 3 84 58 32

a b s t r a c t

The paper describes a novel framework for an assembly-oriented design (AOD) approach as a new

functional product lifecycle management (PLM) strategy, by considering product design and assembly

sequence planning phases concurrently. Integration issues of product life cycle into the product

development process have received much attention over the last two decades, especially at the detailed

design stage. The main objective of the research is to define assembly sequence into preliminary design

stages by introducing and applying assembly process knowledge in order to provide an assembly

context knowledge to support life-oriented product development process, particularly for product

structuring. The proposed framework highlights a novel algorithm based on a mathematical model

integrating boundary conditions related to DFA rules, engineering decisions for assembly sequence and

the product structure definition. This framework has been implemented in a new system called

PEGASUS considered as an AOD module for a PLM system. A case study of applying the framework to a

catalytic-converter and diesel particulate filter sub-system, belonging to an exhaust system from an

industrial automotive supplier, is introduced to illustrate the efficiency of the proposed AOD

methodology.

& 2010 Elsevier Ltd. All rights reserved.

1. Introduction

The current ultra-competitive market in terms of quality–cost–time exacerbated by the global financial crisis leads companies toset up even more focused research and development strategies, inorder to improve their competitiveness. Companies in theautomotive industries are particularly facing their own indus-try-sector specific problems and they are re-organising by bring-ing geographically distributed teams and expertise networkscloser. Thus, the current industrial need requires the tackling ofthe traditional sequential engineering phase based approaches bytaking a concurrent engineering approach to overlapping life cycleengineering activities. Indeed, the inclusion and integration of

ll rights reserved.

fax: +33 0 3 84 58 32 07.

oly@gmail.com (F. Demoly),

(B. Eynard),

m.fr (S. Gomes).

.

07.

product life cycle activities into the product development processhave received much attention over the last two decades,especially at the detailed design phase. Pondering over thesesevere constraints, methods, tools, and techniques for suchan integration of product life cycle activities is the focus ofcommon research interest. One particular area is to research intohow to undertake product design and assembly sequenceplanning (ASP) in a concurrent way [1]. The main objective ofthe paper focuses on the early generation of assembly sequencedefinitions based on the initial evolving design and at the sametime applying this information as early as possible into thepreliminary design stages by introducing relevant assemblyprocess information, in order to provide a better assemblyoriented design guidance. Therefore, considering assembly issuessuch as an ASP phase in early preliminary design stagesrepresents an emergent and novel research approach in thebroader context of an assembly-oriented design (AOD). Consid-ered as a top–down design methodology [2], AOD approach aimsat tackling the lack of life cycle context in the productdevelopment process, which is traditionally based on productfunctional definitions [3]. An AOD is a promising approach tobring out an assembly context for product structuring and

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–4634

modelling by taking into account the resulting preliminaryassembly sequence during the product development process.

Based on the previous work on assembly sequence definition[4], the paper describes an initial effort towards an AOD frame-work for product structure engineering using an ASP information.In contrast to other work on an AOD, this paper presents a novelapproach to bringing and integrating assembly information inearly design stage, so that product designers have all assemblyrelated information to make informed assembly oriented designdecisions. The proposed approach highlights a novel algorithmcalled Assembly Sequence Definition Algorithm (ASDA) and basedon a mathematical model integrating boundary conditions relatedto DFA rules, engineering decisions for assembly sequence and theproduct structure definition.

The paper begins by presenting in Section 2 a literature survey,in which current research state of the art and challenges in termsof broad ASP and DFA are depicted. In Section 3, the novelmethodology describes an AOD framework for product structureengineering and related ASP in the early product developmentprocess. The proposed approach is implemented in a new systemcalled PEGASUS, which is developed as an AOD module for a PLMsystem. Finally, a case study based on a catalytic-converter anddiesel particulate filter sub-system, which forms an exhaustsystem from an industrial automotive supplier, is introduced toevaluate the AOD approach and its efficiency.

2. Literature survey

This section gives an overview on the significant amount ofpublished research work on an ASP and DFA. It also gives aconclusion of the key findings of the current research status andchallenges in the fields.

2.1. Assembly sequence planning

An ASP is considered as the sub-domains of an assemblyprocess planning field. Research interest on an ASP has receivedmuch attention from manufacturing industries and researchprojects over the past 20 years. Such research has led to variousapproaches for assembly sequences generation through algo-rithms, representation using graphs and diagrams, and evaluationwith decision criteria to reduce combinatorial complexity.Bourjault was the first to cover this issue through directed graphsand recurring questions [5]. Later on, De Fazio and Whitneyreduced the number of expert questions in Bourjault’s approach[6]. Then, Homem de Mello and Sanderson showed an AND/ORgraph as a decomposition graph, thus giving a description of theassembly [7]. Santocchi and Dini analysed subassemblies andassembly sequences based on a mathematical model of acandidate product by introducing three matrices, namely theinterference matrix, the contact matrix and the connection matrixin the Cartesian coordinate system [8]. In addition, Mascleimproved the functional model integrating mechanical bondsand related half-degree bonds [9]. More recently, Gu et al.

presented an ordered binary decision diagram (OBDD) torepresent and manipulate all the possible assembly sequences[10]. These attempts highlighted the added value of graph- andmatrix-based approaches only for around ten components.However, for products with more components, these approacheshave not been tested and will increase the reasoning complexityof an assembly planner.

Several authors proposed to work on detailed productgeometry as an input for their approaches. Starting from a CAD(B-Rep) model, Laperri�ere and ElMaraghy carried out a generativeassembly sequences approach from a geometric, stable, and

accessible point of view in order to improve the procedureefficiency [11]. Moreover Gottipolu and Ghosh described amatrix-based methodology for automatically generating assemblysequences in a CAD system, starting from contact and interferencevectors [12,13]. On the other hand, Zhang et al. described amathematical model based on connection matrix to definefeasible assembly sequences for automotive parts in the body inwhite context [14]. Recently, Lin et al. proposed a contact relationmatrix to generate assembly sequences and to aid engineers fordesign alternative identification [15]. In a closer connection withthe CAD system, the authors have also emphasised the impor-tance of understanding relational information and geometrycharacteristics to optimise the assembly sequence planning.

During the last decade, most of research work has focused onoptimisation methods such as metaheuristics algorithms [1].Genetic algorithms (GA) have also been used to generate varioussolutions, some of them being finely optimised. Thus, Bonnevilleet al. used a GA to generate and evaluate assembly plans [16]. DeLit et al. developed a GA to generate assembly trees, and used anexpert system to choose an optimal sequence [17]. In addition,Smith and Smith provided an automated assembly planning basedon an enhanced GA which was more reliable and quicker thanusual GA [18]. Tseng et al. used an improved GA called memeticalgorithm to generate feasible assembly sequences with largeconstraints using the connector concept [19]. More recently, Suproposed a hierarchical ASP approach, so that geometric andengineering admissible assembly sequences could be reasonedout automatically and that the optimal sequences could beselected easily [20]. Although GA based approaches have beenevaluated as more efficient methods to generate and chooseassembly sequences, the global optimum corresponding to thebest assembly sequence is not necessarily reached.

Based on the above findings and limitations, other approachesin the field of knowledge-based engineering and artificialintelligence have been explored. Zha and Du promoted a knowl-edge-based system using multi-agents systems (MAS) and PetriNet to support assembly design and an ASP by considering astart from part relations information [21]. In a separate paper,Dong et al. described a knowledge-based ASP approach, in whichthe assembly was modelled as a connection-semantics-basedassembly tree (CSBAT) [22]. As a result of these efforts, theintroduction of semantic and knowledge in assembly modelsproved to provide a better integration and understanding ofassembly planner’s intents, especially with the use of the agent-based techniques.

Furthermore in the context of PLM, Bowland et al. integratedcomputer-aided assembly process planning (CAAPP) systemfunctionalities into a product data management (PDM) system,providing a data management framework and a high-level datastructure system to form the basis of an APP [23]. Ming et al.

proposed a collaborative approach through process planning(CAPP) and manufacturing (CAM) in the broader context of PLM[24]. Starting from contact and precedence constraints in CADmodels, Demoly et al. laid out a matrix-based approach togenerate feasible assembly sequences into the PLM system andsimulated them in a CAD system [4].

2.2. Design for assembly

Dated from the seventies, design for assembly (DFA)approaches have required a multidisciplinary team to evaluateand validate product design and to make it suitable for anassembly engineering process. First published work and contribu-tions introduced design guidelines and heuristic rules as qualita-tive evaluation. In such a way, Andreasen et al. [25], Redford and

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–46 35

Chal [26] framed rules as design principles for assembly process.Using a qualitative analysis, Hitachi Company developed theassemblability evaluation method (AEM) focusing mainly oninsertion operations, in which each component is allowed tohave only one movement during the assembly process [27]. Then,Yamagiwa proposed the Sony DAC method based on design rulesor requirements in order to adjust product design for automaticassembly [28]. More recently, Stone et al. introduced a conceptualDFA method using a functional basis and heuristic rules, in orderto identify and produce modular product architectures [29].Although these approaches have obtained interests from anindustry, subjective nature has been highlighted as a potentialhindrance because of the required user understanding.

As a result, other approaches have been proposed and denotedformal analysis with quantitative evaluation. Boothroyd andDewhurst presented their experience-based method consideringsystematic handling and insertion time from a motion–time–measurement (MTM) study, for each part of the product and part-to-part difficulties in order to reduce part number, and thereforesimplifying assembly operations [30,31]. Swift proposed Lucas’procedure based on an assembly sequence flowchart and afunctional analysis of the product [32].

Moreover Whitney et al. introduced a strategic concept torationalise product design by considering manufacturing andassembly processes [33]. This first attempt promoted ‘‘Assembly-Oriented Design’’ philosophy as it took into account assemblyconstraints in the product development process [34]. Lee and Shindefined a liaison graph to identify the merge rules for parts [35].

Considering the interrelationships between assembly design,assembly process, and assembly operations, Su proposed theassembly oriented product design and optimisation approach,in which assembly structure evaluation, assembly process plan-ning and assembly system design were integrated accordingto the principle of concurrent engineering [20]. Pu andSu presented an algorithm to aid designers in the design ofassembly plans based on case-based reasoning (CBR) paradigm[36,37].

De Fazio et al. highlighted the weaknesses of traditional DFAapproaches and proposed to consider choice and partitioning ofsub-assemblies and assembly sequence choices, in order to dwellon combinatorial aspects in DFA approaches [38]. Recently,Mascle proposed an approach for computer-aided assembly systemusing assembly features supported by an agent paradigm [39].Barnes et al. highlighted the need of decision support for assemblysequence and product structure generation. A two-stage decisionsupport procedure to define parts and their assembly sequenceswas introduced by Barnes et al. [40]. Coma et al. have presenteda fuzzy decision support system based on Boothroyd andDewhurst DFA methodology [41]. More recently, a DFA approachhas been initiated based on System Modelling Language (SysML)paradigm in the PLM context considering an assembly orientedproduct structure based on preliminary assembly sequence[42,43].

2.3. Synopsis

Among various identified approaches in the field of ASP andDFA as summarised in Sections 2.1 and 2.2, the authors havereviewed each research work on their suitability of the engineer-ing applications, the abstraction level, and the automation degree.The majority of these approaches reviewed are typically semi-generative, based either on heuristic rules or algorithm proce-dures, in order to enable more assembly friendly product design,and to generate all admissible assembly sequences. Integrationattempts in CAD models have meant to take into consideration of

geometric data as an input for assembly sequence generation inan efficient way. However, these research works focused ondetailed product geometry, which essentially introduces combi-natorial problems in an ASP phase due to the product designcomplexity. Besides, traditional well-known DFA approaches canbe considered as reactive, since they work on a detailed productgeometry after all design decisions made, hence they lead toredesigns and delays due to late decisions changes. Theseapproaches therefore result in late decision support and missinga true opportunity for an efficient product development process.

These deficiencies point to the need of developing an emergent,proactive, generative, and Web-based approach to design forassembly in the broader context of PLM [44]. In such a system, itis important to introduce efficient assembly information basedreasoning as early as possible into the preliminary productdevelopment process, in order to define assembly sequences. Atthis stage, combinatorial complexity can be reduced by introducingassembly logical information embedded into directed graphs andrelated matrices to influence the product structuring and modelling.Studies have shown that the decisions made during the productdevelopment process impact the great majority of product costs, andit is imperative to externalise these impacts as early as possible.Based on the reviews and the understanding of the problems, themain objective of this research is therefore to bring assemblyprocess knowledge into an early design stage by developing aconcurrent approach taking into account assembly engineeringinformation into preliminary design stages. This approach willenable significant benefits by applying AOD philosophy to an entireproduct life cycle, and bridging the PLM gaps between proceduresand available company solutions, and intelligent data managementframework in an AOD context. To become more competitive in themodern global manufacturing competition, enterprises must ad-dress these current identified needs and integrate new solutions tothe AOD framework as PLM enablers.

3. Assembly oriented design framework

This section presents an AOD framework by introducing amathematical model, in which boundary conditions related to DFAheuristic rules and assembly process engineering decisionsare described. The proposed framework considers the dynamicaspect of the model, and provides guidance to assembly sequenceplanning and product structure engineering before product geome-try is being specified. The information handling during engineeringdesign such as information modelling, processing, checking, andpropagation are detailed starting from the early product develop-ment process.

3.1. Assembly oriented design methodology

The proposed AOD methodology (Fig. 1) involves several productlife cycle stakeholders related to assembly considerations, whichstarts with a high level of abstraction into the preliminary designstage. The first step consists of setting up a multidisciplinary andmultifunctional working team within a company integrating specificroles of the stakeholders, consisting of:

�

the product architect, who manages product structure andmodelling context for designers in consistency with functionalrequirements, � the designer who specifies and defines product geometric

information according to product geometric requirements andlife cycle engineering constraints,

� the assembly planner, who plans assembly sequences,

assembly operations and allocates assembly resources,

Fig. 1. Assembly-oriented design framework as functional part of PLM strategy.

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–4636

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–46 37

�

the process engineer, who produces and brings out allrelevant information for a specific/standard assembly process, � the ergonomist, who defines and analyses the assembly

operator activity during the assembly of the product.

As a result, this integrated framework (Fig. 1) brings thefollowing added value to an AOD approach:

1.

Starting from a product structure and its associated break-down fulfilling functions and previously defined requirements,the product architect defines a liaison graph by describingcontact relations and assembly pairs between product compo-nents, in which estimated mass is described.

2.

Based on the liaison graph defined by the product architect,the assembly planner assigns an assembly operation for eachcontact relation regarding assembly pairs. As a result, theplanner defines a directed graph by introducing the prece-dence constraints.

3.

Based on the directed graph defined by the product architectand the assembly planner, the ergonomist identifies anddescribes the identified assembly operations. Besides, he/sheintroduces component mass and time constraints and limita-tions as decision making criteria for selecting future sub-assemblies.

4.

The directed graph is converted into an adjacency matrix asrelevant input information for the algorithm. The proposedframework is based on a mathematical model which will bedescribed in Section 3.2. The algorithm integrates the defini-tion of sub-matrix types related to sub-assembly types, andgenerates several sub-assembly alternatives with severaldimensions. The assembly planner manages sub-assemblylayers by taking into account criteria related to the ergonomicsand assembly process engineering. Once the assembly se-quence is defined, an ASDA algorithm creates XML files foreach engineering domain, namely product structure, assemblyoperations structure, and assembly activity structure. This inturn enables information propagation to others PLM systemssuch as CAx, PDM, and Manufacturing Process Management(MPM) systems.

5.

Starting from each generated structure, designers, processengineers and assembly operators can work concurrently on alocal and single product perspective definition, integratingcontextualised information. For instance, designers could workon a skeleton geometry model, which integrates geometricconstraints related to DFA rules and assembly sequence.

It is necessary to highlight the novel aspect of the proposedframework. Whilst the traditional approaches tend to performtasks 1, 2, and 3 in sequential fashion, the novel aspect of thiswork lies in that a concurrent execution of tasks 1, 2, and 3 is fullysupported as indicated by bidirectional arrows among them. Thisis enabled because all these three tasks provide inputs to thecritical adjacency matrix, from which anyone of the threestakeholders can access life cycle related design information atany time. At the same time, the information on assembly planningand ergonomics of product design and use is made available to theengineer designers who undertake the design activities within theproduct architecture task (box 1 of Fig. 1) and associated resultantproduct structure information (box 5). Similarly, product designinformation is also available to assembly planners concurrently toenable a concurrent execution of design activities and planning,and this is shown through bidirectional arrows between boxes 1and 2, boxes 2 and 3, boxes 1 and 3 in Fig. 1. It is clear that theproposed arrangement of design tools and assembly tools withinthe life cycle system, facilitated by the proposed adjacency matrix

and its interface functions. With the above availability ofinformation on both design and assembly planning, an assemblysequence definition algorithm (ASDA) has been developed toenable an effective assembly design process. This illustrateshow the design process interacts with an assembly in aconcurrent manner. Due to space limit, similar interactionbetween design and assembly driven by assembly informationwill be described in another paper. The following section willdescribe the underlying ASDA algorithm, which enables the abovecapabilities.

3.2. Mathematical modelling for the proposed ASDA algorithm

To address the above dynamic nature and context in thesuggested framework, the authors have derived a mathematicalmodel describing an ASDA algorithm of generating assemblysequences for product structure engineering. This algorithm isconsidered as the core of the AOD framework, considering thedefinition of sub-assembly types, DFA heuristic rules in mathe-matical equations, and the resulting domain-specific view gen-eration based on the defined assembly sequence. A flowchart isintroduced to represent the algorithm, in which each step isdescribed into the following sub-sections (Fig. 2). A summary ofnotations used in this model is given in Table 1.

3.2.1. Definition of directed graph and adjacency matrix (steps 1 and

2 of Fig. 2)

A directed graph model with directed edges is set up torepresent the abstract information from each engineering domain,which takes into account all the preliminary information, and ismuch useful for visual analysis. This model is converted into amatrix form called ‘‘adjacency matrix’’, which is used forcomputer processing in the proposed framework. The adjacencymatrix of a directed graph is always anti-symmetric about thediagonal line running from the upper left entry to the lower right.

Let C¼{c1, c2, c3, y, cn) be the finite set of all productcomponents with a cardinality of nAN*. Let R¼{r1, r2, r3, y, rm)be the finite set of all relational constraints between twocomponents with a cardinality of an mAN*. Suppose a directedgraph G¼ ðC, RÞ, where vertices or nodes represent componentsand ordered pairs of vertices called directed edges (or bonds)represent relational information. Among these relations, wedefine various abstraction levels of input information from eachstakeholder’s point of view, as described into the proposedframework. At first, these graph models include two types ofinformation:

�

contact relation as a physical relation between two components(in solid line), and � precedence relation defining an assembly logical order for two

components in contact (solid arrow) and in non-contact (dotpattern arrow).

Let Pn be the adjacency matrix defined from the directed graph Gas a fixed anti-symmetric n�n matrix entries Pi,jA{0, 1, �1, l, �l}(Fig. 3).

Described in the directed graph G, a comprehensive productlife cycle relationships can be defined into an adjacency matrix.These relations about the contact and precedence information aredefined as follows:

Pi, j ¼ ½pi, j�i, jAN*�N* ¼

pij ¼ 1

pij ¼ lpij ¼ 0

8><>: ð1Þ

Fig. 2. Flowchart for the proposed ASDA algorithm.

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–4638

Table 1Notations used in the mathematical model.

Symbol Description

C Set of components

R Set of a relational constraint between two components

N* Set of all positive natural numbers

Z Set of all integers

G Directed graph

Pn Adjacency matrix

P* Contracted matrix

SA Sub-assembly

CMSA Companion matrix related to Sub-assembly matrix

cv Column vector of the matrix

n Number of components

m Pre-determined mass of component

To Acceptable limit of manual handling of loads for one person

CT0 Pre-defined cycle time (expected average total production time per

unit produced)

TT Design Takt time (the rate at which the end product must be

produced to meet customer demand)

Wc Work content

Th Handling time for one component

Ti Idle time (fixed assets)

AW Available working time per day

D Forecasted or customer demand per day

C1 C3

C4

C2

Cn-1

Cn 0..

::0:

00

..0

pn2pn1

p2np21

p1np12

Pn =

Fig. 3. Directed graph G and related adjacency matrix Pn.

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–46 39

where the entries in row i and column j are

pij ¼ 1 if

(there is an oriented contact relation between component i and component j

the component i must be assembled before component j is assembled;

pij ¼ l if

( there is a precedence relation between component i and component j

Component i and component j are not in contact

the component i must be assembled before component j is assembled

;

pij ¼ 0 if

(Self-relationship

No relation

8ði, jÞAN*�N*, pij ¼�pji avec pijAZ

����������������������

In this framework, the adjacency matrix is decomposed into

three layers with different information from a stakeholder’s point ofview. The first layer contains contact relation information, thesecond one defines precedence relations and assembly operationtime related to assembly operations, and the third describes workcontent of activities. Assembly operation time and work content areassociated to each Pij¼1 from the first layer. The following sectionsdetail the algorithm proposed and implemented in this research.

3.2.2. Sub-assembly identification rules (steps 3–6 of Fig. 2)

Starting from the previous adjacency matrix, it is important toidentify possible sub-assemblies with the definition of assemblytypes. As a result, four kinds of assemblies can be identified, anddirected graph, matrix form, pseudo-vector, assembly time andwork content are listed in Table 2. For each type, a pseudo-vectorcan be generated to represent the assembly sequence. The authorsuse a ‘‘,’’ between two components to describe a serial assembly

operation and a ‘‘-’’ for a parallel assembly operation. Assemblytime and work content are associated to each directed edges andtherefore computed for each kind of assemblies. For instance, aserial assembly will have one contact relation between twoconsecutive components only, whereas a parallel assembly willhave all other components connected only to a base component.

A mathematical expression for each kind of assemblies isdescribed in Table 3 to be used in an ASDA algorithm.

For each assembly, an ASDA algorithm generates all thepermutations in order to facilitate sub-matrices recognition fromthe adjacency matrix Pn. This approach avoids time-consumingprocess mainly by investigating subassemblies through simplifiedand smaller sub-matrices as shown in Table 2.

Let XkA{Bk, Ck, Dk, Fk} be the set of all combinations of eachmatrix related to assembly types

Xk ¼ [3rkrn�1

AkkXij 8 i, j, kAN*

�N*�N*

ð6Þ

Let Vk be the set of all combinations of sub-assembly matrixvariants

Vk ¼ [3rkrn�1

AkkXk 8kAN*

ð7Þ

Once the definition and generation of assembly variants arecompleted, the total number of valid matrix combinations to betested is described below. Hk here represents the set of allcombinations of k elements taken from n elements of C in order torepresent combinations of k� k matrices into Pn.

Hk ¼Xk ¼ n�1

k ¼ 3

n!

k!ðn�kÞ!¼

Xk ¼ n�1

k ¼ 3

n

k

� �8k, nAN*

�N*ð8Þ

LetM0ðN*Þ be the set of candidate sub-assemblies matrices of

k dimension in N*

M0ðN*Þ ¼ Vk \ Hk

8ðhi, jÞAHk, ðhi, jÞAM0ðN*Þ3(ðvi, jÞAVk9ðhi, jÞ ¼ ðvi, jÞ ð9Þ

As soon as the full match is found, each sub-matrix represent-ing possible SA can be tested in order to check the engineeringfeasibility through interference with the remaining components.

3.2.3. Sub-assembly validation rules (steps 7–9 of Fig. 2)

For each identified SA from the generated adjacency matrix, acompanion matrix CM is associated and extracted regarding therelations between SA components and the remaining components ofthe product, in order to determine if the SA is feasible from aninterference point of view (Fig. 4). As a result, a rule must be appliedwith the associated companion matrix (10). Let Mk,n�kðN

*Þ be the

set of companion matrices related to candidate sub-assemblies of k

dimension in N*. Let MkðN*Þ be the set of fixed anti-symmetric

k� k matrix in N* related to feasible sub-assemblies.

ðhkÞAMkðN*Þ3

(!ðcmk, n�kÞAMk, n�kðN*Þ9

8ðcvðcmk, n�kÞÞ,cmk, n�kZ03cmk, n�kr0

����� ð10Þ

Table 2Assembly Types of three components and related relevant information for computer processing.

Assembly type Directed graph Adjacency matrix Pseudo-vector Assembly time Work content Example

Interconnected serial (1)

1t12

t23

2

3t13

1 2 3

1

2

3

0 1 1

�1 0 1

�1 �1 0

264

375

[1, 2, 3] t12+t23 wc12+wc23

12

3

Serial (2) t12 t231 2 3

1 2 3

1

2

3

0 1 0

�1 0 1

0 �1 0

264

375

[1, 2, 3] t12+t23 wc12+wc23

3

21

Constrained serial (3) t12

t13 3

211 2 3

1

2

3

0 1 1

�1 0 l�1 �l 0

264

375

[1, 2, 3] t12+t13 wc12+wc13

1

3

2

Parallel (4)t12 t13

1

2 3

1 2 3

1

2

3

0 1 1

�1 0 0

�1 0 0

264

375

[1, 2, �3] Max(t12, t13) wc12+wc13 2 3

1

Table 3Mathematical description of each kind of assembly matrix.

Assembly type

matrix

Mathematical description

Interconnected

serial matrixBij ¼ ½bi,j�i,jAN*

�N* ¼bij ¼ 1 8j4 i

bij ¼ 0 otherwise

(ð2Þ

Serial matrixCij ¼ ½ci,j�i,jAN*

�N* ¼cij ¼ 1 8j¼ iþ1

cij ¼ 0 otherwise

(ð3Þ

Constrained

serial matrix Dij ¼ ½di, j�i, jAN*�N* ¼

dij ¼ 1 8kAN*, (!iAN*=di, k ¼ 1

dij ¼ l 8j¼ iþ1

(ð4Þ

Parallel matrixEij ¼ ½fi, j�i, jAN*

�N* ¼eij ¼ 1 8kAN*, (!iAN*=ei, k ¼ 1

eij ¼ 0 otherwise

(ð5Þ

C1

C3

C2

Cn-2 Cn-1

CnSA 0..

::0:

00

..0

pn2pn1

p2np21

p1np12

Pn

CMSA SA

=

Fig. 4. Sub-assembly SA identification and related companion matrix CMSA.

(a) Directed graph and (b) adjacency matrix.

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–4640

So if the signs of non-zero elements in kth column are neitherall positive nor all negative, there are interferences between thesub-assembly matrix set and other remaining components.Therefore, this set cannot be considered as a feasible sub-assembly from an engineering point of view. Once the engineering

feasible sub-assemblies identified, further decision supportsbased on specific criteria such as stability, mass and so on, canbe used to provide guidance to choose suitable sub-assemblies.

3.2.4. Decision support (step 10 of Fig. 2)

In order to provide a decision support for the sub-assemblyselection, an initial assembly context previously defined by theassembly planner will have to be taken into account. Startingfrom the forecast (or customer) demand per day and the availablework time per day, the design Takt time TT is defined to indicatethe assembly frequency of sold product and is calculatedautomatically.

TT ¼available working time per day

forecasted demand per day¼

AW

Dð11Þ

Decision support variables and the procedure use informationthat has been modelled into the several graph layers, as describedin the proposed AOD framework. Firstly, pre-determined mass ofproduct components and assembly time associated to assemblyoperations are used as decision making criteria. In addition, thecriterion related to the assembly stability with kinematics pairs isexplored. The processing rules to establish these engineeringcriteria are described below.

Let m be the pre-determined mass associated to each productcomponent. Let To be the acceptable limit of manual handling ofloads for one person. For each engineering feasible sub-assembly,the mass Eq. (12) and the acceptable limit of manual handling ofloads limitations (13) are calculated automatically and limitedby pre-determined constraints such as mlimit and TT defined inthe system complying with NF X35-109 and NF EN 1005-5ergonomics standards [45,46].

mSA ¼Xn

c ¼ 1

mc rmlimit ð12Þ

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–46 41

ToSA ¼mSA

TTrTolimit ð13Þ

Concerning assembly operations, the authors have defined foreach assembly type a calculated and pre-determined cycle timeCT0 using standard assembly operation time Ta, idle time Ti andwork content Wc.

CT0 ¼ TaþWcþTirTT ð14Þ

TSA ¼Xn

op ¼ 1

Top ð15Þ

WcSA ¼Xn

op ¼ 1

Wcop ð16Þ

For the stability criteria, the main rule defines an SA stable ifno mobility exists between its own components, in other words, ifall of its components are held in position.

3.2.5. Adjacency matrix concatenation (step 11 of Fig. 2)

Once the assembly planner has chosen engineering feasiblesub-assemblies in the first procedure cycle, an ASDA algorithmprocesses each selected SA as a single component to bere-introduced in the adjacency matrix Pn. Based on Pn, a newsmaller adjacency matrix called contracted matrix P* is generated(Fig. 5). Thus, other sub-assemblies can also be detected andselected in another assembly layer. This matrix P* depends on thek� k selected SA matrices, and the size of P* is: n–k+1. To build P*,several rules regarding SA matrix and related companion matrix

−1

−1

0

00

00

110

*

pSA3pSA2pSA1

p3SA

p2SA

p1SA

Pn−k+1 =

C2

C1

C3

CSA

Fig. 5. (a) Contracted graph and (b) related contracted matrix.

Product

SA (Base Part)

SA (Base Part)

Part (Base Part)

Part

Part

Part

Part

Skeleton

Skeleton

Skeleton

….

Product

Skeleton

Sub-Assembly (SA)

Part

Fig. 6. Setup product structure defined by the assembly sequence.

must be followed in order to assign entries. These rules describehow SA inherits the relational information of its internalcomponents with the remaining components.

Ifðall elements in the jth columns of companion matrix¼ 0Þ

Then pSA, j ¼ 0

Ifðall non zero elements in the jth columns of companion matrix40 and ¼ lÞThen pSA, j ¼ lOtherwise pSA, j ¼ 1

Ifðall non zero elements in the jth columns of companion matrixo0 and ¼�lÞThen pSA, j ¼�lOtherwise pSA, j ¼�1

���������������������ð17Þ

3.2.6. Assembly sequence and resulting information (step 12 of Fig. 2)

At the end of the procedure, an assembly sequence iscomputed as a master result for specific product structure foreach engineering domain. Consequently, the setup productstructure (Fig. 6) based on the assembly sequence will beconsidered as a core contextual support for an AOD, in whichseveral elements are highlighted including:

�

TabPar

N

1

2

3

4

5

6

Sub-assembly: this is the sub-assembly selected by theassembly planner regarding assembly type, assembly time,work content, mass, and stability,

� Skeleton: describing, for each assembly level, the context in

which parts are assembled,

� Part: elementary component composing the sub-assembly, e.g.

base part, fastener, etc.,

� Assembly parameters: managing assembly data in the local

view.

At each SA level in the product structure, a base part will beidentified according to the SA pseudo-vector previously defined.The base part is considered as the reference component for the SA,that is to say the component onto which the other componentswill be assembled.

4. Case study in an industrial application

The above concurrent approach and an ASDA algorithm havereal industrial relevance, as manufacturing companies have a realneed in integrated methods for product design and assemblyprocess engineering, especially at their interfaces. This approachhas been applied to a company’s real product development as acase study. The chosen part is a catalytic-converter and dieselparticulate filter sub-system (cat-converters and DPF) belongingto an exhaust system from an industrial automotive supplier, andthis case study was designed to illustrate the efficiency of theproposed AOD methodology.

le 4ts list for the case study.

o. Name Type No. Name Type

DPF Sub-assembly 7 Bracket right Sub-assembly

Cat-Converter Sub-assembly 8 Bracket left Sub-assembly

Insulating right Sub-assembly 9 Pressure bracket Part

Insulating left Sub-assembly 10 Left half-shell link Part

Inlet Sub-assembly 11 Right half-shell link Part

Outlet Sub-assembly

Fig. 7. Catalytic-converters and diesel particulate filter. (a) ISO view and (b) exploded view.

11

10

98

7

65

4

3

2 1

00000001100000010100000000000000000000000000000000000000000000000000000000011

0000000-1111110111010001010

λ

P11= -1

-1

-1

-1

-1

-1-1

-1

-1

-1-1-1-1-1

-1-1

−λ

Fig. 8. (a) Directed graph G and (b) related adjacency matrix P11 defined for the industrial case study.

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–4642

4.1. Case study: catalytic-converters and diesel particulate filter and

their design problem definition

Catalytic converters and diesel particulate filter sub-assemblyis located at the exhaust system hot end. It includes eightsub-assemblies and additional three parts (Table 4), in order tofulfil two main functions: minimising gas emissions by redox andeliminating particles by filtration and combustion (Fig. 7). Theauthors have chosen to implement the AOD framework into anAOD system called Product dEsign enGineering based onAssembly SeqUenceS Planning (PEGASUS) in connection withother related systems as described in Fig. 1.

The company is an automotive supplier working on exhaustsystem development for car manufacturers. Engineeringdepartments are facing information exchange issues, mainlybetween engineering design and assembly process engineeringdepartments, because of the numerous information systemsimplemented in the company and the lack of informationintegration at the various design stages. Assembly issues areoften identified in the prototype phases, in which the detaileddesign is finalised and to be validated. As a result of such a designapproach, several problems have been highlighted:

�

ASP phase is performed in a sequential way and thereforeintroducing rework in the product development, � Numerous Bill of Materials (BOM): E-BOM (engineering), CAD-

BOM (CAD), M-BOM (manufacturing) defined in a sequentialand independent way into various systems (CAD, PDM, MPM,ERP, etc.), and

� Assembly issues are often neglected during the design process,

especially at the preliminary design stage.

4.2. The approach based on the framework

The authors focus on three roles and associated actors: theproduct architect, the assembly planner and the ergonomist

working into different departments of the company. Variousinitial conditions must be introduced, in order to launch an ASDAalgorithm. First of all, the company applies a ‘‘one piece flow’’strategy to achieve just-in time manufacturing. As a result, thedesign Takt time TT was calculated with a forecast demand of 500units per day and 450 min of available working time, in order tolimit each pre-determined cycle time CT0 as follows:

D¼ 500 units,

AW¼ 1 shift=day¼ 8:5h-0:5hðlunchÞ�0:5hðbreaksÞ ¼ 450min:

�����TT ¼

450

500¼ 54s=unit

CT0rTT

Next, the ergonomist also introduces mass and the acceptablelimit of loads for manual handling for one person in the referencecondition for manual assembly according to [45]

mlimit ¼ 25kg

Tolimit ¼ 50kg=min

�����All these initial conditions are considered as decision making

support constraints for sub-assemblies selection. A directed graphand the corresponding adjacency matrix, including three layershave been generated in PEGASUS (Fig. 8). These integrate allrelevant input information from all stakeholders involved forassembly sequence and product structure definition.

4.3. System execution

With relevant information as inputs for the algorithm ofassembly sequences definition, the assembly planner introducesassembly time regarding assembly operation for each contactrelation in a second layer. The third layer integrates work contentrelated to assembly time. Indeed, work content is defined by theergonomist to bring required time to perform other relatedoperations such as handling, insertion, and inspecting operations.

Fig. 9. Assembly planner view integrating directed graph and choice of feasible sub-assemblies.

2

5

10

11

3

4

1

6

7

8

9

SA1

SA4

SA2

SA3

Product

Fig. 10. Representation of the resulting assembly sequence. Fig. 11. Assembly-oriented product structure based on the previous assembly

sequence and defined by a CATscript file into CATIA v5.

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–46 43

For the proposed experimentation, the authors focus on sub-assembly identification with k¼3, in order to have a finegranularity level for the assembly sequence definition. Once all

these initial conditions and relevant information are identified,the assembly sequence of the product will be defined. The ASDAalgorithm computed three cycles, therefore representing three

Fig. 12. Functional skeleton model of the catalytic-converters and diesel particulate filter considered for the experimentation.

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–4644

assembly layers, in order to assemble all the product componentsshown in PEGASUS (Fig. 9). The result for the case study ispresented below (Fig. 10).

Assembly sequence¼ fððð2, 5, 10Þ, 11, 3Þ, ðð4, 1, 6Þ, 7, 8Þ, 9Þg

with five serial sub-assemblies :

SA1 ¼ ð2, 5, 10Þ,

SA2 ¼ ð4, 1, 6Þ,

SA3 ¼ ðSA2, 7, 8Þ ¼ ðð4, 1, 6Þ, 7, 8Þ,

SA4 ¼ ðSA1, 11, 3Þ ¼ ðð2, 5, 10Þ, 11, 3Þ,

SA5 ¼ ðSA4, SA3, 9Þ ¼ ððð2, 5, 10Þ, 11, 3Þ, ðð4, 1, 6Þ, 7, 8Þ, 9Þ:

The component 2, 4, SA2, SA1, and SA4are consideredas base part in each generated SA:

���������������������Using the PEGASUS system, the product and the assembly

operation structures are automatically generated. The resultinginformation of the product and assembly structure could also beused by other related systems such as PDM, MPM, and CADsystem through XML files. Fig. 11 shows a setup product structuredefined in CATIA v5 through an import of CATscript fileautomatically generated from PEGASUS.

To illustrate the approach to concurrently generate designsolution and assembly sequence, this paper uses a past designsolution by the company shown in Fig. 7, as the original designrequirements. A new design has been undertaken to demonstratethe potential benefits. The functional requirements of the designare the same. Figs. 9 and 10 show the partial assembly sequencegenerated by the PEGASUS system during the design processwhen a partial design definition is completed as shown in Fig. 12.From the figure, it is clear that there are at least four assemblyfunctional axes proposed in this design and this violates theassembly principle of minimum assembly axes, which is stored inthe proposed PEGASUS framework (box 2 of Fig. 1) among manyother design for assembly rules. Based on the assembly sequencedefined for the above design solution proposal, from which fourassembly axes can be identified, applying this design and designfor an assembly principle to the design proposal, the PEGASUSsystem points to a suggestion of eliminating some of the assemblyaxes. Applying another rule in the PEGASUS system, which states

that if possible, eliminate the assembly axis with fewest numberof parts to be assembled along the candidate axis, PEGASUSidentifies in particular that original top and bottom parts of thecover only have one part each to be assembled along theirfunctional axes, and they are good candidates to be eliminated.Based on this suggestion, the designer can then explore thefeasibility of merging these two cover parts together to realise thesame function. Based on the material used and manufacturingtechnique available, it is possible to merge these two cover partstogether and this leads to a simplified and assembly orienteddesign. Current design requires significant effort in welding thesetwo cover parts together and the new design applying assemblyinformation has significantly reduced this effort.

4.4. Discussions

ASP is a crucial step to enable feasible engineering productsolutions. The authors have focused on extracting and applyingassembly sequence definition at the preliminary design stage togenerate an assembly oriented product structure solution byconstraining the product design with assembly process informa-tion. This approach differs significantly from the traditionalengineering design approaches that define assembly sequencesafter the detailed design stage. In addition, the proposed approachoffers a gain in productivity and efficiency by avoiding reworkcaused by a lack of consideration of assembly, as the preliminarydesign decisions are now made based on assembly considerationsand requirements. An AOD framework takes into account themanagement of information status, relationships and changes, inorder to keep traceability and consistency of decision making tomeet the overall PLM requirements and challenges [47].

Traditional design process does not support the designer tohave access to information of the assembly process. For life-cycleoriented design, this is critical to help designer’s decision making.Such an unavailability of assembly information can be avoided bycreating a link between the product development process and anASP phase and this case study demonstrated such a feasible andan effective approach.

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–46 45

5. Conclusion and future work

Current status and challenges in DFA and ASP approaches havebeen highlighted in this paper. Build on these, an assemblyoriented design framework based on a rigorous and novelmathematical model capturing DFA rules and engineeringdecision making process is proposed and implemented in aWeb-based PEGASUS system. The framework approach enablesthe concurrent generation of preliminary design solution in-formation and the assembly sequence information. Such anapproach hence follows a designer to access to this criticalassembly information even at the preliminary design stage andconsequently enables the designer to make assembly oriented andinformed decisions. Design decisions made in such a mannerallow the designer to foresee any potential assembly difficultiesand issues at much early stage, hence avoiding any rework. This isthe fundamental difference of the proposed work to other designfor an assembly work, and this is also where the authors argue thekey contribution this work made to the field. The above idea hasbeen proved and demonstrated through an industrial automotiveexhaust case study which has been implemented to evaluate theproposed approach and address advantages and drawbacks of theAOD framework.

Considering product design and assembly sequence planningin a concurrent way is a huge challenge as there are challenges tobe addressed in providing right information of different life-cyclephase at the right time from different stakeholders. It is importantto focus on several strategic and tactical aspects of the productthat the issue can be used as a natural launch pad for integratedproduct-process design in the PLM context. This work paved theway for future research and development in this field byproviding a framework and associated Web-based tool; however,the authors will need to address further research issues. One areaof future research is to enable proactive support feature of theframework further by providing timely assembly relevant guide-lines and heuristics knowledge into the preliminary designprocess at the right time [48]. Another aspect is to improve therobustness of assembly sequences generated concurrently duringthe early design, as there is certain information missing and thismay be addressed by deploying full product functional definitionand introducing assembly functional axis definitions. The futurework will also support the product modelling phase by introdu-cing the definition of geometric skeletons related to the assemblysequence. A multiple-view model will also be implemented intoPEGASUS, in order to manage information and knowledgethroughout the product life cycle and from various identifiedviews according to stakeholders’ interests and concerns.

Acknowledgements

The research activity is a part of the CoDeKF Research Project(Collaborative Design and Knowledge Factory), which has beenfunded by the French Automotive Cluster ‘Pole de CompetitiviteVehicule du Futur’. The authors would like to thank BernardMignot for Algorithm Development, Faurecia Emissions ControlTechnology for this collaboration, and all the financial contributorsof this research and technology program: DRIRE de Franche-Comte,Communaute d’Agglomeration du Pays de Montbeliard, ConseilGeneral du Doubs, and Conseil Regional de Franche-Comte.

References

[1] Wang L, Keshavarzmanesh S, Feng H-Y, Buchal RO. Assembly processplanning and its future in collaborative manufacturing: a review. Int J AdvManuf Tech 2009;41:132–44.

[2] Yu-liang L, Wei ZO. Development of an integrated-collaborative decisionmaking framework for product top–down design process. Robot CIM-IntManuf 2009;25:497–512.

[3] Rehman F, Yan XT. Supporting function-means mapping using design contextknowledge at conceptual design, J Des Res, Inderscience publishers, 6(1/2)(2007) 169-189.

[4] Demoly F, Gomes S, Eynard B, Rivest L. PLM based approach for assemblyprocess engineering, Int J Manuf Res, 2010;5(4), in Press.

[5] Bourjault A. Contribution �a une approche methodologique de l’assemblageautomatisee: elaboration automatique des sequences operatoires. Besanc-on,France: Th�ese d’Etat, Universite de Franche-Comte; 1984. November.

[6] De Fazio T, Whitney D. Simplified generation of all mechanical assemblysequences. IEEE J Robotic Autom 1987;3(6):640–58.

[7] Homem de Mello LS, Sanderson AC. Representations of mechanical assemblysequences. IEEE Trans Robotic Autom 1991;7(2):211–27.

[8] Santochi M, Dini G. Computer aided planning of assembly operations: theselection of assembly sequences. Robot CIM-Int Manuf 1992;9(6):439–46.

[9] Mascle C, Approche methodologique de determination de gammes pardesassemblage, Th�ese de l’Ecole Polytechnique de Lausanne, 1990.

[10] Gu T, Xu Z, Yang Z. Symbolic OBDD representations for mechanical assemblysequences. Comput Aided Des 2008;40:411–21.

[11] Laperri�ere L, ElMaraghy HA, GAPP A. Generative assembly process planner.J Manuf Syst 1996;15(4):282–93.

[12] Gottipolu RB, Ghosh K. Representation and selection of assembly sequencesin computer-aided assembly process planning. Int J Prod Res 1997;35(12).

[13] Gottipolu RB, Ghosh K. A simplified and efficient representation forevaluation and selection of assembly sequences. Comput Ind 2003;50:251–64.

[14] Zhang YZ, Ni J, Lin ZQ, Lai XM. Automated sequencing and sub-assemblydetection in automobile body assembly planning. J Mater Process Tech2002;129:490–4.

[15] Lin M-C, Tai Y-Y, Chen M-S, Chang CA. A Rule based assembly sequencegeneration method for product design. Conc Eng 2008;15:291–308.

[16] Bonneville F, Perrard C, Henrioud J-M. A genetic algorithm to generateand evaluate assembly plans. IEEE Symp Emerg Tech Fact Autom 1996;2:231–9.

[17] De Lit P, Latinne P, Rekiek B, Delchambre A. Assembly planning with anordering genetic algorithm. Int J Prod Res 2001;39(16):3623–40.

[18] Smith GC, Smith SS-F. An enhanced genetic algorithm for automatedassembly planning. Robot CIM-Int Manuf 2002;18(5–6):355–64.

[19] Tseng H-E, Wang W-P, Shih H-Y. Using memetic algorithms with guided localsearch to solve assembly sequence planning. Exp Syst Appl 2007;33:451–67.

[20] Su Q. A hierarchical approach on assembly sequence planning and optimalsequences analyzing. Robot CIM-Int Manuf 2009;25(1):224–34.

[21] Zha XF, Du H. A PDES/STEP-based model and system for concurrentintegrated design and assembly planning. Comput Aided Des 2002;34(14):1087–110.

[22] Dong T, Tong R, Zhang L, Dong J. A knowledge-based approach to assemblysequence planning. Int J Adv Manuf Tech 2007;32(11–12):1232–44.

[23] Bowland NW, Gao JX, Sharma R. A PDM- and CAD-integrated assemblymodelling environment for manufacturing planning. J Mater Process Tech2003;138:82–8.

[24] Ming XG, Yan JQ, Wang XH, Li SN, Lu WF, Peng QJ, Ma YS. Collaborativeprocess planning and manufacturing in product life cycle management.Comput Ind 2008;59(2–3):154–66.

[25] Andreasen MM, Kahler S, Lund T. Design for assembly. Springer, Verlag, UK:IFS Publications Ltd.; 1983.

[26] Redford A, Chal J. Design for assembly—principles and practice. England:McGraw-Hill Inc; 1994.

[27] Miyakawa S,Shigemura T, The Hitachi assemblability evaluation method(AREM), In: Proc Jpn Soc Mech Eng, Conf Manuf Syst Environ—lookingToward 21st Century, (1990) 277–282.

[28] Yamagiwa Y. An assembly ease evaluation method for product engineers:DAC. Tech Jpn 1988;21(12).

[29] Stone R, McAdams D, Kayyalethekkel VJ. A product architecture-basedconceptual DFA technique. Design Studies 2004;25(3):301–25.

[30] Boothroyd G, Dewhurst P. Product design for assembly. Wakefield, RI, USA:Boothroyd Dewhurst, Inc.; 1990.

[31] Boothroyd G. Product design for manufacture and assembly. Comput AidedDes 1994;26(7):505–20.

[32] Swift KG. Design for assembly handbook. UK: Salford University IndustrialCentre Ltd.; 1981.

[33] Whitney DE, Nevins JL, De Fazio TL. The strategic approach to product design.Des Anal Integrated Manuf Syst 1988:200–23.

[34] Whitney DE, et al. Designing assemblies. Res Eng Des 1999;11:229–53.[35] Lee S, Shin YG. Assembly co-planner: cooperative assembly planner based on

subassembly extraction. J Int Manuf 1993;4(3):183–98.[36] Pu P, An assembly sequence generation algorithm using case-based search

techniques, IEEE Int Conf Robot Autom, Nice, France, (1992) 2425–2429.[37] Su Q. Applying case based reasoning in assembly sequence planning. Int J

Prod Res 2007;45(1):29–47.[38] De Fazio TL, Rhee SJ, Whitney DE. Design-specific approach to design for

assembly for complex mechanical assemblies. IEEE Trans Robotic Autom1999;15(5):869–81.

[39] Mascle C. Feature-based assembly model for integration in computer-aidedassembly. Robot CIM-Int Manuf 2002;18:373–8.

F. Demoly et al. / Robotics and Computer-Integrated Manufacturing 27 (2011) 33–4646

[40] Barnes CJ, Jared GEM, Swift KG. Decision Support for sequence generation inan assembly oriented design environment. Robot CIM-Int Manuf 2004;20(4):289–300.

[41] Coma O, Mascle C, Balazinski M. Application of a fuzzy decision supportsystem in a design for assembly methodology. Int J Comput Intell Manuf2004;17(1):83–94.

[42] Demoly F, Gomes S, Eynard B, Sagot J-C, Towards a design for assemblyapproach based on SysML paradigm and PLM systems, In: Proc Second CIRPInt Conf Assem. Technol. Syst. (CATS), 21–23 September 2008, Toronto,Canada, ISBN:978-0-9783187-1-0, pp. 100–113.

[43] Demoly F, Gomes S, Eynard B, Rivest L, Sagot J-C, Assembly-oriented productstructure based on preliminary assembly process engineering, In: Proc IntConf Eng Des, ICED’09, 24–27 August 2009, Stanford, CA, USA.

[44] Wiendahl HP, ElMaraghy HA, Nyhuis P, Zah MF, Wiendahl HH, Duffie N, et al.Changeable manufacturing—classification, design and operation. CIRP Ann2007;56(2):783–809.

[45] X35-109, Acceptable limits of manual load carrying for one person, FrenchNorm, 1989, ISSN:0330-3931.

[46] Norme europeenne NF EN 1005-5, Securite des machines, Performancesphysique humaine. Partie 5: appreciation du risque relatif �a la manipulationrepetitive �a frequence elevee, Mai, 2007.

[47] Srinivasan V, An integration framework for product life cycle management,Comput-Aided Des, (2010) in Press.

[48] Yan XT, Borg J, Juster NP, Concurrent modelling of components andrealisation systems to support proactive design for manufacture/assembly, JEng Manuf, In: Proc Inst Mech Eng Part B, 215 (2001) 1135–1141.