PROCESS PLATFORM-BASED PRODUCTION CONFIGURATION

Proceedings of IDETC/CIE 2005 ASME 2005 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference September 24-28, 2005, Long Beach, California, USA

DETC2005-85559 PROCESS PLATFORM-BASED PRODUCTION CONFIGURATION Lianfeng Zhang School of Mechanical & Aerospace Engineering Nanyang Technological University Singapore

Jianxin (Roger) Jiao School of Mechanical & Aerospace Engineering Nanyang Technological University Singapore

ABSTRACT In nowadays’ changing manufacturing environment, designing product families based on product platforms has been well accepted as an effective means to fulfill product customization. The current production practice and academic research of platform based product development mostly focus on the design domain, whereas limited attention is paid to how production can take advantage of product families for realizing economy of scale through enormous repetitions. This paper puts forward a concept of process platforms, based on which an efficient and cost saving production configuration for new members of a product family can be achieved. A process platform implies three aspects, including generic representation, generic structures and generic planning. The issues and rationale of production configuration based on a process platform are presented. A multilevel system of nested colored object-oriented Petri Nets with changeable structures is proposed to model the configuration of production processes. To construct a process platform from existing process data, a data mining approach based on text mining and tree matching is introduced to identify the generic process structure of a process family. An industrial example of high variety production of vibration motors for hand phones is also reported. 1. INTRODUCTION One of the pressing needs faced by manufacturers nowadays is quick response to the requirements of individual customers while keeping high quality and near mass production efficiency, namely mass customization [1]. Developing multiple products in product families based on a common platform has been well recognized as a successful approach in many industries [2]. Current practice in designing product families encompasses the design domain only, i.e., dealing with the transformation of diverse customer needs to functional requirements and subsequently the fulfillment of these

Shaligram Pokharel School of Mechanical & Aerospace Engineering Nanyang Technological University Singapore

requirements through a variety of design parameters [3]. It seldom, if not at all, explicitly considers the input from the backend of product realization, viz., production processes. While seeking technical solutions is the major concern in design, it is at the production stage that product costs are actually committed and product quality and lead times are determined per se. For a given design, the actual cost depends on how the production is planned and to what extent the economy of scale can be achieved within the existing manufacturing capabilities. This implies that the claimed rationale of product family design can only be fulfilled at the production stage [4]. The direct consequence of product customization on production is evidenced by an exponentially increased number of process variations (referred to as process variety), such as diverse machines, tools, fixtures, setups, cycle times, and labors [5]. Process variety introduces significant constraints to production planning and control, preventing e.g., make-to-order systems from building up customization capabilities [6]. Regardless of the negative impact of process variety, the common components and the same basic product structure assumed by the set of customized products in a family introduce similarities in the associated production processes. Consequently, companies are eagerly interested in configuring existing operations and processes (referred to as production configuration) by exploiting similarities among product and process families so as to take advantage of repetitions [7]. Corresponding to a product family, a process family comprises a set of similar production processes that share a common process structure (referred to as a process platform). In addition to leveraging the costs of delivering variety, exploiting process families around process platforms can reduce development risks by reusing proven elements in a firm’s activities [8]. Process platforms entail the conceptual structure and overall logical organization of producing a family of products,

1

Copyright © 2005 by ASME

thus providing a generic umbrella to capture and utilize commonality, within which each new product fulfillment is instantiated and extended so as to anchor production planning to a common process structure [9]. The rationale of such process platforms lies in not only unburdening the knowledge base from keeping variant forms of the same solution, but also in modeling the production of a class of products that can widely variegate the operations and process sequences in accordance with specific design changes within a coherent framework [10]. Therefore, developing process platforms becomes an important task for fulfilling product families in mass customization production [11]. 2. LITERATURE REVIEW The importance of managing variety has been well recognized [12]. To overcome the limitations of traditional bills-of-materials (BOMs) in variant handling, the generic BOM concept has been developed [13]. Hasting and Yeh of [14] propose to combine the routing with traditional BOMs to provide material requirement data for each scheduled operation, resulting in a time-phased material requirement plan derived from a feasible schedule. Blackburn of [15] demonstrates how combining routings and BOMs in one document can support just-in-time manufacturing in a traditional MRP-implemented production environment. The generic bill-of-materials-andoperations (GBOMO) is put forward in [16] to unify BOMs and routings for the purposes of easing production planning and control as well as accommodating large number of product and process variants. Focusing on reducing subassembly proliferation and the cost of offering product variety, Gupta and Krishnan of [17] propose a methodology for designing product family-based assembly sequences. While attempting to create common assembly processes, their method neglects the link between product and process families. The group of [18] put forward the integrated design of product families and the corresponding assembly systems. Their focus is on new product families with little attention to the reuse of existing assembly plans. The coauthors of [19] discuss the design of assembly systems for modular products. An assembly line is divided into the basic and variant subassembly lines, such that the basic subassembly line is used for common and basic operations, whereas the variant ones for variant operations. The concept of process platforms is introduced in [20] to facilitate the coordinated product and process variety management. To support computerized production process derivation, the principle of group technology is adopted to build coding schemes for the set of process variants in relation to product variants in [21]. Jiao et al. of [22] study the modeling of process variety using object-oriented Petri-Nets with changeable structures for supporting production configuration. Shierholt of [7] presents the concept of process configuration that combines the principles of product configuration and process planning and is de facto an alternative of computer aided process planning.

3. PROCESS PLATFORM The general gist of platform thinking in [8] is to develop product families and the associated process families so as to produce high variety while maintaining economies of scale and scope. A comprehensive review of platform-driven development can be found in both [23] and [3]. Genetic Structures (a) Genetic Product Structure (GPdS)

(b) Genetic Routing Structure (GRS) AP4

(W AP4, TAP4,SAP4)

,SAO

) )

AP1

MP2

AP4

AP4

AO1

AP4 1

AP4 2

AP4 2

Variety Parameters

MP1 ( WMP1,

(W MP1, TMP1,TMP1)

(W MP2, TMP2,T MP2)

(W ,T (

(W AP1, TAP1,SAP1)

(W AP2, AP2 TAP2,TAP2)

TAP4,SAP4)

AO1AP 4 WAP4 ,TAO ,SAO

AO

(W AP3, AP3 TAP3,TAP3)

AP 4 2

End EndProduct Product(P) (P)

(WAP4,

AP4

XQA1 A1 A1

XQSA12 A12 A12

XQ2 I2

XQCI XQA11 XQC5 C1 A11 C5 R2 A11

XQC6 C6

XQI1 I1

TMP1,SMP1)

MP1 1

MO

(W ,T MP1

MP1 MO 1

,SMO

MP1 1

Customer Order:

)

XQC2 XQC3 XQC4 C2 C3 C4

MP1

R1

Variety Parameters C3.include=0 I2.include=0 I1.shape=round

Material Requirements

Genetic Planning

Product Variant * End End Product Product(P (P*))

(c) Genetic Process Structure (GPcS) End EndProduct Product(P) (P)

X1 * A1 A1*

AP4 (W AP4,TAP4,TAP4) XQI2 I2

XQA1 A1 A1 (W AP3,TAP3,TAP3)

MP2 XQCI C1

XQA11 A11 A11

XQC5 C5

XQI1 I1

(W MP1,T MP1,SMP1) MP1

XQC2 XQC3 XQC4 C2 C3 C4

Raw Material (R)

X4 I1*

R1

(d )

GR S In st an rie tia tio C 3 ty P a n .in ra I1 I2.in clud me .sh clu e te r ap de =0 s e= =0 ro un d

R1

(e) Process Variant

Sequence

MO Machining Operation

Material Requirement AO

Purchased Part (C)

AP Assembly Process

Intermediate Part (I)

MP Manufacturing Process

Assembly Operation

X – Quantity Per

(W AP4*, AP4* 6.4,SAP4*)

AP4* (W AP4*,

6.4,SAP4*)

,2.8,SAO

)

4 WAP4* ,3.6,SAO AO1AP *

(W

)

AP 4 2*

AO

AP3 (WAP3, 14.0,SAP3)

Assembly Assembly(A) (A)

X3 C6

X4 C4

Va

AP2 (W AP2,T AP2,SAP2)

X4 C5

AP1

XQC6 C6

R2

X1 * A11 A11*

X2 C2

(W AP1,T AP1,T AP1) (W MP2,T MP2,TMP2)

AP3

X2 C1

XQA12 A12 A12

X1 * A12 A12*

AP2* (WAP2*, 7.8,S AP2*)

(WAP1, 11.5,SAP1) AP1

(W MP1*, 25.6,SMP1*) MP1*

AP4*

(

AP4 1*

AP4 2*

MP1* (WMP1*,

25.6,SMP1*)

(

1 W ,25 .6,SMO MO1MP MP1* *

MP1 1*

)

S – Setup

WC – Work Center T – Cycle time

Figure 1: Concept Implication of a Process Platform A process platform involves de facto three aspects: (1) a common process structure shared by all process variants; (2) derivation of specific process variants from the common structure; and (3) correspondence between product and process variety, which resembles the correlation between the generic product and routing structures. In this research, we approach to the above three issues by generic structures, generic planning, and variety parameters, respectively. Figure 1 illustrates the concept of a process platform. As noted, each generic or specific process, may it be a manufacturing type or an assembly one, contains one or more than one ordered operations. For example, AP4 , the generic assembly process for forming the family of end products, involves two generic assembly operations AO1AP 4 and AO2AP 4 , while MP1 , a generic manufacturing process, only has one generic machining operation MO1MP 1 . The cycle time and setup of a process are the aggregation of these of its operations. In case, each process only contains one operation, the processes can be replaced by operations. 3.1 Generic Variety Representation

2

Copyright © 2005 by ASME

To describe a large number of product item variants (end products, assemblies, intermediate parts, and raw materials) and process item variants (operations, sequences and processes) with minimal data redundancy, the generic representation is adopted [13]. An item is generic in the sense that it represents a set of similar items (i.e., variants) of the same type (i.e., a family). Instead of using part numbers (so called direct identification), the identification of individual variants of a generic item is based on variety parameters and their instances (a list of parameter values). This is referred to as indirect identification [24]. Such an indirect identification entails a type of class-member relationships (exhibiting a meta-structure) between a family and its variants [16]. In this way, generic variety representation facilitates the specification of feasible variations of the items (products and processes) with respect to optional and alternative values of variety parameters. 3.2 Generic Process Structure Product data can be represented by a BOM that is used for an end product to state raw materials, intermediate parts and assemblies required for making the product. On the other hand, production information is concerned with how a product is built, that is, the specification of processes, operations and their sequences to be performed along with related resources such as work centers/machines, labors, tools, fixtures and setups. Similar to describing a product structure using a BOM, an operations routing is usually used to represent the production structure for a given product [16]. A process platform is underpinned by two generic structures. While a generic product structure (GPdS) [25] shown in Figure 1(a) represents the set of product variants in the family, the related production processes can be generalized as a generic structure of standard routings (GRS) shown in Figure 1(b). These standard routings form the basis of various process variations in consequence of product variety. The relationships between the product structure (i.e., BOMs) and the routing structure are embodied in the materials required by particular production operations. The link between BOM and routing data can be established by specifying each component material in the BOM as required by the relevant operation of the process for making its parent product [26]. Through these links, the GPdS and the GRS can be synchronized into a unified generic structure, called generic process structure (GPcS). Therefore, as conceptually described in Figure 1(c), the GPcS distinguishes the common structure of the process platform, from which process variants are derived in the same way according to given product data. While the GPdS associates each component material directly with its parent product, a component material in the GRS is associated with the relevant process or operation in the GPcS for producing its parent component. For each manufactured end or intermediate product component, a singlelevel GPcS can be derived by specifying the sequence of processes or operations required for producing that component

in connection with materials and resources including workcenters, cycle times, and setups required for each operation. The multi-level GPcS can be composed by linking the single-level GPcSs of lower-level intermediate parts through the processes or operations that require them. For example, in Figure 1, assume a variety parameter, shape, and its value set, {normal, special}, are associated with a generic component, I1. The generic identification of I1 family is described as a set, I 1 ≡ {I 11* , I 1*2 }. Thus two I1 variants can be identified using this variety parameter, i.e., and I 11* = {I 1| I 1.shape = "normal"} I12* = {I1 | I 1.shape =" special" }. The corresponding process variation for producing I1 family involves two variants identified from a generic manufacturing process MP1 that has a generic machining operation MO1MP 1 . To make I 12* , a particular workcenter (WMP1 ), other than that (WMP1) in the standard routing for making I 11* , has to be employed, in which the cycle time and the related setup (same as these of the operation variant MO1MP1 which is identified first) become different from those for making I 11* (change from 12.5 and FMP1 to 25.6 and FMP1*, respectively). *

*

3.3 Generic Planning Generic planning is introduced to determine specific process variations in standard processes in order to accommodate diverse product variants in a family. Within a process platform, the variation of an operation thus the related process results from the differences in product item variants to be processed by this operation. Therefore, derivation of process variants from the GPcS becomes the major concern in generic planning. Taking advantage of the meta-structure inherent in the generic variety representation, variant derivation can be implemented through the instantiation of a GPcS with respect to the given values of particular variety parameters transformed from customers’ individual needs, as shown in Figure 1(d). Under the umbrella of a GPcS, not only the GPdS and the GRS are unified by the material requirement links, but also they employ exactly the same set of variety parameters and their values to handle variety [16]. Thus the class-member relationships between generic items and their variant sets can be consistently used for process variant derivation. In addition, the correspondence between product and process variety can be maintained throughout the variation of both product structures and process structures. As shown in Figure 1, the same set of parameters, {C3.include, I2.include, I1.shape}, is used in generic planning for deriving the process variant in response to the specified product of a particular customer order, that is, Variety parameters and their values: {C3.include=0, I2.include=0, I1.shape=”special”}; Product variant specification: I 1    → I 1 , I 2 → ∅ ,  ;   I1.shape =”special”

C3.include = 0 C3  → ∅

3

*

I2.include = 0



Copyright © 2005 by ASME

Each PI i , ∀i = 1,L , N PI consists of a set of ordered

Process variant derivation: MP1 → MP1 , MP2 → ∅,  . =0   AP2 C3.include  → AP2* , AP4  → AP4*   I1.shape=”special

*

I2.include= 0

Prior to the process variant derivation, the product variant is specified, e.g., based on the product platform of the product family, according to the customer order. During the derivation process, every generic item, more precisely generic process items, involves an instantiation process, thus giving rise to a coordination issue among different types of instances of multiple generic items. Jiao et al. of [16] propose a generic variety structure to coordinate multiple variants in regard to parameter values when exploding a GPdS. Accordingly, rules and constraints are introduced to define relationships between generic items (in and between product and process types), and between parameter values of child and parent product items for the specification of variants of machines, operations and processes. The rules and constraints should guarantee for given specifications of each product item, valid variants of related generic operations and the process variant are generated through the derivation process. 4. PRODUCTION CONFIGURATION A process platform, Ω , contains a set of production process variants, {Pi }P , for producing the set of product variants in a family. It is defined as a tuple Ω = SPI , » .

SPI = {PI i }N is a set of process classes each of which is for producing a family of product items, may it be a part type or an assembly. Different valid configuration of these items forms product members of the family. » is the sequence relation between two process classes in SPI such that PI i » PI j , ∀i ≠ j = 1,L , N PI indicates PI i should be completed PI

before the commencement of PI j . Furthermore, a transitive closure of » is reflexive so that with SPI , the set of » forms trees. With respect to the associated product item families, SPI can be further classified into two sets, i.e., is a set of SPI = SPI M ∩ SPI S , where SPI M = {PI iM }N master process classes that are compulsory to all process variants, SPI S = {PI iS }N is the set of selective process classes that are optional to process variants. A process class, PI i , ∀i = 1,L , N PI , contained in the process platform for producing an item family, can be classified into one of the following three categories: (1) a type of manufacturing process consisting of a series of submanufacturing processes (including machining operations and non-machining operations, e.g., material transfer) for manufacturing a part family, (2) a type of assembly process consisting of a series of subassembly processes (including assembly operations and non-assembly operations) for producing an assembly family, and (3) a mixed process involving the above two for making an assembly family. PI M

PI S

SO PI = {OmPI

operation classes

i

i

}

NO

PI i

PI i = SO PI , f ,

, i.e.,

i

where f is the precedence relation between two classes in SO PI such that OmPI f OnPI , ∀m ≠ n = 1,L , N O suggests PI i

i

that OmPI

i

i

i

should be performed before OnPI . In addition, a i

transitive closure of f is reflexive so that with SO PI , the set of f forms trees. In SO PI of PI i , ∀i = 1,L , N PI , two sets are i

i

where distinguished, i.e., SO PI = SMO PI ∩ SSO PI , PI PI is the set of master operation classes SMO = {MOm }N i

i

i

i

i

MO PI i

, SSO PI = {SOmPI }N selective operation classes optional to {Pi }P .

{P }

necessary to all

i

i

i P

SO

PI i

is the set of

A process variant, Pi , ∀i = 1,L, P , consists of a series of

sequenced processes, i.e., Pi = spii* , » , where spii* = {piij* }P×M is

a set of processes each of which is to produce a specific item of the product variant, » is the sequence relation between two processes in spii* , piia* , piib* , piia* ≠ piib* ∈ PI i , ∀i = 1,L , N PI , a ≠ b = 1,L , M , such that piia* » piib* indicates the process piia* for an item, e.g., Ia, must be completed before producing item Ib by piib* . Each piij* , ∀j = 1,L , M

of

a

production

process

variant,

Pi , ∀i = 1,L, P , contains of a set of operations, may it be a machining type or an assembly, i.e., piij* , ∀j = 1,L , M = SO , f , where SO piij*

piij*

{ }

= Ok

piij*

N SO

pi* ij

is a set

of specific operations, f is the precedence relation between pi operations in Ok N . If piij* , ∀i = 1,L , P , j = 1,L , M is for

{ } * ij

pi* SO ij

{ } piij*

manufacturing a part, then Ok

N SO

pi*ij

is a set of machining

operations, if pi , ∀i = 1,L , P , j = 1,L , M is for producing an * ij

pi*ij

assembly, then Ok , ∀k = 1,L , N SO can be a machining operation or an assembly one. A type of manufacturing or assembly process producing a family of item variants (either a part type or an assembly one) employs a set of machine classes, each of which, in turn, have a number of similar machines, a number of material handling devices (i.e., material handlers) and a number of buffers. In the production of product items or final products, transferring materials, semi-finished items (i.e., WIP), and finished items from one location to another, the involved material handlers may be of AGVs, robots, or human operators. To keep the smoothness of manufacturing and assembly processes, a number of buffers, including input buffers, WIP buffers and output buffers, are used to store materials, WIP and finished items/products, respectively. The underlying principle of process platform-based production configuration is to select the set of process concepts first. The selection is accomplished referring to the hierarchy of the given product. Production knowledge, i.e., rules and pi*ij

4

Copyright © 2005 by ASME

constraints, guides the process selection. Meanwhile, the precedence relations among selected processes are also determined. Then, recursively decompose each assembly item till machined parts are reached. For any assembly at each decomposition level, processes, execution sequences and required resources for its child assemblies and parts are specified at the immediate lower level. Corresponding to parts at the lowest level of the decomposition paths of each assembly, the set of appropriate manufacturing resources, machining operations and their execution order are determined. The decisions made on the selection of process elements are in agreement with the set of variety parameter values of the parts. Figure 2 illustrates the set of elements of a process platform in relation to production configuration. Master Process

Master Operation Machine

Necessary Machine Selective Machine

Selective Operation

Input Buffer

Process Platform configure

Operation

Process of End Product

Process of Specific Item

Precedence

produce

A type of

Buffer

xxx

Sequence Assembly Process

Precedence

: family

: association

End Product

Assembly Process Manufacturing Process

Common Item

Part produce

Item Optional Item

WIP Buffer Legends:

Manufacturing Process

Process of Item Family

Operation

use

produce

A type of

Material Handler

Operator

Output Buffer

Master Sequence

Selective Sequence

Resources

AGV Robot

Selective Process

: a number of classes : generalization

Assembly

: a number of single entities

: composition

: a type of

: instance of

Figure 2: Platform Elements associated with Production Configuration 5. MODELING SUPPORT TO PROCESS PLATFORMBASED PRODUCTION CONFIGURATION The successful implementation of mass customization necessitates the automation and computerization of process platform-based production configuration. Therefore, it is necessary that the process of configuration be transparent for developing such computerized applications. Hence, it raises the importance in the formal modeling of production configuration, i.e., to clarify how a complete production process can be configured based on a process platform. One important issue in modeling is the understanding of the characteristics of systems or processes to be modeled so as to design or select proper modeling tools. In production configuration, the major characteristics are summarized as follows. (1) Variety involvement. For fulfilling, as many as possible, customers’ expectations, the number of product components is rather high. Inevitably, their production creates a large variety of operations, precedence relations and manufacturing resources. Thus, in process platform-based production configuration, a great attention requires to be paid to diversified varieties regarding process, product and resources. (2) Process changes. The frequent design specification changes to the customized products cause recurrent variations

in production. Exhibited by changeovers in operations, sequences and resources, process differences, especially structure changes, must be explicitly considered in production configuration in the way that a more appropriate production process can be obtained. (3) Granular to different levels of specification. To lessen difficulties in focusing on all details at one time, which is not possible especially in large scale configuration, production configuration adopts the strategy of problem decomposition. The production process to be configured for an end product is broken down into a number of process concepts according to the hierarchy of the given product, which are again subdivided, etc. These process concepts are specified for the associated product items at each level of the product hierarchy. Refinement of each process is made at the lower level of decomposition. (4) Constraint handling. Due to the involvement of heterogeneous varieties, compatibility issue or constraint satisfaction is a major concern in production configuration. Three types of constraints are observed in production configuration. The first type constraints, i.e., inclusion conditions, specify the circumstances under which the processes and operations are to be included in a configuration. Constraints of the second type tackle the interrelations among processes (operations) and determine which processes (operations) to be completed before the commencement of others. The constraints of the third type, i.e., execution rules, specify the operation details, e.g., machines to be used and estimated cycle times with respect to specific items. The essential requirements for the modeling tools are raised from the above characteristics of production configuration and summarized as follows. (1) The modeling tool should have the ability to accommodate the involved large varieties in the way that a concise and easy understandable model can be built. (2) The modeling tool should have the ability to handle process changes so that the built model can be adapted, without any difficulties, to different configurations. (3) The modeling tool should have the ability to address issues concerning concept selection first, and then granular refinement of these selected concepts till all details are worked out. (4) The modeling tool should have the ability to deal with multiple constraints at different granularity. Bearing in mind the issues involved in production configuration and the requirements of modeling tools, we propose a multilevel system of nested colored object-oriented Petri Nets with changeable structures (NOPNs-cs) as the modeling formalism. The principles of colored Petri Nets (CPNs) [27], object-oriented PNs (OPNs) [28] and the mechanism for handling structure changes in PN models [29] are adopted to define the nets in the proposed modeling formalism. The relevant data regarding product item, process elements and manufacturing resources is attached to colored tokens in CPNs to tackle multiple configuration constraints. Besides, together with OPNs, they deal with the large and

5

Copyright © 2005 by ASME

various varieties involved. The change handling mechanism is intended to address the modeling of process variations. A concept of net nesting is introduced for addressing the issues of specifying process details at different levels, that is, lower level nets are nested in the places of higher level nets. In the proposed formalism, a resource net ( RNet ) is specified to reflect the internal behaviors of physical objects (i.e., the set of manufacturing resources); a manufacturing net ( MNet ) is defined to reflect both the manufacturing process of parts and parts themselves when it is nested in a place of the higher level net; an assembly net ( ANet ) is introduced to represent the process of producing assemblies, similar to the MNet , it is also used to indicate the produced assemblies when it is nested in a place of the higher level PN; the process net ( PNet ) is used to describe the abstract production process of the final product, it includes a set of conceptual processes selected for major product items at the first level of the product hierarchy, the precedence relations among them and the required manufacturing resources. p4 p1

Level 1 {PNet}

t4

t1

t2

p2 t5

t3 im1

p3

p5

at1

sp1

at 2

om1

sp2

p1 p2

 ANets, 

, Level 2  MNets  RNets 

p4

dt 11

dt 43

dt 13 dt 23

dp1

p6

dt 42

dt 41

dt 21

p5

im1

at1

dt 22

t3

sp1

at2

sp3

at 4

om2

request can be passed to the object represented by the place, pi , unless the object is not occupied, i.e., there is no token in the place. The multilevel nested net system ( NNSys ) is specified to reflect the complete production configuration based on a process platform. It provides abstraction mechanisms for process engineers focusing on selected conceptual processes to work out details while without being distracted by other details of the remaining. In an NNSys , the highest level is the PNet . A number of RNets , MNets and ANets are located at the second level. Each of these nets provides more details for the corresponding places in the PNet . Similarly, at all the following levels, nets in the lower levels provide further descriptions of the assembly and manufacturing processes nested in places in immediate higher level nets. At the lowest level of each path originating from the places representing ANets in the PNet , all nets are RNets , whilst a mixture of RNets , MNets and ANets can be found at any arbitrary level in between the highest level and the lowest level. Figure 3 gives an example of an N+2 level nested net system for production configuration of an end product with an N level hierarchy. Due to the space issue, not all of the nested MNets , ANets and the encapsulated RNets are given in the figure.

om1

sp2

p3 p8

at 3

im2

dp4 dp2

t5

dt 12

resource objects. Inh( pi , t j ) = 1 implies that no operation

p7

im2

at 3

sp3

at 4

om2

p4 p1 Level i an ANet

p3 dt25

p1 at1

im1

at 2

sp1

Level N+1

{MNets , RNets}

om1

sp3

dt12

at 4

om2

p2

at1

dt41 dt24

sp1

at 2

at3

sp3

at1

sp1

om1

:decomposed state place :input AND relation

at1

om2

sp1

at 2

t3

p3

p5

p3 at 2

om1

at 3

sp3

at 4

at1

om2

sp1

at 2

om1

sp2

om1 im2

at3

sp3

at 4

t5

Level i+1 a nested MNet

p4

transition

:output AND relation

:place and the represented ANet

dt11

dt23

dt13

dp1 dt12

dt 21

px

:port place

px px

:place

t x :transition

:socket place

:place and the represented MNet

at x:activity

:inhibitor arc

dt42

dt 22

t3

p2 p6

:colored token

dp4 dp2 dt41

om2

sp2

dt xx :decomposed

dt43

p1 im1

im2

im1 at 4

t5

om1

p7

om1

sp2

im2

at2

dt22 dt 23

sp2 at1

sp1 sp2

dt 42 dt43

sp2 im1

at1

dt44

p5

im1

at 2

sp1

im1

dp4

dp 2

t2

p2

p6

p8

Level N+2

dp x

dt 21

p4

im1

{RNets}

dt13

dp1

t5

sp2 at 3

im2

dt11

t4

t1

:arc

sp x

p5

:state place

:place and the encapsulated RNet

px

transition

:place

t x :transition

:Input AND

:inhibitor arc :Inhibitor arc

:arc dp x :decomposed state place :Colored token

dt xx:decomposed transition

:place and the represented MNet

:Socket place

:Port place

:Message sending & receiving between port and socket places

Figure 3: A System of NOPNs-cs for Production Configuration modeling To clarify the firing conditions for transitions with respect to firing colored tokens in an MNet , ANet or PNet , each of such transitions that input arcs have OR relations is decomposed into several input transitions, a state place and an output transition. In the nets, a single resource object (i.e., the number of such object is one) may have more than one input arc. Thus, conflict may occur when multiple objects or subprocesses require such a single object to perform operations at the same time. To maintain 1-bounded property and the safeness of an object place, the inhibitor arcs [30,31] are applied to these

Figure 4: Illustration of Port and Socket Places To enable the communication through sending and receiving messages between objects at two adjacent levels, the port places in the lower level nested nets and socket places in the higher level nets are introduced. They are only defined for resource objects. The specification of port and socket places attempts to address the connection between lower and higher level nets and thus the continuity of the modeling from the lowest level to the highest level. For example, as shown in Figure 4 ( two levels in a NNSys ), when a token representing a part is produced in the MNet at level i+1, which is nested in

6

Copyright © 2005 by ASME

place p 2 in the ANet at level i, and loaded into the output buffer represented by place p 6 , a token with the same color appears in place p 2 in the ANet at level i. Meanwhile, a message requesting machine setup from the output buffer p 6 in the nested MNet is sent to the place p 3 representing an assembly machine in the ANet at level i. 6. PROCESS PLATFORM CONSTRUCTION In manufacturing practice, a large amount of production information and process data are available in an organization’s databases. In addition to reducing development risks thus saving time and cost, the knowledge reusing from historical data also facilitates the handling of process variety and tradeoffs between design changes and process variations. Recently, due to the rapid increasing of data amount, data mining has been well recognized for decision support by efficient knowledge discovery of previously unknown and potentially useful patterns of information from past data. Thereby, the construction of a process platform, de facto, is to identify the underlying GPcS from existing production process data. Taking advantages of widely used data mining tools, data mining techniques is thus propsed to solve the GPcS identification problem. The sequenced operations suggest production processes (PROCs) can be represented by tree like precedence graphs, i.e., tree representations [9]. While operations details are embedded in nodes of a PROC, the sequences information is reflected by the tree structure. Thus two types of data with respect to textual data and structural data are included in the GPcS identification. For this reason, we put forward a systematic data mining methodology by integrating text mining and tree matching techniques to solve this unique data mining problem. The methodology includes three stages: (1) PROC similarity measure, (2) PROC clustering, and (3) PROC unification. 6.1 PROC Similarity Measure This step deals with measuring similarity of the set of PROC variants in a family. The PROC similarity measure can be simplified as node content similarity measure and tree structure similarity measure because of the classification of textual data (i.e., operation details) and structural data (i.e., operations sequence). 6.1.1 Node Content Similarity Measure In a process family with P members, each PROC variants is composed of a number of instances of a set (or subset) of N operation types. Accordingly, the node content similarity measure of two PROCs is given as the sum of their operation similarity measure. Because an operation is described by three aspects, namely materials, product and resources, comparison of two operations can be measured as material similarity, product similarity and resource similarity.

In an operation, materials (excluding connector, jointers and other connecting devices) contribute differently to the functionality of their parent items. To reflect the different importance levels of materials, a weight value should be assigned to each material component. Thus, the material similarity of two same type operation variants is given as the weighted sum of these of all their material components. With respect to the two kinds of material components, including primitive components (i.e., raw materials and purchased parts) and compound components (i.e., machined parts and assemblies), different approaches are adopted to measure their similarity. For the primitive components, we propose the use of text mining techniques, while for compound components, weighted bipartite matching is employed. The major procedure of comparing primitive components can be summarized as follows. Data file preparation and component description deals with pre-processing raw data so that the mining tools can work on. The different types of primitive and compound components extracted from operation nodes are saved in different files. Each file is created for each family. For describing a component, two types of attributes can be distinguished: nominal type and numerical one. For an easy identification of the right attributes, a single value of nominal attribute may be used, while the attribute-value pairs of numerical attributes are adopted to describe a component. The raw materials are described exactly the same as the parts to be machined. The subsequent data file analysis concerns the processing of prepared data by mining tools. Each time only one file is input to the mining tool for analysis. The result is a list of extracted keywords, such as attribute values, along with the occurrence counts. The aim of quantifying nominal attribute values is to convert their text format values to numerical values (ranging from 0 to 1) for similarity comparison. Attribute weight calculation addresses the obtainment of relative importance of each attribute (i.e., weight) based on the extracted occurrence frequencies. The following attribute similarity measure attempts to calculate similarity of two attributes based on the distance of their value instances in two components. With attribute similarity available, the similarity of two components is measured as the weighted sum of attribute similarity. Finally, with the presence of the pairwise comparison of same type components, a PxP matrix can be established to record the obtained similarity. Repeating the above procedure for other primitive component files, a number of PxP matrixes is constructed in the same way. After the construction of the primitive component similarity matrix, weighted bipartite matching [33] is carried out to measure similarity of compound components of the same type. Similarly, after the calculation, a number of N compound component similarity matrixes can be constructed. Products of operations are of compound components. Therefore, product similarity is the same as that of compound components. As described by machines, cycle times and setups, the resource similarity measure of two operation variants is computed as a weighted sum of similarity measures regarding the three

7

Copyright © 2005 by ASME

attributes. To obtain machine similarity, cycle time similarity and setup similarity, the text mining procedure same as that of measuring component similarity is applied to the extracted variants of the three attributes, respectively. At last, a total number of N resource similarity matrices are constructed for all the operations. With the availability of similarity of materials, products and resources, the similarity of operations is computed as the sum of the three. Subsequently, node content similarity of two PROCs is calculated as the sum of the similarity of same type operations. For a relative measure (i.e., between 0 and 1), the node contend similarity measure is normalized using the maxmin normalization method. After the normalization, a PxP node contend similarity matrix is established to document the pairwise comparison of PROC variants.

distances are measured. For a consistent comparison, the maxmin normalization method is employed to normalize the above absolute tree structure distance values. Subsequently, tree structure similarity can be calculated. Finally, a PxP matrix can be established for pairwise tree structure similarity of the PROC family.

6.1.2 Tree Structure Similarity Measure Tree structure similarity measures the degree of commonality of two PROCs in terms of their operations sequences (i.e., the arcs of precedence graphs). To deal with such structural data, the tree matching technique is applied. The procedure proceeds as follows. The first step is to determine the base PROCs between two PROCs being compared. Owing to the symmetric property of distance measure and cyclic representation of a partial order (an PROC is a partial order) [9], the pairwise comparison of all tree pairs of two PROCs can be simplified to merely compare an arbitrary tree of one PROC (referred to as a base PROC) with all representation trees of the other one being compared. For reducing the total number of tree comparisons, the one with the higher number of representation trees between two PROCs is specified as the base PROC. For easing the pairwise comparison, a table should be established for recording PROCs according to the ascending order of the numbers of their representation trees. In the PROC table, except the first one, all of the representation trees of the following P-1 PROCs are generated for the succeeding comparison by the construction of tree edit graphs, in that in the pairwise comparison, one process only needs to be compared with processes that follows it. The basic principle of tree matching is to compare two trees based on tree transformation – to transform one tree to exactly the same as the other one. For the most accurate tree structure similarity measure of two PROCs, the tree edit graph is employed through providing an indirect way of tree transformation. To facilitate comparisons based on a consistent common ground, the same cost value is assigned to each tree editing operations represented by arcs in the graph. From the top-left corner to the bottom-right corner in a graph, the shortest path with the minimum number of arcs takes fewest editing operations and thus the minimal transformation cost, i.e., the distance of two trees. The distances of tree pairs of two PROCs can be obtained by executing the above process. The tree distance measure of two PROCs defined as the minimum distance among all obtained from the comparison of tree pairs. Repeating the process for all PROC pairs, their structure

6.2 PROC Clustering PROC clustering aims to group a set of individual processes into classes of similar ones. Considering the complex data types involved in processes, this research adopts a fuzzy clustering approach. In comparison with the k-means method, fuzzy clustering partitions PROC instances based on the similarity degree that is derived from the real data of production processes, rather than based on subjectively pre-defined clusters. The procedure of PROC clustering is as follows. The first step is to define a fuzzy compatible relation R as similarity measures for the given PROC set Ω . R should be constructed in a matrix form such that it is identical to a PROC similarity matrix, that is, R is a compatible matrix. The second step is to construct a fuzzy equivalence matrix. The fuzzy compatible relation R is a fuzzy equivalence matrix if and only if the transitive condition can be met. For converting a compatible matrix to an equivalence matrix, the continuous multiplication method is implemented. Thirdly, a λ-cut of the equivalence matrix should be determined. The λ-cut is a crisp set that contains all the elements of the set Ω , such that the similarity grade of R is no less than λ .Then each λ-cut is an equivalence relation representing the presence of similarity among PROC instances to the degree. Finally, the PROC clusters can be identified based on the equivalence matrix by adopting the netting graph method.

6.1.3 PROC Similarity Measure As note content similarity and tree structure similarity are two independent measures, the overall PROC similarity is thus suggested to be measured by an Euclidian distance rather than a simple sum. Repeat PROC similarity calculation for all the PROCs in the family. Then normalize the obtained PROC similarity value. A P × P matrix is established for pairwise PROC similarity.

6.3 PROC Unification PROC unification attempts to unify all members of an PROC cluster into a GPcS. The GPcS is formed by maintaining a valid tree structure through a tree growing process. The formation of a GPcS involves four major steps, including assorting basic process elements, identifying master and selective process elements, forming basic trees, and tree growing, as discussed below. Basic process elements refer to operations and precedence. The assortment of elements by breaking down each PROC leads to a lead node set, an intermediate node set, a lead node arc set and an intermediate node arc set. Accordingly, the associated four types can be distinguished. The second step is to identify

8

Copyright © 2005 by ASME

the set of master elements, i.e., elements that are common to all processes, and the set of selective elements, i.e., elements that are optional to PROCs. The basic trees are defined as the trees with such structures that are assumed by groups of PROC variants in a cluster. The generalization of basic trees can simplify the tree unification as less number of trees to be unified, that is, the tree unification from individual PROC variants is converted to that of basic trees. Tree growing aims to form the generic tree by pasting all basic trees one by one. Thus, an initial generic tree, i.e., a seed, should be selected for growing. The basic tree with the longest path and the maximal number of intermediate nodes should be specified as the seed, in that such a comprehensive tree encompasses most production conditions occurring among the process family members. Then the initial generic tree starts to grow by unifying with the other basic trees. While all nodes representing operations are to be included in the final GPcS, arcs (i.e., operations precedence) cannot be directly added into the growing tree, because the addition of some arcs may damage the tree structure. Their addition is based on the result of evaluation, which is performed between the arc of the tree being unified and the associated arc in the growing tree. Upon the completion of the tree growing process, the formed GPcS consists of a generic tree structure and an additional arc set. Repeating the procedure, the GPcS for other clusters are obtained. Treating such formed GPcS of each cluster as member trees and applying the unification process to them leads to the GPcS for the entire process family. Similarly, the final GPcS includes a unified generic process structure and an extended additional arc set resulted from these of each cluster. Due to the presence of selective arcs in the generic tree, the GPcS is by no means the union of all member trees. 7. CASE STUDY The proposed concept of process platforms has been tested on a family of vibration motors for hand phones produced by a local company. Since every customer has their own needs and wants on the design of hand phones, the vibration motors matched with these diverse hand phone products are typical customized products. Further more, the slightly difference in motor’s design results in the production characterized by a huge number of variations including the changes to work centers, machines, tools, fixtures, and setup activities.

Operation Sequence

x1

x1

Purchased Part

x1

x1

Magnet

Shaft x1

Coil

x1

Tape

Rubber Holder

Mainbody

Weight

Amb

(WC-Aaa,5.30,ST)

Bracket Assy (WC-Aba,4.04,ST)

(WC-A fa,4.48,ST)

Aba

Afa

x1

x1

x1

Frame Assy

Aaa

Shaft

(WC-Amb,9.25,ST)

x1

x1

Armature Assy

Coil Assy

x1

x1

x1

Frame Magnet

Mba

Mf

Aca

(WC-M f,5.07,ST) x1

x1

Tape Commuter

x1

x1

Bracket A Bracket B Terminal

(WC-Mba,5.25,ST)

(WC-Aca,5.18,ST)

(WC-Mt,4.93,ST)

Mbb (WC-Mbb,5.25,ST)

Mt

RM

RM

x1

Coil Mc

RM

(WC-M c,4.52,ST) RM

RM

Figure 6: The GPcS of the Motor Family Table 1: Illustration of generic item and variety parameters Hierarchy Generic Level Parent Item .2 Bracket Assy (BAssy)

Generic Child Item Bracket A (BA)

Variety Parameter

PBA 0 Cardinal flag

PBA 1 Color

PBA 2 Shape

PBA 3 width .2

Bracket Assy Bracket B (BAssy) (BB)

PBB 0 Cardinal flag

PBB 1 Color

PBB 2 Shape

PBB 3 width .2

Bracket Assy Terminal (BAssy) (TL)

PTL 0 Cardinal flag

PTL 2

Rubber Holder

Width

Parameter Value Set

PBA* 01 =" not includeed " PBA* 02 =" includeed "

PBA* 11 = blue PBA* 12 = red PBA* 13 = black PBA* 21 =" T " PBA* 22 =" U " PBA* 23 =" L" PBA* 31 =" 5" PBA* 32 =" 6 " PBA* 33 =" 7 " PBB* 01 =" not includeed " PBB* 02 =" includeed " PBB* 11 = blue PBB* 12 = red PBB* 13 = black PBB* 21 =" T " PBB* 22 =" U " PBB* 23 =" L" PBB* 31 =" 5" PBB* 32 =" 6 " PBB* 33 =" 7 " PTL* 01 =" not includeed " PTL* 02 =" includeed " PTL* 11 =" 6 " PTL* 12 =" 8" PTL* 21 =" 5" PTL* 22 =" 7 "

x1

Bracket Assy

x1

Coil Assy

Operation x1

x1

Intermediate Part

x1

x1

x1

x1

Length

Armature Assy

Frame

Avm

Subassembly

PTL 1

Mainbody

x1

Frame Assy

ST – Setup X – Quantity Per WC – Work center

(WC-Avm,9.25,ST)

Raw material (RM)

Vibration Motor

Weight

Vibration Motor

Material requirement

x1

Bracket A

x1

Bracket B

x1

Terminal

x1

Commuter

The main parts of a vibration motor are rubber holder, weight, and mainbody, which in turn consists of armature assy, bracket assy, and frame assy. The BOM structure for a vibration motor is shown in Figure 5. The manufacturing process for vibration motors involves six assembly operations (Avm, Amb,

Figure 5: The BOM Structure of Vibration Motors

9

Copyright © 2005 by ASME

Aaa, Aca, Afa, Aba) and five machining operations (Mt, Mba, Mbb, Mf, Mc). Unifying product and production data, the underlying GPcS of the process platform for the motor family is presented in Figure 6. As shown, each process to form a product item only includes one operation type either machining type or assembly one. So we directly use operations rather than processes in the figure. To illustrate the concept of generic variety representation, Table 1 shows the generic item bracket assy, its child components. For a given customer order as shown in Table 2, the configured production process is illustrated in Table 3.

The production configuration model for the motor variant specified in Table 2 has been built. Due to the space issue, only the first two levels including the assembly processes of the final motor and mainbody as well as the level for the manufacturing process of coil is shown in Figure 7. Applying the deadlock detection algorithm in [30] to the model, the firing of the sequence of enabled transitions reaches the goal state. Therefore, the model is live and deadlock free. p4 p1

t2

p2

Level i the PNet

t5

Table 2: An individual customer order Order #: xxxxx

Customer Info.: xxxxxxx

Due date: xxxxx Volume: xxx

Delivery: xxxxxxx

p1 Level 2

Description: xxxxxxxxxx

p2

ANet & RNets

t5

* Vibration motor assembly WAvm 2

9.00

S 2* (TF2* )

40

Mainbody assembly

* WAmb 1

9.25

S1* (TF1* )

30

Armature assembly

* WAaa 1

5.00

S1* (TF1* )

20

Coil assembly

WAca* 2

5.12

S 2* (TF2* )

30

Coil fabrication

WM c*4

4.87

S 4* (TF4* )

* fa 3

4.36

S (TF )

Product Quantity Material (Comp. Item) (Comp. Item) per Weight 2* Rholder3* Mbody1* AAssy1* FAssy3* BAssy1* CAssy 2* Shaft 3* Coil 4* Tape2* Commuter3* Raw material

* ba1 * bb 6

5.10 5.12

S (TF ) S 6* (TF6* )

Magnet 2* Frame3* Raw material BA1* BB6* TL*4 Raw material Raw material

5.18

S 4* (TF4* )

Raw material

* 3

20

Frame assembly

WA

10

Frame fabrication

WM *f 3

5.08

S 3* (TF3* )

20

Bracket assembly

WAba* 1

4.04

S1* (TF1* )

10

BA fabrication

10

BB fabrication

WM WM

10

TL fabrication

* WM TL 4

sp x

:state place

om2

at x

:Output AND

t4

t1

t2 t3

:activity transition

:inhibitor arc

om1

sp1 sp2

at2

p2

p4

:Colored token

im1 at1

t5

at4

om2

:arc

:Port place

:place and the represented MNet

:Socket place

:Input or output message place

:place and the represented ANet

:place and the encapsulated RNet

Figure 7: The Model for the Production Configuration Using NOPNs-cs

50

* 1

p1

:Message sending & receiving between port and socket places

Setup

* 3

p3

at4 sp3

im2

om1

sp3

im2

t x:transition

at 2

at3

p4

at 2

at3

:Input AND

om1

sp1 at1

t3

sp1 at 1

:place

im1

sp2

sp2

px

at2

sp2

t2

im1

Level 6

MNet & RNets

om1

sp1 at1

t4

p6

Table 3: The production process for motor variant specified in Table 2 Work Cycle Time Center Sec./item

p3

im1

p5 t1

p3

Coil.Length = 4mm Coil.WindingMode = BFT Frame.Thickness = 4mm Frame.Length = 11mm Bracket A.Shape = O Bracket A.Color = Red Bracket A.Diameter = 4.5mm Bracket B.Shape = U Bracket B.Color = Red Bracket B.Width = 5mm Terminal.Length = 6mm Terminal.Pitch = 2.5mm Tape.Width = 3mm Tape.Color = Red Commuter.Thickness = 2mm Shaft.Length = 12mm Shaft.Material = PVC Magnet.OutDiameter = 3.5mm Weight.Radius = 2.5mm

Operation

t3

p5

Variety parameter and value pairs:

Sequence no.

t4

t1

* 1

Wmotor2*

1 1 1

Mbody1*

1 1 1

AAssy1*

1 1

CAssy 2* Coil 4* FAssy3* Frame3* BAssy1* BA1* BB6* TL*4

For a simple illustration, the data mining methodology for identifying the GPcS is applied to 30 variants in the motor family. The obtained node content similarity matrix, tree structure similarity matrix and process similarity matrix are shown in figure 8(a), 8(b) and 8(c), respectively. Based on the process similarity matrix in Figure 8(c), 4 groups are clustered as shown in Table 4. Applying the proposed tree unification approach to the RC1 in Table 4, the formed GPcS is shown in Figure 9. Table 4: The Identified Clusters for a Motor Family

1 1 1 N/A

PROC Cluster PC1 PC2 PC3 PC4

1 1 N/A 1 1 1 N/A

PROC Variants P1, P3, P10, P13, P14, P17, P20, P22, P25 P2, P4, P5, P6, P7, P8, P9, P11, P16, P18 P23, P26, P27, P28, P29, P30 P12, P15, P19, P21, P24

N/A N/A

* 1): Same sequence numbers indicate parallel operations 2): Setup means the preparation of different tools/fixtures, materials for making particular parts at particular work centers.

10

Copyright © 2005 by ASME

 1 .81 .88 .84 .80 .88 .81 .79 .75 .89 .79 .21 .84 .92  .11 .87 .89 .90 .18 .90 .83 .15 .09 .17 .77 .17 .15 .23 .07 .15

.81 1 .83 .85 .89 .83 .82 .76 .79 .86 .81 .21 .75 .75 .20 .78 .86 .85 .22 .88 .06 .80 .13 .14 .79 .36 .46 .38 .34 .26

.88 .83 1 .78 .82 .90 .79 .81 .77 .91 .79 .19 .82 .82 .13 .85 .91 .99 .20 .88 .13 .85 .11 .19 .77 .15 .13 .13 .09 .13

.84 .85 .78 1 .86 .85 .67 .81 .92 .83 .81 .15 .70 .78 .25 .80 .83 .82 .18 .83 .01 .85 .26 .11 .79 .50 .32 .48 .22 .48

.80 .89 .82 .86 1 .82 .78 .77 .80 .87 .88 .20 .74 .74 .21 .77 .94 .86 .23 .87 .12 .80 .14 .15 .86 .36 .44 .44 .36 .44

.88 .83 .90 .85 .82 1 .79 .89 .84 .91 .79 .27 .82 .82 .12 .92 .91 .92 .19 .88 .13 .92 .18 .19 .76 .34 .52 .46 .48 .34

.81 .82 .79 .67 .78 .79 1 .76 .73 .79 .79 .15 .90 .90 .05 .90 .87 .81 .11 .87 .05 .85 .14 .22 .74 .34 .36 .36 .32 .50

.79 .76 .81 .81 .77 .89 .76 1 .87 .92 .76 .22 .76 .83 .19 .76 .85 .84 .17 .82 .07 .85 .24 .05 .84 .44 .32 .54 .48 .36

.75 .79 .77 .92 .80 .84 .73 .87 1 .81 .79 .18 .73 .73 .22 .79 .82 .89 .20 .90 .05 .83 .12 .12 .77 .20 .18 .26 .14 .08

.89 .86 .91 .83 .87 .91 .79 .92 .81 1 .77 .24 .80 .87 .14 .75 .85 .84 .21 90 .07 .80 .12 .18 .90 .07 .09 .08 .13 .14

.79 .81 .79 .81 .88 .79 .79 .76 .79 .77 1 .20 .73 .73 .20 .67 .86 .70 .23 .83 .14 .80 .04 .04 .85 .28 .21 .13 .17 .07

.21 .21 .19 .15 .20 .27 .15 .22 .18 .24 .20 1 .15 .15 .90 .14 .18 .17 .87 .15 .89 .14 .87 .84 .15 .90 .78 .86 .82 .72

.84 .75 .82 .70 .74 .82 .90 .76 .73 .80 .73 .15 1 .90 .24 .90 .80 .81 .11 .90 .04 .87 .06 .08 .81 .06 .12 .06 .12 .20

.92 .75 .82 .78 .74 .82 .90 .83 .73 .87 .73 .15 .90 1 .20 .90 .80 .81 .11 .90 .18 .78 .09 .05 .69 .08 .14 .11 .14 .13

 1 .65 .95 .67 .64 .69 .65 .64 .67 .88 .69 .24 .85 .90   .21 .74 .88 .76 .82 .32 .32 .78  .51 .32 .74 .42  .41 .43 .40  .41

.65 1 .66 .79 .81 .78 .72 .74 .83 .64 .84 .43 .54 .54 .42 .82 .64 .86 .53 .63 .50 .57 .13 .51 .56 .11 .03 .03 .10 .03

.95 .66 1 .63 .66 .71 .64 .65 .68 .89 .69 .23 .84 .84 .21 .73 .89 .81 .33 .81 .31 .79 .51 .33 .74 .41 .41 .41 .40 .41

.67 .79 .63 1 .94 .93 .86 .92 .89 .62 .84 .62 .53 .59 .63 .83 .62 .84 .52 .59 .50 .61 .20 .51 .56 .07 .06 .05 .14 .05

.64 .81 .66 .94 1 .92 .90 .90 .83 .65 .87 .62 .56 .56 .62 .82 .70 .86 .53 .62 .51 .58 .13 .51 .61 .10 .03 .03 .11 .03

.69 .78 .71 .93 .92 1 .91 .95 .86 .68 .83 .63 .62 .62 .61 .90 .68 .90 .52 .63 .51 .67 .16 .52 .55 .15 .08 .08 .05 .13

.65 .72 .64 .86 .90 .91 1 .89 .80 .60 .83 .61 .67 .67 .61 .88 .65 .84 .51 .62 .50 .61 .13 .53 .54 .09 .10 .11 .08 .06

.64 .74 .65 .92 .90 .95 .89 1 .87 .68 .81 .63 .57 .62 .62 .81 .64 .86 .52 .59 .51 .61 .10 .50 .61 .02 .08 .08 .11 .05

.67 .83 .68 .89 .83 .86 .80 .87 1 .65 .91 .52 .60 .60 .53 .91 .66 .95 .62 .67 .61 .62 .21 .61 .59 .12 .15 .20 .14 .11

.88 .64 .89 .62 .65 .68 .60 .68 .65 1 .62 .34 .76 .80 .31 .63 .93 .67 .43 .89 .40 .84 .61 .40 .88 .51 .51 .51 .51 .50

.69 .84 .69 .84 .87 .83 .83 .81 .91 .62 1 .52 .60 .60 .52 .86 .68 .87 .63 .62 .61 .63 .20 .61 .64 .09 .17 .13 .15 .10

.24 .43 .23 .62 .62 .63 .61 .63 .52 .34 .52 1 .13 .13 .96 .51 .32 .52 .87 .22 .88 .22 .65 .86 .22 .65 .57 .62 .59 .52

.85 .54 .84 .53 .56 .62 .67 .57 .60 .76 .60 .13 1 .96 .10 .71 .76 .65 .21 .89 .20 .87 .40 .20 .84 .50 .50 .50 .51 .52

.90 .54 .84 .59 .56 .62 .67 .62 .60 .80 .60 .13 .96 1 .10 .71 .76 .65 .21 .89 .23 .83 .40 .20 .78 .50 .51 .50 .50 .50

.11 .20 .13 .25 .21 .12 .05 .19 .22 .14 .20 .90 .24 .20 1 .13 .21 .13 .86 .08 .75 .15 .77 .89 .06 .82 .74 .88 .91 .83

.87 .78 .85 .80 .77 .92 .90 .76 .79 .75 .67 .14 .90 .90 .13 1 .80 .86 .11 .87 .05 .84 .20 .20 .75 .20 .17 .32 .28 .28

.89 .86 .91 .83 .94 .91 .87 .85 .82 .85 .86 .18 .80 .80 .21 .80 1 .94 .29 .87 .08 .92 .16 .13 .89 .05 .17 .15 .12 .03

.90 .85 .99 .82 .86 .92 .81 .84 .89 .84 .70 .17 .81 .81 .13 .86 .94 1 .06 .85 .05 .87 .13 .13 .92 .10 .12 .19 .15 .11

.18 .22 .20 .18 .23 .19 .11 .17 .20 .21 .23 .87 .11 .11 .86 .11 .29 .06 1 .06 .86 .12 .98 .97 .22 .88 .81 .86 .90 .90

.90 .88 .88 .83 .87 .88 .87 .82 .90 .90 .83 .15 .90 .90 .08 .87 .87 .85 .06 1 .08 .91 .10 .08 .92 .10 .11 .08 .11 .12

.15 .06 .13 .01 .12 .13 .05 .07 .05 .07 .14 .89 .04 .18 .75 .05 .08 .05 .86 .08 1 .17 .87 .88 .24 .95 .86 .90 .87 .91

.83 .80 .85 .85 .80 .92 .85 .85 .83 .80 .80 .14 .87 .78 .15 .84 .92 .87 .12 .91 .17 1 .17 .10 .90 .07 .07 .13 .07 .07

.09 .13 .11 .26 .14 .18 .14 .24 .12 .12 .04 .87 .06 .09 .77 .20 .16 .13 .98 .10 .87 .17 1 .11 .11 .75 .83 .91 .77 .73

.17 .14 .19 .11 .15 .19 .22 .05 .12 .08 .04 .84 .08 .05 .89 .20 .13 .13 .97 .08 .88 .10 .11 1 .12 .75 .83 .91 .84 .73

.77 .79 .77 .79 .86 .76 .75 .84 .77 .90 .85 .15 .81 .69 .06 .75 .89 .92 .22 .92 .24 .90 .11 .12 1 .10 .21 .14 .13 .14

.17 .36 .15 .50 .36 .34 .34 .44 .20 .07 .28 .90 .06 .08 .82 .20 .05 .10 .88 .10 .95 .07 .75 .75 .10 1 .90 .91 .77 .80

.15 .46 .13 .32 .44 .52 .36 .32 .18 .09 .21 .78 .12 .14 .74 .17 .17 .12 .81 .11 .86 .07 .83 .83 21. .90 1 .88 .86 .83

.23 .38 .13 .48 .44 .46 .36 .54 .26 .08 .13 .86 .06 .11 .88 .32 .15 .19 .86 .08 .90 .13 .91 .91 .14 .91 .88 1 .85 .90

.07 .34 .09 .22 .36 .48 .32 .48 .14 .13 .17 .82 .12 .14 .91 .28 .12 .15 .90 .11 .87 .07 .77 .84 .13 .77 .86 .85 1 .86

.15  .26  .13  .48  .44  .34  .50  .36  .08  .14  .07  .72  .20  .13  .83  .28  .03  .11 .90  .12  .91 .07  .73  .73  .14  .80  .83  .90  .86  1 

.74 .82 .73 .83 .82 .90 .88 .81 .91 .61 .86 .51 .71 .71 .50 1 .64 .94 .61 .65 .61 .63 .24 .62 .57 .17 .15 .12 .10 .10

.88 .64 .89 .62 .70 .68 .65 .64 .66 .93 .68 .32 .76 .76 .33 .64 1 .73 .45 .87 .40 .90 .61 .41 .88 .50 .52 .51 .51 .50

.76 .86 .81 .84 .86 .90 .84 .86 .95 .67 .87 .52 .65 .65 .51 .94 .73 1 .61 .64 .61 .65 .19 .61 .69 .11 .12 .16 .14 .12

.32 .53 .33 .52 .53 .52 .51 .52 .62 .43 .63 .87 .21 .21 .86 .61 .45 .61 1 .30 .94 .31 .77 .92 .34 .66 .61 .64 .67 .67

.82 .63 .81 .59 .62 .63 .62 .59 .67 .89 .62 .22 .89 .89 .20 .65 .87 .64 .30 1 .30 .97 .51 .30 .91 .61 .61 .61 .61 .61

.32 .50 .31 .50 .51 .51 .50 .51 .61 .40 .61 .88 .20 .23 .81 .61 .40 .61 .94 .30 1 .32 .69 .95 .34 .71 .65 .67 .65 .68

.78 .57 .79 .61 .58 .67 .61 .61 .62 .84 .63 .22 .87 .83 .22 .63 .90 .65 .31 .97 .32 1 .52 .30 .96 .61 .61 .61 .61 .61

.51 .13 .51 .20 .13 .16 .13 .10 .21 .61 .20 .65 .40 .40 .58 .24 .61 .19 .77 .51 .69 .52 1 .31 .51 .81 .85 .89 .82 .80

.32 .51 .33 .51 .51 .52 .53 .50 .61 .40 .61 .86 .20 .20 .88 .62 .41 .61 .92 .30 .95 .30 .31 1 .30 .57 .62 .68 .63 .56

.74 .56 .74 .56 .61 .55 .54 .61 .59 .88 .64 .22 .84 .78 .20 .57 .88 .69 .34 .91 .34 .96 .51 .30 1 .61 .61 .61 .61 .61

.42 .11 .41 .07 .10 .15 .09 .02 .12 .51 .09 .65 .50 .50 .59 .17 .50 .11 .66 .61 .71 .61 .81 .57 .61 1 .96 .96 .90 .91

.41 .03 .41 .06 .03 .08 .10 .08 .15 .51 .17 .57 .50 .51 .53 .15 .52 .12 .61 .61 .65 .61 .85 .62 .61 .96 1 .95 .94 .93

.43 .03 .41 .05 .03 .08 .11 .08 .20 .51 .13 .62 .50 .50 .63 .12 .51 .16 .64 .61 .67 .61 .89 .68 .61 .96 .95 1 .94 .96

.40 .10 .40 .14 .11 .05 .08 .11 .14 .51 .15 .59 .51 .50 .66 .10 .51 .14 .67 .61 .65 .61 .82 .63 .61 .90 .94 .94 1 .94

.41 .03 .41 .05  .03 .13 .06  .05  .11 .50  .10  .52  .52  .50  .60  .10  .50  .12  .67  .61 .68  .61 .80  .56  .61 .91 .93 .96  .94  1  30 x 30

30 x 30

(a) .21 .42 .21 .63 .62 .61 .61 .62 .53 .31 .52 .96 .10 .10 1 .50 .33 .51 .86 .20 .81 .22 .58 .88 .20 .59 .53 .63 .66 .60

 1  .43  1  ..43  .43 43  .43  .43  .57  .86  .57  .29  .86  ..86 29  .57  .86  .57  .43  .71  .43  .71  .71  ..43 71  .57  .57  .57  .57  .57

       .14  .71  .14  .14  .71  .71  .14  .14 .71  .14  .29  .86  .29  .86  .86  .29  .86  1 1  1  1  1   30 x 30

.43 1

1 .43

.43 .71

.43 .71

.43 .71

.43 .71

.43 .71

.57 . 86

. 86 .29

.57 . 86

.29 .57

.86 .14

.86 .14

.29 .57

.57 .86

.86 .29

.57 .86

.43 .71

.71 .14

. 43 .71

.71 .14

.71 .14

.43 .71

.71 .14

.57 0

.57 0

. 57 0

. 57 0

. 57 0

.43 .71 .71 .71 .71 .71

1 .43 .43 .43 .43 .43

.43 1 1 1 1 1

.43 1 1 1 1 1

.43 1 1 1 1 1

.43 1 1 1 1 1

.43 1 1 1 1 1

.57 . 86 . 86 . 86 . 86 . 86

. 86 .29 .29 .29 .29 .29

.57 . 86 . 86 . 86 . 86 . 86

.29 .86 .86 .86 .86 .86

.86 .29 .29 .29 .29 .29

.86 .29 .29 .29 .29 .. 29

.29 .86 .86 .86 .86 .86

.57 .86 .86 .86 .86 .86

.86 .29 .29 .29 .29 .29

.57 .86 .86 .86 .86 .86

.43 .71 .71 .71 .71 .71

.71 .14 .14 .14 .14 .14

. 43 .71 .71 .71 .71 .71

.71 .14 .14 .14 .14 .14

.71 .14 .14 .14 .14 .14

.43 .71 .71 .71 .71 .71

.71 .14 .14 .14 .14 .14

.57 0 0 0 0 0

.57 0 0 0 0 0

. 57 0 0 0 0 0

. 57 0 0 0 0 0

. 57 0 0 0 0 0

.86 .29 .86 .57 . 14

.57 .86 .57 .29 .86

.86 .29 .86 .86 .29

.86 .29 .86 .86 .29

.86 .29 .86 .86 .29

.86 .29 .86 .86 .29

.86 .29 .86 .86 .29

1 .43 1 . 71 .43

.43 1 .43 .43 . 71

1 .43 1 .71 .43

.71 .43 .71 1 .14

.43 .71 .43 .14 1

.43 .71 .43 .14 1

.71 .43 .71 1 .14

1 .43 1 .71 .43

.43 1 .43 .43 .71

1 .43 1 .71 .43

.86 .57 .86 .86 .29

.29 .86 .29 .29 .86

.86 .56 .86 .86 .29

.29 .86 .29 .26 .86

.29 .86 .29 .29 .57

.86 .59 .86 .86 .29

.29 .86 .29 .29 .86

.14 .71 .14 .14 .71

.14 .71 .14 .14 .71

.14 .71 .14 .14 .71

.14 .71 .14 .14 .71

. 14 .57 .86 .29 .86 .71

.86 .29 .57 .86 .57 .43

.29 .86 .86 .29 .86 .71

.29 .86 .86 .29 .86 .71

.29 .86 .86 .29 .86 .71

.29 .86 .86 .29 .86 .71

.29 .86 .86 .29 .86 .71

.43 . 71 1 .43 1 . 86

. 71 .43 .43 1 .43 .57

.43 .71 1 .43 1 . 86

.14 1 .71 .43 .71 .86

1 .14 .43 .71 .43 .29

1 .14 .43 .71 .43 .29

.14 .43 .71 1 .71 .43 1 .71 .43 1 .43 .43 1 .71 .43 .86 0.86 .57

.43 .71 1 .43 1 .86

.29 .86 .86 .57 .86 1

.86 .29 .29 .86 .29 . 43

.29 .86 .86 .57 .86 1

.86 .29 .29 .86 .29 . 43

.57 .29 .29 .86 .29 .43

.29 .86 .86 .57 .86 1

.86 .29 .29 .89 .29 .43

.71 .14 .14 .71 .14 .29

.71 .14 .14 .71 .14 .29

.71 .14 .14 .71 .14 . 29

.71 .14 .14 .71 .14 . 29

. 14 .71 . 14 . 14 .71 . 14

.71 .43 .71 .71 .43 .71

.14 .71 .14 .14 .71 .14

.14 .71 .14 .14 .71 .14

.14 .71 .14 .14 .71 .14

.14 .71 .14 .14 .71 .14

.14 .71 .14 .14 .71 .14

.29 . 86 .29 .29 . 86 .29

. 86 .57 . 86 . 86 .57 . 86

.29 . 86 .29 .29 . 86 .29

.26 .86 .29 .29 .86 .29

.86 .29 .86 .86 .29 .86

.86 .29 .86 .86 .29 .86

.29 .29 .86 .86 0.86 .57 .29 .29 .86 .29 .29 .86 .86 0.86 .57 .29 .29 .86

.29 .86 .29 .29 .86 .29

.43 1 .43 .43 1 .43

1 . 43 1 1 . 43 1

. 43 1 . 43 . 43 1 . 43

1 . 43 1 1 . 43 1

.71 .43 .71 1 .43 1

.43 1 .43 .43 1 .43

1 .43 1 1 .43 1

.86 .29 .86 .86 .29 .86

.86 .29 .86 .86 .29 .86

.86 . 29 .86 .86 . 29 .86

.86 . 29 .86 .86 . 29 .86

0 0 0 0 0

.57 .57 .57 .57 .57

0 0 0 0 0

0 0 0 0 0

0 0 0 0 0

0 0 0 0 0

0 0 0 0 0

.14 .14 .14 .14 .14

. 71 . 71 . 71 . 71 . 71

.14 .14 .14 .14 .14

.14 .14 .14 .14 .14

.71 .71 .71 .71 .71

.71 .71 .71 .71 .71

.14 .14 .14 .14 .14

.14 .14 .14 .14 .14

.29 .29 .29 .29 .29

.86 .86 .86 .86 .86

.29 .29 .29 .29 .29

.86 .86 .86 .86 .86

.86 .86 .86 .86 .86

.29 .29 .29 .29 .29

.86 .86 .86 .86 .86

1 1 1 1 1

1 1 1 1 1

1 1 1 1 1

1 1 1 1 1

.14 .14 .14 .14 .14

.71 .71 .71 .71 .71

(b)

(a): similarity matrix of node content (b): similarity matrix of tree structure (c): similarity matrix of routing

(c)

Figure 8: Obtained Similarity Matrixes for Node Content, Tree Structure and Process

Inclusion Conditions:

FinalMotorAssembly Cluster1(N=9)

*

1) IF Bassy3 AND Bassy*11 THEN Tl Machining7* f BracketAssy Assembly*4 = 0

Rh Shaping

MotorAssy Assembly

*

2) IF CoilAssy5 AND CoilAssy*16 THEN Tp Shaping *1 f CoilAssy Assembly9* = 1 Execution Rule: *

Wt Machining

Mainbody Assembly

*

1) IF Mtassy 8 AND Rtassy14 * THEN FassyAssemblyW5 AND FassyAssemblyS 2* AND * FassyAssemblyT11

MagnetAssy Assembly

RotatorAssy Assembly

ArmatureAssyAssembly

CommutatorAssy Assembly

Ct Machining

St Machining

BracketAssy Assembly

FrameAssy Assembly

Ws Machining

Fm Machining

CoilAssy Assembly

Cl Forming

Tp Shaping

Tl Machining

MagnetHousing AssyAssembly

Mt Machining

Bracketa/b Assembly

Ba Machining

Bb Machining

Mh Shaping

Master Operation

Master precedence

Selective Operation

Selective precedence

Figure 9: Identified GPcS for Process Cluster “PC1” in Table 4 8. CONCLUSIONS For realizing mass customization, process platforms have been proposed to configure production processes for new members of product families efficiently and cost-effectively. The concept implication of a process platform is introduced from three aspects with respect to generic variety

representation, generic structures and generic planning. The issues regarding process platform based production configuration are described. For a formal representation of production configuration, a multilevel system of nested colored object-oriented PNs with changeable structures is proposed taking the characteristics of production configuration into account. With the availability of large amount of production process data and advanced data mining tools, a data mining methodology to identify the underlying GPcS of a process platform is put forward. The selection of different λ values produces different GPcSes for the same process family. Therefore, in the possible future work, it raises the evaluation issue of the identified process platforms. Besides the evaluation of identified process platforms, the configured production processes should also be assessed, in that for one given product, more than one processes can be obtained based on a process platform. Furthermore, more research efforts should be put on a mathematical formulation of process platform based production configuration for a more rigorous definition. REFERENCES [1] Pine, B.J., 1993, “Mass Customization: The New Frontier in Business Competition”, Harvard Business School Press, Boston.

11

Copyright © 2005 by ASME

[2] Sanderson, S.W. and Uzumeri, M., 1997, “Managing Product Families”, McGraw-Hill Management & Organization Series, Singapore. [3] Simpson, T.W., 2004, “Product Platform Design and Customization: Status and Promise”, AIEDAM, vol. 18(1), pp. 3-20. [4] Jiao, J. and Tseng, M.M., 2004, “Customizability Analysis in Design for Mass Customization”, Computer-Aided Design, vol. 36(8), pp. 745-757. [5] Wortmann, J.C., Muntslag, D.R., and Timmermans, P.J.M., 1997, Customer-Driven Manufacturing, Chapman and Hall, London. [6] Kolisch, R., 2000, “Integration of Assembly and Fabrication for Make-to-Order Production”, International Journal of Production Economics, vol. 68(3), pp. 287-306. [7] Shierholt, K., 2001, “Process Configuration: Combining the Principles of Product Configuration and Process Planning”, Artificial Intelligence for Engineering Design, Analysis and Manufacturing, vol. 15(5), pp. 411-424. [8] Sawhney, M.S., 1998, “Leveraged High-Variety Strategies: from Portfolio Thinking to Platform Thinking”, Journal of the Academy of Marketing Science, vol. 26(1), pp. 54-61. [9] Martinez, M.T. Favrel, J. and Ghodous, P., 2000, “Product Family Manufacturing Plan Generation and Classification”, Concurrent Engineering: Research and Applications, vol. 8(1), pp. 12-22. [10] Meyer, M. and Lehnerd, A.P., 1997, “The Power of Product Platform – Building Value and Cost Leadership”, Free Press, New York. [11] Jiao, J.X., Zhang, L.F. and Pokharel, S., 2005, “Process Platform Planning for Variety Coordination from Design to Production in Mass Customization Manufacturing”, accepted by IEEE Transaction on Engineering Management. [12] Ho, T.H. and Tang, C.S., 1998, “Product Variety Management: Research Advances”, Kluwer Academic Publishers, Boston/Dordrecht/London. [13] van Veen, E.A., 1992, “Modeling Product Structures by Generic Bills-of-Materials”, Elsevier, New York. [14] Hastings, N.A.J. and Yeh, C.H., 1992, “Bill of Manufacture”, Production and Inventory Management Journal, vol. 33(4), pp. 27-31. [15] Blackburn, J.H., 1985, “Improve MRP and JIT Compatibility by Combining Routings and Bills of Material”, Proceedings of the American Production and Inventory Control Society, pp. 444-447. [16] Jiao, J., Tseng, M.M., Ma, Q. and Zou, Y., 2000, “Generic Bill of Materials and Operations for High-Variety Production Management”, Concurrent Engineering: Research and Application, vol. 8(4), pp. 297-322. [17] Gupta, S. and Krishnan, V., 1998, “Product Family-Based Assembly Sequence Design Methodology”, IIE Transactions, vol. 30(10), pp. 933-945. [18] De Lit, P., Delchambre, A. and Henrioud, J.M., 2003, “An Integrated Approach for Product Family and Assembly

System Design”, IEEE Transactions on Robotics and Automation, vol. 19(2), pp. 324-333. [19] He, D.W. and Kusiak, A., 1997, “Design of Assembly Systems for Modular Products”, IEEE Transactions and Automation, Vol. 13(5), pp. 646-655. [20] Jiao, J.X, Zhang, LF. and Pokharel, S., 2003, “Process Platform Planning for Mass Customization”, 2nd Interdisciplinary World Congress on Mass Customization and Personalization, Munich, Germany. [21] Zhang, L.F., Jiao, J.X. and Pokharel, S., 2004, “InternetEnabled Information Management for Configure-to-Order Product Fulfillment”, Asia Pacific Management Review, vol. 9(5), pp. 851-876. [22] Jiao, J.X., Zhang, L.F. and Prasanna, K., 2005, “Process Variety Modeling for Process Configuration in Mass Customization: An Approach Based on Object-Oriented Petri-Nets with Changeable Structures”, accepted by International Journal of Flexible Manufacturing Systems. [23] Halman, J.I.M., Hofer, A.P. and Vuuren, W.van, 2003, “Platform Driven Development of Product Families: Theory versus Practice”, Journal of Product Innovation Management, vol. 20(2), pp. 149-162. [24] Hegge, H.M.H., 1992, “A Generic Bill-of-Material Processor using Indirect Identification of Products”, Production Planning & Control, vol. 3(3), pp. 336-342. [25] Du, X., Jiao, J. and Tseng, M.M., 2001, “Architecture of Product Family: Fundamentals and Methodology”, Concurrent Engineering: Research and Application, vol. 9(4), pp. 309-325. [26] Mather, H., 1987, “Bills of Materials”, Dow Jones-Irwin, Homewood. [27] Jensen, K., 1992, “Colored Petri Nets: Basic Concepts, Analysis Methods and Practical Use”, vol. 1, SpringerVerlag Berlin Heidelberg , Germany. [28] Wang, L.C., 1996, “Object-Oriented Petri Nets for Modeling and Analysis of Automated Manufacturing Systems”, Computer Integrated Manufacturing Systems, vol. 26(2), pp. 111–125. [29] Jiang, Z., Zuo, M.J., Tu, P.Y. and Fung, R.Y.K., 1999, “Object-Oriented Petri Nets with Changeable Structure (OPNs-CS) for Production System Modeling”, International Journal of Advanced Manufacturing Technology, vol. 15, pp. 445-458. [30] Wang, L.C. and Wu, S.Y., 1998, “Modeling with Colored Timed Object-Oriented Petri Nets for Automated Manufacturing Systems”, Computers and Industrial Engineering, vol. 34(2), pp. 463-480. [31] Romanowski, C.J. and Nagi, R., 2005, On Comparing Bills of Materials: A Similarity/Distance Measure for unordered Trees”, IEEE Transactions on Systems, Man, and Cybernetics, Part A, Forthcoming.

12

Copyright © 2005 by ASME