You are on page 1of 15

Advanced Engineering Informatics xxx (2017) xxxxxx

Contents lists available at ScienceDirect

Advanced Engineering Informatics


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

Knowledge-based design for assembly in agile manufacturing by using


Data Mining methods
R. Kretschmer b, A. Pfouga a, S. Rulhoff a, J. Stjepandic a,
a
PROSTEP AG, Head of Business Unit 3D Product Creation, Dolivostrabe 11, D-64293 Darmstadt, Germany
b
Miele & Cie. KG, Segment Professional Laundry, Director Segment Professional Laundry, Technology Lehrte, Industriestrae 3, 31275 Lehrte, Germany

a r t i c l e i n f o a b s t r a c t

Article history: Decision making in early production planning phases is typically based on a rough estimation due to lack
Received 15 June 2016 of a comprehensive, reliable knowledge base. Virtual planning has been prevailed as a method used to
Received in revised form 16 December 2016 evaluate risks and costs before the concrete realization of production processes. The process of product
Accepted 22 December 2016
assembly, which yields a high share in total production costs, gets its particular importance. This paper
Available online xxxx
introduces a new approach and its initial implementation for knowledge-based design for assembly in
agile manufacturing by using data mining (DM) methods in the field of series production with high vari-
Keywords:
ance. The approach adopts the usage of bulk data with old, successful designs in order to extrapolate its
Design for assembly
Agile manufacturing
scope for assembly processes. Especially linked product and process data allow the innovative usage of
Digital factory DM methods in order to facilitate the front loading in the product development. The concept presents
Assembly an affordable assistance potential for development of new products variants along the product emer-
Process planning gence process (PEP). With this approach an early cost estimation of assembly processes in series produc-
Data mining tion can be conducted using advanced DM methods as shown in an industrial use case. Furthermore,
design and planning processes can be supported effectively.
2016 Elsevier Ltd. All rights reserved.

1. Introduction feasibility of series production must be assured with vague infor-


mation on product and given general conditions, e.g. shift model
Today globally operating companies face additional challenges [6]. This is a great challenge especially to planning cost-intensive
due to turbulent fluctuations in market demand, increasing vari- assembly of a product [41,53].
ability of products, shortened product lifecycles and corresponding Front loading has been seen as an appropriate means to tackle
complexity of processes [1]. It results in higher flexibility expec- such challenges [43] and must be supported by adequate pro-
tance in production system and economically reasonable dispatch cesses, methods and tools. Subsequently, production planning
of new products in an existing production line. The pivotal link should start as early as possible in phase with product develop-
between design and manufacturing lies in process planning. Pro- ment. The research and development project Prospective Deter-
cess planning deals with the selection of necessary manufacturing mination of Assembly Work Content in Digital Manufacturing
processes and determination of their sequences to transform a (Pro Mondi) was initiated to link the early product design with
designers creation (namely the shaped part) into a physical com- the early production planning using methods of data modeling
ponent economically and competitively. In the modern product and data mining (DM) to generate information with focus on the
emergence process (PEP), production planning gains in importance product assembly planning for new products in early production
and has to be executed as parallel as possible to the product devel- planning phases. Aim of this project is to achieve accurate estima-
opment according to concurrent engineering principles [10]. In this tion of expected assembly work and resulting costs in an early
early product creation phase, a first step for planning processes is a stage of the product development. This should be realized in sync
cost calculation for the industrialization of the product in existing with design through provision of additional support with assembly
production lines regarding basic conditions [5]. The cost-effective knowledge for an underlying design. The approach, thus, contains
reuse of existing planning data in order to extrapolate assembly
Corresponding author. processes. Especially linked product and process data allow the
E-mail addresses: ralf.kretschmer@miele.de (R. Kretschmer), Alain.Pfouga@ innovative usage of Data Mining methods. As proof of concept this
PROSTEP.com (A. Pfouga), Stefan.rulhoff@prostep.com (S. Rulhoff), josip.stjepandic@
prostep.com (J. Stjepandic).

http://dx.doi.org/10.1016/j.aei.2016.12.006
1474-0346/ 2016 Elsevier Ltd. All rights reserved.

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
2 R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx

approach has been evaluated with different manufacturing often summarized by the concept Design for Manufacturing and
companies. Assembly (DFMA) [7]. The normal result of DFMA, as an integral
part of the design process, is simpler and more reliable products
that are less expensive to manufacture and assemble [37]. How-
2. Related work ever, products designed in this way tend to have a smaller number
of complex components, making maintenance and upgrading diffi-
2.1. Design for X cult and expensive. The emphasis on reducing manufacturing costs
has, therefore, been at the detriment of in-service costs. This may
The literature offers besides Design for X also the term not be a particular problem for mass-produced (typically minimal
Design to X, both of which are often grouped under the concept maintenance, low priced, short life span) products such as the
Design for X, but have different meanings. Many definitions majority of domestic appliances [22]. Meeting manufacture crite-
highlight the fact that there are different objectives that come into ria, not only anticipated the manufacturing engineering auto-
consideration in this work as well. The terms are interpreted body design activities reducing the time to market, but reduces
according to the definition of Vielhaber [62]: investments and structural cost of the plant are the findings of
the industrial study [36] [39]. These strategies have generated
1. According to Chen [14], Design for X describes the reactive, resources with the mechanical presses selling and create a new
methodological assistance in development of a product to a platform with fewer operations, on average, comparing similar
property determined by a set of features which is necessary, parts with previous designs [57]. Other approaches focus on the
to make a follow-up process in the product lifecycle successful complexity of design rules, especially in semantic assembly deci-
(e.g., adaptability to manufacture). sion [15]. A major challenge of the complexity is a computational
2. Design to X describes the proactive, methodical assistance in issue associated with a very sophisticated and time-consuming
development of a product for a target property (e.g., cost- task with respect to semantic reasoning for ontology-based pro-
effective products). duct design. By using disparate attributes algorithm, computer-
3. X-Oriented Design refers to the collection of all Design for X aided systems become more easily able to understand and to dis-
and Design to X criteria. cern joining types.
4. The terms features and characteristics are understood An excerpt of typical DFMA guidelines is provided in Fig. 1 [9].
according to the view of the Characteristics-Properties Model- An updated version comprises the broad, long-term experience
ing (CPM)/Property-Driven Development (PDP) approaches in DfMA by a set of generic guidelines [6].
by Weber [66] in the definitions.
2.2. Assembly planning
The main objectives (whether Design for X or Design to X) are
often repeatedly considered during the development process and Assembly is the installation of geometrically defined objects by
therefore decisions have to be brought about at a time. For exam- joining, handling, calibration and control operations. The task of
ple, the criteria are taken into account in early stages of product assembly planning refers to the creation of necessary precondition
development, but may need to be focused again later during the for economic installation procedure. This includes identification of
elaboration of the product shape and have to be adjusted [50]. needs, targeted and efficient use of staff as well as resources.
Therefore, it can be stated in principle that the application of meth- Assembly is defined as joining or assembling manufactured items
ods and tools of Design for X is a way to consider the numerous into a functional unit. Within the installation planning, a distinc-
restrictions in the development process [65]. Both analysis and tion is made between operations and detailed planning. Function
synthesis steps must be included. Weber explains the poor integra- of operations planning is to develop an assembly system and to
tion of Design for X-methods and tools in the design methodology create a rough schedule [41,53].
with the required characteristic by Design for X (e.g., material The detailed planning is divided in concretization of the assem-
strength or assemblability), which cannot be easily described by bly system and preparation of a detailed schedule. Table 1 shows
terms such as functional structure or solution principle [67]. functions and planning phases of assembly planning as well as
According to Bossmann [9], DfX methods can be subdivided in its required input information and results to be achieved [9,11].
five main groups (user, environment, functional reliability, costs, As product geometrical shape is main, but not only criterion for
and production). The latter is composed of the following sub- assembly process, many relevant product properties are defined
groups: Machining, Reshaping, Primary forming, Component man- in CAD systems [3,8].
ufacturing, and Assembly. In classic Assembly Planning the assembly sequence has to be
In the ideal case all guidelines are considered in designing a determined by a product analysis. Thereby, geometric relation-
new product, but that case is improbable. Therefore, the design ships between product structure components are examined in par-
department has to be supported by other departments which pro- ticular. For this purpose the product is dismantled into separate
vide their expertise in order to meet the Design for X parts. In some cases, the assembly sequence can be sufficiently
requirements. characterized by the joining sequence of the components [51].
There are several design for or design to criteria, which are For the determination of the assembly sequence Lotter and
summarized with Design for X. Design for Disassembly and Wiendahl evince that 40% of the time is to be invested in order
Design for Recycling, for example, are covered by the same group, to determine the order of the components in the assembly while
which is headlined with Design for Environment. Boothroyd and performing the planning activities [41]. In recent years, some
Alting concentrated on design for assembly [7]. In their concept expenditure shift could be registered due to an increased use of
the reduction of the number of components is particularly computers. The largest effort is still located in determining the
addressed to reduce assembly costs. A consistent implementation assembly sequence, which is why this aspect focused in research
in Swedish companies led to convincing results [61]. In 50% of activities [18]. There are numerous approaches which can be
these companies a reduction of development time and develop- based, for example, on the disassembly sequence to determine an
ment costs by 33% was established. The assembly costs could be appropriate assembly sequence [25,38,47].
reduced by up to 85% with the DFA approach by Boothryd and Alt- One result of the assembly planning is the assembly graph,
ing [61]. Design for assembly and Design for Manufacturing are which represents the possible assembly sequences in a network

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx 3

Fig. 1. DFMA guidelines [9].

Table 1
Functions and planning phases of the assembly planning.

Task Input Information Output


Assembly system development
Operations planning  Work system design  Assembly task  Assembly structure
 Developing system structures  Production program  Rough layout
 Capacity planning  Square measures  Capacity requirements
 Concatenation principle
 Structuring product  Parts list  Precedence graph
 Rough flow planning  Drawing  Rough flow structure
Assembly system refinement
Detailed planning  Principle solution planning  Parts list  Detailed layout
 Workstations planning  Drawing  Workstations
 Concentration resource planning  Concentration resources
 Determining assembly content  Assembly structure  Assembly plan
 Creating assembly plan and assembly documents  Rough layout  Work instructions
 Spreadsheet

structure [19,20]. This network can be seen as a tree structure in predict assembly times from detailed assembly models, low fide-
which individual parts are forming leaves, inner branches are seen lity part models are used in a series of predictive performance
as sub-assemblies, which in turn are taken up by complex ramifi- experiments. Results reveal that this tool can predict the assembly
cations and, finally, end up in the root, the base element [28,29]. time of a product to within 40% of the target as built time using a
For this purpose feature models are used by the Feature Oriented high fidelity neural network and a low fidelity CAD model. The tool
Domain Analysis (FODA) approach to represent the assembly itself is based on structural complexity, representing the assembly graph
and its variance [54]. Similarly, a new approach for mining large as complexity vector of 29 metrics. A neural network is then used
quantities of machining features, Computer-Aided Design (CAD) to build a relationship between the complexity vector (input) and
models and manufacturing data is based on DM and uses the assembly time (output).
learning-logic classification techniques for mining 3D CAD data. Various computer-aided planning approaches are compared
The proposed approach was evaluated in CAD model analysis, with an overweight of knowledge-based and STEP-based
specifically in classification tasks. The experimental results proved approaches in recent years [71]. Advanced planning approaches
that the method is effective in terms of classification accuracy and in assembly planning can be grouped in a number of categories,
can be used as an efficient data mining tool for CAD model analysis i.e. feature-based technologies, knowledge-based systems, artifi-
and classification [2]. cial neural networks, genetic algorithms, fuzzy set theory and
Low-fidelity CAD models can be used to predict assembly times, fuzzy logic, Petri nets, agent-based technology, Internet-based
thereby supporting earlier inclusion of design for assembly meth- technology, STEP-compliant CAPP and other emerging technolo-
ods in the design process [45]. Expanding on previous work to gies [69].

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
4 R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx

A paper gives a framework and organization of the conceptual demonstrate that the proposed model explains 58% of the variance
assembly design system with help of knowledge-based engineer- [32] what indicates disparate usability of this approach. Three fac-
ing. It thereby takes assembly sequence, joint configuration, and tors in the task-oriented dimension, output quality, response time,
tolerance allocation in the auto-body assembly design process and format, provide valuable points of reference for the designers
planning into integrated consideration [40]. Heuristic and empiri- and developers of DM tools and relevant software [46]. Typical
cal knowledge play active roles in the generation of assembly context of the commercial DM procedure is shown in Fig. 2.
sequence, dimension chain, configuration of joint types, allocation Phase 1 (Business Objectives) comprises the definition of a DM
of tolerance limits and qualitative simulation which is knowledge project, including its initial concept, motivation, business objec-
or rule reasoning. It is provided with various rule, criterion and tives, viability, estimated costs, and expected benefit (returns).
principle. Knowledge-based vehicle assembly design system can Phase 2 (Data Preparation) deals with possible sources of data
improve assembly concept quality by means of qualitative simula- and information, different ways of conceptualization, defines the
tion to fulfill the concurrent integration between conceptual required data quality, considers the topics of variable selection
design and detail design stages. Further example of a military sys- and factor derivation, sampling and partitioning methods, which
tem is presented to demonstrate the feasibility of a three-stage is often done when the volume of data is too great to process.
integrated approach with heuristic working rules [12,14]. Phase 3 (Data Analysis) provides a selection of the most common
Similarly, a planning system may include an assemble advisory types of data analysis for data mining. It includes the clustering
module, a knowledge-based module and a user interface to enables in combination with visualization techniques, transactional analy-
design engineers to find and pick-up the best and fastest assembly sis and time series analysis. Phase 4 (Modeling) begins with the
method [60]. definition of a data model, followed by discussion of concepts such
as supervised and unsupervised learning, cross-validation. It con-
2.3. Data mining siders various techniques for modeling data, from AI (artificial
intelligence) approaches, such as neural networks and rule induc-
Concepts for data mining are projected to be useful to provide tion, to statistical techniques, such as regression. Finally, in Phase 5
historical insight and best-practices for an evolutionary assortment (Deployment), the results of data mining can be fed into the
of successful design solutions in the context of this paper. This decision-making and operative processes of the business in differ-
chapter is devoted to an introduction to this particular domain. ent ways (query/reporting, executive information systems and
Data mining is a process of discovering valuable information expert systems).
from observational data sets, which is an interdisciplinary field A further article presents a data-miningaided optimal design
bringing together techniques from databases, machine learning, method that is able to find a competitive design solution with a rel-
optimization theory, statistics, pattern recognition, and visualiza- atively low computational cost [33]. The method consists of four
tion. Introduction of data mining concepts, implementation proce- components: (1) a uniform-coverage selection method, that
dures and application cases in engineering is described by [70,59]. chooses design representatives from among a large number of
The overall goal of the data mining process is to extract informa- original design alternatives for a nonrectangular design space; (2)
tion from a data set and transform it into an understandable struc- feature functions, of which evaluation is computationally econom-
ture for further use. One of the greatest expected benefits of DM ical as the surrogate for the design objective function; (3) a cluster-
methods is the ability to link seemingly disparate disciplines, for ing method, that generates a design library based on the evaluation
the purpose of developing and testing hypotheses that cannot be of feature functions instead of an objective function; and (4) a clas-
approached within a single knowledge domain. Methods are sification method to create the design selection rules, eventually
reviewed by which analysts can navigate through different data leading us to a competitive design.
resources (e.g. historical data) to create new, merged data sets New processes suitable to assemble the given new product shall
[17]. Significant factors are efficient knowledge utilization and be designed based on this existing, historical data (product linked
knowledge exchange on an interdisciplinary level. An empirical to corresponding process). Successful designs from the past are
study involving 206 Data Mining tools users was conducted to expected to be a good envelope for the solution space of the future
evaluate the model using structural equation modeling. Results designs [27]. Automatic analysis with a specific DM method shall

Fig. 2. Typical context of data mining procedure [46].

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx 5

Fig. 3. Design optimization with additional time data.

Fig. 4. Emergence of product assembly information PAI.

be used to create a first draft of the assembly process and estimate engineer can be supported at the selection of appropriate joining
the expected costs. Following production planning processes can elements. With this approach, an assembly knowledge based sup-
be supported by automatic proposals of adequate assembly pro- port of the designer in series production can be achieved using
cesses, which then can be customized [52]. Moreover, the design innovative DM methods.

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
6 R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx

2.4. Agile manufacturing concept focuses on concurrent product design and production
assembly planning [42]. Product designer and production planer
Agile manufacturing has been defined with respect to the agile are thereby required to be more efficiently assisted by various
enterprise, products, workforce, capabilities and the environment applications to improve design or planning process through infor-
that gives impetus to the development of agile paradigm [72]. It mation gathered by data mining [55]. As the assembly process con-
assimilates the full range of flexible production technologies, along sumes a considerable share of production costs, its optimization
with the lessons learned from total quality management, just-in- seems to be a promising strategy.
time production and lean production. Agility is the ability of a In this context, the following requirements for research ques-
business to grow in a competitive market of continuous and unan- tions arise:
ticipated change, to respond quickly to rapidly changing markets
driven by customer-based valuing of products and services. In 1. The assembly planning should start as soon as possible (ideally,
other words agile organizations are able to compete on the basis front-loaded within product design).
of time-compression. Finally, agile manufacturing is a system that 2. To secure re-use of parts and processes, in particular of joining
shifts quickly (speed and responsiveness) among product models elements, planning should be based on past empirical values,
or between product lines (flexibility), ideally in real-time response the product assembly information (PAI).
to customer demand (customer needs and wants). 3. Users should be supported by an intelligent assistant which
either pulls the empirical PAI by data mining in the historical
data or adds new PAI based e.g. on recent recommendations.
3. Background and requirements
Because no commercial application is known supporting afore-
Worldwide available, Miele is the global premium brand of mentioned requirements, demonstrators need to be developed,
domestic appliances and commercial machines in the field of laun- which in first step, focuses on the selection and design of appropri-
dry care, dishwashing and disinfection [44]. A continuous stream ate assembly connections.
of innovations is the foundation of Mieles business success since
1899. In terms of quality, Miele appliances are considerably better 3.1. Preparation and requirements
than those of the competition; otherwise they would not have
been able to compete successfully on such a fiercely competitive Fig. 3 illustrates the underlying approach to realize design opti-
market. To facilitate agile manufacturing, Miele is looking for mization enriched with fundamental data required from Manufac-
new approaches and methods for product and production plan- turing. It describes the overall product emergence process, which
ning. One of the main areas of interest is the assembly process combines process relevant attributes from design (bottom left)
which is cost-intensive and, therefore, has high potential for and process patterns from manufacturing (bottom right). The inte-
improvement by prospective planning. Consequently, fast search grated product model is thereby enriched to an assembly-oriented
in the existing databases promises good assistance for the product product model, having a good conjunction between these phases.
designers. The main input for this research was the request for reli- With this in place, a prospective estimation of the assembly work
able information about the already successfully used connections and costs in series production can be achieved. Thus, design task
which shall be extracted from historical data. are effectively aligned with time requirements linked to assembly
In order to address the challenges of data mining, the integra- planning process.
tion of various planning tasks within the PEP, new concepts are The emergence of Product Assembly Information is shown in
necessary. Though, as a part of integrated product and process Fig. 4 and runs through the following steps:
development there are different definitions for various phases Enrich CAD data with assembling information: Derived from
and aspects of planning activities along the PEP [68]. Regardless previously similar designed products, assembly information such
of the specific definition of these phases and aspects; and based as time data about the actual design situation can be identified
on analysis, it is certain that great amount of their containing infor- and provided in order to support the designer. This additional
mation and knowledge are either utilized insufficiently and inef- information can be used to enrich the CAD data to support the cur-
fectively or remain unused [26]. In this regard, the presented rent design and be updated in later assembly planning processes.

Fig. 5. Workflow of the design optimization with additional time data.

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx 7

Suggesting assembly connections: An assisting option for the to the conventional systems for design and stacking product parts
designer is a suggestion list of similar previously designed assem- and assemblies, systems for process planning and time measure-
bly connection variations. These lists give a quick overview of pos- ment were also taken into account. They sustain a comprehensive
sible and already implemented connection types in the assembly. portfolio of information and therefore can be used to distinguish
Assembly process estimation: The focus is on the creation of different product parts and assemblies. The results of this analysis
an assembly process for a new product. Based on existing product are capsulated as an object oriented data model, further described
and process data compilation of a first approximated assembly in the data model section.
process for a new product could be developed. From this, the pro- The necessary enrichment of product and process data on the
duction planner can specify further details and thus determine a fly (without holding the workflow) for the presented concept
first estimation regarding assembly time. Based on the assembly requires additional efforts in design. This additional expenditure
time and associated calculation scheme, the planner can perform also relates to assembly connections and includes the acquisition
the first cost estimation in a very early planning phase. of new information form the designers know how. The designer
The information in production planning and design processes usually defines assembly connections either implicitly through
including historical insights mutually enrich each other. Addition- formed locked joints by the shaping of the parts or explicitly by
ally, intelligent interconnecting information from both areas cre- connecting elements such as in screwed fasteners.
ates added value. The newly obtained information supports the
workflow throughout the PEP. Therefore, as part of this concept
some requirements need to be met. Thus, the pre-conditions 3.3. Data collection and availability
attached to both systems as well as their respective processes have
to be fulfilled [56]. The designer of the assembly connection considers all these
information in the design but cannot store them in the CAD model
3.2. Attributes and data sources because the CAD tools for the most part are not able to define the
necessary attributes.
Data Mining methods can be used for data clustering and clas- To overcome this problem as part of the concept presented in
sification, however criteria for comparison of data sets have to be this paper, the designer will be provided with an additional tool
identified [30]. To determine the criteria for comparison, within in the CAD system. It can be used to create assembly connections
the scope of this project, a survey of users as well as an analysis and gives additional information (PAI) and explicit design possibil-
of various tools of the DM was performed. The objective of this ities. Thus, data will be collected in the source system, the CAD sys-
analysis was to identify attributes relevant for assembly processes tem in particular. Since the defined objects are not part of PDM
that could be assigned to products and parts in CAD [31], PDM and systems, an extension is necessary in order to implement connec-
production planning systems. In CAD systems, attributes assigned tions as objects and to store them after the transfer in the PDM sys-
to parts contain mainly geometric information including meta tem persistently. In the further processing, the product
information such as volume and weight. The PDM systems contain information will be linked to the planning processes. Unless the
organizational information, such as creator, version and revision as storage of product data are in the same system as for production
well as the mentioned parts information form CAD [24]. In addition planning, the information flow from the PDM system to the plan-

Fig. 6. Assembly time estimation using PAI.

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
8 R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx

Fig. 7. Data model overview.

ning system as well as the DM tool, for further analysis, has to be the PDM database to determine the respective associated and
ensured. related sub processes (Fig. 6) where eBOM and mBOM describe
In current planning systems it is often possible to directly link the engineering and manufacturing BOM respectively. Each pro-
processes to products [49]. Thus, an allocation of to be assembled duct assembly information record represents an assembly connec-
product and the associated assembly processes can be realized. tion. By multiple connections within the assemblies, multiple sub-
In the assembly, however, parts are joined with other parts or processes for the assembly can be determined. These processes
products. These assembly connections with their additional infor- contain the time data relevant for the new product design. Corre-
mation have no digital equivalent object yet. However by means sponding time information and, if requested, along with an alter-
of an object such as the product assembly information, it is pos- native proposal list is transferred and displayed in the CAD
sible to store useful additional connection information, which System. This assembly time information of the existing product
relates directly to the respective assembly connection. As part of represents a first approximated assembly time for the new prod-
this concept, the combination of the products and processes does uct. So the designer is provided with this additional information
not take place directly but through this object product assembly regarding the assembly time and with an enterprise specific factor
information. The linking of product and process does not neces- the corresponding cost of the current design solution. In the final
sarily need to occur at the part level. step, the designer is able to optimize the product iteratively on
the basis of anticipated assembly time and costs for each design.

4. Proposed concept
4.1. Data model
The concept presented in this paper describes an assisting
workflow to support the designer (Fig. 5). As part of a new or mod- Based on determined assembly characteristics, a range of attri-
ified design, the designer creates new product data. A software butes is derived to classify the assembly of the parts. Fig. 7 shows
assistant supports the designer in creating assembly connections an excerpt of the generated data model for the data mining
and enriches the CAD model with product assembly information analysis.
for each connection. This product assembly information includes The ProductAssemblyInformation (PAI) is the central element in
additional connection information including e.g. the torque this data scheme and represents the assembly of the product parts.
screwed fasteners or the type and the form of a welded joint and References for time analysis, assembly requirements, designed
information about other connection types. In the ongoing design parts or products as well as a wide range of meta data including
process, the designer can trigger an evaluation regarding the the assembly department and other information are lodged. This
assembly connections in the model. element is supplemented with attributes of different connection
For this purpose, the characteristics of the CAD model are first types (Fig. 8).
prepared and analyzed with Data Mining. The analysis focuses on The second fundamental object in the data model is the Item. It
the product assembly information and their properties. The parts contains references to existing sub-assembly units, geometrical
associated with the product assembly information and their geo- characteristics as well as ProductAssemblyInformation. Each Item
metric properties are also included in the analysis as additional refers to the ProductAssemblyInformation, which also refers to fur-
information set. Furthermore, an extended database is also pro- ther used Items. This construct is chosen to enable Data Mining
vided and consists of product and process data of existing prod- methods to determine exact similarities between new parts and/
ucts, which are linked via the product assembly information. The or products and other existing parts. Furthermore, it makes com-
characteristics of the product assembly information of the new parison parts and products the new and existing ones, in any order
product are compared with the properties of the product assembly and combination interchangeably possible.
information of the existing product in the extended database. Then, In the first approach the attributes for screw connections are
the most similar product assembly information is determined from clustered and evaluated regarding the relevance for assembly oper-
the existing products. This analysis can be restricted by a class of ation. The identified attributes are classified in the categories fas-
the connection types (screwed, weld, rivet) or deliberately left teners, installation/assembly situation, tools/equipment,
open to widen the solution space and to provide the designer with installation regulations and additional assembly elements. These
information about other assembly connections. attributes are represented in the data model in different object cat-
A limitation on the particular type of connection yields as a egories (here illustrated by color1, c.f. Fig. 8). The evaluation regard-
result of the closest realized assembly connection of the same kind. ing the influence on the assembly time provides a first indication for
Depending on the properties of the parts, other mounting connec- the relevance in the data mining analysis. Production planner
tions can also be found and proposed to the designer in a proposal weighted the attributes with low medium and high in order to deter-
list.
The presented application for the support of the design process 1
For interpretation of color in Fig. 8, the reader is referred to the web version of
uses the product assembly information identified in the analysis of this article.

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx 9

Fig. 8. Structure of Product Assembly Informations (PAI) for weighting.

mine a first selection of attributes relevant for the data mining. time type and unit. The application model supports the distinction
Which attributes really are significant for the similarity of assembly of simple time values or time ranges.
connection have to be determined in a data mining analysis with a The time module occurrence represents the specific incident of
large quantity of product data. The columns design and work plan- an operation. For example if in an assembly operation several
ning shall indicate in which department the attributes can be col- screws have to be tightened, in the time module definition there
lected either by IT system (column System) or experienced are all time types with value and units for the single operation of
users (column Know-How). one screw and in the time module occurrence is the factor for
Not all of these attributes can be identified in the mechanical the number of screws used in this specific process.
design in the CAD system. Some can be determined in production Additionally the object operation definition includes a reference
planning workshops in order to optimize the current design. Expe- container. This object connects the operation to any resources and
rienced designers and production planner can pre-allocate some parts used in the operation. With this structure the application
parameters with estimated values, which can be reviewed later. model represents the important triangle of product, process
Other parameter and the corresponding value data can be and resource. This application model is then translated in the final
extracted out of other systems e.g. the attributes of standard parts. data exchange format in XML. For this mapping the static and
Supplementary connection types can be added to the data dynamic data exchange has to be considered.
model. To provide required information for time analysis, a stan- The implementation concept of the ADiFa protocol provides
dardized data model is applied. In this regard, ADiFa projects capabilities for static and dynamic data exchange (see Fig. 10).
Application-specific data models, so called ADiFa Application The static data exchange, which is in fact file-based, occurs in form
Protocols, were used. It offer the integration of processes and data of XML files, which are compiled and read from the interface pro-
for different DM systems [49]. cessors. The dynamic data exchange transmits the same data
The application data model of the ADiFa Application Protocol is though an SOA (Service-oriented Architecture) concept by using
specified in an UML class diagram illustrated roughly in Fig. 9. It web service technologies. In those web services the requests and
reflects additional requirements of the user, needed to be defined the corresponding data from the application interpreted model
in the application process model. are defined in order to be exchanged between the programs in
The core objects in the application data model are the operation engineering and manufacturing domains. Applications involved
definition, the time module definition and the time module occur- thereby will act both as a server to provide data to another
rence. In Fig. 9 they are highlighted in blue. The operation object requesting application as well as a client to retrieve data from
includes hierarchical structure of processes and activities with other systems.
the definition of successor and child operations. The object time
module definition is designed for the encapsulation of all time 4.2. Data mapping and data mining
management data. It is a general definition of every describing
attribute and time value of an operation. The time module defini- After aggregating and appending data subsets from different
tion can contain numerous time values, each with the classification sources and systems, it is necessary to remove redundant ones

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
10 R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx

Fig. 9. Key objects of the ADiFa Application protocol.

[48] for the DM process. The next step is converting and porting The implementation of the presented approach is challenging
data in the presented data model. Depending on data source the due to high requirement for interconnection and overall quality
conversion is either fully automated or partially automated with of the existing data in different source system. In particular the
further manual adjustment. Often, value and scale of different pure number of realized and existing assembly connections and,
attributes are heterogeneous. In these cases, a normalization of rat- thus, necessary instances of a product assembly information as
ings prevents the undesired high or low impact of certain attri- well as the quality of the data regarding their attributes are
butes on the results and evaluation process. In this regard, a [0, important.
1] linear normalization has been used. Additionally, a further attri- At this point of time, the fulfillment of these high requirements
bute prioritizing via weighting can be necessary to define the has to be verified. In particular, the quality of the linkage of pro-
importance of each attribute for the evaluation [58]. An automated duct data with the corresponding assembly processes poses a chal-
learning of the weights via machine learning methods depends on lenge. The selection of the properties and attributes for the DM
the existing data sets and their quality. Otherwise, they are deter- analysis also has to be determined based on production data to
mined based on expert knowledge or a combination of both meth- ensure the fidelity of generated results (Fig. 11). In this scope a spe-
ods. To prevent further expansion of scope and the complexity of cial focus is on characteristics of parts and of the connection itself.
existing problem, expert knowledge was applied to determine In conformity with the presented objectives and concept, an appli-
the attribute weights. This support was provided by an expert from cation of the methodology is described as follows.
the vendor of the DM tool. Furthermore, it is possible to have more Suggesting assembly connections and enrichment of CAD data:
than a single weight vector. This approach is useful, if there are The designer with expertise in assembly process creates a new
various object types or parts, which have different prioritization module with already known and new assembly connections in
for their attributes [73]. To identify the objects with most similar the CAD system. He designs assembly connections and comple-
product assembly information for a new object, the classification ments these with properties in context of a new module. Via an
algorithm k-nearest neighbour (kNN) [21] with Euclidean distance automated DM process, he is provided with various data about
as evaluation function is used. From the identified objects, a list is assembly connections. Moreover, for each assembly connection a
generated and the most related one can be manually chosen, which list of alternative or ever realized connections can be created.
passes its assembly process data to new object. To assure the reli- Depending on the product properties, the five most similar product
ability of the presented method and prevent over fitting, a cross assembly configurations are made available to the designer as a
validation [34] is used. prepared proposal list, which is generated through cluster analysis

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx 11

Fig. 10. Process for implementation of the ADiFa Application protocol.

Fig. 11. Formation of product clusters and process agglomerations.

of existing product data. These configurations can be used directly, the right product assembly information and, thus, the assembly
later or in an extended context of the product thus enhancing the processes are found. For new unknown connections, the most sim-
CAD model. If the analysis is dispensed with filtering of associated ilar product assembly information and related assembly processes
connections of the product assembly information, the designer can from the database are determined and duplicated. Each of the
also be provided with other not associated connections as founded product assembly information represents a single connec-
alternatives. tion and the linked process represents precisely the assembly work
Estimation of assembly process and information: The produc- instructions for this connection. The sum of the individual connec-
tion planner drafts an initial assembly process for a new assembly tions for the new product is its first assembly process. Thereby, an
at an early stage of product development. Similar to the use case of initial draft of an assembly process of the new module can be gen-
the designer, for known assembly connections that are imple- erated. Founded individual connections, individual process, as well
mented in the new product as well as in the old product data, as the overall process can be used in different ways to assist

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
12 R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx

Fig. 12. Assessment schema for validation.

designer and production planner. Planner and designer addition- considered. To estimate a suitable number for the product cluster
ally get a first estimation for expected assembly time and costs structural data from part lists of assemblies were used. Then, the
in the automated process. In addition, the production planner attributes such as size, material and weight which were shown
can increase the quality of the process by manual intervention. previously were used in the first approach. In this case, the calcu-
On the one hand, he adapts the product assembly information cre- lation of the component dimensions of the bounding box proved to
ated by the designer before the DM analysis. On the other hand, he be sufficiently accurate. This way, 5 product clusters with respec-
can complete the product assembly information with practical tive reference to the component category could be formed. With
knowledge. At this, he has an impact on the input of the DM anal- regard to the associated assembly processes, a first result was that
ysis and increases the quality of the result thereby. Furthermore, components with similar design characteristics also need similar
the designer has a first draft for the assembly process at ones dis- assembly processes. This seemingly trivial statement however val-
posal and a first estimated assembly time in the current CAD sys- idates an important requirement for the applicability of existing
tem. By a company-specific factor, the designer receives also assembly processes for current products to future assembly pro-
information about the cost of the connection in the assembly. By cesses for new-developed products.
verifying this information, the designer can evaluate and compare
the alternatives for different connections. 5.2. Building of process clusters

5. Use case evaluation The clustering for the processes is based on time analysis in
Methods Time Measurement (MTM). Basic principle is the determi-
The general assessment schema is based on ISO/IEC 9126 for nation of target times by combining the time measurement units.
software quality as shown in Fig. 12. The validation criteria are Depending on the containing time blocks, 7-process cluster with
subdivided in 4 weighted categories (methodological, organiza- explicit reference to the component category were formed. Addi-
tional, technical and data validation), which comprise a plethora tional accuracy of clustering can be achieved with different simi-
of singular criteria (e.g. precision). Each evaluation assigns one of larity searches within the process data. The different similarity
3 performance levels in 2 stages. The total sum yields the overall searches are used:
evaluation score for dedicated use case.
1. Similarity search via process parameter: Single time blocks such
5.1. Building of product clusters as pick and place are composed of individual movements.
Each single time block has attributes like distance to pick.
Currently, several individual Use Cases are considered for vali- These attributes are used in the Data Mining method K-Means
dation purposes. First, the formation of the product cluster was to divide similar assembly processes into clusters.

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx 13

Fig. 13. Similarity search for process data.

2. Similarity search via description text: The similarity of descrip- an assessment. However, to produce reliable outcomes, the pro-
tion texted is evaluated based on the text mining. In particular, duct data have to fulfil high requirements in regard to connection
key words as switch panel base or steering obtain a high elements. Concurrently, the necessary data model and some tool
weighting. sets are provided to make the data integration easier. In the future,
3. Similarity search via sequences: Structural characteristics of further development of tool sets and methods could help to reduce
individual assembly processes are considered. In addition, the the high initial effort for adjustment of the data even more. Besides
question of how many same sequences of individual time the evaluation of the results based on product data, it is important
blocks are used in a parent sequence is analyzed. The more to investigate the behavior and results of the methodology for new
identical or similar sequences of time blocks, the more similar and innovative assembly technologies. Furthermore, for analyzing
the considered assembly processes. more complex data sets as well as obtaining better results, it is
important to develop and refine the concept and to apply further
Fig. 13 presents the automated similarity search with the tool Data Mining methods.
Data Miner. In different process blocks, data are read, processed
and analyzed with different similarity searches.
Acknowledgement

6. Conclusion and outlook The research project Prospective Determination of Assembly


Work Content in Digital Manufacturing (ProMondi) is supported
Through utilization of Data Mining tools, the efficient design of by the German Federal Ministry of Education and Research (BMBF)
assembly connection, the quality of planning results and planning within the Framework Concept Research for Tomorrows Produc-
processes can be increased, while time and cost reduction can tion (funding number 02PJ1110) and managed by the Project
simultaneously be achieved. With this approach working sched- Management Agency Karlsruhe (PTKA). The authors are responsi-
ules as planning results are based on field-tested assembly pro- ble for the contents of this publication.
cesses and contain the implicit knowledge used in similar
assembly planning processes. The automatic generation of an
References
adapted assembly process enables fast customization to a concrete
setting at the shop floor. In this regard, the presented approach [1] S. Alguezaui, R. Filieri, A knowledge-based view of the extending enterprise for
contributes an important added value to production design and enhancing a collaborative innovation advantage, Int. J. Agile Syst. Manage. 7 (2)
(2014) 116131.
planning through usage of knowledge in existing systems. Thus,
[2] H. Al-Mubaid, E.S. Abouel Nasr, A.K. Kamrani, Using data mining in the
reduction of planning time, increasing availability of information manufacturing systems for CAD model analysis and classification, Int. J. Agile
in product design as well as making the cooperation between the Syst. Manage. 3 (1/2) (2008) 147162.
designer and product planning teams easier are the consequences. [3] A. Armillotta, G. Moroni, M. Rasella, Computer-aided assembly planning for the
diemaking industry, Robot. Comput.-Integr. Manuf. 22 (2006) 409419.
The feasibility of this concept with productive data is in evaluation. [5] H. Bley, C. Franke, Integration of product design and assembly planning in the
New approaches with clustered data to improve data quality are in digital factory, Ann. CIRP 53 (1) (2004) 2530.

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
14 R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx

[6] G. Boothroyd, Assembly Automation and Product Design, third ed., Taylor & [42] E.B. Magrab, S.K. Gupta, F.P. McCluskey, P.A. Sandborn, Integrated Product and
Francis Group, Boca Raton, 2011. Process Design and Development: The Product Realization Process, second ed.,
[7] G. Boothroyd, W. Knight, P. Dewhurst, Product Design for Manufacture and Taylor & Francis Group, Boca Raton, 2010.
Assembly, second ed., Marcel Dekker, New York, 1994. [43] A. McLay, Re-reengineering the dream: agility as competitive adaptability, Int.
[8] G. Bornet dit Vorgeat, P. Pu, R. Clavel, A. Csabai, F. Sprumont, P. Xiroochakis, M.- J. Agile Syst. Manage. 7 (2) (2014) 101115.
T. Ivorra, MicroCE: Computer-Aided Support for DFMA Conceptual Design [44] Miele, <https://www.miele.com/de/com/philosophie-2095.htm>, Accessed
Phase, CiteseerX: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1. May 16th, 2016.
16.1358, Accessed: Sep 15, 2015. [45] E.Z. Namouz, J. Summers, Complexity connectivity metrics predicting
[9] M. Bossmann, Feature-basierte Produkt- und Prozessmodelle in der assembly times with low fidelity assembly CAD models, in: M. Abramovici,
integrierten Produktentstehung (Ph.D. thesis), University of Saarland, 2007. R. Stark (Eds.), Smart Product Engineering, Springer-Verlag, Berlin-Heidelberg,
[10] U. Bracht, T. Masurat, The Digital Factory between vision and reality, Comput. 2013, pp. 777786.
Ind. 56 (4) (2005) 325333. [46] D. Nettleton, Commercial Data Mining: Processing, Analysis and Modeling for
[11] A. Bryan, J. Ko, S.J. Hu, Y. Koren, Co-evolution of product families and assembly Predictive Analytics Projects, Morgan Kaufmann, Waltham, 2014.
systems, Ann. CIRP 56/1/2007 (2007) 4144. [47] B.A. Nicholds, J. Mo, S. Bridger, Determining an action plan for manufacturing
[12] R.-S. Chen, K.-Y. Lu, P.-H. Tai, Optimizing assembly planning through a three- system improvement: a case study, Int. J. Agile Syst. Manage. 7 (1) (2014) 1
stage integrated approach, Int. J. Prod. Econ. 88 (2004) 243256. 25.
[14] W.-C. Chen, Y.-Y. Hsu, L.-F. Hsieh, P.-H. Tai, A systematic optimization [48] L. Ohno-Machado, H.S. Fraser, A. hrn, Improving machine learning
approach for assembly sequence planning using Taguchi method, DOE, and performance by removing redundant cases in medical data sets, Proc. AMIA
BPNN, Expert Syst. Appl. 37 (2010) 716726. Fall Symposium (1998) 523527.
[15] K. Choi, K.-Y. Kim, H.-J. Yang, Disparate attributes algorithm for semantic [49] D. Petzelt, J. Schallow, J. Deuse, S. Rulhoff, Anwendungsspezifische
assembly design rule management, Adv. Eng. Inform. 27 (2013) 5165. Datenmodelle in der Digitalen Fabrik, ProduktDaten J. 16 (1) (2009) 4548.
[17] W.W. Chu, Data Mining and Knowledge Discovery for Big Data, Methodologies, [50] J. Ponn, U. Lindemann, Konzeptentwicklung und Gestaltung technischer
Challenge and Opportunities, Springer-Verlag, Berlin, Heidelberg, 2014. Produkte Systematisch von Anforderungen zu Konzepten und
[18] A. Coralo, A. Margherita, G. Pascali, Digital mock-up to optimize the assembly Gestaltlsungen, Springer, Berlin, Heidelberg, 2011.
of a ship fuel system, J. Modell. Simula. Syst. 1 (2010) 412. [51] M. Putz, A. Richter, M. Pfeifer, Adaptive planning and optimization of joining
[19] F. Demoly, X.-T. Yan, B. Eynard, L. Rivest, S. Gomes, An assembly oriented and assembling sequences using parallel acting working units, CIRP Ann.-
design framework for product structure engineering and assembly sequence Manuf. Technol. 59 (2010) 5760.
planning, Robot. Comput.-Integr. Manuf. 27 (2011) 3346. [52] R.V. Rao, Advanced Modeling and Optimization of Manufacturing Processes.
[20] F. Demoly, A. Matsokis, D. Kirisits, A mereotopological product relationship International Research and Development, Springer-Verlag, London, 2011.
description approach for assembly oriented design, Robot. Comput.-Integr. [53] B. Rekiek, A. Delchambre, Assembly Line Design The Balancing of Mixed-
Manuf. 28 (2012) 681693. Model Hybrid Assembly Lines with Genetic Algorithms, Springer-Verlag,
[21] S. Dhanabal, S. Chandramathi, Review of various k-nearest neighbor query London, 2006.
processing techniques, Int. J. Comput. Appl. 31 (7) (2011) 1422. [54] G. Rock, K. Theis, P. Wischnewski, Variability management, in: J. Stjepandic
[22] K.L. Edwards, Towards more strategic product design for manufacture and et al. (Eds.), Concurrent Engineering in the 21st Century: Foundations,
assembly: priorities for concurrent engineering, Mater. Des. 23 (2002) 651 Developments and Challenges, Springer International Publishing,
656. Switzerland, 2015, pp. 491519.
[24] M. Eigner, R. Stelzer, Product Lifecycle Management Ein Leitfaden Fr [55] S. Rulhoff, R. Jalali Sousanabady, J. Deuse, C. Emmer, Concept and data model
Product Development Und Life Cycle Management, Springer, Berlin, for assembly work content determination, in: J. Zh (Ed.), Enabling
Heidelberg, 2009. Manufacturing Competitiveness and Economic Sustainability Proceedings of
[25] T.-H. Eng, Z.K. Ling, W. Olson, C. Mclean, Feature-based assembly modeling 5th CIRP Conference on Changeable, Agile, Reconfigurable and Virtual
and sequence generation, Comput. Ind. Eng. 3 (1999) 1733. Production (CARV2013), Springer, Mnchen, Germany, Berlin, Heidelberg,
[26] O. Erohin, P. Kuhlang, J. Schallow, J. Deuse, Intelligent utilisation of digital New York, 2012, pp. 353360.
databases for assembly time determination in early phases of product [56] J. Schallow, K. Magenheimer, J. Deuse, G. Reinhart, Application protocols for
emergence, Procedia CIRP 45th CIRP Conference on Manufacturing standardising of processes and data in digital manufacturing, in: H.A.
Systems vol. 3 (2012) 424429. ElMaraghy (Ed.), Enabling Manufacturing Competitiveness and Economic
[27] L. Graening, B. Sendhoff, Shape mining: a holistic data mining approach for Sustainability Proceedings of 4th CIRP Conference on Changeable, Agile,
engineering design, Adv. Eng. Inform. 28 (2014) 166185. Reconfigurable and Virtual Production (CARV2011), Springer, Montreal,
[28] S. Gupta, Computer-aided generation of modularized conceptual designs with Canada, Berlin, Heidelberg, New York, 2011, pp. 648653.
assembly and variety considerations (M.Sc. thesis), The Pennsylvania State [57] Robert B. Stone, Daniel A. McAdams, Varghese J. Kayyalethekkel, A product
University, 2008. architecture-based conceptual DFA technique, Des. Stud. 25 (2004) 301325.
[29] S. Gupta, G.E. Okudan, Computer-aided generation of modularized conceptual [58] B. Strug, G. Slusarczyk, Reasoning about designs through frequent patterns
designs with assembly and variety considerations, J. Eng. Des. 19 (6) (2009) mining, Adv. Eng. Inform. 23 (2009) 361369.
533551. [59] D. Talia, P. Trunfio, Service-Oriented Distributed Knowledge Discovery, CRC
[30] J. Han, M. Kamber, J. Pei, Data Mining: Concepts and Techniques, third ed., Press, Boca Raton, 2013.
Morgan Kaufmann Publishers, Waltham, 2012. [60] C.F. Tan, L.S. Wahidin, S.N. Khalil, An architecture framework of design for
[31] J. Hartung, J. Schallow, S. Rulhoff, Moderne Produktionsplanung Integration assemble expert system, in: Proc. of the Intl. Conf. on Advances, in: Civil,
in der Produktentstehung, ProduktDaten J. 19 (1) (2012) 2021. Structural And Mechanical Engineering CSM, 2014.
[32] T.C.-K. Huang, C.-C. Liu, D.-C. Chang, An empirical investigation of factors [61] L. Trygg, Concurrent engineering practices in selected swedish companies: a
influencing the adoption of data mining tools, Int. J. Inf. Manage. 32 (2012) movement or an activity on the few?, J Prod. Innov. Manage 10 (1993) 403
257270. 415.
[33] P. Kim, Y. Ding, Optimal engineering system design guided by data-mining [62] M. Vielhaber, Design to knowledge - a root design principle, in: P. Scharff, P.
methods, Technometrics 47 (3) (2005) 336348, August. Kurtz (Eds.), Proceedings of the 56th International Scientific Colloquium
[34] R. Kohavi, A study of cross-validation and bootstrap for accuracy estimation (IWK), TU Ilmenau, Ilmenau, 2011.
and model selection, Proceedings of the 14th International Joint Conference on [65] S. Wartzack, Predictive Engineering Assistenzsystem zur multikriteriellen
Artificial intelligence, (IJCAI95), vol. 2, Morgan Kaufmann Publishers Inc., San Analyse alternativer Produktkonzepte, Forschritt-Berichte VDI, Reihe 1, Band
Francisco, 1995, pp. 11371143. 336, Dsseldorf, VDI-Verlag, 2001.
[36] F.Z. Krumenauer, C.T. Matayoshi, I.B. da Silva, M.S. Filho, G.F. Batalha, [66] C. Weber, CPM/PDD an extended theoretical approach to modelling products
Concurrent engineering and DFMA approaches on the development of and product development processes, in: H. Bley, et al. (eds.), Proceedings of the
automotive panels and doors, J. Achievements Mater. Manuf. Eng. 31 (2) 2nd German-Israeli Symposium on Advances in Methods and Systems for
(2008) 690698. Development of Products and Processes, Stuttgart, Fraunhofer-IRB-Verlag,
[37] T.C. Kuo, S.H. Huang, H.C. Zhang, Design for manufacture and design for X: 2005.
concepts, applications, and perspectives, Comput. Ind. Eng. 41 (2001) 241 [67] C. Weber, Produkte und Produktentwicklungsprozesse mit Hilfe von
260. Merkmalen und Eigenschaften eine kritische Zwischenbilanz, in: D. Krause,
[38] A.J.D. Lambert, Optimal disassembly sequence generation for combined K. Paetzold, S. Wartzack (eds.), Beitrge zum 23, Symposium DfX, Hamburg,
material recycling and part reuse, in: Proceedings of the 1999 IEEE TuTech, 2012.
International Symposium on Assembly and Task Planning, 1999, pp. 146151. [68] D.E. Whitney, Mechanical Assemblies: Their Design, Manufacture, and Role in
[39] M.C. Leu, H.A. ElMaraghy, A.Y.C. Nee, S.K. Ong, M. Lanzetta, M. Putz, W. Zhu, A. Product Development, Oxford University Press, Oxford New York, 2004.
Bernard, CAD model based virtual assembly simulation, planning and training, [69] X. Xu, L. Wang, S.T. Newman, Computer-aided process planningA critical
CIRP Ann. Manuf. Technol. 62 (2013) 799822. review of recent developments and future trends, Int. J. Comput. Integr. Manuf.
[40] Y.-B. Li, G.-L. Chen, X.-M. Lai, S. Jin, Y.-F. Xing, Knowledge-based vehicle body 24 (1) (2011) 131.
conceptual assembly design, Proc. I Mech E Part D J. Automob. Eng. 222 (2) [70] Y. Yin, I. Kaku, J. Tang, J.M. Zhu, Data Mining: Concepts, Methods and
(2008) 221234. Applications in Management and Engineering Design, Springer-Verlag,
[41] B. Lotter, H.-P. Wiendahl, Montage in Der Industriellen Produktion, Ein London, 2011.
Handbuch Fr Die Praxis, 2. Auflage, Springer-Verlag, Berlin-Heidelberg, 2013. [71] Y. Yusof, K. Latif, Survey on computer-aided process planning, Int. J. Adv.
Manuf. Technol. 75 (2014) 7789.

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006
R. Kretschmer et al. / Advanced Engineering Informatics xxx (2017) xxxxxx 15

[72] Y.Y. Yusuf, M. Sarhadi, A. Gunasekaran, Agile manufacturing: the drivers, Engineering Conference IDETC/CIE 2010, August 1518, 2010, Montreal,
concepts and attributes, Int. J. Prod. Econ. 62 (1999) 3343. Quebec, Canada.
[73] D. Zhang, P.L. Yu, P.Z. Wang, State-dependent weights in multicriteria value [23] K. Ehrlenspiel, Integrierte Produktentwicklung Denkablufe,
functions, J. Optim. Theory Appl. 74 (1) (1992) 121. Methodeneinsatz, Zusammenarbeit, Hanser, Mnchen, Wien, 2009.
[35] R. Kretschmer, S. Rulhoff, J. Stjepandic, Design for assembly in series
production by using data mining methods, in: J. Cha et al. (Eds.), Moving
Further reading Integated Product Development to Service Clods in Global Economy, IOS
Press, Amsterdam, 2014, pp. 617626.
[4] R.C. Beckett, Functional system maps as boundary objects in complex system [63] R. Wallis, J. Stjepandic, S. Rulhoff, F. Stromberger, J. Deuse, Intelligent
development, Int. J. Agile Syst. Manage. 8 (1) (2015) 5369. utilization of digital manufacturing data in modern product emergence
[13] W.-C. Chen, P.-H. Tai, W.-J. Deng, L.-F. Hsieh, A three-stage integrated processes, in: J. Cha et al. (Eds.), Moving Integated Product Development to
approach for assembly sequence planning using neural networks, Expert Syst. Service Clods in Global Economy, IOS Press, Amsterdam, 2014, pp. 617626.
Appl. 34 (2008) 17771786. [64] H. Wang, Y. Rong, D. Xiang, Mechanical assembly planning using ant colony
[16] M.-C. Chiu, G. Okudan, Evolution of design for X tools applicable to design optimization, Comput. Aided Des. (2014).
stages: a literature review, in: Proceedings of the ASME 2010 International
Design Engineering Technical Conferences & Computers and Information in

Please cite this article in press as: R. Kretschmer et al., Knowledge-based design for assembly in agile manufacturing by using Data Mining methods, Adv.
Eng. Informat. (2017), http://dx.doi.org/10.1016/j.aei.2016.12.006