You are on page 1of 23

Available online at www.sciencedirect.

com

Computers and Chemical Engineering 32 (2008) 320–342

An ontology-based approach to knowledge


management in design processes
Sebastian C. Brandt a,∗ , Jan Morbach b , Michalis Miatidis a ,
Manfred Theißen b , Matthias Jarke a,c , Wolfgang Marquardt b
a Informatik 5 (Information Systems), RWTH Aachen University, Ahornstr. 55, 52056 Aachen, Germany
b Process Systems Engineering, RWTH Aachen University, Turmstr. 46, 52056 Aachen, Germany
c Fraunhofer FIT, Schloss Birlinghoven, 53754 St. Augustin, Germany

Received 24 January 2007; received in revised form 5 April 2007; accepted 10 April 2007
Available online 24 April 2007

Abstract
Engineering design processes comprise highly creative and knowledge-intensive tasks that involve extensive information exchange and commu-
nication among distributed teams. In such dynamic settings, traditional information management systems fail to provide adequate support due to
their inflexible data structures and hard-wired usage procedures, as well as their restricted ability to integrate process and product information. In
this paper, we advocate the idea of Process Data Warehousing as a means to provide a knowledge management and integration platform for such
design processes. The key idea behind our approach is a flexible ontology-based schema with formally defined semantics that enables the capture
and reuse of design experience, supported by advanced computer science methods.
© 2007 Elsevier Ltd. All rights reserved.

Keywords: Process Data Warehousing; Engineering design; Ontologies; Information management; Knowledge management; Design repository

1. Introduction that emphasize the collection and organization of knowledge


(McMahon, Lowe, & Culley, 2004). We only consider the latter
Knowledge about engineering design processes constitutes approach, stressing the automated recording of the application
one of the most valuable assets of a modern enterprise. Normally, of implicit design knowledge. Thus, experience knowledge is
this knowledge is only known implicitly to the participating captured that can be shared and reused later on. This poses the
designers, relying heavily on the personal experience back- challenges on knowledge management systems (KMS) that are
ground of each designer. To fully exploit this intellectual capital, described below.
it must be made explicit and shared among designers and Typically, a vast amount of design knowledge is manipulated
across the enterprise. Consistent and comprehensive knowledge by legacy tools and stored in highly heterogeneous sources, such
management methods need to be applied to capture and inte- as electronic documents and databases. Thus, a KMS should
grate the individual knowledge items emerging in the course of (a) provide a comprehensive representation of the contents of
an engineering design project. Knowledge management (KM) these sources, thereby correlating the scattered knowledge items
is a scientific discipline that stems from management theory and providing a single point of access to design knowledge.
and concentrates on the systematic creation, leverage, shar- As such a comprehensive representation cannot be complete
ing and reuse of knowledge resources in a company (Awad & (for reasons of scaling, maintainability, practicability, etc.), a
Ghaziri, 2003). Knowledge management approaches are gener- KMS should (b) employ mechanisms to easily locate the orig-
ally divided into personalization approaches that focus on human inal knowledge sources, where more detailed information can
resources and communication, and codification approaches be retrieved. To this aim, (meta) information about the sources
(e.g., type, structure, version history, storage location) should be
combined with information about their contents, and the KMS
∗ Corresponding author. Tel.: +49 241 80 21514. should be integrated with existing tools and data stores to pro-
E-mail address: sbrandt@cs.rwth-aachen.de (S.C. Brandt). mote easy access to the original sources. In addition, the KMS

0098-1354/$ – see front matter © 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2007.04.013
S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342 321

should (c) enable the capture and archival of work processes in processes in these domains are relatively well-documented,
order to provide information about the circumstances in which strict, and deterministic, allowing a prescriptive definition of
the individual knowledge items have been created. In particular, the activities to be executed. In contrast, chemical engineering
recording of the decision-making procedures allows to recall is concerned with the development of individual and highly cus-
the design rationale applied at that time. This allows to sys- tomizable products (i.e., chemical plants); therefore, the design
tematically retrieve experience knowledge that is suitable for a processes are more dynamic and do not allow for deterministic
particular situation, and evaluate its applicability to the current approaches.
working context. Moreover, best practices abstracted from the In this context, two different types of information resources
recorded work processes can be exploited for direct process sup- need to be distinguished. On the one hand, the products created
port (i.e., the partially automated enactment of routine tasks in as part of the design processes (e.g., process flow diagrams, sim-
design tools). ulation models, design calculations, cost estimates, etc.) should
This paper presents a novel approach for supporting creative, be considered. These may be organized into documents, which
non-deterministic1 design processes that addresses the above act as logical units to enable work distribution or version con-
demands: based on the concepts introduced by Jarke, List, and trol. On the other hand, there are the work processes or activities
Köller (2000), our Process Data Warehouse (PDW) allows the themselves, in which the products are created, used or manip-
capture and reuse of design knowledge – product data, doc- ulated. Any kind of integrated support needs to take both the
uments, work processes, and decision-making procedures, as product and the process aspect into account. In the following,
well as their interdependencies – through ontologies. In this con- we discuss to which degree the different types of software solu-
text, the term “ontology” denotes a conceptual data schema that tions available today support the management of product and
represents the relevant domain entities and their relationships process knowledge.
by means of classes and relations. Ontologies are flexible data Document Management Systems such as Windream
structures that can be changed and adapted at run time; more- (Windream GmbH, 2006) or Documentum (EMC2 , 2006) are
over, they allow representing the semantics of design knowledge widely used in industrial practice for the storage, maintenance,
in a formal way the computer can interpret. These characteristics and distribution of documents. Their process support is restricted
enable the provisioning of advanced computer science methods to micro-level processes like versioning and to the execution of
for managing, enriching, and searching the knowledge within simple workflow definitions, offering a subset of the function-
the PDW. ality of standard workflow management systems. Some of these
The paper is organized as follows: the next section investi- systems provide collaborative extensions, such as the Documen-
gates the possibilities of supporting technical design processes tum eRoom. These offer additional functionalities for intra- and
by experience management. In Section 3, we introduce our inter-enterprise collaboration, and for the flexible definition and
approach of the Process Data Warehouse for knowledge cap- execution of cooperative and coordinative workflows.
ture and experience reuse. Section 4 illustrates how the PDW A step further, Product Data Management (PDM) systems
is applied in process engineering design. Methods and tools for provide extended facilities for the handling of detailed product
knowledge capture are described in Section 5. In Section 6, we information, ranging from design to production stage. Nowa-
discuss how our work correlates with the research activities of days, such systems are routinely applied for the manufacturing
other groups. Finally, conclusions are drawn, including further of mechanical and electronic devices. They are being succeeded
on-going and future research. by Product Lifecycle Management (PLM) systems, such as
Windchill (PTC, 2006a), TeamCenter (UGS, 2006) or CATIA
2. Supporting creative design processes (Dassault Systemes, 2006). The aim of these systems is to inte-
grate information on the manufacturing processes (usually CAM
Experience that can be gathered and recorded using knowl- systems) with design data (CAD systems) on the one hand,
edge management methods, is intended to be used for supporting and information about Enterprise Resource Planning (ERP) pro-
designers when executing creative design processes (Jarke cesses on the other hand.
& Marquardt, 1996). Those conceptual design processes, as The PDM/PLM systems available today adequately support
found in chemical engineering, are highly dynamic and non- information exchange between developers, especially in the later
deterministic, and therefore hardly predictable (Marquardt & phases of the engineering lifecycle which are characterized by
Nagl, 2004; Westerberg, Subrahmanian, Reich, Konda, & the more deterministic and well-known processes. Nowadays, they
n-dim group, 1997). Any software solution has to cope with also provide improved possibilities to integrate a company’s
the continually changing requirements and the many degrees legacy systems, provided that the company is willing to tackle
of freedom within these processes. Appropriate process support the required integration effort (Mesihovic, Malmqvist, & Pikosz,
for the chemical engineering domain is even harder to achieve 2004). However, they lack essential capabilities for the manage-
than for domains like automotive engineering where stan- ment and reuse of design knowledge (Bilgic & Rock, 1997;
dardized, well-understood products are developed: the design Gao, Aziz, Maropoulos, & Cheung, 2003; Maropoulos, 2003)
and are less suited for the conceptual design stage (Gao et al.,
2003; Mesihovic et al., 2004).
1 That is, work processes that cannot be completely planned and/or scheduled A significant shortcoming of existing PDM and PLM systems
in advance due to initially incomplete knowledge, and uncertainties. criticized by many authors, is their lack of adequate informa-
322 S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342

tion models for product representation (e.g., Szykman, Sriram, it has been envisioned in the previous section. Since each of
& Regli, 2001). These models would be needed to effectively these tools is based on its own proprietary information model
capture, exchange, retrieve, and reuse design knowledge. In and contains only some generic import and/or export functions,
particular, formal and well-structured information models for only part of the product knowledge created during the design
the conceptual design stage are missing that reflect different process can be captured for reuse. Reuse of product knowledge
perspectives such as function, behavior, and structure (e.g., is further hindered by the lack of adequate information models
Bilgic & Rock, 1997; Mesihovic et al., 2004; Szykman et al., and the tools’ insufficient retrieval capabilities. As for the pro-
2001). cess aspect, none of the tools offers process support beyond
Regarding process support, current PDM systems have prescriptive workflow management and micro-level services.
largely focused on the support of micro-level processes on the As mentioned before, these deterministic approaches are not
administrative level, such as versioning or engineering change adequate for conceptual design processes which are not suffi-
management (Mesihovic et al., 2004). Some attention is paid ciently well understood and involve activities of multi-objective
to project management, however without reaching the capa- decision-making based on incomplete information.
bilities of full-fledged project management systems (Bilgic & Thus, to move beyond the approaches described so far, there
Rock, 1997; Mesihovic et al., 2004). They lack the functional- is a need for integrated technical support in the early phases of
ity to capture complex work processes and decisions, as well as conceptual design. Therefore, the authors’ work groups have
capabilities for direct process support. examined the possibilities offered by recording and reusing
While PLM systems are widely used in the manufacturing product and process knowledge. To stress the integration and
industries, they are less common in the chemical engineer- interleaving of product and process data, the underlying concept
ing domain. Instead, specialized Computer Aided Engineering has been named the Process Data Warehouse.
(CAE) systems are used, which resemble classical PDM solu-
tions but are specifically adapted to the needs of the chemical 3. The Process Data Warehouse
process industries. Examples are Aspen Zyqad (AspenTech,
2006), Comos PT (Innotec, 2006), SmartPlant P&ID In this section, the concept of the Process Data Warehouse as a
(Intergraph, 2006a), and VANTAGE (AVEVA, 2006a). Like traces-based knowledge management system will be described.
PDM systems, these CAE systems focus on geometrical data and It records the traces of design processes in its experience repos-
flowsheets; other plant-related information are also considered itory and thus allows them to be reused for supporting creative
but to a much lesser degree. Information models of CAE systems design processes. Section 3.1 introduces the so-called Core
suffer the same deficiencies as PDM systems, that is the lack of Ontology, and Section 3.2 describes some ontological exten-
well-structured product representation, especially for the early sion modules. Section 3.3 shows how the ontological model is
design phases (e.g., Bayer, 2003; Bayer & Marquardt, 2003). used for the implementation of the PDW.
Like PDM systems, CAE systems offer process support on
the micro-level but fail to support more complex development 3.1. A Core Ontology for engineering knowledge
activities. They lack capabilities for the capture of work pro- management
cesses and decisions, nor do they offer direct process support.
Project management is supported to some degree. The concept of the Process Data Warehouse has been
The CAE systems available today do not cover the entire derived from the concept of Data Warehousing (Jarke, Lenzerini,
lifecycle of a chemical plant (Schneider & Marquardt, 2002). Vassiliou, & Vassiliadis, 2003) where large amounts of fixedly
Therefore, vendors offer information integration environments structured data (e.g., from sales or accounting) are stored, aggre-
such as Optegra (PTC, 2006b) to integrate and consolidate gated, and then presented. For those conventional tools and
information from heterogeneous sources. Some solutions have warehouses, fixed schemas are used for data storage. Yet to
been developed specifically for the domain of chemical engi- support design and other creative work processes, dynamic
neering, such as the Process Engineering Data Warehouse extension of the existing data structures and integration of addi-
(PEDW, Bayer Technology Services, 2006), SmartPlant Foun- tional, suitable domain models2 must be supported (Jarke et al.,
dation (Intergraph, 2006b), and VNET (AVEVA, 2006b). These 2000).
solutions aim to correlate design data and technical specifica- The PDW uses ontologies to represent the domain models.
tions obtained from multiple sources (CAE systems, process In computer science, an ontology is usually understood as a
simulators, and other engineering tools) with other informa- “formal, explicit specification of a shared conceptualization”
tion, such as process studies, material balances, property data, (Studer, Benjamins, & Fensel, 1998); that means, an ontology is
and project-related data. Data import is mainly realized through formal model which explicitly represents the consensual knowl-
XML and is often restricted to the proprietary formats of the edge of a domain. The domain entities are modeled through
vendors and their cooperation partners. The tools offer services classes and relations. By instantiating these ontological con-
such as version management, change management, notification cepts, concrete facts and information items can be stored in the
(if some changes have been committed), and simple navigation ontology.
and retrieval functionality.
In summary, none of the commercial solutions currently
available is able to provide knowledge management support as 2 A domain model is an information model for a particular application domain.
S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342 323

Fig. 1. Simplified view of the Core Ontology and some peripheral ontology modules.

Ontologies have two major advantages over conventional mutual dependencies and their structural decomposition. The
data schemas: firstly, they are highly flexible, enabling mod- Product class denotes all kinds of information elements, such
ifications and extensions of the data structures even during as data items or decision representation objects, which are
project execution. Secondly, they are represented in a machine- created or modified during a work process. Products can be
readable, logic-based language, which allows to formally define aggregated into DocumentVersions. The different versions of
the semantics of the ontological concepts. This enables the com- a Product can be bundled by a VersionSet. A specialization of
puter to interpret and reason with the information stored in the VersionSet is the (logical) Document, which bundles different
ontology. In consequence, advanced support for information DocumentVersions.
management and retrieval can be provided: for example, it is pos- • Storage Area. The Storage Area (right) describes at which
sible to perform a semantic search on the ontology. In contrast to location (StoragePlace) inside a Store a particular Version-
other, simpler forms of search (e.g., keyword based), a semantic Set is stored. Examples of stores are databases, Document
search allows for a more systematic retrieval of information. Management Systems, and external tools. Instances of the
The PDW has been constructed on top of loosely connected Transformation class control the integration of external data,
ontology modules which are held together by a central Core e.g., by defining XML document transformation based on
Ontology (Brandt, Morbach, et al., 2006; Brandt, Schlüter, & XSLT (W3C, 1999).
Jarke, 2006a). The Core Ontology introduces top-level concepts • Descriptive Area. The Descriptive Area (left) contains
that describe products and processes, as well as their interre- basic concepts for describing the content or the role of
lations and dependencies, independently from any particular ProductObjects on a high semantic level. This includes Con-
domain or application. These fundamental concepts are then tentDescriptions and Categorizations, which are grouped into
refined and concretized within peripheral ontology modules. CategorizationSchemes. Unlike Categorizations which sim-
Different ontology modules can be added to the Core Ontol- ply classify ProductObjects according to certain categories,
ogy to flexibly adapt the PDW to the requirements of a specific a ContentDescription is a placeholder for a term (or even a
application domain; some exemplary extensions are presented in composite expression) from a controlled vocabulary that char-
Section 3.2. Fig. 1 displays a simplified view of the Core Ontol- acterizes the contents of the associated ProductObject. Thus,
ogy. Four prominent areas of conceptualization are arranged the descriptive area provides content-specific meta informa-
around the Object as the central concept. tion for the annotation of documents and data stores.
• Process Area. The Process Area (bottom) contains the con-
• Product Area. The Product Area (top) contains concepts for cepts needed to represent the ProcessObjects that manipulate
the description of the type and version history of electronic (i.e., create, use, or modify) ProductObjects. Process-
documents and other information resources, as well as their Elements comprise general process definitions (Activities)
324 S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342

such as coarse-grained guidelines for designers or fine- a large-scale ontology for the domain of computer-aided pro-
grained process steps for direct process support. They also cess engineering. OntoCAPE was originally developed in the
comprise ProcessTraces that describe the concrete actions COGents project (Braunschweig et al., 2004) and has been
performed in a project by Users or (software) Tools. If refined in the context of this work. It consists of a set of intercon-
ProcessTraces have been executed according to a process nected partial models which are hierarchically organized across
definition given by an Activity, they are related by charac- several levels of abstraction. Important areas covered by Onto-
terizedBy. CAPE are substances and their thermophysical properties, unit
operations and plant equipment, as well as mathematical mod-
Knowledge is not only contained in information, but also eling. The design of OntoCAPE emphasizes extensibility and
in the relationships among information items (Rus & Lindvall, ease of customization. Accordingly, only the higher level domain
2002). Thus, many relationships have been explicitly modeled entities are specified in the ontology at deployment time. The
that associate classes inside one area, or between several of the users can then easily refine and extend the ontology according
conceptualization areas. The following may serve as examples: to the needs of their respective organizations. The customization
process is guided by several mechanisms, such as the specifica-
• describedBy connects the ProductObjects with descriptive tion of a default structure for the partial models, the provision of
elements that describe their content or category; design patterns to suggest best-practice solutions for common
• storedAt represents the storage of products and documents in design problems, and the definition of a meta model that repre-
physical locations such as file systems or repositories; sents the underlying design principles of the ontology and is thus
• manipulatedBy describes how the ProductObjects are created, able to enforce design consistency when changing or extending
used, and modified by the users during process activities; the ontology.
• a generic dependsOn relation allows to introduce and derive In the PDW, OntoCAPE extends the Descriptive Area by
specialized relations between Objects of any type. refining the ContentDescription class, thus providing a domain-
specific vocabulary for the description of ProductObjects and
ProcessObjects.
Some of the concepts described here, especially those in
the Product and Process Areas, have been derived from the
3.2.2. Chemical engineering documents
traceability reference model introduced by Ramesh and Jarke
The ontology module Chemical Engineering Documents
(2001). Abstracted from a large number of industrial case stud-
refines the generic Document class in the Product Area of the
ies, the model describes the conceptual relations of product- and
Core Ontology. It provides a taxonomy of the document types
process-related traces in more detail.
that are typically used in chemical process design, such as Pro-
For reasons of interoperability, the Web Ontology Language
cess & Instrumentation Diagrams, equipment lists, or model
OWL (Bechhofer et al., 2004), a W3C standard for the Seman-
files. A detailed description of this ontology module can be found
tic Web (Berners-Lee, Hendler, & Lassila, 2001), would be the
in Morbach, Hai, Bayer, and Marquardt (2007, chap. 2.3). In an
first choice for the representation of ontologies. However, since
analogous manner, alternative taxonomies can be developed by
current OWL-based ontology repositories do not offer an effi-
the users of the PDW to represent the document types that exist
ciently searchable storage in relational databases, the KAON
in the users’ organization.
system (Oberle, Volz, Motik, & Staab, 2004) is used instead.
KAON is based on the same set of base standards (RDF, RDFS,
3.2.3. Document Management
see Manola & Miller, 2004) as OWL. KAON enables efficient
The Storage Area offers the basic models for file storage
search and retrieval mechanism directly on the database back-
inside a Document Management System (DMS), correlating file-
end, at the cost of loosing some of the expressiveness of OWL
based documents with their conceptual representation inside the
(cf. discussion in Section 7).
PDW. This allows to locate the documents in their physical
storage places. Thus, the original documents can be accessed.
3.2. Selected extension modules Additionally, change notifications from such DMS can be cap-
tured.
The Core Ontology only contains high-level concepts The Document Management extension module contains the
required to establish, organize, and integrate the four fundamen- Store refinements for various kinds of DMS, including Docu-
tal Areas. In addition, each Area offers some extension points mentum (EMC2 , 2006), Jakarta Slide (Slide, 2006), and simpler
where ontology modules for specific application domains or systems that do not support version control, such as network
other specializations can be added. These modules refine the file systems. For each Store, specialized classes of StoragePlace
domain-independent concepts introduced in the Core Ontology exist that define the physical location of a document inside the
(refinement is indicated by dashed arrows in Fig. 1). Exemplar- Store, e.g., by giving a path and file name.
ily, some extensions are presented in the following. Other kinds of repositories – relational databases, interfaces
to external tools, etc. – also refine the Store class, and use refine-
3.2.1. OntoCAPE ments of StoragePlace to define uniquely identifying attributes
The most elaborate of these extensions is OntoCAPE (e.g., primary keys or id tags) to represent the physical storage
(Morbach, Yang, & Marquardt, 2007; Yang & Marquardt, 2004), elements.
S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342 325

To achieve an integration of the external systems, it is of of the Product class of the Core Ontology. Instances of classes
course necessary to develop specific program code for each of this module can be further specified using the concepts of the
external system, and to connect this program code with the onto- descriptive area. For instance, the goal of designing a chemi-
logical class that represents the Store refinement (cf. Section cal reactor with a product stream of 40,000 t/year can easily be
5.2). modeled by connecting (i.e., refining) an instance of Goal with
concepts from OntoCAPE.
3.2.4. Work Process & Project Management
The Work Process & Project Management module covers pro-
3.2.7. Application tools
cess models which prescribe (part of) the work processes to be
The extension application tools contains a representation of
performed by a designer or a design team confronted with a cer-
the software tools that can support a designer when executing a
tain task. Typically, these process models contain patterns, such
work process. Activities can be associated with the Tools that are
as sequences of Activities or alternative sub-processes. Coarse-
needed to enact them. Thus, tool environments may be prepared,
grained process models are represented using a process ontology
preconditions like tool usability may be checked, and users may
based on the C3 modeling language (Eggersmann et al., 2007,
be guided in using the appropriate tool. ProcessTraces can be
chap. 2.4), a modified version of UML activity diagrams (OMG,
related to the Tool the changes have been executed in.
2004) which provides the required expressivity for the weakly
Additionally, most software tools only work with certain
structured design processes in chemical engineering.
document types (or other resource types). By relating this
At a more fine-grained level, the designer performs his/her
information to a Tool instance itself, the information resources
assigned tasks following various methodologies while using
resulting from tool use or needed for using the tool can be
interactive software tools. Because of design creativity, only few
identified.
of the methodologies followed are well understood and can be
adequately described using a fine-grained process model. We
refer to such well-understood methodologies as method frag- 3.3. Architecture of the PDW
ments. In our case, we have extended the Process Area of the
Core Ontology with concepts for formalizing method fragments In the preceding sections, the general problems of supporting
using the NATURE process meta model (Rolland, Souveyet, & creative processes have been described, and the Core Ontology
Moreno, 1995). Briefly, the NATURE meta model promotes the has been introduced. In this section, the architectural structure
contextualization of process knowledge and represents method of the PDW experience repository as a trace recording and reuse
fragments as tuples of arbitrary contexts. A context relates the framework is shown. Furthermore, it is described how the PDW
current process situation to the goal of the designer. Thus, a con- interacts with the software environment it is realized in. Follow-
text might either represent a specific method fragment that has to ing the authors’ concept of a posteriori integration, the existing
be followed, or the need for a choice among several alternative software tools are augmented by integrating them with the PDW
contexts, or even a simple step operationalized by a tool action. framework. While the engineers can go on working with the
tools they are accustomed to, additional support functionality is
3.2.5. Design traces offered by the integrated environment.
Whereas prescriptive process models capture an ideal Fig. 2 shows how the experience repository of the PDW is
abstraction of the reality, design traces give a faithful descrip- embedded into its usage environment. The PDW server itself
tion of the real process that provides evidence how the process can be seen on the right of the figure. The KAON ontology
evolves over time (Ramesh & Jarke, 2001). This information framework (Oberle et al., 2004) is responsible for managing the
can be used, for example, to backtrace poor design decisions, ontology classes of the PDW and the corresponding instances.
or to crystallize knowledge out of successful ones. The Design All this data is stored in an SQL2-compatible relational database
Traces module, based on the ProcessTrace class, comprises con- (PostgreSQL, 2006), mapping the flexible ontology structure
cepts for structuring traces recorded during process execution. onto a fixed relational schema. Additionally, KAON offers a
More specifically, concepts are provided for the description of query mechanism that is able to transform requests in a high-
the sequence of performed process steps, the states of the prod- level semantic query language into SQL queries which can be
uct being worked upon by the process steps, as well as the performed directly on the relational storage back-end. Thus, only
relationships between process and product traces. those instances that either result from the query or are navigated
to need to be loaded from storage.
3.2.6. Decision documentation Around this ontology framework, the PDW-Model forms
The decision documentation module, based on the Deci- an implementation wrapper that represents the Core Ontol-
sion Representation Language (DRL) by Lee and Lai (1996), ogy and offers other specialized functionality. It gives access
provides concepts for representing the rationale on which a to the ontological concepts both in a generic way (as classes,
design decision is based during a particular project (Theißen & instances, association types, and relations) and by specialized
Marquardt, 2007, chap. 2.5). For example, a Goal can represent interfaces.
a constraint to be met for a particular DecisionProblem, while The realization of the server as described here is placed in an
an Alternative denotes a design which is to be evaluated based application server based on the J2EE programming language and
on Goals. The decision documentation module is a refinement platform (Sun, 2006). This enables the PDW to offer services
326 S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342

Fig. 2. Overview of the PDW architecture.

based on various standardized communication protocols: work with customized interfaces that offer appropriate solutions
for their problems. Semi-specific interfaces can be used to uni-
• Web services are nowadays often used for service-based formly visualize different concepts of the ontology that have
integration of enterprise applications, thus offering company- similar characteristics. This allows to define general rules for
spanning workflows that can be dynamically composed. presentation as model annotations, e.g., to display composition
Especially in a cooperative setting, this allows to extend the hierarchies or networks of blocks and streams of various kinds,
PDW with well-defined and standards-based service inter- using the same user interface.
faces. The PDW system is in several ways integrated with its
• Access for thin clients (i.e., web browsers) is provided by external environment. As already mentioned, this external envi-
web applications that are using the technologies of Servlets ronment is mainly kept unchanged, it is only extended. A
and Java Server Pages (Sun, 2006). Document Management System, such as Documentum (EMC2 ,
• Access to the server from rich clients (i.e., stand-alone client 2006) or Jakarta Slide (Slide, 2006), is responsible for support-
applications) is realized via Enterprise Java Beans, EJB (Sun, ing document-based storages, offering version management and
2006). change notification support. When an expert checks in a new
version of a document into the DMS, the PDW front-end run-
In the center of Fig. 2, such a rich client application can be ning on his or her workplace is notified. The ontology instances
seen. This application, the PDW front-end, forms the primary corresponding to the document, its new version, and its storage
means of accessing the PDW experience server and repository. place are created or updated and shown to the expert. The doc-
The front-end contains a client-side representation of a part of the ument instance can then be enriched semi-automatically with
PDW’s classes and instances, thus reflecting the current state of the current organizational context. Possibly, the content of the
information inside the server. Changes to the client-side data are document can be transformed from the external tool’s format
transferred to the server and persisted in a database transaction. into a representation inside the PDW’s repository, provided
All those changes are then communicated to the running clients, that the format can be automatically interpreted (cf. Section
allowing the clients to synchronize with the current state of 5.2).
information, and, possibly, display the newly created or changed Other external tools, mainly those that use repositories or
information. databases for storage purposes instead of simple files or docu-
Three different kinds of graphical user interfaces are offered ments, can be integrated as external data sources as well. This
by the PDW front-end. The generic display is based on UML often applies to ERP or CAE systems. Their data can usually be
notation (OMG, 2004) of the classes and instances of the onto- accessed by the programming interfaces offered by these tools,
logical models. It displays classes and instances as blocks of read and converted into PDW storage, enriched with semantic
different colors with their attribute types and values, and rela- information, and transferred to and from other tools for subse-
tions as links between them. This view is mainly aimed at the quent design steps. A generic mechanism for integrating external
ontology engineer responsible for modeling and integrating a data sources has been implemented using XML (W3C, 2006)
particular domain into the system, and for rapid prototyping of as an exchange format, and XSLT (W3C, 1999) to transform
possible use cases. Specific user interfaces need to be created the intermediate files into a form that matches the conceptual
for the various application cases, so that the domain users can representation of the PDW (for details, see Section 5.2).
S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342 327

Fig. 3. Query for polyamide-6 reactor equipment.

4. Using the PDW 2. Review and reuse of retrieved experience knowledge: The
retrieved information needs to be reviewed by the user, who
In Section 4.1, we describe an exemplary session with the can then decide on reusing some or all of the experience
PDW to illustrate its usage in an engineering design project; knowledge in the tools that form the current work context.
usage is demonstrated by means of an application scenario from 3. Knowledge capture: The design knowledge created by the
the area of conceptual process design. As a single scenario can- user when processing the current task is recorded and stored
not cover all the application possibilities of the PDW, we will in the PDW in order to increase the knowledge base.
discuss other use cases in Section 4.2 in a systematic manner.
In the following, we discuss how the three steps are carried
4.1. A scenario in process design out in this particular application scenario.
As application scenario, we consider the conceptual design 4.1.1. Step 1a: Manual search for experience knowledge
of a chemical reactor for polyamide-6, to be performed by To gain a first insight into the new task, the engineer wants
a single engineer who is part of an interdisciplinary design to search the PDW for experience knowledge created by former
team. The chosen scenario is part of a large case study on the design projects. In particular, he is interested in the reactor equip-
design of a polyamide-6 production facility. This case study has ment that is typically used for the production of polyamide-6. To
been elaborated in the Collaborative Research Center (CRC) this end, he formulates a query to the PDW by selecting appro-
IMPROVE (Marquardt & Nagl, 2004) in cooperation with indus- priate concepts from the PDW ontologies. A graphical query
trial partners (e.g., Bayer AG). A detailed description of the editor supports the composition of such queries in a “query-by-
entire case study can be found elsewhere (Bayer, Eggersmann, example”-based fashion. Ontological classes can be employed as
Gani, & Schneider, 2002; Eggersmann, Hackenberg, Marquardt, query variables; that way, all knowledge items that are (seman-
& Cameron, 2002; Nagl, Westfechtel, & Schneider, 2003). tic) instances of the respective class will be retrieved. The query
At the outset of the application scenario, the engineer is variables can be further specified by using relations and already
assigned the task of developing a conceptual design for this known instances from the ontology.
polyamide-6 reactor. The only specifications given at this point As it is shown in Fig. 3, the class ChemicalReactor is chosen
are the feedstock – polyamide-6 is to be produced from water and as a first query variable.3 However, since the semantics of this
caprolactam – and the desired amount and material properties class is fairly general – it is defined as “any apparatus within
of the polyamide-6 product. To come up with a suitable pro- which a chemical reaction takes place” – the query would pro-
cess design, the engineer can draw on the experience knowledge duce a large number of unwanted results. Thus, the query needs
stored in the PDW. to be refined—for instance, by constraining the type of chemical
Typically, a sample session with the PDW consists of three reaction that takes place in the reactor. This is done by means
intertwined steps, each reflecting a different mode of operation: of a second query variable, namely ChemicalReaction, which is
linked to ChemicalReactor via the ontological relation takesPla-
1. Retrieval of experience knowledge: The user searches the
ceIn (cf. Fig. 3); takesPlaceIn defines the semantic relationship
PDW for experience knowledge matching his/her current
between the two query variables. Finally, the ChemicalReaction
task. To this aim, the PDW supports two modes of operation:
is specified by indicating the desired reactants (water and capro-
a. The user may either actively search the PDW by compos-
lactam) as well as the product (polyamide-6) of the reaction; the
ing queries manually, or
b. he may rely on the PDW’s ability to automatically derive
queries from the current work context of the application 3 ChemicalReactor is a class defined in the OntoCAPE ontology (cf. Section

tools (context-sensitive knowledge retrieval). 3.2); the prefixed question mark indicates that the class is used as a query variable.
328 S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342

Fig. 4. Query for P&IDs containing information about reactor equipment for polyamide-6 production.

chemical substances are made available by the PDW in form of 4.1.2. Step 1b: Context-sensitive retrieval of experience
instances, their roles in this context (i.e., as reactants or product) knowledge
are defined via ontological relations. The manual construction of queries is highly flexible, but
The query now reads: “Find all chemical reactors within as it is a rather complex task it might discourage non-expert
which a reaction takes place that has caprolactam and water users. Therefore, a more convenient way of knowledge retrieval
as reactants and polyamide-6 as reaction product”. It should be is offered by the PDW: Queries are derived from the current work
noted that, since the semantics of all query terms are formally context of the application tools, thus providing reusable product
defined within the PDW ontologies, the computer can inter- or process knowledge that matches the current situation. To this
pret the meaning of the query and is therefore able to retrieve end, the PDW needs to be integrated with the (legacy) applica-
the appropriate knowledge items, even if they are represented tion tools used by designers. To demonstrate the feasibility of
through different (but semantically equivalent) ontological con- this approach, a prototypical experience reuse framework has
cepts within the PDW. This retrieval mechanism, which is called been realized, consisting of the Process Data Warehouse, the
“semantic searching”, will be further explained in Section 4.2. process-integrated development environment PRIME (Pohl et
The query is passed to the PDW to search for matching experi- al., 1999) and a proprietary Flowsheet Editor (Bayer, Marquardt,
ence knowledge. Several knowledge items (ontology instances) Weidenhaupt, & Jarke, 2001), which is an advanced prototypical
are found that match the query; all of them can be recognized design environment for process flowsheets.
as instances of a particular class, VK Tube, a specialization of Assume that the engineer needs to perform simulation exper-
ChemicalReactor representing a special type of column reactor iments to determine certain design parameters of the VK tube
for polyamide-6 production (e.g., Agrawal, Devika, & Manabe, reactor. To this aim, he wants to query the PDW for mathemat-
2001). The engineer concludes from these facts that a VK tube ical models that can be reused to simulate the physicochemical
reactor is an appropriate choice for this chemical process. behavior of the VK tube. Instead of composing such a query
To find out more about the construction of a VK tube, the manually, as it was done in Step 1a, the query can be derived
engineer decides to search for associated documents. For that from the work context of the Flowsheet Editor: An initial Pro-
purpose, the above query is extended by concepts from the Core cess Flow Diagram has been created in the Flowsheet Editor,
Ontology’s Product Area. First, ChemicalReactor is linked with reflecting the main process steps and interconnecting material
the query variable DocumentVersion via a describedBy relation, streams of the polyamide-6 process (Fig. 5, item 1). The search
as shown in Fig. 4. Next, the query is further refined by speci- for an appropriate mathematical model is triggered by clicking
fying the desired document type: to this end, DocumentVersion on a menu item “find simulation model” and selecting the sec-
is linked with the class P&ID, a subclass of Document which tion of the flow diagram that is to be simulated (Fig. 5, item 2).
is introduced in the ontology extension Chemical Engineering Again, the query is composed of concepts from several areas of
Documents (cf. Section 3.2) to represent Process & Instrumen- the Core Ontology:
tation Diagrams. By choosing the relation currentVersion for the
linkage of DocumentVersion and P&ID, only the latest version • The product part of the query defines the type of document
of the P&ID document will be retrieved. (i.e., a model file) to be retrieved as well as optional further
Now the query reads: “Find the latest version of any Pro- constraints such as the desired format (e.g., an Aspen Plus
cess & Instrumentation Diagram that describes a reactor for file).
polyamide-6 production from caprolactam and water”. After • The descriptive part of the query specifies the content of the
processing the query, several P&ID documents are retrieved. model by using concepts from OntoCAPE. In this particular
From these, the engineer learns about the specific details of case, the specification comprises the type of model (steady-
VK tube design. He now possesses enough information to start state, rigorous), the type of the modeled reactor (VK tube),
working on his design task. and the material streams flowing into and out of the reactor.
S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342 329

Fig. 5. The application scenario of chemical plant design with PDW support.

• The most important process element is the user’s intention detailed design calculations and was therefore validated against
in this situation, i.e., “find simulation model”, given by the experimental data. Thus, in accordance with the present task,
menu item. This also determines the desired document type, the engineer decides to reuse the latter model.
a model file (see above). The model is then retrieved from the PDW and adapted
according to the specifications of the given design task. Sim-
This situation definition can then be passed on to the Process ulation experiments are performed, from which the engineer
Data Warehouse to search for matching experience information derives design parameters such as temperature, flow rates, or
(Fig. 5, item 3). dimensions. Based on these parameters, the reactor equipment
(apparatus, heat transfer equipment, etc.) is specified. Option-
4.1.3. Step 2: Review and reuse of retrieved experience ally, the rationale underlying this specification process can be
knowledge documented using the concepts of the decision documentation
In the example scenario, several different mathematical mod- module. For instance, assume that the engineer is notified about
els for the simulation of a polyamide-6 reactor are found and a still functional VK tube available from a decommissioned
returned. The models are combinations of different standard plant. The capacity of this tube is not sufficient for the new
model blocks, such as flash units, Plug Flow Reactors (PFR), or plant, but it can be used to realize an alternative plant design
Continuous Stirred Tank Reactors (CSTR), which are intercon- comprising two VK tubes in parallel. In contrast to the orig-
nected via process streams. This information is then presented to inal design, the alternative would require the acquisition of
the user via the front-end of the PDW (Fig. 5, item 4). Two dif- a considerably smaller and cheaper VK tube, which finally
ferent visualizations, as described in Section 3.3, can be applied makes the engineer opt for the alternative design. This deci-
here: The generic representation shows the class instances, their sion is likely to become inscrutable when the knowledge about
attributes and relations in UML instance notation, while the spe- the already existing VK tube becomes lost. Thus, the engi-
cific representation in this case shows a graphical snapshot of neer explicitly documents the underlying constraints to make
the reactor models (see Fig. 6). sure that his decision in this particular case will not affect
Now the expert needs to decide which of the simulation other design projects when knowledge from this project may be
models to reuse. The PDW supports this decision-making reused.
by providing additional information about the models in the
Description Area, such as the reaction kinetics and material 4.1.4. Step 3: Knowledge capture
properties data used in the model. Moreover, by browsing Up to now, the scenario process comprises specifying a reac-
associated concepts from the Process Area (refinements of Pro- tor design, simulating its behavior, analyzing the results, and
cessTrace) and related decision documentation, the user may finally the decision for a specific design. This process is traced,
find out more about the former usage of the models. For exam- and some design knowledge is extracted, recorded, and stored in
ple, he may learn that a particular model was developed for the PDW in order to augment the knowledge base. Apart from
an initial feasibility study (and is therefore likely to be rather the product information (the adapted simulation model as well as
simple and inaccurate), while another model was employed for the design specification in the Flowsheet Editor), the task-related
330 S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342

Fig. 6. The generic (left) and the specific (right) visualizations of the search results.

working steps and situations are traced (semi-)automatically. DecisionProblem, two instances of the class Alternative have
The technical realization of this knowledge capture is discussed been evaluated with respect to the sub-goal EconomyGoal (only
in Section 5. the Evaluation of the TwoTubesAlternative is shown in the fig-
Fig. 7 illustrates how (part of) the application scenario is ure), taking into account the existence of an old VK tube.
captured within the PDW. Grey boxes represent instances of Based on these Evaluations, the final Decision is in favor of
classes, which are defined in the peripheral ontology mod- the TwoTubesAlternative.
ules. For the sake of illustration, the representation has been In the Descriptive Area, the ProductObjects are described by
simplified. domain-specific concepts: ModelFile v1.0 is characterized as a
The Product Area represents the ProductObjects that are MathematicalModel that models the behaviour of a VK Tube;
created in the scenario. ReactorBlock is an instance of Prod- VK Tube, in turn, describes both the ReactorBlock and the
uct representing the VK reactor within the Flowsheet Editor; SingleTubeAlternative. MathematicalModel and VK Tube are
it is contained in the ProcessFlowDiagram v0.7, which is further specified by the indication of the ChemicalReaction,
an instance of the DocumentVersion class. ModelFile v1.0 is which has Caprolactam and Water as reactants and Polyamide-6
another instance of DocumentVersion representing the math- as reaction product.
ematical model used in the final calculations for the original In the Storage Area, all file versions of the Aspen Mod-
design featuring a single tube. ModelFile v1.0 and Process- elFile are storedAt the location “<project>/AspenFile#1” (a
FlowDiagram v0.7 are the current versions of ModelFile and StoragePlace) inside the DocumentManagementSystem “Doc-
ProcessFlowDiagram, respectively, which are instances of the umentum”. The Flowsheet Editor uses a database storage of
Document class. The upper part of the Product Area shows its own that can be accessed via certain programming inter-
instances of the decision documentation module representing faces. Therefore, this store has been integrated via this interface
the rationale to use a design with two VK tubes. For the (ToolInterface), using parameterized StorageQueries to get cur-
NumberOfVK TubesProblem, which is an instance of the class rent and historical flowsheet data.
S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342 331

Fig. 7. Representation of the application scenario within the PDW.

The Process Area describes the usage and/or manipulation of edge (i.e., documents and data) is discussed, while Section 4.2.2
the Products. The tasks that the engineer performs in the scenario addresses the reuse of process knowledge, including decisions.
are represented by instances of ProcessTrace (e.g., Modeling to The support of intra-organizational and cross-organizational
denote the creation of the reactor model, and Deciding for the cooperation is addressed in Sections 4.2.3 and 4.2.4, respec-
final decision to use the two-tube design). Appropriate instances tively.
of the classes User and Tool describe the human and computer
agents involved in the work process. The ProductObjects pro- 4.2.1. Reuse of product knowledge
duced during the engineering activities are associated with the Retrieval of product knowledge (as well as process knowl-
ProcessTraces via the relation createdBy, a specialization of the edge) is supported by three major retrieval mechanisms: (1)
manipulatedBy relation introduced in Fig. 1. semantic search, (2) browsing ontological categories, and (3)
navigating through a semantic network of interconnected, the-
4.2. Additional use cases matically related resources. All these retrieval mechanisms rely
on the description of the resources through meta information. As
This section summarizes the different usage possibilities of explained before, this meta information is represented by instan-
the PDW, some of which have already been mentioned in the tiating the classes and relations specified in the Core Ontology
section above. In Section 4.2.1, the reuse of product knowl- and their various extensions. The PDW does not only support
332 S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342

conventional, but also content-specific meta information to char- information since this enables the precise description of
acterize the contents of the different documents and data stores. knowledge items and therefore increases the probability of
The terminology for the content descriptions is provided by the relevant retrieval. However, such a specialized vocabulary
ontology modules that extend the Descriptive Area of the Core restricts the access to the knowledge for a novice (Hinds,
Ontology, particularly by the domain ontology OntoCAPE. 1999) as well as for a specialist from another area (Dougherty,
In the following, the usage of the different retrieval mecha- 1992). The PDW supports users of both groups: while domain
nisms offered by the PDW will be discussed in detail. experts will prefer a specialized terminology to annotate and
query the repository, non-specialists may use more generic
4.2.1.1. Semantic search. As explained before, semantic search terms for information retrieval. The special terms are derived
retrieves information based on the types of the information from the more general ones, and these specialization rela-
items and the relations between them, instead of using sim- tions are automatically inferred. Therefore, the knowledge
ple string comparisons. Queries to the PDW can be composed resources will be found either way. For example, if an expert
from arbitrary concepts (classes, instances, relations, attributes) chooses the term “Reactor” to annotate a resource, it can actu-
of the PDW ontologies. A major advantage over conventional ally be retrieved by the generic query term “Vessel” since
keyword search is the possibility to connect query terms via Reactor is defined as a subclass of Vessel in the ontology.
ontological relations instead of using Boolean operators. In
conventional information retrieval, Boolean operators produce 4.2.1.2. Browsing by categories. A second way to retrieve
ambiguous queries. Consider, for example, a query for a reac- product knowledge from the PDW can be achieved by brows-
tor that produces Caprolactam. If the logical operator AND ing the available meta information. To this aim, the ontology
was used to combine the query terms “Reactor” and “Capro- modules that extend the Descriptive Area of the Core Ontology
lactam”, the query could refer to a reactor for the production provide categories to structure the meta information.
of Caprolactam as well as to a reactor that takes Caprolac-
tam as a feed. The ambiguity can be resolved by replacing the • Some Products are explicitly categorized by means of the Cat-
generic AND by the more specific relation “hasOutput” (which egory concept. A well-known example for this type of retrieval
produces the composite query “Reactor hasOutput Caprolac- mechanism is the Yahoo! Directory, http://dir.yahoo.com/.
tam”). That way, the intended meaning of the query is clearly Classification under more than one category is of course pos-
defined. sible.
The queries are processed by the PDW’s built-in semantic • The ContentDescriptions of Products can be searched in a
search engine which tries to match the query expression against similar way. As explained, ContentDescriptions are given by
the meta information describing the various knowledge items. the OntoCAPE ontology. OntoCAPE has a concise structure
Since the semantics of the ontological concepts are formally which can be browsed easily: the ontology is thematically par-
defined, the search engine can interpret the meaning of both titioned into sub-ontologies (called “partial models”) which
query and meta information in order to find a semantic match. already constitute a coarse categorization of the domain. The
This approach has the following advantages: partial models are consistently structured: on the entrance
level, the top-level classes are organized according to the
principles of general systems theory (e.g., van Gigch, 1991);
• The search engine has the ability to skip homonyms4 during a
the classes are then refined on the lower levels, thus creating
search in order to avoid some of the unwanted results produced
taxonomies of specialized subcategories. To find informa-
by conventional search engines. For instance, consider search-
tion about a specific topic, one firstly selects the appropriate
ing for information about the reaction of a certain substance.
partial model and then browses its class hierarchy for meta
A conventional search using the keyword “reaction” would
information that describe the respective Products.
yield not only the desired information but also information
about allergic reactions to this substance. The semantic search
4.2.1.3. Navigating through a semantic network. Another pos-
engine, on the other hand, can distinguish between a chemical
sibility to explore the PDW is given by the ontological relations
reaction and an allergic reaction since their respective mean-
that interlink the individual ContentDescriptions. They form a
ings are formally defined in the PDW ontologies. Thus, as
directed graph, consisting of nodes which represent instances of
only information with the desired meaning is retrieved, the
ContentDescription and arcs which represent semantic relations
precision5 of the search is largely improved.
between these ContentDescriptions. Such a semantic network
• Based on the inference functionality offered by the ontology
(e.g., Sowa, 1991) of related meta information can be navigated
language, the semantic search engine takes the specialization
by means of custom tools like the PDW front-end.
relations between ontological classes into account. This helps
Semantic relations cannot only be used to connect Content-
to resolve the following dilemma of knowledge retrieval:
Descriptions describing similar contents but also to interrelate
domain experts prefer to use domain terminology as meta
ContentDescriptions with complementary contents, thus point-
ing to additional product knowledge that would otherwise
4 Homonyms are terms with the same spelling but different meaning. remain undetected. For instance, the documents ModelFile
5 Precision is defined as the proportion of relevant information out of all the and ProcessFlowDiagram created in the application scenario
retrieved information. (cf. Figs. 5 and 7), provide complementary information about
S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342 333

the chemical reactor. Therefore, their ContentDescriptions are the rationale is only recognizable when looking at the process
linked via the modelsBehaviorOf relation. A user may navigate history. Together with other aspects of the Process Area, they
from the simulation model of the reactor to the flowsheet object document the organizational context in which a certain project
describing its realization, and vice versa. Such navigation is use- has been or is being conducted.
ful in daily work when questions arise like: “Where can I find Beyond this documentation of organizational context, poten-
the simulation model that was used to design this apparatus!”. tial applications for reusing this process knowledge are (1)
Furthermore, ontological relations may be used to link Pro- process advice, and (2) the automatic execution of processes.
ductObjects to concepts from the other areas of the Core
Ontology in order to provide some additional information about (1) One can systematically search for a work context similar
the respective products: to the current situation to find some procedural advice that
is applicable. Moreover, by analyzing the ProcessTraces of
• Via relations, it is possible to point from a document to its already accomplished projects, the user can find out who
author or other potential knowledge holders that might be already solved a certain type of problem, and he can con-
relevant in this context. tact this expert directly to draw on the person’s implicit
• By connecting ProductObjects to ProcessObjects, it becomes knowledge.
possible to provide information about the organizational con- (2) This also offers the possibility of analyzing the data to
text (i.e., the work processes and decision-making procedures) detect recurring fragments. Fine-grained process support
in which a product was created, used, or modified. This allows can then be offered to the user by guiding him or her through
to answer questions such as “What has this model file been uti- these fragments, or even enacting those traces in the process
lized for?” or “Which decisions have been taken on the basis engine of the process integrated environment (see PRIME
of this data?”. That is, by analyzing the organizational context in Section 5.5).
associated with some design knowledge, one can evaluate the
applicability of the retrieved knowledge to the current work 4.2.3. Support of intra-organizational cooperation
situation. As pointed out by Carlile and Rebentisch (2003), The members of design teams need to communicate effi-
the circumstances in which the knowledge has been produced ciently to achieve the common goal of completing a project.
are highly relevant for its successful reuse: if the context of Typically, the team members are distributed across disparate
the information item changes between storing and reusing locations and use different software tools to create and manip-
this knowledge, new requirements or novel conditions arise. ulate product information. The PDW guides and drives the
In this case, the stored knowledge is no longer relevant and transfer of development product information between team
can even become harmful when it is retrieved for reuse and members and across software systems. The participating experts
applied under wrong conditions. directly access the repository of the PDW, which contains the
product and process traces of the project, thus reflecting its
Concluding the discussion about the reuse of product knowl- current status. Access can be achieved by extending the com-
edge, we would like to emphasize the PDW’s ability to centralize mon design tools of the domain with additional, integrated
access to product information. Such a central point of access functionality that accesses the PDW’s repository. The PDW par-
greatly facilitates the retrieval of product knowledge. The PDW ticularly improves the communication in interdisciplinary teams
allows to simultaneously search the contents of all kinds of doc- where the team members have heterogeneous professional back-
uments and data stores, regardless of their respective formats or grounds, as the ontologies provide access at different levels of
storage locations, which reduces the search effort significantly. expert knowledge (cf. discussion in Section 4.2.1).
Ease of use is further improved by the fact that the individual In later stages of the plant lifecycle, the PDW can be used as
sources can be explored via a single, familiar user interface. an information base to support tasks like change management,
Also, the PDW contains relationships between elements of dif- plant expansion, or claims management.
ferent sources, which allows to find information based on these
relations that are not contained in any of the data sources them- 4.2.4. Support of cross-organizational cooperation
selves. Moreover, the PDW allows to search information sources Information often needs to be passed on to a coopera-
that could not be accessed otherwise, as the user might not tion partner, based on strict rules of intellectual property and
have the required software or access rights to open the origi- need-to-know. Besides the problems of business agreements,
nal sources. And lastly, the search considers even sources of the communications, and the technologies used for information
existence of which the designer was not even aware. Thus, the transfer, it must be ensured that only specific information may
scope of knowledge retrieval is widened while the search effort be passed across organizational boundaries. Consequently, the
is reduced. cooperation partner may only be allowed to have access to a
restricted and well-controlled view of the repository.
4.2.2. Reuse of process knowledge To define such a view on the side of the contractee (i.e.,
The PDW contains coarse-grained and fine-grained process the delegating company), cooperative discussions among the
traces and decision-making procedures. The rationale leading to contractee’s experts are necessary. They need to agree which
a decision commonly spans multiple working steps, each with information items (ontology instances), relations between these
different products being manipulated. Hence, the full scope of items (ontology relationships), and which of their attributes may
334 S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342

be transferred to the external contractor (“released”). The dis- tion of processes and key decisions, as these do not count among
cussion results are recorded in the PDW, including the decisions the core activities of a project and are therefore not considered
taken with their arguments, and the data to be transferred. This in the project budget and schedule.
enables the reuse of these traces in later cooperations. In the While this lack of motivation can be counteracted by appro-
front-end of the PDW (see Section 3.3), the instances that corre- priate reward systems (e.g., Rus & Lindvall, 2002), a second
spond to the released information can be marked. Furthermore, problem is that knowledge capture is both haphazard and error-
view definitions may be abstracted from a concrete view to define prone: empirical studies (e.g., Furner, Ellis, & Willett, 1999;
which information items are to be usually released jointly. Furnas, Landauer, Gomez, & Dumais, 1987; Leonard, 1977)
The marked information can then be exported into an ontolog- have shown that little agreement exists among different per-
ical format (e.g., OWL, Bechhofer et al., 2004) and transferred sons on the sets of annotation terms to be assigned to individual
to the contractor. This ontological information is defined by resources. Thus, meta information about products is chosen
a precise semantic structure; thus, the contractor can draw on inconsistently and, similarly, work processes and decision-
computer support (e.g., transformations or matching rules) to making procedures are captured by different users in different
integrate the information into his own information systems. ways (e.g., using dissimilar structuring mechanisms on different
While the task is being solved at the contractor’s site, infor- levels of detail). Consequently, knowledge retrieval is severely
mation may be recognized to be missing. In this case, it has to be limited, as only part of the available knowledge items would be
inquired from the contractee. The PDW allows finding this data properly integrated into the ontology structure of the PDW and
and, possibly, additionally retrieving the decision that led to its thus could be found by some query.
exclusion. This decision may now either be revised (allowing A solution to these problems is to automate knowledge acqui-
the data to be released) or confirmed. sition as much as possible. Depending on the nature of the
As a last part of the cooperation process, the information knowledge, different techniques can be applied, some of which
returned from the contractor needs to be reviewed, discussed, will be described in the following sections.
and then integrated into the project flow through the PDW’s
repository. Similarly, the PDW can support the contractor, 5.2. Connecting external data sources
instead of the contractee as described above. More information
about the cross-organizational support functionality of the PDW The recording of product information is normally realized
can be found in Brandt, Fritzen, Jarke, and List (2007, chap. 4.1) inside the different software tools of a development environment.
and Brandt, Schlüter, and Jarke (2006b). Thus, an important task of the PDW is to transfer information
from the external data sources into its repository. This concerns
5. Knowledge acquisition document-based tools, application tools that use a database store
themselves (such as CAE systems), external databases (such as
This section describes how information, and thus knowledge, material databases), and complete software systems that provide
from existing sources can be acquired by the PDW. Section 5.1 certain data or programming interfaces (such as ERP systems).
discusses the problems of knowledge acquisition in general. The architectural concepts of integrating these data sources have
Section 5.2 gives an overview on how the PDW handles the already been introduced in Section 3.3. Here, the underlying
integration of external data sources. The subsequent sections mechanisms for connecting external data sources and converting
provide examples of these integrations. their contents to and from the information storage of the PDW
will be described in more detail.
5.1. General problems of knowledge acquisition For this goal, a flexible and extensible mechanism had to
be developed for integrating all these different kinds of data
Knowledge acquisition – that is, the process of capturing sources. External stores and their elements can be represented
and storing knowledge in an appropriate format in a knowledge through ontological concepts from the Storage Area, particularly
base – is recognized as a major bottleneck when it comes to the Store and StoragePlace. For accessing the external stores and
deployment of a knowledge management system in an organi- converting their data into the PDW format, “connectors” and
zation (e.g., Abecker, Bernardi, Hinkelmann, Kiihn, & Sintek, “transformers” have been designed which will be described in
1998; Carlile & Rebentisch, 2003; Thomson, Adams, Cowley, the following. The two cases of file-based storage and external
& Walker, 2003; Zeng, Pan, Zheng, & Peng, 2006). In indus- tool interfaces will be distinguished.
trial practice, documentation and archival of design knowledge
is often neglected or not performed at all. The reasons for this 5.2.1. Importing documents
are two-fold. When working with documents or files, those are usually
Employees are reluctant to invest much time in sharing their stored inside a Document Management System, which auto-
knowledge since this requires extra effort on top of their daily matically offers cooperative features like shared folders, file
workload, from which they do not profit personally. Even worse, locking, access control, and workflow management. The Doc-
by making their knowledge explicit, employees could weaken umentum (EMC2 , 2006) system and the Jakarta Slide system
their position as indispensable knowledge holders in the orga- (Slide, 2006), which is based on the WebDAV standard (Web
nization, and thus facilitate their replacement. Likewise, project Distributed Authoring and Versioning, RFC, 1999), have been
leaders have little interest in enforcing a thorough documenta- integrated with the PDW for this purpose.
S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342 335

When a new version of such a document is checked into the based methodical guidance for developers. To this end, the
DMS, the PDW is notified of this change and can then locate the PRIME process-integrated modeling environment closely inter-
corresponding ontology instances for the Store, the place where weaves with the PDW (cf. Section 5.5).
the document is stored (StoragePlace), and the logical document This kind of external source integration described so far is
concept (the Document) with its various versions (DocumentVer- usually change- or source-driven: that is, when a user changes
sion). A connector which needs to be specifically developed for a document with the corresponding tool, the new information is
each kind of source document, is then responsible for reading the stored in the PDW. This stands in contrast to the query-driven
information from the external file into an intermediate, XML- integration as described in the following.
based format. This format is usually specific for each application
the document stems from. Sometimes, the information from sev- 5.2.2. Integrating databases and tools
eral tools may be treated the same way, or needs to be integrated Often, information needed for the PDW is stored by an
to construct a comprehensive intermediate document (see Sec- external application in some repository or database format that
tion 5.3 for an example). In most cases, only part of the external cannot be accessed directly, but only through the programming
file needs to be read, as some specific details or fine-grained interfaces offered by the external tool. Sometimes, information
aspects are often not necessary for the “overview” aspect of the from such a tool is required in the PDW, e.g., when browsing
PDW. through the experience base and navigating from an instance
This intermediate XML is then transformed into an input that is related to an external storage representation. In this case,
format that directly represents instances of the PDW’s con- a connector as described above, is responsible for retrieving
cepts. The appropriate Transformation instance is looked up in the required data and writing it into the intermediate XML
the PDW, and the corresponding transformer is activated. RDF format. Again, this XML data is transformed by the appropri-
Schema (W3C, 2004) is used here as output format from the ate transformer and then imported into the PDW as instances,
transformers and thus, as input format to the PDW. This has attributes and relationships. Similar queries can be executed on
the advantage that, on the one hand, the KAON system which database back-ends, often by using SQL. The execution of these
forms the base of the PDW, can directly work with it; on the other queries can also be triggered by external events (source-driven,
hand, many cases allow to use powerful XML Schema Transfor- as above), or done regularly.
mations (XSLT; W3C, 1999) to convert the intermediate XML As an example, the ERP system SAP R/3 (SAP, 2006) has
into the XML-based RDF Schema representation. The Product been integrated into certain PDW-based workflows. This allows
instances corresponding to the entities defined in the input files, to retrieve the transport history of certain goods by calling the
and their related descriptive meta information, can be created appropriate SAP interface method, and use this information
from this data; furthermore, relations to or from existing identi- for information visualization and semantic queries in the PDW
fiable instances can be created or modified. It is also possible to (Kopecky, 2006). Usually, the relatively unstructured design pro-
find the differences between the existing and the new version of cesses are not supported by deterministic systems like ERP. The
the document, thus importing only the changes into the PDW. PDW, in contrast, allows to relate these creative design processes
Sometimes, such a transformation is not possible, as the on the one hand, with the administrative or business perspective
document is too weakly structured or the necessary imple- on the other hand. This can be achieved by integrating the pro-
mentation has not yet been realized. In such a case, it is also cess traces of the PDW with the project management sub-system
possible to annotate and/or enrich the Document and Docu- of the ERP system.
mentVersion instances of the PDW manually, for example, by
classifying them into categories or by providing other kinds of 5.3. Conversion of simulation models
content description. Several tools are available for this purpose:
existing markup tools like the OntoMat-Annotizer (Annotation Conversion of document formats with well-defined syntax
Portal, 2006) or SemanticWord (Teknowledge, 2006) support and semantics (databases, XML files, model files, etc.) can
the annotation of informal text documents with ontological con- automatically be performed by connectors that extract meta
cepts via simple drag and drop operations. The tool TRAMP information from the documents and transformers that map
(Jarke, Miatidis, Schlüter, & Brandt, 2006) has been especially it onto the ontologies of the PDW. For demonstration pur-
developed for structuring and annotating multimedia contents poses, a prototypical converter has been developed, which
resulting from three-dimensional simulations in plastics engi- can automatically create content-specific meta information for
neering. Also, the organizational aspect (especially the current model files of the process simulator Aspen Plus (AspenTech,
user and workflow activities) can be recorded in relation to the 2006).
imported information: traces of coarse-grained work processes The converter follows a step-wise approach: in the first step
can be captured by the Workflow Modeling System WOMS (cf. of this conversion process, a mathematical model represented in
Section 5.4). For supplementary information like decisions taken the Aspen Plus input format (.inp), is converted to a so-called
and their arguments, specialized tools (e.g., a decision editor) are raw XML file, which gives a more structured representation of
being developed. (selected portions of) the Aspen Plus model. In the next step,
Additionally, environments for fine-grained process support the raw XML is transformed into CapeML, an XML-based
at the engineering workplaces can store traced experiences in the model exchange language specifically developed for the CAPE
PDW and reuse them when demanded, by providing experience- domain (von Wedel, 2002). This second conversion step com-
336 S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342

Fig. 8. Step-wise conversion of Aspen Plus files into an OWL file.

prises a further structural as well as a semantic enrichment to XML, the further conversion into OWL is realized by using
of the raw XML data. Additionally, material property infor- the XML2OWL converter described in Section 5.3; the OWL
mation derived from an Aspen Plus report file (.rep) may be representation can then be imported by the PDW. Depending on
included in the CapeML file. In the final step, the CapeML the type of the work process model (prescriptive or descriptive),
file is converted to an OWL file; this transformation comprises the OWL representation is based either on concepts defined in
both a format conversion and a further semantic enrichment. the extension module Work Process & Project Management or
The resulting file can then be imported by the PDW (simi- on concepts defined in Design Traces (cf. Section 3).
lar to the import based on RDF Schema described in Section The difficulties in knowledge acquisition described in the
5.2). beginning of Section 5 are in particular significant for capturing
A graphical illustration of the conversion chain is shown and documenting the rationale of design decisions. Document-
in Fig. 8: the Aspen2CapeML Converter integrates the first ing design rationale is often seen as an unacceptable overhead
two conversion steps, which are indicated by the blocks Aspen by designers (e.g., Lee, 1997), while the beneficiaries of the doc-
Inp(ut) Handler and Report File Handler. The final conversion umentation are typically to be found later in the design process
is performed by the XML2OWL Converter (Amin & Morbach, or even later in the life cycle of the artifact (Grudin, 1996), for
2007). example, on the occasion of a retrofit of an existing plant. A fur-
The converters implementing the different steps of the con- ther obstacle is the reluctance to explicitly document discarded
version chain may also be used independently. In particular, the alternatives as they might be conceived as an indication of not
XML2OWL converter is not limited to handling CapeML files; working in an optimal way. In order to support designers as
rather, it is capable of mapping arbitrary XML data onto any far as possible during the creation of decision documentation,
existing OWL ontology. The conversion from XML to OWL is an extended version of WOMS is under development, which
controlled by so-called mapping rules specified in an external will support the integrated documentation of work processes
rules definition file. These rules can be formulated easily and and decisions.
without deep knowledge of the converter’s program logic. Thus,
the converter can be easily configured to a number of differ-
ent mapping problems, such as the WOMS to OWL conversion 5.5. Process-integrated tracing and guidance
described in the next section.
The considerable heterogeneity of the software tools
supporting various design tasks greatly complicates their inter-
5.4. Process and decision documentation operability as well as the exchange of data (cf. Section 2). As
a remedy to this problem, integrated tool environments consti-
The Workflow Modeling System WOMS was created for tute an ambitious endeavor for integrating individual tools into
modeling design processes on a coarse-grained level (cf. Hai, a coherent whole, whose interplay is much easier to coordinate.
Theißen, Schneider, Morbach, & Marquardt, 2007, chap. 5.1). Typically, such environments employ application and process
The tool, based on Microsoft Visio, supports a variant of the integration techniques in order to orchestrate the services pro-
C3 modeling language. Its graphical user interface (see Fig. 9) vided by disparate legacy systems using some sort of middleware
provides the C3 modeling elements which can easily be dragged and agreed-upon protocol.
into the drawing area. Several macros support the easy and intu- In the context of the IMPROVE project, we aimed at devel-
itive creation of work process models. WOMS is suitable for oping such an approach in order to integrate the heterogeneous
both prescriptive models (e.g., guidelines or best practices for tools employed during the reference polyamide-6 design pro-
designers) and descriptive models of executed design processes cess, and provide method guidance to the designer based on
(i.e., design traces). As the tool provides an export functionality previously traced experiences. To this end, we adapted the
S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342 337

Fig. 9. Workflow Modeling System WOMS.

Process-Integrated Modeling Environment (PRIME) approach can potentially exploit the following four extension modules of
to build an integrated design support environment for chemi- the Core Ontology:
cal engineering (Jarke, List, & Weidenhaupt, 1999; Pohl et al.,
1999; Weidenhaupt, 2001). PRIME is especially well suited for
• process models which are described by using fine-grained
the realization of tool environments for creative processes, like
concepts of the Work Process & Project Management module,
engineering design processes. It builds on the key idea of a pos-
can be enacted by process-integrated tools in order to provide
teriori process integration that offers the potential to flexibly
method guidance to the designer;
couple different tools. At the same time, it provides flexible,
• the Design Traces module can be used to build a concrete
situated method guidance to the designer directly from inside
traceability structure, according to which the design history
his tools. More specifically, PRIME is able to record all the
inside process-integrated tools is recorded;
actions executed inside a process-integrated tool, as well as the
• inside PRIME, process-integrated tools are modeled with
products worked upon by these actions. The recorded infor-
respect to their services and user interface capabilities. Thus,
mation is structured according to a concrete traceability meta
the application tools module can be employed for their
model abstracted from a number of case studies (Ramesh &
description;
Jarke, 2001). On the other hand, recorded process or product
• lastly, descriptions of the products manipulated by process-
experiences from the past can be reused, on request, inside
integrated tools, can be represented using elements from the
process-integrated tools. Then, the tool is able to automatically
Chemical Engineering Documents module.
execute the steps comprising a requested method fragment, or
directly load products for reuse support, corresponding to the
current design situation (Miatidis, Jarke, & Weidenhaupt, 2007, 6. Related work
chap. 3.1).
The overall PRIME-IMPROVE environment has been imple- As already discussed in Section 2, most of the existing PDM
mented as a flowsheet-centered architecture that operationally and PLM systems still lack essential aspects needed for sup-
couples the fully process-integrated Flowsheet Editor (cf. Sec- porting phases of conceptual design, such as capabilities for
tion 4.1) with other domains-specific simulation tools (Bayer et knowledge management (Gao et al., 2003), or flexible data struc-
al., 2001). It has a flexible repository layer that can be fairly tures that allow runtime extensions and adaptations (Marquardt
easily adapted in order to exploit the PDW infrastructure for the & Nagl, 2004). Some recent research projects try to extend
persistent storage of information manipulated or generated by the capabilities of PDM/PLM systems. Two examples from the
process-integrated tools. More specifically, the PRIME approach domain of automotive engineering are given in the following.
338 S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342

Kim, Kang, Lee, and Yoo (2001) have integrated concepts mation sources, their structure, access permissions, and format
of artificial intelligence into commercial PDM systems. The properties; Domain Ontologies that describe the content of the
software is based on a dynamic and flexible workflow model, information sources; and Enterprise Ontologies that model the
as opposed to the deterministic workflows seen in most com- organizational context, such as the work processes or the struc-
mercial PDM applications. A built-in workflow management ture of an enterprise. Thus, the Information Ontologies cover
system enables the integrated management of task processes the scope of both the Product Area and Storage Area of the
and information flows. The system can be searched by means Core Ontology, the scope of the Domain Ontologies correspond
of a semantic query and manipulation language. However, the to that of the Descriptive Area, and the Enterprise Ontologies
system relies on prescriptive process definitions which require resemble the Process Area. The paper gives a good overview on
relatively well-understood and well-documented processes. As the process of knowledge management itself and the concepts
this does not apply to conceptual process engineering, this of ontological support, from the perspective of computer and
approach cannot be used here. information science. The application of these concepts to engi-
Gao et al. (2003) describe an integration of a PDM system neering domains does not fall into the scope of this particular
with ontological methods and tools. The Protégé ontology edi- research group.
tor (Stanford Medical Informatics, 2006) is combined with a In the area of chemical engineering, several prototypical
commercial PDM system to provide knowledge management design environments have been developed based on ontological
capabilities for the conceptual design stage. In an associated models and tools. Unlike the PDW, these environments focus
research project, a graphical interfaces for user-friendly knowl- on other aspects than knowledge management and experience
edge acquisition has been developed. In contrast to the PDW reuse. Some recent examples are given here.
described here, knowledge needs to be entered manually and Bañares-Alcántara and Lababidi (1995) have developed
explicitly, based on ontology instances and static rules; expe- KBDS, a prototypical design support system for process
rience knowledge is not recorded. Again, this approach relies engineering. The implementation is based on CLOS, an object-
on a domain where the processes are better understood than in oriented extension of Common Lisp, and additionally makes
process engineering. use of an assumption-based truth maintenance system (ATMS).
Some research teams are currently developing design repos- KDBS is not a full-fledged KM system, but incorporates some of
itories based on semantic technologies. Two representative its features: it supports (1) the representation and management
examples from the domain of electromechanical engineering of different design alternatives (i.e., alternative flowsheets) and
are described below. their relations to mathematical models, (2) the documentation
Kopena and Regli (2003) designed an ontology for the rep- of the design objectives, (3) the evaluation of a design alter-
resentation of product knowledge. A Core Ontology defines native against a previously stated set of design objectives, and
the basic structure to describe products from a functional view. (4) the evolution of these aspects during a design project. The
Vocabulary extensions describe the domain of electromechani- representation of design rationale in KBDS (Bañares-Alcántara
cal devices. The model representation is based on a Description & King, 1996) is based on IBIS (Issue-Based Information Sys-
Logic (DL) system with low expressivity to achieve good com- tems, Kunz & Rittel, 1970), a graphical notation similar to DRL,
putability and scalability (cf. discussion in Section 7). but less expressive. KBDS supports decision-making in the con-
Szykman, Sriram, Bochenek, Racz, and Senfaute (2000) ceptual design stage and facilitates the cooperation between the
describe a design repository to enable storage and retrieval of members of a project team. Compared to the PDW, KBDS has
design artifact knowledge. It is based on a proprietary knowledge a narrower focus in so far as it covers only information about
base implementation that can be accessed via a web interface. flowsheet alternatives and design objectives.
All queries need to be developed manually and coded as spe- Batres and Naka (2000) and Batres, Naka, and Lu (1999)
cific algorithms. The system uses technologies that predate the have developed the prototypical design environment P-DOSS.
Semantic Web approach. Consequently, it tries to mimic the The major objective of the P-DOSS system is to enhance inter-
functionality of ontological models and semantic technologies operability between different application tools. Within P-DOSS,
without reaching their potential. information are described through the ontology MDF (Batres,
Both described design repositories are limited to the storage Aoyama, & Naka, 2001; Batres & Naka, 2000), which is repre-
of product data and documents. They do not offer support for sented in the Knowledge Interchange Format, KIF (KSL, 2006).
the recording and management of the associated work processes Comparable to OntoCAPE (cf. Section 3.2), the MDF ontology
and decision-making procedures. A review of other recent devel- describes products and processes in the chemical engineering
opments in the area of representation, capture and retrieval of domain, however with the focus on the later design stages (pri-
design knowledge is given also by Szykman et al. (2001). marily detail engineering and operations planning).
An ontological architecture for knowledge management that Kitamura and Mizoguchi (2003) describe a design envi-
resembles the Core Ontology described here has been pro- ronment called Functional Way Server which is based on the
posed by Abecker et al. (1998). Similar to our approach, the prototypical ontology editor Hozo (Kozaki, Kitamura, Ikeda,
proposed ontology-based system it is intended to enable both & Mizoguchi, 2000). In the conceptual design stage, the user
content-specific and task-specific retrieval of resources. The needs to specify the desired function of the engineering artifact
authors distinguish three types of interconnected ontologies: to be designed; the system then suggests alternative “functional
Information Ontologies that model the different kinds of infor- ways” (e.g., different physical principles) how to achieve the
S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342 339

artifact’s desired function. It also determines the manufacturing these frameworks can handle only a relatively small amount of
steps that are necessary for the realization of each functional way. instance data. Thus, to develop a system usable for real-world
The Functional Way Server relies on a knowledge base consist- application, it is necessary to use a back-end for database storage
ing of two major ontologies: the Functional Ontology, which that is directly able to interpret at least some inference rules and
provides a comprehensive representation of generic design semantic queries. The KAON system which the PDW uses has
knowledge, forms the core of the knowledge base. For the sup- been realized this way.
port of conceptual process engineering, the Functional Ontology On the other hand, the full expressive power of ontology
is extended by a domain-specific Plant Ontology (Mizoguchi, and description-logic (DL) based languages (Baader, Horrocks,
2001; Mizoguchi, Kozaki, Sano, & Kitamura, 2000). The knowl- & Sattler, 2004, chap. 1) would offer capabilities not avail-
edge base of the Functional Way Server must be developed a able in the current PDW. While the PDW is only able to infer
priori; unlike the PDW, there are no mechanisms to record and supertype and subtype relations, a full-fledged description logic
incorporate experience knowledge during project execution. system would be able to infer certain characteristics of classes
The Process Information Repository developed by Zhao, and instances, based on semantic constraints and rules. This
Hailemariam, et al. (2006), Zhao, Jain, et al. (2006), Zhao, goes beyond the capabilities of semantic search currently imple-
Joglekar, Jain, Venkatasubramanian, and Reklaitis (2005), and mented in the PDW, as can be seen from the following example
Venkatasubramanian et al. (2006) aims at the systematic reuse which extends the use case described in Section 4.2.1: if an inex-
of knowledge in the pharmaceutical industries. Like the PDW, perienced user enters information about a reactor, annotating it
the repository stores product knowledge (meta information as an instance of the generic class Vessel only, the current PDW
describing the contents of documents and databases) as well as will not return this instance when queried for a Reactor. In con-
process knowledge (deterministic guidelines) in form of ontolo- trast to that, a DL system might contain an inference rule “A
gies. However, the scope of representable (and thus retrievable) Reactor is a Vessel in which a ChemicalReaction takes place”.
knowledge is limited, as the ontologies of the repository are The system could then conclude that the above Vessel is in fact a
specialized on describing particular aspects of pharmaceutical Reactor, and return it when queried for such. Thus, the recall of
products and processes, and are consequently not as generic and the search (i.e., the proportion of relevant information retrieved
broadly applicable as those of the PDW. out of all the relevant information in the collection) may be
Moreover, the system does allow to perform inference and significantly enhanced.
search operations directly on the repository, but requires load- As we did not find a way to combine scalability and full
ing all data into memory first. This is not a practicable option logic expressiveness, we decided to focus on the former for the
due to scalability and performance problems (cf. discussion in development of the PDW. This decision will have to be reviewed
Section 7). A further shortcoming is that the crucial point of once efficient and scalable DL systems have become available.
automated knowledge capture (cf. discussion in Section 5) is not In addition to this technical aspect, we are planning to extend
addressed. the models and application cases of the PDW. Experience knowl-
edge from the full plant life cycle is going to be examined, from
7. Discussion and outlook the design phase through plant operation (production), modifi-
cations and redesign or reengineering. As a first step, we have
The Process Data Warehouse has been developed as an considered to extend the knowledge acquisition to the produc-
ontology-based repository for the support of design processes. tion stage. One potential benefit of this could be to incorporate
Its key features are: (1) flexible data structures and work process selected data from production processes into the PDW and relat-
models for the support of creative and dynamic work processes, ing it to the simulation models used to design the respective
(2) improved navigation and retrieval mechanisms based on production facility. This allows to compare the simulation mod-
the use of Semantic Web technologies (e.g., semantic search), els with the actual plant performance; thus, when reusing the
(3) integrated representation of resources, content descriptions, models, the model quality can be easily determined. Expe-
and organizational context to enable experience reuse, and (4) riences in this direction have been gained by recording and
functionality for (semi-)automatic capture of experience knowl- supporting rubber extrusion processes (Raddatz, Schlüter, &
edge during the execution of design projects. These features Brandt, 2006), and by feeding these traces back into process
are achieved by means of a specific architecture: a central design.
Core Ontology which is flexibly refined by extension ontol- To remedy the difficulties in documenting design and deci-
ogy modules. The application of the PDW has been illustrated sion rationale mentioned in Section 5.4, an extended version of
by several use cases, including a major application scenario WOMS is under development, which supports the integrated
about the conceptual design of a polyamide-6 production facil- documentation of work processes and decisions. In particu-
ity. In the following, some of the shortfalls of our approach are lar, the tool will enable the reuse of fragments of generalized
described, together with possible counter-measures and some decision models (decision templates) as part of the decision
planned extensions. documentation in concrete projects. This approach will not only
A major problem of current ontology-based frameworks is reduce the effort to create decision documentation but also help
the trade-off between implementation efficiency and descriptive the designer to consider the relevant aspects in a particular case,
logical power. Most existing ontology frameworks need to load such as potential alternatives and their pros and cons under
the entire information into memory to work with it. As a result, certain constraints.
340 S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342

Acknowledgements Bayer Technology Services. (2006). Engineering software—PEDW. Available


at http://www.bayertechnology.com/eng/products/43 1488.php. Accessed
October 2006.
Most of the research described in this article has been per-
Bechhofer, S., Harmelen, F., van Hendler, J., Horrocks, I., McGuiness, D.,
formed in the context of the Collaborative Research Center on Patel-Schneider, P., et al. (2004). OWL Web Ontology Language Reference.
Cooperative Computer-Aided Process Engineering (CRC/SFB Available at http://www.w3.org/TR/owl-ref/. Accessed September 2006.
476 IMPROVE, Marquardt & Nagl, 2004), funded by the Berners-Lee, T., Hendler, J., & Lassila, O. (2001). The Semantic Web—A new
German Research Foundation (DFG). The authors would like form of Web content that is meaningful to computers will unleash a revolution
of new possibilities. Scientific American, 17.
to thank their student workers, especially M.A. Amin, M.
Bilgic, T., & Rock, D. (1997, September). Product data management systems:
Comanns, M.T. Ikram, J. Kopecky, A. Passen, and J. Renner, State-of-the-art and the future. In Proceedings of the 1997 ASME design
who implemented major parts of the software systems described. engineering technical conferences
Brandt, S. C., Fritzen, O., Jarke, M., & List, T. (2007). Goal-oriented information
flow management in design processes. In M. Nagl & W. Marquardt (Eds.),
References Collaborative and distributed chemical engineering design processes: From
understanding to substantial support. Berlin, Heidelberg: Springer-Verlag.
Abecker, A., Bernardi, A., Hinkelmann, K., Kühn, O., & Sintek, M. (1998). Brandt, S. C., Morbach, J., Miatidis, M., Theißen, M., Jarke, M., & Marquardt,
Toward a technology for organizational memories. IEEE Intelligent Systems, W. (2006). Ontology-based information management in design processes.
13(3), 40–48. In W. Marquardt & C. Pantelides (Eds.), 16th European symposium on com-
Agrawal, A. K., Devika, K., & Manabe, T. (2001). Simulation of hydrolytic puter aided process engineering and 9th international symposium on process
polymerization of nylon-6 in industrial reactors: Part I Mono-acid-stabilized systems engineering (pp. 2021–2026). Elsevier.
systems in VK tube reactors. Industrial & Engineering Chemistry Research, Brandt, S. C., Schlüter, M., & Jarke, M. (2006b). A process data warehouse for
40(12), 2563–2572. tracing and reuse of engineering design processes. International Journal of
Amin, M. A., & Morbach, J. (2007). XML to OWL converter. Technical report Intelligent Information Technologies, 2(4), 18–26.
(LPT-2007-02), Lehrstuhl für Prozesstechnik. RWTH Aachen University. Brandt, S. C., Schlüter, M., & Jarke, M. (2006c). Process data warehouse models
Annotation Portal. (2006). OntoMat Homepage. Available at http://annotation. for cooperative engineering processes. In Proceedings 9th IFAC symposium
semanticweb.org/ontomat/index.html. Accessed October 2006. on automated systems based on human skill and knowledge.
AspenTech. (2006). Homepage. Available at http://www.aspentech.com/. Braunschweig, B., Fraga, E., Guessoum, Z., Marquardt, W., Nadjemi, O., Paen,
Accessed October 2006. D., et al. (2004). CAPE web services: The COGents way. In A. Barbrosa-
AVEVA. (2006a). VANTAGE Plant Engineering. Available at http://www.aveva. Povoa & H. Matos (Eds.), European symposium on computer aided process
com/products/plant/vpe.html. Accessed October 2006. engineering, Vol. 14 (pp. 1021–1026). Elsevier.
AVEVA. (2006b). VANTAGE Enterprise Net. Available at http://www. Carlile, P. R., & Rebentisch, E. S. (2003). Into the black box: The knowledge
aveva.com/products/plant/vnet.html. Accessed October 2006. transformation cycle. Management Science, 49(9), 1180–1195.
Awad, E. M., & Ghaziri, H. M. (2003). Knowledge management. Prentice Hall. Dassault Systemes. (2006). PLM Solutions. Available at http://www.3ds.com/
Baader, F., Horrocks, I., & Sattler, U. (2004). Description logics. In S. Staab & R. products-solutions/plm-solutions/. Accessed October 2006.
Studer (Eds.), Handbook on ontologies. Berlin, Heidelberg: Springer-Verlag. Dougherty, D. (1992). Interpretive barriers to successful product innovation in
Banares-Alcantara, R., & King, J. M. P. (1996). Design support systems for large firms. Organization Science, 3, 179–202.
process engineering—III. Design rationale as a requirement for effective Eggersmann, M., Hackenberg, J., Marquardt, W., & Cameron, I. (2002). Applica-
support. Computers and Chemical Engineering, 21, 263–276. tions of modelling: A case study from process design. In B. Braunschweig &
Banares-Alcantara, R., & Lababidi, H. M. S. (1995). Design support systems R. Gani (Eds.), Software architectures and tools for computer aided process
for process engineering—II. KBDS: An experimental prototype. Computers engineering (pp. 335–372). Elsevier Science.
and Chemical Engineering, 19, 279–301. Eggersmann, M., Hai, R., Kausch, B., Luczak, H., Marquardt, W., Schlick, C.,
Batres, R., Aoyama, A., & Naka, Y. (2001). A life-cycle approach for model reuse et al. (2007). Work process models. In M. Nagl & W. Marquardt (Eds.),
and exchange. In R. Gani & S. B. Jørgensen (Eds.), European symposium Collaborative and distributed chemical engineering design processes: From
on computer aided process engineering, Vol. 11 (pp. 87–92). Elsevier. understanding to substantial support. Berlin, Heidelberg: Springer-Verlag.
Batres, R., & Naka, Y. (2000). Process plant ontologies based on a multi- EMC2 . (2006). Documentum. Available at http://software.emc.com/products/
dimensional framework. In M. F. Malone, J. A. Trainham, & B. Carnahan product family/documentum family.htm. Accessed September 2006.
(Eds.), Proceedings of the fifth international conference on foundations of Furnas, G. W., Landauer, T. K., Gomez, L. M., & Dumais, S. T. (1987). The
computer-aided process design, AIChE symposium series no. 323, Vol. 96 vocabulary problem in human-system communication. Communications of
(pp. 433–437). the ACM (CACM), 30, 964–971.
Batres, R., Naka, Y., & Lu, M. L. (1999). A multidimensional design framework Furner, J., Ellis, D., & Willett, P. (1999). Inter-linker consistency in the man-
and its implementation in an engineering design environment. Concurrent ual construction of hypertext documents. ACM Computing Survey (CSUR),
Engineering: Research and Applications, 7(1), 43–54. 31(4). Article No. 18
Bayer, B. (2003). Conceptual information modeling for computer aided sup- Gao, J. X., Aziz, F. L., Maropoulos, P. G., & Cheung, W. M. (2003). Appli-
port of chemical process design. Ph.D. thesis. Lehrstuhl fur Prozesstechnik, cation of product data management technologies for enterprise integration.
RWTH Aachen University. Published in: Fortschritt-Berichte VDI: Reihe 3, International Journal of Computer Integrated Manufacturing, 16(7–8), 491–
No. 787. Düsseldorf: VDI-Verlag. 500.
Bayer, B., Eggersmann, M., Gani, R., & Schneider, R. (2002). Case studies in Grudin, J. (1996). Evaluating opportunities for design capture. In T. P. Moran &
design and analysis. In B. Braunschweig & R. Gani (Eds.), Software archi- J. M. Carroll (Eds.), Design rationale: Concepts, techniques, and use (pp.
tectures and tools for computer aided process engineering (pp. 591–634). 453–470). Mehwah, New Jersey: Lawrence Erlbaum Associates, Inc.
Elsevier Science. Hai, R., Theißen, M., Schneider, R., Morbach, W., & Marquardt, W. (2007).
Bayer, B., & Marquardt, W. (2003). A comparison of data models in chemi- Scenario-based analysis of industrial work processes. In M. Nagl & W. Mar-
cal engineering. Concurrent Engineering Research and Applications, 11(2), quardt (Eds.), Collaborative and distributed chemical engineering design
129–138. processes: From understanding to substantial support. Berlin, Heidelberg:
Bayer, B., Marquardt, W., Weidenhaupt, K., & Jarke, M. (2001). A flowsheet Springer-Verlag.
centered architecture for conceptual design. In R. Gani & S. B. Jørgensen Hinds, P. J. (1999). The curse of expertise: The effects of expertise and debias-
(Eds.), European symposium on computer aided process engineering, Vol. ing methods on prediction of novice performance. Journal of Experimental
11 (pp. 345–350). Elsevier. Psychology: Applied, 5, 205–221.
S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342 341

Innotec. (2006). Homepage. Available at http://www.innotec.de/index.php?id= Collaborative and distributed chemical engineering design processes: From
14&L=l. Accessed September 2006. understanding to substantial support. Berlin, Heidelberg: Springer-Verlag.
Intergraph. (2006a). SmartPlant P&ID. Available at http://www. Mizoguchi, R. (2001). Ontological engineering: Foundation of the next gener-
intergraph.com/smartplant/pid/. Accessed October 2006. ation knowledge processing. In Proceedings of web intelligence: Research
Intergraph. (2006b). SmartPlant Foundation. Available at http://www. and development, Lecture notes in computer science, Vol. 2198 (pp. 44–57).
intergraph.com/smartplant/foundation/. Accessed October 2006. Springer.
Jarke, M., Lenzerini, M., Vassiliou, Y., & Vassiliadis, P. (2003). Fundamentals Mizoguchi, R., Kozaki, K., Sano, T., & Kitamura, Y. (2000). Construction and
of data warehouses (2nd ed.). Springer-Verlag. deployment of a plant ontology. In R. Dieng & O. Corby (Eds.), Proceedings
Jarke, M., List, T., & Köller, J. (2000). The challenge of process data ware- of the 12th European workshop on knowledge acquisition, modeling and
housing. In Proceedings of the 26th international conference on very large management, Lecture notes in computer science, Vol. 1937 (pp. 113–128).
databases—VLDB Springer.
Jarke, M., List, T., & Weidenhaupt, K. (1999). A process-integrated conceptual Morbach, J., Hai, R., Bayer, B., & Marquardt, M. (2007). Document models.
environment for chemical engineering. In Proceedings of the 18th interna- In M. Nagl & W. Marquardt (Eds.), Collaborative and distributed chemical
tional conference on conceptual modeling—ER. engineering design processes: From understanding to substantial support.
Jarke, M., & Marquardt, W. (1996). Design and evaluation of computer-aided Berlin, Heidelberg: Springer-Verlag.
process modeling tools. In J. F. Davis, G. Stephanopoulos, & V. Venkata- Morbach, J., Yang, A., & Marquardt, W. (2007). OntoCAPE—A large-scale
subramanian (Eds.), Intelligent systems in process engineering, AIChE ontology for chemical process engineering. Engineering Applications of
symposium series no. 312, Vol. 92 (pp. 97–109). Artificial Intelligence, 20(2), 147–161.
Jarke, M., Miatidis, M., Schlüter, M., & Brandt, S. C. (2006). Process inte- Nagl, M., Westfechtel, B., & Schneider, R. (2003). Tool support for the manage-
gration and media-assisted traceability in cross-organisational engineering. ment of design processes in chemical engineering. Computers and Chemical
International Journal of Business Process Integration and Management, Engineering, 27(2), 175–197.
1(2), 65–75. Oberle, D., Volz, R., Motik, B., & Staab, S. (2004). An extensible ontology soft-
Kim, Y., Kang, S., Lee, S., & Yoo, S. (2001). A distributed, open, intelligent prod- ware environment. In S. Staab & R. Studer (Eds.), Handbook on ontologies
uct data management system. International Journal of Computer Integrated (pp. 311–333). Springer.
Manufacturing, 14, 224–235. OMG. (2004). Unified Modeling Language (UML), version 2.0. Available at
Kitamura, Y., & Mizoguchi, R. (2003). Ontology-based description of functional http://www.uml.org/. Accessed October 2006.
design knowledge and its use in a functional way server. Expert Systems with Pohl, K., Weidenhaupt, K., Dömges, R., Haumer, P., Jarke, M., & Klamma, R.
Applications, 24(2), 153–166. (1999). PRIME: Towards process-integrated environments. ACM Transac-
Kopecky, J. (2006). Model-based integration of operating and ERP systems data. tions on Software Engineering and Methodology, 8(4), 343–410.
Master’s thesis. RWTH Aachen University. PostgreSQL. (2006). Homepage. Available at http://www.postgresql.org/.
Kopena, J., & Regli, W. C. (2003). Functional modeling of engineering designs Accessed October 2006.
for the semantic web. IEEE Data Engineering Bulletin, 26(4), 55–61. PTC. (2006a). Windchill. Available at http://www.ptc.com/appserver/
Kozaki, K., Kitamura, Y., Ikeda, M., & Mizoguchi, R. (2000). Development mkt/products/home.jsp?k=37. Accessed October 2006.
of an environment for building ontologies which is based on a fun- PTC. (2006b). CADDS 5i Optegra. Available at http://www.ptc.com/products/
damental consideration of “relationship” and “role”. In Proceedings of cadds/optegra.htm. Accessed October 2006.
the sixth pacific knowledge acquisition workshop (PKAW2000) (pp. 205– Raddatz, M., Schlüter, M., & Brandt, S. C. (2006). Identification and reuse
221). of experience knowledge in continuous production processes. In 9th IFAC
KSL, Stanford University. (2006). Knowledge Interchange Format (KIF). symposium on automated systems based on human skill and knowledge
http://www.ksl.stanford.edu/knowledge-sharing/kif/. Ramesh, B., & Jarke, M. (2001). Toward reference models for requirements
Kunz, W., & Rittel, H. W. J. (1970). Issues as elements of information sys- traceability. IEEE Transactions on Software Engineering, 27(1), 58–93.
tems. Working paper no. 131. Institute of Urban and Regional Development, RFC—Request for Comment. (1999). HTTP Extensions for Dis-
Berkeley. Available at http://www-iurd.ced.berkeley.edu/pub/WP-131.pdf. tributed Authoring—WEBDAV. IETF RFC 2518. Available at
Accessed October 2006. http://www.ietf.org/rfc/rfc2518.txt. Accessed October 2006.
Lee, J. (1997). Design rationale systems: Understanding the issues. IEEE Expert, Rolland, C., Souveyet, C., & Moreno, M. (1995). An approach for defining
12(3), 78–85. ways-of-working. Information Systems, 20(4), 337–359.
Lee, J., & Lai, K.-Y. (1996). What’s in design rationale? In T. P. Moran & J. M. Rus, I., & Lindvall, M. (2002). Knowledge management in software engineering.
Carroll (Eds.), Design rationale: Concepts, techniques, and use (pp. 21–51). IEEE Software, 19(3), 26–38.
Mehwah, New Jersey: Lawrence Erlbaum Associates, Inc. SAP. (2006). SAP—Business software applications and services (homepage).
Leonard, L. E. (1977). Inter-indexer consistency studies, 1954–1975: A review Available at http://www.sap.com/. Accessed October 2006.
of the literature and summary of study results. Occasional papers series, Schneider, R., & Marquardt, W. (2002). Information technology support in the
no. 131. University of Illinois Graduate School of Library Science, Urbana- chemical process design life cycle. Chemical Engineering Science, 57(10),
Champaign, IL. 1763–1792.
Manola, F., & Miller, E. (Eds.) (2004). RDF Primer. Online available at Slide. (2006). The Jakarta Slide project. Available at http://jakarta.
http://www.w3.org/TR/rdf-syntax-grammar/. Accessed January 2007. apache.org/slide/. Accessed October 2006.
Maropoulos, P. G. (2003). Digital enterprise technology—Defining perspec- Sowa, J. (Ed.). (1991). Principles of semantic networks: Explorations in
tives and research priorities. International Journal of Computer Integrated the representation of knowledge. San Mateo, CA: Morgan Kaufmann
Manufacturing, 16(7/8), 467–478. Publishers.
Marquardt, W., & Nagl, M. (2004). Workflow and information centered support Stanford Medical Informatics. (2006). The Protégé ontology editor and knowl-
of design processes—The IMPROVE perspective. Computers and Chemical edge acquisition system. Available at http://protege.stanford.edu/. Accessed
Engineering, 29, 65–82. October 2006.
McMahon, C., Lowe, A., & Culley, S. J. (2004). Knowledge management in Studer, R., Benjamins, V. R., & Fensel, D. (1998). Knowledge engineering:
engineering design, personalisation and codification. Journal of Engineering Principles and methods. IEEE Transactions on Data and Knowledge Engi-
Design, 15(4), 307–325. neering, 25(1–2), 161–197.
Mesihovic, S., Malmqvist, J., & Pikosz, P. (2004). Product data management Sun. (2006). Java 2 Enterprise Edition Homepage. Available at
system-based support for engineering project management. Journal of Engi- http://java.sun.com/javaee/. Accessed October 2006.
neering Design, 15(4), 389–403. Szykman, S., Sriram, R. D., Bochenek, C., Racz, J. W., & Senfaute, J. (2000).
Miatidis, M., Jarke, M., & Weidenhaupt, K. (2007). Using developers’ experi- Design repositories: Engineering design’s new knowledge base. IEEE Intel-
ence in cooperative design processes. In M. Nagl & W. Marquardt (Eds.), ligent Systems, 15(3), 48–55.
342 S.C. Brandt et al. / Computers and Chemical Engineering 32 (2008) 320–342

Szykman, S., Sriram, R. D., & Regli, W. C. (2001). The role of knowledge in next- W3C-World Wide Web Consortium. (1999). XSL Transformations (XSLT), ver-
generation product development systems. ASME Journal of Computation sion 1.0, W3C Recommendation. Available at http://www.w3.org/TR/xslt.
and Information Science in Engineering, 1(1), 3–11. Accessed September 2006.
Teknowledge. (2006). Semantic word. Available at http://mr.teknowledge. W3C-World Wide Web Consortium. (2004). RDF Vocabulary Descrip-
com/daml/SemanticWord/SemanticWord.htm. Accessed October tion Language 1.0: RDF Schema. W3C Recommendation. Available at
2006. http://www.w3.org/TR/rdf-schema/. Accessed October 2006.
Theißen, M., & Marquardt, W. (2007). Decision models. In M. Nagl & W. Mar- W3C-World Wide Web Consortium. (2006). Extensible Markup Lan-
quardt (Eds.), Collaborative and distributed chemical engineering design guage (XML) 1.0 (Forth Edition), W3C Recommendation. Available at
processes: From understanding to substantial support. Berlin, Heidelberg: http://www.w3.org/TR/xml/. Accessed October 2006.
Springer-Verlag. Yang, A., & Marquardt, W. (2004). An ontology-based approach to conceptual
Thomson, J., Adams, D., Cowley, P. J., & Walker, K. (2003). Metadata’s process modelling. In A. Barbarosa-Póvoa & H. Matos (Eds.), European
role in a scientific archive. IEEE Computer Magazine, 36(12), 27– symposium on computer aided process engineering, Vol. 14 (pp. 1159–1164).
34. Elsevier.
UGS. (2006). Teamcenter 2005. Available at http://www.ugs.com/products/ Zeng, A., Pan, D., Zheng, Q. L., & Peng, H. (2006). Knowledge acquisition
teamcenter/. Accessed October 2006. based on rough set theory and principal component analysis. IEEE Intelligent
van Gigch, J. P. (1991). System design modeling and metamodeling. New York: Systems, 21(2), 78–84.
Springer. Zhao, C., Hailemariam, L., Jain, A., Joglekar, G., Venkatasubramanian, V.,
Venkatasubramanian, V., Zhao, C., Joglekar, G., Jain, A., Hailemariam, L., Morris, K., et al. (2006). Information modeling for pharmaceutical prod-
Suresh, P., et al. (2006). Ontological informatics infrastructure for pharma- uct development. In W. Marquardt & C. Pantelides (Eds.), 16th European
ceutical product development and manufacturing. Computers & Chemical symposium on computer aided process engineering and 9th international
Engineering, 30, 1482–1496. symposium on process systems engineering (pp. 2147–2152). Elsevier.
von Wedel, L. (2002). CapeML—A model exchange language for chem- Zhao, C., Jain, A., Hailemariam, L., Joglekar, G., Venkatasubramanian, V.,
ical process modeling. Technical report (LPT-2002-16), Lehrstuhl für Morris, K., et al. (2006). A unified approach for knowledge modeling in phar-
Prozesstechnik. RWTH Aachen University. maceutical product development. In W. Marquardt & C. Pantelides (Eds.),
Weidenhaupt, K. (2001). Anpassbarkeit von Software-Werkzeugen in prozess- 16th European symposium on computer aided process engineering and 9th
integrierten Entwicklungsumgebungen. Ph.D. thesis. Informatik V, RWTH international symposium on process systems engineering (pp. 1929–1935).
Aachen University. Elsevier.
Westerberg, A. W., Subrahmanian, E., Reich, Y., Konda, S., & the n-dim group. Zhao, C., Joglekar, G., Jain, A., Venkatasubramanian, V., & Reklaitis, G. V.
(1997). Designing the process design process. Computers & Chemical Engi- (2005). Pharmaceutical informatics: A novel paradigm for pharmaceutical
neering, 21(S), S1–S9. product development and manufacture. In L. Puigjaner & A. Espuña (Eds.),
Windream GmbH. (2006). Managing documents by Windream. Available at European symposium on computer aided process engineering, Vol. 15 (pp.
http://www.windream.com/. Accessed October 2006. 1561–1566). Elsevier.

You might also like