Conference PaperPDF Available

A Meta Model for Artefact-Orientation: Fundamentals and Lessons Learned in Requirements Engineering


Abstract and Figures

Requirements Engineering (RE) processes are highly volatile due to dependencies on customers' capabilities or used process models, both complicating a standardised RE process. A promising solution is given by artefactorientation that emphasises the results rather than dictating a strict development process. At such a basis one is able to incorporate domain-specific methods for producing artefacts without having to take into account the variability of process definitions. Although artefacts are known to support customisable development processes, there still is no common agreement about the structure and semantics of artefact-based methodologies. In this paper we discuss different interpretations of the term artefact considering aspects like process integration capabilities and necessities within individual project environments. We contribute a meta model for artefact-orientation that is inferred from two RE models elaborated within industrial cooperation projects of our research group. We conclude with a discussion of performed case studies and ongoing work.
Content may be subject to copyright.
A Meta Model for Artefact-Orientation:
Fundamentals and Lessons Learned in
Requirements Engineering
and Manfred BROY
Technische Universität München, Germany
Abstract. Requirements Engineering (RE) processes are highly volatile due to de-
pendencies on customers’ capabilities or used process models, both complicating a
standardised RE process. A promising solution is given by artefact-orientation that
emphasises the results rather than dictating a strict development process. At such a
basis one is able to incorporate domain-specific methods for producing artefacts
without having to take into account the variability of process definitions. Although
artefacts are known to support customisable development processes, there still is
no common agreement about the structure and semantics of artefact-based meth-
odologies. In this paper we discuss different interpretations of the term artefact
considering aspects like process integration capabilities and necessities within in-
dividual project environments. We contribute a meta model for artefact-orientation
that is inferred from two RE models elaborated within industrial cooperation pro-
jects of our research group. We conclude with a discussion of performed case stud-
ies and ongoing work.
Keywords. artefact-orientation, model-based development, requirements engineer-
ing, meta-modelling
1. Introduction
Modelling systems during the development process has become a widely accepted
paradigm in the development of software and embedded systems. One goal is, finally,
to generate source code from models. Not only in such an approach, where system
models are systematically derived from requirements, the precise specification of re-
quirements is crucial. Obviously, if the requirements are to be precisely specified, the
structure, syntax and semantics of requirements documentation, that capture various
aspects, have to be described, too.
Typically, requirements are collected for a specific family of systems (like busi-
ness information systems), so a systematic interpretation of the basic concepts that re-
spect the needs of the application domain is required. This leads to the tendency of de-
fining a systematic artefact-based requirements engineering (RE) process, where arte-
fact models are used as reference models that capture the domain-specific results of the
development steps. In addition to the need of capturing the basic concepts of a domain
of application, early stages of development are characterised by volatile or changing
project environments due to the dependency to customers’ capabilities, to used process
models and to produced specification documents. Also this barrier can be tackled by
artefact models since they guide the systematic elaboration of measurable results inde-
pendent of the chosen development process for the artefacts, i.e. the development result
is described independently of the process producing it.
In a model-based design and development paradigm, however, an artefact is seen
as a structured abstraction of modelling elements used as input, output, or as an inter-
mediate result of a process. An artefact has particular properties (structure, behaviour,
etc.) and may be precisely described using standardised (semi-)formal modelling con-
cepts. Upon such a description one is able to incorporate different techniques and no-
tions, define clear responsibilities, and support a progress control for the production of
artefacts. Hence, as the definition of an artefact model as a backbone supports a cus-
tomisable process, it has become a well-accepted approach for defining development
processes [11]. Internal (micro-)processes are transparent for contract-parties. They
agree upon deliverables (artefacts) and do not have to care about the detailed realisation.
Furthermore, if artefacts are defined for exchange, heterogeneous development envi-
ronments can easily be coupled [12], also abstracting from concrete methods or tools.
However, a first step towards the definition of an artefact-based methodology for a
company-wide use consists in the precise definition of the artefacts and their relations
being of interest for a particular development process, in our case for RE. This guaran-
tees that all parties in a project have the same understanding of what will be exchanged
within this RE process. Once an artefact model is defined, it is obligatory for customers
and all team members working on it.
The next step consists in the integration of the artefact-based RE approach into the
development process by establishing the associations of requirements artefacts with
development contents (process integration). A mandatory prerequisite for this integra-
tion is the definition of interfaces. Thereby, compatibility of the artefact-based RE pro-
cess and the used development process has to be achieved. This means that the RE pro-
cess has to be described by following the rules of the envisioned development process.
Furthermore, the roles relevant for RE have to be harmonised with the roles given by
the development process. This achieves finally a seamless integration of RE as an
interconnected discipline to architecture, design, code, and quality assurance.
Problem Statement. So far, there is no common understanding and agreement about
artefact-orientation in general and the term artefact in particular. There are interpreta-
tions in the granularity range from general document structures to data models defined
using domain-specific languages (DSLs [3]). We believe all perspectives have to be
considered in order to address the necessary process integration capabilities and at the
same time precision of the artefacts produced within single projects. Although, this
problem is understood, so far there exists not enough guidance to tackle it since the
definition of an artefact model and the view taken strongly depends on the exact pur-
pose of the artefact model.
Contribution. We contribute a meta model for artefact-based RE that results from les-
sons learned in different cooperation projects of our research group. This meta model is
suitable to cover the RE domain, its usual artefacts and relations ready to be integrated
into the process. Applying meta modelling on RE means to define a language (respec-
tively an abstract syntax) to model (1) requirements artefact types, (2) artefact struc-
tures and dependencies/associations and (3) content items and content relations. Fur-
thermore, operations can be introduced to build concrete exemplars (instances) of par-
ticular artefacts in a consistent frame and to provide capabilities for customisation and
express variability. To cover all necessary elements of a project, the RE meta model
contains structure, information and semantics of not only artefacts, but also processes
and roles. The meta model is compared to DSL-based modelling [3] a concrete DSL. In
contrast to the multi-layered abstraction hierarchy of the Object Management Group
(OMG) [8], we position the meta model according to Henderson-Sellers [28] as an ab-
straction of an artefact-based methodology and not as an abstraction of a modelling
language for generating code (like the UML).
In general, meta models capture only basic types, therefore a domain-specific in-
terpretation is needed to come up with concrete RE artefacts, e.g. by establishing an
artefact-based methodology for the application domain of business information systems.
Using a meta model, it is possible to integrate operations to support such interpretations
systematically, e.g. commonalities and variabilities according to product line ap-
proaches [2], or explicit relations to enhance variability etc. However, meta models can
be created in different ways: One option is to create a meta model constructively “from
scratch”. This option is usually used in completely new domains by developing DSLs.
Another option is to infer a meta model by analysis of existing elements from given
domain-specific models. In this paper we follow the second variant.
We present two RE models that have been elaborated in research projects for two
different domains of applications: for the one of embedded systems and for the one of
business information systems. We infer in a second step a meta model by analysing and
abstracting the results and concepts of the two developed RE models. Hence, we con-
struct a unified framework that benefits the future elaboration of artefact-based RE
models. We use this contribution as a basis for tasks, such as domain-specific model-
ling, process integration and tool-support. We finally discuss the advantages of the ap-
proach and its applicability with respect to several performed case studies.
Outline. The remainder of this work is organised as follows: In the following Sect. 2
we describe the fundamentals and related work in the areas of our contribution. Section
3 describes two sample RE models that have been elaborated within research cooper-
ations for two different domains of application. Both serve as a basis for analysing
similarities in Sect. 4. The meta model for artefact orientation is then inferred in Sect. 5.
In Sect. 6 we discuss the results for example with respect to performed case studies,
before finally giving in Sect. 7 concluding remarks.
2. Foundation and Related Work
Method engineering [26] contributes important approaches for the design of domain-
specific RE processes. Method engineering deals with the construction of domain-
specific methods to deploy modelling techniques for the elaboration of artefacts. The
selection of methods according to individual project settings is known as “situational
method engineering”, the integration of methods into a development process model is
known under the term “method weaving”. Compared to method engineering, where the
focus lies on the definition, the selection and the integration of methods into a devel-
opment process, artefact-orientation gives a more detailed view onto the corresponding
results structure with the purpose of enabling seamless modelling of consistent results
without having to take into account the variability of volatile processes and the com-
patibility of methods.
Still, while the area of method engineering provides with meta models a clear
structure of method-based methodologies, there still is no common agreement on the
structure and semantics of artefact-based methodologies. There exist different interpre-
tations of same and similar terms, like artefact, work product or product type. Terms
like these often co-exist in different process models, such as the referred work products
that are described in the rational unified process or the product types that are in scope
of the V-Modell XT [24], a German artefact-based development standard for software
and systems development projects. Available approaches differ in their understanding
towards artefacts and artefact models, whereby artefact models are defined considering
different views (Fig. 1): structure models, generic content models, domain-specific
content models, and finally integrated modelling theories.
Figure 1 Variations of Artefact Models and Implications
The variations of artefact models as shown in Fig. 1 are characterised in the following:
Structure Models. Structure models include a basic description of general topics and
terminology to be considered within a structure reference for documents or data sets.
One example is given by the IEEE Std. 830-1998 recommended practice for software
requirements specifications [1] that defines a taxonomy as a reference model for the
construction of specification documents. Another example is given by the mentioned
V-Modell XT, where artefacts are defined from a comprehensive process viewpoint as
an abstract result set of an overall development process including for each artefact a
taxonomy of general topics and content to be covered. Structure models in general em-
phasise the structure of the results and aim at a common, standardised understanding on
what should be produced in general within individual projects.
Generic Content Models. Generic content models define contents, using for example
semi-formally structured checklists. Such checklists describe independent of the chosen
structure the content to be considered and give guidance for choosing exemplary meth-
ods. The Volere requirements specification templates [4] is an example. Still, the se-
mantics of the contents is not precisely defined within such taxonomy-based standards.
Hence, they give no guidance with respect to the elaboration of the contents and their
interdependencies and fail therefore at guiding the elaboration of (syntactically) consis-
tent artefacts. One step towards this direction in the context of RE has been made with
the requirements engineering reference model (REM) [5] of the Technische Universität
München and Siemens Corporate Research. REM defines the structure of goals, re-
quirements and specifications within a proposed taxonomy-based guideline and infor-
mally describes dependencies between the elements of the guideline based on proposed
refinement principles. REM was not developed for a specific domain of application.
Existing approaches like the introduced ones use models only to establish a generic
reference model of the general content and relations to be considered. Still, those mod-
els provide no concrete information about the possible concepts like the ones for con-
structing use case models and thereby they do not formally capture the interdependen-
cies between the concepts like the dependency of use case models to actor descriptions.
In fact, the description of concrete concepts strongly depends on a chosen domain of
application what leads to domain-specific content models.
Domain-specific Content Models. The next step in the direction of formalisation are
domain-specific content models that abstract from the content of documents and data
sets of a particular domain of application. In such approaches, modelling concepts, in-
cluding types and dependencies, are used to describe artefacts and complex artefact-
based systems. Here also a formal syntax and bits of semantics are used to define arte-
facts’ structure, their creation, and content. On this level of abstraction, (domain-
specific) modelling languages can be defined [3,6,7] using, e.g., data models for their
representation. These data models provide the possibility to specify the desired amount
of redundancy of contents. Furthermore, one is able to establish a comprehensive tool
support on this level, as data models are available for computing. Schätz describes in
[18] the possibilities of additionally enriching domain-specific concept models with
conformance constraints in order to guide the elaboration of RE models at project-level
(as known from the area of automated model-driven approaches). Thereby, he ensures
quality within those models in terms of ensuring conformance of the models being pro-
duced within a project to the reference model. In particular, the application of conform-
ance constraints to the concept model benefits the domain-specific awareness of the
results in terms of ensuring syntactic completeness and consistency in a tool-supported
manner. How conformance constraints are defined, depends on the chosen quality as-
surance techniques. One possibility is to define the conformance constraints in terms of
logic-based formalisms, as done in [9].
Integrated Modelling Theory. Mathematical models give the most formalised view
onto artefact models and are often used to precise a comprehensive modelling theory.
For instance, by defining properties of modelling concepts like “services” and math-
ematical relations to other concepts like “business processes” and, thus, they create a
semantic foundation for commonly used terms. Mathematical models mostly offer their
own syntax or are used as a basis for the definition of a syntax with a precise math-
ematical meaning (see e.g. Focus, [9]). Since the elements to be produced within a pro-
ject are all defined according to the mathematical model, the syntactic consistency is
Summarised, the definition of an artefact model and the understanding of the term
artefact strongly depends on the purpose of the model [19], e.g. concerning the cou-
pling of different tools or concerning the definition of a flexible process model. Some
of the resulting views onto an artefact exclusively focus on structure of artefacts and
some (although implicating also a notion of structure) emphasise the content. Each of
the given views implies going in Fig. 1 from left to right a higher degree of operation-
ability and precision in the design of an artefact model. Precision considers the struc-
ture and the content of specifications from which the models abstract for a particular
domain of application. The higher the degree of precision (as given by the latter exam-
ples), the higher the enforcement of syntactic consistency as artefact models are used as
a domain-specific reference model for a particular syntax. At the same time the less the
degree of flexibility. For practical usage, the establishment of domain-specific concept
models has become a wide accepted technique, mostly arising from the benefit of ena-
bling seamless modeling. Seamless modelling is enabled because such models do not
only describe how a RE model is structured, but also how valid instances are created.
For valid instances, all artefacts based on the defined reference model are created the
same way and all artefacts of the same type have the same basic structure (e.g. each use
case has at least one actor assigned, a detailed description of what this use case is good
for, …). This information is a basic component to build up interfaces to other (process)
disciplines and their artefacts and to provide finally tool-support in a project.
Therefore, defining artefact models from this perspective, a process-integrated,
seamless and tool-supported modelling of requirements is possible, what finally is not
in scope of the area of method engineering.
3. Requirements Engineering Approaches
This section describes two concrete samples of requirements engineering approaches
that have been elaborated in recent research projects. We first describe the REMsES
project considering a RE model for embedded systems and afterwards the REMbIS
project considering the domain of business information systems. Similarities of both
models are analysed in detail in Sect. 4 in order to infer a meta model for artefact orien-
tation in Sect.5.
3.1. REMsES
The REMsES project was a research collaboration with partners from academia and
industry1. Goal of the project was the elaboration of a practical guide for systematic
requirements engineering and management of embedded systems, especially in the
automotive domain [13]. The result is an approach with a reference artefact model (see
Fig. 2). This reference model is based on two key concepts: support for abstraction
levels and coverage of three content categories. The structure supports requirements
engineers in determining which type of model they should use and what kind of ab-
stractions they should create in a particular project.
Coverage of Abstraction Levels. Requirements at different levels of detail, ranging
from business goals to fine-grained technical requirements (e.g. concerning the system
hardware), need to be included in the requirements document. High-level requirements
provide a justification for detailed requirements and support the understandability of
the requirements. Low-level requirements are needed to provide enough information
for implementing the system correctly.
In the REMsES project, a hierarchy of three abstraction levels was adopted: system
level, function groups level, and HW/SW level. The abstraction levels form the vertical
dimension of the structure shown in Fig. 2.
1 The REMsES guide is available for free download at
Figure 2 REMsES Artefact Model
At the System Level, the stakeholders have a black box view of the system. RE ar-
tefacts modelled at this levels focus on the usage of the system by human beings and
other systems but no system-internal or auxiliary functions are considered.
The Function Groups Level represents a whitebox view onto the system. At this
level, the system is viewed as a network of interacting, logical units obtained by a func-
tional decomposition of the system. These function groups have defined interfaces and
can interact with each other as well as with the system environment.
At the Hardware/Software Level, a coarse-grained partitioning of the system’s
functionality into HW and SW is defined. For this purpose, a system is decomposed
into a (coarse-grained) HW topology and SW architecture.
Coverage of 3 Content Categories. By analysing RE documents in the automotive
domain, three main content categories of such a document were identified: context,
requirements, and (high-level) design. Therein, the categories context and design con-
tain important information for the RE process and therefore have a considerable influ-
ence on the requirements. The content categories relate to the horizontal dimension of
the structure shown in Fig. 2. The three categories are defined orthogonally to the
abstraction layers.
The Context of the system is the part of the real world that influences the require-
ments for the system. Context artefacts are, for example, laws, business goals, general
constraints, environmental conditions, etc. The context sets the frame within which the
system is developed.
Requirements are expressed using modelling concepts. Three types of such con-
cepts, i.e. goal models [14], scenario models [15], and function models [16], are identi-
fied as the most important models needed to support the RE process of an embedded
Design as the third part plays an important role. In the development of embedded
systems, requirements and design are tightly intertwined, because the knowledge about
major system components is inevitable to specify detailed requirements. By introducing
design explicitly as a content category, we support developers in documenting re-
quirements and design as separate models, rather than intermingling the two.
3.2. REMbIS
REMbIS results from a research corporation between the Technische Universität
München and Capgemini sd&m AG. REMbIS is a model-based RE approach for the
application domain of business information systems. It consists of (1) an artefact ab-
straction model that defines horizontal abstraction and modelling views, (2) a concept
model that defines those aspects dealt with during construction of models including the
definition of possible notions for producing the models and finally (3) a method de-
scription that defines the activities and tasks of the RE process. Each sub-model is sub-
sequently described.
Figure 3 Excerpt of the Artefact Models and the Concept Model of REMbIS
Artefact Abstraction Model. The artefact abstraction model defines horizontal levels
of abstraction that represent the stages of refinement of different RE-specific results.
Modelling views structure each level according to environment, behaviour, structure
and information. The levels and the views serve as a basis for traceability among the
results. The left side of Fig. 3 depicts the artefact abstraction model of REMbIS, while
embedding exemplary concepts (denoted by the boxes). The levels are divided into two
major areas concerning the description of an organisation’s and business structure (the
upper three layers) and the compliant description of requirements for the underlying IT
infrastructure (the last depicted levels).
The organisation’s context defines the high-level, long-term steering principles
and objectives of a company. The business process hierarchy describes the organisa-
tion’s structure, including a taxonomy of the main offered business activities. Derived
from this structural view, the business process logic emphasises the internal realisation
of the activities in terms of a business process model that defines a workflow descrip-
tion including interchanged information and participating roles.
The information system service hierarchy defines the external behaviour of single
systems by defining what parts of the business process model shall be supported by the
system. It defines user-visible characteristics of the system by means of e.g. an (infor-
mation system) service description and a use case model. The information system’s
constraints capture single quantified requirements that address the system in its appli-
cations, architecture and environment.
Finally, to define the scope of the different levels that address the business and sys-
tem descriptions, REMbIS makes use of visions, e.g. the system vision that defines the
system’s context and summarizes all relevant information system services.
Concept Model. The concept model defines all elements and relations of used model-
ling techniques for the construction of domain-specific RE models in individual project
environments (such as structure and content of use case models and their relation to
service models). For each of the concepts, the approach includes a description of rec-
ommended modelling languages. The right side of Fig. 3 shows an exemplary excerpt
of the concept model that defines the contents of the information system service hier-
archy. The concepts shall not be discussed in detail, but illustrate how the model serves
as a basis for the definition of terminology and consistency rules of all produced enti-
ties of a project.
Methods. Based on the concept model, REMbIS offers a description of all the tasks
performed in order to use, modify and produce selected concepts, enriched with best
practices that provide a set of chosen notions for constructing the concepts.
4. Analysis and Inference
Based on the two project samples described above, we analyse both approaches with
respect to their similarities. This first step is aimed at abstracting from the specialities
of the variants to identify similarities. Based on the results, we infer a basis for defining
a meta model for artefact orientation in the following section.
Table 1. Analysis of the Example Models with respect to core Issues
Embedded Systems
Business Information Sys-
Generic, domain-
independent approach
Refinement over abstrac-
tion layers and content
Refinement over abstraction
levels and modelling views
Abstraction levels and
Focus on structural aspects
and notations
Concept models abstracting
from description techniques
and notions
Artefact types separating
structure and content with
Specification techniques
and process close to RUP
Set of specific tasks (meth-
ods), checklists and mile-
stones inspired by RUP
Process interface to couple
methods and artefacts
Coarse-grained definition
Roles coupled to concepts
and tasks
Interface to assign roles
and artefacts
Table 1 illustrates a summary of both RE models, REMsES and REMbIS, which
are subsequently compared according to different scopes.
Domain. The two approaches have been elaborated for different domains of applica-
tion: REMsES for the one of embedded systems, REMbIS for the one of business in-
formation systems. As a consequence, the meta model shall not be described using any
domain-specific aspects.
Refinement. REMsES and REMbIS both offer means to refine the results of RE over
different domain-specific degrees of abstraction. For each of the described degrees of
abstraction, a subset of the results is grouped according to either content category or
view, where both describe common characteristics of different contents. Hence, the
least common denominator is the necessity of describing vertical and horizontal ab-
straction for RE models.
Results. Regarding the results, REMsES emphasises the structure of the results and
syntactical possibilities for producing these results. A major benefit of this is easy inte-
gration within process structures, because workflow entities can be coupled to cohe-
sively described results. Instead, REMbIS defines a concept model based on the ele-
ments of domain-specific modelling techniques with focus on consistency among the
results rather than on process integration. As a consequence, REMbIS has been defined
in a process-agnostic manner and features an indirect overall structure that results from
the dependencies between the concepts. Each content concept lays the foundation for
choosing a specific syntax. The explicit distinction between structure and content en-
ables a process integration and at the same time consistency among the results.
Methods. Both approaches offer a set of specification techniques and follow a process
integration in terms of defining an abstract sequence of producing the results. In par-
ticular, REMsES defines concrete tasks and a coarse-grained integration into a general
development process. REMbIS defines milestones and tasks both coupled to the results,
but does not define any specific process. Hence, the meta model shall offer a descrip-
tion of a generic process model that allows to couple methods to results without dictat-
ing a concrete process.
Roles. Roles are defined both in REMsES and REMbIS, but REMbIS additionally pro-
vides their assignment to concepts and tasks.
5. Meta Model for Artefact-Oriented RE
As we identified a common superset for the two RE models, we are now able to infer a
meta model that describes a language for the artefact-oriented paradigm. Fig. 4 il-
lustrates the artefact-oriented RE meta model, which is structured into single sub-
models. These sub-models define major areas of concern with a specific set of connec-
tors (association classes) in between these areas. This modular structure fits to the as-
pects identified in Sec. 4 and an ease modification of the single identifies areas. We
subsequently describe the single sub-models of the meta model: the artefact model, the
artefact abstraction model, the generic process model and finally the generic role model.
Artefact Model. The artefact model builds the backbone by defining structure and con-
tent of domain-specific results of RE activities. The structure addresses the aspects of
hierarchically ordered documents or data sets being produced during development tasks
in which single content items serve as containers for the concepts, respectively concept
models. The concept models (the artefacts’ content) define according to Schätz those
aspects of a system under consideration dealt with during the development process [17]
and reflected in the elements of the description techniques and their relations [18]. How
the models are exactly defined depends on the chosen application domain from which
the selection of description techniques arises (thereby often referred to as domain mod-
els). Based on these differentiating angles from which an artefact is tackled, we define
an artefact as follows:
An artefact is a deliverable that is produced, modified, or used by a sequence of
tasks that have value to a role. Artefacts are subject to quality assurance and version
control and have a specific type. They are hierarchically structured into content items
that define single areas of responsibility and that are the output of a single task. Each
content item encompasses at its lowest level of decomposition:
1. Concepts: a concept defines the elements and their dependencies of domain-
specific description techniques used to represent the concern of a content item.
Concepts have a specific type and can be decomposed to concept items. The
latter differentiation is made if different items of a concept can be described
with different techniques.
2. Syntax: the syntax defines a concrete language or representation that can be
chosen for a specific concept.
3. Method: The method (or task) describes the sequence of steps that is per-
formed in order to make use of a concept.
Figure 4 Meta Model for Artefact-Orientation
Note that when discussing artefacts, we usually refer to artefact types (“require-
ments specification”) rather than to an artefact (“requirements specification ’Travel
Ordering System’”). Similarly, when discussing concepts, we refer to concept types
(“Use Case”) instead of to a concept (“UML Activity Diagram: ’Search Hotel’”). Fur-
thermore, each artefact has dependencies. We distinguish between content dependen-
cies and structural dependencies. The structural dependencies are defined by compo-
sition, as an artefact type is decomposed into several content items (e.g. representing
chapters within the specifications) and, finally, the dependencies that result from the
content dependencies. For reasons of complexity, these dependencies are not restricted
by specific types within the meta model. Hence, as the content dependencies are not
typed, the domain- specific instances need a definition of conformance constraints. One
possibility for the definition of such constraints is given by the use of the Object Con-
straint Language (OCL) for supporting the conformance of the models being created in
a project to the reference models in a tool-supported manner. Note that the conform-
ance of domain-specific instances to the meta model at hand is given, if these are de-
scribed in the language defined by the meta model.
Artefact Abstraction Model. The artefact abstraction model defines the domain-
specific abstraction that refers to the construction and refinement (and / or decomposi-
tion) of the artefacts. We distinguish between (1) horizontal abstraction, that reflect
aspects like structure, behaviour or roles and referred to as modelling views [20] and
(2) vertical abstraction that describes, for each of the views, the refinement and / or
decomposition of corresponding artefacts, also referred to as levels of abstraction
[21,22]. These levels of abstraction have specific characteristics that match with the
degree of detail of an artefact’s underlying concept and its concern.
Generic Process Model. The generic process model defines all entities of a workflow
description that can be used to define a specific process. In particular, we define (1)
milestones in terms of a point in time in which an artefact or content item should be
completed and (2) tasks that build a part of phases and activities. A phase (e.g. re-
quirements engineering) defines a repeatable set of activities which, in turn, define ma-
jor areas of concerns (e.g. requirements elicitation). For such an area of concern, there
exist several tasks where each consists of a sequence of atomic steps that are performed
by a role in order to produce, modify and / or use an artefact as input or output. Hence,
a task is a method in the sense as it is defined in the area of method engineering [23]. A
task supports the selection of which artefacts are needed as basis for modifying or pro-
ducing further artefacts by the choice of a specific syntax. The generic process model
also describes the process interface to be used for integrating the RE meta model into a
development process model. Model elements such as milestones are underspecified, as
the precise descriptions have to be delivered by the process model.
Generic Role Model. The generic role model describes a logical representation of the
individuals who directly participate in or indirectly contribute to the development pro-
cess. A role has the responsibility for at least one artefact while performing a set of
activities. Similar to the process model, the generic role model is underspecified, as the
instance role model has to be provided by the development process.
6. Discussion
With the meta model, presented in the previous section, we define a language for arte-
fact-oriented RE. This language provides applicable guidelines for domain-specific RE
models that support consistency and completeness within the results across different
projects, continuity within the process and variability in the process definitions.
Syntactic Consistency and Completeness. Using the proposed language for the
description of domain-specific RE models supports syntactic and semantic consistency
and completeness among the results of RE. Syntactic consistency and completeness
include the restriction of possible description techniques (by defining concept models)
and thereby lay the foundation for consistency and completeness conditions with re-
spect to the choice of a syntax. In explicit, the RE model answers the question “How is
the syntax of artefact A related to the syntax of artefact B?”. Semantic consistency
means consistency between project-specific contents, i.e. it provides an answer to the
question “Does the statement of artefact A fit to what the model of artefact B de-
clares?”. This can not be handled by an artefact model, but verification of semantic
consistency requires syntactical consistency.
Continuity. Artefact-orientation fundamentally enables continuity within the develop-
ment process. As the concept model provides the basis for seamless modelling, the pro-
cess that is coupled to the artefact model provides a continuous workflow when elabo-
rating the artefacts.
Variability supporting Customisation. Artefact-orientation tackles the problem of
variability in designing a process within volatile environments. The meta model offers
a modular structure and supports process integration capabilities, i.e. the customisation
of artefact-based RE models at organisational levels by integrating them into a devel-
opment processes of particular companies. This process integration can for example be
performed with the V-Modell XT [24], a meta model-based German standard for soft-
ware & systems development, by integrating the single elements over the association
classes into the organisation-specific process models (see also Sect. 5). In general, the
connections that are given by several association classes state what connection points
have to be available (obviously, for integrating the RE meta model, a meta model-
based development process model is a mandatory prerequisite). The artefact-centric
view unifies the definition of taxonomy-based structures of documents or data sets and
the definition of domain-specific modelling concepts (content). In a second step, we
couple all elements of a process model to the artefact model. Therefore, while method-
based methodologies face problems with respect to syntactic variability during the
integration of new methods, artefact-oriented approaches support a systematic integra-
tion of methods by innately restricting the results’ contents and therefore indirectly the
possibilities of selecting methods producing them. As each domain-specific RE model,
derived from the meta model, exhibits the same modular structure, it satisfies in addi-
tion to needed process integration capabilities the needs of customisation at project
level. Therefore, it serves as a reference model that guides through the elaboration of
the results of RE being compliant to the RE model. Moreover, it can be customised
according to varying project parameters that strongly affect the elaboration of single
artefacts. Therefore, it tackles the deficiencies of process-driven approaches.
Trade-Off between Precision and Flexibility. When establishing an artefact-based
approach supporting an efficient and effective RE process, there is a trade-off between
flexibility and precision: The more options are provided for syntax or concepts, the
higher is the variability with respect to the relations between artefacts and the more
complex is the verification of consistency. The higher the restrictions for syntax, the
more suffers the flexibility for applying the models at variable project-levels due to
making knowledge explicit. Referring to Fig. 1, there is a conflict between a general
and a precise view. The more precision there is in the chosen way of defining an arte-
fact model, the more capabilities there are in gaining consistency and completeness
within the models considered compliant to the artefact models. Following the provided
meta model of Sect. 5, artefact models perform therefore a unique combination of two
views, the view onto content and dependencies via concept models embedded into a
second taxonomy-based structure view. Hence, the necessary degree of flexibility is
ensured while the degree of variability when constructing and representing the content
decreases since we still have restrictions on the content and the used syntax.
Evaluation in Case Studies. The introduced meta model is suitable to cover the stated
needs since it abstracts from two concrete RE models that both have been evaluated in
industry. The REMsES approach was evaluated in case studies, student experiments,
and pilot projects at the sites of the industrial partners (see also [27]). The REMbIS
approach is fully integrated into the development process of Capgemini sd&m AG.
Their engineers have been trained on REMbIS in seminars and constantly apply it in
custom software development projects in which an evaluation can not be published for
reasons of confidentiality.
The meta model was the basis for the development of BISA (artefact-based approach
for Business Information Systems’ Analysis) that offers customisation capabilities for
process integration and for project-specific customisation. The latter is performed by
customising the artefact model according to project parameters that influence the
possibilities and necessities of elaborating single artefacts to which the parameters are
coupled (see also [29]). The process integration has been performed with the V-Modell
XT, a meta model based standard process including a comprehensive tool chain. The
resulting integrated, customisable VM BISA [25] has successfully been applied in a
project hosted by Siemens.
7. Conclusion
This paper concentrates on the paradigm artefact-orientation in requirements engineer-
ing and presents a meta model. This meta model is inferred from two concrete domain-
specific RE models of our research group: one for the application domain of embedded
systems and one for the application domain of business information systems. Both RE
models have been evaluated in real-life projects and thereby validated with respect to
We outlined the advantages of artefact-orientation: Taking an artefact model as
backbone of project execution benefits (1) syntactic consistency and completeness of
the results being compliant to the domain-specific reference model, (2) seamless mod-
elling of the results and continuity within the development process chain and, (3) can
be customised to individual needs. Such a customisation can be performed at organisa-
tional level in terms of process integration and at project level by adapting domain-
specific artefact models according to the influences of individual project environments.
The integration into a comprehensive process framework and the definition of a cus-
tomisation approach considering the project level has been performed in [25] and suc-
cessfully applied in a pilot project hosted by Siemens.
[1] IEEE: IEEE Recommended Practice for Software Requirements Specifications. IEEE Std 830-1998,
[2] Rombach, D.: Integrated software process and product lines. In Li, M., Boehm, B., Osterweil, L., eds.:
International Software Process Workshop. Lecture Notes in Computer Science 83–90, 2005.
[3] Greenfield, J., Short, K.: Software Factories. Wiley and Sons, 2004.
[4] Robertson, J., Robertson, S.: Volere Requirements Specification Templates-Edition 11. Available On-
line at, August, 2007.
[5] Geisberger, E., Broy, M., Berenbach, B., Kazmeier, J., Paulish, D., Rudorfer, A.: Requirements
engineering reference model. Technischer Bericht, Technische Universität München, 2006.
[6] Gronback, R.: Eclipse Modeling Project. Addison-Wesley, 2009.
[7] Steinberg, D., Budinsky, F., Paternostro, M., Merks, E.: EMF -Eclipse Modeling Framework. Addison-
Wesley, 2009.
[8] OMG: Unified modelling language (uml) -infrastructure and superstructure 2.1.1. Specification, Object
Management Group, 2007.
[9] Schätz, B.: The odl operation definition language and the autofocus/quest application framewoek aqua.
Technical report, Technische Universität München, 2001.
[10] Ameluxen, C., Rötschke, T., Schürr, A.: Graph transformations with mof 2.0. In: 3rd International Fu-
jaba Days, 2005.
[11] Managing Successful Projects with Prince 2. O!ce of Government and Commerce, 2009.
[12] Kuhrmann, M., Kalus, G., Chroust, G.: Tool-Support for Software Development Processes. In: Social,
Managerial, and Organizational Dimensions of Enterprise Information Systems. IGI Global, 2009.
[13] Penzenstadler, B., Sikora, E., Pohl, K.: Guiding requirements modelling in the embedded systems do-
main with an artefact reference model. In: REFSQ, 2009.
[14] van Lamsweerde, A.: Requirements engineering: From craft to discipline. In: SIG¬SOFT ’08/FSE-16:
Proceedings of the 16th ACM SIGSOFT International Symposium on Foundations of Software Engi-
neering, New York, NY, USA, ACM 238–249, 2008.
[15] Cockburn, A.: Writing Effective Use Cases. Addison-Wesley Longman Publishing Co., Inc., Boston,
MA, USA, 2000.
[16] Davis, A.M.: Software requirements: objects, functions, and states. Prentice-Hall, Inc., Upper Saddle
River, NJ, USA, 1993.
[17] Schätz, B., Pretschner, A., Huber, F., Philipps, J.: Model-Based Development of Embedded Systems.
In: Advances in Object-Oriented Information Systems, Lecture Notes in Computer Science. Volume
2426., Springer-Verlag 298– 311, 2002.
[18] Schätz, B.: Model-Based Development of Software Systems: From Models to Tools. Habilitation,
Technische Universität München, 2008.
[19] Endres, A., Rombach, H.: A handbook of software and systems engineering: empirical observations,
laws and theories. Addison-Wesley, 2003.
[20] Broy, M., Schätz, B., Wild, D., Feilkas, M., Hartmann, J., Gruler, A., Penzenstadler, B., Grünbauer, J.,
Harhurin, A.: Umfassendes Architekturmodell für das Engineering eingebetteter software-intensiver
Systeme. Technical Report TUM-I0816, Technische Universität München, 2008.
[21] Gorschek, T., Wohlin, C.: Requirements abstraction model. Requir. Eng. 11(1) 79–101, 2005.
[22] Ramesh, B., Jarke, M.: Toward reference models for requirements traceability. IEEE Trans. Softw. Eng.
27(1) 58–93, 2001.
[23] Braun, C., Wortmann, F., Hafner, M., Winter, R.: Method construction -a core approach to organiza-
tional engineering. In: SAC, ACM 1295–1299, 2005.
[24] V-Modell XT., 2008.
[25] Mendez Fernandez, D., Kuhrmann, M.: Artefact-based requirements engineering and its integration into
a process framework. Technical Report, Technische Universität München, 2009.
[26] Brinkkemper, S., Joosten, S.: Method Engineering and Meta-Modelling. Special Issue. Information and
Software Technology, 1996
[27] Braun, P., Broy, M., Houdek, F. Kirchmayr, M., Müller, M., Penzenstadler, B,, Pohl, K., Weyer, T.:
Entwicklung eines Leitfadens für das Requirements Engineering softwareintensiver Eingebetteter Sys-
teme. Technical Report. Technische Universität München, 2009.
[28] Gonzalez-Perez, C. and Henderson-Sellers, B.: A powertype-based metamodelling framework. Soft-
ware and Systems Modeling, Vol. 5, pp. 72-90, 2006.
[29] Mendez Fernandez, D. Wagner, S., Lochmann, K., Baumann, A.: Field Study on Requirements Engi-
neering Artefacts and Patterns. Proceedings of 14th International Conference on Evaluation and As-
sessment in Software Engineering, Staffordshire, United Kingdom, 2010.
... The need to create a classification scheme (ISO/IEC TR 11179-2:2019, 2019) for requirements that is both generic and adaptable has been identified more than 20 years ago (Hochmüller, 1997). Several approaches to classifying or modeling requirements have been created since then (e.g., Gorschek and Wohlin (2006);Méndez Fernández et al. (2010)). For instance, viewpoints can be used to classify requirements, considering perspectives of different stakeholders (Finkelstein et al., 1992;Sommerville and Sawyer, 1997). ...
... Epics, Features, and Stories are more specialized entity types of Backlog Item. Other terms for RIM are requirements metamodel, reference model, or artifact model (Méndez Fernández et al., 2011;Méndez Fernández et al., 2010). A RIM for agile enterprises focuses on backlog items to organize teams' tasks (Leffingwell, 2011). ...
Full-text available
In large-scale automotive companies, various requirements engineering (RE) practices are used across teams. RE practices manifest in Requirements Information Models (RIM) that define what concepts and information should be captured for requirements. Collaboration of practitioners from different parts of an organization is required to define a suitable RIM that balances support for diverse practices in individual teams with the alignment needed for a shared view and team support on system level. There exists no guidance for this challenging task. This paper presents a mixed methods study to examine the role of RIMs in balancing alignment and diversity of RE practices in four automotive companies. Our analysis is based on data from systems engineering tools, 11 semi-structured interviews, and a survey to validate findings and suggestions. We found that balancing alignment and diversity of RE practices is important to consider when defining RIMs. We further investigated enablers for this balance and actions that practitioners take to achieve it. From these factors, we derived and evaluated recommendations for managing RIMs in practice that take into account the lifecycle of requirements and allow for diverse practices across sub-disciplines in early development, while enforcing alignment of requirements that are close to release.
... The need to create a classification scheme (ISO/IEC TR 11179-2:2019, 2019) for requirements that is both generic and adaptable has been identified more than 20 years ago (Hochmüller, 1997). Several approaches to classifying or modeling requirements have been created since then (e.g., Gorschek and Wohlin (2006);Méndez Fernández et al. (2010)). For instance, viewpoints can be used to classify requirements, considering perspectives of different stakeholders (Finkelstein et al., 1992;Sommerville and Sawyer, 1997). ...
... Epics, Features, and Stories are more specialized entity types of Backlog Item. Other terms for RIM are requirements metamodel, reference model, or artifact model (Méndez Fernández et al., 2011;Méndez Fernández et al., 2010). A RIM for agile enterprises focuses on backlog items to organize teams' tasks (Leffingwell, 2011). ...
... Mendez et al. have contributed a reference artifact model for requirements engineering [5,75] based on their fundamental positioning on artifact orientation [76,77]. The AMDiRE approach constitutes a domain-agnostic reference for artifact types and serves the purpose requested by Femmer et al. [53] in that it can be tailored toward any industry context to model an artifact structure. ...
Full-text available
High-quality requirements minimize the risk of propagating defects to later stages of the software development life cycle. Achieving a sufficient level of quality is a major goal of requirements engineering. This requires a clear definition and understanding of requirements quality. Though recent publications make an effort at disentangling the complex concept of quality, the requirements quality research community lacks identity and clear structure which guides advances and puts new findings into an holistic perspective. In this research commentary, we contribute (1) a harmonized requirements quality theory organizing its core concepts, (2) an evaluation of the current state of requirements quality research, and (3) a research roadmap to guide advancements in the field. We show that requirements quality research focuses on normative rules and mostly fails to connect requirements quality to its impact on subsequent software development activities, impeding the relevance of the research. Adherence to the proposed requirements quality theory and following the outlined roadmap will be a step toward amending this gap.
... Examples of such models are the work of Broy [16] and Silva et al. [62]. Moreover, there are ontologies and meta-models to classify artefacts in specific software development areas (e.g., Idowu et al. [29] Mendez et al. [51], Zhao et al. [79], and Constantopoulos and Doerr [19]). ...
... The resulting artifact structure (also called infrastructure) supports communication between different roles and departments inside an organization, e.g., to which product platform a certain feature belongs, what requirements a certain feature consists of, what test cases that belong to a certain requirement, which release a certain requirement should be implemented in, or what artifacts patches that represent the implementation of a certain requirement. The communication schema should be altered dependent on the firms' needs and processes [43], e.g. to follow-up what requirements are contributed. In this study, we introduce an information meta-model that proposes how a set of repositories may be set up to support the above-mentioned communication and decision-making. ...
Full-text available
Open Source Software (OSS) ecosystems have reshaped the ways how software-intensive firms develop products and deliver value to customers. However, firms still need support for strategic product planning in terms of what to develop internally and what to share as OSS. Existing models accurately capture commoditization in software business, but lack operational support to decide what contribution strategy to employ in terms of what and when to contribute. This study proposes a Contribution Acceptance Process (CAP) model from which firms can adopt contribution strategies that align with product strategies and planning. In a design science influenced case study executed at Sony Mobile, the CAP model was iteratively developed in close collaboration with the firm's practitioners. The CAP model helps classify artifacts according to business impact and control complexity so firms may estimate and plan whether an artifact should be contributed or not. Further, an information meta-model is proposed that helps operationalize the CAP model at the organization. The CAP model provides an operational OI perspective on what firms involved in OSS ecosystems should share, by helping them motivate contributions through the creation of contribution strategies. The goal is to help maximize return on investment and sustain needed influence in OSS ecosystems.
... Roles define responsibilities for executing certain activities along with the skills, competences or experiences necessary for their execution [1]. Artifacts are abstractions of input, output or intermediate results [1], [33]. From our point of view, there are various levels of detail in agile practices. ...
Conference Paper
Agile methods require constant optimization of one's approach and leading to the adaptation of agile practices. These practices are also adapted when introducing them to companies and their software development teams due to organizational constraints. As a consequence of the widespread use of agile methods, we notice a high variety of their elements: Practices, roles, and artifacts. This multitude of agile practices, artifacts, and roles results in an unsystematic mixture. It leads to several questions: When is a practice a practice, and when is it a method or technique? This paper presents the tree of agile elements, a taxonomy of agile methods, based on the literature and guidelines of widely used agile methods. We describe a taxonomy of agile methods using terms and concepts of software engineering, in particular software process models. We aim to enable agile elements to be delimited, which should help companies, agile teams, and the research community gain a basic understanding of the interrelationships and dependencies of individual components of agile methods.
... Examples of such models are the work of Broy [14] and Silva et al. [60]. Moreover, there are ontologies and metamodels to classify artefacts in specific software development areas (e.g., Mendez et al. [50], Zhao et al. [78], and Constantopoulos and Doerr [17]). ...
Full-text available
Context: Developing software-intensive products or services involves utilising many artefacts that are either part of the offering or part of enabling their development. These artefacts, if valuable and used more than once by the development organisation, can be seen as assets such as test cases, code, requirements, and documentation. As such, assets are often reused, updated, and become a base or even necessity for product development and evolution over time, constituting an investment. Assets are not well understood beyond code-based ones in the area of technical debt. Thus most asset types are often overlooked, and their state, which is subject to degradation over time, has not been much studied. Method: To address the problem, we conducted industrial focus groups with five companies and a literature review. The results were analysed qualitatively and summarised in a taxonomy. Results: We created a structured, extendable taxonomy of assets. The taxonomy contains 57 assets. Conclusions: The taxonomy serves as foundations for defining and identifying assets as a concept. It provides the foundation for studies into asset degradation and subsequent asset management. The taxonomy also includes code-based assets and thus can be helpful in further research into investigating degradation and debt concepts.
In the process of development and maintenance of a software product, many things are created and used that are called software artefacts. Software artefacts are reused, changed, including their relationships, in the process of software product development and maintenance. The complexity and variety of software artefact relationships requires adequate means of their description and management. Software ecosystem research methodology can be used to investigate software artefacts in the context of a software product creation processes. The research subject will be a software artefacts ecosystem. Such an ecosystem is concerned with a more detailed level as compared to a software ecosystem. However, most of the approaches, methods and tools proposed in the software ecosystem research methodology can be also very well used at this level. In the article, the concept of a software artefacts ecosystem is proposed. Concept describes a generic model. According to the software ecosystem research methodology, it is the Cornerstone ecosystem type model, and consists of three actors—platform, software, and artefact. The SD model of the software artefacts ecosystem is described based on the generic model. The roles of the actors in the ecosystem are indicated, the relationships between actors are described. Application of the concept is illustrated by an example of case study of a programming style artefact. When styles (standards) are used, developer’s activities will be more efficient, software is easier to understand, and development and maintenance is cheaper. In a case study, based on the generic model of the software artefacts ecosystem, a declarative model of a programming style ecosystem has been developed. Described a three-level model of a programming style artefact. Tools and processes for creating and using a programming style artefact are developed and described.
Formal methods have provided approaches for investigating software engineering fundamentals and also have high potential to improve current practices in dependability assurance. In this article, we summarise known strengths and weaknesses of formal methods. From the perspective of the assurance of robots and autonomous systems (RAS), we highlight new opportunities for integrated formal methods and identify threats to the adoption of such methods. Based on these opportunities and threats, we develop an agenda for fundamental and empirical research on integrated formal methods and for successful transfer of validated research to RAS assurance. Furthermore, we outline our expectations on useful outcomes of such an agenda.
Full-text available
Requirements engineering (RE) builds a crucial part in software evolution. Nowadays, industries are more then ever facing the problem that the RE process is highly volatile due to the closeness to customer’s capabilities, to used process models and to produced specifications. The missing integration of RE into the development life cycle and the missing support of project-specific influences complicate the use of available RE approaches as a standardised, company-specific RE process that fits to the needs of individual (variable) project environments. In this report we provide an artefact-based RE reference model for business information systems. It serves as an orientation for producing precise specification documents being conformant to the reference model. Based on this reference model, we define a mechanism for a systematic and transparent customisation of the reference approach. The customisation mechanism defines how to customise the reference model at organisational environments concerning process-integration. This process integration is performed with the V-Modell XT, the standard development process for IT-projects within the German government. Furthermore we show how to customise it at project-levels according to variable project influences. We provide therefore a RE model that is highly customisable and deeply integrated into the development life cycle and prepare its application in real life projects. We lay the foundation for a systematic and integrated RE that fits to the needs variable project environments.
Software development projects are complex. The more complex a project is, the higher are the requirements related to the software development process. The implementation of a process is a great challenge. This, in part, has to do with human factors (acceptance, etc.) as the benefits of a formal development process might not be obvious immediately and it may take a while until the process becomes the lifeblood of a team. A crucial step towards implementing, enacting and enforcing a process is to provide tool support for the many activities the process asks for. Tool support is necessary to guarantee efficiency in the project, to do the housekeeping and to minimize the "overhead" of the process. This chapter describes challenges and options for supporting process models by tools. Furthermore it describes concrete samples and shows how tool chains can be created with commercial tools as well as with open source tools.
Not only since the introduction of the Unified Modeling Language and Model-Based Ar- chitecture, model-based development has been considered a central approach to deal with the increase in software complexity. Often, however, it is misunderstood as substitut- ing code-based by graphic-based implementation, adding platform-independence through code-generation. Applied in a broader understanding, model-based development is the central engineering paradigm allowing to tackle the increasing complexity of today’s software systems. Its core principles are adequate models focusing on domain-specfic core concepts and providing a concise interpretation as well as supporting mechanisms. Using these kind of models allows to improve the quality of the developed product as well as the efficiency of the development process. To elaborate these principles and thus to establish a general framework for model-based approaches, concept models - providing the concepts for the syntactic description of a sys- tem - and content models - providing the contents for the semantic interpretation - are introduced, explicating their characteristics and benefits in general, and illustrating their instantiation for the domain of reactive systems. For these classes of models, their applica- tion in a tool-based development process is demonstrated by providing operationalizations for the construction of concept models as well as for the analysis and synthesis for both classes of models. The potential of models for a development process is demonstrated by illustrating their implementation through model-based techniques for the analysis - via structured requirements elaboration and modular function composition – and design – via constraint-conformant construction and architectural refactoring – as well as test – via integrated test-case generation – of reactive systems.