-
As for many other comp le x systems, designing an airc raft is not an easy task; be it the
new version of an existing model or c reating fro m the scratch. Generally, the stage
following the collect of customer needs is the require ments definit ion, starting by
elic itation and analysis. This process continues with the cascading of require ments
that often is a mirror o f the product breakdown structure. Then, for each system level,
require ments set are used to define the system ele ments. The other require ments are
allocated to sub-system at the lower leve ls and the process continuing till the
development of co mponents. The ma rket driven incre mental p roduct development and
delivery (re lease) is becoming increasingly co mmonp lace in software industry [
3
].
Incre mental product development is planned and e xecuted with the goal o f de livering
an optima l subset of require ments in a certain release (version of a product that is
distributed to customers) [
2
]. In pa rallel, ‘Agile ’ methodologies are an alternative to
traditional sequential development, addressing unpredictability response through
incre mental and iterative work cadences, known as sprints [
4
]. Incre mental Design
Processes is a pro mising e me rging discipline related to, but not specifically a subset of,
the ma rket driven incre mental product development and ag ile methodologies. These
processes combine both approaches, trying to merge the agility methods and the
incre mental process to address the specificity of designing based variants products. A
system design process could be said to be “ incre mental design process” if he is based
on the use of high quality generic require ments instantiated into specific require ments
dedicated to the definition of variants or increments of a system.
Our interest is to identify whether the incre mental design process could be applied for
the development of mult iple variants of comple x products (based on product line
engineering), considering the current methods for creating high quality generic
require ments. But idea of using incre mental design processes raise s challenges that
could be interesting for the community of requirements engineering researchers.
2
Problem statement and experiences in industry
Our p roble m lies in the way we could formu late the high quality system require ments
to be added in the incre mental design processes. As the sources of the require ments
vary it is not surprising that require ments come in different shapes and forms, at mu
ltiple leve ls of abstraction, and described on varying levels of refine ment [
1
].
Require ments are specified for many diffe rent purposes and fro m many d ifferent
engineering activit ies [
5
]. Du ring authoring, natural language rema ins a universal means
of e xp ressing require ments and studies indicates that 89% of engineers [
7
] shown
their preference to use of natural language require ments. However a mbiguity or
vagueness of require ments are the two ma in proble ms arising fro m the over-fle xib ility
of the natural language [28], ma king their interpretation challenging fo r any natural
language processing systems to reasonably understand the subject-matter.
The require ment writer in many occasions is skewed by their own personal e
xperiences hence semantics, vocabulary and terms differ wide ly fro m one person to anot
her as illustrated by Dickerson [
6
]. Not much research has investigated whether
different domains need different kinds of semantic tools displaying different kinds of
semantic relat ions. To address the challenge of writ ing the requirements right, we have
identified two main industrial problems.
2.1
Poor quality of requirements while authoring
Fro m the statistic survey established by Fanmuy and Foughali [
7
], it was found that
the most common leading defects in the natural language require ments falls under:
semantic contradiction, not verifiable , not complete, a mbiguous, not understandable
and not precise enough. On the conclusion of the survey, it was stated that the pro
ble ms still persist despite the use of several require ment engineering approach like
writing ‘SMART ’ require ments right fro m the very first attempt. Dedicated to assist
the system engineers, this approach leads to eliminate unnecessary informat ion in
requirements, to improve readability (i.e. text length, number of punctuation marks,
etc.) and reduce complexity. Another approach: the formal notation and graphical
representation based on models albeit offers alternative means to natural language.
Go rschek proposes a Requirement Abstraction Model [
1
]. Ho wever, these approaches
remain not so convenient to cover wide range of concepts, to manage co mpliance and
to address the needs creativity of system engineers. In practice, the requirements do
not exhib it all the acceptable characteristics of good requirements. As an example,
this bad requirement “New and modified air distribution components shall be
designed to minimize noise levels ” should be replaced by “New and modified air
distribution components installed in cabin areas shall emit less than 30 dB(A) of acoustic
noise.” The consequence of these mistakes is felt during all downstream act ivities
such as architecting, design, imp lementation, and testing. Organizations struggle to
bring consistency to their project and 47% of unsuccessful projects fail to meet goals
due to poor requirements management [
20
].
2.2
The problems of requirements verification
Requirement Verificat ion is about verifying sufficiently early in the develop ment
process whether the requirements have sufficient quality (i.e. they are well-formed
according to the ISO/IEC/IEEE 29148 standard) to avoid many negative impacts
subsequent activities. When eventually discovered, these defects will be significantly
more expensive and take mo re t ime to fix them. The following picture shows the cost
to extract defects.
Industrial projects handle up to thousands of requirements where human based verif
ication process during peer rev iews beco mes extremely tedious , time consuming and
expensive for the organization.
3
Proposed research activities
To address the hereunder issues, the following research tasks were identified.
3.1
Boilerplate for syntactic analysis of requirements
First of all, improve ment of require ments quality is given through a better structuring
of require ments sentences. This structuring goes through the definition of models
allo wing the creat ion of one or several types of require ments. To do so Mavin
considers a simple and effic ient set of sentence structures to improve drastically the quality
of require ments [
8
]. Each sentence structure can be based on full sentence models as
proposed by Al-Safadi [
9
] and in the Cesar pro ject [
10
]. This study proves the interest
to use sentences model based on predefined (frozen) or progressive, adaptive stru
ctures to support the authoring of require ments . This structure is generally called
‘patterns’ or ‘boilerplate’. The concept of using boilerp lates for writ ing statements of
require ments is quite simp le: choose an appropriate predefined pattern, and fill in the
gaps. Each statement of requirement is then based on a boilerp late where the selected
attributes have specific terms. Exa mple of boile rplate: The <system> shall be able to
<capability_verb> at a ma ximu m rate of at least <quantity> times per <time unit>.
The corresponding require ment could be: ‘The light shall be ab le to flash at a ma
ximu m rate of at least 5 t imes per second.’ The benefits of using boilerplates to
structure sentence for the syntactic analysis of require ments are nume rous. The most
important ones are an aid in the articulation of sentence, a uniformity of language and a
one-stop control over expression.
3.2
Ontologies for Semantic analysis of Requirements
The improve ment of require ments is also possible thanks to an analysis of their
meaning. Indeed, even if a require ment could be correctly written with boilerp late, this
solution does ensure neither its intrinsic quality (i.e .: what is the require ment mea
ning?) nor its global quality (is the require ment redundant, contradictory or similar
with regard to the other require ments?). The consistency quality of a require ment may
rely on a semantic analysis of its sentences. To carry out this analysis we use
outcomes given in the Cesar project [
10
], by Kof [
11
] and Jureta [
12
] wh ich have already
tackled these issues through different approaches, in particular with a domain
ontology (like in some of the early works by Lin [
13
] Yu [
14
] and Cadih lac [
16
]). Fro m a ll
available definitions of ontologies , the best suitable for our purpose is given by
Gruber [
15
]. In simp le terms, ontology represents a doma in of knowledge. To build
and manage our ontologies, diffe rent tools exist on the market. Our aim is not to ma ke
a benchmark of these tools but to define a method to apply the m. The process
supporting this method of do main ontology definition is summarized in the ne xt figure.
Starting fro m the collection of corpus and require ments sets, the global process is
based on five ma in sub-processes (the terminology, thesaurus, pattern definition,
pattern matching and formalization) grouped together into two main phases.
The result of this process is double; a domain terms-based ontology (i.e. light
ontology) and a set of structured formu lation of sentences (i.e. bo ilerplates). Both are lin ked
together, therefore the ontology is mapped on the boilerplates that support the analysis
of quality. Indeed, only the comb ination of structured sentences and well known te
rminology ensures the definition of better quality requirements.
Future directions focus on e xtending our understanding of boilerp lates and ontology
techniques imple mented for aircraft industry. Currently, we are in the process of
constructive generic method to define clearly the process of ontology and boilerplate
creations. So far we identified different require ment error ta xono my, se mantic based
require ment engineering concepts [
17
], formal e xpression language used in ontology
[
18
], ru les to construct domain ontology and issues co ncerning maintainability and
interoperability of ontology [
19
]. Practica l work covers the construction of th esaurus
and its rules for ontology. Ne xt activity will be to apply the process dedicated to the
imple mentation and use of the boilerplates and ontology. However, the next proble m
will be the comple xity of incre mental design processes that requires the creation of
high quality generic require ments. This drives to research issues: how ensuring
require ments consistent and complete in incre mental design processes? Which methods
and techniques drive to require ments quality for product line proces ses? Rather than
affirming at this stage, what shall be done in the ne xt years down the line, we can only
expect some ground breaking results thanks to new research activities.