Towards a Generic Multidisciplinary Models Composition Tool

Anas Abouzahra, Ayoub Sabraoui and Karim Afdel

Laboratory of Computer Systems and Vision LabSIV, Faculty of Science, Ibn Zohr University, Agadir, Morocco

1 RESEARCH PROBLEM

Emergent engineering domains like healthcare

engineering, neural engineering, financial

engineering and many other domains get developed

more and more quickly, pushing the systems

complexity to an inexorable growth. In this

particular context, modeling and especially MDE

(Model Driven Engineering) approach (Mellor,

2003), plays an important role today. In fact,

modeling reduces the increasing complexity of

systems, mainly by reducing the gap between the

problems treatment and the technologies spaces

(France, 2007) (Shmidt, 2006). It provides domain-

specific software tools to non-programmer engineers

to take advantage of using computer computing

power, by building themselves specific softwares for

their domains.

A project’s complexity can be defined in terms

of variability and interdependency (Baccarini, 1987).

These two dimensions are particularly important and

cover a number of current industrial challenges.

When constructing a rocket for example, one has to

switch between thermal, mechanical, electrical, and

many more aspects. Each of these aspects is

complex by itself. But even more complex are the

numerous relations between these various sub-

systems. It seems obvious that the need to express

relations between multidisciplinary models is a

reality that we can no longer escape.

In this paper we present a generic

multidisciplinary models composition tool project; a

new contribution to address multidisciplinary MDE

development needs. We start in Section 2 by giving

a quick overview of the outline of the objectives.

Then we present a concise state of the art listing the

related works in Section 3. Section 4 describes our

methodology. After this, in Section 5, we present the

expected outcomes. Section 6 exposes the actual

stage of the research. The concluding section

summarizes the usefulness and provides future

perspectives of our work.

2 OUTLINE OF OBJECTIVES

When you want to develop an application with a

MDE approach, it became practically a necessity to

use tools provided by the Eclipse Modeling

Framework (EMF) project (EMF, 2015) or tools

around these technologies, especially, if you want to

respect the standards defined by the Object

Management Group (OMG). The overall goal of this

project is to propose a new solution to contribute to

open modeling tools based on the EMF to

multidisciplinary development context. Our

contribution will consist on three complementary

modules. The first one is called KM4M (Kernel

Metametamodel for Modellaborate). It is a small and

concise textual language for describing metamodels

accompanied by a rich text editor. It will provide

requisite expressions to define relations between

heterogeneous metamodels. These relations express

the potential interlinks between models conform to

these metamodels. The second module is called

CG4M (Code Generation for Modellaborate). It is a

code generation facility which is capable of

generating every needed Java interfaces and

implementations for all the classes in a metamodel

described by KM4M language, including methods to

instantiate and manage model interlinks. This

generation will be done on the fly or on demand in

connection with the metamodel editor. The last

module is called MR4M (Metamodels Repository

for Modellaborate). It is a metamodel repository that

communicates with the other two modules to

facilitate editing of metamodels and provides the

necessary information for the automatic code

generation. The three modules will work together in

order to be able to set up links between EMF

heterogeneous models (by heterogeneous we mean

conforming to distinct metamodels). All of that with

a synchronisation mechanism that diffuses models

modifications each time that a model has been

changed.

Abouzahra, A., Sabraoui, A. and Afdel, K.

Towards a Generic Multidisciplinary Models Composition Tool.

In Doctoral Consortium (DCMODELSWARD 2016), pages 9-15

3 STATE OF THE ART

The impact of MDE can be measured solely through

its actual applications in concrete use cases in

software industry. These measures are highly

correlated with the power and the limits of

technology available. Actually, the OMG has

defined some standards around the MDE: among

other, the Meta Object Facility (MOF 2.5). The

MOF 2.5 (MOF 2.5, 2015) is at the top layer of

modeling pyramid as a meta-metamodeling language

for defining metamodels. The MOF 2.5 plays

exactly the role that EBNF (a notation to define

grammars) (Pattis, 1980) plays for defining

programming language grammars. The OMG has

defined also the EMOF 2.5 (Essential MOF 2.5)

(MOF 2.5, 2015), a simplified version of the MOF

2.5. In parallel, The OMG has defined the XML

Metadata Interchange (XMI 2.5); a standard for

exchanging metadata (XMI 2.5, 2015). It is

thereafter the by default persistence format of

models conforming to MOF 2.5. In this wake, the

EMF project has provided Ecore (EMF, 2015); the

core metamodel at the heart of EMF that implements

the EMOF 2.5. The Ecore is today the main

reference implementation of the EMOF 2.5. The

EMF project provides in addition a set of tools and

runtime support to produce code generation from a

model specification described in XMI (Steinberg,

2009) (EMF, 2015). The EMF project became the de

facto set of tools to build applications according to

the MDE approach. However, EMF has shown clear

limits in terms of flexibility (Bagnato, 2014)

(Fouquet, 2012), scalability (Kolovos 2013)

(Barmpis, 2012) (Benelallam, 2014) and

extensibility (Garmendia, 2014), especially when

was actually put to the test on complex systems

context.

In order to overcome these limitations, a

substantial number of research works were carried

out, focusing essentially on building layers on top of

EMF. Model composition has been the subject of

great deal of projects. We could mention some of the

important ones: Epsilon Merging Language; EML

(Kolovos, 2009), Kompose (Fleurey, 2007), Atlas

Model Weaver; AMW (Didonet, 2009) and Virtual

EMF; VEMF (Clasen, 2011). The EML language is

a hybrid rule based language for merging models.

This approach proposes to merge homogeneous or

heterogeneous models through three categories of

rules (Match, Merge, and Transform). It reuses the

syntax and semantics of ETL (Epsilon transforming

language) (Kolovos, 2008) and extends it with

concepts specific to model merging. Kompose

implements a generic structural composition

operator that can be specialized to a particular

modeling language described by a metamodel.

The composition mechanism is done in two steps:

matching the model elements that describe the

same concepts, then merging those elements.

The

Atlas Model Weaver is a model composition

framework that uses model weaving and model

transformation to define and execute composition

operation. The model weaving captures the links

between model elements. Then, a transformation

produces the composed model. Virtual EMF is a

model composition approach based on the AMW

project and providing a virtualization mechanism

that offers a direct and transparent access to the

contributing models used in the composition

process. The weaving model becomes a virtual

model.

To enable a synthesis to assess the relevance of

each approach, we used the set of following criteria:

 Heterogeneity: it means that the solution is able

to provide composition of heterogeneous models.

 Number of Input Models: since we are

interested to multidisciplinary modeling context,

this criterion characterizes the possible limitation

of the number of input models.

 Symmetric/ Asymmetric: These two properties

distinguish symmetric approaches; where no

distinction is made between source models, from

asymmetric approaches; where base models play

the major role in the composition, when aspect

models provide weaving concepts.

 Synchronization: This criterion refers to the

ability of the approach to spread the changes

made in the source models into the composed

ones.

Table 1: Evaluation of composition approaches.

Approach

Criteria

EML Kompose

AMW VEMF

Heterogen Yes No

Yes Yes

Nb of Inputs

limit

Sym / Asym

S A

S S

Synch

No No

No Yes

Table “Tab. 1” provides an overview of

evaluating the approaches against the defined

criteria. Except Virtual EMF, the other approaches

rely on the use of a third model combined with some

data redundancy. In plus, they do not provide

automatic synchronization with source models.

DCMODELSWARD 2016 - Doctoral Consortium on Model-Driven Engineering and Software Development

Virtual EMF has presented a promising approach but

it remained at its prototyping stage. We will not

criticize here these approaches, since they have

demonstrated their usefulness in many use cases.

What we can note is that evaluation shows that they

are less adapted to a multidisciplinary development

context, where heterogeneous models must

collaborate.

4 METHODOLOGY

To illustrate the approach proposed in this paper, we

use an example in which two disciplines interact

together. It is about the calculation of employee

benefits according to the International Accounting

Standard Nineteen IAS 19 (IASB 19, 2004). This

example will be explained below in Section 6.

To prove our work we decide to follow a

methodology in five steps:

 Develop a first prototype of our case study using

conventional MDE tools (EMF models, XMI

serialization ... etc.).

 Analyse the advantages of using modeling

techniques to the development of a Domain

Specific Modeling Tool (DSMT) before showing

the limitations induced inter alia by the lack of a

powerful heterogeneous models composition

mechanism.

 Develop a prototype of our solution for

multidisciplinary models composition.

 Rebuild the first DSMT prototype with our

proposal.

 Finally, demonstrate the improvements that our

solution brings.

5 EXPECTED OUTCOME

The solution we intend to implement is expected to

provide a model composition mechanism capable to

link elements from heterogeneous models. This will

occur through three modules: KM4M, CG4M and

MR4M.

5.1 KM4M: A Rich Language for

Metamodels Specification

The MDE challenges to drive development of rich

domain specific tools that evolve into a complex

multidisciplinary context. Our contribution has been

designed to complete existing tools to provide a

multidisciplinary models composition mechanism.

Since EMOF 2.5 does not include concepts to

express relations between different metamodels, we

propose to overcome this limitation by extending the

EMOF 2.5 concepts by introducing new relation

concepts that can be expressed between metamodels.

Then, we propose a new language named KM4M

that allows specifying domain definition metamodels

and provides some relatively simple features to

describe their relationships. KM4M is therefore a

Domain Specific Language (DSL) to define

metamodels. The KM4M itself has its own

metamodel. It is a meta-metamodel, to which other

domain metamodels conform. This meta-metamodel

may be defined in KM4M, just like EBNF may be

defined in EBNF using only few lines. It uses

concepts like Class, Attribute, and Reference. It is

structurally close to EMOF 2.5 with additional

concepts to express relations between metamodels.

We plan to implement KM4M as an Eclipse plug-in.

It will be accompanied by a rich editor and will

work in collaboration with two other modules that

we present in the paragraphs to follow.

5.2 CG4M: An Automatic Code

Generation Module

The CG4M module will be a code generation facility

able to automatically produce all necessary code to

instantiate and enrich models. It will generate Java

interfaces and implementation classes for all the

classes defined by a metamodel, a factory and

package implementation class, and adapter classes

that adapt metamodel classes to manage inter-model

composition links. The goal is to be able to

instantiate composition links conforming to

relationships defined in the metamodels and use

them in a fully transparent way as if we were using

an intra-model link (links between elements of the

same model).

All entities of EMF models are subclasses of the

super class EObject which provide a set of methods

to access an entity. Among these are reflective

methods eSet() and eGet() to access its attributes and

references. CG4M will provide a specific

implementation of EObject class that allows using

those methods to navigate transparently a

composition link.

We plan to implement CG4M as an Eclipse

plug-in. It will produce code generation

automatically from metamodels written with

KM4M, on the fly by default, or by manual

activation if the user chooses a separate generation

mode.

Towards a Generic Multidisciplinary Models Composition Tool

5.3 MR4M: A Metamodels Repository

The MR4M module will be an Eclipse plug-in for

providing metamodels to other tools and services

dealing with models.

For each platform, we suppose that there is an

associated metadata repository defining the metadata

associated to this platform. Within the content of

such a platform, the metadata repository records all

available resources. In the context of a MDE

multidisciplinary development platform, a repository

of metamodels from several disciplines will make

available, in such a platform (in particular, if it

integrates the two previous modules: KM4M and

CG4M), all the necessary information to build their

metamodels and use composition links between

models.

6 STAGE OF THE RESEARCH

6.1 A First Prototype

As mentioned in the methodology section we begin

our work by implementing a first prototype using

conventional MDE Tools. The example of IAS19

calculation is a simple example representing a real

use case in which two disciplines interact together

creating a redundant use of data: Human Resource

Management and Actuarial Science. “Fig. 1” and

“Fig. 2” show two class diagrams representing two

short extracts from the two metamodels. The

diagrams were designed using Graphical Modeling

EcoreTools (EcoreTools, 2015).

Figure 1: Class diagram representing an excerpt of the

Actuarial study metamodel.

In collaboration with an Actuarial Cabinet

(Actuaria Global, Casablanca, Morocco) we tried to

implement a prototype of a DSMT dedicated to IAS

19 studies. We used classic modeling technologies

to achieve it. The actuarial calculation starts from a

data source (files, database...etc.) containing a

company's employees information. This is provided

by human resource service. The study ends with a

report listing estimates of provisions to ensure the

employee benefits at short, medium and long terms.

Under IAS 19, the staff of a company benefits inter-

alia of the three following benefits:

 Retirement Bonus (IFC).

 Longevity Pay.

 Illness Coverage.

The tool should provide the calculation of these

three benefits.

Figure 2: Class diagram representing an excerpt of the

Human Resource metamodel.

6.1.1 Technical Architecture

Developed on the top of an Eclipse RCP base (The

Eclipse project provides a complete set of tools for

developing general purpose applications named

Eclipse Rich Client Platform), the prototype is

implemented with Java and public libraries under

free license.

6.1.2 An MDE Approach

We applied an MDE approach to drive the

development of the first prototype. We implemented

a metamodel for each discipline (Human Resource

and Actuarial) using Ecore tools (EcoreTools, 2015).

Then we generated metamodels Java classes and

adapters using the EMF Codegen facility (EMF,

2015). We implemented Java classes for the

injection/extraction of data model employing POI

DCMODELSWARD 2016 - Doctoral Consortium on Model-Driven Engineering and Software Development

Apache Apis (Java API for Microsoft Documents)

(Apache POI, 2015). We designed ergonomic

interfaces using the Standard Widget Toolkit SWT

(SWT, 2015) and Eclipse JFACE UI toolkit (JFace,

2015) in order to provide editors for different steps

of the actuarial studies. The evaluation of benefits is

done using a model transformation implemented

using the Atlas Transformation Language ATL

(JOUAULT, 2005). It produces a model (result)

from which we can extract a textual report using a

model to text generation implemented by Java

classes.

6.1.3 Results & Discussion

This MDE approach gave us the means to

implements a rich DSMT for the actuarial IAS 19

studies. A high-level abstraction helped us to better

design the implemented prototype respecting a

successful separation of interests. It is particularly

important to separate the logical layer from the

application layer, the presentation layer from the

persistent layer and the data layer from that of their

transformations. Through the application of MDE

concepts, the prototype tool is composed of

maintainable components, and the result is a

modular application, more stable, more

understandable, more adaptable, more reusable,

more maintainable and easier to evolve.

Within this prototype tool, when the actuary

begins his study for a given company, he chooses a

specific advantage kind (IFC, Illness covertures,

Longevity pay …etc.). Then the tool creates an

instance of an empty model which is persisted in

XMI. Afterward, the tool guides the actuary to

complete this model using a succession of graphical

interfaces. The first step consists of importing a

population of individuals (active employees, retired

employees, widowers, orphans…etc.) from a data

source provided by the company (Excel files). Then

the actuary proceeds to the verification and

validation of data. Once he has finished, he creates

the needed demographic and financial assumptions.

After that, he configures its calculation methods and

finally runs the evaluation of benefits. The result is

the calculation of aggregate benefits consisting on

global figures, but also individual figures for each

individual in the population.

Unfortunately, despite all the advantages

mentioned previously, this tool suffers from certain

limits. We would have liked to be able to share a

population of individuals between several actuarial

studies. For each individual of this population, the

actuarial study should associate, after calculation,

some additional data (benefits). In some other cases,

during the data validation, the actuary may have to

modify a property's value of a given individual.

Thus, he would like that this change impacts all

studies implying this individual. Similar cases can

be invoked for sharing assumptions, calculation

methods ... etc.

Not finding a models composition mechanism

providing necessary agility to implement these

requirements, we choose to duplicate some

properties of the Individual Human Resource

metamodel class into the Actuarial metamodel

Individual class. Thereby, each time the actuary

imports the same employee data, for a new study, he

creates a redundancy of data. Similarly, the

operations of reusing a study configuration,

assumptions or calculation methods, share the same

inconvenient. In addition, to calculate other types of

benefits or those of coming years he can not directly

point some of his older configurations (present in the

older studies). If we have found a models

composition tool meeting the requirements of our

use case, we would have designed the prototype

architecture differently. We would have defined the

assumptions into a separate metamodel, as well as

for the calculation methods and so on. This would

have allowed a better separation of concerns.

Figure 3: Representation of instances of Human Resource

and Actuarial metamodels.

Let’s focus on the sharing of population

individuals for example. We can redefine our

metamodels with a class Population of Actuarial

metamodel that directly point the class Individual of

the Human Resource one. The “Fig. 3” illustrates

three instances of models conforming to these

metamodels (two instances of Actuarial models and

one insatnce of Human Resource model). We can

Towards a Generic Multidisciplinary Models Composition Tool

see that the IFC study evaluated at 2014 is based on

data of the ANWAR company employees without

duplicating them on its own instance. The actuarial

study of next year (2015) will reuse the same

company’s data. Furthermore, if the situation of an

individual of this company, in the meantime, has

been changed like if he has been transferred to

another subsidiary, the change is automatically taken

in account in the two IFC studies. Moreover, this can

be generalized in the same way for assumptions,

calculation methods…etc.

6.2 Solution's First Steps

We started the implementation of the first two

modules: KM4M and CG4M. To create the KM4M

language and its rich editor, we used the XText

framework (XText, 2015); a framework for

development of programming languages and

domain-specific languages. Then we use the XPand

framework (Xpand, 2015) to implement the CG4M

module.

The KM4M language defines a simple grammar

that translates the concepts of EMOF 2.5. In

addition, it introduces new concepts to express

relations between different metamodels; under three

possible forms: equivalence relation, specialization

relation, generalization relation. The first type is

used to reuse a class from a metamodel “A” in a

class of a metamodel “B” exactly as it is. The two

other types of relations allow reusing a class defined

in a metamodel “A” in a metamodel “B” by

expanding or reducing its properties.

Figure 4: An example of two metamodels definition using

the KM4M language and the specialisation relation

concept.

The “Fig. 4” shows an example of a

specialization relation applied to the use case

presented above. In this example the class

Population of the Actuarial metamodel reuses the

class Individual of the Human Resource metamodel

by expanding it with an additional attribute

(representing the IFC benefit value).

7 CONCLUSION

This work tries to contribute to MDE tools with a

new solution for multidisciplinary models

composition. We presented in this paper a

summarized view of our research work, the expected

outcome and the stage of progress. Through a

concrete use case, we have shown the interest of

such a solution. We intend to continue the

implementation of the three modules: KM4M,

GM4M and MR4M. We then plan to prove our

approach by considering the use case again and

rebuild the first prototype using our proposition.

Finally, we will compare the two prototypes by

measuring the progress accomplished and assess the

road which lies ahead. However, we are aware that

our solution provides a model composition

mechanism that overlay EMF implementation of

EMOF 2.5. Therefore, the composed models made

by our solution will still correlated with the use of

our framework. While an extension of the concepts

of EMOF 2.5 in order to provide model composition

will allow a direct implementing of model

composition at the EMF level. This solution could

be shared among all the tools that are based on EMF

at by the same time enabling interoperability

between them.

REFERENCES

Mellor, S. J., Clark, A. N., Futagami, T., 2003. Model-

Driven Development. In Computer (Vol. 20, no. 5, pp.

14-18). IEEE Computer Society.

France, R. Rump, B., 2007. Model-driven development of

complex software: a research roadmap. In Proceeding

of FOSE’07: 2007 Future of Software Engineering

(pp. 37-54). IEEE Computer Society.

Shmidt, D.C., 2006. Guest editor's Introduction: model-

driven engineering. In Computer (Vol. 39, no. 2, pp.

25–31). IEEE Computer Society.

Baccarini, D., 1996. The concept of project complexity - a

review, In International Journal of Project

Management, (pp. 201-204).

EMF, Eclipse Modeling Framework, 2015, url:

https://eclipse.org/modeling/emf/

DCMODELSWARD 2016 - Doctoral Consortium on Model-Driven Engineering and Software Development

MOF 2.5 Core Specification, OMG Document

formal/2015-06-05, 2015, url: http://www.omg.org/

spec/MOF/2.5/PDF

Pattis, R.E., 1980. EBNF: A Notation to Describe Syntax.

url : http: //www.cs.cmu.edu/~pattis/misc/ebnf.pdf.

XMI 2.5, XML Metadata Interchange, Version 2.5.1,

OMG document formal/2015-06-07, 2015, url:

http://www.omg.org/spec/XMI/2.5.1/PDF

Steinberg, D., Budinsky, F., Paternostro, M., Merks, E.,

2009. EMF – Eclipse Modeling Framework 2.0,

Addison-Wesley Professional, 2

edition.

Bagnato, A., Malo, P., Trujillo ,S., Mendialdua, X., De

Carlos, X., Brosse, E., Sadovykh, A., 2014. Flexible

and Scalable Modelling in the MONDO Project:

Industrial Case Studies. In Proceedings of the 2014

Extreme Modeling Workshop (pp. 42–51). ACM.

Fouquet, F., Nain, G., Morin, B., Daubert, E., Barais O,

Plouzeau, N., Jézéquel, J., 2012. An Eclipse

Modelling Framework Alternative to Meet the

Models@Runtime Requirements. In Model Driven

Engineering Languages and Systems. Springer.

Kolovos, D., Rose, M., Matragkas, N., Paige, F., Guerra,

E., Cuadrado, J., De Lara, J., Ráth, I., Varró, D., Tisi,

M., Cabot, J., 2013. A research roadmap towards

achieving scalability in model driven engineering. In

Proceedings of the Workshop on Scalability in Model

Driven Engineering (BigMDE '13). ACM.

Barmpis, K., Kolovos, D.S.,2012. Comparative analysis of

data persistence technologies for large-scale models.

In Proceedings of the 2012 Extreme Modeling

Workshop (pp. 33–38). ACM.

Benelallam, A., G´omez, A., Suny´e, G., Tisi, M., Launay,

D., 2014. Neo4EMF, a scalable persistence layer for

EMF models. In Modelling Foundations and

Applications, of Lecture Notes in Computer Science

(Vol. 8569, pp. 230-241). Springer.

Garmendia, A., Guerra, E., Kolovos, D., De Lara, J., 2014.

EMF Spliter: A Structured Approach to EMF

Modularity. In Proceedings of the 2014 Extreme

Modeling Workshop (pp. 22–31). ACM.

Kolovos, D., 2009. Establishing Correspondences between

Models with the Epsilon Comparison Language. In

Proceedings of the 5th European Conference on

Model Driven Architecture - Foundations and

Applications (pp. 146-157). Springer.

Kolovos, D., Paige, F., Polack, F., 2008. The epsilon

transformation language. In Theory and practice of

model transformations: First International

Conference, ICM (pp. 46-60). Springer.

Fleurey, F,. Baudry, B., France, R., Ghosh, S., 2007. A

Generic Approach for Automatic Model Composition.

In H. Giese, MoDELS Workshops, of Lecture Notes in

Computer Science (Vol. 5002, pp. 7-15). Springer.

Didonet Del Fabro, M., Valduriez, P., 2009. Towards the

Efficient Development of Model Transformations

using Model Weaving and Matching Transformations.

In Software and Systems Modeling (Vol. 8, No. 3, pp.

305–324). Springer.

Clasen, C., Jouault, F., Cabot, J., 2011. Virtual

Composition of EMF Models. In 7èmes Journées sur

l'Ingénierie Dirigée par les Modèles (IDM 2011).

unpublished.

IASB, IAS 19 amendments, 2004. Employee benefits :

actuarial gains and losses, group plans and disclosures.

EcoreTools, 2015, url: http://www.eclipse.org/ecoretools/

Apache POI, The Java API for Microsoft Documents,

2015, url: https://poi.apache.org/

SWT,The Eclipse Standart Widget Toolkit, 2015, url:

https://www.eclipse.org/swt/

JFACE, The JFace UI framework, 2015,

http://help.eclipse.org/juno/index.jsp?topic=/org.eclips

e.platform.doc.isv/guide/jface.htm

JOUAULT, F., KURTEV, I., 2005. Transforming Models

with ATL. In J.-M. BRUEL (Ed.), MoDELS'05

Proceedings of the 2005 international conference on

Satellite Events at the MoDELS (Lecture Notes in

Computer Science ed., Vol. 3844, pp. 128-138).

Springer-Verlag Berlin, Heidelberg.

XText, 2015, url : http://www.eclipse.org/Xtext/

Xpand, Eclipse Model to Text (M2T) Xpand , 2015, url :

http://www.eclipse.org/modeling/m2t/

Towards a Generic Multidisciplinary Models Composition Tool