=Paper=
{{Paper
|id=Vol-3144/RP-paper14
|storemode=property
|title=Towards AIDOaRt Objectives via Joint Model-based Architectural Effort
|pdfUrl=https://ceur-ws.org/Vol-3144/RP-paper14.pdf
|volume=Vol-3144
|authors=Bilal Said,Andrey Sadovykh,Brosse,Alessandra Bagnato
|dblpUrl=https://dblp.org/rec/conf/rcis/SaidSBB22
}}
==Towards AIDOaRt Objectives via Joint Model-based Architectural Effort==
https://ceur-ws.org/Vol-3144/RP-paper14.pdf
Towards AIDOaRt Objectives via Joint Model-based
Architectural Effort
Bilal SAID1, Andrey SADOVYKH 1,2, Etienne BROSSE 1 and Alessandra BAGNATO 1
1
Softeam, Docaposte, Paris & Nantes, France
2
Innopolis University, Innopolis, Russia
Abstract
This paper outlines the AIDOaRt project (AI-augmented automation supporting modelling,
coding, testing, monitoring and continuous development in Cyber-Physical Systems)
objectives, the status, and current achievements. In particular we briefly outline one of its
notable results so far, a model-based requirements engineering process that the project adopted
to cope with the challenges of producing coherent joint technical results based on contributions
of 32 partners. We shortly overview the process and give reference to the complete process
that has been already applied and evaluated in several European Research projects. That way
we intend to share the best practices in reaching ambitious objectives as targeted by the
European research agenda. The project has received funding by the European Union KDT JU
Key Digital Technologies Joint Undertaking, and it began in April 2021 for a period of 36
months. The Research Challenges in Information Science conference topics covered by the
paper are the following: Information Systems Transformations, Model-Driven Engineering,
User-Centered Design, Method Engineering
Keywords 1
AIDOaRt, KDT, European Research, Requirements Integration Process, Model-based
Requirements Engineering
1. Introduction
AIDOaRt2 [1] is a KDT JU Key Digital Technologies Joint Undertaking project, and it aims at
supporting systems engineering and continuous delivery activities, namely requirements engineering,
modelling, coding, testing, deployment, and monitoring, with AI-augmented, automated Model-
Based Engineering (MBE) and Development Operations (DevOps). To achieve this goal, AIDOaRt
proposes a model-based framework (cf. Figure 1) that specifies proper methods and tools to enable
design and run-time data collection, ingestion, and analysis to provide tailored and efficient AI/ML
solutions that are to be integrated and evaluated on concrete industrial case studies involving various
Cyber-Physical Systems (CPS).
AIDOaRt aims at fulfilling three main objectives [1]: (1) to provide a model-based framework for
continuous engineering of CPS, benefiting from AI-augmented automation of MDE, (2) to enhance
DevOps processes by integrating AI/ML techniques in their tool chains, (3) to assist the analysis of both
historical and real-time data in combination with design information in order to enhance the overall
engineering process. To measure the fulfilment of these objectives, the consortium defined several
ambitious Key Performance Indicators (KPI) that are evaluated at the level of each case study, as well
as the overall project.
Joint Proceedings of RCIS 2022 Workshops and Research Projects Track, May 17-20, 2022, Barcelona, Spain
EMAIL: bilal.said@softeam.fr (A. 1); andrey.sadovykh@softeam.fr, a.sadovykh@innopolis.ru (A. 2); etienne.brosse@softeam.fr
(A. 3); alessandra.bagnato@softeam.fr (A. 4)
ORCID: 0000-0003-2259-6063 (A. 1); 0000-0003-2384-5447 (A. 2); 0000-0002-3781-4804 (A.3); 0000-0003-2675-0953 (A. 4)
© 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License
Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
2
https://www.aidoart.eu/
The main KPIs measure the reduction of time and effort needed to detect design problems and
system deviation from design, the increase in the number of automated and enhanced DevOps phases
and the increase in the number of data sources used to enable the AI/ML based optimization decisions
in the systems.
Figure 1 AIDOaRt Proposed Framework (inspired from [1])
The project involves 32 industrial and academic partners from different European countries. 10 end-
user partners provide 15 industrial case studies capturing CPS of varying complexity levels and in
different application domains, ranging from automotive to software and communication systems.
Partners have identified in these case studies a quite varied list of 128 functional and data requirements.
The requirements largely differ in terms of abstraction levels, broadness, and coverage. We
conducted a considerable analysis by separating requirements areas and partner’s contributions in five
key SE/DevOps activities - Requirements Engineering, Modeling, Coding, Testing and Monitoring (cf.
Figure 1).
Each category was elicited by a separate working group led by experts in the related domain.
This yielded a set of 30 generic, common and high-level requirements. On a more technical level,
research and industrial partners propose a set of 57 candidate solutions – building blocks to implement
planned case studies. This way the project addresses a broad area of industrial requirements with regards
to engineering data-centric, AI-augmented system in a model-based way. The requirements and
architecture for AIDOaRt must capture 113 services to be offered by proposed solutions, with
input/output data, technical and deployment constraints.
Given the large variety of case study requirements and solutions capabilities, there is an obvious
need for a common framework architecture to enable collaboration between the various partners,
namely among case study and solution providers.
We intended to benefit from our prior experience in previous similar large projects, namely the
predecessor ECSEL project “MegaModelling at Runtime – Scalable model-based framework for
continuous development and runtime validation of complex systems”, or shortly MegaM@Rt2 [2]. The
MegaM@Rt2 Framework proposed several interconnected tool sets for system engineering, modeling
and traceability management as well as runtime analysis.
Those tool sets are interconnected to achieve the goal of linking system models with the runtime
analysis of large-scale industrial systems [2]. A similar approach has been followed in AIDOaRt to
propose a modular architecture with key interconnected component capabilities playing different yet
complementary roles and forming together the AIDOaRt Framework. The notable contribution of the
AIDOaRt approach is specific focus on the specification of data-centric and AI-augmented capabilities
to be interfaced with the MBE and DevOps components.
The nature of our AIDOaRt project, in terms of number of partners and variety in requirements and
proposed solutions calls for the adoption of an agile methodology to gradually define the architecture
and to incrementally collect, refine and clarify the requirements, and map them to candidate solutions.
In fact, as the work in the project progresses, discussions and in-depth negotiations are taking place
between case study providers and solution providers. By gradually refining the requirements in a short
iteration we managed to obtain more precise and concrete requirements that in their turn drive better
choices for selecting candidate solutions. This agile methodology makes it possible to integrate the
newly reported updates in requirements and solutions descriptions from all partners in a flexible, simple,
and traceable fashion (cf. next section). Furthermore, this model-based approach allows us to
automatically generate several significant project deliverables documents with content collected and
consolidated from all the 32 partners in a collaborative, distributed and seamless manner.
This paper presents the proposed methodology used to elaborate the AIDOaRt architecture and
consolidate the partners requirements and solutions. It is destined to share our experience with the larger
community of researchers and industries who are interested in managing the complete requirements
engineering process of large research projects.
2. Application of the Model Based Requirement Engineering (MBRE)
Approach in AIDOaRt
Softeam has a substantial experience of large collaborative research projects where we played a role
of technical lead. In several projects (DataBio3, REVaMP24, MegaM@Rt [2]), we noticed similar
needs and patterns for Requirements Engineering – such as need for integration of contribution of
multiple partners, need for traceability between requirements and architecture as well as
similarities in the architectural concepts to be captured – tool components, interfaces, services,
deployment constraints.
That information had to be captured in a coherent and homogeneous way in multiple
project deliverables. Softeam is a renowned innovator in the MBE world – we consistently
promote model centered engineering principles since 2000s and offer Modelio workbench. In the
above-mentioned research projects, we proposed model-based tool with the goal to address the RE
challenges.
In this context, we explored several modelling standards such as TOGAF/ArchiMate ([3], [4]),
UML [5], and SysML [6]. In the final version, a very small subset of UML has been chosen and
proved to be very efficient while providing great expressiveness and flexibility.
These results are validated in [7] and [8] where partners provided an encouraging feedback on the
approach features such as scalability (e.g. large number of handled requirements), heterogeneity (e.g.
partners with different types of RE background), traceability (e.g. from the requirements to the
software components), and automation (e.g. requirement documentation generation) [7].
As many AIDOaRt partners had previously experience with UML, Modelio [9] and MBRE,
the project technical leaders have naturally selected this approach for coordinating requirements
and architectural efforts. The flexibility has been proven by an ability of the approach to
accommodate new needs for concepts such as data requirements and solution architecture. Below,
we outline several notable details about this approach.
Any engineering effort requires an appropriate process. In AIDOaRt, we opted for an agile
approach with short interactions proposing to edit a joint model, as shown in Figure 2.
To accelerate the requirements gathering, we started from simple text-based tabular forms
that require no prior knowledge or expertise in modelling or requirements engineering. We then
verified, corrected, and auto imported the first iteration of requirements by partners into the
modelling tool. In parallel, we dedicated a special effort in educating partners about MBRE and
Modelio peculiarities.
After this initial phase, we iterated with short cycles of micro-tasks design and
implementation. Every couple of weeks, we collected additional model content, e.g., additional
descriptions on some requirements, new types of requirements or traceability mappings between the
various elements of the model. That way, week after week, by small steps, all together we have
iteratively enriched our model following the evolving needs of the project technical coordinators for
requirements and architectural information.
3
https://www.databio.eu/
4
https://www.revamp2-project.eu/
This incremental process allows us to continuously update the data collected by our partners at any
stage of the deliverable writing, until its internal review and final submission.
Figure 2 AIDOaRt's Agile Iterative MBRE Approach
3. Applying Modelio SaaS
The current version of the presented agile MBRE process is supported by Modelio SaaS [9], a cloud-
based requirements and systems modeling tool developed by Softeam. Instead of spreading effort
in editing bits and pieces of architecture numerous documents, partners work collaboratively on a
single unique model representing AIDOaRt architecture including all the conceptual elements
such as requirements and solution description. This model is shared through a repository where
partners can connect and commit changes.
Multiple project deliverables are automatically created and updated with specifically
developed document generators that export various architectural views. This shared “live” model
represents a single source of truth for all the partners since this model is a unique place where the
updates have to be committed. The model lifecycle covers all stages of the project from requirement
inception to their validation in case study scenarios.
Meanwhile, the automated document generation guarantees a sound, traceable and up-to-date
overview of AIDOaRt’s architecture, case studies requirements and their fulfillment by the proposed
solutions until the end of the project. It is worth noting that a significant number of project-wide
deliverables have been fully created from Modelio.
To illustrate the scalability and usefulness of the approach, we would like to provide several metrics.
Among the 76 users representing all project partners who are currently registered users in the Modelio
repository, 60% are active users with multiple recurrent contributions. This indicates the high
engagement of the partners and their adoption of our proposed approach. After ten months, i.e., almost
one third of the project duration, MBRE has helped to elicit and detail 455 requirements, and a total of
6202 model elements in a raw of 1572 individual commits. This reflects the high scalability of the
approach and its suitability for such large projects.
At the current stage, 3 major project deliverables5 have been released accounting more than 200
generated pages each. The deliverables are automatically generated from the Modelio model with
textual descriptions, various diagrams, and traceability matrices representing different aspects of
AIDOaRt architecture. This automated generation allowed us to obtain a consistent document at every
moment in time regardless of the heterogeneity and timeliness of the contributions. Furthermore, it
allows to capture the most up-to-date content as provided by the partners, which continued to flow into
the shared repository until the last moment before the deliverables’ submission. Any other manual
document aggregation approach would have resulted in considerable delays and potential difficult-to-
handle inconsistencies.
5
https://www.aidoart.eu/aidoart/results/deliverables
In conclusion, the MBRE approach proved to be beneficial since it allowed us faster integration of
various contributions. In the beginning we experienced a brief “learning curve” period when partners
had to install Modelio and get familiar with requirements modeling approach.
Through the iterations of our agile process the partners learned to appreciate the approach that
helped us to drastically improve productivity and easily integrate new features into the model. The
documents were generated on the go without any troubles for managing consistency of the documents
and formatting. At the moment of this writing, three new deliverables are being prepared, and two are
planned for the next months. All of these deliverables will be also auto generated from Modelio with
content being iteratively collected using the agile MBRE approach.
4. Conclusion and Future Work
This paper presents the AIDOaRt project objectives. AIDOaRt builds a framework for AI-
augmented model-based engineering. The project is in the first very important stage i.e., scoping of
the project by analyzing the case study requirements and matching them with consortium
capabilities. During this phase, the project produced a set of fundamental artifacts including the
requirements analysis and the architecture of proposed solutions. We applied a Model-Based
Requirements Engineering approach that helped us to bootstrap collaboration in the project and
produce an integrated view of the project objectives in terms case study requirements and tool
capabilities. This approach enabled us to interactively analyze various aspects of the project and
create coherent and consistent reports with document generation.
For the future, the project intends to develop a great number of engineering solutions to address
the case studies and research challenges. The approach we apply will further help us in validation
and demonstration activities e.g., to specify the roadmap, KPIs and keep traceability of requirements.
Acknowledgements
Special thanks to all AIDOaRt consortium members who have embraced our approach and closely
collaborated during the implementation of the MBRE process. We would like to thank in particular
the AIDOaRt work package leaders for managing and orchestrating the requirements gathering
process in AIDOaRt, and for their valuable input on the specification of the AIDOaRt Architecture.
This research work has received funding through the AIDOaRt project from the ECSEL
Joint Undertaking (JU) under grant agreement No 101007350. The JU receives support from the
European Union’s Horizon 2020 research and innovation program.
References
[1] R. Eramo et al., “AIDOaRt: AI-augmented Automation for DevOps, a Model-based Framework
for Continuous Development in Cyber-Physical Systems,” in 2021 24th Euromicro Conference on
Digital System Design (DSD), Sep. 2021, pp. 303–310. doi: 10.1109/DSD53832.2021.00053.
[2] W. Afzal et al., “The MegaM@Rt2 ECSEL project: MegaModelling at Runtime – Scalable model-
based framework for continuous development and runtime validation of complex systems,”
Microprocess. Microsyst., vol. 61, pp. 86–95, Sep. 2018, doi: 10.1016/j.micpro.2018.05.010.
[3] “TOGAF: The Open Group Entreprise Architecture Framework,” The Open Group Website, May
04, 2018. https://www.opengroup.org/togaf (accessed Dec. 10, 2021).
[4] “The ArchiMate® Enterprise Architecture Modeling Language,” The Open Group Website, Oct.
12, 2018. https://www.opengroup.org/archimate-forum/archimate-overview (accessed Dec. 10,
2021).
[5] “OMG Unified Modeling Language.” https://www.uml.org/ (accessed Dec. 10, 2021).
[6] “OMG Systems Modeling Language (SysML).” https://www.omgsysml.org/ (accessed Dec. 10,
2021).
[7] A. Sadovykh, D. Truscan, and H. Bruneliere, “Applying Model-based Requirements Engineering
in Three Large European Collaborative Projects: An Experience Report,” in 2021 IEEE 29th
International Requirements Engineering Conference (RE), Sep. 2021, pp. 367–377. doi:
10.1109/RE51729.2021.00040.
[8] A. Sadovykh et al., “Model-Based System Engineering in Practice: Document Generation-
MegaM@ Rt2 Project Experience,” in Proceedings of the 14th Central and Eastern European
Software Engineering Conference Russia, 2018, p. 9.
[9] P. Desfray, “Model repositories at the enterprises and systems scale: The Modelio Constellation
solution,” in 2015 International Conference on Information Systems Security and Privacy
(ICISSP), Feb. 2015, p. IS-17-IS-17.