Model-based framework with AI-augmented automation for cyber-physical systems
Sophisticated software is enabling increasingly advanced functionality in domains including industrial automation, healthcare, autonomous vehicles and smart grids. The growing complexity of the resulting cyber-physical systems (CPSs) poses challenges to continuous improvement throughout system design, development, testing, deployment, use and maintenance to ensure reliability, functionality and adaptability. The EU-funded AIDOaRt project developed an innovative model-based framework leveraging machine learning to support the automated continuous development of CPSs in large and complex industrial systems.
Integrating DevOps with model-driven engineering and AI
Over the last couple decades, the use of three innovative approaches supporting software development has grown rapidly: DevOps, model-driven engineering (MDE) and AIOps (AI for IT operations). DevOps is a collaborative software engineering approach in which development and operations teams work together, using automated software delivery pipelines to reduce cycle time and enable quick response to changes. MDE relies on models and transformation rules to simplify and automate the code generation process. AIOps uses data analytics, AI and machine learning approaches to enhance and automate IT operations. According to Gunnar Widforss, project coordinator of Mälardalen University: “AIDOaRt’s real novelty is the combination of the three paradigms DevOps, MDE and AI/machine learning. This allows us to observe and analyse collected data from both design-time and runtime artifacts and enhance the engineering process in rapid computer science-and-engineering cycles.”
Three toolsets with multiple components and capabilities
The AIDOaRt platform includes three main open source toolsets. AIDOaRt’s core toolset provides the generic capabilities of the AIDOaRt framework such as model handling and management and is used by the other toolsets. For example, the MOMoT tool is a framework that combines MDE and search-based software engineering to optimise model transformation. Its data engineering toolset supports collection, management and representation of data. The AI-augmented toolset includes components for AIOps engineering and AI-supported DevOps engineering. Overall, “AIDOaRT introduced 34 key innovations that distinctly advanced the state of the art,” notes Widforss.
Co-design and implementation in a plethora of use cases
AIDOaRt’s use case providers contributed specific scenarios or problems, defining the practical challenges and requirements and ensuring that the research was grounded in real-world needs. Its solution providers designed and implemented the technical or methodological solutions required to address the use cases. The project’s practical demonstrations of how these solutions can help automate modelling tasks when engineering large and complex CPSs – particularly during the design, development and testing phases – have resulted in numerous publications, easily accessible on the project website. “AIDOaRt has been demonstrated in heterogeneous domains and target applications including automotive, aerospace, maritime and railway. It has made a significant contribution to the creation of a cross-domain ecosystem of integrated AI-augmented solutions for system and software engineering support,” concludes Widforss. AIDOaRt’s unique integration of DevOps, MDE and AI/machine learning in its open source toolkit will support continuous development of complex CPSs while increasing its efficiency and effectiveness.
Keywords
AIDOaRt, AI, engineering, DevOps, software, MDE, CPS, machine learning, AIOps, model-driven engineering, continuous development, cyber-physical systems
