CYCLOPS

Engineers of Cyber-Physical Systems of Systems (CPSoS) need to construct models of the systems they intend to build in order to capture, test and validate their ideas with other stakeholders before embarking on a long and costly production process. To accommodate the needs of a wide range of systems, CPS-geared languages such as Simulink and Modelica are large and verbose, and are supported by heavyweight (and often expensive) modelling and model processing (e.g. simulation, verification, code generation) tools. These monolithic approaches to CPS contrast with the heterogeneity of new CPSoS. On one side, the efficiency in developing specific component systems of the CPSoS may be greatly increased by using ad-hoc development environments, methods and runtime support. Instead, using general-purpose CPS languages and tools effectively for all systems requires expertise unavailable to many European SMEs, can be error-prone, and significantly increase the engineering time and cost. Moreover, each CPS system requires in some degree an intelligent behavior, that may be obtained by relying on machine learning libraries, external to the CPS toolbox. External tools and libraries generally appear as black boxes to the CPS model, hampering meaningful analysis, simulation, verification and optimization. Because of these software engineering problems, despite the significant amount of research in the area, obtaining the desired global behavior from the CPSoS is still very difficult. The vision of CYCLOPS is to address these major challenges by providing an integrated open-source platform that will enable the development of CPS-tailored domain-specific modeling languages (DSMLs), and their use throughout the engineering and operation lifecycle of smart CPSs. The platform will be: - Heterogeneous: Tool providers will be able to create DSMLs for different CPS domains and technologies, including machine learning models. A standard protocol for CPS-oriented DSML specifications will be defined, to decouple the DSML from its development environment. Hence, for each component, developers will be able to choose independently the most suitable DSML and the most effective development environment, including state-of-the-art 3D physical modeling tools. - Holistic: Despite the heterogeneity of languages and development tools for component systems, the platform will allow for simulating and reasoning on the global behavior of the CPSoS. This integration will be performed by high-level DSMLs defining model transformations towards common semantic domains. Metamodels of these domains will include physical properties (e.g. time, units) and laws. The DSMLs will span over all the phases of the CPSoS development, from requirements to operation. - High-performance: A library to support the development of efficient code generators for CPS-oriented DSMLs will be provided. In particular, the CYCLOPS platform will allow to define for each DSML a concept of execution trace and autonomic loop, in the form of model updates in response to runtime events. New mechanism for high-performance storage and querying of time-based models at runtime will be necessary. To achieve this timely and ambitious vision, the CYCLOPS proposal already includes R&D partners with a long track-record of conducting cutting-edge research on domain-specific modelling and model-driven engineering, and of delivering research results in the form of widely-used, sustainable and industrial-strength open-source software. The MRSEI will allow the coordinator to extend the consortium with academic partners from the CPSoS and machine learning communities, and to bring together end-user partners with case-studies from different CPSoS domains (e.g., public-facing safety-critical systems, IoT manufacturing, remote industrial control). The development of the CYCLOPS platform will be guided by an advisory board of world-class experts in CPSoS.