European industry has been very competitive on the global markets by utilizing highly efficient Artificial Intelligence (AI) tools and massively producing high quality products. The advent for mass customization has been stressing the capability of modularization and flexibility of production processes through the incorporation of AI technologies. However, the communication between the different automation systems has not been yet accomplished efficiently since they lack interoperability and are restricted to their own system of coordination. The MAS4AI proposal proposes a system that allows the deployment and synchronization of different AI agents in manufacturing for autonomous modular production and human assistance. The MAS4AI system will be heavily driven by large industrial cases and will aim towards digitalising European industry with AI tools according to the Industry 4.0 paradigm. MAS4AI will develop its overall ambition by the means of four Scientific and Technological objectives namely: a) Multi-Agents-System (MAS) for distributing AI components in different hierarchy layers, customers and suppliers for realising refurbishment activities, b) AI agents using Knowledge-based Representation with Semantic Web Technologies, c) AI Agents for hierarchical planning of production processes, d) model-based Machine Learning (ML) AI agent. MAS4AI research and technological activity will be strongly driven by a set of industrial use cases which will be then used as demonstrators. The demonstrators involve important industrial sectors of high value added for Europe, namely AI technologies used for automotive, contract manufacturing, bicycle industry, bearings production and wood processing industry.

Smart factories are characterised by smart processes, smart machines, smart tools and smart products as well as smart logistics operations. These generate large amounts of data, which can be used for analysis and fault prevention, as well as the optimisation of the quality of manufacturing processes and products. DAT4.ZERO is a Digitally-enhanced Quality Management System (DQM) that gathers and organizes data from a Distributed Multi-sensor Network, which, when combined with a DQM Toolkit and Modeling and Simulation Layer, and further integrated with existing Cyber-Physical Systems (CPS), offers adequate levels of data accuracy and precision for effective decision-support and problem-solving – utilizing smart, dynamic feedback and feed-forward mechanisms to contribute towards the achievement of Zero Defect Manufacturing (ZDM) in smart factories and their ecosystems. The aim is to Integrate smart, cost-effective sensors and actuators for process simulation, monitoring and control; develop real-time data validation and integrity strategies within actual production lines; demonstrate innovative data management strategies as an integrated approach to ZDM; & develop strategies for rapid line qualification and reconfiguration. Deployed in 5 distinct industrial pilot lines we address the following primary objective: Develop and demonstrate an innovative DQM system and deployment strategy for supporting European manufacturing industry in realizing ZDM in highly dynamic, high-value, high-mix, low-volume production contexts, by effective selection and integration of sensors and actuators for process monitoring and control, a DQM platform with an architecture that provides reliable and secure knowledge extraction to ensure integrity of data, & strategies for advanced realtime data analysis and modeling in multiple domains and sectors that will increase quality, reduce ramp-up times and decrease time-to-market.

Facts: i) Zero-defect manufacturing and flexibility of production processes are some of the main challenges for European manufacturing; ii) One of the engineering tools with higher potential is the linking of simulation tools with measurement devices for real-time control of applications. The huge potential of this synergistic loop remains untapped for manufacturing processes and could be used for reducing product variability, increase line flexibility and achieve zero defect production. These objectives could be reached by integrating in the production line multi-physics simulation models, able to predict the product quality indicators in response to the values of critical input parameters (components dimensions, material properties, etc), which are unavoidably subject to variability: different batches, different suppliers,... The models will be fed with actual data from online measurements and, based on the model prediction, the critical steps of the line will be controlled to adjust the product to the exact design specifications or to quickly change specifications for producing customised batches. However, doing this in real time is not possible due to the computational cost of models. Reduced Order Modelling is a new generation of techniques which allow us to obtain parametric solutions of complex models that can be particularized in real time for any value of the parameters. The models run so fast that they can be executed on tablets or smart phones. ROM will be used to transform complex models into the real-time capable models that can be integrated in the production line. Moreover, the online deployment of ROMs and data gathering systems will generate big data which will be exploited through data analysis techniques for further improving the process. The project will show proof-of-concept demonstrations in three real process chains of the automotive sector, covering different types of production methods, products, materials and manufacturing processes.