The objectives of GO0D MAN project are set to allow realization of a fully functional, replicable and therefore widely exploitable solution, employing multi-agent systems, smart on-line inspection tools and data analytics, for implementing a Zero defect Manufacturing (ZDM) strategy. Full success will consist in the demonstration in three relevant industrial cases, which represent more than 80% of the manufacturing sector.
Aiming to improve the process performance and product quality, and particularly mitigate the impact of unexpected perturbations and defects, an intelligent, flexible and distributed infrastructure is mandatory to collect, monitor and process data according to the different industrial requirements, namely in terms of response time and optimization. For this purpose, a distributed and decentralized decision making structure will be adopted at different ISA 95 levels, using the multi-agent systems principles, for data collection, monitoring, defect diagnosis and predictive analysis using real-time and historic datasets. A critical issue is proper balancing of local data analysis (that provides automatic and real time monitoring and early detection of deviations and trends) and global data analysis (that provides self-optimization and continuous improvement using also cloud based infrastructures for data storage and processing). The multi-agent system will play a key role to implement the distributed infrastructure able to collect, monitor and process data, in local and global levels, supporting the execution of different types of advanced data analytics algorithms.
A zero defect strategy requires reliable information to support any decision. Reliable information requires measured data from the process and the product. However, in a real production scenario, availability of data does not by itself guarantee an effective information build-up. In particular, measurement uncertainty must be known, kept under control and adequately managed so to keep the confidence level on measured data at the required level.
For this scope the GO0D MAN project will implement quality control systems exhibiting smart behaviours, such as continuous adaptation of the sensing functions to the variable product/process parameters/characteristics and to the environmental changes, sensor self-tuning, self-diagnosis and sensor self-calibration. Furthermore, valid measured data will be processed at local level so to extract meaningful diagnostic information, with associated uncertainty and confidence level; in order to support decision making. Data compressed at local levels are shared at global level with just the right amount of detail.
A lot of information and knowledge can be extracted from the production system, by processing data available from on-line inspection tools and from local process controllers. However today, usually the data are analyzed in separate stages for specific objectives, i.e. quality control of a specific component, process control of a specific machine operation or assembly, data from repairing operation. The GO0D MAN project aims to introduce a data driven quality control solution, capable to extract data, store in data base and/or into the cloud environment and real-time stream processing of huge amount of data. At the shop floor level the CPSs are responsible to extract and pre-process the data, after that pre-processing, the information is streamed to further processing and stored as well. Hence, the information can be used to perform self-learning (using machine learning algorithms, big data processing and data mining). With the self-learning capabilities, the system is capable to analyse trends and predict future behaviours of the manufacturing system at global level, suggesting and triggering the reconfiguration of the system, using again the CPS to do it.
4. Integration and prototyping of Hybrid ZDM strategy
In order to run a multi-stage manufacturing system exploiting the distributed system architecture built on agent-based Cyber-Physical Systems (Objective 1), an agent has to be associated to each resource present on the line, i.e. to each process operation and to each smart inspection system (Objective 2), as well to each component/product flowing through the line. The system has to be then integrated with the data analytics tools (Objective 3), which allow ZDM strategy to be realized. This will be done on two different levels first the technical level and second the conceptual level. The technical level deals with the aforementioned agents that are deployed in place and establish a reporting stream to a database. This database is accessed by the process operations to enable the mining. Traditional algorithms for cleaning and splitting the data sets are performed.
The conceptual level introduces the necessary meta data and semantic to the database scheme, which must be understood by the agents as well as the processing optimization algorithms. Here concept modelling with the meta modelling database is used to setup a user-friendly meta data schema that can be interpreted and used by the agents as well as the process operations.
The agent-based CPS embodying smart on-line inspection tools and advanced data analytics will be installed and tested during normal production in three existing multi-stage manufacturing lines, in order to bring it up to TRL7. Even if the scientific community has put in evidence the advantages of intelligent distributed control systems based on CPS, their adoption by industry is still a challenge. The way to address this challenge is to demonstrate in real production environments the potential of such solutions and to develop replicable solutions applicable to different sectors. A real fulfillment of the project objectives and the realization of a real impact at industrial level will be achieved only if it will be demonstrated that it is possible to implement a ZDM strategy using a common, and moreover industrially replicable, approach based on Multi Agents and CPS technologies in significantly different industrial cases, with different levels of automation and production rates. The three industrial use cases have been selected to be representative of more than 80% of existing multi-stage manufacturing systems and differ significantly in their degree of automation and in production rate.