Multistage manufacturing systems are inherently complex because of their hybrid structures with mixed sequential and/or parallel stages, mixed data types and scales and the existence of feedback/forward loops. In multistage systems the outputs from a given stage are the inputs to another, result from one stage is influenced by local variations in the process, but also by the propagation of defects from the previous stages, hence the overall global performance of these systems is statistically dependent on the stochastic performance of each individual stage.
In a context of a ZDM strategy, the use of innovative approaches is proposed to early detect anomalies and properly implement mitigation actions for preventing the defect generation and propagation to downstream stages and to reduce waste.
As it can be observed in the figure, GO0D MAN will provide a cluster of tools and services conglomerated in a distributed cloud covering several layers of the ISA-95 pyramid from level 1 to level 4. These services and tools encompass Multi Agent based CPS, big data analytics, data mining and machine learning techniques as well as knowledge management along with all the tools required for a successful zero defect strategy.
The overall multi-agent system architecture, integrating process and quality control, and enhanced with advanced data analytics algorithms and learning techniques – represented in the figure below- supports the distributed on-line data acquisition systems, the on-line inspection tools, the on-line defect and recovery management policies as well as inter-stage coordination strategies and selective inspection policies to achieve higher control of the entire multi-stage system.
The establishment of local and global levels is the core architectural axiom by implementing ZDM strategies, since it allows to combine the earlier detection of anomalies locally at each stage, with the possibility to aggregate and correlate data from different individual stages aiming to identify anomalies that can only be detected globally. The challenge is the proper balancing of the local data analysis – providing automatic and real time monitoring and early detection of deviations and trends – and the global data analysis – providing self-optimization and continuous improvement.
As real-world examples from the selected use cases, GO0D MAN will allow:
- A real-time analysis and knowledge extraction from the big data gathered from the shop floor to provide real time detection of anomalies or deviations, at a single stage or resulting from the effects of several stages. This mechanism allows to detect in real time the occurrence of possible defects or deviations, being able to select in real time proper strategies to prevent or mitigate their further occurrence.
- An online inspection system of the evolution of the product quality along the production process. This mechanism permits to detect, in real time and at any point of the production line, when the desired quality for a product is not possible anymore to be achieved, even if all the remaining operations would be performed with a performance of 100%. In this case, agents may decide to stop the production of the product, removing it from the production line, or implement some recover operations to improve the product quality.
- Adjustment of upstream and downstream stages operation parameters. The implementation of feedback control loops and the correlation of gathered data by the distributed agent-based system will allow to influence the operation of upstream and downstream stages by adjusting production processes (e.g. adapting the machining operations) or customizing the on-line inspection tests (e.g. adapting the sequence of tests or calibrating the tests) taking into consideration the historical production data. At this level, the propagation of defects can be mitigated by the inter-coordination among agents.