Workpackage 1 finished with the submission of Deliverable 1.3 “ZDM Management Strategies and Rules” that shows a demonstrator to present GO0DMAN approach, involving all the components of the architecture: MAS, data analytics and knowledge management.
The developments of Task 1.3 aim at delivering an implemented solution of GO0DMAN’s ZDM methodology through the proper integration of all its key modules, allowing the correct flow of information through the different layers that constitute the GO0DMAN overall solution. To this extent, a service-based rule server was developed permitting the communication between the more complex, higher-level functionalities (KM and Data Analytics) and the Multi Agent-based CPS.
To do so, it is necessary to translate the decision rules resulting from the application of complex data analysis algorithms and the domain expert interaction by the KM, into simple, straightforward rules compliant with the common data format adopted by the CPS.
These rules can then be used by the CPS to adequately monitor and control the system’s quality parameters in runtime. This can be done for instance by modeling when to call or trigger maintenance actions, when to adapt production parameters as well as which parameters to tune, or even when to ask for new configurations for a given station.
As such, Task 1.3 defines the dataflow and provides the means for decision rules to be translated from the Decision Model and Notation (DMN) format, into an AutomationML repository, and finally being made available to the agents to be interpreted in runtime. The overall integration approach can be seen in the figure.
There are three key interactions involved in the proposed approach. More specifically, the KM layer can push new or updated rules to be relayed to the CPS, which can in turn consult existing rules, or request an update in case it finds itself with insufficient knowledge to process a given event.
Furthermore, this integration layer also serves as a way to decouple the CPS from the Knowledge Management specifications, allowing for a truly modular approach. This means that even if the KM layer is modified, only the interface to AutomationML needs to be adapted, making this process invisible to the CPS since its common data representation format remains the same, thus requiring no additional programming effort on that front.