AI Components Distribution
A basic concept in MAS4AI is to integrate all Smart Components in a holistic system architecture to enable the easy development and deployment of industrial AI technologies. Service implementations on the Factory Edge level facilitates:
Horizontal control of production modules
Data aggregation
Execution – of higher-level AI routines
Due to the open structure, the Software Agents can be distributed freely on the hardware and software resources. The structure can be scaled with low effort by adding external IoT cloud services using its high-performance resources.
Knowledge Representation
MAS4AI will employ semantics as an integration between various systems for production, coordination and interaction. This in turn requires formalized knowledge representations for products, process, human worker, object/resources, along with a generic asset/agent description.
MAS4AI semantic models will also adhere to certain principles:
Extendable – models adapted and extended, to fit the open-world context, to incorporate a new situation or changing environment
- Deal with uncertainty – due to ignorance, errors, unreliable sources or lack of observations
- Modularization – in order to be able to incorporate new knowledge and refine/extend existing knowledge
Hierarchical Planning of Production Processes
.An intelligent hierarchical planning agent that will evaluate the flow of information from the different abstraction layers of production and will take into account the user defined criteria in order to provide the best scheduling alternatives. It will be based on the following pillars:
Integration of product- and safety-related aspects at system level
- Open allowing interfacing with existing legacy systems
- Service-oriented communicating all information to external systems through dedicated methods
The concept includes the building of a rule-based mechanism that takes advantage of a knowledge base, which is constructed and expanded by accumulating problem-solving expertise. The goal is to find a good – not necessarily optimum – solution aiming to schedule the different production processes.
Model-based Machine Learning
The Model-based machine learning agent will be responsible of the mathematical comprehension of the process in terms of KPI’s, so MAS4AI implementation is able to correlate process parameters with important KPIs. The modelling approach here has three alternatives:
- Machine Learning (ML) data based models
Process simulation-based models
Hybrid models : also known as grey-box
Once the model-based machine learning correlate process parameters with KPIs, and taking benefit of the models and scenarios developed, the system is then ready to implement optimization. MAS4AI will address different optimization solutions in different hierarchical levels, guaranteeing the interaction amongst them.
Testing & Validation
MAS4AI involves challenging use cases that will demonstrate the feasibility, adaptability, scalability and flexibility of the MAS4AI multi-agent-based framework.
Lagging Majority – comprises those manufacturing processes where the maturity level is low and modularity is only achieved at activity level (e.g. logistics, assembly, inspection, etc.)
Early adopters – beginning to adopt and implement flexible modular productions concepts in their manufacturing processes in order to maintain their competitive positioning in the market
Innovators – vendor independent research facility
MAS4AI solutions will be deployed n different hierarchical layer for autonomous modular production and human assistance in a wide range of sectors (automotive, wood, bicycles, bearings and metal) of key economic impact for European economy.
