Software-in-the-loop testbed for multi-agent-systems in a discrete event simulation: Integration of the Java Agent Development Framework into Plant Simulation

Abstract

Today's research projects propose a modular manufacturing environment for production sites, which adapt itself autonomously and makes manufacturing decisions without human interaction. Therefore, it is necessary that the next generations of production lines, especially the intralogistics transportation systems, are designed more adaptable and flexible. The object in this paper is a cyber-physical material flow system with flexible, autonomous and collaborative vehicles combined with centralized sensors to digitize the workspace. For this purpose, an interface was developed which allows a discrete event simulation tool to communicate with a Multi-Agent-System. Thereby, the decision-making of the agents is integrated directly into the simulation process of the discrete event simulation software. The architecture of this interface is presented as well as a test of its functionality. The architecture is implemented with the Java Agent Development Framework and Plant Simulation as the discrete event simulation tool. The result is an interface, which allows to transfer data from the simulation, in case of an event, to the agent platform. The Multi-Agent-System solves the event specific problem due to its ontology and responses it to the simulation. Therefore, it is possible to integrate the ontology implemented in the physical system as software-in-the-loop in the simulation environment. Furthermore, the possibility is given to improve the ontology iteratively based on historical production data. Different strategies of agents can be combined and improved through machine-learning algorithms by using real production data from the task specific hardware. This leads into a continuous improvement process.