An agent approach to flexible automated production systems based on discrete and continuous reasoning

The rising complexity of today's automation systems leads to new challenges for manufacturers of automated production systems (aPS) and the producing industry. An important goal in automation software engineering is coping with complexity by introducing intelligent software components to broaden the flexibility of the overall system. To achieve plug-and-produce abilities, the paradigm of agent-oriented software engineering (AOSE) became popular in the last decade. This paper proposes an industrial agent based on a novel hybrid practical reasoning approach with discrete and continuous models. Therefore, the automation module's behavior is described with an undirected graph and a state-space model to compose the agent's knowledge base. By applying a combination of graph-search and multiple forward simulations of the state-space model, the agent acquires predictive insight about the module's behavior after following control commands by the agent system. Further the resulting trajectories are reasoned with an optimization criterion to evaluate the outcome and conduct decision-making by the resource agent. The approach was implemented in MATLAB/Simulink and evaluated on a modular lab-size plant, showing that the hybrid knowledge base is suitable to optimize throughput dynamically during run-time, even if constraints are introduced.