Modeling evolutionary supervisors for multi-agent manufacturing control with discrete event formalism

We describe the control system of a manufacturing plant as a network of atomic agents (discrete event controllers) of different kinds. Since agents of the same type do not interact, the control leads to low programming effort and data exchange but, at the same time, it can provide poor performance of the plant. To overcome this drawback, we introduce a discrete event supervisor which modifies the decision laws of the members of the agent population, according to a genetic evolutionary mechanism. Simulation shows that, due to the supervisor action, the network of agents can efficiently face a variety of disturbances.