Bottom-up Job Shop Scheduling with Swarm Intelligence in Large Production Plants

In production plants organized by the job shop principle, the factory-wide scheduling problem is NP-hard and can become extremely large. Traditional optimization methods like linear optimization reach their limits in these settings due to excessive computation time. Therefore, we propose this industrial setting as a novel field of application for swarm intelligence using bottom-up algorithms that do not require the infeasible calculation of an overall solution but depend only on local information. We consider the example of the semiconductor industry producing logic and power integrated circuits where a diverse range of highly specialized but low volume products are fabricated in the same plant. This paper shows how to select and model swarm members, swarms, and their interactions for use in real-world production plants. There are multiple possibilities for the modeling of the agents: a swarm member could be a single machine or a set of machines (workcenter), a product or group of products of the same/similar type, or a more abstract agent like a process. In particular, we consider criteria for selecting appropriate swarm members and potential candidate swarm algorithms inspired by hormones and ants.