In job shop-configured production facilities, the overarching scheduling challenge is recognized as NP-hard and can scale to vast complexities. Classical optimization techniques such as linear optimization often fail to be effective in these scenarios due to the prohibitive computational demands. In response, we suggest exploring the use of swarm intelligence in this industrial context, leveraging bottom-up algorithms that rely on localized data rather than computing a comprehensive solution. This paper focuses on the semiconductor industry, where both logic and power integrated circuits are produced. These facilities typically handle a varied assortment of specialized, low-volume products within the same factory. Our research details methods for choosing and structuring swarm entities and their interrelations for real-world application in production environments. Several approaches are considered for agent modeling: a swarm member might represent a single machine or a cluster of machines (a workcenter), an individual product or a collection of similar products, or even a more conceptual agent like a process. Specifically, this document discusses the criteria for the selection of suitable swarm entities and reviews potential candidate swarm algorithms inspired by biological systems such as hormonal and ant behaviors.
M. Schranz, M. Umlauft and W. Elmenreich
