Task rescheduling in multi-agent manufacturing

Abstract

We present a model for task rescheduling in multi-agent manufacturing that is geared toward maximising system efficiency and reliability in an environment with predictable failure patterns. We describe a distributed manufacturing system architecture (http://hms.ifw.uni-hannover.de/public/overview.html) to illustrate the merits of rescheduling. We also develop an action redistribution framework based upon resource cooling, i.e. migrating actions from the most utilised or faulty (hottest) agents' resources to the coldest (least constrained) ones. Furthermore, the paper discusses the costs versus benefits of resource cooling when rescheduling cascades over parallel resources executing a decentralised production task.