Digital twin composition in smart manufacturing via Markov decision processes

Abstract

Digital Twins (DTs) are considered key components in smart manufacturing. They bridge the virtual and real world with the goal to model, understand, predict, and optimize their corresponding real assets. Such powerful features can be exploited in order to optimize the manufacturing process. In this paper, we propose an approach, based on Markov Decision Processes (MDPs) and inspired by Web service composition, to automatically propose an assignment of devices to manufacturing tasks. This assignment, or policy, takes into account the uncertainty typical of the manufacturing scenario, thus overcoming limitations of approaches based on classical planning. In addition, obtained policies are proven to be optimal with respect to cost and quality, and are continuously updated in order to adapt to an always evolving scenario. The proposed approach is showcased in an industrial application scenario, and is implemented as a freely available tool.