Smart Manufacturing Control with Cloud-embedded Digital Twins

The paper presents a model for smart control of large scale manufacturing systems, in which pools of shop floor and computing resources are shared in a dual cloud pattern. The proposed architecture uses the holonic manufacturing paradigm by decoupling the decision layer from the control one. The decision layer uses intelligent agents that reconfigure optimally in real time the resource allocation and scheduling of operations on products at batch level; also, the resource health is monitored continuously. Decisions are taken in the high layer of the MES based on real time machine learning algorithms that predict resource performances and QoS influencing usage costs, classify and cluster resource states to predict anomalies in behaviours and prevent resource failures. The distributed control layer keeps reality awareness during production by using digital twins replicated for all resources. Data is collected in real time streams from physical resource and process twins, aggregated in time series and sent to the intelligent agents in the cloud without delaying production. Experiments discuss the forecast of abnormal pick-and-place robot operations.