Model-Predictive Controllers for Performance Management of Composable Conveyor System

The increasing complexity of Cyber Physical Systems (CPS) found in a variety of domains, and the unforeseen fluctuations in operating conditions caused by the open nature of these systems makes it significantly challenging to appropriately configure and adapt the operating parameters of CPS to ensure reliability and desired quality-of-service (QoS). Material handling and packaging is one such domain experiencing these difficulties where the plants employ composable and reconfigurable conveyor systems that enable adaption to changing demands and business processes. Plant operators, however, face a daunting task of ensuring that the system throughput is maximized in the face of fluctuating workloads, while at the same time operation costs incurred in operating the plant are minimized. Plant operators require that these capabilities continue to be available even after the conveyor layouts change. To address these duo of challenges, this paper develops and compares five different model-predictive controller designs all of which have an inherent quality of being composable. The five controller designs comprise a singlelevel, completely decentralized controller with two alternatives on how belt speed is estimated; a centralized controller; a twolevel controller; and a time-abstracted centralized controller. The composable nature of all the controller designs lends towards addressing the situation when the layouts change. An evaluation of our controller designs in the context of a case study is presented comparing to a baseline case with a static system configuration. Keywords—Model-predictive Control, Composable conveyors, CPS