CONWIP control in the digitized world: The case of the cyber-physical jobshop

A digital transformation is occurring in the operations of systems. This has been enabled by digital technologies developed under the umbrella of Internet of Things (IoT) and cyber-physical systems (CPSs) that automate operations of systems, allowing functioning with changing data. However, many production systems that were hitherto used to significant human intervention and offline computations are struggling to adapt to this digital trend. CONWIP, short for constant work-in-progress (WIP), is a pull mechanism widely used in operations of manufacturing and supply chains to place thresholds on WIP inventory, while delivering high production rates. Traditional approaches for CONWIP control malfunction in the digitized world, as they require optimization models that are slow and further need significant offline human intervention. A digital approach, on the other hand, is expected to be resilient, i.e., use only basic shopfloor data and deliver results quickly and automatically in real time. Hence, this paper seeks a digital approach for deriving and implementing CONWIP thresholds. To the best of knowledge, existing literature does not provide any CONWIP thresholds and/or updating algorithms needed in the digital setting of a CPS jobshop. First, closed-form formulas for CONWIP thresholds, requiring only basic shopfloor data without any simulation/optimization but suitable for incorporation into a digital hardware, are derived. Second, a machine-learning approach is developed for integrating the proposed formulas, thereby enabling seamless data-driven integration and functioning with rapidly changing data. Computational experiments show the proposed digital data-driven approach to closely approximate results from an offline optimization methodology in real time.