Modeling and knowledge management of agile production systems with automation

Despite their wide deployment, automated agile production (AAP) systems have not yet received enough attention in the existing literature. This thesis work attempts to partially fill this void by studying the planning, design, control and management principles of AAP systems. First, we show that automating a manual machine can dramatically improve system performance if the operator is the bottleneck. On the other hand, when a machine becomes the bottleneck the benefits of further automation are greatly reduced. Second we show that, for push AAP systems, automation is most effective when it is placed towards the end of the line, with the operator giving higher priority to downstream jobs and jobs on the automated machine; while for CONWIP (pull) AAP systems, the position of automation does not matter, and the operator only needs to prioritize jobs on the automated machine. Third, we show that CONWIP not only offers greater design flexibility, but also is more efficient and robust than push. Finally, we propose a diagnostic tree, aiming at helping practitioners better manage production systems. The diagnostic tree decomposes a performance objective into successively more concrete subordinate objectives and finally into potential improvement strategies, enabling practitioners better understand the links between policies and performance and organize working-knowledge. We demonstrate the development of the tree and its application to a real-world case study.