Comparison of fully decentralized algorithm with and without coordinator for production control in a reentrant flow shop

Abstract

In this study, we propose a fully decentralized optimization algorithm for production control in a reentrant hybrid flow shop. In this problem setting, each production area plans and controls the production in its own area in accordance with the cooperation with other production areas. Also, the production areas do not share any local information but communicate locally optimized solution with directly connected neighbors. We compare the performance of the proposed fully decentralized control scenario with a fully decentralized control with a coordinator whose responsibility is to force the production areas to agree with a compromised solution. The experiments were performed on a Lego smart factory. The Lego factory consists of three-stage reentrant hybrid flow shop. The Lego machine at each stage operated by LEGO Techniques and BrickPi equipped with NFC reader and wireless communication. Products are prototyped by the NFC tagged 3D printed balls. We analyzed the performance of the two methods based on the production quantities, delayed customer demands, Work-In-Process (WIP) and utilization under the occurrence of dynamic events such as machine breakdown and demand changes. We compared Coefficient of Variance of WIP and Root of Mean Squared Error of production quantities. The result of the experiments indicates that the fully decentralized control without coordinator outperforms the fully decentralized control with coordinator in terms of system WIP variance, production quantity and Customer Demand fulfillment.