Dynamic Job Shop Scheduling Based on Order Remaining Completion Time Prediction

Abstract

Emerging ubiquity of smart sensing in production environments provide opportunities to make use of fine-grained, real-time data to support decision-making. One, currently untapped opportunity is the prediction of order remaining completion time (ORCT) which can be used to improve production scheduling. Recent research has focused on the development of ORCT prediction models however, their integration into scheduling algorithms is an understudied area, especially in job shop environments where processing times can be highly variable. In this paper, an artificial neural network was developed to predict ORCT based on real-time job shop status data which is then integrated with classical heuristic rules for facilitating dynamic scheduling. A simulation study with four scenarios was developed to test the performance of our approach. The results demonstrated improved completion time, however tardiness was not reduced under all scenarios. In moving this research forward, we discuss the need for further research into combining static and dynamic characteristics and priority rule design for satisfying multiple objectives.