A learning coevolutionary framework with property-based knowledge for a bucket brigade Seru scheduling problem

Customized variable demand arises from volatile markets, which challenges conventional production systems and requires mass customization (MC) to accommodate customers’ individual demands at a mass-production cost. Seru production system (SPS) provides an effective solution to deal with personalized demands because of its responsiveness. However, customized products exhibit variability, which poses difficulty in matching production requirements of orders with production capacity of serus. Moreover, inappropriate processing sequences of customized products may increase idle or waiting time, which leads to deteriorated performance for the SPS. This study focuses on a bucket brigade seru scheduling problem (BSSP) which includes three subproblems, i.e., order assignment, order sequencing, and product sequencing, with the objective of minimizing makespan. With mathematical analysis, three properties of BSSP that contain problem-specific knowledge are proposed. These properties enable factory managers to dynamically adjust orders’ or products’ processing sequences without negatively affecting makespan. A learning coevolutionary framework with property-based knowledge (LCF) is developed to solve BSSP. In LCF, collaborators between populations are dynamically chosen by a learning-based collaborator selection method. Problem-specific knowledge derived from the above three properties is incorporated to improve the efficiency of LCF. More than 4,000 instances are generated to evaluate the performance of LCF in addressing BSSP. Experimental comparison among LCF, variants of coevolutionary algorithms, and state-of-the-art algorithms demonstrates the superiority of LCF, which verifies the effectiveness of LCF in solving scheduling problems of the SPSs.