Proactive in-house part-feeding for mixed-model assembly systems with dynamics

Abstract

The rich data collected from smart shop floors necessitate proactive decision-making to relieve system nervousness in a dynamic environment. This proactivity takes real-time actions based on disruption-driven prediction but not on disruption-created negative effects. This paper focuses on investigating a proactive in-house part-feeding (PIP) approach by integrating real-time data collected from shop floors with the physical properties of a mixed-model assembly system. An event-driven proactive prediction for replenishments is first implemented after coupling such data as disruptions, job rescheduling, part re-releasing, and real-time lineside inventories with a widely used reorder-point-based replenishment policy. This prediction can provide sufficient prior information about replenishments and, in turn, trigger tow-train rerouting to simultaneously minimize the costs of distribution and lineside inventories. In terms of limited tow train capacity and the just-in-time part-feeding goal, this rerouting problem should address various constraints on multiple trips, time windows, and the real-time working status of tow trains. An adaptive large neighbourhood search (ALNS) is developed to obtain the best solution for rerouting by designing customized removal and insertion operators, as well as local search heuristics. A case study of a real-life car seat assembly system is presented to verify the efficiency of the PIP. Several rerouting benchmark instances are obtained from the case study to evaluate the performance of ALNS, and the computational results indicate that the proposed ALNS has a satisfactory performance in solving the rerouting instances compared with some of the formerly proposed and published methods. The results of the practical case study illustrate that the PIP may increase the lineside inventory cost slightly but can significantly reduce the distribution cost compared with prevailing reactive approaches