A Reactive Kanban System with a Neural Network

Abstract

As a Just-in-Time(JIT)ordering system for multi-stage production systems, the Kanban system has been proposed. In the Kanban system, the order release for each process is determined on the basis of actual demands, that is, without demand forecasts, and it leads to avoiding the influence of forecasting error. Although the influence of forecasting error can be avoided, the influence of demand changes, especially unstable changes, cannot be avoided even in the Kanban system. Based on this background, a reactive Kanban system that can react to unstable changes in product demand is proposed for realizing agile control in the Kanban system. In the reactive Kanban system, unstable changes in product demand are detected using a control chart, and the number of Kanbans is adjusted to the detected unstable changes. As a technique for detecting changes rapidly, neural networks have been applied to process control, and their effectiveness has been reported. Based on the literature, this paper develops a reactive Kanban system with a neural network for a multi-stage production system with unstable changes in product demand. In the system developed, a multi-layered neural network is applied for detecting pattern changes in time series data on demand, and the detected pattern change is utilized for controlling the number of Kanbans. A mathematical model of the reactive system is formulated, and the performance of the developed system is investigated and compared with that of the previous system by simulation experiments.