Dynamic inventory replenishment strategy for aerospace manufacturing supply chain: combining reinforcement learning and multi-agent simulation

The (I, R, S) policy is a well-known inventory replenishment strategy, where inventory is raised to an order-up-to-level S at the end of each review interval I, if it falls below a reorder-point R. Determining the optimal values for these parameters by mathematical analysis methods are difficult, especially in sectors with complex and uncertain purchasing, manufacturing and delivering processes. The (I, R, S) policy has been shown to result in low supply chain performance (SCP) composed of sales revenue, tardiness fine, manufacturing cost, inventory holding cost, raw material cost, etc. in industries that involve highly-customised orders, such as aerospace industry. In this paper, we develop a multi-agent simulation model combined with a reinforcement learning-based dynamic inventory replenishment strategy to maximise the SCP. The approach has been applied in an aerospace manufacturing case study. It empirically demonstrates that the dynamic strategy yields considerable improvements, and has an additional benefit of adaptivity to changes, such as demand and supply uncertainties.