Reinforcement learning based buffer allocation problem on complex production line

Abstract

We present a model and algorithm of Reinforcement Learning (RL) based Buffer Allocation Problem (BAP) for complex production lines. The BAP has been a critical part of the production line design and inventory policy construction. It determines the buffer size within the production line to achieve the highest production rate considering the uncertainty of the production line such as machine failures and repairs. The existing BAP models are designed to optimize buffers for highly simplified production lines. However, it is still a challenge to evaluate an optimization buffers for actual complex production lines with the existing approaches and algorithms. In this study, we present a model and algorithm to find the optimal buffer space for complex production lines using simulation based Reinforcement Learning (RL). Here, we build a complex production line model with a simulator and integrate the model with the RL based algorithm, seeking an optimal buffer allocation policy. The preliminary results show that the proposed model reliably fines an optimal buffer size for complex production lines.