MATRIX ASSEMBLY SYSTEM SCHEDULING OPTIMIZATION IN AUTOMOTIVE MANUFACTURING: A DEEP Q-NETWORK APPROACH

In response to the demand diversification in automobile production, traditional manufacturing processes are transitioning towards more flexible systems with dynamic scheduling methods. The Matrix System (MS) stands out for its utilization of Autonomous Mobile Robots (AMRs) and multi-purposed workstations, enabling a dynamic production environment. Each AMR is tasked with transporting a partially assembled vehicle through multiple workstations until final assembly, adhering to predefined precedence orders. However, determining operation schedules amidst the complexity of multi-model systems poses a significant challenge in minimizing manufacturing time. To address this, we formalize the problem into a Markov Decision Process (MDP) and propose a Deep Q-Network (DQN) based scheduling optimization algorithm for the Vehicles Production Scheduling (VPS) problem. Our approach utilizes discrete event simulation to assess candidate actions suggested by the DQN, aiming to derive an optimal policy. This paper validated the proposed algorithm by comparing with various dispatching rules.