DHRP-DQN: A Deep Reinforcement Learning Approach for Dynamic Path Optimization in Aircraft Maintenance Material Warehouses

With the rapid development of the aviation industry, efficient retrieval of aircraft parts in maintenance, repair, and overhaul (MRO) warehouses has become a critical bottleneck, accounting for 15% to 25% of total turnaround inspection time. Considering the high value of aircraft parts, stringent airworthiness requirements, time-sensitive maintenance tasks, and the limitations of existing path optimization methods in addressing dynamic material locations and complex constraints, this study proposes an enhanced deep Q-network algorithm termed DHRP-DQN. This algorithm integrates a Dueling Q-Network architecture, prioritized experience replay, hierarchical dynamic reward functions, and dropout regularization to solve warehouse path optimization for aircraft material retrieval. Experimental results demonstrate that DHRP-DQN achieves an average reward value of 727.8, a task success rate of 77.0%, and algorithmic stability of 31.5 (σ). Compared to the Dueling DQN baseline and standard DQN, it achieves improvements of 31.8% and 11.7% in average reward and task success rate, respectively, while reducing algorithm stability to 73.6% and 29.5%. This fills a research gap in MRO warehouse path optimization for aircraft materials and holds significant value for digital transformation in aviation maintenance.