Aircraft maintenance stand scheduling via graph attention network and deep reinforcement learning

Purpose Maintenance stands are the scarcest maintenance resources, providing the essential space and facilities for aircraft maintenance. To expand the maintenance market, maintenance, repair and overhaul companies urgently need to achieve a more rational intelligent aircraft maintenance stand schedule. This paper aims to examine aircraft maintenance stand scheduling problem to enhance safety, efficiency and profitability. Design/methodology/approach A new heterogeneous graph model is first constructed to represent the scheduling problem, incorporating two types of nodes: maintenance item nodes and maintenance stand nodes. Second, the authors propose a novel graph attention network architecture to enable targeted embeddings for both nodes and arcs. Then, an end-to-end proximal policy optimization algorithm with an actor-critic style is designed to interact with the environment iteratively, leveraging a vectorized value function. Findings The extensive experimental results validate the performance of the proposed method, demonstrating its superiority over the state-of-the-art in terms of high-quality solutions, online computation time and generalization capability. Moreover, ablation studies on synthetic instances demonstrate that the proposed method can significantly reduce the relative scheduling error by at least 20.89%. Originality/value To enhance safety, efficiency and profitability in the maintenance process, the authors use an end-to-end deep reinforcement learning method to address aircraft maintenance stand scheduling problem.