Collaborative-sequential optimization for aero-engine maintenance based on multi-agent reinforcement learning

To ensure the safety of aircraft, traditional intensive aero-engine inspections often lead to inefficient resource allocation. Consequently, there has been a growing research effort on optimization of inspection intervals. This study presents a collaborative-sequential optimization method for aero-engine maintenance, leveraging multi-agent reinforcement learning (MARL) techniques. Addressing the challenge of passive agents responding uniformly to rewards, the value decomposition network (VDN) architecture is adopted. Besides, the order requirements between inspection and maintenance determine that there should be timing between multiple agents. Therefore, a sequential simulation environment is constructed to transform multi-agent sequential decision-making into a simultaneous process. The proposed method is capable of dynamically adjusting inspection intervals under varying aero-engine degradation conditions. In comparative analysis against baseline strategies and fixed inspection interval strategies, the cost of the proposed method is up to 33.49 % of intensive inspection and 71.6 % of interval optimization, which verifies the superiority of this method.