This paper presents a novel framework for optimizing aeroengine maintenance strategies, addressing the limitations of traditional approaches that rely on simplified degradation models and fixed inspection intervals. The study proposes a multiagent reinforcement learning (MARL) approach guided by a data-driven health index to dynamically adapt inspection and maintenance decisions according to the engine’s evolving health condition. A novel attention-enhanced unsupervised hybrid model, enhanced attention-residual autoencoder (EARAE), is developed for accurate health index estimation, capturing complex degradation patterns from sensor data through attention mechanisms, residual connections, and a degradation-aware constraint layer. The estimated health index serves as the state for two cooperating MARL agents, which optimize inspection intervals and maintenance actions to minimize total costs. By integrating double deep Q-networks, prioritized experience replay, and upper confidence bound exploration, the framework efficiently learns robust and adaptive maintenance policies. Experiments on the N-CMAPSS data set demonstrate that EARAE-MARL reduces total maintenance costs by 66.05% and 6.55% compared with single-agent inspection and preventive periodic maintenance plans, respectively, while ensuring timely interventions and mitigating engine failure risks. This study provides a practical and effective approach for integrating data-driven health estimation with MARL, enabling intelligent, cost-effective maintenance optimization for complex engineering systems.
Wan et al. (2026) studied this question.