Soft actor-critic facilitated algorithm for maintenance optimization considering reliability

This paper studies a maintenance optimization problem considering reliability for the engines equipped on a fleet of airplanes. During operations, engines inevitably encounter the risk of in-flight alerts, which can significantly impact maintenance decisions and flight scheduling. Therefore, we formulate a mathematical model to jointly optimize flight-to-engine assignments, the timing of shop visits, and the depth of modular maintenance based on in-flight alert reliability analysis. To solve this comprehensive optimization problem, we propose a soft actor-critic (SAC) facilitated algorithm. The algorithm integrates a graph attention network based dual stabilization strategy to accelerate convergence, alongside an improved SAC method incorporated with a reward shaping mechanism to further enhance computational efficiency. Experimental results demonstrate that the proposed algorithm reduces the computational time by 34% compared to the traditional column generation method. Finally, we derive several managerial insights for airline operations. For instance, constructing flight portfolios with moderate heterogeneity can balance engine flight hour and flight cycle consumption to extend operational time between maintenance events, and given the uneven degradation across engine modules, differentiated flight assignments prevent simultaneous shop visits, thereby reducing total maintenance costs.