Cooperative game approach based on agent learning for fleet maintenance oriented to mission reliability

- A cooperative game method is proposed for multi-level fleet maintenance planning.
- The problem scale is reduced significantly by agent learning.
- The effect of complex reliability model is controlled by agent learning.
- The global optimal solution can be rapidly got with high quality.

Fleet maintenance oriented to mission reliability is a multi-level maintenance planning problem that becomes highly difficult due to the various reliability models of equipment and fleet. A three-level decision structure for fleet maintenance is established, the objective is maintenance cost, the constraints is the reliability of fleet, and the variables are the maintenance statuses of line replaceable modules. Then, the fleet maintenance process is translated into game behavior among considerable equipment with different statuses. A cooperative game framework based on agent learning is developed. A convergence condition for optimization is proposed by a simulated annealing approach. In the game method, three types of learning signals and their evaluation rules are introduced to establish the equipment's reduced strategy space. Thus, the computation amount of game can be controlled, and the reliability constraints can be satisfied during the game process. Furthermore, the assessment method for the equipment payoff with a penalty factor is established, and the rapid search algorithm of Pareto optimal solution is provided on the basis of the total revenue of game. A case study is performed on a fleet of 15 aircrafts to prove that the proposed approach can reduce the maintenance cost effectively and can meet the fleet mission reliability requirements.