Condition-Based Maintenance for Large-Scale Fleets under Multiple Constraints: A Constrained MDP Model with Primal-Dual Solution

Managing maintenance activities for large-scale fleets, such as wind farms with numerous wind turbines, presents a significant challenge in condition-based maintenance. In addition to the curse of dimensionality inherent to optimizing dynamic decisions for large systems, prior research has primarily concentrated on individual modeling challenges, such as limited maintenance resources or overall system performance requirements, without fully addressing the need for a comprehensive solution that accounts for both dimensions. In this article, we propose a novel approach in the context of condition-based maintenance planning that integrates all three critical factors: system scale, resource limitations, and performance constraints. Specifically, we develop a constrained multi-agent Markov Decision Process (MDP) model to tackle the maintenance planning problem for a multi-component system, and we solve it using a Primal-Dual algorithm. The system includes more than 50 components with known transition dynamics. At each time step, the planner must decide whether to replace each component, balancing limited maintenance resources with stringent availability requirements. The goal is to find an optimal policy that the minimizes expected discounted maintenance cost while adhering to these constraints. Finally, we compare our method's performance against baseline approaches, demonstrating its ability to achieve superior trade-offs between cost and constraint satisfaction.

Keywords: Condition-based maintenance, Constrained Markov Decision Process (CMDP), Large-scale fleet systems, Resource constraint, Availability constraint, Primal-Dual approach.