Explainable Fault Diagnosis Driven by Maintenance Expenditure in Power Systems

To improve the economic performance and operational reliability of wind-powered integrated energy systems, this paper proposes an explainable fault diagnosis framework. It explicitly embeds maintenance expenditures into the training objective of diagnostic models. Specifically, a cost-sensitive fault diagnosis method is developed based on the Economic Manufacturing Quantity (EMQ) model. Conventional "estimate-then-optimize" approaches prioritize diagnostic and prognostic accuracy while overlooking economic consequences, may resulting in suboptimal maintenance decisions under model misspecification. To bridge this gap, we present an integrated "estimation-and-optimization" strategy that embeds a maintenance cost model into the learning objective. A neural-network-based diagnostic model is trained to minimize both the expected maintenance cost (derived from the EMQbased production-inventory-maintenance formulation) and a standard statistical loss, thereby aligning prediction targets with decision-making goals. A case study demonstrates that the proposed approach reduces maintenancerelated expenditures and improves decision stability while preserving comparable diagnostic accuracy. The results highlight the practical value of cost awareness and explainable diagnosis for enhancing the economic efficiency of prognostics and health management frameworks in modern power and energy systems.

Keywords: Fault Diagnosis, Economic Manufacturing Quantity, Maintenance.