Explainable Artificial Intelligence for Trustworthy Power Plant Maintenance Decisions

Power plant maintenance engineers increasingly distrust artificial intelligence (AI) predictions due to the opaque nature of complex models, limiting real-world adoption despite significant technical advances in predictive maintenance. This article investigates explainable AI (XAI) as a necessary condition for deployment in safety-critical power generation environments. We analyze four major AI techniques used in power plant maintenance, machine learning, deep learning, expert systems, and natural language processing and evaluate their explainability capabilities across three XAI metrics: transparency, interpretability, and actionability. The principal contribution is a novel XAI-readiness matrix that maps power plant asset classes (turbines, boilers, generators, transformers) to appropriate AI techniques based on failure criticality and operator expertise. We conclude that expert systems and rule-based AI offer the highest immediate trustworthiness, while deep learning requires post-hoc explanation methods (SHAP, LIME) for operator acceptance. The proposed framework enables maintenance managers to select AI techniques not only by prediction accuracy but also by explainability requirements, supporting safer and more reliable power plant operations.