[Objective] With the increasing penetration of renewable energy and the growing demand for flexible operation of coal-fired power units, thermal power equipment is experiencing more frequent load variations, deep peak shaving, and complex operating conditions. These changes have introduced new challenges to equipment condition monitoring, fault diagnosis, and intelligent maintenance. Conventional approaches based mainly on fixed alarm thresholds and expert experience are often insufficient for identifying early-stage abnormalities, locating fault sources, and supporting maintenance decisions under variable operating conditions. This paper aims to systematically review the recent progress of artificial intelligence (AI) applications in condition monitoring and fault diagnosis of thermal power equipment and provide insights into the development of intelligent operation and maintenance technologies. 

[Progress] The development and application status of AI-based condition monitoring and fault diagnosis are reviewed from two perspectives: technological evolution and typical equipment scenarios. From the perspective of intelligent methods, the characteristics, application tasks, and limitations of machine learning, deep learning, large language models, and time-series foundation models are analyzed. Traditional machine learning approaches, including principal component analysis, support vector machines, random forests, Bayesian networks, and multivariate state estimation techniques, have demonstrated advantages in small-sample modeling, feature-based diagnosis, and integration with expert knowledge. Deep learning methods, such as convolutional neural networks, recurrent neural networks, autoencoders, Transformers, and graph neural networks, have further improved the capability of extracting nonlinear and high-dimensional features from multi-source monitoring data, including process signals, vibration signals, frequency-domain features, time-frequency representations, and images. Recently, large models and foundation models have extended intelligent diagnosis from pattern recognition toward knowledge-enhanced reasoning, time-series representation learning, diagnostic explanation, and maintenance decision support. From the perspective of equipment applications, representative studies on boilers, steam turbines, generators, coal pulverizing systems, fans and pumps, and flue gas desulfurization and denitrification systems are summarized. The major monitoring data sources, typical fault types, and corresponding AI-based diagnostic methods are discussed. Existing studies indicate that intelligent monitoring has gradually evolved from single-parameter alarm systems to multi-source data fusion, online condition assessment, fault prediction, and integrated diagnosis. However, several challenges remain, including false alarms caused by operating-condition changes, insufficient fault samples, limited model interpretability, weak engineering evidence for diagnostic results, and inadequate integration between diagnostic outputs and practical maintenance actions. 

[Prospect] Future research should focus on improving data quality management, operating-condition identification, and cross-condition generalization capability. The integration of equipment mechanisms, operation and maintenance knowledge, digital twins, knowledge graphs, and time-series foundation models is expected to enhance the robustness, interpretability, and reliability of AI-based diagnostic systems. Through mechanism-data fusion, knowledge-enhanced reasoning, and closed-loop integration with operation and maintenance processes, AI technologies can further promote the transformation of thermal power equipment management from passive fault response toward proactive warning, intelligent diagnosis, and predictive maintenance.