Machine Learning Enabled Fault Detection and Predictive Maintenance in Power Systems

The growing complexity of modern power systems—driven by renewable integration, distributed generation, and advanced power electronics—has reduced the effectiveness of traditional fault detection and maintenance strategies. Conventional threshold-based relays and time-scheduled maintenance struggle to address evolving grid dynamics and diverse fault signatures. Meanwhile, increased deployment of sensors, IEDs, and IoT-enabled monitoring systems generates large volumes of multi-modal data that can support data-driven decision-making. Machine learning (ML) techniques enable early fault detection, anomaly classification, and predictive maintenance by extracting patterns from electrical, environmental, and condition-monitoring datasets. This paper reviews classical ML models, deep learning architectures, and hybrid wavelet–DL approaches applied to transformers, transmission lines, circuit breakers, and current transformers. A predictive maintenance framework covering data acquisition, preprocessing, time-series forecasting, and health-index estimation is also presented. Case studies show that ML improves fault localization accuracy, remaining useful life estimation, and maintenance scheduling compared to traditional methods. Key challenges—including data quality, class imbalance, model interpretability, cybersecurity, and integration with legacy protection schemes—are discussed. Future research opportunities include self-supervised learning, physics-informed models, federated analytics, and edge-intelligent condition monitoring for next-generation resilient grids.