Integration of AI and IoT for Predictive Maintenance in Electrical Power Equipment

Abstract

The reliability of electrical power equipment is paramount to grid stability, economic efficiency, and public safety. Traditional reactive and schedule-based preventive maintenance strategies are increasingly inadequate for aging infrastructure operating under growing renewable energy penetration and rising load demands. This paper presents a fully integrated framework combining Internet of Things (IoT) sensor networks with Artificial Intelligence (AI) algorithms to enable Predictive Maintenance (PdM) of critical electrical power equipment including power transformers, induction motors, high-voltage circuit breakers, underground cables, and busbars. A five-layer IoT-AI architecture spanning sensing, edge computing, fog aggregation, cloud-based machine learning, and operator applications is systematically designed and validated. Python-based preprocessing pipelines including sliding-window feature extraction (FFT, wavelet decomposition, SHAP importance analysis), class-imbalance handling via SMOTE, and seven-algorithm benchmarking are fully described. A CNN-LSTM hybrid model achieved the highest fault detection accuracy of 96.1% (AUC = 0.987) across a composite dataset of 14 months of sensor records from 28 substations, outperforming LSTM, Transformer, Random Forest, Gradient Boosting, and Isolation Forest baselines. Field validation across seven equipment categories and 266 confirmed fault events yielded a mean F1-score of 0.933 and a mean detection lead time of 18.4 days for transformer insulation faults. Eight original figures are presented, including the system architecture diagram, ML benchmarking comparisons, ROC curves, fault detection results, lead-time box plots, economic impact analysis, SHAP feature importance rankings, and a real-time anomaly detection time-series illustration. An economic model demonstrates annual maintenance cost savings of USD 334,000 per substation (58.9% reduction) versus reactive maintenance, with a system payback period of 2.6 years. Cybersecurity threats specific to IoT-AI operational technology are systematically assessed against IEC 62443 and IEC 62351. The framework provides a scalable, standards-compliant pathway for utilities in both mature and developing power system environments.