AI-Driven Predictive Maintenance for Institutional Electrical Systems

High-voltage power distribution networks in institutional settings are increasingly vulnerable to age-related degradation, operational volatility, and unexpected service disruptions. Conventional maintenance practices—dominated by reactive repairs and static time-based servicing—impose significant financial burdens and risk over-servicing intact assets. To address these operational constraints, this study employs a thematic literature synthesis of smart grid prognostics and IIoT asset telemetry (2015–2025) to formulate an end-to-end four-layer predictive maintenance (PdM) framework. The architectural continuum is established by systematically mapping physical multi-sensor acquisition to edge-cloud analytics (GRU and Bi-LSTM engines), dynamic Asset Health Index (AHI) tracking, and automated, cyber-secure Computerized Maintenance Management System (CMMS) dispatch. Synthesizing literature performance baselines reveals that deep bidirectional recurrent networks—notably Bi-LSTM models—achieve superior degradation tracking for Remaining Useful Life (RUL) estimation, attaining an F1-score of 91.0%, a Mean Absolute Error of 4.3%, and a Root Mean Squared Error of 4.2. Furthermore, multi-modal feature fusion significantly sharpens Asset Health Index (AHI) modeling, driving prediction errors down to an RMSE of 0.052, lengthening warning lead times to 37 days, and maintaining false alarm rates at 6.2%. The transition to an AI-enabled prognostic strategy yields up to a 50.0% decrease in overall operational and maintenance costs while extending high-voltage asset longevity by 40%. The inclusion of Explainable AI (SHAP and LRP) alongside robust Operational Technology cybersecurity measures guarantees operational transparency and cyber resilience.