AI-ENABLED PREDICTIVE ANALYTICS AND FAULT DETECTION FRAMEWORKS FOR INDUSTRIAL EQUIPMENT RELIABILITY AND RESILIENCE

This study investigates how AI-enabled predictive analytics and fault detection and diagnosis frameworks relate to industrial equipment reliability and resilience. Using a quantitative cross-sectional case-study design, we synchronize condition-monitoring signals, CMMS event histories, and operations context into per-asset analytical snapshots. Primary outcomes include failure occurrence and counts, failure rate and mean time between failures, downtime hours, availability, and overall equipment effectiveness. Core predictors are AI health indicators such as anomaly score and predicted remaining useful life, and detector quality metrics including F1, AUROC, and PR-AUC computed on temporally separated validation windows. Across an analyzable cohort of N = 412 assets, negative binomial models with operating-hours offsets and robust OLS demonstrate that higher anomaly burden aligns with higher failure intensity and more downtime, while longer predicted remaining useful life and higher detector quality associate with fewer failures and fewer hours lost. Utilization emerges as both a main driver and a moderator, with the anomaly to downtime slope steeper at higher duty cycles; class-stratified contrasts reveal the strongest effects for rotating equipment, moderate for discrete actuators, and attenuated for utilities. The contribution is twofold: a transparent pipeline that links standardized indicators to plant KPIs, and adjusted estimates that quantify the operational value of model discrimination and calibration. Robustness checks varying windows, thresholds, and leverage trimming preserve effect directions and magnitudes within narrow bands, and ethical safeguards include de-identified asset IDs and auditable data lineage. The design is grounded in a structured literature review covering 57 papers that frame constructs, metrics, and governance choices used in the analysis.