Artificial Intelligence and IoT for Predictive Maintenance in Smart Manufacturing: A Framework for Equipment Reliability and Operational Efficiency

Abstract

Predictive maintenance combines equipment condition information with analytical models to support interventions before unacceptable degradation or failure. Artificial intelligence and the Internet of Things can strengthen this process by connecting industrial sensors, historical maintenance records, and decision-support applications. However, accurate model outputs do not automatically improve equipment reliability or production efficiency. This narrative review integrates established literature on condition monitoring, machinery prognostics, machine learning, cyber-physical systems, uncertainty evaluation, and explanation methods. It proposes a framework connecting asset criticality, sensing architecture, data quality, model development, maintenance planning, and continuing oversight. The analysis distinguishes anomaly detection, fault diagnosis, failure prediction, and remaining useful life estimation, because these tasks require different labels, evaluation designs, and operational responses. Particular attention is given to scarce failure data, censored histories, changing operating conditions, information leakage, and the lead time required for useful intervention. An illustrative graph calculates the relationship between assumed failure probability and a simplified maintenance decision under explicit cost assumptions; it does not report measured industrial savings. A comparative table summarizes complementary analytical approaches and their limitations. The principal conclusion is that predictive maintenance should be evaluated as a complete operational system rather than as an isolated algorithm. Reliable sensing, calibrated uncertainty, feasible maintenance actions, and prospective assessment are necessary to establish value. Future studies should test integrated systems across equipment types and production environments while reporting service consequences, decision costs, and limitations in generalization.