MANAGEMENT OF AI SMART MONITORING IN RAILWAY SYSTEMS: A SYSTEMATIC REVIEW OF PREDICTIVE MAINTENANCE, RISK-INFORMED DECISION SUPPORT, AND OPERATIONAL RESILIENCE

Abstract

Artificial intelligence, data-driven condition monitoring and digital twins are increasingly used to anticipate railway degradation, yet their managerial value depends on whether predictions are converted into defensible maintenance actions. This paper provides a structured narrative review of recent peer-reviewed evidence on railway predictive maintenance, focusing on maintenance management rather than algorithmic performance alone. The synthesis examines condition assessment, failure anticipation, maintenance prioritization, scheduling and resource allocation, and closed-loop performance feedback across infrastructure and rolling-stock applications. Recent studies show that machine-learning models can identify high-risk track sections, provide explainable real-time fault warnings, and optimize tamping schedules under operational constraints. However, a persistent translation gap remains between accurate diagnosis and network-level management because of limited field validation, weak interoperability, uncertainty, black-box models, and fragmented system integration. A monitoring-to-management framework is therefore proposed, linking sensing, diagnosis, prognosis, criticality assessment, prioritization, scheduling, intervention and feedback, supported by explainability, data governance and lifecycle costing. The review positions AI monitoring as a decision-support capability embedded within railway asset-management processes.