From Condition Assessment to Verified Intervention: A Dual-Loop Architecture for Intelligent Railway Infrastructure Operation and Maintenance

Abstract Intelligent railway infrastructure operation and maintenance (O&M) must move beyond assessing current condition to managing how infrastructure changes under traffic, environmental exposure, and maintenance. Several gaps prevent this transition. Inspections can miss deterioration between visits or at unmonitored locations. Sensor measurements do not always reveal the underlying physical condition, network-level summaries can conceal local defects, and maintenance often begins only after problems become apparent. This Perspective proposes an engineering-oriented dual-loop architecture to connect condition assessment with verified action. Loop I combines sensing from in-service trains, uncertainty-aware condition estimation, and rule-constrained decision support to identify sites that need further investigation or maintenance. Loop II connects precise localization and controlled intervention with post-maintenance measurements to determine whether the intended change occurred. Cases from the Dazhun heavy-haul line and a kilometer-span high-speed railway bridge illustrate components that can be deployed and the connections that remain unresolved. This perspective outlines a staged research agenda focused on data alignment, uncertainty quantification, safety assurance, human authorization, transfer across railway lines, and learning from verified interventions. The proposed architecture provides a testable path from condition assessment to verified intervention.