Event-Level Visual and Equipment-State Integration with Closed-Loop Governance for Community Smart Elevators

Multimodal sensing provides a mature foundation for elevator monitoring, while visual records, equipment-state records, and operational responses still need to be connected in many deployed community systems. This study develops an event-level closed-loop governance framework for that connection. Continuous visual detections and abnormal equipment states are aggregated by elevator identifier and time to form candidate events with explicit temporal boundaries. Rule-based event-level integration assigns each candidate a platform-confirmed, pending-verification, or excluded state. An event-triggered supervisory loop then maps event state, type, persistence, and potential consequence to monitoring, verification, voice prompting, authorized door-control linkage, work-order dispatch, or emergency escalation; updated equipment, work-order, and on-site evidence determine closure, continued tracking, or escalation. A retrospective evaluation used operational data from October 2024 to March 2025 for a community smart-elevator platform in Shanghai’s Lingang New Area. Among 3657 platform-confirmed risk events, 1754 were faults, 1814 were anomalies, and 89 were passenger-entrapment events, equivalent to 279.4 events per 100 deployed elevators over six months. Ten communities accounted for 57.2% of all events. Scenario, Criteria Importance Through Intercriteria Correlation (CRITIC), minimum-value, weight-perturbation, and screening-threshold checks were used to examine the stability of maintenance priorities. Community A remained the highest-priority community in all comparisons. The three attention bands were further mapped to priority intervention, scheduled targeted inspection, and routine preventive monitoring strategies. This mapping prioritizes limited inspection and diagnostic capacity; it is not an estimate of component remaining useful life. The framework therefore connects risk recognition, supervisory response, recovery verification, and evidence-based maintenance prioritization without extending the available data beyond their operational scope.