Sensor Data Fusion and Condition Forecasting in Tunnel Ventilation Networks: Longitudinal Tracking

Abstract

Tunnel ventilation networks represent critical safety-critical systems within subterranean transportation infrastructure, requiring continuous monitoring and proactive maintenance. Traditional maintenance strategies often rely on reactive responses or simplistic single-sensor threshold alarms, which fail to capture complex, multi-modal degradation pathways. This paper presents a comprehensive study linking multi-sensor data fusion to long-term condition forecasting using longitudinal tracking evidence gathered over a twenty-four-month period from an active highway tunnel. We propose a hierarchical data fusion framework that integrates vibration, acoustic, thermal, and aerodynamic data stream sources from distributed sensor arrays. By employing a robust temporal alignment and noise reduction methodology, the fused data streams feed into a multi-horizon predictive forecasting model. The empirical results demonstrate that our multi-sensor data fusion approach yields a thirty-two percent improvement in forecasting accuracy for mechanical degradation compared to single-sensor baselines. Longitudinal analysis reveals that the fusion framework successfully identifies early-stage impeller imbalance and stator insulation degradation up to ninety days before structural or operational failure occurs. The study underscores the utility of continuous multi-sensor integration in modern infrastructure asset management, providing actionable insights for transition from scheduled to condition-based maintenance policies in underground transit networks.