Artificial Intelligence-Enhanced Connected Structural Health Monitoring of Traffic Signal Mast-Arm Structure Population within Vehicle-to-Everything Infrastructure

Manual inspection of constructed infrastructure assets remains labor-intensive, subjective, and poorly suited for large-scale automation across construction and maintenance lifecycles. Although vibration-based and artificial intelligence (AI)-driven methods offer potential for automated inspection, their adoption has been constrained by sensor deployment complexity, data-management burden, and limited integration with operational digital workflows. This paper presents a methodological framework for AI-enhanced connected structural health monitoring (SHM), whose primary novelty lies in integrating established methodological components, including vibration-based feature extraction, physics-informed modal feature engineering, conventional feedforward neural-network (NN) regressors, edge computing, vehicle-to-everything (V2X) middleware, and telematics-based visualization, into a unified V2X-enabled SHM workflow for automated condition assessment of traffic-signal mast-arm structures. Validation results demonstrate strong predictive performance, with coefficients of determination of 0.91, 0.98, and 0.70 for base, mid-arm, and mast-arm regions, respectively, and mean absolute errors of 0.038, 0.020, and 0.052, respectively, under realistic sensor noise. A data-driven alerting strategy, integrating isotonic regression and change-point detection, enables automated classification of normal, warning, and critical conditions, correctly identifying 86% of undamaged cases and 72% of critically damaged cases while conservatively flagging more than 99% of severe deterioration scenarios for intervention. The complete analytical pipeline, including signal processing, NN inference, and threshold evaluation, executes in under 5 ms per analysis window on central processing unit-only edge hardware, indicating computational feasibility for edge-based SHM without cloud dependency. By reusing existing digital infrastructure for sensing, analytics, and reporting, the proposed methodological framework has the potential to reduce manual inspection burden and provides a practical pathway for supporting construction inspection, post-construction verification, and infrastructure condition assessment. Field validation remains a necessary next step to evaluate long-term performance under real operational conditions.