This quantitative study investigates the implementation and performance of artificial intelligence (AI)-integrated Internet of Things (IoT) sensor networks for real-time structural health monitoring (SHM) of in-service bridges. A dataset of 62 bridges, representing diverse structural types (steel, concrete, composite), environmental exposures (urban, rural, marine), and traffic conditions, was analyzed to understand deployment attributes, system performance, and predictors of structural condition. The study examined sensor modalities (accelerometers, GNSS, vision, and fiber Bragg gratings), data transmission networks (LoRa, ZigBee, 5G, Wi-Fi/Ethernet), and key performance indicators including sensor accuracy, AI detection precision, transmission latency, and the Bridge Health Index (BHI). Analytical procedures followed a rigorous, multi-step process: descriptive statistics profiled the asset base and technology adoption; assumption checks validated data quality and model suitability (normality, homoscedasticity, multicollinearity); correlation analysis explored variable relationships; and multiple regression models tested predictive drivers of bridge health. Results showed strong uptake of AI-enabled systems (61%) and robust sensing performance with accelerometer error ~1.8% and fiber Bragg grating error ~8 με. AI detection precision averaged 92%, while transmission latency varied substantially across networks (median: 21 ms for 5G vs. 180 ms for LoRa). The final regression model explained 64% of BHI variance (adjusted R² = 0.61). Both AI detection precision (β = 0.29, p = .001) and sensor accuracy (β = 0.27, p = .003) were strong positive predictors of BHI, while latency negatively impacted structural condition (β = −0.31, p = .001). Control variables such as bridge age, heavy traffic, and marine exposure were associated with lower BHI scores, highlighting the need to consider environmental and operational stressors. Hierarchical modeling confirmed that AI precision adds significant explanatory power beyond sensing and network performance, and interaction analyses revealed that robust AI can partially offset slower data networks, while sensor calibration is especially valuable in marine contexts. This work advances SHM practice by quantifying AI’s added predictive value, benchmarking network performance under field conditions, and clarifying environment–technology interactions.
Md Ismail Hossain (Mon,) studied this question.