.png){kind=icon}
.png){kind=icon}
Return
Network-level pavement structural health monitoring enabled by a physics-informed hierarchical attention network and vehicle vibrations
.png){kind=icon}
PRE
AI
W
Q
DOI：10.1016/j.aei.2026.104775.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The deteriorating state of extensive road networks demands cost-effective, scalable structural assessment methods to inform proactive maintenance. Moving beyond localized and expensive traditional surveys, this study proposes PhysHA-Net, a Physics-Informed Hierarchical Attention Network, for interpretable multi-scale health monitoring with potential application to crowdsourced vehicle vibration data. The current empirical validation is conducted using a single vehicle to demonstrate the methodology and the physical plausibility of the model. By embedding vehicle-pavement interaction (VPI) dynamics into a hierarchical attention mechanism, the framework bridges the gap between raw acceleration signals and structural performance. Comparative experiments against three baseline models, the proposed PhysHA-Net achieves a high coefficient of determination (R2 = 0.985) and a low Mean Absolute Error (MAE = 0.124). Notably, PhysHA-Net reduces the prediction error by more than 88% compared to standard data-driven attention mechanisms. The results confirm that the physics-informed layer effectively constrains the attention weights within mechanically plausible domains, providing a reliable and interpretable tool for network-level infrastructure management.
Keywords:
Pavement health monitoring
Physics-informed neural networks
Hierarchical attention mechanism
Vehicle vibrations
Structural assessment
Journal
IF:
9.9
Papers:
4.0K
Citations:
1.7W
Organization
Cited Papers