Daneshvar, A., Golroo, A., & Rasti, M. (2025). Incorporating Physics-Informed Neural Networks into YOLOv8 for Pavement Rutting Detection. In L. Eberhardsteiner, B. Hofko, & R. Blab (Eds.), Advances in Materials and Pavement Performance Prediction IV : Contributions to the 4th International Conference on Advances in Materials and Pavement Performance Prediction (AM3P 2025), 7-9 May 2025, Vienna, Austria (pp. 547–550). TU Wien, E230-03 Road Engineering. https://doi.org/10.34726/10885
Advances in Materials and Pavement Performance Prediction IV : Contributions to the 4th International Conference on Advances in Materials and Pavement Performance Prediction (AM3P 2025), 7-9 May 2025, Vienna, Austria
Physics-Informed Neural Networks (PINNs) have shown great promise in embedding domain-specific knowledge into machine learning models. This study proposes an enhancement to the YOLO object detection framework by introducing a physics-informed penalty into its loss function for detecting pavement rutting. A dataset of 4880 labeled images was used to train the model, with modifications ensuring predictions adhere to known physical constraints of rutting dimensions and aspect ratios. Experimental results demonstrate improved bounding box accuracy and alignment with domain-specific rules, reducing false positives in critical infrastructure monitoring. This approach highlights the potential of hybrid PINN-YOLO models in civil engineering applications.
en
Research Areas:
Modeling and Simulation: 35% Structure-Property Relationsship: 30% Computational System Design: 35%
CC BY 4.0
