Keywords: Civil engineering
-
A multi-module deep learning framework with graph-based network and crack attention for tunnel lining crack segmentation from LiDAR point cloud
Cracks in railway tunnels threaten safety and require efficient monitoring. This study introduces PointCrackNet, a deep learning method that detects tunnel lining cracks directly from LiDAR point clouds. By combining graph-based convolution, attention mechanisms, and a crack-enhancement module, the model accurately captures fine crack details while maintaining global structural context. A tailored loss function addresses class imbalance and improves crack continuity. Tested on a large-scale LiDAR dataset from a real railway tunnel, PointCrackNet outperformed existing methods. The approach enables automated tunnel inspection and supports smart, data-driven infrastructure maintenance.
Shanpeng Liu, Koichi Isobe, Junling Si, Diyuan Li, and Daoju Ren
Construction and Building Materials Volume 494, 2025, 143383
SH wave scattering in Eringen’s nonlocal elastic solid using the method of fundamental solutions
Eringen’s nonlocal elastic solid is a mechanical model that enables the analysis of phenomena difficult to describe using classical elasticity. This study analyzed wave scattering in nonlocal elastic solids using the method of fundamental solutions, a meshfree numerical method. An analytical representation of the traction operator specific to nonlocal elasticity was derived, and scattering characteristics relevant to ultrasonic nondestructive testing were evaluated.
Akira Furukawa, Taizo Maruyama, Takahiro Saitoh, Sohichi Hirose, Davinder Kumar, Dilbag Singh, and Sushil K. Tomar