Ultrasonic non-destructive testing of defects in concrete based on convolutional neural network
Fengyi Zhang,Lihua Wang,Yameng Wang,Wenjing Ye,Yongdong Pan
Abstract
Acoustic methods are widely used for non-destructive concrete testing. They detect internal flaws while preserving structural integrity. Nevertheless, current detection methods primarily depend on manual judgment, which is laborious, time-consuming, expensive, and susceptible to misjudgment and oversight. This study established an efficient and accurate detection system for identifying defects in concrete structures. Experimental data is first collected through ultrasonic tomography scans. To compensate for limited test data, numerical simulations generate extensive datasets with varied defect locations, enabling development and optimization of a 1D-CNN model using the dataset. This network structure is then applied to the experimental data for detecting defects in concrete structures. Values of 0.9708 and 0.9440 are achieved by the models which validates its effectiveness in practical applications and generalization ability. This method successfully combines convolutional neural networks with ultrasonic techniques. It enables the automatic detection of internal defects and provides a new effective approach for nondestructive testing of concrete structures.
Fengyi Zhang, Lihua Wang, Yameng Wang, Wenjing Ye, Yongdong Pan — School of Aerospace Engineering and Applied Mechanics, Tongji University, Shanghai 200092, PR China
PDF Viewer
Preview is limited to the first 3 pages. Sign in to access the full PDF.
