1D-CNNs for autonomous defect detection in bridge decks using ground-penetrating radar

Bridges play a pivotal role in modern society, especially when considering the amount of global automotive transportation; therefore, it is essential to protect these structures from deterioration. Engineers and inspectors are searching for efficient subsurface defect detection methods due to the critical nature of these structures. Ground Penetrating Radar (GPR), a Nondestructive Evaluation (NDE) technique, is a well-established method used to locate subsurface degradation in bridges. The GPR method uses radiofrequency electromagnetic waves to create images of subsurface irregularities by detecting dielectric property differences in bridge decks. Using artificial intelligence (AI) to augment manual GPR data analysis can increase the NDE performance in the field by mitigating data interpretation. One dimensional (1D) Convolutional Neural Networks (CNNs) were employed to evaluate concrete bridge decks to classify the GPR data collected from eight laboratory-created concrete specimens with either defect-free or known artificial subsurface defects. We used 1D-CNNs to classify GPR data for accurate flaw identification. The proposed method's accuracy was greater than 84%, outperforming existing Machine Learning (ML) based GPR data classifications and demonstrating the proposed method's effectiveness in detecting subsurface defects.