Advanced Signal Analysis Model for Internal Defect Mapping in Bridge Decks Using Impact-Echo Field Testing.

This study presents an advanced signal analysis model for internal defect identification in bridge decks using impact echo field testing data designed to mitigate signal noise and the variability encountered during real-world inspections. Field tests were conducted on a concrete bridge deck utilizing an automated inspection system, systematically capturing impact-echo signals across multiple scanning paths. The large volume of field-acquired data poses significant challenges, particularly in identifying defects and isolating clean signals and suppressing noise under variable environmental conditions. To enhance the accuracy of defect detection, a deep learning framework was designed to refine critical signal parameters, such as signal duration and the starting point in relation to the zero-crossing. A convolutional neural network (CNN)-based classification model was developed to categorize signals into delamination, non-delamination, and insignificant classes. Through systematic parameter tuning, optimal values of 1 ms signal duration and 0.1 ms starting time were identified, resulting in a classification accuracy of 88.8%. Laboratory test results were used to validate the signal behavior trends observed during the parameter optimization process. Comparison of defect maps generated before and after applying the CNN-optimized signal parameters revealed significant enhancements in detection accuracy. The findings highlight the effectiveness of integrating advanced signal analysis and deep learning techniques with impact-echo testing, offering a robust non-destructive evaluation approach for large-scaled infrastructures such as bridge deck condition assessment.