@article {10.3844/ajeassp.2024.102.115,
article_type = {journal},
title = {Defect Classification of Reinforced Concrete Structures with Nondestructive Tests Using Statistical and Machine Learning Methods},
author = {Sayyar-Roudsari, Sajjad and Shalbaftabar, Armaghan and Damirchilo, Farshid and Taslimian, Rohollah and Abu-Lebdeh, Taher and Hamoush, Sameer and Yi, Sun},
volume = {17},
number = {3},
year = {2024},
month = {Jun},
pages = {102-115},
doi = {10.3844/ajeassp.2024.102.115},
url = {https://thescipub.com/abstract/ajeassp.2024.102.115},
abstract = {In this study, twelve reinforced concrete beams were constructed, each with specified dimensions and initial compressive strength. The beams were divided into four groups: A control group without any defect, a void group featuring a centrally located void, a corrosion group and a debonding group. The impact echo test was used for nondestructive testing, gathering data on compressive and shear wave velocity and frequency. The collected data, including compressive and shear wave velocity, frequency and derived material properties as well as modulus of elasticity, were used for subsequent analyses. To determine the type of defect, artificial intelligence and machine learning methods were utilized. Data collected from the impact echo method were analyzed using RStudio and the MATLAB toolbox for statistical analysis. Linear regression was employed to establish relationships between inputs (wave velocity and frequency) and outputs as shear and compressive modulus. The accuracy of these relationships was assessed through correlation coefficients, p-values and adjusted R-squared error. Additionally, an Imperial Competitive Algorithm (ICA) as part of the artificial neural network method was implemented to predict the variables. The results demonstrated high correlation coefficients and low mean square errors, indicating accurate predictions. Frequency domain defect detection was performed by analyzing frequency-amplitude data. The MATLAB toolbox was used to identify peaks and determine defects based on a 20% boundary condition. The comparison of peaks confirmed the presence of defects in beams with voids, corrosion and debonding. Subsequently, support vector machines were employed to classify defects in reinforced concrete structures, including voids, corrosion and debonding. This study utilized key features of reinforced concrete and assessed SVM performance using precision, recall and F1-score metrics. Overall, this study illustrates the effectiveness of machine learning techniques complied with impact echo tests in assessing and predicting the quality of reinforced concrete beams with various internal defects},
journal = {American Journal of Engineering and Applied Sciences},
publisher = {Science Publications}
}