DEFECT IDENTIFICATION IN CONCRETE USING DEEP LEARNING TECHNIQUE ON NONLINEAR ULTRASONIC WAVE

Abstract

Nonlinear ultrasonic techniques have emerged as a promising tool for detecting defect in concrete. However, detecting the first and higher harmonics of ultrasonic waves in concrete can be challenging due to its complexity and heterogeneity nature characteristics. In this study, a deep learning algorithm was used to improve the accuracy of defect identification. Experiments were conducted on three concrete block samples, including pure concrete and two concrete samples with inclusions. The study utilized an array of R6 sensors as transmitters and an array of R15 sensors as receivers for the measurements. The deep learning algorithm was applied to the wavelet spectrogram of each wave, using 1050 images for training, 116 images for validation, and 292 images for testing. Convolutional neural networks (CNN) were used in the deep learning model. The approach focused on the regions in the first and second harmonic, which are more representative of defect, in the deep learning method. The proposed network consists of several layers that perform different operations to extract relevant features from the input data. The experiments demonstrated the effectiveness of using deep learning algorithms for identifying and classifying defect in concrete. The model achieved an overall accuracy of 94.8% in detecting defect in the concrete samples, with a high precision score for both defect and no-defect identification. This approach successfully detected defect in concrete samples, including the presence of inclusions. As a result, the study showed that deep learning algorithms can be effective in identifying and classifying defect in concrete, with the potential to improve the maintenance and management of concrete structures, enhancing their safety and durability.