Data-driven based ultrasonics analysis for evaluating the bond strength of concrete layers

Abstract

In accelerated bridge constructions, the 3D printing process, and structural retrofitting, the bond condition between concrete layers considerably affects their structural safety and functionality. Nondestructive tests can effectively and efficiently monitor these structures' bond conditions. This paper used the convolutional neural network (CNN) and multilayer perceptron (MLP) to estimate the bond strength of concrete layers based on vital ultrasound parameters. In this regard, ultrasound pulse velocity, attenuation, maximum amplitude, and energy loss of the ultrasound wave were assessed as the inputs of the machine learning models. The target was the bi-surface shear strength of the specimens. Two different approaches were considered to evaluate the energy loss of the wave 1) in the whole specimen and 2) at the bond zone, and the models based on each dataset's accuracy were compared. Parameters resulting from the second approach were more reliable and had better consistency with bond strength variation than the first one. Moreover, the ultrasonic parameters were investigated to determine the most sensitive ultrasound factor to the different bond treatments. Accordingly, attenuation was the most susceptible parameter toward the change of bond strength.