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

Area of Science:

• Civil Engineering
• Materials Science
• Non-Destructive Testing

Background:

• The bond condition between concrete layers is critical for structural integrity in accelerated bridge construction, 3D printing, and retrofitting.
• Nondestructive testing (NDT) offers an efficient method for monitoring these crucial bond conditions.
• Assessing bond strength is essential for ensuring the safety and functionality of concrete structures.

Purpose of the Study:

• To develop and compare machine learning models for estimating concrete layer bond strength using ultrasound parameters.
• To identify the most sensitive ultrasound parameter for detecting variations in bond strength due to different treatments.
• To evaluate the effectiveness of different energy loss calculation methods in predicting bond strength.

Main Methods:

• Utilized Convolutional Neural Network (CNN) and Multilayer Perceptron (MLP) models.
• Input parameters included ultrasound pulse velocity, attenuation, maximum amplitude, and energy loss.
• Compared two energy loss evaluation approaches: whole specimen versus bond zone analysis.

Main Results:

• Machine learning models accurately estimated bi-surface shear strength based on ultrasound parameters.
• Evaluating energy loss specifically at the bond zone yielded more reliable and consistent results than whole specimen analysis.
• Ultrasound attenuation emerged as the most sensitive parameter to changes in bond strength, indicating its effectiveness in detecting bond variations.

Conclusions:

• Machine learning, particularly CNN and MLP, combined with ultrasound testing, provides a reliable method for assessing concrete layer bond strength.
• Focusing ultrasound energy loss analysis on the bond zone enhances prediction accuracy.
• Ultrasound attenuation is a key indicator for monitoring bond quality and detecting debonding in concrete structures.