Automated Defect Classification in Anchor Bolts using Machine Learning and Ultrasonic Signal Features

The structural integrity of anchor bolts in civil structures is critical to ensuring safety and operational efficiency. Traditional visual inspection methods often fail to detect internal defects, necessitating more advanced approaches. This study integrates machine learning and ultrasonic signal analysis to enhance the reliability and precision of defect detection in anchor bolts. Ultrasonic A-scan signals were analysed to extract key features for machine learning models. Gradient Boosting Classifier model emerges as the optimal model, achieving an accuracy of 97% in defect classification. The methodology targets bolt defects such as thinning and corrosion, utilizing normalised signal features to distinguish severity of defects and enable automated classification.