Subsurface Voids Detection of Concrete-Filled Steel Tubular (CFST) Structures Using Percussion in Conjunction with Decision Tree

The detection of interface debonding, such as subsurface voids, is a burdensome and laborious task for concrete-filled steel tubular (CFST) structures. In this study, we present a novel but simple method to detect the voids detects exist between core concrete and steel tubular in CFST components. A machine learning (ML) model based on a decision tree was trained and tested. Three voids were mimicked. The depths of the voids are 3cm, 5cm, and 5cm, respectively. A feature extraction approach based on power spectrum density (PSD) was employed, and nine features were obtained by summarizing the PSDs from 1 kHz to 10 kHz with a step of 1kHz. Experimental results show that the accuracy of the 100 times repeated test is 96.33%.