ABSTRACT
This study examined the applicability of vibration measurement using a lightweight and compact automatic impactor and anomaly detection using machine learning as methods of simple and time-efficient non-destructive testing to evaluate damages in RC-members. In the experimental study, three RC beam specimens of different dimensions were made without stirrups to shear failure, and loading tests were conducted on each specimen. The training data for machine learning was collected before and after the loading tests, respectively, using an automatic impactor. Based on the vibration information obtained, a machine learning model was constructed using the data set and the autoencoder. The results showed that the anomaly values were larger in areas with shear cracks and wider crack widths, indicating the possibility of detecting shear cracks that are structurally significant with these methods of no-destructive testing.
