Abstract
The aging of concrete structures in Japan is becoming increasingly serious. Periodic inspection is necessary to prevent accidents caused by aging. One of the methods used to inspect concrete is the hammering test. In this research, we aim to develop a system to identify the topology of defects in concrete by machine learning based on the acceleration response data obtained from the hammering test. As a machine learning model, we build a neural network based on self-attention. Furthermore, we propose a data augmentation method for this task and test its effectiveness.
