ABSTRACT
In this study, the hammering method for concrete structures using wearable devices was investigated. First, comparative studies of several wearable microphones revealed that the response varied depending on where the microphones were attached. As a result, the helmet type demonstrated the best performance. Next, features of sound pressure and input force waveforms were calculated and evaluated for their importance in defect detection. Then, a fully connected neural network was constructed and the evaluated features were selected as input parameters. Cross-validation was performed using the dataset obtained from a concrete wall with various artificial voids. As a result, the defect detection rate was around 94%, including voids at 10 cm depth at maximum.
