Percussion-based quasi real-time void detection for concrete-filled steel tubular structures using dense learned features

•A percussion-based void detection method using dense learned features was proposed.•The void detector leverages CNN, GAP, and CAM to obtain high accuracy.•The void detector listens to the “echo” part carefully, rather than the “impact” part.•The classification results depend on both the vibration frequency and damping ratio.•High classification accuracies (more than 99%) was obtained.