Two-step detection of concrete internal condition using array ultrasound and deep learning

Abstract

Internal defects in concrete structures pose a great risk and potential hazard to the safety and durability of civil structure. Reliable and accurate detection of the internal conditions of structures is a prerequisite for structural safety assessment. To realize the visualization of the internal condition of concrete structures and the identification of defects, this paper proposes a two-step detection method for the internal condition of concrete structures based on array ultrasound and deep learning. First, to address the limited resolution problems of conventional ultrasound synthetic aperture focusing technology (SAFT) imaging and the time-consuming computation of wavefield reconstruction methods such as reverse time migration (RTM) and full waveform inversion (FWI), an array-SAFT imaging algorithm is introduced to balance the resolution and real-time performance of the inspection results. The variational mode decomposition (VMD) and bilateral filtering methods are used before and after the signal focus imaging for signal noise reduction and image edge processing to further improve the resolution of ultrasound images while satisfying the real-time visualization of the detection results. Secondly, since pre-buried objects such as rebar and internal defects are reflected on the surface in ultrasonic images, a method based on the improved YOLOv5 network target detection model is proposed to achieve reliable identification of internal defects and intelligent interpretation of abnormal features in ultrasonic images. The proposed method is effectively validated in the ultrasonic inspection of concrete specimens with different emulated defects pre-built in the laboratory. Additionally, ultrasonic detection is also implemented in the internal defect detection of tunnel secondary linings.