Ultrasonic-Based Stress Identification of a Reinforced Concrete Beam via USR-Net

In recent two decades, though great progress has been made in stress identification of reinforced concrete structures using the ultrasonic-based coda wave interference technics, the state of art techniques in this field seems to be capable of considering only a single loading pattern. To address this issue, in this paper, a USR-Net based on DL algorithm is proposed for stress identification of reinforced concrete beams under two different load patterns. It is featured by a successive use of S-transform and Residual Blocks, with the latter’s conventional “square” convolution kernel replaced by a “flat” one, which gives play to the pictorial characteristics of the time-frequency matrix from the former for a more efficient and accurate stress recognition. The effectiveness of USR-Net is validated on experimental tests of two concrete I-beams under respectively a three-point bending and an eccentric compression load. Compared with traditional non-DL ultrasonic based algorithms, the USR-Net is more adaptive for different load patterns and robust to noises, providing a potentially novel solution for stress identification of practical concrete structures.