Detection of Grouting Quality in Prestressed Box Girders via Improved MobileNet and Impact-Echo

The impact-echo (IE) method has been widely applied in assessing the grouting quality of prestressed ducts in bridges due to its simplicity, low cost, and requirement of access to only one surface. However, current applications of this method still face challenges such as strong subjectivity and heavy reliance on operator experience in interpreting the results. To address these limitations, this study proposes a grouting compactness detection approach for prestressed box girder webs based on deep learning and the impact-echo technique. The raw time-domain signals are first transformed into two-dimensional time-frequency spectrograms using continuous wavelet transform. These spectrograms are then input into a convolutional neural network (CNN) for feature extraction, model training, and intelligent classification. Laboratory experiments demonstrate that the improved SE-MobileNet model effectively captures and learns the differences in time-frequency features between well-grouted and defective signals, enabling accurate diagnosis of grouting defects in prestressed webs. This approach reduces the subjectivity and dependence on expert interpretation. The feasibility of the proposed method is further validated through field tests.