Deep Learning-Based Guided Wave Method for Semi-Grouting Sleeve Detection

Semi-grouting sleeves are widely used in various types of prefabricated buildings, and their defects have an important impact on the stability of the structure. The current defect detection methods have the disadvantages of low accuracy and complex data processing processes. In this paper, we propose a deep learning-based guided wave detection method. The method consists of two parts: (a) a frequency attention convolutional autoencoder (FACAE), the FACAE is inserted with a frequency attention module (FAM) which makes the model pay more attention to more important channels, and it compresses the guided wave signal to a global feature vector (GFV); (b) an Xgboost regression model which is used to predict the defect rate. A dataset for training was generated by finite element simulation and a set of semi-grouting sleeve experiments were used to verify our method. The results show that our method predicts the defect rate within a 8% error.