Acoustic Emission Monitoring of Concrete Plate-like Structures Using Deep Transfer Learning and Beamforming

The important task of structural health monitoring in civil engineering requires sensitive detection and accurate location of the damage. Acoustic emission (AE) technology is of broad concern because of its potential advantages in damage monitoring and source localization. However, traditional AE localization methods are based on the detection of P-wave arrival, which do not consider the heterogeneity of concrete materials, resulting in poor accuracy in a slightly larger range. With rapid developments in deep learning technologies and air-coupled sensors, this study provides a feasible approach for automated monitoring of concrete plate-like structures based on a new air-coupled MEMS microphone array unit. We adapted a deep neural network (DNN) model trained on seismic data for Brazilian disc tests. The trained model was examined using AE signals that were simulated in artificial AE source localization tests, and AE events were automatically extracted using the transfer-learning (TL) model. A TL-aided beamforming method was proposed to determine the azimuth of the artificial AE source at different positions at a high environment noise level; this accuracy is sufficient for most field monitoring applications. The method in this study is a preliminary study for forecasting failure in large concrete plate-like structures, such as tunnel linings and bridge decks.