Deep learning-based autonomous damage-sensitive feature extraction for impedance-based prestress monitoring

Abstract

In the electromechanical impedance-based technique, the selection of proper impedance features and frequency bands has played a significant role in enhancing the results of structural damage assessment. Using hand-crafted features or inappropriate frequency bands could lead to the false alarm of structural damage and the erroneous estimation of severity and further prevent the usage of the technique for real-time structural health monitoring. This study proposes a deep learning-based autonomous feature extraction approach for impedance-based damage monitoring. A 1-dimensional convolutional neural network (1-D CNN) model is developed to automatically extract and directly learn the optimal features of damage from the raw impedance signals. The feasibility of the proposed approach is demonstrated via monitoring the prestress-loss of a post-tensioned reinforced concrete girder. As the result, it is shown that the proposed technique successfully estimates the true severity of prestress-loss in the girder, even for untrained prestress cases.