ABSTRACT
loss of structural dynamic response measurements at the carefully selected locations of structural health monitoring (shm) systems can happen because of the uncontrollable in-field measurement condition. Investigation of reconstructing the lost response is an urgent need for shm. This paper proposes a response reconstruction method based on a specially designed deep learning model, that is a fully convolutional neural network. It features skip connections and densely connected blocks for improving information flow. The network is trained by a supervised manner with input and output defined as responses of available and desired locations. The effectiveness and robustness of the developed approach are validated by using measured acceleration data from guangzhou new tv tower. The proposed method exhibits a strong capability on structural response reconstruction, which effectively predicts the useful information of unavailable responses and eliminates measurement noise.
