A data-driven based method for damage detection of combining joints and elements of frame structures using noisy incomplete data

Abstract

Most previous studies on damage detection in civil engineering structures have focused on either element damage detection or joint damage detection, separately. However, in practice, both elements and joints may be prone to damage, simultaneously. In addition, the development of effective numerical approaches to detect and quantify these damages in real time within structures is essential to ensure their integrity and reliable operation. Therefore, to deal with these problems, this study proposes an effective data-driven approach by using an attention based convolutional gated recurrent unit network (ACGRU) for real time damage detection of combining joint and element in frame structures using incomplete data. In the proposed approach, the convolutional layers in the network are utilized to extract critical features from the raw input data. Then, these extracted features are fed into the gated recurrent unit layers to learn to predict the desired output data. In addition, by introducing the attention mechanisms into the network, the important information can be effectively learned. The performance and applicability of the ACGRU are validated through two different numerical examples using incomplete data in both noise-free condition and noisy condition. Moreover, the effect of the numbers and placements of sensors on the damage detection results is also investigated. The damage detection results achieved by the proposed approach are compared with those of other state-of-the-art methods to inspect the reliability of the proposed approach.