Joint damage detection of structures with noisy data by an effective deep learning framework using autoencoder-convolutional gated recurrent unit

Recent advances in chip and sensor technology allow a large amount of data to be collected for serving structural damage detection (SDD). However, how to efficiently utilize and transform these complex sensing data into useful engineering information has remained many challenges. One of the major challenges in SDD using sensing data is how to effectively extract features that are sensitive to damages from measured data under the effect of noise factors. In order to deal with this issue, this paper proposes a new deep learning (DL) framework using an autoencoder-convolutional gated recurrent unit (A-CGRU) for structural joint damage detection using noisy data. In the proposed approach, the autoencoder component is used for noise removal for the measured noise data and the output of the autoencoder is then fed into the convolutional component to automatically determine the important features. Finally, the latent features extracted from the convolutional component are fed into the gated recurrent unit to learn to predict the location and severity of damaged joints in structures. The performance and applicability of the proposed A-CGRU are validated through various joint damage scenarios in a two-story planar frame structure and a four-story planar frame structure. The outcomes achieved by the proposed method are compared with those of other methods in order to verify the accuracy and reliability of the proposed A-CGRU method. In addition, the influence of various parameters such as the number and location of sensors, the number of measured frequencies and mode shapes, as well as various noise levels on the capability and stability of the proposed method is also investigated. The obtained results show the accuracy, efficiency, and applicability of the proposed A-CGRU method in structural joint damage detection using noisy data.