AUTHOR=Zhou Xin , He Yuanpeng TITLE=Dynamic displacement estimation of structures using one-dimensional convolutional neural network JOURNAL=Frontiers in Physics VOLUME=11 YEAR=2023 URL=https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2023.1290880 DOI=10.3389/fphy.2023.1290880 ISSN=2296-424X ABSTRACT=

For large infrastructures, dynamic displacement measurement in structures is an essential topic. However, limitations imposed by the installation location of the displacement sensor can lead to measurement difficulties. Accelerometers are characterized by easy installation, good stability and high sensitivity. For this regard, this paper proposes a structural dynamic displacement estimation method based on a one-dimensional convolutional neural network and acceleration data. It models the complex relationship between acceleration signals and dynamic displacement information. In order to verify the reliability of the proposed method, a finite element-based frame structure was created. Accelerations and displacements were collected for each node of the frame model under seismic response. Then, a dynamic displacement estimation dataset is constructed using the acceleration time series signal as features and the displacement signal at a certain moment as target. In addition, a typical neural network was used for a comparative study. The results indicated that the error of the neural network model in the dynamic displacement estimation task was 9.52 times higher than that of the one-dimensional convolutional neural network model. Meanwhile, the proposed modelling scheme has stronger noise immunity. In order to validate the utility of the proposed method, data from a real frame structure was collected. The test results showed that the proposed method has a mean square error of only 5.097 in the real dynamic displacement estimation task, which meets the engineering needs. Afterwards, the outputs of each layer in the dynamic displacement estimation model are visualized to emphasize the displacement calculation process of the convolutional neural network.