An innovative predictive method based on supervised teacher-student learning for forecasting limited structural responses of long-span bridges from satellite images

Forecasting the responses of large-scale civil structures offers an alternative to field measurement. Recently, spaceborne remote sensing technology has been increasingly adopted to monitor complicated and large structures. This approach involves extracting structural displacements from synthetic aperture radar images. To overcome some important restrictions associated with these images, the best solution is to exploit machine learning-aided prediction of displacement responses. For this purpose, it is necessary to measure key external factors, particularly environmental and operational conditions. In some cases, installing sensors for these factors may not be tractable, in which case some unmeasured and unknown conditions, which can affect structural responses, are not incorporated into the prediction process. To avoid poor performances and inaccurate forecasting outputs, this paper proposes a predictive solution using the idea of supervised teacher-student learning. This method consists of two parts of an elaborate regression model via a long-short-term-memory neural network acting as a teacher and a simple model through a single-hidden-layer feedforward neural network behaving as a student. The effectiveness and success of the proposed method are benchmarked by limited information of a long-span bridge. Results show that this method can adequately forecast limited bridge responses in the presence of the impacts of unmeasured predictors. © 2024, NDT. net GmbH and Co. KG. All rights reserved.