No abstract

In actual concrete arch dam engineering scenarios, the dynamic data obtained by the health monitoring system of an arch dam are incomplete. The data acquired typically depend on the state of the dam structure, that is, whether it is intact or incomplete. Besides, the future environmental loads of the structure are unpredictable. Thus, environmental noise is also uncertain. In practical engineering, the use of a damage identification model constructed based on incomplete information is problematic in scenarios with variable loads. Consequently, detecting the water level in actual arch dam projects after an earthquake and determining the impact of environmental uncertainty are necessary. Accordingly, this paper proposes a denoising contractive sparse deep autoencoder (DCS-DAE) model based on domain adaptation. The core idea of the proposed method is to constrain the data probability distribution of feature spaces in the source and target domains using maximum mean discrepancy. This fusion enables the DCS-DAE model to be capable of feature extraction. Moreover, it resolves the problem in which the objective function cannot be applied to other similar scenarios because of the lack of consistency constraints of feature spaces in the source and target domains. Four working conditions are designed to reproduce the uncertainty of structural modeling and the variability of water levels. The conditions are based on the postseismic water level detection requisites of dams in practical engineering. The results show that the proposed anomaly detection model enhances the generalization performance of the DCS-DAE in terms of feature design. Hence, the constructed model can “infer other things from one fact.” The results of this study are meaningful for the real-time cross-domain monitoring of structures under variable load conditions, providing a driving force to apply similar methods to practical arch dam projects.

No abstract