Unsupervised Domain Adaptation Damage Identification Approach of High Arch Dams after Earthquakes

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