Dam safety assessment through data-level anomaly detection and information fusion

Abstract

The anomaly detection and safety assessment of dams have attracted increasing attentions. To assess the safety of dams, a novel dam safety assessment model is proposed. The safety assessment model consists of data-level anomaly detection and information fusion. In the anomaly detection, a novel model based on the generative adversarial network and variational autoencoder is proposed. Through the stepwise training, the anomaly scores with reconstruction and discrimination losses can be obtained. Considering the multi-source and complexity of information system, a two-step fusion model is proposed. In the multi-source and multi-attribute fusion, the neighborhood rough set model is employed to construct the neighborhood granular structure. Through the idea of granular computing, the uncertainty of these sources and attributes is measured. Then the Sup-Inf fusion functions and fusion weights are employed to integrate the information. Finally, the anomaly scores with interval form are obtained. For verification, an arch dam is taken as an example. Through the analysis of detection and fusion results, the proposed model can detect the abnormal signals accurately and have the comprehensive assessment of the dam.

A brief introductory review to deep generative models for civil structural health monitoring

“…The use of VAEs in SHM can go back to the early 2020s (Liu et al, 2019b ; Ma et al, 2020 ), presenting anomaly detection on railways and feature extraction via VAE. Since then, several studies have been presented employing the generative skill of VAEs in civil SHM for various purposes, such as damage and anomaly identification, and condition assessment (Anaissi et al, 2023 ; Pollastro et al, 2022 ; Xu et al, 2021b ; Yuan et al, 2021 ; Zhou et al, 2022 ), and optimal sensor placement (Sajedi & Liang, 2022 ) (Fig. 2 ), addressing data scarcity challenge in the SHM domain in one way or another.…”