Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Bridge damage identification using a small amount of damage labeling data
AbstractThis paper proposes a method for bridge damage identification using a small amount of damage labeling data. This method first trains a deep neural network (DNN) with undamaged bridge inclination responses as inputs and bridge equivalent loads as labels. The ratio curve related to the bridge damage state can be obtained by quantifying the change in the DNN prediction error before and after bridge damage. Then, this method achieves the efficient calculation of ratio curves corresponding to different damage states based on finite element static simulation, and damage index curves calculated based on ratio curves are used to produce bridge damage localization labeling data to achieve bridge damage localization. Finally, the quantification of bridge damage can be achieved by only calculating the ratio curves of different damage degrees at the damage location. The proposed method not only overcomes the limitations of high modeling cost, low efficiency, and poor robustness to measurement noise and modeling errors of the finite element dynamic simulation method in producing damage labeling data to some extent but also can achieve bridge damage localization by using only the damage labeling data of a single damage degree at each damage location, and can achieve the approximate prediction of multi‐damage locations without including multi‐damage localization labeling data. The feasibility of the proposed method under conditions of unknown loads, a small number of sensors, and the presence of modeling errors and measurement noise is verified by numerical simulations.
Bridge damage identification using a small amount of damage labeling data
AbstractThis paper proposes a method for bridge damage identification using a small amount of damage labeling data. This method first trains a deep neural network (DNN) with undamaged bridge inclination responses as inputs and bridge equivalent loads as labels. The ratio curve related to the bridge damage state can be obtained by quantifying the change in the DNN prediction error before and after bridge damage. Then, this method achieves the efficient calculation of ratio curves corresponding to different damage states based on finite element static simulation, and damage index curves calculated based on ratio curves are used to produce bridge damage localization labeling data to achieve bridge damage localization. Finally, the quantification of bridge damage can be achieved by only calculating the ratio curves of different damage degrees at the damage location. The proposed method not only overcomes the limitations of high modeling cost, low efficiency, and poor robustness to measurement noise and modeling errors of the finite element dynamic simulation method in producing damage labeling data to some extent but also can achieve bridge damage localization by using only the damage labeling data of a single damage degree at each damage location, and can achieve the approximate prediction of multi‐damage locations without including multi‐damage localization labeling data. The feasibility of the proposed method under conditions of unknown loads, a small number of sensors, and the presence of modeling errors and measurement noise is verified by numerical simulations.
Bridge damage identification using a small amount of damage labeling data
Computer aided Civil Eng
Sun, Hongshuo (Autor:in) / Song, Li (Autor:in) / Yu, Zhiwu (Autor:in)
25.03.2025
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Bridge Damage Identification Using Rotation Measurement
| ASCE | 2023
Bridge Damage Identification Using Artificial Neural Networks
| British Library Online Contents | 2018
Bridge damage identification through frequency changes
| SPIE | 2021
Model-Based Damage Identification in a Continuous Bridge Using Vibration Data
| Online Contents | 2010