A deep learning-based bridge damage detection and localization method. (15th June 2023)

• Record Type: Journal Article
• Authors: Sun, Hongshuo; Song, Li; Yu, Zhiwu
• Is Part Of: Mechanical systems and signal processing. Volume 193(2023)
• Publication Date: 2023-06-15
• Subjects: Bridge damage detection -- Partial least-squares regression -- Deep learning -- Nodal load; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2023.110277
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: A method for estimating excessive nodal loads is proposed. Damage detection method that does not require damage information to train networks. A real-time detection and localization method for bridge damage. Abstract: Existing studies have utilized highly efficient partial least-squares regression (PLSR) to estimate nodal loads of the entire bridge using a small number of bridge sensors, and when the structure is damaged, the estimated nodal loads include damage information. Based on the ability of convolutional neural networks (CNNs) that can learn the PLSR method to estimate nodal loads, this paper proposes a bridge damage detection and localization method using inclination or deflection measurements. First, this study develops a method for estimating excessive nodal loads and establishes a framework for bridge damage detection and localization utilizing the change in the deviation of excessive nodal loads estimated by a CNN and the PLSR method before and after structural damage. Then, a CNN model is designed in this study, and the CNN model establishes a mathematical relationship between the monitoring point response as input and the estimated excessive nodal load as output through training. Finally, the detection and localization of bridge damage are realized using the proposed calculation method of damage indicator. The proposed method avoids costly finite element modeling and does not require difficult-to-obtain real structural damage information to train network models, and can achieve real-time detection and localization of bridge damage with a small number of sensors installed. Numerical simulations show that the proposed method can detect and locate damage very accurately and reliably in the presence of unknown loads, multi-damage, and measurement errors, revealing its potential in the field of bridge damage detection.