.png){kind=icon}
.png){kind=icon}
Return
Bridge damping ratio identification based on function approximation-guided physics-informed neural networks
.png){kind=icon}
PRE
AI
DOI：10.1016/j.istruc.2025.108540.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Reducing the reliance on large amounts of vibration data can enhance the efficiency and applicability of the damping ratio identification. Therefore, a function approximation-guided physics-informed neural network (FAPINN) is proposed in this study. Firstly, a hybrid output layer is designed and composed of adaptively weighted contributions from both the self-output and the function approximation. Secondly, the challenges in the convergence of the conventional physics-informed neural network (PINN) for solving the bridge-free vibration problem are analyzed, and the identification results of the conventional and improved methods are compared. Finally, the influence of the output layer weight update path on the damping ratio identification result is discussed. The results illustrate that learning the inherent laws of the physical system is the key to network convergence. Based on the designed hybrid output layer, the proposed method accurately identifies the firstorder modal damping ratio of the bridge with an error of 0.07 %. The proposed method utilizes vibration data from only three displacement sensors to identify accurately.
Keywords:
Bridge damping ratio
Physics-informed neural network
Hybrid output layer
Function approximation
Vibration