|
Title |
Finite Element Model Updating Technique Based on Bayesian Optimization Using Bridge Dynamic Characteristics
|
|
Authors |
심형민(Hyeong-Min Shim) ; 송민규(Min-Kyu Song) ; 이종한(Jong-Han Lee) |
|
DOI |
https://doi.org/10.11112/jksmi.2026.30.1.54 |
|
Keywords |
모델업데이팅; 교량; 베이지안 최적화; 유한요소해석; 깊은신경망 Model updating; Bridge; Bayesian optimization; FEM; DNN |
|
Abstract |
In this study, a model updating technique combining a finite element (FE) analysis model with a deep neural network (DNN) is proposed to evaluate the structural safety of deteriorated bridges. Traditional FE models have limitations in reflecting post-construction conditions of bridges, often leading to discrepancies between analytical and actual structural responses. Moreover, conventional optimization-based updating methods are time-consuming. To address these issues, the proposed method utilizes natural frequencies and mode shapes of the bridge as input data, allowing the DNN to predict member stiffness values. The DNN architecture is automatically optimized through Bayesian optimization. The proposed technique was validated through numerical simulations on PSC-I girder bridge models, demonstrating high accuracy with member stiffness prediction errors within 5% under various damage scenarios. Furthermore, its applicability to real structures was verified using measured data from an in-service railway bridge. The results confirm that the proposed method enables rapid and reliable updating of FE models for practical structural assessment.
|