Open Access

ARTICLE

Feature Extraction and Intelligent Model Updating of Cable-Stayed Bridges Based on Multi-Point Dynamic Strain Measurements under Complex Operational Conditions

Yongning Zhang¹, Dongxue Li^1,2,*, Cen Yang³, Yongwang Gui⁴

1 School of Architecture and Civil Engineering, Chongqing Metropolitan College of Science and Technology, Chongqing, China
2 School of Civil Engineering, Chongqing University, Chongqing, China
3 Institute of Engineering Design & Research Chongqing Jiaotong University, Chongqing, China
4 Chongqing Zesheng Engineering Technology Co., Ltd., Chongqing, China

* Corresponding Author: Dongxue Li. Email:

(This article belongs to the Special Issue: Sustainable and Resilient Civil Infrastructure with Intelligence and Digital Transformation)

Structural Durability & Health Monitoring 2026, 20(5), 21 https://doi.org/10.32604/sdhm.2026.081767

Received 09 March 2026; Accepted 14 May 2026; Issue published 24 August 2026

Abstract

To address the challenge that the baseline state of FE models for operational highly statically indeterminate bridges is difficult to evaluate accurately, this paper proposes an intelligent multi-parameter inversion and updating framework driven by measured dynamic strains and a LSTM neural network. First, to tackle the complex environmental interferences coupled within short-term monitoring strain signals, a moving-window baseline detrending and refined thermal effect decoupling algorithm is employed. This successfully strips away long-term dead loads and temperature drift, extracting pure mechanical strain sequences with a high signal-to-noise ratio. Second, to overcome the mode omission issue caused by strain nodes at single measurement points, the multi-channel FDD method, combined with the stabilization diagram validation of the covariance-driven SSI-COV method, is utilized to robustly identify the first three baseline natural frequencies of the bridge under authentic operational conditions. Building upon this, an LSTM based nonlinear inverse surrogate model network is constructed, with macroscopic structural frequencies as inputs and local stiffness parameters as outputs. By feeding the measured frequencies into this network, the updated equivalent elastic moduli are obtained to update the FE model. The results demonstrate that the relative errors of the first three modal frequencies converged substantially from the initial 72.25%, 56.57%, and 46.56% down to 16.87%, 24.45%, and 11.18%, respectively. This framework effectively bridges the deviation in mechanical information between the theoretical model and the actual structure, significantly enhancing the fidelity of the digital twin baseline, and providing a reliable physical model foundation for subsequent seismic performance assessments and lifecycle health management of the bridge.

---

Keywords

---