Open Access

ARTICLE

Identification of Damage in Steel‒Concrete Composite Beams Based on Wavelet Analysis and Deep Learning

Chengpeng Zhang, Junfeng Shi^*, Caiping Huang

School of Civil Engineering, Architecture and Environment, Hubei University of Technology, Wuhan, 430068, China

* Corresponding Author: Junfeng Shi. Email:

(This article belongs to the Special Issue: Sensing Data Based Structural Health Monitoring in Engineering)

Structural Durability & Health Monitoring 2024, 18(4), 465-483. https://doi.org/10.32604/sdhm.2024.048705

Received 15 December 2023; Accepted 23 February 2024; Issue published 05 June 2024

Abstract

In this paper, an intelligent damage detection approach is proposed for steel-concrete composite beams based on deep learning and wavelet analysis. To demonstrate the feasibility of this approach, first, following the guidelines provided by relevant standards, steel-concrete composite beams are designed, and six different damage incidents are established. Second, a steel ball is used for free-fall excitation on the surface of the steel-concrete composite beams and a low-temperature-sensitive quasi-distributed long-gauge fiber Bragg grating (FBG) strain sensor is used to obtain the strain signals of the steel-concrete composite beams with different damage types. To reduce the effect of noise on the strain signals, several denoising techniques are applied to process the collected strain signals. Finally, to intelligently identify the strain signals of combined beams with different damage types, multiple deep learning models are constructed to train and to predict strain signals as denoised and not denoised, allowing for damage classification and localization in steel-concrete composite beams. In this experimental context, residual network-50 (ResNet-50) achieved the highest average accuracy compared to that of the other deep learning models. The average accuracy of the un-denoised and denoised signals is 96.73% and 97.91%, respectively, and wavelet denoising improved the prediction accuracy of ResNet-50 by 1.18%. The strain–time domain signals collected by sensors located farther from the damage zone also contain information about the damage to the composite beam. The deep learning models effectively extract damage features. The results of this experiment demonstrate that the approach used in this paper enhances the intelligence of structural damage identification.

---

Keywords

---