Xudong Qiao
Xudong Qiao has a total of 3 co-authors. Here are the top co-authors by number of shared publications:
- Daming Luo
College of Civil Engineering, Xi’an University of Architecture and Technology, Xi’an 710055, China
1 shared publication - Ditao Niu
School of Civil Engineering, Xi’an University of Architecture and Technology, Xi’an, China
1 shared publication - Fan Li1 shared publication
Latest Publications
- Open Access
- 1 January 2026
- Published by Wiley in Structural Control and Health Monitoring
- Vol. 2026 (1)
Utilizing Machine Learning for Ultrasonic Time–Frequency Spectrogram Recognition, Diagnosis, and Prediction of Concrete Damage Layer Thickness Under Freeze–Thaw Cycles
Traditional methods for diagnosing and predicting freeze–thaw damage in concrete often rely on macroscopic indicators such as mass loss and relative dynamic elastic modulus, which are difficult to reflect the heterogeneity of damage distribution in space. This study establishes an intelligent diagnosis and prediction method for concrete freeze–thaw damage layer thickness by applying deep learning to ultrasound signal processing. Firstly, a variational mode decomposition algorithm based on particle swarm optimization is proposed to denoise concrete ultrasonic signals. Subsequently, by combining 2D ultrasound time–frequency spectrum images, a ConvNeXt image recognition model incorporating CBAM attention mechanism was constructed to achieve efficient discrimination of different damage layer thicknesses in concrete. Finally, with the energy distribution of ultrasonic signals and material parameters as the core input features, the GWO algorithm was used to optimize XGBoost, and a prediction model for the thickness of concrete damage layer under freeze–thaw cycles was established, and SHAP was used for feature contribution interpretability analysis. The results indicate that the concrete ultrasonic signal denoising method proposed in this study can retain the main ultrasonic response characteristics while significantly suppressing noise interference. The accuracy of the ConvNeXt + CBAM model for identifying the thickness of concrete damage layers can reach 97.4%, and all evaluation indicators are significantly better than other classical convolutional neural networks and the model before improvement. The GWO‐XGBoost model performs the best among all indicators, with a correlation index R 2 of 0.9847, and can accurately predict the thickness of the concrete damage layer under freeze–thaw cycles. There is a significant positive and negative correlation between the ultrasonic signal components IMF8 and IMF1 and the thickness of the concrete damage layer, respectively. Compared to reducing the water–cement ratio, the addition of fibers has a more significant effect on improving the frost resistance of concrete, and the influence of polypropylene fiber content on the thickness of the damaged layer of concrete under freeze–thaw cycles is greater than that of basalt fibers.
- 1 January 2026
- Published by Wiley in Structural Control and Health Monitoring
- Vol. 2026 (1)
Open AccessUtilizing Machine Learning for Ultrasonic Time–Frequency Spectrogram Recognition, Diagnosis, and Prediction of Concrete Damage Layer Thickness Under Freeze–Thaw Cycles
Traditional methods for diagnosing and predicting freeze–thaw damage in concrete often rely on macroscopic indicators such as mass loss and relative dynamic elastic modulus, which are difficult to reflect the heterogeneity of damage distribution in space. This study establishes an intelligent diagnosis and prediction method for concrete freeze–thaw damage layer thickness by applying deep learning to ultrasound signal processing. Firstly, a variational mode decomposition algorithm based on particle swarm optimization is proposed to denoise concrete ultrasonic signals. Subsequently, by combining 2D ultrasound time–frequency spectrum images, a ConvNeXt image recognition model incorporating CBAM attention mechanism was constructed to achieve efficient discrimination of different damage layer thicknesses in concrete. Finally, with the energy distribution of ultrasonic signals and material parameters as the core input features, the GWO algorithm was used to optimize XGBoost, and a prediction model for the thickness of concrete damage layer under freeze–thaw cycles was established, and SHAP was used for feature contribution interpretability analysis. The results indicate that the concrete ultrasonic signal denoising method proposed in this study can retain the main ultrasonic response characteristics while significantly suppressing noise interference. The accuracy of the ConvNeXt + CBAM model for identifying the thickness of concrete damage layers can reach 97.4%, and all evaluation indicators are significantly better than other classical convolutional neural networks and the model before improvement. The GWO‐XGBoost model performs the best among all indicators, with a correlation index R 2 of 0.9847, and can accurately predict the thickness of the concrete damage layer under freeze–thaw cycles. There is a significant positive and negative correlation between the ultrasonic signal components IMF8 and IMF1 and the thickness of the concrete damage layer, respectively. Compared to reducing the water–cement ratio, the addition of fibers has a more significant effect on improving the frost resistance of concrete, and the influence of polypropylene fiber content on the thickness of the damaged layer of concrete under freeze–thaw cycles is greater than that of basalt fibers.