A CNN-LSTM Hybrid Model for Subtle Damage Identification in Structural Health Monitoring

Authors

  • He Bai School of Civil Engineering, Changchun Institute of Technology, Changchun, Jilin, 130012, China
  • Heqi Zhang School of Civil Engineering, Changchun Institute of Technology, Changchun, Jilin, 130012, China
  • Jincheng Li School of Civil Engineering, Changchun Institute of Technology, Changchun, Jilin, 130012, China

DOI:

https://doi.org/10.54691/kdk9gp78

Keywords:

Structural Health Monitoring; CNN-LSTM; Deep Learning; Damage Identification; Time-series Analysis; False-alarm Rate.

Abstract

Time-series data in structural health monitoring (SHM) contain both local anomaly features, such as crack jumps and strain sudden changes, and long-term evolution dependencies, such as cumulative deformation, which a single model can hardly extract simultaneously. A CNN-LSTM hybrid damage identification model is therefore constructed: one-dimensional convolutional layers extract transient anomaly features through local receptive fields, while LSTM layers model long-term temporal dependencies through gating mechanisms, and the two are cascaded to form an identification pipeline of "local feature extraction - temporal dependency modeling - damage state discrimination". Model training and validation rely on real field data obtained from 27 measurement points over 8.5 months of continuous monitoring, including 24710 crack records, 55832 tilt records and 86181 strain records, and a labeled identification dataset is constructed by injecting artificial subtle damage signals into normal sequences. Comparative experiments show that the CNN-LSTM hybrid model achieves an accuracy of 96.4%, a recall of 94.7%, an F1 score of 0.956 and a false-alarm rate of 1.8%, comprehensively outperforming standalone CNN, standalone LSTM, SVM and the traditional threshold method; compared with the standalone CNN, the accuracy is improved by 4.3 percentage points, the recall by 9.4 percentage points, and the false-alarm rate is reduced by 2.8 percentage points. Ablation analysis indicates that the complementary mechanism of convolutional and recurrent layers is the key to the performance gain. The validity boundaries concerning artificially labeled damage and the lack of cross-point transfer verification are also discussed, providing a basis for applying deep learning methods to the early diagnosis of subtle structural damage.

Downloads

Download data is not yet available.

References

[1] Tang, D. Y., Wei, X. B., Sun, Z. H., et al. (2023). Review of structural health monitoring systems. Intelligent Building and Smart City, (8), 6–11.

[2] Jiao, K., Lai, H. L., Peng, Z. X., et al. (2021). Research on key technologies of an IoT-based life-cycle safety operation service system for existing buildings. Industrial Construction, 51(10), 201–210.

[3] Bai, B., Chen, Y., Du, C. Q., et al. (2020). Intelligent safety monitoring building information model based on IoT technology. Industrial Construction, 50(4), 175–179.

[4] Luo, Y. Z., & Zhao, J. Y. (2022). Research status and prospects of structural health monitoring of spatial structures. Journal of Building Structures, 43(10), 16–28.

[5] Fadel, Y., & Zhu, J. (2023). Classification and regression-based convolutional neural network and long short-term memory configuration for bridge damage identification using long-term monitoring vibration data. Structural Health Monitoring, 22(6), 4027–4054. https://doi.org/10.1177/1475921723116047.

[6] Tokognon, C. A., Gao, B., Tian, G. Y., et al. (2017). Structural health monitoring framework based on internet of things: a survey. IEEE Internet of Things Journal, 4(3), 619–635. https://doi.org/10.1109/JIOT.2017.2674466.

[7] Azimi, M., Eslamlou, A. D., & Pekcan, G. (2020). Data-driven structural health monitoring and damage detection through deep learning: state-of-the-art review. Sensors, 20(10), 2778. https://doi.org/10.3390/s20102778.

[8] Liu, Y. F., Qi, Y., Li, B. L., et al. (2022). Intelligent identification of sensitive service indices for multi-scenario structural operation and maintenance. Journal of Building Structures, 43(10), 1–15.

[9] Yue, Q. R., Liu, X. G., & Chen, H. B. (2022). Research and application progress of intelligent diagnosis for structural service safety. Journal of Building Structures, 43(10), 41–49.

[10] Xia, Z. Q., Ma, L. Y., Shan, J. Z., et al. (2023). Two-stage detection method for building facade spalling and cracks based on computer vision. Journal of Building Structures, 44(2), 207–216.

[11] Yang, Y., Wang, Z. W., Ling, Y., et al. (2023). Damage identification of frame structures based on statistical moment theory and Bayesian method. Journal of Building Structures, 44(2), 217–226.

[12] Li, R., & Zhang, C. (2021). Structural damage detection method based on one-dimensional dilated convolutional neural network. Industrial Construction, 51(10), 170–176.

Downloads

Published

2026-11-20

Issue

Section

Articles

How to Cite

Bai, H., Zhang, H., & Li, J. (2026). A CNN-LSTM Hybrid Model for Subtle Damage Identification in Structural Health Monitoring. Scientific Journal of Technology, 8(11), 54-60. https://doi.org/10.54691/kdk9gp78