Application of data augmentation techniques in tunnel health assessment

Abstract

Tunnel health assessment is an important part of ensuring the structural safety and extending the service life of tunnels. However, limited by the problems of insufficient data and class imbalance in the monitoring of tunnel defects, the model may face the prediction bias during the training process. Therefore, this study introduces a tunnel health assessment method based on data augmentation to improve the classification performance and generalization ability of the model. First, the defects monitoring data of the left line of the Huilongshan Tunnel in Shaoguan City, Guangdong Province were collected, a true dataset containing 13 defect indicators was established, and preprocessing operations such as feature transformation, outlier detection and handling, missing value filling, and normalization were performed on it. Then, three data augmentation methods, CTGAN, SMOTE, and CVAE, were used to augment the dataset to generate the synthetic datasets, D _s1 , D _s2 , and D _s3 , respectively. The similarity between the synthetic datasets and the true dataset was assessed using statistical methods, including statistical indicators, boxplots, and Q-Q plots. The effectiveness of data augmentation was then validated using three machine/deep learning models, BP neural networks, SVM, and XGBoost. The experimental results show that the synthetic dataset D _s1 generated by CTGAN performed the best in terms of accuracy (98.47%), precision (98.05%), recall (98.10%), and F1 score (98.06%), significantly improving the model’s classification performance while effectively mitigating the problems of insufficient data and class imbalance. Overall, this study demonstrates the superiority of the CTGAN method in tunnel health assessment tasks and provides a reliable data augmentation solution for tunnel health assessment.