Intelligent safety evaluation of tunnel lining cracks based on machine learning

Abstract

Tunnel crack-related issues are a major factor affecting operational safety. Traditional safety assessment methods cannot avoid the influence of subjective human factors, and the adoption of intelligent machine learning algorithms is emerging as a novel and effective approach. This study utilizes statistical data on tunnel cracks from existing projects and performs refined modeling of lining cracks using ABAQUS software and the Extended Finite Element Method (XFEM). The model's validity was confirmed based on laboratory test results. Subsequently, the safety levels of the cracks were assessed using two methods: ’crack tip stability’ and ’sectional bearing capacity’. Next, intelligent evaluation methods, including the Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Extremely Randomized Trees (ETC), and Support Vector Machine (SVC), were introduced. Crack length, depth, volume, area, type, and location were selected as features, and through training and parameter tuning of the four algorithms, a nonlinear mapping relationship between input and output parameters was established. Finally, the developed evaluation method was applied to the Chongqing Nancheng Tunnel. The results showed that the accuracy of machine learning-based intelligent evaluation methods exceeded that of traditional evaluation methods. This approach is expected to enhance the efficiency and accuracy of tunnel lining crack maintenance.