Intelligent Recognition of Joint Positions in Surrounding Rock of Tunnels Based on Interpretable Machine Learning Models

Journal of Ocean Engineering and Marine Energy · Springer Nature · 2026-10-16 · DOI: 10.1007/s40098-026-01639-7

Аннотация

The existence of joints in the surrounding rock of tunnels has a significant impact on the stability and safety of tunnels, and accurate prediction of joint positions is crucial for early warning monitoring and support reinforcement of tunnel engineering. The purpose of this study is to use acoustic emission (AE) technology combined with machine learning (ML) algorithm to propose a new method to predict the position of joints in the surrounding rock of tunnels, so as to provide a scientific basis for the safety assessment of tunnel engineering. In this research, non-destructive testing of jointed sandstone specimens was carried out by AE technology, and four energy characteristics of feature-derived Cumulative energy, Energy rate, Kurtosis energy and Root-mean-square energy were simultaneously used. Four ensemble algorithms of ML—Random Forest (RF), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost)—were used to classify and predict the joint positions. The LightGBM model with the best performance was selected, and the Shapley additive interpretation (SHAP) method was combined to reveal the nonlinear relationship between joint positions and AE parameters. The results show that Hit number is the most important feature affecting the prediction of rock joint position, followed by Kurtosis energy, Root-mean-square energy, Peak frequency, Duration time and Counts. In contrast, Energy rate and Rise count have less influence on the prediction of joint position. In addition, different AE parameter features have different patterns and degrees of influence on the prediction of joints at different positions, and there are obvious dependencies between the features. This study effectively combines the joint position of surrounding rock with the AE parameters during the failure process, providing a new tool with wide application prospects for predicting the joint position of surrounding rock. The research results provide new research ideas for the fields of rock mechanics and geological engineering.

Читать у издателя →

Похожие статьи