Automated Recognition Model of Geomechanical Information Based on Operational Data of Tunneling Boring Machines

• Published in: Rock mechanics and rock engineering Vol. 55; no. 3; pp. 1499 - 1516
• Main Authors: Yang, Haiqing, Song, Kanglei, Zhou, Jiayuan
• Format: Journal Article
• Language: English
• Published: Vienna Springer Vienna 01.03.2022; Springer Nature B.V
• Subjects: Algorithms; Analysis; Boring machines; Civil Engineering; Classifiers; Cluster analysis; Clustering; Data analysis; Decision trees; Drilling & boring machinery; Earth and Environmental Science; Earth Sciences; Geological data; Geology; Geomechanics; Geophysics/Geodesy; Mathematical models; Model accuracy; Model testing; Original Paper; Outliers (statistics); Oversampling; Parameters; Plant cover; Prediction models; Predictions; Safety; Thrust; Torque; Training; Tunnel construction; Tunnels; Ground type; Clustering and classification; TBM operational data; Intelligent prediction; Mixed-face ground
• ISSN: 0723-2632; 1434-453X
• DOI: 10.1007/s00603-021-02723-5

When a tunnel boring machine (TBM) is applied to the tunnel constructed in the mixed-face ground, the ground conditions ahead of tunnel face have a key impact on the operation performance and safety. Aiming to establish an automatic prediction model for geological conditions based on the operational data of TBM, the first step is to conduct clustering analysis using Canopy and K-means algorithms to recognize ground types based on geological data. Then, the ground type obtained by clustering analysis and corresponding operational parameters of tunneling machine are combined to construct a sample set. The outlier detection and synthetic minority oversampling technique (SMOTE) were used to preprocess the sample set. To obtain the best prediction effect, three different classifiers were applied for the model selection. By comparing the prediction performance of these three classifiers models, the gradient boosting decision tree (GBDT) model with accuracy of 0.804 shows the best performance as the geological prediction model. The test results of the prediction model show a low sensitivity when training set is small (set as 20%). The analysis of the importance of the model inputs showed that among the six machine parameters used in this study, the total thrust force, penetration rate and ratio of thrust to torque, are the three most influential inputs on the ground condition prediction results. Hence, the proposed prediction procedure can be applied to characterized and predicted ground conditions to ensure the safety and efficiency of tunneling. Highlights Clustering algorithms are used to distinguish the geological types of tunneling faces. A GBDT-based recognize model for geological conditions in TBMs tunneling process is developed. The proposed model is examined by different sizes of training sets for testing sensibility. The importance analysis of TBM operational parameters is conducted.