Rock Mass Classification in TBM Tunneling using Artificial Neural Network Techniques: A Case Study from Siwalik Region of Nepal

Abstract

The Himalayan region exhibits a highly complex geological setting influenced by tectonic activities, resulting in faulted, folded, sheared, and deeply weathered rock masses. This study focuses on the Siwalik region of Nepal, where the geology is characterized by sandstone, mudstone, siltstone, and clay. Accurate rock mass characterization is crucial for Tunnel Boring Machine (TBM) tunneling projects, particularly in the challenging geological conditions of the Nepal Himalayas. Empirical rock mass classification systems, such as the Rock Mass Rating (RMR) and Q-system, often fall short in TBM operations due to limited access to the tunnel face and the dynamic nature of TBM excavation. To address these challenges, this research employs a machine learning (ML) technique to classify rock mass conditions using operational and geological data collected from the Sunkoshi Marin Diversion Multipurpose (SMDM) project in Nepal, where a double-shield TBM was used to excavate the 13.3 km long headrace tunnel. A comprehensive dataset comprising 3,173 TBM cycles, including parameters such as cutter head speed, torque, thrust, and penetration rates, was utilized for model development. An Artificial Neural Network (ANN) model was developed, trained, and optimized using grid search to identify the best hyperparameters. The Synthetic Minority Oversampling Technique (SMOTE) was applied to address the class imbalance, significantly improving the model's recall for Class V (poor rock mass class). Performance metrics such as accuracy, precision, recall, and F1-score were used to evaluate the model. Additionally, SHAP (Shapley Additive Explanations) analysis was conducted to interpret feature contributions for rock mass Class V, which revealed that torque and thrust had the highest influence on predicting poor rock mass conditions. This study highlights the effectiveness of ML models in improving rock mass classification, especially for underrepresented classes, and provides valuable insights for optimizing TBM operations in complex geological settings.