Machine learning approach for prediction of TBM performance and risk of jamming in Himalayan geology using a cross-project tunnelling database

Tunnel Boring Machine (TBM) excavation in the Himalayan region presents significant challenges due to complex geological conditions. The identification of tunnelling risk under such conditions is crucial for optimizing TBM performance. This study proposes a machine learning (ML)-based framework to predict TBM performance and assess associated jamming risk using a cross-project TBM database from the Himalayan region. The study employs ML approaches, including random forest, bagging, XGBoost, stacking ensemble, and artificial neural network. The combined stratified cross-project database results in improved model performance, with R² values ranging from 0.960 to 0.965. Shapley Additive exPlanations (SHAP) analysis revealed that the TBM net penetration rate (PRnet) is primarily influenced by response parameters such as torque and thrust with corresponding rock mass quality conditions. A combined jamming risk (CJR) scoring system was found to be useful to predict potential TBM jamming events in tunnelling. The CJR system effectively provides early warning signals in at least one ring (~ 1.5 m) in advance of TBM stuck events. The findings demonstrate that the application of ML-based techniques offers a valuable tool for predicting TBM performance and jamming risk, enabling adjustment of tunnelling parameters in the Himalayan region and similar complex geological settings worldwide.