Machine Learning-Based Sensitivity Analysis of Soil Ingredients and Soil Mechanical Properties on Tunnel Boring Machine (TBM) Advance Rate

The prediction of Tunnel Boring Machine (TBM) advance rate is crucial for optimizing tunneling operations and ensuring cost-effective project management. This study investigates the use of machine learning (ML) models, specifically the Random Forest Regressor (RFR) and Support Vector Machines (SVM), to predict TBM performance in heterogeneous soil conditions. By leveraging data-driven approaches, the research provides insights into the factors influencing TBM advance rates, including soil composition, mechanical properties, and operational parameters. A comprehensive sensitivity analysis was performed to identify the key variables affecting TBM efficiency, focusing on soil properties such as cohesion, friction angle, and uniaxial compressive strength. The RFR and SVM were used as the primary ML models to predict the TBM advance rate based on these features. The models were trained on 80% of the dataset, while 20% was held back for testing and validation. Results indicate that these machine learning-based models, particularly the RFR and SVM, offer significant accuracy in predicting TBM performance, outperforming traditional empirical methods. The study contributes to the growing body of knowledge on ML applications in underground construction, with implications for enhancing TBM performance prediction, real-time monitoring, and reducing project risks.