Predicting the performance of slurry TBM through marine deposits using machine learning models

Abstract

Slurry TBMs are essential for large-diameter tunnels in challenging geological environments, but accurate prediction of penetration rate (PR), a key parameter signifying operational efficiency, remains understudied. Existing machine learning (ML) models for TBMs do not consider slurry-related parameters. This paper leverages a comprehensive dataset ($\approx$2000 data points) from the Mumbai Coastal Road tunnel in India to address this gap and develop seven supervised ML models. Among these Extra Tree (ET) and XGBoost (XGB) models demonstrated superior performance, with Root Mean Squared Error (RMSE) values of 1.69 and 1.91, respectively. The study reveals an 11% increase in model performance when slurry parameters are included in the ML model for PR prediction. Results from the SHapley Additive exPlanations (SHAP) analysis are grounded in the geomechanical theory, demonstrating the generalizability of the ML models and providing insights into critical parameter contributions, enabling accurate penetration rate predictions and improved project timelines, outperforming traditional empirical methods.