ABSTRACT
The application of tunnel boring machines, which offers lower noise and vibration levels as well as higher stability compared to drilling and blasting methods, is increasingly being utilized in tunnel projects. The disc cutters continuously interact with the tunnel face, inevitably experiencing wear. Since replacing disc cutters is time-consuming and costly, accurately predicting disc cutter wear is crucial for the economic efficiency of TBM projects. This study quantitatively predicts the wear of slurry TBM disc cutters using geological conditions, TBM operational data, and machine learning techniques. Due to the limited availability of uniaxial compressive strength data across the entire tunnel section, UCS was initially estimated, followed by disc cutter wear prediction. A comparison of the disc cutter wear prediction models indicated that the XGBoost model demonstrated the highest predictive performance on the test dataset. Finally, Shapley additive explanations were used to interpret the complex machine learning models.
