Abstract
Accurate prediction of the learning phase period ( LPP ) for slurry tunnel boring machines (STBMs) is essential for effective scheduling, cost control, and risk management in soft ground and mixed-face tunneling. Traditional models often oversimplify TBM performance and do not fully account for geological variability at the cutterhead depth. This study presents a predictive model that integrates ground conditions, represented by the average standard penetration test N-value ( N (av) ), with operational learning dynamics. The model incorporates ground factor ( Gf ) and a learning parameter ( c ) to describe advance rate ( AR ) growth during the LPP as an exponential function. The model was calibrated using field data from multiple tunnel phases, with c value of 0.25 accurately representing observed AR progression. LPP was defined as the time required for AR to reach 95% of its steady-state value. Validation shows strong agreement with measured data, yielding a mean absolute error (MAE) of 0.61 m/day, root mean squared error (RMSE) of 0.673 m/day, mean squared error (MSE) of 0.453 (m/day) 2 , and mean absolute percentage error (MAPE) of 8.54%. Sensitivity analyses indicate that steady-state AR , c , and Gf strongly influence LPP duration and performance stabilization. By quantitatively linking geotechnical indicators with TBM operational behavior, the model provides a practical decision-support tool for forecasting productivity, optimizing tunneling strategies, and enabling real-time performance adjustments. This framework enhances mechanized tunneling efficiency, especially during the critical early phase of operations.