Existing component-based models have difficulty interpreting the field response of earth pressure balance (EPB) shield tunneling parameters when the excavation face gradually changes from hard rock to sandy strata. To address this problem, this study proposes a mechanism-informed semi-empirical analysis based on a continuous face sand fraction function, $\eta \left(z\right)$, which represents the evolution of the sand-bearing area fraction at the excavation face along the shield advance direction. The function is constructed from the geological profile and is used as a continuous reformulation of existing face-composition descriptors such as rock/sand ratio and composite ratio. Based on $\eta \left(z\right)$, engineering-equivalent models for total thrust and cutterhead torque are developed and evaluated using field tunneling data from Qingdao Metro Line 15 and Line 5. The Dandan right-line data are used for parameter calibration, while the Basi right-line data from Rings 590–650 are used for independent validation without further parameter tuning. The results show that the η-related correction term improves the independent validation performance of the total thrust model, reducing the MAPE from 22.56% to 14.25%. In contrast, cutterhead torque exhibits stronger operational variability and interval-specific baseline offset. After applying a baseline correction determined from the stable hard-rock section of the Basi interval, the ring-scale torque MAPE decreases from 44.07% to 17.25%, but the additional $\eta$-related torque contribution remains limited. These results indicate that total thrust is more sensitive to the gradual increase in face sand fraction, whereas cutterhead torque is more strongly influenced by machine condition, cutter wear, and operational control. The proposed approach provides a field-based semi-empirical reference for interpreting tunneling parameter responses in similar hard-rock-to-sandy-strata transition zones.

When an earth pressure balance (EPB) shield operates in hard rock under air-assisted excavation, the conventional face-pressure term for full-chamber EPB operation is no longer directly applicable, while chamber air pressure and cutterhead–rock interaction become important components of the axial load. This study proposes an axial force-balance model for partially filled air-assisted excavation in hard rock. An effective air-pressure action-area coefficient is introduced to account for uncertainty in the air-pressure reaction, while unit-penetration torque, together with cutterhead rotational speed and advance rate, is used to represent the equivalent hard-rock excavation resistance. Field data from a hard-rock section of the Qingdao Metro are used for parameter identification, air-pressure-term ablation, sensitivity analysis, continuous-block validation, and cross-line validation. At the representative condition of χ_a = 0.75, the left-line model achieves R² = 0.771, an RMSE of 1158.6 kN, and a correlation coefficient of 0.878 between measured and calculated total thrusts. Removing the air-pressure reaction reduces model accuracy and markedly increases the correlation between the residuals and air pressure, confirming that this term cannot be neglected. Refitting the same model structure to the right line yields R² = 0.647, whereas direct transfer of the left-line parameters performs poorly. This indicates cross-line applicability of the model structure but condition dependence of its parameters. The proposed model provides a physically constrained framework for interpreting ring-averaged shield thrust and evaluating deviations between measured and calculated thrusts during air-assisted excavation in hard rock.