Statistical analysis on field penetration index in shield TBM

The prediction of TBM (Tunnel Boring Machine) performance is very important for reliable estimation of construction period and cost in preconstruction stage. TBM performance can be diversely expressed in terms of penetration rate (PR), advance rate (AD) and field penetration index (FPI) etc. To predict the TBM performance more accurately, many researchers have been trying to utilize ground information obtained from field as well as machine data such as RPM, thrust and torque etc. Among the TBM performance, FPI can be used to not only evaluate the rock mass boreability without the influence of TBM operating parameters (Thrust, RPM) in design stage but also to calculate the PR. In this study, various rockmass parameters of shield TBM were analyzed to derive the significant parameters influencing on the FPI and to develop the best predictive model. Rockmass parameters such as water absorption (Wa), deformation modulus (Dm), lugeon value (Lv), rock mass rating (RMR), rock quality designation (RQD) and uniaxial compressive strength (UCS) were obtained from the cable tunnel project in South Korea. After selecting some of rockmass parameters representing the each characteristics of rock and rockmass through a literature review, a simple regression analysis was performed to eliminate some of abnormal data. And then, multiple linear regression analysis was carried out with the selected rockmass parameters (e.g. Lv, RMR and UCS). All of combination of input parameters were analyzed to select the best predictive model which correspond to the lowest PRESS. As a results of the regression analysis, adjusted r squared (R_adj²) and mean absolute percentage error (MAPE) were obtained as 0.574 and 32.7% each other, which means the selected model can be evaluated with explanatory power of 57.4% and prediction error of 32.7%. However, it should be noted that the developed model is very limited resulting from a narrow range of field data used in this study. Thus, quite a more and a wide range of data are required to develop the universal model.