Feature Selection and Multicollinearity-Sensitivity Relationship in Tunnel Boring Machine Performance Assessment

The assessment of tunnel boring machine (TBM) performance parameters is crucial for the successful planning and execution of tunneling projects. This study focuses on predicting the advance rate (AR), penetration rate (PR), and field penetration index (FPI) of shield TBMs using geological and operational parameters. A database was compiled from the literature comprising cutterhead rotation speed (CRS), mean thrust (MF), mean cutterhead torque (MT), upper earth pressure (UEP), torque penetration index (TPI), specific energy (SE), and lower earth pressure (LEP). The variance inflation factor (VIF) method was applied to evaluate the multicollinearity levels of these features, and four combinations were developed accordingly. Performance analysis revealed that empirical models trained with the full feature set (CRS, MF, MT, UEP, TPI, SE, and LEP) achieved the most reliable predictions of AR, PR, and FPI. Furthermore, the feature multicollinearity-sensitivity relationship indicated that the relative multicollinearity of operational and geological parameters had a less significant impact on prediction accuracy and curve fitting.