Hybrid ensemble paradigms for estimating tunnel boring machine penetration rate for the 10-km long Bahce-Nurdagi twin tunnels

This study presents a novel hybrid ensemble strategy based on an inputs-outputs amalgamation technique for estimating the rate of penetration (ROP) of tunnel boring machines (TBMs). Notably, accurate estimation of TBM performance reduces risks that can arise during tunnelling work; thus, assessments of TBM's performance is one of the primary tasks before the construction of tunnels in railway, highway, metro, and hydroelectric projects. In this study, an extensive real-life database of 5334 observations were compiled from an ongoing project of BahceNurdagi twin tunnels near one of the most active seismic faults in Turkey. Initially, a portion of the main database was used to construct five standalone models, namely random forest, support vector regression, extreme learning machine, artificial neural network, and multivariate adaptive regression splines. Subsequently, the estimated outputs (i.e., ROP) of all the models were combined with the primary input parameters, called an inputs-outputs amalgamation technique, and utilized as a new set of inputs to estimate the ROP using each standalone model. Five hybrid ensemble (HENSM) models were constructed and validated based on the proposed approach. According to R2 and MAE criteria, the proposed HENSM-RF (a hybrid ensemble paradigm of random forest constructed with a new set of inputs) attained higher prediction precision (R2 = 0.8637 and MAE = 0.0378) in the validation phase. The performance of the proposed models was also compared as per site conditions. Overall results suggest that the proposed model has a high potential for assisting practitioners in assessing the ROP of TBMs in tunnelling projects. The employed database also attached as a supplementary for future studies.