ABSTRACT
The use of the TBMs (tunnel boring machine) is increased in the last decades because of the efficiency and safety during the excavation process. Moreover, the TBM solutions is widely preferred by Contractors to reduce the risks during the construction, have lesser economical deviation from the forecast costs and time. To ensure lower construction cost and higher project safety, it is necessary to adjust the operating parameters of the TBM depending on the geological conditions. The method for improving the adaptability of the TBM is to utilize real on-time monitoring data to predict the future operating values of the TBM. There have been several studies on the prediction of operating parameters. Among them, prediction of the thrust force and torque have been considered one of the important research projects. A study applying random forest (RF) was also conducted based on heterogeneous in situ data. Recently, some works using machine learning (ML) techniques have been reported to effectively analyse complex field data. Based on data collected from the field, some authors employed gene expression programming (GEP) to predict thrust-related parameters, while others used support vector regression (SVR) in hard rock conditions. The aforementioned works are, however, bases on predictions based on experimental data (normally in one single tunnel) or on empirical dataset with the same geological parameters. The aim of this paper, so, is to provide a prediction of the TBM machine parameters to be used during design stage employing the so-called Unscented Kalman Filter (UKF) and be used by the average engineers. The approach employed will take as starting point the average formulations (semi-empirical and analytical) giving a guide to manage the ranges of values based on stochastic approach. The methodology will be applied to a practical example in four HSR (High-speed railway) tunnels in the northern Italy.
