ABSTRACT
Tunnel Boring Machines (TBM) are the predominantly used for the excavation of metro tunnels in urban areas, as at the same time they can achieve almost zero disturbance to the surrounding rock mass and can attain high advance rates. Nevertheless, the complexity of the geological and geotechnical conditions along with the vulnerability of surface structures can give rise to surface settlements and potential problems. Although there is a number of methods of both empirical and computational nature to make predictions of the anticipated settlements, the introduction of modern machine learning applications can further assist in a more accurate perdition and assessment of risk prone circumstances. In the paper the analysis of the surface settlements is made with machine learning methods utilizing data coming from the construction of Athens metro Line 2, namely an interval part of the Hellinikon extension project. The data taken into account consist of both the geological and geotechnical conditions encountered and the EPB-TBM operating data. Several Artificial Neural Networks (ANNs) were developed utilizing a different set of input parameters and were evaluated against their prediction performance capabilities as compared with real measurements. Four ANNs showed the most promising results having an average prediction accuracy of more than 79% and a consistent behavior across almost the entire range of the test dataset used, closely following the trend and behavior of the measured settlements.
