Prediction of tunnel wall convergences for NATM tunnels which are excavated in weak-to-fair-quality rock masses using decision-tree technique and rock mass strength parameters

Abstract

Tunnel wall convergences should be predicted before the excavation and should be determined accurately in order to ensure safe and economic tunnel excavation media. For this to be possible, rock mass behaviors in the tunnel should also be estimated prior to excavation. This could be ensured by site investigation studies, numerical models and predictive statistics. In this study, the development of convergences during a tunnel excavation in Turkey, excavated by NATM technique, was evaluated by using statistical prediction techniques for weak-to-fair-quality rock masses. For this aim, actual tunnel wall convergence data and rock mass strength parameters were used. For prediction of tunnel wall convergence, multivariable regression analysis, artificial neural networks (ANNs), classification and regression tree (CHAID) and Chi-square automatic interaction detection (C&RT) methods were used and prediction results were compared to each other in terms of superiority and practicality. For this aim, 112 tunnel sections were used for prediction model and 30 different tunnel sections were used for validation. According to the obtained statistical prediction findings, it is seen that C&RT and ANN methods provide good prediction of tunnel wall convergences. However, C&RT method was found more practical in field use when compared to ANN. The results have shown that overburden thickness is the most effective parameter on tunnel convergence when compared to Crm, Φrm, Erm, RMR and Q. However, the best way for determination of tunnel wall convergences is to use in situ measurements, but in case of the lack of in situ measurement instruments, the suggested probability-based statistical approach is proven to be very effective and practical. Ultimate convergence level for any cross section with similar geological and geotechnical parameters to those analyzed before can be predicted by using the suggested method in this study. Yet, it should be kept in mind that the findings of this study are limited with the data used in this study. Although the aim of this study is to ensure a different point of view for determination of convergences in case of lack of convergence measurements and field data, by adding up more convergence measurements and rock mass strength data to the proposed statistical method, prediction power of this method can be improved and then this method can be used as a practical tool for the prediction of tunnel convergences. Besides, the user-friendly and open-to-development structure of this study can be a very useful tool for the geotechnical engineers, engineering geologist and mining engineers, if it can be developed by more field data.