The steel pipe umbrella is a widely used technology when tunnelling in weak soils in order to create pre-support ahead of the tunnel face. The design of steel pipes is frequently done through simplified analytical approaches which are easy to apply but require proper assessment of the loads acting on the pipe. To provide information on this key design aspect, the results of the comparison between a three-dimensional numerical model developed with the code FLAC 3D and an analytical model based on the approach of a beam on yielding supports is presented and discussed. The comparison refers to a shallow tunnel with an overburden of three times its diameter for two different types of weak rock masses. The obtained results provide suggestions about the load that has to be applied in the analytical model for the design phase.

Ground settlement during and after tunnelling using TBM results in varying dynamic and static load action on the geo-stratum. It is an undesirable effect of tunnel construction causing damage to the surface and subsurface infrastructure, safety risk, and increased construction cost and quality issues. Ground settlement can be influenced by several factors, like method of tunnelling, tunnel geometry, location of tunnelling machine, machine operational parameters, depth & its changes, and mileage of recording point from starting point. In this study, a description and evaluation of the performance of the arti?cial neural network (ANN)was undertaken and a comparison with multiple linear regression (MLR) was carried out on ground settlement prediction. The performance of these models was evaluated using the coefficient of determination R2, root mean square error (RMSE) and mean absolute percentage error (MAPE). For ANN model, the R2, RMSE and MAPE were calculated as 0.9295, 4.2563 and 3.3372, respectively, while for MLR, the R2, RMSE and MAPE, were calculated as 0.5053, 11.2708, 6.3963 respectively. For ground settlement prediction, bothANNandMLRmethodswere able to predict significantly accurate results. It was further noted that the ANN performance was higher than that of the MLR.