A universal surrogate model for predicting ground movements resulting from shaft construction in London Clay

The surrogate model was established as an artificial neural network (ANN) using the deep learning framework Keras. A dataset of vertical and horizontal movements due to shaft excavation in London Clay was generated by conducting detailed coupled-consolidation axi-symmetric finite element analyses. A total of 1000 analyses were carried out adopting different combinations of problem geometries, including variations in the thickness of Made Ground and Terrace Gravels layers (both of which adopting values between 0 and 5 m); shaft height (values between 10 and 70 m); shaft diameter (values between 2.5 and 40 m); the proportion of the shaft height in contact with London Clay adopting concrete segments (named PCS and adopting values between 50 and 70%) and the shaft lining thickness (values between 0.2 and 0.4 m).The ANN was trained to predict the vertical and horizontal displacement radially around the shaft up to a distance of 2H (where H is the depth of the shaft) at certain depth (up to the depth of the shaft). The numerical database used to train the ANN and evaluate its performance can be made available upon request (10.5281/zenodo.10762242).

The tool produces two graphs: one of vertical displacements with radial distance from the shaft at the selected depth and a separate graph of the horizontal displacements. If selected by the user, the application exports the data points of those two graphs to a csv file. Note that the tool does not accept input values outside the range utilised to train the ANN. 

The button "save csv" should be clicked before the button "Run" when output to a csv file is to be requested. Otherwise, only the button "Run" needs to be clicked.