Use of artificial neural networks to analyse tunnelling-induced ground movements obtained from geotechnical centrifuge testing

In geomechanics, centrifuge modelling and digital image analysis enable the acquisition of large amounts of high-quality data related to ground movements. In this paper, modern intelligent methods based on a feedforward artificial neural network (ANN) architecture are applied to study tunnelling-induced ground displacements. Soil displacement data obtained from a geotechnical centrifuge test are used to investigate the capabilities of ANNs in this context. Because this work represents a feasibility study, the centrifuge dataset is limited to a single test. The trial-and-error process is used to identify three architectures of varying complexity that achieve a good level of performance. Predictions are evaluated both statistically (R2) and qualitatively (analysing the shape of vertical and horizontal displacement profiles). Results show the applicability of modern intelligent analysis methods for analysing centrifuge datasets and highlight certain strengths and deficiencies of feedforward ANN architectures compared to empirical methods