Applying machine learning to the development of surrogate models for shafts in clay

1. Home
2. Publications
3. Online Library
4. Applying machine learning to the development of surrogate models for shafts in clay

Publication Title
Applying machine learning to the development of surrogate models for shafts in clay

Abstract
Surrogate models are machine learning-based models that approximate aspects of interest of a reference numerical model while being considerably faster to run. This feature makes them very attractive for engineering purposes, such as design optimisation and back-analysis. A surrogate model capable of predicting the short-term vertical movements around a shaft excavated in clay is developed in this paper. The surrogate model consists of an artificial neural network (ANN) which is trained with an extensive database generated from a finite element (FE) model considering variations in shaft geometry (diameter and depth). Correspondingly, the ANN is established to predicting the surface and subsurface vertical displacements around the structure for a given shaft geometry. The procedure followed for handling the data inputs as well as optimising the training of the ANN is described. Furthermore, the predictive capabilities of the ANN are thoroughly assessed against the numerically-generated data demonstrating an overall excellent performance. This suggests that the FE model of the shaft can be confidently replaced by the ANN in preliminary design stages.
Authors
Agustin Ruiz Lopez; David M. G. Taborda; Aikaterini Tsiampousi; A. M. G. Pedro; Stuart Hardy

Published In

18th European Conference on Soil Mechanics and Geotechnical Engineering (ECSMGE2024)

Theme

A - New developments on structural design

Download PDF