Predicting Horizontal Deformations under an Embankment using an artificial Neural Network

Horizontal deformations can cause damage to objects adjacent to a newly constructed embankment. A tool has been developed to determine in an early state of a project if such a risk exists. The basis of this tool is an Artificial Neural Network (ANN). ANNs are particularly suited to find a relationship between an input and an output parameter, even if it is a very complex one. With this in mind, a network has been trained using calculations with the Finite Element Method (FEM) to find the relationship between the maximum vertical and maximum horizontal displacement under an embankment. Clearly, only the total vertical displacement is not enough information to predict the horizontal deformations. To make a good prediction of the horizontal deformation one needs, for a completely drained situation, at least two extra parameters: the thickness of the soft soil layers and the height of the embankment. Given these, an ANN can reproduce the finite element result with an error less than 1 percent. To make a good prediction of the undrained case one needs two extra input parameters: the initial height of the embankment and the initial excess pore water pressure due to this previous load. Again, these five parameters reproduce the results of the FEM analysis within 1 percent. This readily accessible program has the same quality as a complete finite element analysis but is very fast and simple to use.