NEURAL-NETWORK AS A CONSTITUTIVE MODEL OF SOIL

The modeling of a constitutive law for soft soils by neural network is the topic of the present paper. A feed-forward neural network used in predicting the oedometer loading curve has been trained by the 40 of the 46 oedometer curves, and the neural network has been tested on the remaining six oedometer curves not included in the learning process. The neural network has been implemented into a computer program using the finite element method. An example of a test embankment on very sof t subsoil has been calculated using the Cap model and a neural network model, and the comparison between the two methods is presented here.