Soil Profile Prediction Using Artificial Neural Networks   Yousef Nouri PhD student, Civil & Environmental Engineering Department University of Michigan, MI48109, USA e-mail: ynouri@umich.edu Asskar JanAli-Zadeh Choobbasti Assoc. Prof., Faculty of Engineering, Mazandaran University, Babol, Iran e-mail: asskar@umz.ac.ir and Ali Kaveh Professor, Civil Eng Dept, Iran University of Science and Technology, Tehran, Iran e-mail: Alikaveh@iust.ac.ir
			ABSTRACT

Soil profile information plays an important role in geotechnical evaluation in construction of major structures including the selection of core material for construction of rock-fill dams. Inferring the geotechnical characteristics including layering (stratification) and engineering properties of the underlying soil in the cross-borehole region is a human-intensive process and is subject to statistical and systematic errors. Improving the reliability of soil property interpolation may lead to cost reduction and improved operation planning.

Artificial Neural Networks (ANN) is a technology inspired by the brain’s certain information-processing characteristics including the parallel processing, the ability to learn and generalize, disregard data errors and produce meaningful solutions, which fall beyond the reach of conventional computer programming. In this paper ANN systems consisting of several Feed-forward Multilayer Perceptron Networks are developed to predict the soil profile in a specified location, based on the available site investigation data from a 30 square kilometers area of Tehran municipal region and the results are then compared with unused data of actual boreholes to check the ANN model’s validity.

Keywords: Site investigation, Soil-geology, Lithology, Stratification, Spatial variability, Neural Networks

Get the entire paper (zip)    Go back to the TOC