Development of a Soil Parameter Interpolation Based on ANN

摘要

This paper aims at developing a site characterization model based on spatial soil property data processed through an Artificial Neural Network (ANN). A geotechnical database is created, which contains 140 borehole data of the capital of Bangladesh, Dhaka City. A back-propagation ANN has been used successfully to simulate soil strength parameters in three dimensions using the data from the geotechnical database. Data from SPT (Standard Penetration Test), UCS (Unconfined Compression Test) and index properties are used to train the ANN system. So SPT and UCS at any point can be obtained using the ANN system, if soil index properties such as specific gravity water content, dry density, liquid limit, plastic limit and provided. Depending on the prediction accuracy several different SPT models are proposed. The advantages of ANN system over other more conventional data fitting methods are that it provides rapid results and can be retrained as additional data are acquired.