Development of an interactive database for soil strength parameters of Dhaka city based on artificial neural networks

Abstract

This research presents the development of an Interactive Database for different soil parameters for different areas of Dhaka City based on Artificial Neural Networks (ANN). Since it is expensive to obtain extensive information regarding soil parameters of a region, ANN can be used to generalize soil data over threedimensional space. Soil data from the existing soil reports are used to train a hierarchy of artificial neural networks for this purpose. The study has been divided into two phases. In the first phase of the study, soil reports have been collected from different Government and Non-government organization to prepare a relational database based on Microsoft Access Software which is readily available in most of the Personal Computers. This database can be used effectively for the preliminary design of any geotechnical structures. Artificial neural network has the capacity to map a very complex relationship among different parameters of a complex phenomena. It can generalize and interpolate the missing data in any possible direction. Thereby a complete three dimensional geotechnical database of an area can be obtained. In the second phase of the study, back-propagation neural networks has been used to simulate soil strength parameters (SPT and Unconfined Compression Strength) In three dimension using the data from the geotechnical database developed in the first phase of the study. The variables used in the models are Topographical Information, Depth, Specific Gravity (GS), Water Content, Dry Density, Percentage of Sand, Silt and Clay, Liquid Limit and Plastic Limit. The database contains 140 borehole data of Dhaka. The database can be updated easily and data of any place of Bangladesh can be added. The training of the ANN system is performed based on the available data of SPT, UCS and other available soil parameters stored in the database. The modeling approach has been found to be successful. The model predictions are convergent with the observed results. It has been observed that water content and dry density have significant effect on both SPT and UCS. The other soil parameters GS, %Sand, %Silt, %Clay, LL or PL do not have individual effect on SPT and UCS, but together with other variables they can be used to predict SPT and UCS up to a depth of 25ft. Additionally, %Sand, %Silt and %Clay together with Topographical Information and Depth has been used for prediction of SPT up to a depth of 100ft. ii The methodology and. application developed in this research can be extended in many directions. A framework that integrates spatially enhanced GIS systems with 3D graphics representation using a shared database can be developed. Also similar ANN models for predicting soil strength parameters for other areas of Bangladesh can be developed.