MODELLING SHEAR STRENGTH WITH INDEX PROPERTIES OF ZEYSSE ELGO SOILS BY STATESTICAL AND ARTIFICIALNEURAL NETWORK APPROAC

Abstract

II

ABSTRACT This research work seeks to develop models for predicting the shear strength parameters (cohesion and angle of friction) of Zeysse-elgo soil using statistical and artificial neural network modeling technique. A total of twenty (20) soil samples were collected from various locations in Zeysse-elgo town. The potential of SPSS and Neural network to capture the nonlinear interaction between various input and output parameters has attracted many researchers towards investigating it. This paper presents the development of SPSS and ANN as prediction tool for determination of internal friction angle „Φ‟ and cohesion „c‟ from index properties. With this context, an attempt has been made to develop a neural model from the index properties of soil consisting of liquid limit (LL), plastic limit (PL), Plasticity Index (PI), ɸ, c and Ucs as input parameter. A comprehensive set of experimental data were used in developing models based on SPSS and neural network. Linear and Non-linear correlation coefficients R square and adjusted R square analysis was performed using same dataset and statistical parameters to judge the efficiency of proposed model. According to this analysis liquid limit (LL), plastic limit (PL) and plastic index(PI) the most significant variables contributing in estimation of strength parameter “c” and “ɸ֯. However plasticity index (PI), plastic limit (PL) and liquid limit (LL) affects considerably in prediction Ucs. The data indicates the average value of LL is 59.21% which indicates about marginal swelling potential because it‟s between 50%-60%, the average value of PL is 41.48% and PI is 17.72% which indicates clay and low swelling potential soil, the average value of ɸ(%) is 28.33 which indicates that the soil is silty or silty sand and the average value of ucs is 49.08 which is a very soft soil and its consistence is medium. Index and shear properties tests were performed on twenty soil samples in the laboratory, collected from various locations in Zeyssse-elgo town. Based on the test results, the soils are categorized fine soils. Analysis of the experimental data indicated that there exist a good correlation among the measured value and predicted value by SPSS approach but not by neural due to small sample size