VALIDATION OF ASSESSMENT OF SHEAR MODULUS OF SOILS USING NEURAL NETWORK

Abstract

Artificial Intelligence technique, especially Artificial Neural Networks or Neural Networks, are beginning to dominate most of the prediction aspects. The basic advantage in the use of Neural Networks being the capability of handling imprecise, imperfect and incomplete data while yet producing acceptable solutions. Predicting Standard Penetration Test (SPT) N-Value is very difficult because soil properties changes from place to place and varies with depth. Though SPT is simple but expensive and time consuming. Therefore, in the present study the application of Neural Networks has been examined with field/laboratory data to predict the N-Value obtained from the SPT of the particular site and the predicted N-Value can be used to determine the shear modulus of soils using existing relationship with N-Value. Prediction of N-Value has been based on the following input parameters, Soil Classification, Grain size distribution, Depth, Cohesion, Angle of friction and Specific Gravity. The Neural Networks were found to perform satisfactorily and results are in close agreement with measured field data. The Neural Networks capability was utilized to conduct parametric study with regard to relation between depth and N-Value and relation between shear modulus (obtained from existing formula) and predicted N-Value from Neural Networks. The potential of application of Neural Networks to augment/replace expensive and time consuming insitu test procedures is amply demonstrated and scope for further studies has been indicated.