GENERAL REGRESSION NEURAL NETWORK MODELING OF SOIL CHARACTERISTICS FROM FIELD TESTS

Abstract

The Standard Penetrating Test (SPT) can be considered as one of the most common in-situ popular and economic tests for subsurface investigation. Therefore, many empirical correlations have been developed between the SPT N-value, and other properties of soil. The principle objective of the current study is to demonstrate the feasibility and efficiency of using artificial neural networks (ANNs) to predict the soil angle of internal friction (Φ), the soil modulus of elasticity (E) and tip resistance (qc) of cone penetration test (CPT) results from SPT results considering the uncertainty and non-linearity of the soil. In addition, ANNs are used to study the influence of different input parameters that can be used to improve the prediction. A large amount of field and experimental data including SPT/CPT results, plate load tests, direct shear box, grain size distribution was obtained from a project in the United Arab Emirates to be used in the training and the validation of the ANNs. The ANN results are compared with some common traditional correlations. The results show that the developed ANNs can efficiently predict the aimed parameters from the SPT results. The predicted parameters from ANN are in very good agreement with the measured results compared to the predicted values from available traditional correlations.