The Cone Penetration Test (CPT) is a common in-situ

\r\ntest which generally investigates a much greater volume of soil more

\r\nquickly than possible from sampling and laboratory tests. Therefore,

\r\nit has the potential to realize both cost savings and assessment of soil

\r\nproperties rapidly and continuously. The principle objective of this

\r\npaper is to demonstrate the feasibility and efficiency of using

\r\nartificial neural networks (ANNs) to predict the soil angle of internal

\r\nfriction (Φ) and the soil modulus of elasticity (E) from CPT results

\r\nconsidering the uncertainties and non-linearities of the soil. In

\r\naddition, ANNs are used to study the influence of different

\r\nparameters and recommend which parameters should be included as

\r\ninput parameters to improve the prediction. Neural networks discover

\r\nrelationships in the input data sets through the iterative presentation

\r\nof the data and intrinsic mapping characteristics of neural topologies.

\r\nGeneral Regression Neural Network (GRNN) is one of the powerful

\r\nneural network architectures which is utilized in this study. A large

\r\namount of field and experimental data including CPT results, plate

\r\nload tests, direct shear box, grain size distribution and calculated data

\r\nof overburden pressure was obtained from a large project in the

\r\nUnited Arab Emirates. This data was used for the training and the

\r\nvalidation of the neural network. A comparison was made between

\r\nthe obtained results from the ANN's approach, and some common

\r\ntraditional correlations that predict Φ and E from CPT results with

\r\nrespect to the actual results of the collected data. The results show

\r\nthat the ANN is a very powerful tool. Very good agreement was

\r\nobtained between estimated results from ANN and actual measured

\r\nresults with comparison to other correlations available in the

\r\nliterature. The study recommends some easily available parameters

\r\nthat should be included in the estimation of the soil properties to

\r\nimprove the prediction models. It is shown that the use of friction

\r\nration in the estimation of Φ and the use of fines content in the

\r\nestimation of E considerable improve the prediction models.<\/p>\r\n",
"references": null,
"publisher": "World Academy of Science, Engineering and Technology",
"index": "International Science Index 98, 2015"
}