Exploring Artificial Neural Network to Evaluate the Undrained Shear Strength of Soil from Cone Penetration Test Data

In geotechnical design practices, the undrained shear strength of soil is regarded as one of the engineering properties of paramount importance. Over the past years, several theoretical and empirical methods have been developed to estimate the undrained shear strength based on soil properties using in-situ tests such as cone and piezocone penetration tests (CPT and PCPT). However, most of these methods involve correlation assumptions that can result in inconsistent accuracy. In this paper, an artificial neural network (ANN) is used to devise a model with a better and more consistent prediction of the undrained shear strength of soil from CPT data. The ANN algorithm does not require such assumptions as it learns from previous cases/instances. A database was prepared of soil boring data and laboratory test data along with corresponding CPT/PCPT data from 70 test sites located in 14 different parishes in Louisiana. Presenting this data to the ANN, models were trained through trial and error using different network algorithms, such as back propagation method and quasi-Newton method. Different ANN models were trained using corrected cone tip resistance and sleeve friction input data along with some other easily measurable soil properties. The results of ANN models were then compared with a conventional empirical method of determining undrained shear strength of soil from CPT parameters. The results of this study clearly demonstrated that the ANN models outperformed the conventional empirical method, which confirms the applicability of ANN in the evaluation of the undrained shear strength of soil from CPT data.