Determination of Coefficient of Lateral Earth Pressure at Rest for Sandy Soils Using Cone Penetration Test and Artificial Neural Network

The estimation of soil parameters in geotechnical practice is always an important step for design of geostructures. In order to control the stability and to perform a stress-deformation analysis of a geotechnical system, it is essential to know the in-situ stress state of the ground. The in-situ vertical stress can be calculated easily if the depth and the soil density are known. However, determination of the in-situ horizontal stress is a complicated task because it depends on several other soil characteristics, such as stress history and over consolidation ratio (OCR). This has always been one of the most challenging geotechnical problems. In the past, attempts have been made to find reliable methods to determine the coefficient of at-rest earth pressure by means of in-situ or laboratory tests. This is especially true for cohesionless soils, using in-situ test results such as pressuremeter test, blade test and cone penetration test (CPT) to name a few. In this study, the calibration chamber data on CPT tests performed at universities worldwide or well known institutes were gathered. Then using these series of reliable CPT calibration chamber test data and a system consisting of three types of neural networks, the coefficient of at rest pressure (K0) is predicted while it has good agreement with measured data. In this proposed system, a series of neural networks perform some tasks and finally by strategically combining the networks, the system will be able to predict parameter (K0) with reasonable accuracy. The proposed system uses self organizing map (SOM) for clustering data into training, testing and validating sets and probabilistic neural networks for classifying the sands and back propagation neural networks for conclusive function approximation. Details on the development of such a system are described in the present paper and finally results obtained by this system are compared to the available relations suggested by other researchers.