Undrained shear strength of norwegian clays estimated by neural networks

This paper describes an artificial neural network model for estimating undrained shear strength from piezocone penetration test (PCPT) data. The model requires the following five input variables: cone resistance, total vertical overburden pressure, excess pore water pressure at the cone tip, excess pore water pressure measured just above the cone base (reference location), and the hydrostatic pore water pressure. The neural network was trained using partial set of PCPT data from three Norwegian geotechnical test sites (Bakkiandet, Glava and Valoya clays). The remaining data was used to verify the predictive capabilities of the model. The undrained shear strengths at these sites were also estimated using some of the existing interpretation techniques. The neural network model was found to give better predictions of undrained shear strength compared to the methods evaluated.