Neural Network Based Prediction of Cone Side Resistance for Cohesive Soils

Abstract

The assessment of soil properties for the design of structure requires a wide range of tests. Sampling difficulty, time and cost constraints forces the practitioners to adopt correlations existing among the in situ tests and the physical or mechanical properties of soils. This paper presents the application of neural network to predict the cone side resistance (qs) obtained in the cone penetration test (CPT) for the cohesive soil based on plasticity index (PI), consistency index (CI) and the under drained shear strength (Su). Feed-forward back propagation algorithm was used for this purpose for the development of neural network model which was developed using 50 in situ dataset collected from the literature. Finally, the cone side resistance obtained from the developed neural network model was compared with the measured cone side resistance obtained from the CPT tests reported in literature. Further, the sensitivity analysis was performed to study the impact of plasticity index, consistency index and the under drained shear strength on the cone side resistance. The results of this study reveal that the developed neural network model was able to predict cone side resistance accurately.