Simplified neural network models for estimating soil resistance using dynamic pile test

摘要

Neural network formulations are attempted to predict the soil resistance under dynamic conditions. These models are based on simple input data from the dynamic pile test. In this investigation, only HST data and the geometrical properties of the pile are invoked in the models. Thus making the models effective and appropriate for the daily practice. The models involved are simple 3-layer and 4-layer backpropagation networks. Data from more than 90 driven piles, representing different subsurface conditions were utilized in the investigation. The results and comparisons with other methods indicate that the neural network models fulfill the reliability and adequacy desired in the prediction of soil resistance under dynamic loads.