General Regression Neural Networks for Driven Piles in Cohesionless Soils

Abstract

Numerous investigations have been conducted in recent years to predict the load-bearing capacity of driven piles in cohesionless soils. The problem is extremely complex owing to the large number of factors that affect the behavior of piles. The available methods either oversimplify the nature of the problem or improperly consider the effect of certain governing factors. In this paper, a general regression neural network (GRNN) is used for predicting the capacity of driven piles in cohesionless soils. Predictions of the tip, shaft, and total pile capacities are made for piles with available corresponding measurements of such values. This is done using four different procedures as well as the GRNN. Comparisons of capacity component predictions (i.e., tip and shaft capacities) are only achieved for instrumented piles. The actual measurements of tip and shaft pile capacities were adjusted to account for residual stress using the wave equation analysis. For the remaining piles, the predicted total capacities are compared only with the measured value. The GRNN provides the best predictions for the pile capacity and its components. It is also shown that the GRNN is applicable for all types of conditions of driven piles in cohesionless soils.