A Generalized Regression Neural Network Model to Predict CFA Piles Performance Using Borehole and Static Load Test Data

Abstract

This research aims to predict the downward displacement of piles as the axial loading increases and decreases to form the pile static load curves. The generalized regression neural network (GRNN) is used to predict the behavior of continuous flight auger piles through training and testing the network by using field data collected from over 100 field static load tests. The model aims to reduce cost, decrease the dependency on the lengthy field test, and complement the current methods used in pile design. In this research, the authors built a neural network that associates the ground properties, such as friction angle and unconfined compressive strength, as well as the pile’s length and diameter with the CFA pile performance at different sites. More than six thousand field points were collected to train the neural network. Moreover, three independent field tests were used to examine the performance of the network comparing the predicted values with the field data. The statistical measures showed that the proposed model can be generalized. The model can predict the pile displacement curves accurately as long as the testing data fall within the training set. If the testing data fall outside the range, the model was not as accurate. To evaluate the performance of the GRNN, three statistical measures were calculated; the coefficient of efficiency ranged between 0.8362 and 0.9968, the root mean square error ranged between 0.0868 and 0.2760, and the mean absolute error ranged between 0.070469 and 0.23443.