Applicability of a CPT-Based Neural Network Solution in Predicting Load-Settlement Responses of Bored Pile

Abstract

In this article, the results of load-settlement responses in piles bored from cone penetration tests (CPTs) are presented and discussed to present an accurate artificial intelligence (AI) model. Different AI computation methods, including static and dynamic neural networks, namely, feed-forward neural networks (FFNNs) and focused time-delay neural networks (FTDNNs), are presented using an extensive data set of in situ CPTs. Several interpretation diagrams show the performance of the models. The accuracy of the presented models was investigated using the value of root-mean square error (RMSE) and regression (R2) plots. A FFNN model was chosen for CPT result prediction because of its accuracy and simplicity. The results of convergence analysis indicate that the proposed CPT-based design model is promising for predicting load transfer and settlements for axially loaded single bored piles. A simple formula is presented based on neural network parameters. The predicted results were compared with the experimental data, and a good agreement was attained, confirming the reliability of both the FFNN (R2 = .9996) and FTDNN (R2 = .9995) solutions in this study.