Assessment of Driven Pile Ultimate Capacity through Artificial Neural Network Analysis of Cone Penetration Test Data

Abstract

In this research, the application of an artificial neural network (ANN) was employed utilizing cone penetration test (CPT) information to produce an enhanced comprehension of the ultimate load-bearing capacity of piles. The ANN algorithm is independent of correlation assumptions as it uses prior cases/instances to grasp the relationship. A database of eighty pile load tests on squared precast/prestressed concrete (PPC) driven piles and corresponding CPT data was prepared in this regard, in which the ANN models were trained using these data. Feed-forward network techniques such as backpropagation algorithm, Levenberg–Marquardt algorithm were used with trial and error. The cone sleeve friction and corrected cone tip resistance were used to train numerous ANN models. A comparison was made between the prediction of ANN models and three pile-CPT methods, that is, Laboratoire central des pontes et chaussées (LCPC), probabilistic, and University of Florida (UF) methods. The findings of this research exhibited that ANN excels in the evaluation of ultimate capacity of squared PPC piles. A comparison was also made with LCPC, probabilistic, and UF method on the basis of reliability-based load and resistance factor design analysis, which also demonstrates higher resistance factors, ϕ, and superior efficiencies of ANN models over the traditional pile-CPT methods. Consequently, these discoveries reinforce the efficacy of utilizing ANN for assessing the ultimate load-bearing capacity of piles through the interpretation of CPT data.