Estimation of pile stiffness in non-homogeneous soils through Artificial Neural Networks

Many international standards highlight the relevance of studying the compatibility of forces and displacements between the structure and the foundation that supports it. In the case of pile foundations, some authors use continuum models that rigorously reproduce the interaction of the pile with the surrounding soil. However, their high computational cost justifies the use of simplified methodologies that allow obtaining sufficiently accurate results in significantly less time. This work presents a surrogate model based on Artificial Neural Networks (ANNs) trained from a synthetic dataset generated by a continuum numerical model. A good regression capacity is observed by the proposed model, also requiring very short evaluation times. Two use examples are presented to illustrate the smooth behaviour of the ANNs model and its ability to determine the critical pile length. This surrogate model allows introducing soil–structure interaction in problems with large volume of evaluations in a feasible way without significantly compromising the confidence of the results.