Artificial neural networks for the evaluation of impedance functions of inclined pile groups

Recent works provided insights into the dynamic response of inclined piles. However, there is a lack of design tools that can be implemented to perform inertial soil-structure interaction analyses, in the framework of the sub-structure approach, bypassing the numerical modelling of the kinematic interaction problem. For this purpose, this paper presents the application of ANNs to the impedance functions of inclined pile groups. Firstly, the dataset required for the training and testing of the ANNs is obtained from a parametric investigation in which the frequency-dependent impedance functions of 22 pile groups rigidly connected at the head and embedded in homogeneous soil deposits are considered. A dimensional analysis of the problem is performed and soil-foundation systems with different geometric characteristic and properties (e.g. inclination angles, pilesoil stiffness ratios, pile length-to-diameter ratios) are investigated. Data, representing the target of the neural networks, are obtained through a numerical FE model developed by the authors. The main trends of impedances with the dimensionless parameters of the pile groups are inferred from the presented results. Subsequently, several ANNs consisting of two hidden layers are trained, tested, and validated to optimise the generalisation performance of the model. Results show that ANNs are able to capture the trends of the impedance functions of inclined pile groups on the basis of the adopted dimensionless parameters. In addition, the complexity of the ANN models achieving good performance confirms the need for advanced regression models.