Time‐Dependent Settlement Model for Driven Piles Using Hybrid Finite Element‐Artificial Neural Network and Regression Approaches

Abstract

The prediction of time‐dependent pile settlement remains challenging due to the nonlinear behavior of saturated clay soils, consolidation, and pile setup effects. The complexity of this phenomenon is not fully captured by classical elasticity theories. This study develops a hybrid model for time‐dependent pile settlement based on the mobilized stiffness of an instrumented field pile. A Finite Element Method (FEM) is combined with an Artificial Neural Network (ANN) to predict load‐settlement curves at different times. Advanced performance metrics, such as the A20, Scatter, and Agreement indices, confirm the robustness of the model. A regression‐based methodology is subsequently used to derive the mobilized stiffness of the pile‐soil system from the initial portion of the curves, revealing a logarithmic trend. A formulation is proposed by extending the elastic shaft stiffness of Randolph and Wroth's solution with a time‐dependent amplification factor accounting for pile installation, geometric and scale effects. The rate of stiffness evolution is related to soil compressibility parameters, including void ratio, modified swelling and compression indices. Validation through laboratory model pile tests and comparison with existing analytical frameworks shows that the proposed model can reliably capture the time‐dependent settlement and stiffness evolution within the working‐load range.