Development and Calibration of a Simplified Nonlinear Spring Model for Predicting the Lateral Response of Caisson Foundations Using a Multiobjective Genetic Algorithm

This study introduces a simplified one-dimensional nonlinear spring model to predict the lateral behavior of caisson foundations under combined lateral and moment loading considering nonlinearities at both the soil and the soil–caisson interface. This model extends the conventional Winkler-based soil spring approach by adopting a modified hyperbolic formulation for spring stiffness incorporating distributed translational, rotational, and base translational nonlinear springs. A multiobjective genetic algorithm was implemented to optimize the simplified model’s parameters, calibrating the predicted lateral load–deformation (p–y) and moment–rotation (m–θ) behavior against the results of three-dimensional finite-element analysis (FEA). This study focused on caissons embedded in both cohesive (clay) and cohesionless (sand) soils with varying diameter-to-depth (D/B) ratios. Numerical simulations were performed using the FEA tool ABAQUS/CAE (version 2019), while MATLAB (version R2024a) was used to simulate the simplified model and optimization algorithm. Multivariate nonlinear regression analysis was used to develop the proposed empirical equations for calculating the key parameters of the simplified model. The model was extensively validated against three-dimensional continuum-based analysis using homogeneous and layered soil profiles, demonstrating excellent agreement. Furthermore, the model was tested to generate shear force and bending moment diagrams, which provide valuable insights for design and performance assessment. This simplified approach is computationally efficient, making it practical for design offices, while remaining sufficiently robust to capture the complexities of nonlinear soil–caisson interactions.