Parameter Identification and Uncertainty Analysis of Stone‐Column‐Reinforced Composite Foundations Constrained by Static Load Tests

ABSTRACT Accurate identification of constitutive parameters for stone‐column‐reinforced composite foundations is essential for deformation analysis and settlement prediction of clay‐core rockfill dams. However, the parameters of the Duncan‐Chang model are difficult to determine directly under field conditions, whereas conventional inversion methods relying on repeated finite element (FE) analyses are computationally expensive. To address this issue, this study proposes a surrogate‐assisted parameter inversion framework constrained by static load tests in the stone‐column‐reinforced composite foundation of a planned clay‐core rockfill dam in Rwanda. Response samples under staged loading were generated using FE simulations, and a surrogate model was introduced into the inversion loop as a computational substitute for repeated FE analyses, thereby improving inversion efficiency. By jointly calibrating settlement responses under multiple load levels, the Duncan‐Chang model parameters governing the overall load‐settlement behaviour of the composite foundation were identified. Furthermore, uncertainty analysis was conducted for a representative test column to obtain the posterior distributions of the key parameters and the parameter‐induced posterior prediction intervals of the load‐settlement response. The identified parameters were further validated through FE back‐analysis and dam‐scale settlement prediction. The results show that the proposed framework can satisfactorily reproduce the load‐settlement response observed in static load tests while substantially reducing the computational cost. This study demonstrates that surrogate‐assisted parameter inversion constrained by static load tests provides a practical and robust approach for parameter identification in stone‐column‐reinforced composite foundations, while maintaining a favourable balance between efficiency and engineering applicability.