Identifiability-Aware Swarm Calibration of Clay Models from Sequential Triaxial Paths

Abstract

The accurate calibration of constitutive models for clayey soils remains a formidable challenge in geotechnical engineering due to the complex, non-linear, and highly coupled nature of the mathematical formulations used to describe soil behavior. Traditional calibration procedures often rely on single-path triaxial tests, which frequently result in an ill-posed inverse problem characterized by severe parameter non-uniqueness and poor practical identifiability. This paper introduces an identifiability-aware swarm optimization framework specifically designed to calibrate advanced elastoplastic clay models using sequential triaxial paths. By integrating local sensitivity analysis and the Fisher Information Matrix directly into the objective function and velocity update equations of a particle swarm optimization algorithm, the proposed methodology actively guides the search agents toward regions of the parameter space that guarantee unique and robust solutions. Sequential triaxial paths, encompassing alternating phases of consolidation, compression, and extension, are utilized to persistently excite the internal state variables of the constitutive model, thereby maximizing the informational content of the experimental data. Comprehensive numerical investigations using synthetically generated, noise-corrupted stress-strain datasets demonstrate that the identifiability-aware swarm approach significantly outperforms standard metaheuristic algorithms in terms of convergence reliability and parameter precision. The framework effectively mitigates the phenomena of parameter compensation and topological valleys in the objective function landscape.