Inverse Identification of Surface Elastic Parameters in Soft Solids Using GA-ANN Surrogate Model

Abstract

Surface elasticity plays a crucial role in the mechanics of soft solids at the scale of micrometers and may become significant even at the scale of millimeters in exceptional cases. However, despite the efforts made for the identification of surface elastic parameters, accurately determining these values remains challenging due to the complex interplay with bulk elasticity and nonlinearity. To address this issue, a novel GA-based optimization framework is developed by employing an ANN-based surrogate forward model for efficient identification of surface parameters from the force-displacement response of a cylindrical specimen. The ANN is trained on force-displacement data of a soft cylindrical specimen, generated by nonlinear FE model predictions incorporating model elastic surfaces. The data used for training is generated for non-dimensional values of surface tension in the range of 0-5 and surface shear modulus in the range of 0-50. The accuracy and reliability of the trained ANN are established. The key novelty lies in replacing analytical forward models, which rely on idealized boundary conditions, with a data-driven surrogate trained on FE simulations under realistic constraints. The parameter identification is performed for a number of surface parameter sets using numerically generated force-displacement data as well as for noisy data obtained by adding 5% error in simulated data. The maximum error in identified values of parameters in all the cases is less than 8%. Repeatability and uncertainty analyses based on multiple noisy realizations further demonstrate the robustness of the approach, yielding narrow confidence intervals for the predicted parameters. The proposed framework provides a novel, efficient, and robust alternative to FE-based inverse identification of surface parameters, particularly for experimentally relevant boundary conditions.