Material modeling through inverse analysis

This thesis extends the capability of a new data-driven inverse analysis method, SelfSim (self-learning simulations), for extracting material behavior of large and complex problems. SelfSim uses load-displacement measurements from structural tests whereby the material experiences non-uniform stresses and strains to extract the material constitutive behavior in the form of a stress-strain database. This method is unconstrained by a pre-defined material model unlike the conventional inverse analysis approaches to find the optimized parameters for material models.;SelfSim frameworks consist of massive computations in Neural Networks (NN) trainings and Finite Element (FE) simulations. To reduce the computational costs four parallelization schemes are developed and verified with benchmarking examples; FEM A/B parallelization, multiple NN parallelization, single NN parallelization, and FEM forward separation.;Extended SelfSim is applied to extract the material behavior of large and complex problems using simulated and physical experiments in the field of structural, geotechnical and bioengineering. First, SelfSim successfully extracts the anisotropic response of aluminum from multiple tests. The method simplifies the laborious and lengthy process of developing a conventional material model for metals whenever a new material constitutive behavior is to be characterized.;Then, SelfSim is conducted to extract the three dimensional (3D) behavior of the deep excavation from inclinometer measurements around the excavation. Numerical development required for 3D modeling of excavation using SelfSim inverse analysis is demonstrated.;SelfSim is also introduced to extract Red Blood Cells (RBCs) material stress-strain behavior from measurements of forces and displacements obtained by optical tweezers techniques. Deformation characteristics of RBCs are closely linked to disease (e.g. malaria) progression and hold promise as a tool for disease diagnosis. SelfSim reveals that in order to capture the interrelationship between measured axial and transverse deformations the stress-strain relationship for healthy RBC has to be anisotropic and thus differs from commonly assumed isotropic hyperelastic response. The deformability and anisotropic stress-strain behavior of healthy RBC decrease for mature stages of malaria.