Improved SelfSim for Inverse Extraction of Nonuniform, Nonlinear, and Inelastic Material Behavior under Cyclic Loadings

Abstract

In this paper, an improved self-learning simulation (SelfSim) method is proposed for the inverse extraction of nonuniform, inelastic, and nonlinear material behavior under cyclic loadings. The SelfSim has been used to inversely extract local inelastic and nonlinear behavior of materials using limited global boundary responses. However, the SelfSim with conventional artificial neural network (ANN) models needs ad hoc data processing that frequently interrupts SelfSim training even in training monotonic constitutive behavior. To overcome this problem, an improved SelfSim with a new ANN-based hysteretic model is proposed. In addition, the ANN material model is implemented with considerations of large volume changes and geometric nonlinearity. The new SelfSim shows superior performances in the inverse modeling of complex material behavior under “multiaxial” and “cyclic” stress states. Two simulated numerical tests using a laminated rubber bearing with reinforcing steel shims are used to demonstrate the proposed SelfSim performance. A detailed comparison of the SelfSim with conventional ANN model is also presented. Finally, the improved SelfSim is experimentally verified by extracting nonuniform, nonlinear, and inelastic behavior of metallic material (low carbon SAE 1006 specimen) under cyclic loading. It shows very promising performances for the inverse material characterization of various engineering materials.