Identification and Identifiability Analysis of Material Parameters Using Hybrid Optimization Algorithms and Optimized Artificial Neural Networks

Accurate data sets for material behavior model simulation are necessary for realistic numerical simulation. The true material parameter’s identification is still a difficult task. When the optimization process is combined with a finite element simulation, this is typically constrained by the computational time. In this study, a procedure for the identification of material parameters based on a hybrid approach is described. This methodology suggests a strategy for decreasing the computing cost by substituting the FEM simulations by an artificial neural network (ANN) model in the optimization loop. For this reason, a parametric study of the FE simulation is carried out to generate an ANN training database. A high-predictive-performance ANN model is also created by optimizing the hyperparameters. To quantify the conditioning of the inverse problem and to justify the replacement of the FE model with an ANN model, an identifiability analysis based on an identifiability indicator (I-index) is also proposed. The classical characterization tensile test is used to apply this optimization approach. Finally, numerical and experimental stress-strain tensile curves are compared to evaluate the effectiveness of this methodology.