Surrogate-Assisted Optimization for Augmentation of Finite Element Techniques

The application of finite element techniques for the analysis and optimization of complex thermo-mechanical structures typically involves highly nonlinear models for material characterization, tribological contact, large deformation, damage, etc. These nonlinearities usually call for a higher-order Spatio-temporal discretization, including a large number of elements and time-steps in order to provide good convergence and sufficiently accurate simulation results. This inevitably leads to many expensive simulations in terms of cost and time if an optimization or adaption of model parameters has to be done. In this work, a FEM simulation modeling approach is proposed, which uses radial basis function interpolations (RBF) as efficient surrogate models to save FEM simulations. Also, a surrogate-assisted optimization algorithm [3] is utilized to find the parameter setting, which would lead to maximum damage in a simple tensile testing scenario involving a notched specimen with as few FEM simulations as possible. The relatively high accuracy of the utilized surrogate models showcases promising results and indicates the potential of surrogate models in saving time-expensive simulations.