Search the digital library catalog

Abstract: The basis of this thesis is research and development of a genetic algorithm for material parameters optimization. It is written in collaboration with AVL, which already has a solution for this problem, but is looking into better alternatives. Chaboche material model is a nonlinear isotropic and kinematic hardening model which can describe elasto-viscoplastic constitutive relations. Parameters of such complex nature do not have a physical interpretation in the real-world and must be defined with inverse analysis. Genetic algorithms (GA) are a promising tool to help with such tasks. They have been widely used and recognized for various optimization problems. Material data available are low cycle fatigue (LCF), creep, and tensile experiments. For each experiment a corresponding finite element model in Abaqus is prepared. Comparing experimental and simulation data is the objective function GA will try to minimize. For this reason, a corresponding fitness function was developed to score each individual. It makes use of similarity measure algorithm proposed in this paper [10]. GA was implemented in Python with Pygad library. Instead of bits, genes are represented with real-valued numbers with defined limits. Performance of developed GA was tested based on various population sizes, mutation probabilities, and crossover operators. The main parameter that impacts algorithms performance is population size. Paired with right mutation probability the algorithm can find a global minimum of described optimization problem. Making it a viable alternative to existing approach used at AVL.
Keywords: Chaboche material model, parameter optimization, genetic algorithm, finite element method
Published in DKUM: 16.12.2022; Views: 872; Downloads: 0
Full text (1,90 MB)