Critical Analysis of the Suitability of Surrogate Models for Finite Element Method Application in Catalog-Based Suspension Bushing Design

Abstract

This work presents a critical analysis of the suitability of surrogate models for finite element method application. A case study of a finite element method (FEM) structural problem was selected in order to test the performance of surrogate algorithms. A simple design of experiments (DoE) approach, based on 1D kernel density estimations, is employed to construct a representative pool of real FEM simulations, which becomes the dataset for five different surrogate models, two linear and three non-linear, whose most relevant hyperparameters were tuned (model selection). Results in a real bushing case study show that surrogate models can accurately mimic FEM simulations outcomes, in this case four types of stiffnesses (axial, radial, torsion, and cardanic).