Finite Element Modeling of Indentation Tests and Neural Networks Model Applications

Abstract

The diploma work investigates the Instrumented Indentation Test (IIT) modelling with Finite Element Method (FEM) and Neural Network (NN). The final goal is to create a NN that can evaluate the experimental data from a physical IIT in such a way to directly calibrate a suitable plasticity model. The study is conducted on an austenitic stainless steel SS304L which is a challenge for IIT. To achieve the goal, IIT is performed on a SS304L specimen to obtain the indenting depth versus the reaction force the material exerts on the indentation ball. Besides, tensile tests with Digital Image Correlation (DIC) on the same material are performed to obtain the tensile stress-strain curve, from which parameters of a constitutive model can be identified. The results from IIT and tensile test are then used to establish a FEM model. Using the FEM model, several IITs can be simulated to create a dataset for NN training in the range of austenitic steels. This dataset is fed into a NN and numerical experiments are done to find an optimal architecture for the final goal. The thesis proposes a framework and proves its feasibility considering Armstrong-Frederick hardening model with three parameters. All the data and identified challenges are reported and discussed.