Aplicação de redes neurais artificiais na estimativa de propriedades mecânicas de materiais a partir de ensaios de indentação

Abstract

Indentation testing is a non-destructive method used to evaluate the mechanical properties of materials. This test consists of applying a controlled force to the surface of the material through a tip with high hardness and a specific geometric shape. With the depth and force data, the determination of the mechanical properties of the material under analysis can be carried out in different ways. In this work, an Artificial Intelligence (AI) method was implemented to identify the parameters of the Ludwik equation that describes the plastic phase of the stress-strain curve of materials. For this purpose, first, simulations were performed to generate a database with 10 360 different combinations of hardening exponent $n$, strength coefficient $K$, and yield strength $\sigma_y$. Then, this database was divided into three distinct sets: Training, validation, and testing. The first was used to train three Artificial Neural Networks (ANNs) used to estimate $K$, $n$, and $\sigma_y$ from the data of the simulated indentation. The validation set was used to optimize the architecture of these networks. Finally, the test set evaluated the generalization ability of the developed ANNs. The results showed that the AI method was able to identify the coefficient of resistance and yield strength parameters of the Ludwik equation with good accuracy. However, it was observed that the method was not able to determine the hardening exponent with good results.