Banca de DEFESA: Paulo Guilherme Marques Flávio

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
DISCENTE : Paulo Guilherme Marques Flávio
DATA : 02/05/2023
HORA: 14:00
LOCAL: Plataforma Microsoft Teams
TÍTULO:

SUPERVISED MACHINE LEARNING FRAMEWORKS FOR THE ANALYSIS OF CONTACTING SOLIDS UNDER INELASTIC STRAINS

PALAVRAS-CHAVES:

Artificial Neural network; Finite Element; Visco-elastoplastic model; Mulliken-Boyce; indentation

PÁGINAS: 135
RESUMO:

The study about the mechanical behavior of materials has always been a constant target of research, the most prominent being experimental and numerical approaches. Tensile and compression tests are the most common characterization tests; however, other tests are a viable alternative. Concerning numerical approaches, modeling using constitutive equations coupled to a Finite Element discretization of the solids is another widely accepted option. Artificial Neural Networks (ANN) can be understood as a complex computing system made up of a series of interconnected processing elements inspired by the Biological Neural Networks that have the purpose of processing information aiming to generate a response. Its use is constantly growing, being found in the most diverse fields. Thus, its implementation in parallel with other commonly used tools, such as the case of the Finite Element, can result in several gains, especially regarding the description of behaviors that are difficult to characterize and evaluations that demand high computational cost. This study aims to combine the capabilities of an Artificial Neural Network, developed using a supervised learning algorithm for predicting the mechanical behavior and material properties of the respective constitutive models adopted for metallic and polymeric materials The evaluation of the polymers is restricted to thermoplastic polymers and a viscoelastoplastic model based on the constitutive Mulliken-Boyce model is adopted. The implementation of the model is carried out via VUMAT subroutines (explicit integration) in the commercial software ABAQUS for the study of uniaxial compression in large deformations. The desired predictions for the study of the evaluated polymers are: (1) Prediction of the material parameters of the modified constitutive model of Mulliken-Boyce implemented, from the experimental stress-strain curve and (2) prediction of the stress-strain curve from the material parameters. The study of metals is based on indentation tests at large deformations. Ludwik equation is adopted for the description of the mechanical behavior and is implemented via implicit integration in the same commercial software. The desired predictions for metals are: (1) prediction of the stress-strain curve from the displacement-force curve obtained through indentation test and (2) prediction of the displacement-force curve, residual indentation and contact pressure from the main mechanical properties of metals. All predictions are performed through the implementation of deep neural networks and supervised training. The results achieved are satisfactory, the performance of the predictions made for all evaluated cases show good accuracy with a significant reduction in processing time compared to implementations via Finite Element.

MEMBROS DA BANCA:
Presidente - 2143651 - THIAGO DE CARVALHO RODRIGUES DOCA
Interno - 1722212 - EDER LIMA DE ALBUQUERQUE
Interno - 1646090 - FABIO COMES DE CASTRO
Externo à Instituição - MARCELO GRECO - UFMG
Externo à Instituição - RAIMUNDO CARLOS SILVÉRIO FREIRE JÚNIOR - UFRN