Previsão do gradiente térmico e dimensões do cordão de solda no processo de soldagem GMAW: abordagem numérica e experimental

• Ano de defesa: 2024
• Autor(a) principal: Schauenberg, Aquiles da Silva
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• dARK ID: ark:/26339/001300001brm6
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Engenharia Mecânica; UFSM; Programa de Pós-Graduação em Engenharia Mecânica; Centro de Tecnologia
• Palavras-chave em Português: Soldagem com arame sólido sob proteção gasosa (GMAW); Método dos elementos finitos (MEF); Simulação de soldagem; Modelo de Goldak; Redes neurais artificiais (RNA); Gas metal arc welding (GMAW); Finite element method (FEM); Welding simulation; Goldak model; Abaqus welding interface (AWI); Artificial neural networks (ANN); CNPQ::ENGENHARIAS::ENGENHARIA MECANICA
• Link de acesso: http://repositorio.ufsm.br/handle/1/33825

The objective of this work is to compare different numerical models already used in the literature and develop a new application to predict the thermal gradient generated by GMAW welding in a "T" joint, reconciling the potential of each model. This more comprehensive approach aims to improve the computational processing time of the model while also providing more accurate results for the fusion zone (FZ) and heat-affected zone (HAZ). The numerical models were created using the finite element method (FEM), where the first model represents the welding process through a volumetric heat flow, presented in this study as Goldak's double ellipsoid. The geometry and intensity of the flow were implemented in the software through the DFLUX subroutine. For the second model, a deactivation or activation method of elements with prescribed temperatures was used. Finally, the new model is created by combining the strengths of the previous ones. Additionally, a comparison is made for the new proposed model regarding different starting points for the volumetric heat flow, showing a significant impact on the temperature gradient and processing time. The results indicate that the proposed new model has better agreement between the temperature distribution along the plate and the shape of the fusion zone, as well as improved processing time. Furthermore, an artificial neural network (ANN) is developed to predict the geometric parameters of the weld bead, aiming to reduce preliminary experimental tests to identify the geometric parameters of the heat source. The ANN is trained using 34 prior experimental data, with the input data being the process parameters: wire feed, voltage, current, and welding speed. The output data considered are the geometric parameters of the weld bead: leg length and penetration. The ANN uses the scaled conjugate gradient backpropagation algorithm. The other parameters of the weld bead are found trigonometrically through the values predicted by the ANN. The results demonstrate that the neural network is capable of predicting the geometry of the weld bead within a specific range of input parameters, with an average error of 10.74%.