Aplicação de redes neurais no apoio à tomada de decisão para a aprovação de componentes estruturais mecânicos

Abstract

The Finite Element Method (FEM) is a virtual calculation resource used around the world to determine approximate numerical solutions to engineering problems. It is common for an analysis carried out by the MEF to be delayed according to its complexity, making a project expensive and prolonged. Given this scenario, the present work sought to develop an Artificial Neural Network (ANN) capable of predicting finite element responses and fatigue life for a brake spider, as a support tool in real-time decision making for structural validation of mechanical components. For this, a numerical virtual model was calibrated by the FEM based on theories of mechanics, results obtained from bench tests, and instrumentations. Based on virtual calibration, a finite element database was generated for the ANN application, containing 130 data from a total of 4.800 possibilities. ANN training, validation, and testing were determined using an ANN performance analysis algorithm. Finally, the results obtained with the ANN were compared with the results of finite elements and computational fatigue life, and the efficiency of the real-time response prediction method was measured using the Mean Squared Error (MSE). [resumo fornecido pelo autor]