Redes Neurais incorporadas a métodos numéricos para solução de problemas de engenharia estrutural

Abstract

The development of new materials and technologies, associated to increasingly complex problems in structural engineering, require more robust and accurate numerical analyses with an affordable computational cost. Computer modeling has been evolving by improving existing numerical methods and also developing new ones. Among the methods widely applied, finite elements have become a powerful tool in the search for solutions in structural engineering. Other approaches also emerged as alternative modeling strategies, such as artificial intelligence techniques. One of these techniques, Artificial Neural Networks (ANN), is designed for simulating human conexionism paradigm. ANN has been applied in several areas to solve a range of problems concerning classification, function fitting, forecasting, among others. This work explores ways to coupling ANN to traditional approximation methods such as the Finite Element Method (FEM) for applications in plane elastostatics, considering geometrically nonlinear aspects.