Implementation of a hybrid strategy for addressing mechanical structural models: AI and finite element analysis

Resumo

FEM is one of the main numerical methods in engineering to solve boundary problems; however, the computational cost associated with this method is a challenge to be faced when using it in real cases. The objective of this paper is to evaluate the performance of Neural Networks (NN) in the approximation of results obtained by FEM while maintaining accuracy and reducing computational time. The results show an accurate model for the case of a flat plate subjected to tension, altering the elastic modulus (E), applied force (F), and plate thickness (h) in the trainig data. This work is a starting point for the optimization of computational time in the calculation of the approximate solution of engineering problems of greater complexity. In the future, hybrid strategies of FEM with neural networks could benefit mathematical modeling and computational simulation, making its use viable in critical areas such as medicine, where the presence of tissues with nonlinear characteristics is a common challenge to be solved or where real-time decision making is crucial.