An approach for displacement prediction in truss structures combining the Finite Element Method and Deep Learning Techniques

Resumo

Recent advancements in machine learning have facilitated groundbreaking applications across various
knowledge domains. This paper introduces a promising application of Deep Learning Techniques (DL) in
conjunction with the Finite Element Method (FEM). The aim of this study is to evaluate a convolutional neural
networks ́s (CNN) ability to predict 2D truss displacement fields. This assessment leverages prior knowledge of
the structure's global stiffness matrix (K) and applied external load vector (f). The technique's advantage lies in
circumventing complex numerical approaches for solving linear and nonlinear systems, which often entails
extended processing times and substantial computational effort depending on model complexity. The employed
methodology involves constructing a dataset comprising varied structural configurations, processed through a
finite element solver. This dataset then trains a CNN composed of residual, dense, and fully connected blocks. The
outcomes underscore the significance of this approach; the proposed model demonstrates commendable
performance in this task, exhibiting a maximum absolute mean error of approximately 2% and a maximum mean
squared error of 0.2%, when contrasted with classical FEM solutions.