Stability of feed forward artificial neural networks versus nonlinear structural models in high speed deformations: A critical comparison

Abstract

In recent years, artificial neural networks have been proposed for engineering applications, such as predicting stresses and strains in structural elements. However, the question arises, how many complex influences can be included in an artificial neural network (ANN) and how accurate these predictions are in comparison to classical finite element solutions. A weakness of finite element predictions is that they can behave sensitive and unstable to changes in material parameters. An ANN does not need an underlying model with parameters and uses input variables, only. In the present study the stability of numerical results obtained by ANN and FEM are compared to each other for a problem in structural dynamics. The result gives new insight about the possibilities to predict accurately structural deformations by means of ANNs. As an example for highly complex geometrically and physically nonlinear structural deformations, the response of circular metal plates subjected to shock waves is investigated.