Deep convolutional neural networks in structural dynamics under consideration of viscoplastic material behaviour

The aim of the present study is to develop a deep convolutional neural network (DCNN) to predict geometrically and physically nonlinear structural deformations. Training data is obtained by short-time measurements in shock tubes, wherein metal plates are subjected to impulsive loadings, leading to viscoplastic vibrations and inelastic deflections. Due to the fact that, in literature, feed forward neural networks (FFNN) are more distributed for applications in structural mechanics, comparative calculations are presented between structural deformations based on DCNNs and FFNNs. Special attention is focused on the ability of DCNNs to capture also path-dependent deformations inside the network, which is an essential feature for inelastic material behaviour. (C) 2020 Elsevier Ltd. All rights reserved.