Marcus Stoffel

In this study, we introduce a Generative learning‐based approach to accelerate Finite Element simulations. The aim is to explore the ability of a Generative learning‐based approach to predict the output of Finite Element simulations. Often, the drawback of classical regression models is that they are bound to the distribution of training data and cannot extrapolate outside of the training domain. In the current study, to overcome this, we propose a Generative Adversarial Network (GAN) based approach for learning to infer structural mechanics simulations. GANs have the ability to learn the underlying distribution of data and can efficiently extrapolate outside the training range. We train the proposed network on data obtained from classical Finite Element simulations based on linear elastic and Johnson‐Cook plasticity models. The goal here is to understand the ability of a GAN to replace FEM by incorporating the underlying mechanics knowledge into the network to enhance the generalizing ability of the network for various materials on various loading and boundary conditions. We propose an encoder‐decoder‐like U‐Net generator architecture with activated skip connections and a PatchGAN‐based discriminator that discriminates based on local patches of data leading to a more robust critique for the generator. The advantage of the proposed framework is that it takes inputs such as the nodal information, their corresponding edges, nodal coordinates and the boundary conditions for each particular node from a Finite Element pre‐processor and computes the von Mises stress at each node as output that can be read by a Finite Element post‐processor. Also, the novelty of the proposed GAN is that, unlike the existing literature, it takes nodal values such as applied Boundary conditions and coordinates as inputs instead of images and performs nonlinear regression with Finite Element data and gives von‐Mises stresses at each node as output.