A novel self-adversarial training scheme for enhanced robustness of inelastic constitutive descriptions by neural networks. (June 2022)

• Record Type: Journal Article
• Authors: Stöcker, Julien; Fuchs, Alexander; Leichsenring, Ferenc; Kaliske, Michael
• Is Part Of: Computers & structures. Volume 265(2022)
• Publication Date: 2022-06
• Subjects: Data-driven modeling -- Machine Learning -- Neural network constitutive description -- Recurrent neural network -- Multiscale modeling -- Self-Adversarial Training; Structural engineering -- Data processing -- Periodicals; Electronic data processing -- Structures, Theory of -- Periodicals; 624.171
• DOI: 10.1016/j.compstruc.2022.106774
• Languages: English
• ISSNs: 0045-7949

Abstract

Highlights: Accuracy of neural network constitutive model drops for output-input data feedback. Neural networks own prediction error generates an adversarial attack during testing. Training data with perturbed samples representing predictions improves performance. Adaptive sampling to account for increased robustness due to training progress. Abstract: This contribution presents a novel training algorithm to increase the robustness of recurrent neural networks (RNN) used as a constitutive description that are subjected to perturbations induced by its own prior output. We propose to extend the data obtained from Numerical Material Tests (NMT) on Representative Volume Elements (RVE) by generating adversarial examples based on the prediction errors. This method introduces new hyperparameters like the training length before reevaluating the errors and the fraction of adversarial examples contained in the dataset. Therefore, numerical investigations of an RVE, considering two different sets of materials with elasto-plastic behavior, are conducted and a set of hyperparameters for the Self-Adversarial Training, that result in high prediction robustness, are identified. The capabilities and limitations of the application of a neural network based constitutive description with enhanced robustness are evaluated and discussed for a numerical simulation on structural level.