Physics-informed two-tier neural network for non-linear model order reduction

Abstract In recent years, machine learning (ML) has had a great impact in the area of non-intrusive, non-linear model order reduction (MOR). However, the offline training phase often still entails high computational costs since it requires numerous, expensive, full-order solutions as the training da...

Full description

Saved in:
Bibliographic Details
Main Authors: Yankun Hong, Harshit Bansal, Karen Veroy
Format: Article
Language:English
Published: SpringerOpen 2024-11-01
Series:Advanced Modeling and Simulation in Engineering Sciences
Subjects:
Online Access:https://doi.org/10.1186/s40323-024-00273-3
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1846165070963277824
author Yankun Hong
Harshit Bansal
Karen Veroy
author_facet Yankun Hong
Harshit Bansal
Karen Veroy
author_sort Yankun Hong
collection DOAJ
description Abstract In recent years, machine learning (ML) has had a great impact in the area of non-intrusive, non-linear model order reduction (MOR). However, the offline training phase often still entails high computational costs since it requires numerous, expensive, full-order solutions as the training data. Furthermore, in state-of-the-art methods, neural networks trained by a small amount of training data cannot be expected to generalize sufficiently well, and the training phase often ignores the underlying physical information when it is applied with MOR. Moreover, state-of-the-art MOR techniques that ensure an efficient online stage, such as hyper reduction techniques, are either intrusive or entail high offline computational costs. To resolve these challenges, inspired by recent developments in physics-informed and physics-reinforced neural networks, we propose a non-intrusive, physics-informed, two-tier deep network (TTDN) method. The proposed network, in which the first tier achieves the regression of the unknown quantity of interest and the second tier rebuilds the physical constitutive law between the unknown quantities of interest and derived quantities, is trained using pretraining and semi-supervised learning strategies. To illustrate the efficiency of the proposed approach, we perform numerical experiments on challenging non-linear and non-affine problems, including multi-scale mechanics problems.
format Article
id doaj-art-c1fa0e3cca7b41b18d1c89541b51f514
institution Kabale University
issn 2213-7467
language English
publishDate 2024-11-01
publisher SpringerOpen
record_format Article
series Advanced Modeling and Simulation in Engineering Sciences
spelling doaj-art-c1fa0e3cca7b41b18d1c89541b51f5142024-11-17T12:38:41ZengSpringerOpenAdvanced Modeling and Simulation in Engineering Sciences2213-74672024-11-0111114910.1186/s40323-024-00273-3Physics-informed two-tier neural network for non-linear model order reductionYankun Hong0Harshit Bansal1Karen Veroy2Centre for Analysis, Scientific Computing and Applications, Eindhoven University of TechnologyCentre for Analysis, Scientific Computing and Applications, Eindhoven University of TechnologyCentre for Analysis, Scientific Computing and Applications, Eindhoven University of TechnologyAbstract In recent years, machine learning (ML) has had a great impact in the area of non-intrusive, non-linear model order reduction (MOR). However, the offline training phase often still entails high computational costs since it requires numerous, expensive, full-order solutions as the training data. Furthermore, in state-of-the-art methods, neural networks trained by a small amount of training data cannot be expected to generalize sufficiently well, and the training phase often ignores the underlying physical information when it is applied with MOR. Moreover, state-of-the-art MOR techniques that ensure an efficient online stage, such as hyper reduction techniques, are either intrusive or entail high offline computational costs. To resolve these challenges, inspired by recent developments in physics-informed and physics-reinforced neural networks, we propose a non-intrusive, physics-informed, two-tier deep network (TTDN) method. The proposed network, in which the first tier achieves the regression of the unknown quantity of interest and the second tier rebuilds the physical constitutive law between the unknown quantities of interest and derived quantities, is trained using pretraining and semi-supervised learning strategies. To illustrate the efficiency of the proposed approach, we perform numerical experiments on challenging non-linear and non-affine problems, including multi-scale mechanics problems.https://doi.org/10.1186/s40323-024-00273-3Physics-informed machine learningNeural networksNon-linear model order reductionHyper-reduction
spellingShingle Yankun Hong
Harshit Bansal
Karen Veroy
Physics-informed two-tier neural network for non-linear model order reduction
Advanced Modeling and Simulation in Engineering Sciences
Physics-informed machine learning
Neural networks
Non-linear model order reduction
Hyper-reduction
title Physics-informed two-tier neural network for non-linear model order reduction
title_full Physics-informed two-tier neural network for non-linear model order reduction
title_fullStr Physics-informed two-tier neural network for non-linear model order reduction
title_full_unstemmed Physics-informed two-tier neural network for non-linear model order reduction
title_short Physics-informed two-tier neural network for non-linear model order reduction
title_sort physics informed two tier neural network for non linear model order reduction
topic Physics-informed machine learning
Neural networks
Non-linear model order reduction
Hyper-reduction
url https://doi.org/10.1186/s40323-024-00273-3
work_keys_str_mv AT yankunhong physicsinformedtwotierneuralnetworkfornonlinearmodelorderreduction
AT harshitbansal physicsinformedtwotierneuralnetworkfornonlinearmodelorderreduction
AT karenveroy physicsinformedtwotierneuralnetworkfornonlinearmodelorderreduction