A neural network based non-intrusive surrogate modelling framework for fluid-structure interaction

Fairchild, D; Şugar-Gabor, O

Abstract

The use of surrogate modelling in Multiphysics simulation is becoming increasingly attractive due to large computation time and hardware demands of modern simulations. These methods aim to reduce the computational time and complexity of the determination of the response of a system by constructing a simple model which approximates the input-output behaviour of the underlying model. This paper investigates the application of surrogate modelling to multiphysics unsteady parametric non-linear simulations, specifically fluid-structure interaction. The proposed framework, composed of a fully non-intrusive Proper Orthogonal Decomposition and Neural Networks scheme is developed and applied to both a one and two-dimensional fluid structure interaction cases, with the prediction accuracy and reduction in computational time being investigated in respect to the high order simulation. The proposed surrogate framework shows promise in accurately and efficiently predicting the response of the more complex fluid structure interaction problems.

Citation

Fairchild, D., & Şugar-Gabor, O. (in press). A neural network based non-intrusive surrogate modelling framework for fluid-structure interaction. International Journal of Multiphysics, 18(3),

• Journal Article Type: Article
• Acceptance Date: Jun 10, 2024
• Deposit Date: Jul 25, 2024
• Journal: International Journal of Multiphysics
• Print ISSN: 1750-9548
• Publisher: International Society of Multiphysics
• Peer Reviewed: Peer Reviewed
• Volume: 18
• Issue: 3
• Keywords: Surrogate modelling; fluid-structure interaction; artificial neural networks