Artificial Neural Network augmented hyper-reduced order model for transient flow optimization problems
Year: 2026
Type: article
Abstract: This paper introduces an Artificial Neural Network (ANN) augmented, Variational Multiscale (VMS) stabilized, hyper-reduced order model (HROM) for time-dependent fluid flow shape optimization applications. The method extends the authors VMS-based ROM with Discrete Empirical Interpolation Method (DEIM) hyper-reduction to achieve further computational efficiency. A fully discrete level nonlinear ANN correction term is applied to account for errors associated with both ROM projection and hyper-reduction that improves the stability, as well as the accuracy required for unsteady flows. The HROM approach we obtain here is implemented into a Sequential Least Squares Quadratic Programming (SLSQP) based gradient-based aerodynamic shape optimization problem. NACA airfoils at high angles of attack are chosen to get the transient effects like separation and vortex-shedding. The ANN correction is trained on snapshots from full-order simulations and used directly in the optimization loop at no additional cost. Numerical results show that the HROM with ANN-correction successfully captures the full order model accuracy for both transient prediction and optimization and achieves more than 95% reduction in computational cost. The results validate the effectiveness of the developed ANN-augmented HROM as an accurate and efficient counterpart for transient-CFD-based optimization problems.
Authors: Zulkeefal Dar, Joan Baiges, Ramon Codina
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,830
OpenAlex ID: W7162400234