Abstract:This work deals with the investigation of bifurcating fluid phenomena using a reduced order modelling setting aided by artificial neural networks. We discuss the POD-NN approach dealing with non-smooth solutions set of nonlinear parametrized PDEs. Thus, we study the Navier-Stokes equations describing: (i) the Coanda effect in a channel, and (ii) the lid driven triangular cavity flow, in a physical/geometrical multi-parametrized setting, considering the effects of the domain's configuration on the position of the bifurcation points. Finally, we propose a reduced manifold-based bifurcation diagram for a non-intrusive recovery of the critical points evolution. Exploiting such detection tool, we are able to efficiently obtain information about the pattern flow behaviour, from symmetry breaking profiles to attaching/spreading vortices, even at high Reynolds numbers.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in computational fluid dynamics (CFD), how to effectively study bifurcation phenomena using reduced - order models and artificial neural networks (ANN). Specifically, the authors focus on the bifurcation behavior in fluid phenomena described by nonlinear parameterized partial differential equations (PDEs), which include: 1. **Coanda effect**: Viscous, steady, and incompressible flow occurring in a channel, and this phenomenon is related to a heart disease - mitral regurgitation. 2. **Triangular cavity - driven flow**: Complex flow behavior occurring in a triangular cavity. ### Research Background Although traditional high - fidelity discretization methods (such as the finite element method) can accurately simulate these complex fluid phenomena, they face the following challenges when dealing with bifurcation problems: - **High computational cost**: A large amount of computational resources are required to solve nonlinear systems. - **Multi - solution problem**: When the parameters reach the critical value, the system may lose uniqueness and multiple branch solutions may appear. - **Real - time and multi - query requirements**: In practical applications, solutions under different parameters need to be quickly evaluated, and traditional methods are difficult to meet this requirement. ### Solution To solve the above problems, the authors propose a method based on POD - NN (Proper Orthogonal Decomposition - Neural Network), which combines the advantages of reduced - order models (ROM) and artificial neural networks. The main objectives of this method are: - **Efficient reconstruction of bifurcation diagrams**: Represent the physical characteristics of the system by constructing a low - dimensional manifold and provide the ability to quickly evaluate solutions under new parameters. - **Non - invasive restoration of key point evolution**: Use neural networks to map from the parameter space to the reduced - order coefficients, thereby avoiding the complex assembly and solution processes in traditional methods. ### Specific Steps 1. **Dataset construction**: Use POD technology to extract the main information from high - fidelity solutions and construct a training dataset. 2. **Neural network training**: Train a feed - forward neural network (MLP) to map input parameters to reduced - order coefficients. 3. **Non - invasive solution**: Directly obtain the reduced - order coefficients for any given parameters by querying the trained neural network, without the need for initial guesses or complex nonlinear solvers. 4. **Error evaluation**: Evaluate the relative errors of the POD - Galerkin and POD - NN methods through the test set to ensure the accuracy and reliability of the model. ### Application Cases The paper analyzes two specific computational fluid dynamics cases: - **Coanda effect**: Study the influence of sudden expansion in the channel on the flow behavior. - **Triangular cavity - driven flow**: Explore the influence of changes in geometric parameters on the flow pattern. Through these cases, the authors demonstrate the potential of the POD - NN method in efficiently capturing bifurcation phenomena, especially under high Reynolds number conditions, and it can reveal complex flow behaviors such as symmetry breaking, vortex attachment, and expansion. ### Summary This paper proposes an innovative method that combines the advantages of POD and ANN to solve the research problems of bifurcation phenomena in computational fluid dynamics. This method not only improves computational efficiency but also can accurately capture the complex behaviors of the system in a multi - parameter environment, providing new ideas and tools for research in related fields.

An artificial neural network approach to bifurcating phenomena in computational fluid dynamics

Non-intrusive reduced order models for the accurate prediction of bifurcating phenomena in compressible fluid dynamics

An optimisation-based domain-decomposition reduced order model for parameter-dependent non-stationary fluid dynamics problems

A Comparison of Reduced-Order Modeling Approaches Using Artificial Neural Networks for PDEs with Bifurcating Solutions

Flow Sensing Method for Fluid-Structure Interaction Systems Via Multilayer Proper Orthogonal Decomposition

A stochastic perturbation approach to nonlinear bifurcating problems

Computational reduction strategies for the detection of steady bifurcations in incompressible fluid-dynamics: applications to Coanda effect in cardiology

On the Application of Reduced Basis Methods to Bifurcation Problems in Incompressible Fluid Dynamics

Neural networks in feedback for flow analysis and control

Neural networks in feedback for flow analysis, sensor placement and control

Proper Orthogonal Decomposition and Physical Field Reconstruction with Artificial Neural Networks (ANN) for Supercritical Flow Problems

FEM-based Neural Networks for Solving Incompressible Fluid Flows and Related Inverse Problems

An artificial neural network framework for reduced order modeling of transient flows

Physics-informed neural networks with domain decomposition for the incompressible Navier–Stokes equations

Solving free-surface problems for non-shallow water using boundary and initial conditions-free physics-informed neural network (bif-PINN)

An optimisation-based domain-decomposition reduced order model for the incompressible Navier-Stokes equations

Predictive Limitations of Physics-Informed Neural Networks in Vortex Shedding

Network-theoretic modeling of fluid-structure interactions

Solution multiplicity and effects of data and eddy viscosity on Navier-Stokes solutions inferred by physics-informed neural networks

Learning the Latent dynamics of Fluid flows from High-Fidelity Numerical Simulations using Parsimonious Diffusion Maps

Artificial Neural Networks in Fluid Dynamics: A Novel Approach to the Navier-Stokes Equations