Multi-domain multi-model formulation for compressible flows - Conservative interface coupling and parallel implicit solvers for 3D unstructured meshes

Marius Paraschivoiu, Xiao‐Chuan Cai, Marcus Sarkis, David P. Young, David E. Keyes37th 1999 · DOI

TL;DR

This paper introduces a novel multi-domain, multi-model approach for simulating compressible fluid flows on unstructured meshes. By intelligently coupling simpler, faster models (like the full potential equation) with more complex ones (like the Euler equations) in different regions of the flow, they achieve significant computational savings without sacrificing accuracy, particularly for transonic flows.

What Problem Does This Paper Address?

Simulating complex compressible fluid flows, especially for aerodynamic design, is computationally very expensive. While different mathematical models exist (full potential, Euler, Navier-Stokes), each has its limitations and computational cost. The full potential equation is fast but inaccurate for flows with strong shocks (like transonic flows). The Euler equations are more accurate but significantly more computationally demanding. The Navier-Stokes equations are even more so, required for viscous effects. The challenge is to leverage the speed of simpler models where they are valid, while using more accurate models only where necessary, thereby reducing the overall computational burden for large-scale simulations. This paper addresses the need for an efficient and accurate method to couple these different models on complex, unstructured 3D meshes.

Methods

The core of this work is a "multi-domain, multi-model" formulation. The computational domain is split into subdomains, each assigned a specific flow model. The key innovation lies in the conservative interface coupling between these domains. They employ a finite volume approach and a Defect Correction (DeC) method to solve the coupled nonlinear systems.

Specifically:

  • Domain Decomposition: The flow domain is partitioned into regions where the full potential equation (for irrotational, isentropic flow) and the Euler equations (for inviscid compressible flow) are solved.
  • Conservative Interface Coupling: A novel finite volume based conservative interpolation is developed to ensure smooth and accurate transfer of information across the interface between the full potential and Euler domains. This is crucial for maintaining the overall solution's integrity.
  • Implicit Solvers: To achieve reasonable computation times, especially for parallel processing, implicit solvers are used for both the full potential and Euler equations.
  • Defect Correction (DeC) Method: This iterative technique is used to solve the coupled nonlinear systems. It allows for the use of a first-order accurate Jacobian while achieving second-order spatial accuracy in the residual.
  • Parallel Implementation: The formulation is designed for parallel computing to further reduce computation time.

Datasets

Not discussed in the paper. The paper uses standard aerodynamic test cases (NACA0012 airfoil, AGARD wing 445.6) for validation.

Key Insights and Ideas

  • Smart Model Selection Pays Off: By strategically using the computationally cheaper full potential model in regions where it's valid (e.g., subsonic areas, away from shocks), significant computational savings can be achieved compared to using the Euler equations everywhere.
  • Interface is Key: The success of multi-model approaches hinges on robust and conservative coupling at the domain interfaces. The proposed finite volume interpolation method appears to achieve this, leading to smooth solutions across domain boundaries.
  • DeC for Efficiency: The Defect Correction method, combined with implicit solvers, provides a powerful framework for solving the coupled systems, allowing for efficient convergence.
  • Unstructured Meshes Offer Flexibility: The use of unstructured meshes is vital for handling complex geometries and for adaptive refinement, which is essential for accurate aerodynamic simulations.

Key Results

The paper demonstrates the feasibility of their approach through simulations of:

  • Subsonic flow over a NACA0012 airfoil: The coupled solution shows good agreement with the TRANAIR code (a well-established aerodynamic analysis tool) and captures the numerical entropy generated by the Euler solver near the stagnation point, which is absent in the pure full potential solution.
  • Transonic flow over a NACA0012 airfoil: The coupled solution shows a smooth transition of Mach number contours across the interface, and the pressure distribution is comparable to the Euler solution.
  • Flow over an AGARD wing 445.6: For this 3D case, the Mach number contours and pressure distributions from the coupled solution are nearly identical to the Euler-only solution, indicating that the coupling does not degrade accuracy.

While the paper highlights computational savings, specific quantitative speedup figures are not provided, as load balancing for parallel processing was still under development. However, they note that the full potential solver is "very fast" and that the coupled solver requires more iterations than the individual solvers, suggesting ongoing work on convergence acceleration.

Limitations and Future Work

  • Convergence Rate: The authors acknowledge that the convergence rate of the coupled solver is slower than individual solvers, particularly for transonic flows. They are investigating factors leading to this and exploring preconditioners (like restricted additive Schwarz) to improve linear convergence.
  • Load Balancing: For effective parallel performance, load balancing procedures are identified as crucial and are planned for future implementation.
  • Dynamic Zonal Configuration: Automating the placement of interfaces based on flow variables (dynamic zonal configuration) is a key area for future work to further enhance efficiency.
  • Extension to Navier-Stokes: The current work focuses on coupling full potential and Euler equations. The authors explicitly state their intention to extend this multi-model approach to include Navier-Stokes equations for flows with viscous effects.
  • Multi-level Model Approach: They are investigating a novel approach that uses different solvers at different computational steps to achieve faster solutions.

Summary updated

BibTeX
@inproceedings{paraschivoiu1999multi,
  author = {Marius Paraschivoiu and Xiao‐Chuan Cai and Marcus Sarkis and David P. Young and David E. Keyes},
  title = {{Multi-domain multi-model formulation for compressible flows - Conservative interface coupling and parallel implicit solvers for 3D unstructured meshes}},
  booktitle = {37th},
  year = {1999},
  doi = {10.2514/6.1999-784},
}
Figure