Abstract
A numerical code for solving axisymmetric compressible Navier-Stokes equations using the spectral collocation method was developed and applied to supersonic flow over a sphere. A novel Residual-Dependent Filter was introduced to significantly improve accuracy and stability. The study investigates the influence of filtering parameters, time step size, and grid density, demonstrating excellent agreement between the numerical results and experimental data.
Introduction
Spectral methods are considered the "gold standard" for accuracy in computational fluid dynamics (CFD) because their error decreases exponentially as grid points are added. However, applying these methods to compressible flows with shock waves is notoriously difficult due to the Gibbs phenomenon—high-frequency oscillations triggered by discontinuities. These oscillations can lead to numerical instability or the "smearing" of critical physical features like boundary layers. While filters are commonly used to suppress these oscillations, standard fixed filters often compromise accuracy. This research introduces a Residual-Dependent Filter designed to maintain spectral precision while stabilizing simulations of supersonic, viscous flows.
Methodology
The researchers solved the axisymmetric compressible Navier-Stokes equations using a spectral collocation (pseudospectral) method. The approach incorporates several key techniques:
- Coordinate Transformation and Shock-Fitting: Moretti’s transformation was employed to map the physical space between the sphere and the bow shock into a computational square. A shock-fitting technique treated the shock as a moving boundary, ensuring sharp resolution.
- Spatial Discretization: Chebyshev polynomials were used for spatial derivatives, allowing for non-uniform grid spacing that naturally clusters points near the wall to resolve thin boundary layers.
- Temporal Integration: A second-order predictor-corrector scheme was used for time stepping.
- Residual-Dependent Filter: This core innovation adjusts the filtering strength based on the local residual. By distinguishing between developing physical features (low-frequency) and numerical noise (high-frequency), the filter aggressively damps errors without smoothing out physical gradients.
Results and Discussion
The implementation of the Residual-Dependent Filter allowed the simulation to achieve high stability and precision even with relatively coarse grids.
As shown in the temperature distribution (Figure 2), the method successfully captures the thermal gradients near the sphere's surface. The authors found that while a $9 \times 9$ grid could capture the velocity boundary layer, a $17 \times 17$ grid was necessary for accurate temperature recovery factor ($R_c$) calculations, matching experimental trends from Hickman and Giedt (1963).
Numerical stability was a primary focus. The Residual-Dependent Filter reduced the maximum density residual significantly compared to standard fixed filters. As seen in Figure 4, the residual stabilizes after an initial transient period, which is crucial for long-duration simulations and high-fidelity results. Furthermore, the calculated pressure distribution on the sphere surface showed excellent agreement with experimental data, outperforming Modified Newtonian Theory in downstream regions.
Conclusion
The spectral collocation method, enhanced by a Residual-Dependent Filter, proves to be a powerful tool for simulating supersonic viscous flows. It achieves high accuracy with significantly fewer grid points than traditional finite-difference methods. Future work will focus on extending this approach to 3D complex geometries and developing more robust shock-capturing techniques for internal discontinuities.
Summary updated