TL;DR
This paper introduces a novel computational fluid dynamics (CFD) framework that combines the Shifted Boundary Method (SBM) with incomplete adaptive octree meshes to simulate incompressible fluid flows over highly complex geometries. By shifting boundary conditions to an optimized surrogate boundary, the method eliminates the need for tedious body-fitted mesh generation while completely avoiding the numerical instabilities typical of traditional immersed boundary methods.
What Problem Does This Paper Address?
Simulating fluid flow around complex or irregular shapes—such as airfoils, porous structures, or intricate industrial components—traditionally requires "body-fitted" meshes that perfectly conform to the geometry. Generating these meshes is a massive bottleneck in engineering workflows: it is labor-intensive, requires extensive manual tuning, and struggles with highly complex topologies.
"Immersed Boundary Methods" (IBMs) offer an alternative by embedding the complex geometry within a simple, fixed background grid. However, when grid lines cut arbitrarily through a geometry, they create tiny, awkward "sliver cells." These sliver cells cause severe numerical ill-conditioning, ruining the stability and accuracy of the simulation unless complex, computationally heavy fixes are applied.
The authors address this gap by extending the Shifted Boundary Method (SBM) to work with incomplete adaptive octree grids for incompressible Navier-Stokes simulations, providing an efficient, stable, and automated pipeline for complex flow physics.
Methods
The framework integrates several advanced numerical techniques:
- Shifted Boundary Method (SBM): Instead of enforcing boundary conditions directly on the true complex surface ($\Gamma$), SBM "shifts" them to a nearby surrogate boundary ($\tilde{\Gamma}_h$) that naturally aligns with the faces or edges of the background mesh. To prevent loss of accuracy, a Taylor expansion correction is applied to map the true boundary conditions onto this surrogate surface.
- Incomplete Adaptive Octree Meshes: The background grid uses an octree structure, allowing for massive local refinement where physics demand it (e.g., boundary layers and wake regions). Crucially, the octree is "incomplete," meaning cells completely outside the fluid domain are "carved out" to eliminate wasted memory and computation.
- Optimal Surrogate Boundary ($\lambda = 0.5$): The surrogate boundary is constructed using a volume-fraction threshold ($\lambda$). By setting $\lambda = 0.5$, the algorithm selects cells that have at least 50% of their volume inside the true domain, which mathematically minimizes the root-mean-square distance between the true and surrogate boundaries.
- Variational Multiscale (VMS) Formulation: The incompressible Navier-Stokes equations are discretized using a VMS approach, which simultaneously handles numerical stabilization and acts as a Large Eddy Simulation (LES) turbulence model.
Datasets
The framework was rigorously validated using a variety of canonical CFD benchmarks and complex geometric datasets:
- 2D Benchmarks: Unsteady flow past a cylinder ($Re=100$), lid-driven cavity flow with a circular obstacle ($Re=5000$), and aerodynamic flow past a NACA 0012 airfoil ($Re=1000$).
- 3D Benchmarks: Flow past a sphere ($Re=300$ and $3700$), flow past a 3D cylinder ($Re=3900$), and internal flow through a contracting/expanding nozzle.
- Complex Geometries: Non-parametric shapes (such as plane, cat, and cattle profiles) sampled from the SkelNetOn dataset, and a highly intricate 3D gyroid topology embedded inside a pipe flow configuration.
Key Insights and Ideas
- Bypassing the Sliver Cell Problem: Because SBM integrates over standard volume Gauss points on the surrogate boundary rather than performing complex surface quadratures on cut-cells, it completely bypasses the ill-conditioning caused by tiny sliver cells.
- The Magic of 50% Volume ($\lambda = 0.5$): The authors demonstrate that setting the cell inclusion threshold to exactly 50% volume yields the smoothest pressure profiles. Choosing $\lambda = 0$ or $\lambda = 1$ introduces significant artificial pressure oscillations along the surface due to larger variations in the distance function.
- Pixelated but Accurate: Even though the surrogate boundary looks "pixelated" or staircase-like to the grid, the Taylor-series correction ensures that the fluid experiences the true, smooth geometry accurately, preserving global mass and momentum conservation.
Key Results
- Massive Error Reduction: The SBM approach achieved boundary velocity errors approximately two orders of magnitude smaller than standard "strong" boundary condition methods on the same pixelated grids.
- High Quantitative Accuracy: Key aerodynamic metrics like drag coefficients ($C_d$) and Strouhal numbers ($St$) closely matched high-fidelity DNS, LES, and experimental literature baselines across all test cases (e.g., capturing the complex horseshoe-shaped vortices in flow past a sphere at $Re=3700$).
- Excellent Parallel Scalability: Built on top of the Dendro-KT and PETSc libraries, the framework demonstrated near-ideal parallel scaling up to 392 processors on the TACC Frontera supercomputer.
Limitations and Future Work
- Thermal Flows: The current framework is restricted to isothermal incompressible flows. The authors plan to integrate the convection-diffusion equation to handle thermal transport and buoyancy-driven flows.
- Fluid-Structure Interaction (FSI): Future work will focus on coupling this octree-SBM fluid solver with structural shell formulations to simulate moving, flexible boundaries.
- Higher-Order Elements: The current implementation relies on piecewise-linear (first-order) FEM basis functions. Expanding the framework to support higher-order basis functions could significantly increase accuracy per grid node.
Summary updated