CFD-101: The Basics of Computational Fluid Dynamics Modeling
What You Should Know About Computational Fluid Dynamics Modeling
Software packages for fluid flow and heat transfer analysis come in many forms. These packages differ greatly in their physical approximations and numerical solution techniques, which makes the selection of a suitable package a challenging proposition. The following discussion covers some important items to consider when choosing flow simulation software. Topics include:
- Meshing and Geometry
- Momentum Equation vs. Approximate Flow Models
- Liquid-Solid Heat Transfer Area
- Control Volume Effects on Liquid-Solid Heat Transfer
- Implicitness and Accuracy
- Implicit Numerical Methods For Convective Transport
- Relaxation and Convergence Parameters
- Free-Surface Tracking
Solution methods that employ finite-element or "body-fitted coordinates" require the generation of a solution grid that conforms to the geometry of the flow region. It is a non-trivial task to generate these grids with acceptable element sizes and shapes for accurate numerical approximations. In complicated cases this type of grid generation may consume days or even weeks of effort. Some programs attemptto eliminate this generation problem by using only rectangular grid elements, but then they must contend with "stair-step" boundaries that alter flow and heat-transfer properties. FLOW-3D solves both problems by using easy-to-generate rectangular grids in which geometric features are smoothly embedded using the FAVOR™ (fractional area/volume) method. A simple and powerful solids modeler is packaged with FLOW-3D or users may import geometric data from a CAD program.
An accurate treatment of fluid momentum is important for several reasons. First, it is the only way to predict how fluid will flow through complicated geometry. Second, the dynamic forces (i.e., pressures) exerted by the fluid can only be computed from momentum considerations. Finally, to compute the convective transport of thermal energy, it is necessary to have an accurate picture of how individual fluid particles move in relation to other fluid particles and confining boundaries. This implies an accurate treatment of momentum. Simplified flow models that only crudely approximate the conservation of momentum are not used in FLOW-3D because they cannot be used to predict realistic fluid configurations and temperature distributions.
Heat transfer between a liquid and a solid (e.g., metal-to-mold) requires an accurate estimate of the interfacial area. Stair-step boundaries over-estimate this area; for example, the surface area of a cylinder would be over-estimated by a factor of 27%. Accurate interfacial areas are automatically computed by the FAVOR™ method for each control volume in the FLOW-3D pre-processor.
The size of control volumes can influence the rate and amount of heat exchanged between a liquid and solid because heat must also flow in the control volumes containing the liquid/solid interface. In FLOW-3D control volume sizes and their conductivities are accounted for when computing heat transfer rates across liquid-solid interfaces.
Implicit methods for nonlinear and coupled equations require iterative solution methods that have the character of an under-relaxation in each iteration. This behavior can cause significant errors (or very slow convergence) in some situations, for example, when using control volumes with large aspect ratios or when the implicitness is used in anticipation of an effect that is not actually significant. In FLOW-3D explicit numerical methods are used whenever possible because they require less computational effort, and their numerical stability requirements are equivalent to accuracy requirements.
Implicit Versus Explicit Numerical Methods
Implicit numerical techniques that allow arbitrarily large time-step sizes to be used in calculations are a popular way to reduce CPU time requirements. Unfortunately, these methods are not accurate for convective processes. Implicit methods gain their time-step independence by introducing diffusive effects into the approximating equations. The addition of numerical diffusion to physical diffusion, e.g., to heat conduction, may not cause a serious problem as it only modifies the diffusion rate. However, adding numerical diffusion to convective processes completely changes the character of the physical phenomena being modeled. In FLOW-3D time steps are automatically controlled by the program to ensure time-accurate approximations.
Numerical methods that use implicit approximations also require the selection of one or more convergence and relaxation parameters. Making poor choices for these parameters can lead to either divergences or slow convergence rates. Only one convergence and one relaxation parameter are used in FLOW-3D, and both parameters are dynamically selected by the program. Users are not required to set any parameters controlling the numerical solver.
There are two methods used to model liquid-gas interfaces (i.e., free surfaces). One of these is to compute flow in both the liquid and gas regions and to treat the interface as a sharp change in fluid density. Typically, the density discontinuity is modeled using higher-order numerical approximations. Unfortunately, this treatment allows the interface to smooth out over a few grid cells and does not account for a corresponding sharp change in tangential flow velocity that generally exists at such interfaces. This technique must also be supplemented with escape ports or sinks for the gas if it is to be replaced by liquid entering a computational region. Further, such methods must typically work harder to satisfy the incompressibility of the fluids. This happens because gas regions must have nearly uniform pressure adjustments which tend to slow down the solution convergence rate. A different technique, the Volume-of-Fluid (VOF) method, is used in FLOW-3D. This is a true three-dimensional interface tracking scheme in which the interface is closely maintained as a step discontinuity. Moreover, normal and tangential stress boundary conditions, including optional surface tension forces, are applied at the interface. Gas regions are not computed unless the user requests these regions to be included in the model.