Overview of CFD Techniques

1.0 Introduction

Is there anything that computational fluid dynamics (CFD) can’t do?  At DMS, clients ask variations on this question when discussing new projects.  Theoretically, CFD remains capable of nearly any fluid dynamics problem.  Practically speaking . . . it can’t work for free.  And several categories of CFD problems become prohibitively expensive.  We can achieve the desired result, but you may regret paying for the answer.

The rates for CFD costs get driven not by the problem to solve, but by the CFD tools required for the solution.  Don’t get caught unprepared.  Review the major CFD methodologies and catch the warnings to know when combinations may lead to cost overrun.

2.0 Hierarchy of Methodology

Figure 1‑1 lists the hierarchy of CFD methods.  These generally increase your project costs as you select from lower down in the hierarchy.  Selecting combinations from this hierarchy really compounds the costs.  The trick requires understanding typical expectations for normal CFD methods and which tasks compound to increase costs.  In some cases, careful selection of combinations can actually decrease costs.

Each of the following sections lists the typical methods available in each category, with the common default option highlighted in bold.

Hierarchy of CFD Methods

Figure 1-1: Hierarchy of CFD Methods

 

3.0 Mathematics

  • Boundary element method
  • Finite volume method

The normal options for mathematics are the finite volume method.  This is not an especial burden on cost or schedule. 

The alternative of the boundary element method reduces run times down to 1% or less, compared to the run times of the finite volume method.  But you sacrifice quality for boundary element method.  Consider that for early stage design when you need to iterate through thousands of design alternatives.

4.0 Dimension

  • 2D
  • 3D

We exist in a 3D world.  For all real world problems, the normal is 3D.  The cost depends on the size of your problem, and the detail required.  For example, flow around a 300 m long ship may be a large size, but require relatively coarse details.  In terms of CFD effort, that could be equivalent to flow from an inkjet printer with tiny details. 

When modeling tiny details or trying to validate specific flow physics, consider 2D space as an alternative.  Many structures can be simplified down to 2D dimensions without losing the essential flow physics. (Figure 1‑2)  This greatly reduces costs and allows you to focus more budget on the questions that you set out to answer.  You can explore more design variations, more optimization, better performance, or just save some budget for later in the design.

Example of 2D CFD Flow NACA Foil Section [1]

Figure 1-2: Example of 2D CFD Flow
NACA Foil Section [1]

 

5.0 Time Domain

  • Steady
  • Unsteady

 

Picture of pocket watch

Figure 1-3: How To Measure Time [2]

The time domain captures variation in CFD flow, changes in the input conditions.  This may be necessary for items such as a rotating propeller, a column of water collapsing, or bubbles rising to the surface.  All these changes require unsteady flow in the time domain.  That implies larger, more expensive runs. 

CFD engineers prefer to stay with steady flow if possible.  It keeps budgets down and maintains high simulation quality.  To that end, they employ many alternatives to recast an unsteady problem into a steady-state CFD simulation.  Ask you CFD engineer about alternatives when looking at modeling in the time domain.

 

 

6.0 Turbulence

  • Laminar
  • RANS
  • LES / DES
  • DNS

Turbulence may be the largest technical question for each CFD engineer.  Each category of turbulence employs several methods and alternatives.  You engineer spends a fair amount of time selecting the best turbulence model for your case.  The normal approach are RANS turbulence models, and they work for 85% – 90% of all marine CFD applications. 

Laminar turbulence models are specialized and only work for extremely low speed applications.  They imply a complete lack of turbulence and require even less effort than RANS models.

LES / DES methods take an exponential jump in effort.  Theoretically more accurate than RANS methods, but also easier to get them wrong.  Do not switch to LES / DES methods just because you liked the sales pitch.  Take a critical assessment to ensure you absolutely need them for accuracy.

Graph showing different scales of eddies in large eddy simulation models

Figure 1-4: A View of Some Theory Behind LES / DES Methods [3]

 

DNS methods are flat out not possible with modern computers (circa 2019).  If a CFD engineer promises DNS turbulence models, throw them out of your office.

7.0 Motion / Mesh Deformation

  • No motions
  • Prescribed motion
  • DFBI (dynamic fluid body interaction)

In the category of motion, we move the object in the simulation, and this requires us to shift and deform the mesh around the object.  The mesh was fundamental to the simulation, and now we need to alter that mesh while using it to perform calculations.  The quality of our calculations hinges on the shape of the cells in our mesh, and those cells now change with each update to the calculations.  You see the resulting potential for complication. 

Normally, we prefer to avoid this complication and create a simulation with no motion at all.  The next step up involves prescribed motion.  Imagine a piston in a cylinder.  Some type of mechanical motion independent of the fluid flow conditions.  The ultimate challenge invokes dynamic fluid body interaction (DFBI). 

DFBI gets employed for rigid body motion; imagine a ship turning in a channel.  In DFBI, the CFD solver calculates the fluid forces on your object, the body.  Those forces drive the motion of that body.  The motions of the body then feed back and influence the flow patterns of the fluid.  If unchecked, this feedback loop goes haywire.  Done correctly, it leads to almost autonomous predictions of maneuvering and motion control.  DFBI is definitely advanced fluid modeling.  It requires loops within loops within loops for the CFD solver.  Major potential for cost overrun and project risk.

8.0 Conclusion

Taken individually, each CFD method becomes an important decision.  Combined together, they evolve into a major project risk.  Table 7‑1 compares the various CFD methods.  The blue bar in the middle shows the combination for a normal CFD project with reasonable cost expectation.  The rest of the table shows alternatives from that normal CFD project, with red text highlighting the relative increase or decrease in project budget for each change.

The tricky part comes from combining changes in multiple categories.  Consult your CFD engineer when considering a complex CFD project.  They will estimate the relative risk, and suggest alternatives.  CFD engineers do more than just run software.  We understand your project and plan a strategy to minimize project risks so that you don’t get caught by combining unknown cost increases.

 

Comparison of CFD Methods

Table 7 1: Comparison of CFD Methods

 

 

9.0 References

[1]

Michael Belisle, “Streamlines Around a NACA 0012,” Wikimedia Commonds, 23 April 2008. [Online]. Available: https://commons.wikimedia.org/wiki/File:Streamlines_around_a_NACA_0012.svg. [Accessed 31 12 2018].

[2]

C. Walkins, “Pocket Watch with Chain,” Wikimedia Commons, 6 Oct 2005. [Online]. Available: https://commons.wikimedia.org/wiki/File:Pocket_watch_with_chain.jpg. [Accessed 31 Dec 2018].

[3]

Youtube Author: alpha754293, “Cylinder Valve IC Engine CFD Test,” YouTube, 3 Dec 2009. [Online]. Available: https://www.youtube.com/watch?v=2Ql8IPG6iTY. [Accessed 31 Dec 2018].

[4]

Nick Jenkins, “Turek Fluid-Structure Interaction Benchmark,” YouTube, 28 Oct 2011. [Online]. Available: https://www.youtube.com/watch?v=-lDbseYB7Sc. [Accessed 31 Dec 2018].