When Good CFD Goes Bad

You've checked everything, and I mean everything, but still your Computational Fluid Dynamics (CFD) simulation does not resemble reality. Then it's time to consider the deeper underlying basis of your CFD software.

Mandelbrot Set From Chaos TheoryMandelbrot Set From Chaos TheoryCFD Simulations are not immune from descending into chaos
License: CC BY-SA 3.0, Wolfgang Beyer

Check the Obvious

Your geometry doesn't have any inconsistencies or slivers. Your mesh metrics look nominal (NASA speak for in bounds) and you have appropriate sized elements throughout your flow domain. You've checked and rechecked your boundary conditions and they are all good. The solver settings all seem fine and have worked well in previous simulations. It doesn't matter whether you try a transient or steady setup, neither helps. Laminar or turbulent, it still doesn't want to cooperate. This is a good moment to step away and think about the underlying assumptions and models that your CFD software uses.

Breaking the Continuum Assumption

Most industrial CFD is based on a continuum assumption, whereby the fluid in each element is squishy and irreducible. If your CFD simulation doesn't conform to this assumption then you will be in trouble.

Microfluidics is one such example, where the flow is typically laminar and slow. At this extreme the fluid behaves more like a plastically deforming solid (barely squishy) and a more appropriate (and significantly faster) simulation strategy could be to use something similar to non-linear stress analysis software.

At the other extreme to microfluidics is a very low density gas (e.g., interstellar gas) where molecules are dispersed and rarely interact. This breaks the continuum assumption (molecules are assumed to be relatively tightly packed) and will cause an industrial CFD solver major issues.

Missing Models

In between the two extreme cases where the continuum assumption clearly breaks down are shades of gray where the appropriate models might be missing. For example, if you want to simulate cavitation with industrial CFD, then your software will have to explicitly have a cavitation model. Without the appropriate model the CFD simulation will often converge well, but the results will be unrealistic.

Alternate Models

The final level of unrealistic behavior you might see is that you have a model (e.g., Moving Reference Frame) that will invariably make assumptions, but for your particular case the assumptions are borderline satisfied. It might be that a different model is more appropriate (e.g., moving mesh), but then you have to use good judgement as to whether the added complexity and increase in turnaround time are worth the increase in accuracy.