<p>Does the accuracy then come down to people meshing everything an an optimal fashion? I have personally never used any commercial software so I don’t really have any real experiences with it being inaccurate or not.</p>
<p>Also clandry, you can do computational solid mechanics, as an example.</p>
<p>clandry:
If it is governed by an equation, there is probably a computational field that models it.</p>
<p>aerokid1491:
If you are doing a DNS, then yes, it practically all comes down to the mesh and settings for things like how aggressive the solver is at seeking a solution (e.g. iteration size). For any other CFD technique, it has everything to do with the assumptions inherent in the code as well. No code save a DNS can fully solve a turbulent flow, for example, so they all have models of varying degrees of accuracy and complexity to describe turbulent flows. These models are accurate in some regards and inaccurate in others. The user needs to know what model he or she is using so that they are aware of which quantities in a turbulent flow are reasonable and which are going to be inherently poorly handled.</p>
<p>There are other examples of course, but the important thing is that most codes seem take one of two approaches. They either strive to accurately capture the physics in a particularly type of flow and don’t bother with others and excel in their one area or they take the general route and try to be pretty good at a lot but have trouble with anything too specialized, even in the regimes they presumably cover. The latter seems to be the route most commercial packages take. As long as the user understands how the commercial code works and can adequately set up the problem (including proper BCs, proper meshes, proper solver settings and a problem that is well-suited to that solver) and goes to the trouble of actually obtaining an iteratively converged solution, then even the general solvers can get good results, particularly from a design engineering standpoint.</p>
<p>The problem is that they often aren’t the most physically accurate, even if they get the basic properties like lift and drag correct. The nature of these codes means that lots of physics tends to be averaged out (e.g. RANS) in order to save time and the averaged out physics get modeled using mostly empirical data. This is great for getting certain quantities of interest, but never works for all of them at once. This simply isn’t suite for research, which primarily focuses on either researching the physics or developing better models, both of which pretty much dictate that the commercial codes not be used.</p>
<p>Cool.
Is CFD considered a “subfield” of fluid mechanics?</p>
<p>Not really. It’s just one way to attack a problem, along with experiments and theory. Subfields would be like turbulence modeling, boundary layer stability, flow control, or turbomachinery, to name a few.</p>
<p>Theory? Can you elaborate? I thought CFD was considered a theory as a purely mathematical method and with physics of testing it combines in two ways of solving problems. You seem to insinuate there are three methods. Thanks.</p>
<p>There are three methods. There are people out there who just work through theoretical models and solutions for fluid mechanics. Then, there are those out there who just do real experiments to try and draw conclusions about certain aspects of fluid mechanics. Then CFD combines the two and uses theoretical models to do experiments on a computer that otherwise would be near impossible or too costly to do in real life and draws conclusions from the resulting simulation. </p>
<p>These are all three ways to attack fluid mechanics problems and all have their advantages and disadvantages.</p>
<p>I’ve always thought CFD was a subfield.
Would it sound stupid if you told someone that “I like fluids mechanics, especially the CFD subfield?”</p>
<p>Possibly if the person was well educated in what CFD is. I would just rephrase that and make sure you just say “I like fluid mechanics, especially CFD”. No need to get too specific, especially if it isn’t right.</p>
<p>This is probably a dumb question, but is there an experimental side to CFD?
I know it’s ‘computational’ fluid dynamics, but is it ‘all’ computational?</p>
<p>It can be experimental if you are simulating things we haven’t/can’t test in the real world right now. The professor I am working for is simulating the human voice so we can start understanding how our vocal chords work to generate our voice. It is totally experimental and will shed light on how the vocal chords work since we don’t really know right now. It’s cool!</p>
<p>But to kind of answer one of your questions… It is all computational, hence it’s name, but it can be used for experiment simulations.</p>
<p>That sounds like a fun project!
Is experimental more hands on lab experience?</p>
<p>It depends on how you define “experiments” and “experimental”. To me, experiments are observations and analysis of the physics of the real world, so in that sense, CFD is never truly “experimental” since nearly all CFD make assumptions that satisfy the physics. Of course, if you are doing a DNS you are no longer assuming out portions of the physics, so you will sometimes see them referred to as numerical experiments. Even that is borderline because you are still discretizing space and time.</p>
<p>So again, it comes down to what you mean right now by “experimental”. You can certainly still learn new things from CFD simulations, but it is never experimental strictly speaking. Experimental work is more along the lines of taking measurements in a wind tunnel.</p>
<p>Oh I see. I don’t have a clear cut definition for those terms. </p>
<p>Is there any hands-on work in CFD research?</p>
<p>Define “hands on”. If you mean in the sense of turning nuts and bolts, then no. Do experimental work for that since there is a whole lot of that that goes into running, maintaining and upgrading wind tunnels. With CFD you are going to be working on a computer.</p>
<p>For CFD, the experiments are virtually done on the computer, as boneh3ad said. However, it seems to be common for there to be an actual experimental group that does real tests to compare it to numerical results from CFD and other computational experiments, but the experimental group and the CFD/computational group still do way different things.</p>