Howdy Jasir Jawad,

Is the water vapor condensing in the cylinder during the simulation?

Happy Trails, Len

That's a very common problem and is covered in a first course on CFD. I'm pretty sure Versteeg and Malalasekra do cover it in their text book.

The general solution is to increase the length of your cylinder because the flow must be "developed" physically to be correctly simulated numerically.

As you know, for a regular water flow in a pipe, the developing length is 20 diameters. When I did my simulations in ANSYS FLUENT for a water flow with bubbles in a cylindrical pipe I had to go to, at least, 22 diameters to get rid of the "reverse flow at the outlet" issue.

With sonic flows of gases it is more difficult to pick the right length of the pipe because we don't have a good rule of thumb for the developing region as well as we don't, normally, define developing regions in sonic flows (as far as I know). But the general idea of increasing the computational domain is still valid. You just have to pick the right length empirically.

I suspect the "right length" will be way bigger than 20 diameters. Thus, you are stuck with a lot of cells and long computational times. But that is a general problem with sonic flows as far as I am concerned.

Also, do your problem in "baby steps". Start with a small difference in pressures, because CFD can "freak out" when dealing with high pressure differences. Then pick a proper cylinder length and cell size. Pay attention that in transient turbulent simulations the limiting factor is k and epsilon convergence (if you're using k-epsilon turbulent model). And - it turns out - the most important thing that affects k and epsilon convergence is the time step, not the cell size. Also, pay big attention to the cell size near the wall. You must achieve y+ = 5 condition (if I remember this condition correctly). Once you're done with this baby step, increase the pressure difference slightly and repeat the procedure. Keep iterating till you get to your desired pressure difference.

P.S. I'm talking about sonic flows because you said you have compressible flow and we, normally, define comprehensibility of the flows via Mach number.

Howdy Folks,

Old guy observations are coming up here, but you are using computers to simulate nature and the reverse flow during sonic simulations seems to be well known. It has been half-a-century since I was involved in tuning two-stroke exhaust systems to enhance performance and four decades since I wrote a program for non-linear thermally driven flow in the mesosphere and also reviewed soil moisture modeling computer programs. So, I am not well versed in current CFD, but I'm only out-of-date, not out-of-thought. By the way, I really appreciate learning about these issues in sonic flow CFD, and offer a couple observations that may be helpful.

Firstly, new may not be new: back in '83 a colleague was quite excited about a friend's discovery that harmonic analysis could be used for simulation of soil temperature profiles and general treatment of the energy budget at the surface. I pointed out that he was right and the papers on the topic published in '49 were well worth reading. Fun observation, but not irrelevant.

Secondly, and more significantly: The gradients of soil moisture and diffusivity across a wetting front are extreme and create a serious problem for computer simulations. As one successful method, algorithms using implicit models instead of explicit models turned out to be very stable even for predictions of wetting front behavior. That method was used in the upper atmosphere successfully as well. If Fluent, or some other CFD package, has an implicit programing option, I suggest it be considered. It is not a valid claim that one must use explicit methods in such cases.

And be very careful with a vapor in case that has enhanced what turns out to be a computational stability problem in the general case.

Happy Trails, Len

Howdy Dr. Hall,

Thank you for your insights. I am simulating a piston cylinder compression of water vapor with injection technology. In my case, it's possible that certain conditions may occur that leads to vapor condensing, however it's not favorable (also I am not using any 2 phase models) . At this point in my simulation, I am really just trying to flow pressurized vapor from the cylinder to the outlet.

I was able to fix this issues by making a smaller outlet rather than considering the whole face of the cylinder as an outlet.

Jasir

Hi Mr. Ivan,

Thank you for your response. Because of the dimensions of the piston cylinder system that I am simulating, I cannot place the outlet further downstream. Therefore, I managed to make a small opening on the face of the cylinder outlet and set that as the new outlet. Thanks again for your detailed answer.

Howdy Jasir Jawad and Ivan Nepomnyashchikh,

It is good to read that you have solved the computational issue with a smaller opening, Jasir Jawad, just as it was good to read your excellent discussion of programming modifications that address the reverse flow calculation, Ivan Nepomnyashchikh.

Actually, I realized in the night that the reverse flow may be physical given the abrupt start of the simulation with a pressure difference across the end of an open tube. Your simulation starts with a plane pressure surface as real as an acoustic/sonic plane wave that has propagated to that surface. It is reflected in the simulation at the tube open end as well as propagated away from the tube, and the simulation calculates the streaming motion of the reflection back into the tube!

Hmm, let's check this out. Physics without calculus back in '59; professor asks how reflection occurs at the open end of a tube; I suggest that the impedance change from a plane to a spherical wave is the cause of the reflection; he was surprised and delighted a student would suggest that. Trying to understand my trombone had helped, but the answer came in response to the question. Now, this very significant physical condition may be addressed by changing the strength of the reflection or by modifying the algorithm so the computer is able to track the very strong event, i.e., make it a weak event some way or other.

It turns out if this view is valid, the reverse flow is not a problem, it is a revelation of the richness of the case you are treating - "more things in Heaven and Earth . . ." idea. Well, it's a thought,

Happy Trails, Len

Hi Dr. Hall,

Absolutely love the way you discussed the problem and it's physical significance. I will definitely take the pointers into account going forward. Thanks again.

Howdy Jasir Jawad,

It is quite kind of you to be so affirming, thank you! If you like this awareness, you might enjoy the Lettau Effect colloquium as illustration of a person seeking to explain.

Happy Trails, Len

Howdy Folks,

Yes! It is a physical response to find reverse flow into the open end of the cylinder during a transient adjustment to start-up of your simulation.

"It is easy to see" (eventually) that the physical response of streaming is produced by the affect of a spherical sonic wave propagating away into the ambient pressure field at start-up, leaving a rarefaction of nearly equal pressure deficit in the end of the cylinder. In response to this pressure gradient, radial flow toward the cylinder end would produce the "reverse flow" into the cylinder as a natural response.

Actually, it would be a trivial problem for the hybrid analog-digital computer of molecular unrest to accomplish this result in an actual case since it is calculating at about 1035 operations per second per cubic meter. No weakening of the event would be necessary. Even modern computers are ambling along like lumber wagons by comparison.

Anyway, this is no longer a speculation/possibility for me, it has become a firm result with this more complete analysis. Thank you for bringing it up, this stuff is great fun.

Happy Trails, Len