[yansheng]

Dear expert,

I am doing a RANS simulation modelling turbulent channel flow using STAR-CCM+. The desired velocity field in the streamwise direction should be like figure 1 within the channel. However, my results indicated a totally different velocity field as shown in figure 2.

The interesting part is that the skin friction of my own simulation has relatively good result compared to empirical formula for skin friction and ratio of centreline velocity to mean velocity separately. Actually the Reynolds number of my simulation is 5600 based on the Reynolds equation: Re=2δ*ū/ν, where δ is half channel width, ū is bulk velocity, and ν is kinematic viscosity. Normally, for a channel flow, if the Reynolds number is bigger than 1800, it will be fully turbulent. Moreover, to make sure the flow is fully turbulent, I apply the boundary conditions with the type of fully developed for the inlet and outlet boundaries. The grid size is Δx+=18, Δz+=2.8, Δymin=2.4 and Δymax =10. But from the scalar scene of velocity field, it doesn’t seem to be a good representation of turbulent channel.

The desired turbulent channel flow is like that shown in the following video link:
https://www.youtube.com/watch?v=ezp7QVUBYSg
I am really grateful if any advice could be provided to get the desired velocity field.

Best regards,

Yansheng

[yansheng]

The relevant figures are attached as below

[fluid23]

Image 1 is an LES solution and you stated that you have a RANS setup... My guess is that you don't know what either one of these imply and that's why you are getting the 'wrong' answer.

Essentially RANS is a time-averaged solution so you won't see an instantaneous snap shot of the flow field like you show in Image 1, you will get a smooth uniform field that is the average each point sees over time. Also, typically the RANS turbulence models use a characteristic length scale that does not capture all the effects of turbuelnce on your flow.

A Large Eddy Simulation (LES) is not time averaged and is more akin to direct numerical simulation (DNS), only instead of resolving all turbulent length scales you only resolve the large turbulent eddies which are influenced by model geometry and you account for small intrinsic eddies through a sub-grid scale model.

[yansheng]

Thanks a lot for your reply. I totally agree that the RANS is totally a modelling method for turbulent flow. However, only sub-grid scale part is modeled in LES while large scale motions are resolved.

Actually, I did carry out some LES simulations myself using STAR-CCM+, and they present some kind of similar velocity field as shown above in figure 2. Since the skin friction I got from my own LES case is far smaller than that from empirical formula, I did not say that I did LES.

Apart from that video link I attached below, there are also some videos that shows the transition to fully turbulent status. Since my velocity field is same as the initial status of the channel flow in the video below, I was really confused how it could be possible to get that kind of fully turbulent status. Actually, I also read some articles (The potential of Large Eddy Simulation for the modelling of wall bounded flows - by Eugene de Villiers) which indicate that a initial perturbation should be added to induce a fully turbulent status.

The video link is:
https://www.youtube.com/watch?v=lAt1DF-esDI

Since the velocity field in my case is not as typical as that in figure 1, I was really worried if I correctly modeled the turbulent flow in channel flow by RANS even though the skin friction and other results fit well with empirical formula.

I will really appreciate it if any suggestion or advice could be offered, and if there is anything I didn't state clearly. Please tell me.

Best regards,

Yansheng

[fluid23]

"I totally agree that the RANS is totally a modelling method for turbulent flow. However, only sub-grid scale part is modeled in LES while large scale motions are resolved. " - I believe that is what I said as well.

What exactly is your problem, it's not very clear?Skin friction is too low for LES but reasonable for RANS? However, LES solution results in similar velocity field to the RANS solution? Am I reading that correctly?

[yansheng]

Dear Matt,

Thanks a lot for your reply, the problem that exists in my RANS case is that I am not even sure if the simulation is turbulent or not. From the result of skin friction, centerline velocity to bulk velocity ratio, it fits well with the work done by Kim, Moin, and Moser in the paper - “Turbulence statistics in fully developed channel flow at low Reynolds number”.

However, from the velocity field as attached above, it looks like laminar flow since it looks so ordered. Also, I try to post-process my data and get the turbulent intensity around 2.5E-6, which is far smaller than the desired value 0.054 derived from empirical formula 0.16Re^(-1/8), where Re is half channel width based Reynolds number and the formula is for fully developed pipe flow. Also the velocity fluctuation I got from post-processing is as small as 1.0E-6 - 1.0E-7, which makes me feel like that the flow I modeled is actually not turbulent when the velocity fluctuation is so small.

I hope I have explained my problem well and if any suggestion could be provided, I will really appreciate it. And I am sorry for any point that is poorly explained.

Best regards,

Yansheng

[fluid23]

Can you show a screenshot of your physics model selections? I am willing to bet that you have a turbulent model, but your settings are probably off somehow.

What are your turbulence settings at the inlet and for the domain? Are these default or did you take them from the paper you cite?

As for your comments about the velocity field, a turbulent RANS analysis will result in a structured, laminar looking flow like you have shown. Remember, this is a time averaged analysis. So the local instant velocity can rise and drop as eddies pass through a given point but as t approaches infinity it will time averaged velocity will be a structured profile like you are seeing. If you were to do a LES analysis of this and run it for long enough and sampling the velocity profile at each (properly sized) time step then you would end up with roughly the same profile you get from a converged RANS, all things being equal.

[yansheng]

Dear Matt,

Thanks a lot for your reply. For the inlet boundary, I set the type to velocity inlet, and velocity magnitude to 2.8 m/s. But I create a interface between inlet and outlet which is fully developed periodic. The periodic boundary conditions is used for almost every single paper I have read modelling turbulent channel flow. I plan to use this periodic boundary condition to make my simulation fully turbulent without using an infinite long channel.

Just to make sure that I fully understand what you mean in LES analysis of this problem. Is it correct that the result of a converged LES will give the similar structured, laminar looking flow like what I have got in RANS case.

Best regards,

Yansheng

[yansheng]

The screenshot of physics model is attached as below.

[fluid23]

I am more concerned with your turbulence settings for the inlet and the domain... if these are left as default values then this is likely why you do not match some of the turbulence related parameters. You should try to duplicate the setup in the other paper if you are trying to match it (whether this is a model or experiment)... you should always specify the correct turbulence parameters at the inlet if they are known.

No... a converged LES will give you an instant snap shot for one time step, then you move to the next step and converge it, and so on.... The RANS analysis is what you would get if you added up all the snap shots and divided by the number of time steps.

[yansheng]

Dear Matt,

Thanks a lot for your reply. I will try to match the turbulent parameter with what I could get or calculate from the paper like turbulent intensity, turbulent kinematic energy and so on. In addition, I was curious about how to obtain that kind of velocity field which looks fully turbulent as shown in Figure 1 simulated by LES. I did read some papers which also has such kind of figures, some of them managed it with initial perturbation ("The potential of large eddy simulations for the modeling of wall bounded flows" by Eugene de Villiers) and some of them managed it by modeling the transition from laminar to turbulent (http://engineering.jhu.edu/zaki/rese...ss-transition/). I am grateful if you could provide any feasible ways to obtain such figures.

Best regards,

Yansheng

[harry123]

You will have to do LES instead of RANS if you need to get turbulent structures. I guess this was made clear by Matt in his first reply.