Fluids

Fluids

A Pragmatic study: turbulent flow over BFS in FLUENT

    • José Mantovani
      Subscriber

       


      Hello everyone!


      So as you could saw, I dedicated my time in order to understand the flow over BFS which have four interesting behavior about a turbulent flow over solid wall: separation, recirculation, reattachement and recovery. I modeled this flow first like the Jovic and Driver (1994) experiment, and shared here my doubts and results through this thread in the link (https://forum.ansys.com/forums/topic/understanding-the-behavior-of-the-solution-results-by-fluent/). 


      An overview of my results for this approach was: I got the overall behavior of this flow, without getting unwanted non-physical results. The question of obtaining results that are equal to or close to that of the experiment, with validation effect, requires some techniques such as non-uniform velocity profile and detailed conditions equivalent to those present in the experiment. In the available paper of this experimental work JD (1994) has all relevant information for such, but I found in a NASA validation site, more information of easy access (in txt file already ready to plot) without the need of renormalization or things of the type, for the BFS stream studied by Driver and Seegmiller. So I decided to model this flow over BFS of this experiment of Driver and Seegmiller (1985), which differs in some things like Reynolds number and ratio of E.R. channel expansion in relation to the Jovic and Driver experiment. 


      As I had already studied using Jovic and Driver incessantly, I started to do a study of the development of velocity profiles in channels by FLUENT, I already had the knowledge that to obtain a developed velocity profile an input channel of minimum 40 times the channel inlet diameter. So I created the BFS computational domain (according with Driver and Seegmiller tunnel geometry) along with an extensive inlet channel, to simplify the "job" of simulating a channel twice and to remove the profiles to use as inlet to a BFS domain with reduced inlet channel, I even tried this as discussed in this thread (https://forum.ansys.com/forums/topic/segmentation-default-error-what-is-this/), but due to a divergence of locale I decided to test everything in a single domain. 


      I created a higher density mesh near the wall in order to solve fine gradients, basing myself on the value of y + for first cells through a calculator on the internet and obtained a mesh with values of y+ = 0.765 for the first cell adjacent to the wall, which was later confined by the Wall Yplus contour in FLUENT. My mesh was dense just in the y direction. In the x direction there was no locality with greater density than another. The mesh created was orthogonal structured.


      Due once again, because I had tested the BFS flow several times for the Jovic and Driver experiment, I knew that the sensitivity of defining ke epislon or omega was better in order to improve the convergence (agility and ease) of the which define intensity and viscosity ratio. So based on the FLUENT theory guide, I calculated values for k and epsilon or omega (since I would test epsilon and omega models) based on the input conditions. Then in the input I used the values k and epsilon or omega, and in the output I used the scale values of turbulent length and intensity. This time I obtained results closer to the experimental one because of the input channel to develop the speed profile and to obtain a better mesh in relation to the previous approach. This can be seen in the agility and ease with which FLUENT achieved convergence, without oscillations as before and in fact achieving convergence below 10 ^ -7 for all residues, which before some reached a maximum of 10 ^ -5 and continuity 10 ^ -3.


      The big question was, I for the time, did not extract the values and compared in excel with the experiment, but I could see that they are close, and what I found "funny" was that in the recirculation region FLUENT I calculated only a large bubble and not a large bubble preceded by a small one as before in approaching me based on the study of JD. The graph Cf that has a double change of signal indicating this, started to have a "quadratic" behavior in the region of the smaller bubble, not demonstrating its curve.


      From the vectors it was possible to see the absence of the primary bubble, only the secondary bubble configuring a single bubble in the recirculation region. As in the image below. Together in the image has the graph of Cp that is very close to the experimental, the Cf with this behavior that I described, both for the k-omega model. I tested std and SST k-omega and std k-epsilon, all presented similar results.



       


      I do not have the Driver and Seegmiller paper and I do not know if for the configuration of his experiment, the recirculation region has two bubbles, I think so. I think that maybe the mesh in the region of the wall of the step (vertical wall) may be favoring this result, which at the same time I think not because before in the approach of Jovic and Driver the two bubbles were captured and the graphs of Cf showed the double change and different behaviors for different models of turbulence. 


      I think the problem may be what is seen in the contour of Wall Yplus, maybe high wall values can generate these results. I have not yet tested calculate a profile on an input channel and set it as input in a reduced BFS domain. And I have not yet tested a mesh with a higher density in the x direction near the step. 


      Sorry for the long thread, but for understanding I should explain where I came from and where I am. Thank you immensely for those who devote themselves to reading and helping. Thank you!

    • José Mantovani
      Subscriber

      Someone can help me? With some tips, comments?


      Thanks.


      Mantovani. 

    • DrAmine
      Ansys Employee

      Long thread but what are your questions now? In Germany we say something like "Brevity is the soul of wit". Perhaps you can add below your long introduction precise questions so can community members might chime in easily. 

    • José Mantovani
      Subscriber

      So, I'm dumb Amine? hahahah


      The question is around the Cf chart and recirculation zone. As I said in thread I tested several ways to get numerical results close to experimental. Now I get close results, but in recirculation zone my solution computate just one bubble in the recirculation zone and through the Cf chart we can see that no have two changes of sign, what indicates one more time just one bubble. So, Why?


      Sorry for the long thread. I thought if I fully explain my approach could facilitate in understanding. Thanks for reading and replying, thank you for your patience.


      Thanks one more time,


      Mantovani. 

    • Karthik R
      Administrator

      Hello Mantovani,


      Just to clarify, did you solve the model using fully developed turbulent profiles as input conditions (vel. inlet BC)? I am pretty sure you must have double-checked all your input conditions to your model? Also, how big are the two recirculating regions in the paper? What is your mesh resolution like? Can you plot the y+ values from your simulation?


      Thank you.


      Best Regards,


      Karthik

    • José Mantovani
      Subscriber

      Hello my dear Kremella. Thanks for reply.


      So, I dont use a fully developed, but I create the domain with the inlet channel of 40D (D is the length of inlet) in order to obtain a fully developed profile. To set my conditions I used a "Turbulence Calculator" in web and I got this values in image below. I used k and omega as inlet turbulence parameters and turb. intensity and turbulence length as outlet conditions. The recirculation zone length measured by experiment is around Xr/h= 6.26 +- 0.1. 




      About the Wall Yplus Kremella, have in a image in the thread post. The value for cell adjacent to the wall is around y+ = 0.765. I'm guessing I'm having this result without the initial bubble because of the length of the cells from the step. Maybe they have to be half that size. I'll double the number of cells in the x-direction and see what happens.


      Thanks.


      Mantovani. 

    • Karthik R
      Administrator

      Hello,


      That is the next thing I was going to ask you - mesh density in the streamwise direction. Please let us know what you find when you add more cells near the step in the flow direction. You might want to make use of the bias factor in meshing to reduce the overall cell count.


      Best,


      Karthik

    • José Mantovani
      Subscriber

      Hi Karthik, thanks to reply.


      As you can see, I got a satisfatory reattach length and a good velocity profile at station x/H= -4. I will try to make this, soon I share here the results. 


      Thank you one more time.


      Mantovani. 

Viewing 7 reply threads
  • You must be logged in to reply to this topic.