Turbulent flow convergences in a pipe / High Re-Number

classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|

Turbulent flow convergences in a pipe / High Re-Number

Leon Robers
Dear all

I set up a 2D simulation which models the flow behavior of air in a pipe of apprx. 0.005m diameter. With a Reynoldsnumber of ca. 1000 (inlet velocity of 3m/s) and a fine grid, it converges.

When I go up to a higher Reynold number of 3000 (10m/s) or also 1500 (5 m/s inflow velocity), the simulation doesn't converge anymore. I know that the critical reynolds number for pipes is around 2000 and the flow becomes turbulent, but shouldn't FEATool be able to handle turbulent flows?

So how can I set up a stable simulation for flows with Re>2000?

Thanks!
Reply | Threaded
Open this post in threaded view
|

Re: Turbulent flow convergences in a pipe / High Re-Number

Andrew
Have you tried a finer mesh?  It could be that with the higher flow rate, the courant number is too high, that is, the element of fluid does not sit in each mesh volume long enough for the calculation to be meaningful.
Reply | Threaded
Open this post in threaded view
|

Re: Turbulent flow convergences in a pipe / High Re-Number

Precise Simulation
Administrator
This post was updated on .
In reply to this post by Leon Robers
Few if any simulation codes today can manage to solve turbulent flows without including any form of turbulence modeling. A turbulence model essentially introduces an additional (turbulent) viscosity in regions where the grid is unable to resolve the small scale turbulent fluctuations. FEATool does not currently include built-in turbulence models, but it is fairly straigt forward to manually add an algebraic mixing length model as in the attached example [1].

For more accurate and advanced turbulence models it is recommended to use the built-in support for the OpenFOAM CFD solver, as for example in the turbulent flow over a backwards facing step tutorial model.


[1] turbulent_channel_flow.fes: Simulation of stationary and incompressible turbulent flow between two flat parallel plates using an algebraic mixing length turbulence model. A fully developed turbulent velocity profile is expected to form at the outflow boundary.

Ref. Laufer, J. (1951) Investigation of turbulent flow in a two-dimensional channel. NACA Report 1053, NACA Technical Note 2123.
Reply | Threaded
Open this post in threaded view
|

Re: Turbulent flow convergences in a pipe / High Re-Number

Leon Robers
This post was updated on .
In reply to this post by Andrew
Yes, this works to some extend but at some point, the grid would have to be so fine that the calculation takes ages. (And since I am running the calculation in an optimization loop with changing designs, there is a limit to my accetable calculation time)
Nevertheless, thanks for your answer and calculating the courant number to check maximal grid size is also a valuable input

Later Edit: I just figured out that theoretically, a coarser mesh reduces the courant number and ensures convergences, not the other way around. This is strange, since the CFD calculation only converges with a rather fine mesh :O
Reply | Threaded
Open this post in threaded view
|

Re: Turbulent flow convergences in a pipe / High Re-Number

Leon Robers
In reply to this post by Precise Simulation
Thanks, using the Openfoam solver works!
Nevertheless, also for laminar flows it takes significantly longer to converge than the integrated FEATool solver (especially if one would choose the same stopping criteria). But I guess this is normal?

To simulate turbulent flow, can I confidently trust the calculated parameters for the k-omega/k-epsilon model? (I am not very experienced with turbulent flow at all)

Thanks again
Reply | Threaded
Open this post in threaded view
|

Re: Turbulent flow convergences in a pipe / High Re-Number

Precise Simulation
Administrator
Yes, the default OpenFOAM solvers all use (pseudo) time-stepping, even for steady-state simulations so they will require significantly more iterations and time to converge than a dedicated steady-state solver.

Unfortunately, turbulence modeling is a science all by itself and notoriously hard to completely automate. Turbulence models are often tuned to specific flow regimes and situations, such as internal/external/separated flows etc. and can be very sensitive to mesh quality. However, the employed k-epsilon/omega models are historically the most widely used and therefore also tested so using them you should be able to at least get a rough idea of the flow behaviour as a starting point.


Leon Robers wrote
Thanks, using the Openfoam solver works!
Nevertheless, also for laminar flows it takes significantly longer to converge than the integrated FEATool solver (especially if one would choose the same stopping criteria). But I guess this is normal?

To simulate turbulent flow, can I confidently trust the calculated parameters for the k-omega/k-epsilon model? (I am not very experienced with turbulent flow at all)