Hi all,

I'm a student specialized in CFD, and I often heard that a big problem for CFD convergence is low viscosity. It seems that programs which adds artificial (numerical or whatever) viscosity has less convergence problems.

In that case, we could simply increase the value of mu (viscosity) and increase another value, as incident velocity (V_inf) or model size (L wing depth for example) in order to keep the right Reynolds Number and converge easily.

Is this technique used in commercial codes, and if not, why ?

Thanks in advance !

LS

Since my first post, I realized that for big Reynolds numbers, the problem will remain numerically stiff. The problem does not come from viscosity but from the Re number.

But if you have comments or other ideas, please make me know.

The main problem of CFD code is not low viscosity but stability!

Numerical viscosity is a way to solve the problem (not changing the Re but adding other terms such as jameson's that stabilize the code).

Another way is to use upwind schemes that are "intrinsically stable" and overcome the probelm of adding viscosity.

Mar

Hi Mar, thank you for answering

"Numerical viscosity is a way to solve the problem"

Sure, but numerical viscosity makes a poor solution quality, isn't it ?

"Another way is to use upwind schemes that are intrinsically stable "

I don't know any upwind sheme applicable to finite volume methods on coupled approach. Dou you know one ?

Would you have some links to some more theory about these ways to stabilize (finite volume coupled) codes ?

I am speaking from my experience in Coupled solvers, Coupled solvers, eddy viscosity (initial turbulence) should be varied to of the order of 10 times of kniematic viscosity, rather than playing with the velocity. For Coupled Solvers I do not have Upwind Scheme available with me. I may be wrong. Please correct me. Cheers.

"eddy viscosity (initial turbulence) should be varied to of the order of 10 times of kniematic viscosity"

Does it mean that I should add a "viscous turbulence model" term to Navier-Stokes equations and set it to 10 times the real fluid viscosity to get stability (and a good precision of solution) ?

The solver I am using has the parameter Eddy Viscosity with unit m2/sec i.e. Eddy Kinematic Viscosity and requires Initial Turbulence to be specified in the range of 1 to 10 times that of the Kinematic Viscosity of the Real Fluid Kinematic Viscosity. I am not aware which solver you are refering. Very less knowledge about the Fuly Coupled Solvers. Cheers.

In fact I'm working on a non commercial solver, and I have some convergence problems for low viscosity. But I will try to make some analysis based on taylor's developements to compute numerical viscosity.

Bye