CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   Convergence of Radial Turbine Simulation (

Hugh July 15, 2004 09:31

Convergence of Radial Turbine Simulation
I am undertaking CFX analyses of a nozzleless radial turbine but am having great difficulty in achieving convergence. If I use a high fluid time step then I get a 'fatal overflow' error and if I use a small fluid time step then the error is a problem with a computed static pressure less than or equal to zero with recommendations to use an even smaller time step. Are there any suggestions on how this problem can be solved? Many Thanks!

Glenn Horrocks July 15, 2004 18:17

Re: Convergence of Radial Turbine Simulation
Hi Hugh,

Is your mesh fine enough? Are your physical models appropriate for the physics of the simulation? I'll bet the answer to one of those questions is no.


Karthi June 25, 2010 01:48


I'm new to CFX. I'm trying to simulate a base case for a radial inflow turbine only the runner. The runner is of a turbocharger, inlet area 49mm*26mm
Here are my settings:
Fluid domain, rotating, 68K RPM, Axis Z, clockwise rotation, ref pressure 1 atm
Rotating, Velocity initialized in cyl. coord. as -30m/s(radially inward) in radial direction in rotating frame, 400'C temp and 0 atm rel. press.
Rotating frame, 600'c and 2 bar inlet pressure, flow direction cyl coord: 0,-1,0(a,r,t)
Rotating frame, 0 atm rel pressure(static).
Periodicity is ensured 1 to 1

Meshed in ICEM, quality above .25 min angle 18'

About solver: High res, 1st order and Phy timescale .5 sec

I know a little about most of to params i have given......I am giving all settings so that someone can help me....

Coming to actual prob.....The solver either oscillates very badly and flow reversal error appears to close 80% of both inlet and outlet boundaries.
Or sometimes the solver doesnt even start saying a param DOMVEL is uninitialised

Someone pls help me....i'm running outta time

ghorrocks June 25, 2010 06:39

Attesz June 25, 2010 11:09

My suggestions:
- do you calculated the initial values correctly? if not sure, use automatic initialisation instead
- first run a simple steady state simulation, but use high resolution.
- you can use aggressive time step, and if there isn't any convergence problems, you can increase the scale factor, to 5 or 10.
- first run a semi-load operation point, for example on the desired rotating speed but much lower pressure ratio. the convergence of these points are more stable and fast. from the results you can increase the pressure ratio using interpolation.

Send picture about the mesh and the conv. curves if you can.


Karthi June 26, 2010 01:43


thanks for reply

Automatic initialisation does not work. an error: subroutine getVar is not able to retrieve parameter DOMVEL in ZN1 occurs. The values for velocity i have given are tentative only.

However 1 of the trials did converge though with high reversal of flows at the outlet. From the post i found out the mass flow rate at inlet and changed the inlet boundary as mass-flow inlet and stat press outlet....

BUT.....this worked for coarse mesh but in the fine mesh....the RMS-P mass oscillated very vigorously after 90 iterations and did not converge......?

I also have confusion abt the sign of velocity init. I have given -30m/s radial.....because this right? or should i jus give 30m/s ??????

Wall formation at the outlet due flow reversal occurs very frequently upto 80%....

How should i define the outlet boundary frame type: ????

Rotating or stationary?? because i am not defining any flow direction or vectors at the outlet

thanks for reply once again

ghorrocks June 26, 2010 07:28

Attila - the FAQ I linked to does not discuss initial guesses or ramping up to full flow. These are both very useful techniques - can you add them to the FAQ? Questions like this get asked very often so let's make the FAQ as good as possible.

Karthik - It is common for a coarse mesh to diverge and a fine mesh to diverge. The finer mesh is capturing finer flow features which are more difficult to converge and may also start being transient and therefore require careful consideration.

The coordinate system is defined in the documentation. Just output your initial guess and look at it in Post if you don't trust it.

The wall formation at outlet indicates you have backflow at the outlet. Can you move your outlet downstream? This is by far the best way to deal with it.

Karthi June 29, 2010 10:18

Thnks all....ghorrocks.....i found the was with time stepping and inlet total pressure conditions.........

Now i need help in one more aspect.........i'm modelling an internal channel to cool the leading edge of the blade... channel dia is 1/3rd blade thickness....i'm using air at 2 bar 150'C as coolant.....i've meshed in ICEM and imported into CFX and made the domains......

My doubt is once i have created a domain interface, an interface boundary is automatically attached to both the domains.......but is it necessary is explicitly define the common wall boundary as CHT boundary........?

Secondly.....which parameter that we set in solid domain definition defines the solid to conduct heat inward..... should i make the surface heat on the blade surface to interract with the cooling effect of the fluid in internal passage?

if u can help me......any help....pls do so.....

thnks in advance

ghorrocks June 29, 2010 18:50


is it necessary is explicitly define the common wall boundary as CHT boundary........?
If automatic detection of interfaces is on it will have a go at doing it automatically. Always check that the interface it generates is sensible before using it. Otherwise just generate it manually.


which parameter that we set in solid domain definition defines the solid to conduct heat inward
No such thing. You sent the initial and boundary conditions to have a temeprature field and the heat flows accordingly. You do not define the heat flow to be in a certain direction.


how should i make the surface heat on the blade surface to interract with the cooling effect of the fluid in internal passage?
That is automatically handled in a CHT solution.

These are pretty basic CHT questions. Sounds like you should do the CHT tutorial in the CFX tutorial manual. Have a look at the heater coil example, that will help you.

Karthi June 30, 2010 03:07


So basically wat you tht domain interfaces are optional .....So i can avoid creating domain interfaces.......instead i can define the blade's interior channel wall as a non-adiabatic wall and give some heat transfer prop to it....right?

ghorrocks June 30, 2010 08:08

In that case remove the mesh for the cooling channel and just impose your boundary condition on the coolant channel wall.

All times are GMT -4. The time now is 19:11.