CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   FLUENT (http://www.cfd-online.com/Forums/fluent/)
-   -   Setting convergence criterion (http://www.cfd-online.com/Forums/fluent/41898-setting-convergence-criterion.html)

Vidya July 31, 2006 22:11

Setting convergence criterion
 
Hi,

I have always used the default convergence criterion of 1e-3 to determine the convergence. I got convergence within 110 iterations with the k-e model that I'm using. Is there any necessity to tighten this criterion to 10e-5 or lower? How do we know what is appropriate for a particular case?

My other question is regarding the creation of a rake in the flow domain. I made a rake with 1000 points (the max value) spanning the flow domain. But when i tried to display the vectors on that rake, all I get is one single arrow. What could be wrong? Also, I was determining the trend during computation on that rake. The value remained zero throughout the computation...... what does this mean? I'm puzzled by this. I was also monitoring the trend on the Xz, Yz planes along the flow field, and these showed some values. Only the values along the rake and the inlet zone showed a value of zero and stayed there.


Swarup August 1, 2006 04:50

Re: Setting convergence criterion
 
I shall try first question only. 1.0e-03 is just one indication that errors are under control. You should continue iterating with still lower residual criteria to a point where they go flat. This means that with the current set-up, machine can calculate only upto that level and a finer grid may likely allow you even further reduction. As is said infinitely many times, "check other variables of interest". It may not be possible to specify "appropriate" for a class of problems but I guess somebody here should tell whether such "acceptable" values/criteria do exist for k-epsilon model.

Regards. Swarup.

Jason August 1, 2006 08:10

Re: Setting convergence criterion
 
You had sent me plots before where you kept claiming it was 0 because the scale on the left was all zeros... but when I changed the axis from float to scientific, that showed that it wasn't actually zero... typically if the value really is zero the plot will have a max of 1 a min of 1, and your plot will be a straight line. Also, what exactly are you monitoring? On a rake, you're not monitoring 1000 individual points... you can either monitor an average, a maximum, or a minimum along the rake... the plots you sent me previously were monitoring inegrals on a small surface, and because of the units you were getting values on the order of 1e-5... but that was purely because of the monitor you chose (integral becomes (m/s)(m^2)... and since your velocity was low, and your area was very small, that's why you were getting such low values...). You need to pay attention to what you're actually looking at. If you're plotting minimum velocity, and there's even one point along the rake where the velocity goes to zero, then you're going to get zero. In the plots you sent me before it wasn't just a flat line... it oscillated about a value and converged on a value... that was a big hint that it wasn't just 0.00 like your scale was claiming.

And once again, you can't judge convergence based on residuals alone. You need to monitor other properties and make sure they level off(like Swarup said), even if that means dropping your residuals to 1e-6 or lower!

Hope this helps, and good luck, Jason

Vidya August 2, 2006 23:42

Re: Setting convergence criterion
 
Hi Jason, What you say makes absolute sense, and indeed I was monitoring the integrals. How do we monitor area weighted variables like pressure or velocity? And also, please explain how you changed the scale from float to scientific?

What I want to do is to monitor the average velocity along the rake? the velocity should be the centerline velocity? How do we choose this parameter, particularly the centerline velocity?

Each time I run the simulation, I still get the values as zero...... just like the plots I sent you earlier.

Thanks, Vidya

Jason August 3, 2006 08:36

Re: Setting convergence criterion
 
Don't monitor integrals unless you have a reason to monitor the integral. (i.e. the integral of pressure over an area would be a pressure force (N/m^2*m^2=N) or the integral of a velocity over an area is a volume flow rate (m/s*m^2=m^3/s)).

If your rake is along the centerline of the pipe, then you can monitor the average (area-weighted), the maximum, or the minimum (or setup multiple monitors on the same rake and monitor all three quantities). If you're monitoring the average, it's going to be the average along the centerline... the maximum is going to occur at the constriction (i.e. where the maximum velocity occurs), and the minimum will most likely occur near your BCs.

If your rake is radial, then you want to monitor either the average (area-weighted) or the maximum. If you monitor the minimum on a radial rake, it's going to be 0 at the wall!

I would think you would want to monitor the maximum velocity... this is where the flow is most likely going to be changing the most, and if you do an average then the variation in the maximum velocity may be cancelled out (or at least washed out if it occurs over a small area) and you may think your value is converged when it isn't.

Now, for changing the scale... when you're defining your monitor (in the window where you setup which surface to monitor and what quantities to monitor) there is an "Axes" button... click this... the window that shows up is how you change the scales, and set ranges if you want.

Again, be VERY careful about saying the values are 0, because based on the plots you sent me before they aren't... if the accuracy of your scale is 0.00, then anything below 0.005 will not show up on your plots. The reason you know it's not actually zero is because there is a shape to the plot... if it were actually zero, then it would be a flat line, and I believe the defaults in Fluent for this situation is to have a minimum of -1, a maximum of 1, and then you'll just have a horizontal line at 0... if the plot shows any shape to the curve, and the range of the plot isn't from -1 to 1, then your values aren't 0, you're just not plotting enough digits to see what the value is. This is why I recommend switching to Exponential notation... it will show however many significant figures you ask for, but because the decimal point is allowed to float, you can see values for small numbers!

Hope this helps, and good luck, Jason

Vidya August 6, 2006 16:46

Re: Setting convergence criterion
 
I plotted the shear stress values on the rake created with a 1000 points. It gives me a zero value and then a few points here and there with some value. I expect high shear stresses at those regions. Is this something wrong with the rake creation or is it that the shear stresses are so low?



All times are GMT -4. The time now is 02:00.