Yet another fluctuating residual problem
 

Hi, everyone.

My domain is for a two stage axial turbine.I have set SST-kw model with auto-time scale.
I have a problem where the residuals fluctuate with high amplitudes(RMS residuals fluctuate around 1e-4 and MAX residuals fluctuate around 1e-2), as shown in the figures attached.This is highly undesirable for me.I have checked domain wise and found which domains the fluctuations are observed.The mass-imbalance is 0.0000%.

All the mesh metrics are in accordance to that of the CFX requirements.

https://www.sharcnet.ca/Software/Flu.../i1323480.html

But I am being skeptical about my results and its too late to re run.Are there any methods that I can check correctness of my solution??

I do have the locations of max residuals from the ".OUT" file. It shows around 28 nodes having this high value of residuals. Is it OK? Please find the files attached.
Yes!!...I have read the FAQs,corresponding links of convergence criteria and also previous posts...But still I am skeptical and also a little optimistic that my results are not erroneous... Kindly go through the attachments.

Thanks in advance...

Some comments:

There is no such thing as CFX mesh requirements. Some simulations are very tolerant of poor mesh (eg solid domains, low Re flows) and some are very intolerant (eg shock waves, surface tension). For instance if you are trying to do a surface tension model for reasonable accuracy you need a hexagonal grid with adjacent element size ratio of about 1.2 or better.

The reference you quote is for Fluent anyway. Those numbers are not necessarily relevant for CFX.

So my point is there is no universal mesh quality requirement. In fact I bet the root cause of your current problem is mesh quality (it is mesh quality in about 90% of cases).

Yes: Do a mesh, time step and convergence sensitivity study. Also consider boundary proximity.

You say you have read the FAQ, but this FAQ (http://www.cfd-online.com/Wiki/Ansys...gence_criteria) explains exactly what to do in your case.

Monitor the quantities you are interested in, and then decide the convergence, instead of waiting for smooth sloping residuals. Often, "Local timescale" option helps with unstable simulations. Start with a smaller timescale factor of 5 and then gradually ramp it up during the run. This is a smoother way of reducing the residuals. Once you increase the local timescale factor to say 100 and the monitors still remain flat, you are closer to the convergence. From then on it is upto you to decide how long you wish to run and how accurate you want the results to be.

Also, to decide mesh independence, you can use following guide:
http://journaltool.asme.org/Template...umAccuracy.pdf