[PabloXX37]

I am simulating the flow through a transonic axial compressor stage.

I am having some difficulties getting consistent results. These inconsistencies are related to the use of a custom gas model, especially due to the Min/Max values used for the Table Generation (photo for clarity).




It seems that prescribing different limits to the maximum temperature yield very different simulation results (in massflow, pressure ratio, efficiency). This would be fine, except for the for the fact that the temperature limits that I’m prescribing (5400R, 9000R, etc.) are actually way above any values that will realistically appear in my simulation solution. The same principle applies when I try to apply clippings (temp extrapolation - off). At those limits the clippings are never engaged, but the solution changes nonetheless.

What is the exact way that CFX handles the Table Generation? Are there some aspects of the Min/max values selection that could cause a fundamental difference in the solution? And the clipping?



Some extra info and additional questions:

INFO.
- As the working fluid I am using a custom material. It is being modeled with an Ideal Gas equation of state with and CEL expression for the Specific Heat Capacity (Cp), which is a function of temperature.

ADDITIONAL QUESTIONS
- How can 2 different temperature limits for the table generation change the solution, especially if the temperature range of the solution is fully contained within these limits in both cases?

- How can the clipping of the variables fundamentally change the solution, even when the clipping is not even needed/activated/used?



Also from the manual I've seen that CFX uses an adaptive algorithm to select the table points. My initial guess was that changing the table ranges leads to a loss of precision (if the number of points is not automatically adjusted). So I tried running two more cases with the option "Error Tolerance= 0.01" active. This setup had very little effect when the variables were clipped, and none at all when no clipping was present. So I think that the differences in the solution are not due to a bad interpolation of the table values.

- I would appreciate any insight on the Table Generation mechanism of CFX to understand the reasons for the different solutions, and to try and have more consistent results. Are there any recommended best practices?

-Lastly, out of curiosity, if I set my table generation limits to something that is less than the actual range of temperatures of the problem (eg. [500-600K]-> very narrow range) and do not touch any extrapolation/clipping settings, how would CFX handle the values that go outside of this predefined range? Would it clip them or extrapolate them? What would be the default mode for CFX in this case?

[ghorrocks]

Table generation rarely signicently affects results. So the fact that it does in your case means you have an unusual material properties. They are likely to be highly curved. I would check whether they are correct.

I don't know the details of exactly how CFX chooses the points (because it has never caused me a problem before), but it is obviously fitting points to the properties curves to simplify obtaining the properties and their gradients during run-time. This means normal curve-fitting procedure would be to use more points and/or a tighter tolerance until it does not affect the results - then you have sufficient accuracy on your properties that you know you are resolving them accurately.

If you have inadequate resolution on the table then changes to the table will cause large changes in results - which appears to be exactly what you are seeing.

[PabloXX37]

Originally Posted by ghorrocks

Table generation rarely signicently affects results. So the fact that it does in your case means you have an unusual material properties. They are likely to be highly curved. I would check whether they are correct.

I don't know the details of exactly how CFX chooses the points (because it has never caused me a problem before), but it is obviously fitting points to the properties curves to simplify obtaining the properties and their gradients during run-time. This means normal curve-fitting procedure would be to use more points and/or a tighter tolerance until it does not affect the results - then you have sufficient accuracy on your properties that you know you are resolving them accurately.

If you have inadequate resolution on the table then changes to the table will cause large changes in results - which appears to be exactly what you are seeing.

Thanks for the feedback.

I have double checked the material properties, and they are very standard. I would argue that the Cp curve that I'm using is not very curved, and could actually be approximated by a linear relation in the working temperature range that I am using.

It also surprises me that I might have an inadequate resolution of my table, since by default the error tolerance for the table generation is 1%, which should be enough for most applications. I might try to do a run setting lower tolerances to see if this has any effect.


In any case, I have just opted for a specific temperature range that covers all of my possible temperatures with a bit of leeway (following the CFX manual best practices). I'll at least try to be consistent and use the same settings for all further simulations.

If I figure any other possible causes for this variability I'll post some of my conclusions as a follow up to this message.

[TWA]

PabloXX37, did you ever conclude on this matter? I am seeing similar behavior in my analysis.

[PabloXX37]

Hey TWA. Sadly, I could not follow up on the issue, so I had to compromise. I can offer the following suggestions:
1. Set limits for your temperature and pressure that are wider than what you expect in your final solution. This extra space is also to accomodate some outlying temperatures and pressures that may arise in the transient before the final solution is reached.
2. I recommend that you do not extrapolate. It is better to clip your variables. This is done by selecting Temp/Pressure Extrapolation, but then unchecking everything inside those options.
3. I recommend that you run with the 0.01 tolerance option enabled. That way you will get good accuracy if your polynomial is highly non linear.
4. Finally, if you are using equations in other places (for example, Sutherlands law for viscosity), then be careful that your reference values for temperature and other values are inside the limits you have set previously (see point 1)

If you can, bring this up with Ansys customer support (it helps if this is part of a professional setting), they are usually very helpful.

If you need anything else, let me know. Best of luck!