Convergence Plot Spike After Simulation Re-Start

[pwilkins]

Sometimes during simulations I need to stop the study for whatever reason.  On a few occassions now I have gotten a spike in convergence values upon re-start of the study, even when the convergence plots seemed to be steadying out before the simulation stop.  What could be causing this to happen?

For instance, pressure will be at a steady-state and then spike and settle at a higher steady-state value.  The spike can take as few as 10 iterations for some plotted values and many more for other values.

[OmkarJ]

The spike occurs because the momentum terms are recalculated. And since SimCFD uses seggregated algorithm, which uses velocity values to calculate pressure/turbulence quantities, you may see a spike in all these. After few iterations, things should be back to normal, although, I agree this is a strange behaviour in the newer versions. 

[napoleonm]

This spike did not appears in old versions of CFDesign (v10 and V9)

[Royce_adsk]

This 'spike' has always been around, at least for as long as I've been using the software, and is more pronouced in some simulations compared to others.  Many times I don't notice much of a spike at all.  Technically, after a continue you should run for at least another 50-100.

Omkar explained the source of the issue perfectly.  Either way, it is something you can generally ignore.

Best regards,

[awiertek]

So I also have a question regarding this spike. It was said previously that after the spike values should get back to previous level after some amount of iterations, but in my case they did not. The solution converged finally with the values on totally different level. This was a natural convection simulation with Joule heating applied on copper parts. Please see attached plot for average values. What is a cause of it? Do I have to run the simulation once again? It's almost 15 000 000 elements simulation, so it takes about 2 days to run.

For this simulation I used mixing length turbulence model - and here is my second question. Although avg values got converged max and min values plot looks very bad for me. It's very jagged. The solution stop bcs with the output: time avg values flatten and solver automatically stopped. So the solution is converged anyway even if max and min plot is not really flat. Is it a nature of this turbulence model? Is there a way to get this plots flat as well? Please see attached plot.

Regards,

Artur

[Jon.Wilde]

Hi Artur,

Can I suggest starting up a new thread next time? 🙂

To try to help, are you running steady state or transient? If transient, what is your timestep size? Small enough to allow CFD to converge on the changing flow simulation?

What changed when you stopped and re-started? Anything that could cause the change in results?

Thanks,

Jon

[awiertek]

Hi Jon,

Ok, next time I will start a new thread 🙂 It's steady state simulation. I did not change anything before restarting. Set up is exactly the same as at the very beginning. The reason I stopped the simulation is I had to save result files of the other scenario.

Regards,

Artur

[Jon.Wilde]

No problem, just helps keep our forum nice and tidy 🙂

A spike is to be expected but your expectations are correct, we should see the same results.

Seeing what you are showing, I would be inclined to run it again. It is possible that the restart threw the results - are you happy with the mesh that you have? Is the Nodal Aspect Ratio (NAR) nice and low (below 150)? (You can show this through Solve -> Result Quantities.

Thanks,

Jon

[awiertek]

Jon,

To answer your question about mesh quality - I think my mesh is not the worst one... I refined mesh in some regions and I still have very small regions with NAR above 150 (even around 1500), but these are some singular elements, or groups of elements with NAR 150-200A. I can say that it is less than 0,01% of all elements and my total element number is 15 000 000. Additionally it is very irregular geometry of electrical module (even entire structure) so you can imagine it's hard to go below NAR 150 for totally all elements. All bad NAR elements are far away from the regions with highest temperature and velocity, so I believe they should not impact on e.g. permanent temperature rise after the spike.

So after refining my mesh I ran simulation once again until average plot where almost flat, than I stopped the solution and ran once again without any additional changes. Spike appeared and once again solution is going to converge with values on the totally different level.

Which level is correct? How I can eliminate this issue?

Please note that max and min values plots are sharp and only avg is flat. What is the cause of it? How I can get max and min plots flat? Is it connected with turbulence model I use (mixing length)?  

Please see attached plots.

Regards,

Artur