Learning Forum

[zss]

Hello! Working with some models on our customer PC with pretty fast professional video card we've noticed that depending on position of the Speed-Fidelity we have wrong result sometimes. Some loads on symmetric model sometimes doesn't seem to act at all as on the pictures in attachment. I can't reproduce the issue on my two computers but I've seen it on the customer's one.          The model is loaded only by 5 displacements on the radial directions.

[Gaurav Sharma]

Zagrebelny

Based on the pictures you shared, it seems that at low fidelity, you are observing some randomness with the results. Either the results are not symmetric or at some locations, it appears as if no load has been applied, though a radial displacement is there. If I look at the higher fidelity plots, the results look all good.

If my understanding is correct, and with the known limitation of thin geometries, this might be because the thickness of component at few locations appears to be relatively small compared to the major dimension. For such cases, a low fidelity solution might fail to respect the geometry at times and it is advisable to use higher fidelity.

Also, the resolution used by the solver depends on the GPU capabilities. So, its quite possible that even at high fidelity, the resolution used at your machine is different then the one at customer's end and hence, you are not able to reproduce the problem.

Nevertheless, as the max fidelity solution appears to yield expected results, everything seems to be as expected.

Please let me know if I have missed anything.

Best Regards,

Gaurav

[zss]

Gaurav_ANSYS thank you for reply.

I know that the solver depends on the GPU but in this case (on different PCs) depending on the position of Speed-Fidelity different loads don't work (new position - new load may not work). This model is symmetric and the problem can be easily noticed. But on one PC the correct (symmetric) result on this model comes only with some middle position of Speed-Fidelity and turning it to max Fidelity make it unsymmetric. Unfortunately it is not possible now to send you a screenshot of this situation but I give you my word that it did happen.

The main question here is: is there some procedure to check the correctness of acting of the loads (if they act or not)? In this particular model it is easily noticed but for unsymmetric model with unsymmetric loads for non-engineers it is not clear how to understand that the model is solved correctly and all loads actually act. 

[Gaurav Sharma]

Zagrebelny 

As I said earlier, results can be different on different PCs even if run at same fidelity. However, the observation you have reported that results are expected at middle fidelity and goes bad at max fidelity, is a bit strange. Unfortunately, since these observations are not captured, its not possible to discuss it further, but I would request you to please share with us, if you are able to reproduce any such observation in future.

Coming to your question: The graphics easily convey if the problem is properly discretized at a particular fidelity. There will be missing surfaces on the results contour for a badly solved problem. At this instant, it is recommended to increase the fidelity and if the problem, persists even at max fidelity, a higher capability graphics card is required to solve the problem. However, I agree at times (for some rare problems), it may still be difficult to identify this visually. For such cases, standard post processing techniques can be useful.

This is in sync with what you are already doing, visually analyzing if the deformations are inline with the loads applied, playing animations for deformed shape can be useful here. Even further, for a structural analysis, the best practice is to use a reaction resultant calculator and extract reaction forces at applied BCs. Force balance must be ensured for equilibrium and any deviation can indicate a problem with the solution.

I hope that explains, please let me know if I need to elaborate further.

Thanks,

Gaurav