[MachZero]

Has anyone tried ansys discovery live? It looks very beautiful but I have to imagine it doesn't deliver much in terms of accuracy. You can't just magically speed up all those numerics.

Anyone have thoughts?

[evan247]

Tried the demo using their GPU cluster a month ago. The type of simulation you could perform seemed rather 'basic'. At least I couldn't find anywhere to setup periodic BC, non-rectangular fluid domain, moving interface, etc.

I'm not too surprised with the speed, given the documentations recommended high-end GPU cards. Moreover, you don't really get the 'converged solution' until a few seconds after you click the start button. It looks like a time-marching code which shows you how the flow 'develops' as it gradually settles down.

I'm rather interested in the background CFD code (Fluent? CFX?), as I remember the GPU acceleration of Fluent isn't very satisfactory.

[MachZero]

Interesting. I didn't realize it was built for gpus. I also remember not being impressed with gpus in fluent so they must have built the numerics for that.

A few seconds run time based on a single high end gpu? That seems too good to be true, right? Is it a really coarse grid it's using or is the answer not very accurate?

I get the impression it's not meant to replace fluent but rather be a tool a designer can hit the run button on and get either a pretty picture or (if it is as good as they make it sound to be) a warm and fuzzy for trends.

[Gert-Jan]

It is interesting meshless CFD. I tried the trial version of Discovery Live to generate a nice animation. That worked pretty well. The LES-like velocity field always impresses the audience.

However, when I compared the pressure drop results with CFX, i was flabbergasted, see picture.

Another issue. I had a duct with a thin obstacle in the ceiling. This obstacle was not 'felt' by the software. The air flows right through it. The thickness of the obstacle and the so-called 'fidelity' determines whether the software takes the obstacle into account or not.

It is certainly not developed to replace CFX or Fluent. It is a no-budget tool for e.g. draftsmen. But if the pressure drop does not make sense at all, I think it is worthless, for whoever will use it.

I don't want to bash ANSYS right now, but expect them to narrow the gap in the future. I'm following it closely to see how it will develop.

If someone has better or similar results to share, please post it.

[MachZero]

Your experience is along the lines of what I was expecting.

[hpvd]

Originally Posted by Gert-Jan

...
However, when I compared the pressure drop results with CFX, i was flabbergasted, see picture.
...

Many thanks for posting these comparison!


Of course I wouldn't have expected perfect results, but not even getting the "order" of these geometries matching is pretty worse.



Just to dive somehow deeper:
1) how did you determine the "final" results in discovery live?
- how long does the calculation run?
- on which type of GPU?
- did you use the "average function" to make it "steady"?
2) Did you try to use the slider speed vs fidelity?
- would this help to come closer to the cfx results?
- or at least to get the same "order" of the 3 geometry types?
3) how sure are you about the results of cfx?
- would you mind giving some details regarding mesh setup and solver?


Many many thanks !

[Gert-Jan]

It has been a while a go. I used the evaluation version. Maybe things have improved.
1)
- Calculation is within a minute or so
- nVidia Quadro M4000 8GB 4 x displayport
- Just monitor the total pressures and wait until you get a reasonable average.
2)
- I can't recall slider speed. Certainly I played around with Fidelity.
- I put Fidelity as far as possible, but don't remember there was any improvement
- I used the evaluation version. And did not purchase it. I have CFX and Fluent, so why should I?
3)
- I'm not sure. I don't have experimental values. But my customer thought these were logical. And in line what he knows from practice.
- Tet/prism and Coupled solver of CFX. Never lets me down.
4) If I find the time, I can add Fluent results.

[siw]

Have you read the links to Discovery Live here: https://blog.pointwise.com/2017/09/1...ek-in-cfd-289/. In those articles it states that the simulation methodology is proprietary so unknown to the User - so presumably you cannot construct a mesh, choose a turbulence model etc. so how can the results be sensible as how does the software know what is most suitable.

It's not something that would be used where I work. We stick with using CFX/Fluent/Mechanical. It all looks very nice in their demos but from my experience these are not good tools for a design engineer, as they are aimed at them rather than CFD/FEA specialists.

[Gert-Jan]

I'll read the Blog.

I have been told it is meshless CFD. The accuracy depends on the fidelity, which has an effecton on which details are captured or not. That is the only control you have.
So, there is no input required, you don't know what is behind the screen. I would not trust it unless you can compare it with experiments.

I used it for a fancy animation. For presentation puroposes. Never had a LES-like animation at such low cost. For that it is great.

Look on sheet 14 of this presentation: http://investors.ansys.com/~/media/F...esentation.pdf
Current users of CFX/Fluent/Mechanical are on the top of the pyramid. ANSYS wants the whole pyramid to be filled with DL-users.

It is also possible to perform mechanical calculations. Did anyone test and compared this? Please share your findings........

[hpvd]

@Gert-Jan: many thanks for your detailed answer !

[Gert-Jan]

And now comparison with Fluent added

[hpvd]

Originally Posted by Gert-Jan

And now comparison with Fluent added

really interesting! thanks again!

[asoltoon]

Hello dears,

I am a bit late to speak about the topic. I have got familiar with the software recently. As I have understood, and as you can watch in the official video(minute 19), Discovery Live is not an alternative for ansys mechanical or fluent etc. It is just a preliminary software in the first steps of the design process to help the engineers to make better approximations and make the geometry more efficient in a very short time. After that using analytical common softwares like ANSYS or Abaqus should be used to make more accurate results.

I have not worked with the software yet, so can anybody tell me if my sentences is right about the software or not.

thanks.

[ElenaMa]

Good day!

Could you tell me, please, can I use a surface body as geometry in discovery life?

I tried to sew the model into a solid body in NX and in discovery itself, but apparently the geometry is not very clean, so the stitching does not work .

I remember that at the presentation in discovery was loaded model of the earth's surface in stl format and there were huge holes, but discovery normally calculate the air flow , simulating the air on the earth's surface. In my case, if you load a surface, the program ignores it and air through through these surfaces.

Or maybe the discovery there are settings for this. I did not find this information in the help and manuals.

[Gert-Jan]

According to my experience a year ago: no. It was not even able to model thin solids.
But things can change significantly in a year. So, you'd better ask support.

[ElenaMa]

Originally Posted by Gert-Jan

According to my experience a year ago: no. It was not even able to model thin solids.
But things can change significantly in a year. So, you'd better ask support.

Thank you!

[CrackerJacks]

Late reply, but I think the input is worth it. We have recently acquired Discovery Live Standard and I have tested it against Ansys Fluent. All the issues Gert-Jan encountered are still present. The pressure drop is roughly double what Fluent and theoretical calculations predict. Furthermore, thin parts of our geometry are not being resolved even at the highest fidelity.


My understanding is that they are using a "proprietary" version of Lattice Boltzmann - which, if the resolution were high enough - would be fine. But they don't give you enough control over fidelity or operating conditions.


It makes pretty pictures but is otherwise useless.

[CLEB44]

I've tried discovery live in a mechanical way to have a Von Mises stress.
I've test the same model (built in spaceclaim) in ANSYS Mechanical.
In Discovery live, the stress value are 80% more that the values calculated via Mechanical.
Discovery live is easy to use, but if the calculations are not good, there's no utility to use it.

[Gert-Jan]

Is it an extra built-in safety factor? To be on the safe side when users are applying the Discovery outcome in real life?

[ersanzop]

I had a duct with a thin obstacle in the ceiling. This obstacle was not 'felt' by the software techzpod mobdro download