Learning Forum

[zou.xf]

 I have a twisted tube heat exchanger problem. The result by Discovery Live is wrong. I guess the reason is that DL has fidelity upper limit, so that it cannot catch the tiny gap in some place. 

 

This is the model, the arrows show the expected flow

 

Discovery Live give a wrong result that penetrate the wall

 

 

the whole model is a 6000mm long cylinder. Its diameter is 3000mm

 

 

It has huge number of small structures from 2mm to 10mm. It seems that discovery live does not capture the details enough to give reliable results, even with the highest fidelity.

 

system configuration

 4 GTX 1080TI graphics cards server, each card has 11GB gpu memory. The system has 96GB memory.

windows 10 64bit, Discovery Live 19.1

 

Anyone can help me on this problem? Thanks!

[Naresh Patre]

Hello  davidzou

The thickness of 2mm is very small as compared to the cylinder length which is 6000 mm, so it's very difficult to capture this small thickness even with highest fidelity. You may want to increase the thickness of the walls in this case. Or, if possible reduce the length of the cylinder. Please check out below post on thin geometric features.

Discovery Live: "Thin" geometric features and relative GPU adjustments

[zou.xf]

Naresh I have already tested with one shorter cylinder (1000mm), it failed. (the graphics window crashed). Then I remove 90% complicate structures, and I get the above results.

 

As mentioned above, my system is  4 GTX 1080TI graphics cards server, each card has 11GB gpu memory (total 44GB gpu memory). The system has 96GB memory.  CPU is intel i7 6850K, 12 cores.

 

I want to solve the problem, I do not need it to be fast as 1 second to get the result. I may accept that if Discovery Live give the right result after 24 hours computation. I could extend the gpu and cpu and system memory as it need to get the right result.

 

Can you recommend a system configuration with much larger gpu memory to solve my problem?

What is the highest configuration Discovery live have tested? Have Discovery Live ever tested on a problem that need 100GB GPU memory? 1000GB gpu memory?

[Naresh Patre]

davidzou  Even if you have 4 Graphics cards, only one card will be used for solving a problem in Discovery Live. 

Discovery Live relies on GPU memory and not CPU memory. You may want to check out the hardware requirements posted on forum at this link . 

I doubt whether a 16 GB graphics card will be also to resolve the thin wall structure of 2mm present in your model. If you can share the model with me, I can test it on one such machine and let you know.

I would recommend you to try solving your model in Discovery AIM which will not only help you capture all geometry details but also give you accurate results.

[papp]

Naresh Patre I'd like to point out that the boundary layers are rather asymmetric on the 2 sides of the inlet tube in the 2nd image.

I think it has to do with the fact that the whole geometry is way larger than this exact part and therefore DL cannot capture it correctly. However, even smaller parts are meshed (same figure), so there might be another type of misbehaviour.

Can you suggest any explanations to this phenomenon? Correct me if I'm wrong, but this seems to be a significant flaw that might be worth looking into.

[zou.xf]

I also got answer from Aric Meyer: DL is not meant for late structures with small features.  General rules is not smaller than 0.5% of the full size.

 

Thank you all. I'm trying Discovery AIM, hoping AIM do the job.

[Naresh Patre]

Hello  Bálint Papp

As  davidzou rightly said, very small features will not get captured inside Discovery Live if the domain size is too big. 

Regarding the asymmetric boundary layer issue, if we can get a sample geometry to reproduce the issue, we can test it and report it to development.