DNS using Fluent 15.0

adnan.nweilati · October 17, 2014, 17:40

Hello fellow Fluent users,

I'm currently doing my final year in Mechanical Engineering. The dissertation topic I choose is basically a project aiming at direct simulation of turbulent flow without using turbulence models.

In other words, I have to preform DNS on Fluent (I was told by my adviser that I have to work with Fluent).

After going through this website's threads, I can say I have a good idea about DNS and how crazy it is in term of computational power and time consumption.

However, there is a question that I'm really confused about. How am I going to perform the DNS without the model itself. Does that mean I don't have to create a model in ICEM? (the software I was taught how to use for creating mesh grids)

Or do I create a really simple geometry (like a rectangle) with a really really fine mesh inside it and choose laminar flow as viscous model?

Please help this confused student

Thanks!

ghost82 · October 18, 2014, 11:05

It is theoretically possible to perform DNS in fluent: "all" you have to do is to create a very fine mesh (able to resolve the kolmogorov scale), choose the laminar model and make a transient simulation with a very small time step (turbulent time scale).
I hope you have a very big cluster and a lot of memory......

adnan.nweilati · October 18, 2014, 11:26

Correct me if Im wrong but this is how Im going to approach this:

- Create a simple rectangle in ICEM and specify the inlet and the walls
- Create the fine mesh
- Go to fluent and choose the laminar model and make a transient simulation with a very small time step
- Hope my computer wont explode

ghost82 · October 18, 2014, 11:31

I think you will not be able to perform such an analysis without a big cluster and (with a big cluster) with a lot of time...
What's your Reynolds number?

However, basically the approach you described is correct.

adnan.nweilati · October 18, 2014, 11:46

Around 68000. My inlet speed is 10 m/s and my diameter is 0.1 m

ghost82 · October 18, 2014, 11:50

There is an old post in cfd online about fluent and DNS:
http://www.cfd-online.com/Forums/mai...e-dns-les.html

From post #5 you can understand that you cannot perform this simulation without a very big cluster

You should have a mesh > several hundreds milions cells and time steps of the order of 10^-10 s

Who is the crazy person who told you about this project???

adnan.nweilati · October 18, 2014, 12:01

I choose it from a long list of topics

I didnt know about how complicated DNS was at that time!

Im getting really worried about this now

ghost82 · October 18, 2014, 12:02

Can't you change the topic?
It is not possible to assign these topics to students without providing at least hardware to solve them...

adnan.nweilati · October 18, 2014, 12:15

Its not possible to change it since its been a month since we started.

Worst case scenario is that I have to at least perform RANS and LES instead of DNS.

Or maybe do the DNS but with an extremely small case and compare the results to RANS and LES.

ghost82 · October 18, 2014, 12:16

Good luck my friend..

venkatesh92 · November 21, 2014, 09:25

An interesting study would be to compare performance of different turbulence models. There is plenty of literature out there for validation. Good luck!

HyperNova · November 21, 2014, 15:40

hi my friend
i read a paper few months ago , they modelled a very small geometry about 24cm^3 and Re = 180 , the mesh was about 10million and the computer had 11000 cores! this the paper you can find it on google scholar :
title : Influence of Bubbles on the Turbulence Anisotropy
Igor A. Bolotnov
Department of Nuclear Engineering, North Carolina State University, Raleigh, NC 27695

adnan.nweilati · November 25, 2014, 08:47

Hi, would you mind telling me the best literatures for validation? I tried searching for weeks for something similar to my dissertation topic but I had no luck.

A possible reason for this is that my researching skills are still poor, so I don't really know how to look a good paper.

Thanks

venkatesh92 · November 27, 2014, 05:37

You could probably use Science direct or google scholar. Unless you have a clear idea of what you want to get out of your simulation, it is very hard to find a proper literature. Once you do that, you will discover that there are more than sufficient experiments available to validate your cfd predictions....

Shamoon Jamshed · March 13, 2023, 09:52

Dear Adnan,

DNS actually does not require any turbulence model. Your title was misleading since DNS does not use any turbulence model. If it had to, then what was the need to do DNS? Modeling was enough. Anyways, DNS wants a very very fine grid. From Turbulent Flows by Pope, DNS simulation has cost Re^(9/4) in each direction. I mean your each direction cells must be proportional to Re^(9/4) for capturing a DNS flow. This will make the computation extremely heavy.

Regards,

SJ

October 17, 2014, 17:40	DNS using Fluent 15.0	#1
adnan.nweilati New Member Adnan Nweilati Join Date: Oct 2014 Posts: 18 Rep Power: 12	Hello fellow Fluent users, I'm currently doing my final year in Mechanical Engineering. The dissertation topic I choose is basically a project aiming at direct simulation of turbulent flow without using turbulence models. In other words, I have to preform DNS on Fluent (I was told by my adviser that I have to work with Fluent). After going through this website's threads, I can say I have a good idea about DNS and how crazy it is in term of computational power and time consumption. However, there is a question that I'm really confused about. How am I going to perform the DNS without the model itself. Does that mean I don't have to create a model in ICEM? (the software I was taught how to use for creating mesh grids) Or do I create a really simple geometry (like a rectangle) with a really really fine mesh inside it and choose laminar flow as viscous model? Please help this confused student Thanks!

October 18, 2014, 11:05		#2
ghost82 Senior Member Rick Join Date: Oct 2010 Posts: 1,016 Rep Power: 27	It is theoretically possible to perform DNS in fluent: "all" you have to do is to create a very fine mesh (able to resolve the kolmogorov scale), choose the laminar model and make a transient simulation with a very small time step (turbulent time scale). I hope you have a very big cluster and a lot of memory...... Jomid likes this.

October 18, 2014, 11:26		#3
adnan.nweilati New Member Adnan Nweilati Join Date: Oct 2014 Posts: 18 Rep Power: 12	Correct me if Im wrong but this is how Im going to approach this: - Create a simple rectangle in ICEM and specify the inlet and the walls - Create the fine mesh - Go to fluent and choose the laminar model and make a transient simulation with a very small time step - Hope my computer wont explode Stevie_1200 likes this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
How to start Fluent with Matlab??	Jay Hu	FLUENT	8	November 9, 2022 07:30
DNS in FLUENT	rajani	FLUENT	9	May 3, 2021 01:55
Two questions on Fluent UDF	Steven	Fluent UDF and Scheme Programming	7	March 23, 2018 04:22
Abaqus - Fluent Coupling WITHOUT MPCCI	s.mishra	FLUENT	1	April 5, 2016 07:47
What the differences flow equation of Fluent 6.3 and Fluent 12.1	opehterinar81	FLUENT	0	August 19, 2011 12:55

October 18, 2014, 11:31		#4
ghost82 Senior Member Rick Join Date: Oct 2010 Posts: 1,016 Rep Power: 27	I think you will not be able to perform such an analysis without a big cluster and (with a big cluster) with a lot of time... What's your Reynolds number? However, basically the approach you described is correct.

October 18, 2014, 11:46		#5
adnan.nweilati New Member Adnan Nweilati Join Date: Oct 2014 Posts: 18 Rep Power: 12	Around 68000. My inlet speed is 10 m/s and my diameter is 0.1 m

October 18, 2014, 11:50		#6
ghost82 Senior Member Rick Join Date: Oct 2010 Posts: 1,016 Rep Power: 27	There is an old post in cfd online about fluent and DNS: http://www.cfd-online.com/Forums/mai...e-dns-les.html From post #5 you can understand that you cannot perform this simulation without a very big cluster You should have a mesh > several hundreds milions cells and time steps of the order of 10^-10 s Who is the crazy person who told you about this project???

October 18, 2014, 12:01		#7
adnan.nweilati New Member Adnan Nweilati Join Date: Oct 2014 Posts: 18 Rep Power: 12	I choose it from a long list of topics I didnt know about how complicated DNS was at that time! Im getting really worried about this now

October 18, 2014, 12:02		#8
ghost82 Senior Member Rick Join Date: Oct 2010 Posts: 1,016 Rep Power: 27	Can't you change the topic? It is not possible to assign these topics to students without providing at least hardware to solve them...

October 18, 2014, 12:15		#9
adnan.nweilati New Member Adnan Nweilati Join Date: Oct 2014 Posts: 18 Rep Power: 12	Its not possible to change it since its been a month since we started. Worst case scenario is that I have to at least perform RANS and LES instead of DNS. Or maybe do the DNS but with an extremely small case and compare the results to RANS and LES.

October 18, 2014, 12:16		#10
ghost82 Senior Member Rick Join Date: Oct 2010 Posts: 1,016 Rep Power: 27	Good luck my friend..

November 21, 2014, 09:25		#11
venkatesh92 New Member Join Date: Nov 2014 Posts: 9 Rep Power: 12	An interesting study would be to compare performance of different turbulence models. There is plenty of literature out there for validation. Good luck!

November 21, 2014, 15:40		#12
HyperNova Senior Member B_Kia Join Date: May 2014 Location: Ir Posts: 123 Rep Power: 12	hi my friend i read a paper few months ago , they modelled a very small geometry about 24cm^3 and Re = 180 , the mesh was about 10million and the computer had 11000 cores! this the paper you can find it on google scholar : title : Influence of Bubbles on the Turbulence Anisotropy Igor A. Bolotnov Department of Nuclear Engineering, North Carolina State University, Raleigh, NC 27695

November 25, 2014, 08:47		#13
adnan.nweilati New Member Adnan Nweilati Join Date: Oct 2014 Posts: 18 Rep Power: 12	Hi, would you mind telling me the best literatures for validation? I tried searching for weeks for something similar to my dissertation topic but I had no luck. A possible reason for this is that my researching skills are still poor, so I don't really know how to look a good paper. Thanks

November 27, 2014, 05:37		#14
venkatesh92 New Member Join Date: Nov 2014 Posts: 9 Rep Power: 12	You could probably use Science direct or google scholar. Unless you have a clear idea of what you want to get out of your simulation, it is very hard to find a proper literature. Once you do that, you will discover that there are more than sufficient experiments available to validate your cfd predictions....

March 13, 2023, 09:52		#15
Shamoon Jamshed Senior Member Shamoon Jamshed Join Date: Apr 2009 Location: Karachi Posts: 381 Rep Power: 18	Dear Adnan, DNS actually does not require any turbulence model. Your title was misleading since DNS does not use any turbulence model. If it had to, then what was the need to do DNS? Modeling was enough. Anyways, DNS wants a very very fine grid. From Turbulent Flows by Pope, DNS simulation has cost Re^(9/4) in each direction. I mean your each direction cells must be proportional to Re^(9/4) for capturing a DNS flow. This will make the computation extremely heavy. Regards, SJ