What is the advantage of LES over SAS?

Time4Tea · November 25, 2019, 12:02

Hi, I have recently been exploring the use of scale-resolving turbulence models for predicting incompressible internal flows through control valves. In particular, I have mainly been using the SAS turbulence model in FLUENT. It seems to give significantly better results than a simple RANS model, whilst also being quite robust and I like that way that you can 'target' it by only applying a fine scale-resolving mesh to the region of the model where you need to resolve the large-scale turbulent eddies, which makes the simulation more efficient.

However, from what I can see, it seems to be relatively uncommon to see it used in academic research, which seem to generally prefer LES/DES or other methods.

So, I am wondering, what is the advantage of using LES/DES over SAS? Is SAS generally considered to be less academically rigorous than those other scale-resolving models?

FMDenaro · November 25, 2019, 14:24

Quote:

Originally Posted by Time4Tea

Hi, I have recently been exploring the use of scale-resolving turbulence models for predicting incompressible internal flows through control valves. In particular, I have mainly been using the SAS turbulence model in FLUENT. It seems to give significantly better results than a simple RANS model, whilst also being quite robust and I like that way that you can 'target' it by only applying a fine scale-resolving mesh to the region of the model where you need to resolve the large-scale turbulent eddies, which makes the simulation more efficient.

However, from what I can see, it seems to be relatively uncommon to see it used in academic research, which seem to generally prefer LES/DES or other methods.

So, I am wondering, what is the advantage of using LES/DES over SAS? Is SAS generally considered to be less academically rigorous than those other scale-resolving models?

SAS is somehow a DES approach wherein the LES filtering is not based directly on the computational grid. In academic research we work on basic aspects of turbulence and DNS and pure LES are more common.
Have a look here
https://www.google.com/url?sa=t&rct=...OD3RsbKLmYAcOO

LuckyTran · November 25, 2019, 14:28

LES is a spatially resolved (up to the filter width) and temporally resolved approach. Supposing that the LES is run on a super duper fine grid, it is almost as good as a DNS.

The SAS model is still a (U)RANS model. You can run SAS on a infinitely fine grid and it'll still act like a (U)RANS. At the end of the day, it is a slightly better RANS turbulence model.

Depending on what literature you are consuming... SAS is relatively new. It was only introduced into commercial codes recently (after LES). Not that many people use it. Even moreso, SAS fills a very small niche. When you are okay with RANS you'd typically go RANS. When RANS models are no good and you need to brute force the turbulence, you'd go LES. In my opinion, the best uses of SAS models is for visualizing large scale structures, like wingtip vortices. SAS model does this much more accurately than traditional RANS models and LES is not needed for this problem.

At least, this is true for the SST-based SAS model. Maybe in the future someone will develop a more advanced SAS approach that can mimick LES better. This would make it a much better hybrid LES/RANS, similar to "DES". Then we'll probably drop one of the two names, between SAS/DES and keep only one.

Time4Tea · November 25, 2019, 17:00

@LuckyTran: my personal experience with SAS doesn't really jive with your description of it being "a slightly better RANS model". I have been using SAS for around 3-4 months now to analyze flow capacity in industrial control valves and I have found it to compare very well with the results of lab flow tests (within ~5%). Previously, I was using RANS, which seemed to be very hit and miss - SAS seems to perform much better.

Also, the results shown for the SAS model in the document that FMDenaro linked seem to show it performing well. For example, on page 11 it seems to compare very closely to LES for decay of isotropic turbulence. It also seems to give good results in the industrial application examples that are shown (i.e. the airfoil Cp plot on p.13).

Tbh, (as an Engineer in industry) I can't see why I would want to go to the extra trouble of meshing and running LES, given the performance I am seeing from SAS.

This is why I am asking the question: I am curious as to what further benefit LES could give over SAS and why SAS is not used more in academic research. Perhaps it is, and it is my perception that is wrong. Are there any particular examples or situations you know of where SAS is known to be inferior to LES?

FMDenaro · November 25, 2019, 17:06

Quote:

Originally Posted by Time4Tea

@LuckyTran: my personal experience with SAS doesn't really jive with your description of it being "a slightly better RANS model". I have been using SAS for around 3-4 months now to analyze flow capacity in industrial control valves and I have found it to compare very well with the results of lab flow tests (within ~5%). Previously, I was using RANS, which seemed to be very hit and miss - SAS seems to perform much better.

Also, the results shown for the SAS model in the document that FMDenaro linked seem to show it performing well. For example, on page 11 it seems to compare very closely to LES for decay of isotropic turbulence. It also seems to give good results in the industrial application examples that are shown (i.e. the airfoil Cp plot on p.13).

Tbh, (as an Engineer in industry) I can't see why I would want to go to the extra trouble of meshing and running LES, given the performance I am seeing from SAS.

That is why I am curious as to why SAS is not used more in academic research. Perhaps it is, and it is my perception that is wrong. Are there any particular examples or situations you know of where SAS is known to be inferior to LES?

The main reason is that SAS (as same as DES) is based on a RANS formulation close to the body. In academic research that is not useful as all the details and feature of wall turbulence is only modelled and no useful exploration of the physics can be deduced. Generally such formulations are used only for getting funds in specific projects.

Time4Tea · November 25, 2019, 18:06

Quote:

Originally Posted by FMDenaro

The main reason is that SAS (as same as DES) is based on a RANS formulation close to the body. In academic research that is not useful as all the details and feature of wall turbulence is only modelled and no useful exploration of the physics can be deduced. Generally such formulations are used only for getting funds in specific projects.

Ah, ok, thanks. That makes sense. So the SAS model will give an improvement over SST k-omega in terms of the turbulent viscosity in the bulk flow away from the wall, but not close to the wall. So, the accuracy improvement over RANS might be good enough for industrial applications, if they can make do with the RANS-style wall modelling.

I guess that also explains why SAS seems to compare well to LES for the decay of isotropic turbulence in the document you linked - because it looks like there are no walls involved there?

FMDenaro · November 25, 2019, 18:11

Quote:

Originally Posted by Time4Tea

Ah, ok, thanks. That makes sense. So the SAS model will give an improvement over SST k-omega in terms of the turbulent viscosity in the bulk flow away from the wall, but not close to the wall. So, the accuracy improvement over RANS might be good enough for industrial applications, if they can make do with the RANS-style wall modelling.

I guess that also explains why SAS seems to compare well to LES for the decay of isotropic turbulence in the document you linked - because it looks like there are no walls involved there?

Clearly in homogenous and isotropic turbulence you can think of a unique characteristic lenght for the turbulent viscosity, a very simple case to be studied.

sbaffini · November 26, 2019, 07:36

I would not say that SAS is in the same arena as LES and DES.

The latter are based on a length scale that is either user prescribed (in the very uncommon scenario of explicitly filtered LES) or (more commonly) linked to the underlying grid.

In contrast, SAS is based on a "physically" prescribed length scale (which is based on derivatives that, in the end, depend on the grid). So, in theory, SAS belongs to the URANS arena, and indeed it can do much better than URANS.

Comparison to RANS is definitely no sense or, however, not fair, because RANS is steady.

How sound is the SAS approach, I guess, is matter of debate. Also, the intsability triggering mechanism behind SAS is much more complex than that in DES and LES, so understanding the results is clearly a daunting task.

My experience is that there are flows that, for a given coarse mesh, might be better represented in LES than SAS (or URANS and DES), so the overall cost would be lower in LES. Of course, as everything in CFD, you need to know what you're doing and make tests/benchmarks for your use case.

November 25, 2019, 12:02	What is the advantage of LES over SAS?	#1
Time4Tea Senior Member Lee Strobel Join Date: Jun 2016 Posts: 133 Rep Power: 10	Hi, I have recently been exploring the use of scale-resolving turbulence models for predicting incompressible internal flows through control valves. In particular, I have mainly been using the SAS turbulence model in FLUENT. It seems to give significantly better results than a simple RANS model, whilst also being quite robust and I like that way that you can 'target' it by only applying a fine scale-resolving mesh to the region of the model where you need to resolve the large-scale turbulent eddies, which makes the simulation more efficient. However, from what I can see, it seems to be relatively uncommon to see it used in academic research, which seem to generally prefer LES/DES or other methods. So, I am wondering, what is the advantage of using LES/DES over SAS? Is SAS generally considered to be less academically rigorous than those other scale-resolving models?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Evaluation of length scale ratio for LES, DES, SAS etc.	siw	CFX	12	May 19, 2016 03:25
Grid difference between wall modeled LES and wall resolved LES	hityangsir	Main CFD Forum	9	April 17, 2015 12:59
OpenFOAM LES capability questions	siw	OpenFOAM Running, Solving & CFD	3	January 6, 2015 09:50
Turbulence dampening due to magnetic field in LES and RAS	eelcovv	OpenFOAM	0	June 8, 2010 12:35
Some Questions about LES.	Bin Li	Main CFD Forum	2	February 20, 2004 10:58

November 25, 2019, 14:28		#3
LuckyTran Senior Member Lucky Join Date: Apr 2011 Location: Orlando, FL USA Posts: 5,754 Rep Power: 66	LES is a spatially resolved (up to the filter width) and temporally resolved approach. Supposing that the LES is run on a super duper fine grid, it is almost as good as a DNS. The SAS model is still a (U)RANS model. You can run SAS on a infinitely fine grid and it'll still act like a (U)RANS. At the end of the day, it is a slightly better RANS turbulence model. Depending on what literature you are consuming... SAS is relatively new. It was only introduced into commercial codes recently (after LES). Not that many people use it. Even moreso, SAS fills a very small niche. When you are okay with RANS you'd typically go RANS. When RANS models are no good and you need to brute force the turbulence, you'd go LES. In my opinion, the best uses of SAS models is for visualizing large scale structures, like wingtip vortices. SAS model does this much more accurately than traditional RANS models and LES is not needed for this problem. At least, this is true for the SST-based SAS model. Maybe in the future someone will develop a more advanced SAS approach that can mimick LES better. This would make it a much better hybrid LES/RANS, similar to "DES". Then we'll probably drop one of the two names, between SAS/DES and keep only one.

November 25, 2019, 17:00		#4
Time4Tea Senior Member Lee Strobel Join Date: Jun 2016 Posts: 133 Rep Power: 10	@LuckyTran: my personal experience with SAS doesn't really jive with your description of it being "a slightly better RANS model". I have been using SAS for around 3-4 months now to analyze flow capacity in industrial control valves and I have found it to compare very well with the results of lab flow tests (within ~5%). Previously, I was using RANS, which seemed to be very hit and miss - SAS seems to perform much better. Also, the results shown for the SAS model in the document that FMDenaro linked seem to show it performing well. For example, on page 11 it seems to compare very closely to LES for decay of isotropic turbulence. It also seems to give good results in the industrial application examples that are shown (i.e. the airfoil Cp plot on p.13). Tbh, (as an Engineer in industry) I can't see why I would want to go to the extra trouble of meshing and running LES, given the performance I am seeing from SAS. This is why I am asking the question: I am curious as to what further benefit LES could give over SAS and why SAS is not used more in academic research. Perhaps it is, and it is my perception that is wrong. Are there any particular examples or situations you know of where SAS is known to be inferior to LES?

November 26, 2019, 07:36		#8
sbaffini Senior Member Paolo Lampitella Join Date: Mar 2009 Location: Italy Posts: 2,192 Blog Entries: 29 Rep Power: 39	I would not say that SAS is in the same arena as LES and DES. The latter are based on a length scale that is either user prescribed (in the very uncommon scenario of explicitly filtered LES) or (more commonly) linked to the underlying grid. In contrast, SAS is based on a "physically" prescribed length scale (which is based on derivatives that, in the end, depend on the grid). So, in theory, SAS belongs to the URANS arena, and indeed it can do much better than URANS. Comparison to RANS is definitely no sense or, however, not fair, because RANS is steady. How sound is the SAS approach, I guess, is matter of debate. Also, the intsability triggering mechanism behind SAS is much more complex than that in DES and LES, so understanding the results is clearly a daunting task. My experience is that there are flows that, for a given coarse mesh, might be better represented in LES than SAS (or URANS and DES), so the overall cost would be lower in LES. Of course, as everything in CFD, you need to know what you're doing and make tests/benchmarks for your use case.