[julieng]

Hello,


I am interested in calculating the heat exchange between solid and fluid. Most of the time, I have to estimate the heating time of a cold solid with hot air (internal forced convection with fans in ovens).


So I start with 2D laminar cases, only few question remains without answer now for simple cases. CHT cases pre and post processing is ok for me now, I can go to the next step: turbulence.


In the CHT tutorials, all turbulent cases are kEpsilon. So I try this first.


First, I try to estimate turbulence parameters according to my readings:
Tell me if I am wrong, I am totally new

- Turbulent intensity:2 or 5% http://www.cfd-online.com/Tools/turbulence.php
- Turbulent lenght scale = 0.0038 x Dh (Dh the hydraulic diameter)
- u' = U0 x Turbulent Intensity (U0: velocity away from wall) [m/s]

- k = 3/2 x u'^2 [m2/s2]

- e = 0.09^(3/4) x k^(3/2) / Turbulence lenght scale
- w = e / (0.09 x k)
Understood that it is not so important, only initial conditions



Bloerb still indicates the way to follow for accurate heat transfert in this post:

Heat transfer problem


The first question is about the wall functions:

I read that I need fine meshes close to walls y+ close to 1.

Do I have to use wall functions with a so small mesh ? (kEpsilon is usually only used with wall functions no?)

Of course, I am interested in low computation times, so wall functions could be very helpfull, but do I have the choice ? What is your feedback about that ?
I need a robust turbulence model with a medium/10-20% accuracy. What is your recommendation ?



Bloerb thinks that komegaSST model is a good choice. Whith or without wall functions? max y+? Which ones ? Please give me an example?


Sorry for the number of questions but I am really interested in this subject.


Best regards



Julien

[dasith0001]

Hi,

Just wondering if you have found some clarity with the issue or anyone has a good explanation to these questions?

Thank you

[alexvaleije]

Hi,

As far as my knowledge go, there are two ways of solving the flow near the walls in terms of velocity:

The first is to solve the flow inside the viscous sublayer, for which you need to have small y+ values, typically lower than 8.

The second approach is solving the first element using wall functions, which typically require y+ values between 30 and 300 for a good accuracy.

If your first element falls in the buffer layer (y+ values between 10 and 30) the wall function won't adjust as well as the other paths, and you will might add some errors in your model. In my experience, falling in this layer usually reproduces results with less pressure drops than being in the other two regions, so it can be a bit dangeruous if you are dimensioning equipment, for example, but you can try to reproduce it in your own field.

However, if you need to capture thermal effects, the first approach is usually the better choice, since you will need to recreate also the thermal gradients between wall and flow. If your gradients are not too relevant or not too high, you might find in the second path a good companion, since it will require much less computational effort to recreate.

Hope this helps a bit,
Regards

Originally Posted by dasith0001

Hi,

Just wondering if you have found some clarity with the issue or anyone has a good explanation to these questions?

Thank you

[dasith0001]

Originally Posted by alexvaleije

Hi,

As far as my knowledge go, there are two ways of solving the flow near the walls in terms of velocity:

The first is to solve the flow inside the viscous sublayer, for which you need to have small y+ values, typically lower than 8.

The second approach is solving the first element using wall functions, which typically require y+ values between 30 and 300 for a good accuracy.

If your first element falls in the buffer layer (y+ values between 10 and 30) the wall function won't adjust as well as the other paths, and you will might add some errors in your model. In my experience, falling in this layer usually reproduces results with less pressure drops than being in the other two regions, so it can be a bit dangeruous if you are dimensioning equipment, for example, but you can try to reproduce it in your own field.

However, if you need to capture thermal effects, the first approach is usually the better choice, since you will need to recreate also the thermal gradients between wall and flow. If your gradients are not too relevant or not too high, you might find in the second path a good companion, since it will require much less computational effort to recreate.

Hope this helps a bit,
Regards

Hi,

Thank you for your descriptive answer; I think I have no option but to use the second option by using a 'wall function'.

One quation though, how high is a thermal gradient categorised as a' high gradient' ?

My model basically dictates by natural convection heat transfer mechanism at high temperatures where the flow have to go through both macro and micro channels; Could you suggest an appropriate turbulence model to in corporate that ?

Thank you
Dasith

[alexvaleije]

Originally Posted by dasith0001

Hi,

Thank you for your descriptive answer; I think I have no option but to use the second option by using a 'wall function'.

One quation though, how high is a thermal gradient categorised as a' high gradient' ?

My model basically dictates by natural convection heat transfer mechanism at high temperatures where the flow have to go through both macro and micro channels; Could you suggest an appropriate turbulence model to in corporate that ?

Thank you
Dasith

Hio Dasith,

To characterize whether your model requires for a turbluence model or it has a laminar flow, you should use the Reynolds number. Since you say you have both micro and micro channels, I would recommend to calculate the Reynolds number based on the characteristic lengths on both sides, and see in which type of flow your are in both of them. Depending on the type of flow, you could use either a laminar model or a turbulent one. In case your flow is turbulent, I usually have good results with a SST k-w model.

Another way to characterize your model is with the Rayleigh number. For very high numbers (>1e8) the flow is usually turbulent.

Once you know in which regime you are, you can try to stimate the length of both your velocity and thermal boundary layers. You can get some basic information about this topic just by looking at Wikipedia:

https://en.wikipedia.org/wiki/Therma...ness_and_shape

Using the 99% layer approach, you can get your boundary layer thickness and, with this value and a reasonable expansion ratio (1.1-1.6) get your first approach in the boundary layer. Once you get this, you can get a first simulation, evaluate your results and y+ and modify your case from that point.

If you find out you are in a laminar regime, you don't really need a boundary layer, but then you will have to check if you capture your velocity profile correctly in the flow bulk.

Hope this info helps you. Regards,

[dasith0001]

Originally Posted by alexvaleije

Hio Dasith,

To characterize whether your model requires for a turbluence model or it has a laminar flow, you should use the Reynolds number. Since you say you have both micro and micro channels, I would recommend to calculate the Reynolds number based on the characteristic lengths on both sides, and see in which type of flow your are in both of them. Depending on the type of flow, you could use either a laminar model or a turbulent one. In case your flow is turbulent, I usually have good results with a SST k-w model.

Another way to characterize your model is with the Rayleigh number. For very high numbers (>1e8) the flow is usually turbulent.

Once you know in which regime you are, you can try to stimate the length of both your velocity and thermal boundary layers. You can get some basic information about this topic just by looking at Wikipedia:

https://en.wikipedia.org/wiki/Therma...ness_and_shape

Using the 99% layer approach, you can get your boundary layer thickness and, with this value and a reasonable expansion ratio (1.1-1.6) get your first approach in the boundary layer. Once you get this, you can get a first simulation, evaluate your results and y+ and modify your case from that point.

If you find out you are in a laminar regime, you don't really need a boundary layer, but then you will have to check if you capture your velocity profile correctly in the flow bulk.

Hope this info helps you. Regards,

This helps heaps, Thank you very much for very descriptive answer. I think I have a clear idea on a starting point now.