[Freeman]

Dear all

I am studying rather basic problems of heat transfer though pipes with different sand roughness (ks) sizes.

In my models I have a circular pipe of Ø11mm, a mass flow of helium of 30g/s entering at 300°C at 8 MPa and an homogeneous heat flux in the pipe of 0.5MW/m². I did a parametric analysis with different roughness sizes, ranging from smooth channel to a ks=400µm.

Taking the most extreme cases of smooth and rough pipe with 400µm, I wanted to know the flow velocity at the outlet of the pipe. For this I did

smooth pipe
massFlowAve(Velocity)@outlet = 51.30m/s
areaAve(Velocity)@outlet = 49.66m/s

rough pipe 400µm
massFlowAve(Velocity)@outlet = 51.74m/s
areaAve(Velocity)@outlet = 49.63m/s

Which evaluation is more accurate? the one with the areaAve sounds the one to take, as it matches the value that one obtains by doing a simple calculation taking into account the density, area and mass flow of the fluid at the outlet. However, I have read in the CFX theory manual that the boundaries are virtually moved the half of the roughness height in channels with rough walls for the near-wall treatment. Therefore I am not sure if the roughness makes an influence on the flow speed due to this virtual shift of the wall, or if this shift is only used as an internal operation to make the calculations for the near-wall treatment.

thanks!

[Gert-Jan]

If you use the mass flow average, the variable that you average is weighted by the massflow through each element. So if you take the mass flow average of velocity, you are weighing the velocity with velocity*density. If density is constant, you are weighing velocity with velocity. Doesn't sound right. Therefore use area average.

[ghorrocks]

Gert-Jan, that logic appears flawed. If the density is constant then there is no weighting and the mass flow average is the same as the area average. They are only different if the density is not constant, and in this case you should use massflow average to weight it with variable density.

In short, do you want the integral of density*normal velocity or the integral of the normal velocity? The first one requires a massflow integral, the second area integrals. But the first one is what you want most of the time.

[Gert-Jan]

Maybe I'm wrong, but I have performed a calculation with plain water. I get different results for the areaAve and massFlowAve of Velocity at the outlet. So, the results are not the same.

The massFlowAve weighs each variable by the massflow per element (see CFX-help). In my case, the constant Density cancels out, so only weighing by Velocity remains. Therefore, weighing Velocity by Velocity, results in a (unwanted) bias to higher velocities..........

For this reason, I use the areaAve of the velocity for an average velocity. For averaged temperature, pressures, or whatsoever, I always take massFlowAve-results.

[Freeman]

Dear Gert-Jan and Glenn,

thanks for your answers. To your points:

In short, do you want the integral of density*normal velocity or the integral of the normal velocity? The first one requires a massflow integral, the second area integrals. But the first one is what you want most of the time.

as you say, it is the density*normal velocity what I am interested in, so I guess the massFlowAve is the way to go. I didnt say before, but I am using real gas material properties (I defined them on my own) and as the tube is subjected to a high heat flux, the temperature difference between the bulk and the wall is significant and it may be the reason for the difference between areaAve (not sensitive to material properties) and massFlowAve: is that correct?

For this reason, I always take the areaAve of the velocity, if I want an average velocity. For averaged temperature, pressures, or whatsoever, I always take massFlowAve-results.

Gert-Jan, are the material properties of the water variable in your simulations?

In any case, a couple of questions that arose from this conversation also:

• what is doing CFX in the case of a rough pipe? I have read in the CFX theory manual that the walls are virtually moved half of the roughness height in channels for the near-wall treatment. Is the roughness really partially "blocking" the flow by reducing the pipe cross section by half of the roughness height or this wall shift is only used for the near wall calculation?
• what is happening to the mesh cells below the roughness height? or should the first layer thickness be larger than the roughness height because otherwise has no sense (but then the Y+ will be significantly larger)?

thanks again

[Opaque]

Perhaps we are missing the point of massFlowAve(quantity)@locator.


massFlowAve(quantity) = massFlowInt(quantity)@locator / massFlow()@locator


It is a convenient quantity to refer to something more important, the total amount of an advected quantity, i.e.

quantity = 1 --> total mass flow

quantity = velocity --> total linear momentum flow

quantity = total enthalpy --> total energy flow

quantity = mass fraction --> total mass flow of species k

I so far never seen any reason to use massFlowAve/Int unless I am working with conserved quantities.

For example, I would not use massFlowAve/Int(Pressure)@locator. Not sure what that means in practice. However, areaAve/Int(Pressure)@locator is the force at the location.

Total Pressure is a different issue since the interpretation, via Bernoulli, is similar to energy flow (incompressible flow of course).

[Gert-Jan]

Originally Posted by Freeman

Dear Gert-Jan and Glenn,

Gert-Jan, are the material properties of the water variable in your simulations?

No. Density is constant. It is just a pressure drop calculation for cold water.

Originally Posted by Freeman

Dear Gert-Jan and Glenn,

In any case, a couple of questions that arose from this conversation also:

• what is doing CFX in the case of a rough pipe? I have read in the CFX theory manual that the walls are virtually moved half of the roughness height in channels for the near-wall treatment. Is the roughness really partially "blocking" the flow by reducing the pipe cross section by half of the roughness height or this wall shift is only used for the near wall calculation?
• what is happening to the mesh cells below the roughness height? or should the first layer thickness be larger than the roughness height because otherwise has no sense (but then the Y+ will be significantly larger)?

thanks again

I have little experience with roughness. I would ask ANSYS Support.

[ghorrocks]

what is doing CFX in the case of a rough pipe? I have read in the CFX theory manual that the walls are virtually moved half of the roughness height in channels for the near-wall treatment. Is the roughness really partially "blocking" the flow by reducing the pipe cross section by half of the roughness height or this wall shift is only used for the near wall calculation?

It treats the boundary layer calculation as if the first element is half the roughness height, but the momentum and mass equations still use the wall at the modelled location. There is no effect on cross section area.

what is happening to the mesh cells below the roughness height? or should the first layer thickness be larger than the roughness height because otherwise has no sense (but then the Y+ will be significantly larger)?

Nothing special happens to those cells, the roughness model gets applied to them as well. But the issue is that when the mesh gets smaller than the roughness height it is not an appropriate model as the roughness now extends over multiple elements. So you are applying an invalid model in your simulation and that is not recommended.

[ghorrocks]

what is doing CFX in the case of a rough pipe? I have read in the CFX theory manual that the walls are virtually moved half of the roughness height in channels for the near-wall treatment. Is the roughness really partially "blocking" the flow by reducing the pipe cross section by half of the roughness height or this wall shift is only used for the near wall calculation?

It treats the boundary layer calculation as if the first element is half the roughness height, but the momentum and mass equations still use the wall at the modelled location. There is no effect on cross section area.

what is happening to the mesh cells below the roughness height? or should the first layer thickness be larger than the roughness height because otherwise has no sense (but then the Y+ will be significantly larger)?

Nothing special happens to those cells, the roughness model gets applied to them as well. But the issue is that when the mesh gets smaller than the roughness height it is not an appropriate model as the roughness now extends over multiple elements. So you are applying an invalid model in your simulation and that is not recommended.

[Freeman]

Originally Posted by ghorrocks

It treats the boundary layer calculation as if the first element is half the roughness height, but the momentum and mass equations still use the wall at the modelled location. There is no effect on cross section area.



Nothing special happens to those cells, the roughness model gets applied to them as well. But the issue is that when the mesh gets smaller than the roughness height it is not an appropriate model as the roughness now extends over multiple elements. So you are applying an invalid model in your simulation and that is not recommended.

perfect, thanks for the explanation, it was chrystal clear!

Just FYI, I recently got to know from ANSYS support that since version 18 CFX has implemented a blending function bridging the viscous sublayer and the log-law without the need for virtual wall shifts and similar "tricks". This allows (and even forces) the user to model the boundary layer with full resolution down to Y+~1, regardless of the roughness height.

For this, one must enable the beta feature "Blending for near wall treatment (Beta)" in the fluid model. This feature will appear only if the wall is defined as "smooth" or "high roughness (icing)". I did a couple of checks with high roughness again and now the Y+ came back indeed to typical values even if he first thickness lays below the roughness height. I observed that the heat transfer is slightly enhanced with respect to a calculation without this blending function and results correlate better with the available correlations for fully rough regime.