There are so many possibilities to get differences between experimental and numerical measurements we can write pages about that...

You could have set different boundary conditions, difference in the geometry, wrong model assumptions, not enough grid resolution, lack of full convergence, etc.

But what if your numerical solution is acceptable and your experimental measures are affected by errors?

Thus, I strongly suggest to provide the full details of your modelling, set-up and exerimental devices.

Dear Filippo,

Thanks for your answer. As you already mentioned, there are various source of errors that can lead to inconsistencies between numerical and experimental results.

I'm quite sure about the experimental results.

If you look at the geometry, the issue is the interface between fluid and internal structure, in which I defined a coupling interface between shared boundaries in Fluent. What else should I do to have a good definition for this interface?

The first think to check is if the boundary conditions of the numerical model matches those of the experimental one. This is likely to be the most common problem. Can you prescribe the same inlet and outlet conditions of your experiment

Are you assuming some further assumptions like steady turbulence and using RANS? The grid resolution is sufficient?

Finally, I suggest also to ask in the Fluent section of the cfd-online forum.

I wonder if you have simulated enough long pipe in order to be sure that the flow is fully developed.

Best regards

Dear Filippo & Mounir,

As Mounir also mentioned, our experimental needs to be done with a larger tube length, because of our measuring instruments accuracies. Since, the measured pressure drop was less than 1 mmHg.

Thanks for your help.

You can adjust the wall roughness. Also properties of the liquid.

Boundary conditions.

Head change during the experiment. etc.. ,matters

Dear Mohammad,

again you did not addredd any useful information about your simulation. This way is useless to immagine any answer.

Dear Filippo,

The inlet condition for experimental was velocity inlet (laminar regime) with inlet pressure as well, so I defined the same condition for simulation and put the pressure outlet boundary condition. The mesh resolution is fine enough (about 1 million).

Dear Jubin,

Our flowing fluid is water, so the properties is well known.

The BCs were defined similar to the experimental setup.

How do you ensure that the flow inlet in the experiment is really a parabolic laminar profile? And what about the Reynolds number at the inlet? The flow around these bluff bodies becomes separated and eventually turbulent. The grid resolution depends strongly on what formulation you are using RANS/LES/DNS and type of boundary conditions setting at the walls. If you are using RANS it will be unable to do the representation of the transition from the laminar inflow to the separated flow. If you use LES/DNS the 1 million-node grid seems too coarse

I don't know if you have experience in CFD but your answers are not sufficient to understand your flow problem

The Reynolds number is around 1500 and the inlet pressure is around 8000 Pa, so it's a laminar flow at the inlet. However, the velocity increased around the internal structure boundaries, see the pictures (velocity and pressure distributions along a middle plane).

I used laminar model, should I change that by other models, like LES/DNS models?

For the walls, I used no-slip condition.

First, I would better understand the flow you have in the internal structure. Does it pass around or through or both? I would understand if separation occurs and if the flow can become transitional and eventually turbulent.

From the velocity field it is not possible to understand. Then, I see in the solution the onset of oscillations, is your solution at a steady state? What about a grid convergence analysis? Have you checked if your grid is enough fine to solve the boundary layer? Otherwise the no-slip conditions are useless.

If you look at the geometry, the flow goes both around and through (from six small holes) the structure, but mainly around it.

The solution is solved under steady state condition. How can I check if the grid size can solve the boundary layer?

Dear,

at First I agree with Mr. Fillipo. He put excelent questions. Before one month I have laboratory experiment. Part of them is flow throught tube (in. diam 17 mm). Re number was small (800). Flow is not laminar!. In my case it is transitional. At this small velocities of water natural convection play very important role. Because of this appear swirl flow.

See literature about it (Metai, Eckert (1964), Depew, C. A., August S. E., Heat Transfer due to Combined Free and Forced Convection in a Horizontal and Isothermal Tube, Trans. ASME. Ser. C. J. Heat Transfer, vol. 93, pp. 380–384, 1971. )

Difference between laminar and swirl flow is twice ...

Good luck

The resolution of the boundary layer is (for wall bounded flow) dictated by the value of y+. A grid resolving the BL has at least 3-4 nodes within y+

.... my comment was for heat exchanged...

This is always connected with pressure drop ...

Yes I agree, however we have no details to know if heat transfer has some role in this problem

... usualy experimental pressure drop is higher than numericaly obtained.

In your case is oposite?

In velocity plane is visible swirl. This mean greater pressure drop than for laminar flow. This back problem at begining. Maybe problem is in measurement of pressure drop.

... or in wrong model assumptions.

Dear Filippo and Mladen,

Sorry for late reply, I was stucked with the simulation.

The experimental results are higher than numerical, almost twice. My aplogies, since I mistakenly wrote in the question that experimental is lower than numerical!

I have tried to run the problem with LES model with around 2 millions cells, but still the same results.

The tube size is around 120mm with inner diameter of 9.5mm.

Well, twice is quite high difference that makes me suspect your bc.s do not match correctly the experiment. Have you tried fixing at the inlet the thermodinamic properties, the total temperature and let free the stream-wise velocity?

My model doesn't involve the temperature, do I have to include that?

How can I do the "free stream-wise velocity" simulation?

I was thinking to a full compressible model that you can run at low Mach number. The set of variables in inlet could be used without prescribing the inlet profile as a Dirichlet condition but letting it developping from the flow solution. However, you need to prescribe the energy condition at the inlet, for example total pressure or total temperature that match your experiment.

Put in numerical model same parameters as in experiment (inlet pressure, temperature). Take care that tube inlet in experimental model is geometricaly identical as in numerical model.

Inlet has room temperature with a pressure of 8500 Pa. Why use compressible flow?

For low Re numbers (Re=1500) it is not suprise to obtain twice greater pressure. In my simulation (k-o SST turbulence model) for Re=2000 pressure drop in experiment is 80% greater.

What is difference in your results for LES and previous model?

I obtained almost the same results for LES and the previous laminar model.

However, when I run the simulation for tube only (without internal structure), the obtained pressure drop is very close to the following formula:

delta_P = 64 * L * Ro * v^2/(2 * Re * D)

with around 15% error.