[lzhaok6]

Dear all,

I understand the effect of non-reflective boundary (NRB) when it is used to prevent the outgoing wave from reflection. However, I got confused when it is used for inlet condition for incoming waves. Let's suppose the field variable is velocity potential and the governing equation is the linear acoustic equation to be spatially discretized by FEM. In order to let the acoustic wave pass from the exterior into the computational domain, the normal fluid particle velocity is generally prescribed on the boundary superposing the outgoing wave NRB. The normal-to-boundary velocity component is required by the discretization of the FEM (divergence theorem). At the same time, the tangential part of the velocity is not prescribed on the boundary. By doing that, would the oblique wave (propagation direction is not normal to the boundary) be distorted by the NRB? If the NRB is not there (which is the situation we want to simulate), both the normal and tangential particle velocity would be preserved. If the distortion does happen, is there any way to circumvent that?

Thanks for the attention!

[LuckyTran]

I don't quite follow what the issue is...

First you have the non-reflecting boundary (usually in velocity). Ok. Good.

To have incoming waves, you still do the non-reflecting part, but now you add an additional normal velocity fluctuation on-top of the non-reflecting part (super position). That's it.

Traditionally, you only touch the normal velocity and the tangential velocities are undisturbed. Btw we don't really do this on purpose, you should also do something to the tangential velocities, but the NRBC's are not well developed.

If you don't have a NRBC, and you impose an incoming normal velocity, then you get reflections of the outgoing wave.

[FMDenaro]

First, that issue applies for compressible flows. Only in case of a subsonic inflow you get into the problem of a wave coming from the interior to the inflow. What variable at the inflow must be let free to adapt its value is an object of several proposal. However, I suggest to work with energy variable.
The simple case of the linear acustic problem has an exact solution and you can see how to work with in the book of Leveque.

[lzhaok6]

Originally Posted by LuckyTran

I don't quite follow what the issue is...

First you have the non-reflecting boundary (usually in velocity). Ok. Good.

To have incoming waves, you still do the non-reflecting part, but now you add an additional normal velocity fluctuation on-top of the non-reflecting part (super position). That's it.

Traditionally, you only touch the normal velocity and the tangential velocities are undisturbed. Btw we don't really do this on purpose, you should also do something to the tangential velocities, but the NRBC's are not well developed.

If you don't have a NRBC, and you impose an incoming normal velocity, then you get reflections of the outgoing wave.

Thanks for clarifying the question and pointing out that the tangential velocity is an imperfect part of current NRBC. My real doubt is actually how to deal with that tangential part of the velocities. Is there any general way to add that tangential disturbance on to the boundary?

[lzhaok6]

Originally Posted by FMDenaro

First, that issue applies for compressible flows. Only in the case of a subsonic inflow you get into the problem of a wave coming from the interior to the inflow. What variable at the inflow must be let free to adapt its value is an object of several proposals. However, I suggest to work with energy variable.
The simple case of the linear acoustic problem has an exact solution and you can see how to work within the book of Leveque.

Thanks for your reply! The wave source is actually outside the domain. So it's a problem of the exterior flow as inflow to the domain.

[LuckyTran]

Originally Posted by lzhaok6

Thanks for clarifying the question and pointing out that the tangential velocity is an imperfect part of current NRBC. My real doubt is actually how to deal with that tangential part of the velocities. Is there any general way to add that tangential disturbance on to the boundary?

If you are so worried about the tangential components of the waves, you really should not be putting the boundary there! That would be my recommendation.

Basically, current state-of-the-art NRBC is developed assuming 1D plane waves. I'm not aware of any general method or advances beyond this. You might find something has been done out there, but at least I'm not aware of it.

A general method would have to extend 1D plane-wave formulation to non-normal plane waves and then other wavefronts such as spherical waves, and this is not straightforward. Unless you are specifically interesting in developing such techniques, I would take it as-is. The NRBC is already much better than nothing.

Besides, the purpose of the NRBC is to prevent the build-up of acoustic energy in the domain and let you get reasonable results. If you want to kill all reflections, you would have to design your simulation specifically for this purpose using the so-called sponge layer approach to absorb acoustics in special regions. That is, you construct the numerical equivalent of an experimental anechoic chamber.

One reason it is hard to find a general way forward is that, even for 1D plane waves, the non-reflecting boundary condition in time domain isn't perfectly non-reflecting. You struggle between imposing the mean condition and the non-reflecting condition. The more you are closer to the mean condition, the more reflecting the boundary is. And the more you are non-reflecting the more deviation you are from your mean condition. It works well for high-frequencies but struggles with low frequencies. And this is only for the 1D plane wave!

[lzhaok6]

Originally Posted by LuckyTran

If you are so worried about the tangential components of the waves, you really should not be putting the boundary there! That would be my recommendation.

Basically, current state-of-the-art NRBC is developed assuming 1D plane waves. I'm not aware of any general method or advances beyond this. You might find something has been done out there, but at least I'm not aware of it.

A general method would have to extend 1D plane-wave formulation to non-normal plane waves and then other wavefronts such as spherical waves, and this is not straightforward. Unless you are specifically interesting in developing such techniques, I would take it as-is. The NRBC is already much better than nothing.

Besides, the purpose of the NRBC is to prevent the build-up of acoustic energy in the domain and let you get reasonable results. If you want to kill all reflections, you would have to design your simulation specifically for this purpose using the so-called sponge layer approach to absorb acoustics in special regions. That is, you construct the numerical equivalent of an experimental anechoic chamber.

One reason it is hard to find a general way forward is that, even for 1D plane waves, the non-reflecting boundary condition in time domain isn't perfectly non-reflecting. You struggle between imposing the mean condition and the non-reflecting condition. The more you are closer to the mean condition, the more reflecting the boundary is. And the more you are non-reflecting the more deviation you are from your mean condition. It works well for high-frequencies but struggles with low frequencies. And this is only for the 1D plane wave!

"If you are so worried about the tangential components of the waves, you really should not be putting the boundary there!"

That makes a lot of sense! I suppose the right way to do it would be put NRB fairly away from the source so that the wave becomes fairly planar when it reaches the boundary.

[Sanda]

Originally Posted by LuckyTran

If you are so worried about the tangential components of the waves, you really should not be putting the boundary there! That would be my recommendation.

Basically, current state-of-the-art NRBC is developed assuming 1D plane waves. I'm not aware of any general method or advances beyond this. You might find something has been done out there, but at least I'm not aware of it.

A general method would have to extend 1D plane-wave formulation to non-normal plane waves and then other wavefronts such as spherical waves, and this is not straightforward. Unless you are specifically interesting in developing such techniques, I would take it as-is. The NRBC is already much better than nothing.

Besides, the purpose of the NRBC is to prevent the build-up of acoustic energy in the domain and let you get reasonable results. If you want to kill all reflections, you would have to design your simulation specifically for this purpose using the so-called sponge layer approach to absorb acoustics in special regions. That is, you construct the numerical equivalent of an experimental anechoic chamber.

One reason it is hard to find a general way forward is that, even for 1D plane waves, the non-reflecting boundary condition in time domain isn't perfectly non-reflecting. You struggle between imposing the mean condition and the non-reflecting condition. The more you are closer to the mean condition, the more reflecting the boundary is. And the more you are non-reflecting the more deviation you are from your mean condition. It works well for high-frequencies but struggles with low frequencies. And this is only for the 1D plane wave!

Dear LuckyTran,

Can you please explain to me regarding NRBC? Is it used in ANSYS fluent?
I am looking for NRBC which can be used in pressure outlet of 2way FSI in ANSYS.
It may be the old thread but I want to follow up the discussion.

Best Regards,

[LuckyTran]

Fluent has an nrbc for a pressure outlet also based on the LODI approach.


The discussion in this thread is for inlets, which is similar but inlet also come with forcing.

[Sanda]

I already posted the discussion for NRBC for pressure outlet.
Could you please kindly show how to apply it?

Reflection Free Boundary Condition

[FMDenaro]

Originally Posted by Sanda

I already posted the discussion for NRBC for pressure outlet.
Could you please kindly show how to apply it?

Reflection Free Boundary Condition

But your previous post was for incompressible flows, no matter about pressure wave propagation, they are at infinit speed.