[huangxianbei]

Hi,all
I'm a little confused about the difference between steady and transient solution.As we all know, the transient N-S and Reynolds time-averaged equations both contain the term corresponding to the time(t). Usually, we depart the value such as Velocity U into Uave(time average) and u(the fluctuation).

Q1:Is that means the steady state solution represents the Uave? While only the transient solution can get u. Does there any difference in the equations been solved?

Q2:If it's required to specify the Pressure gradient force(ave) in a transient simulation, how can we implement this? Through momentum source or some way else?

[ghorrocks]

Good question

For a steady state run the Reynolds averaging assumes the flow has an averaged component and a time varying component, where the time varying component averages over time to zero. Then it is obvious to simulate the averaged flow component and use a turbulence model to handle the time varying component.

But in a transient flow with RANS things are a bit more subtle, as the simulated flow also has a time varying component as it is transient. The fundamental assumption being made is that there is a distinction in the spectrum of the turbulent variations and the average flow variations. As long as that distinction can be made and the turbulence model is correctly scaled to that distinction you are OK. But in many flows there is not a clear distinction - for instance IC engine flows, flows off bluff bodies. That is one key reason why RANS models do not do very well in these applications and LES based approaches are often more accurate.

You have implement a momentum source to do whatever you want, but the pressure gradient is already part of the momentum equation so why do you want to include it again?

[huangxianbei]

Originally Posted by ghorrocks

Good question

For a steady state run the Reynolds averaging assumes the flow has an averaged component and a time varying component, where the time varying component averages over time to zero. Then it is obvious to simulate the averaged flow component and use a turbulence model to handle the time varying component.

But in a transient flow with RANS things are a bit more subtle, as the simulated flow also has a time varying component as it is transient. The fundamental assumption being made is that there is a distinction in the spectrum of the turbulent variations and the average flow variations. As long as that distinction can be made and the turbulence model is correctly scaled to that distinction you are OK. But in many flows there is not a clear distinction - for instance IC engine flows, flows off bluff bodies. That is one key reason why RANS models do not do very well in these applications and LES based approaches are often more accurate.

You have implement a momentum source to do whatever you want, but the pressure gradient is already part of the momentum equation so why do you want to include it again?

Hi,ghorrocks:

According to your description, the steady state gets the average solution, ,and the equations been solved have no difference from the transient case, is my understanding right?

As for the second question, well ,I'd like to describe it in details:
First, in the rotating tunnel flow(rectangular intersection), many researches were done with DNS.
In the DNS calculation, the N-S equations under rotating frame were used which including the centrifugal force and Coriolics force terms. Usually, the centrifugal force term is included into the pressure gradient term and the pressure gradient term Grad(P)becomes effective pressure gradient term, Grad(Peff).
In order to drive the flow, the mean effective pressure gradient was used as a constant value, that is Grad(Peff)ave, so I don't know how to implement this.

[ghorrocks]

A better way to think of it is that both approaches use the RANS approach. In the steady state situation what is the avergae flow and what is the turbulent fluctuations is obvious as anything transient is assumed turbulent, but in transient simulation the RANS approach assumes a distinction between turbulent and average flow time scales. In this case some transient flow is assumed avergae flow and some turbulent.

Second question> CFX already has a rotating frame of reference model built in which already has the centripetal and coriolis forces applied. Why not use that?

[huangxianbei]

Originally Posted by ghorrocks

A better way to think of it is that both approaches use the RANS approach. In the steady state situation what is the avergae flow and what is the turbulent fluctuations is obvious as anything transient is assumed turbulent, but in transient simulation the RANS approach assumes a distinction between turbulent and average flow time scales. In this case some transient flow is assumed avergae flow and some turbulent.

Second question> CFX already has a rotating frame of reference model built in which already has the centripetal and coriolis forces applied. Why not use that?

Thank you for your explanation. Well, as for the second question, I tried to use the rotating frame method in which the rotation speed was specified and the momentum source was used in this way:

according to the classical research, the Grad(Peff)ave is specified as a constant as referred before. Just as you say ,in the rotating frame in CFX, the centrifugal force is applied already, so in the momentum source we are going to specify is needed to exclude these effects, but in the N-S equations, the Grad(P) and F(centrifugal) both exist, so I don't know what do the formal researchers mean in the specification of Grad(Peff)ave.

[ghorrocks]

I have no idea what you are talking about. If you are modelling a rotating region then why is the default rotating frame of reference model not suitable? It includes all necessary terms, you should not need to add anything.

[huangxianbei]

Originally Posted by ghorrocks

I have no idea what you are talking about. If you are modelling a rotating region then why is the default rotating frame of reference model not suitable? It includes all necessary terms, you should not need to add anything.

In general,the problem is to set the Grad(Peff)=const, yes, in rotating frame, the Grad(P) and centrifugal force both exist, but I should set the plus ,that is
-Grad(P)+centrifugal force=const. But in fact, this is impossible to guarantee this condition(I think)
So I think a momentum source is needed instead,which can be written like this:
S=const+Grad(P)-centrifugal force
This means to use a momentum source to make sure that the
S-Grad(P)+centrifugal force=const

[ghorrocks]

Are you saying the implementation of rotating frames of reference in CFX is wrong?

[huangxianbei]

Originally Posted by ghorrocks

Are you saying the implementation of rotating frames of reference in CFX is wrong?

No, I agree with you that rotating frame is a simpler way to solve this problem. And the problem I'm facing is how to set
Grad(P)ave + centrifugal force(ave)=const,
as in the steady state, the Grad(P)ave=Grad(P), centrifugal force(ave)=centrifugal force
so the problem turns to how to set
Grad(P) + centrifugal force=const

[huangxianbei]

Originally Posted by ghorrocks

Are you saying the implementation of rotating frames of reference in CFX is wrong?

Hi,ghorrocks:
I reviewed many papers in this field of rotating channel and find that their main method is to set the pressure gradient adjusted to keep the flow rate constant. I referred to the CFX-help and got to know how to get the pressure gradient through Fortran. But I still don't know how to make the pressure gradient to adjust the flow rate. Can you give me some advice?

[ghorrocks]

I still have little idea of what you are talking about, and even less idea why you are doing this when CFX already has a rotating frame of reference model.

If you are talking about the pressure gradient at each control volume - you cannot set that. The solver does not work that way.

If you are talking about the pressure gradient from the inlet to outlet - that is easy, you simply set the boundary conditions to the pressure drop you wish.

[huangxianbei]

Originally Posted by ghorrocks

I still have little idea of what you are talking about, and even less idea why you are doing this when CFX already has a rotating frame of reference model.

If you are talking about the pressure gradient at each control volume - you cannot set that. The solver does not work that way.

If you are talking about the pressure gradient from the inlet to outlet - that is easy, you simply set the boundary conditions to the pressure drop you wish.

I'm sorry I've tried many ways to do this. The problem is like this: a rotating channel, with rectangular intersection. The rotating axis is spanwise, say z axis for example, and the stream wise direction, x axis, under the effect of the Coriolics force, the mean flow along the y direction should be asymmetry. In the DNS simulation, usually the mean effective pressure gradient grad(Peff)(Peff=P(ave)+centrifugal force(ave)) is set as a constant(while the equations to be solved is transient), and the x and z directions are set as periodic to get a fully developed flow.

In order to do this in CFX, I chose the RANS, which in the equations to be solved is time-averaged, so if I want to get the same result, I should also set the grad(Peff) to be a constant. The problem I'm facing is how to ensure this? In the RANS equation, how can I guarantee the sum of the pressure gradient term and the centrifugal force term being a constant?

[ghorrocks]

The rotating domain model in CFX already includes coriolis and centripetal components. The effect you are talking about should be already in the model. It is not imposed as you describe (a constraint on grad(Peff)) but as source terms on the momentum equations to account for the rotation. This is all described in the documentation for further details.

You cannot impose a constraint in CFX such that grad(Peff) is a constant. You would need to write a custom solver to handle this constraint.

My suggestion is you use the built in rotating domain model and check that grad(Peff) is constant (or nearly so) in post processing. This will be a check on the accuracy of your simulation - and of the assumption that grad(Peff) is in fact constant.

[huangxianbei]

Originally Posted by ghorrocks

The rotating domain model in CFX already includes coriolis and centripetal components. The effect you are talking about should be already in the model. It is not imposed as you describe (a constraint on grad(Peff)) but as source terms on the momentum equations to account for the rotation. This is all described in the documentation for further details.

You cannot impose a constraint in CFX such that grad(Peff) is a constant. You would need to write a custom solver to handle this constraint.

My suggestion is you use the built in rotating domain model and check that grad(Peff) is constant (or nearly so) in post processing. This will be a check on the accuracy of your simulation - and of the assumption that grad(Peff) is in fact constant.

Thank you. I think I've misunderstood the problem. You are really kind for answering so many questions. Thank you again.