[MovinJ]

Hi,

I'm trying to run a transient simulation that is of a cyclonic separator for a chemical separation project. I'm trying to see the effectiveness of this separator with aerosol-type droplets, and so have modelled the liquid state chemical (NH3), that needs to be separated from the gas, as a dispersed solid with mean diameter of 50 microns. I have turned buoyancy on and have also applied a solid pressure model so that the droplets once fallen to the bottom don't become packed on top of each other.

I also included a drag coefficient model (the Gidaspow model) as this is the behaviour that I'm most interested in to see, how it interacts with the gas and gets separated due to circular motion. However, the solver keeps failing and refers to the "Droplet|Syngas. drag coefficient" as the defective variable and the error type is ENFORCE BOUNDS.

I have had a look at the theory guide for CFX and it seems that the gidaspow model has good definitions for volume fractions between 0-0.8 and 0.8 to 1, but it apparently "Linearly interpolates" between the two functions when the value is between 0.7 and 0.8, my intuition at the moment is this is what's causing it. But I really don't know for certain.

Something to note is that the solver solves very well for the first few fractions of seconds, maybe 60-70 timesteps, and then fails, and it usually says overflow, but I sometimes see this error too.

Any help on this would be really appreciated, I have been stuck on this for weeks now and it would mean alot to me if there are others out there that have had issues with the Gidaspow model as well, and what was done to fix it.

Many thanks

[ghorrocks]

I think your problem is you are using an inappropriate model for your physics - which means it will never run properly no matter what you do.

The Gidaspow model is for solid pressure caused by solid particles bouncing off each other. You do not have solid particles (they are liquid) and they will not bounce off each other (I suspect they will merge into a larger drop). So the Gidaspow model is not appropriate for what you are modelling.

I presume you are using a Eularian multiphase model for the particles. If this is the case then the only model I can see which models wall deposition is with the Algebraic slip model (see CFX Theory manual, 7.16.1.3), and I suspect the simplifications inherent in the Algebraic slip model may make it inappropriate for you.

As this is the only wall deposition model I am aware of in a Eularian multiphase model and it is unlikely to be useful in your case, this means you will have to look at the Lagrangian particle model.

If you use a Lagrangian particle model there is a wall film model in CFX which might be suitable for what you are doing. Also you could use simple coefficients of restitution terms to define whether the particle bounces or disappears.

I have been doing some cyclone work recently as well and I have taken the approach that there is no good model for what happens to the drops once they impact the wall. Therefore I have assumed they stick to the wall and disappear, implemented by just using a simple restitution coeff=0. I know they accumulate into drops on the wall and then are blown across the surface by the air flow but I do not have a good model for that so I am not modelling it.

[MovinJ]

Hi Glenn!

Thanks for replying. I just wanted to say that I have modelled the Liquid NH3 as a 'Dispersed Solid' and I've set the dispersed solid particles to have a density of a liquid. So they act as solid particles in the model, but represent spray particles in a liquid.

Originally Posted by MovinJ

... I'm trying to see the effectiveness of this separator with aerosol-type droplets, and so have modelled the liquid state chemical (NH3), that needs to be separated from the gas, as a dispersed solid with mean diameter of 50 microns. ...

I'm not entirely sure why I took this approach, maybe at the time I thought modelling two fluids with a fluid-fluid pair model would be too complicated, as I'm new to CFD and I had been put off from modelling two fluids like this due to my colleagues advice (They were talking more about modelling condensation).

Is it wrong to model a 'work-around' like this for real behaviours, as it isn't a 'real' problem (i.e. the particles are never going to be solid in reality) and therefore this model would never have a solution?

I've just spent a few hours having a look at the Algebraic Slip Model on CFX theory guide and reading up on it, and I think it's an interesting but also quite complicated model to understand, and the theory guide doesn't really give too much of details on it.

You mentioned lagrangian and euler multiphase models. How do I implement those models? Which material morphology option do those tend to appear in? (continuous fluid/dispersed fluid/particle transport fluid?)

If possible, would you be willing to share your .cfx file or the CCL code of the cyclonic separator project that you have been working on, so that maybe I could understand how to implement ASM or wall film model, or even how you included the 0 coeff. of restitution.

Many thanks

[ghorrocks]

The Algebraic Slip model is designed for particles which have little inertia, like small air bubbles in a liquid. It is not appropriate for heavy particles in a light carrier fluid - which is what you have. That is why I said I do not think the ASM is appropriate.

And yes, when you use a model outside of its intended use (like what you have done with the Gidaspow model) then you risk running into strange behaviour and poor convergence. It is buyer beware. But more importantly you are unlikely to be capturing the intended physics correctly, so even if you do get it to converge I would be very suspicious of the results.

Eularian = "continuous fluid" and "dispersed fluid", Lagrangian = "particle transport"

All the cyclone work I have been doing is using the Lagrangian model (particle transport). It has the advantage that it is easy to put a range of particles in and you can see the individual particle tracks, but it is much more computationally expensive than the Eularian models.

For a particle tracking model - see the tutorials for that. The coeff of restitution is under the wall settings (as it applyies to walls).

[MovinJ]

Thank you so much for the support. This has been really helpful. It would have been nice to know how to get the Gidaspow model to work with the Gidaspow solid pressure model, which I think might have been the underlying problem I was having (just my beginner assumption), but this has shown me that there are better ways of modelling cyclonic behaviour, and that I should probably look into it.

Have you come across any design formulas for the design of cyclonic separators? Just a separate discussion from CFD as since you have worked on cycloning separators before, I was wondering if you may know any design standards/codes that take the design by formula approach to design a cyclonic separator. Just in case, as I'm worried the CFD might be too hard for me to wrap my head around and get it to work.

Many thanks

[ghorrocks]

We have a few in-house references, but publicly available and seminal works include:
Lapple, C F: "Fluid and Particle Mechanics" University of Delaware 1951
Stairmand: 1951
Swift: 1969