Learning Forum

[kiran.purushothamakeshavan]

Hey guys,

I want to simulate a DPM setup on ANSYS Fluent to count number of particles passing through a boundary in the domain using the trap BC for a varied set of input parameters like: velocity, particle size distribution, relative humidity (by inducing water droplets as injections in the domain) etc.
I'm looking to perform a parametric study on my a domain on Fluent.
Any tips on how I can begin to attempt this?
I tried checking the UDF manual for this but didn't find much help there.
I'd love to hear what you think about this.

Thanks in advance!
Kiran

[Rob]

Getting a parcel count out as a parameter may need a UDF and then figuring out how to turn that into a parameter. However, the DPM reports (mass flow or mass escape - check the options) can be used as an output parameter. If the only source of humidity is the droplet evaporation can you just monitor vapour mass? 

[kiran.purushothamakeshavan]

Hi,

Thank you for the answer.

I get the idea of how I can make this work to observe output parameters. What options would I have to explore to parametrise an array of inputs to get standard set of observations as outputs

[kiran.purushothamakeshavan]

Hi,

Thank you for the answer.

I get the idea of how I can make this work to observe output parameters. What options would I have to explore to parametrise an array of inputs to get standard set of observations as outputs.

[Rob]

It depends on what you want to alter to see the result. Ie do you want to optimise something, if so what can you alter to acheive this. The tools will allow most settings to be changed, but you need to determine the limits: if I want to cool a room I could just spray a lot of cold water into it, it'll be cool but very wet.....  

Have a look at the Design of Experiments functions in Workbench but also Adjoint in Fluent. There are also the new AI based tools,  https://www.ansys.com/technology-trends/artificial-intelligence  to consider. Most of my work tends to be around making the base model do something complicated, and I leave optimsation etc to colleagues who specialise in that field. 

[kiran.purushothamakeshavan]

Many thanks for pointing out to these particular areas that I can focus on.
I just want to setup a virtual environment that takes in multiple inputs (parameters) in a DPM/ similar multiphase model. I guess the design of experiments (DOE) function does seem like a promising option but it's got almost no literature (with DPM), it has only left me wondering if such a thing is possible in using DOE with DPM.

The adjoint method mostly deals with shape optimisation, I think it's a nice addition to what I'm doing right now
Insights from people who specialise on optimisation is most welcome.

Thanks in advance!!

[Rob]

DPM results can be very non-linear relative to observed outcomes (just look at the drag correlations in a text book) so it's not a common approach: trajectory may be linked to diameter, but evaporation rates are surface area dependent. Instead we use some engineering knowledge and approximations to narrow down what we need and then just run the variations in a semi-manual way. 

[kiran.purushothamakeshavan]

I'm trying to find the outcomes (here, number of particle trapped) by taking a statistical approach, like varying a set number of inputs (mentioned above) and then observing the outcomes as number of particles trapped. Do you think the options I've got is just going to be Design of experiments (DOE) through designXplorer? For this again I might need to purchase a license.
Do you think we I can explore some other route to achieve my objective?

[Rob]

Have a look at the report definitions. You may find that monitoring escaped DPM mass to be a better option. Ie mass is injected and some is caught in the system, the remainder escapes. Rather than trying to find what is caught you look at what isn't. It's still potentially a nonlinear correlation but the output parameter is (I think) a result from the flow side of the solver and not the Lagrangian DPM part.