[montera]

For the exactly same shape geometry, Why simulation is very slow when it's small(centimeter), but fast when large(meter)?

Xflow to simulate a fan.

[AtoHM]

How is it the same geometry, if its once small and once big?
A too general answer to a too general question:

Probably it results in different mesh sizes with whatever parameters you use for the meshing and thus, simulation time differs.

[montera]

Originally Posted by AtoHM

How is it the same geometry, if its once small and once big?
A too general answer to a too general question:

Probably it results in different mesh sizes with whatever parameters you use for the meshing and thus, simulation time differs.

Yes, once imported the geometry by centimeter unit, then simulated.
Second time, imported the geometry by meter unit, then simulated.

Every parameters are the same. Except Resolved scale times 100.

[AtoHM]

My reply was aimed to get you to investigate your setup more thoroughly and tell us what you are doing. With the information provided, it is just guessing:
* what do the meshes look like (cell count, quality, ...)
* what equations are solved, models used
* are the results similar, are important quanities converging?

Have a look how other people here describe their cases.
Guide: How to ask a question on the forums

[montera]

I attached a very simply example to explain this question.(if I attached them correctly)

Xflow.zip contains 2 project files:
"Cylinder100.xfp, Cylinder100.lay" is the large one, "dm.png" tells its time.
"Cylinder.xfp, Cylinder.lay" is the small one, "mm.png" tells its time.

I just wish to learn the reason for the obvious difference.

[AtoHM]

Most people, including me, on this forum are not in possession of that particular software, so unfortunately this does not help.

[arjun]

Originally Posted by montera

For the exactly same shape geometry, Why simulation is very slow when it's small(centimeter), but fast when large(meter)?

Xflow to simulate a fan.

Assuming x flow is lattice boltzmann method, by changing the grid length ie from meter to cm you are now solving the problem at different Reynolds number because viscosity is computed from the size of the domain.

Thats why now the time to simulate has changed,

[montera]

Originally Posted by arjun

Assuming x flow is lattice boltzmann method, by changing the grid length ie from meter to cm you are now solving the problem at different Reynolds number because viscosity is computed from the size of the domain.

Thats why now the time to simulate has changed,

I try to catch your point.

Does it mean domain size ↓, viscosity ↑, Reynolds number ↓, time ↑ ?

[arjun]

Originally Posted by montera

I try to catch your point.

Does it mean domain size ↓, viscosity ↑, Reynolds number ↓, time ↑ ?

Should be domain size down viscosity goes down and increase in Re gives to more computational efforts (reduction of time step).

There is a paper about it:

https://www.semanticscholar.org/pape...7321d72aeceaaf

Numerical Viscosity of Finite Difference Lattice Boltzmann Method

S. Mizutani, M. Tsutahara
Published 2006

Physics, Mathematics
Transactions of the Japan Society of Mechanical Engineers. B

The difference of the influence of numerical viscosity between the Navier-Stokes based finite difference method and the finite difference lattice Boltzmann mathod is shown. In order to stabilize calculation, upwind schemes with numerical viscosity is generally used in the finite difference lattice Boltzmann method. The influence of numerical viscosity is dependent on a Mach number. The dependence to the direction of the flow of numerical viscosity is small as compared with the Navier-Stokes based finite difference method.


Edited to add: Interestingly this was pointed to me by Tsutahara san himself when he was trying to clear some of my doubts.

[montera]

Originally Posted by arjun

Should be domain size down viscosity goes down and increase in Re gives to more computational efforts (reduction of time step).

There is a paper about it:

https://www.semanticscholar.org/pape...7321d72aeceaaf

Numerical Viscosity of Finite Difference Lattice Boltzmann Method

S. Mizutani, M. Tsutahara
Published 2006

Physics, Mathematics
Transactions of the Japan Society of Mechanical Engineers. B

The difference of the influence of numerical viscosity between the Navier-Stokes based finite difference method and the finite difference lattice Boltzmann mathod is shown. In order to stabilize calculation, upwind schemes with numerical viscosity is generally used in the finite difference lattice Boltzmann method. The influence of numerical viscosity is dependent on a Mach number. The dependence to the direction of the flow of numerical viscosity is small as compared with the Navier-Stokes based finite difference method.


Edited to add: Interestingly this was pointed to me by Tsutahara san himself when he was trying to clear some of my doubts.

Really? you happened to meet the author?

Frankly speaking, the paper is some difficult for me.
So, this difference of simulation speed is normal, neither xflow problem, nor my mistake?

[arjun]

Originally Posted by montera

Really? you happened to meet the author?

Frankly speaking, the paper is some difficult for me.
So, this difference of simulation speed is normal, neither xflow problem, nor my mistake?

We were living in same city Kobe Japan that time and he sometimes used to visit the place I was working.

I had difficulty understanding the viscosity since in lattice boltzman you really do not specify it. It is an outcome of three main things

1. Time Step

2. Relaxation Parameter

3. The advection method used.


You are not doing anything wrong, what you are seeing is what is to be expected.

[montera]

Thank you, arjun and AtoHM, very much!
I learned much these days.
And I hope get along with xflow.