Realtime CFD ?

[freebeard]

here's some things it would incorporate:

Word salad reflects cognative dissonance.

For one example: 'Navier-Stokes Equation'. Mooted by lattice Boltzman methods.

And you omitted OpenVDB. https://www.openvdb.org/

Maybe Vekke will show interest?

[aerohead]

Originally Posted by freebeard

Word salad reflects cognative dissonance.

For one example: 'Navier-Stokes Equation'. Mooted by lattice Boltzman methods.

And you omitted OpenVDB. https://www.openvdb.org/

Maybe Vekke will show interest?

You just showed your hand and by your own doing revealed the vacuity of any command of the subject.
I reduce ten pages of highly technical science text from an authority on the subject to bullets, and this is my reward.

[freebeard]

You just showed your hand and by your own doing revealed the vacuity of any command of the subject.

"Showed your hand"?

I don't claim 'command of the subject'. An unsorted keyword list is not informative. Which would be the salient line item?

Why aren't lattice Boltzman and OpenVDB pertinent?

I shan't experiment with Fluid X3D until my 3D printer is back up and running. Then we can compare notes.

edit:: I looked at the system requirements.

Compatibility
works in Windows, Linux and Android with C++17
runs on any hardware that supports OpenCL 1.2, from any vendor (Nvidia, AMD, Intel, ...)

So to run it on an M1 Mac Mini I'd have to install a new operating system. There was one Linux distro I was going to watch. If I remember the name....

2nd edit:

https://screenrant.com › install-linux-m1-mac-shouldnt-why
You Can Now Install Linux On An M1 Mac, But You Probably Shouldn't
25 Mar 2022 Asahi Linux For M1 Macs The Asahi Linux installer allows setting up a separate partition to boot directly into Linux on an Apple Silicon Mac and runs well with a few limitations. The GPU card isn't enabled and admittedly will take quite some time to get working. That means gaming will not work and it can't even handle video streaming.

Maybe Ubuntu in a VM?

[Piotrsko]

Ubuntu would probably choke on the data. Some other Linux might not.

Oddly , some sort of COBOL variant might handle the data massaging

[aerohead]

This solver :
* uses Very Large Eddy Simulations
* it does not perform Large Eddy Simulations
* it does not perform Small Eddy Simulation
* it requires either a CAD-CAM data cloud, or 3D-scan
* it cannot solve 'holes' ( partial cooling systems ), cabin ventilation, lug nuts, windshield wipers, must be 'airtight'
* requiring CAD clean-up
* it requires a meshing program to generate the mesh
* it requires a turbulence model
* it requires post processing after simulation to create coefficients and flow data
* it's scan may be required to be reduced in resolution in order to 'quicken' computer time, using something like PowerDELTA.
* scan malformations must be 'fixed' with something like ANSA
--------------------------------------------------------------------------------
* Accuracy between (LBM) results and wind tunnel have been reported at a 13% deviation.
* (LBM) has been used for 'qualitative' 'trend' investigations, with no pretense of achieving 'accurate' numerical data.
* (LBM) cannot predict flow separation locations
* (LBM) cannot predict flow reattachment locations
* (LBM) cannot accurately predict an accurate wake geometry
* (LBM) cannot accurately predict wake size
* (LBM) 'reruns' can produce variability in results in spite of zero changes introduced into the 'test' parameters.
* (LBM) is capable of registering zero trends at all for some modification
--------------------------------------------------------------------------------------
NOTE:
* the image of the Space Shuttle is associated with aeronautical CFD, with which attached flow is presumed, and aeronautical engineers are not concerned with 'solving' for it.
--------------------------------------------------------------------------------------
CAUTION:
In automotive work, CFD is required to 'solve' the entire flow domain, not only the near-field around the vehicle, but also the far-field up to 2-meters away, in all directions, except underneath. This is where the computing penalty rears it's ugly head.
As of 2014, Dr. Theodore Wolf, of Mercedes-Benz reported that their ( industry-grade) CFD required 48-hours for a single iteration, whereas, they could get 25-measurements/day in their wind tunnel ( 50-results in 48-hours ).
Yes, Moore's Law marches on, with computers growing more powerful all the time.
Perhaps a member, or guest will step forwards with updated information about the current, complete, 'total cost' of University-grade CFD.
Dassault Corp. refused to communicate with me about Ecomodder.com crowd-sourcing their POWERFLOW software, as used by Tesla Motors.

[freebeard]

Sounds like you've been doing some homework.

CAUTION:
In automotive work, CFD is required to 'solve' the entire flow domain, not only the near-field around the vehicle, but also the far-field up to 2-meters away, in all directions, except underneath. This is where the computing penalty rears it's ugly head.

This is where I think OpenVDB shows benefit.

https://github.com › AcademySoftwareFoundation › openvdb
GitHub - AcademySoftwareFoundation/openvdb: OpenVDB - Sparse volume ...
OpenVDB is an open source C++ library comprising a novel hierarchical data structure and a large suite of tools for the efficient storage and manipulation of sparse volumetric data discretized on three-dimensional grids. It was developed by DreamWorks Animation for use in volumetric applications typically encountered in feature film production.

[aerohead]

Originally Posted by freebeard

Sounds like you've been doing some homework.

This is where I think OpenVDB shows benefit.

The only CFD that will 'solve' an automobile completely is the 3D, Full Navier-Stokes Equations for viscous flows, plus a turbulence model.
The billions of cells required are presented in a spherical-coordinate system, with the vehicle at the 'center of the universe.'
The vehicle, along with the entire domain within the virtual test section must be 'solved' simultaneously, each cell affected by the dynamics of each cell bordering it, plus the boundary of the vehicle and 'floor.'
Cartesian grids can't do it.
X, y ,z, time, velocity, pressure, friction, inertia,, etc., for each cell is being solved as a function of what all it's surrounding neighbors are doing simultaneously.
As of 2011, computers were still not powerful enough to accommodate the requirements of the full Navier-Stokes equation.
They had it by 2014.
Industrial Light & Magic employed 3D- scans of actual 'real' dynamics to create the algorithms used in their CGI creations.
Take what they used and multiply by a thousand, and you might begin to represent what the full Navier-Stokes equations must navigate to predict, within 99% accuracy, what air is 'doing' around an automobile and it's flow domain. It makes 'rocket-science' look like 'Tinker-Toys'. I'm quite serious!

[freebeard]

The only CFD that will 'solve' an automobile completely is the 3D, Full Navier-Stokes Equations for viscous flows, plus a turbulence model.
The billions of cells required are presented in a spherical-coordinate system, with the vehicle at the 'center of the universe.'...
Cartesian grids can't do it.

That's what I'm saying. With OpenVDB more distance cells are 'larger'. Each volex can have n attributes attached and available for computation.

Point Clouds and Tools
OpenVDB can natively store point data within its hierarchy using PointDataGrids. These grids can store the points with attributes directly in the VDB Grid. Points are spatially-organized into VDB voxels to provide faster access and a greater opportunity for data compression compared with linear point arrays.

https://www.openvdb.org/about/

'CFD' encompasses the algorithmic pseudocode to implement Botlzman or Navier-Stokes [whichever] and the the dataspace where the computation takes place. Open VDB offers the latter.

[aerohead]

Originally Posted by freebeard

That's what I'm saying. With OpenVDB more distance cells are 'larger'. Each volex can have n attributes attached and available for computation.

https://www.openvdb.org/about/

'CFD' encompasses the algorithmic pseudocode to implement Botlzman or Navier-Stokes [whichever] and the the dataspace where the computation takes place. Open VDB offers the latter.

I'm unsure what the term 'grid' means in terms of a 'spherical coordinate system.'
All the 'cells' are the same dimension.
There exist 'NO' a priori 'attributes' to any of the cells. They are all unknowns.
The 'word salad' that I provided includes some of the math required to 'define' each cell as the simulation proceeds. Just for giggles, let's think in the direction of solving for 1-billion unknowns. It's exponential compared to an aircraft.
'The Navier-Stokes equations represent, in principle, the true simulation of the physics of viscous flow.' Dr. Wolf Heinrich Hucho

[aerohead]

It seems like this is limited to animation.
They're not measuring anything.
Their algorithms have been cribbed from actual kinematic phenomena recorded and stored in front a green or blue screen.
'Air' cannot be 'observed.'