Is CFD software useful?

[sgtlethargic]

It says "... shaping still has to be carried out almost exclusively by experiment [versus numerical methods]"
in "Aerodynamics of Road Vehicles."

A solar car article said they used CFD software.

[JulianEdgar]

Originally Posted by sgtlethargic

It says "... shaping still has to be carried out almost exclusively by experiment [versus numerical methods]"
in "Aerodynamics of Road Vehicles."

A solar car article said they used CFD software.

Professional level CFD is used by all major car manufacturers, solar car designers, etc, to develop shapes. However, from what I understand, there is a major difference between CFD of the sort available free and at low cost - and the professional level stuff.

Professor Joe Katz was scathing about low cost CFD. He said to me: "It's just kids with pretty pictures."

Rob Palin (ex Tesla) is currently using a professional version of a low cost CFD to do some modelling. It was taking his PC (he gave the spec - overclocked and high quality) three days to churn out 10 seconds of airflow modelling.

If you look around the web you'll find people modelling current cars with low cost CFD - and their calculated Cd values are often very different to the manufacturer's quoted Cd value for the car.

I am no expert in this area, but as far as I can tell, CFD isn't yet viable at amateur level for results that you'd trust. I'd love to see some comparison testing between low cost CFD and full-size testing of the car.

[aerohead]

Originally Posted by sgtlethargic

It says "... shaping still has to be carried out almost exclusively by experiment [versus numerical methods]"
in "Aerodynamics of Road Vehicles."

A solar car article said they used CFD software.

The Full-Navier-Stokes Equation is the only numerical path to an accurate 1:1 scale representation of the real world, including all separation, turbulence, and the wake.
Hucho wrote in 1986 that this procedure was only then bordering on the possible, with the advent of supercomputers, and then, even requiring days to navigate a single iteration.
Daimler and BMW were using a watered down variety a decade or so ago, requiring almost 48-hours run time for a single iteration, whereas in a full-scale wind tunnel you can get a result in four minutes.
The beauty of CFD is that power requirements are that of only a 'village', as with a full-scale climatic wind tunnel they require the electricity of a 'city.'
The director for Lockheed's Marietta, Georgia wind tunnel told me in 1991 that every time they turned the tunnel on, the lights in the neighboring town all dimmed.

[freebeard]

Whelp, they're both gone; but both right? Or they agree and they're both wrong?

CFD is useful. The problem is in the software. The big corps throw money at the problem that 'free' software won't resolve. But computational science progresses. It's a modeling problem. The memory and speed constrained solutions are obsolete.

https://en.wikipedia.org/wiki/OpenVDB

https://www.openvdb.org/

OpenVDB is an Academy Award-winning open-source C++ library comprising a novel hierarchical data structure and a 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 and is now maintained by the Academy Software Foundation (ASWF).

August 13, 2020
OpenVDB 7.1.0 is now available to download.

Highlights of this release:

new fast sweeping methods that outperform existing techniques for computing signed distance fields in addition to supporting velocity extension

I know what they are saying, but I couldn't say it as well myself. I used to work elbow-deep in the MacOS file structure, so I have an understanding of sparse, shallow b-trees. They use some brilliant optimizations.

So what does this mean for the unwashed masses? It has to do with free and open-source software. The best, fastest moving example I know of is Blender. It evolves on a rapid schedule, but with landmark long-term support releases so you can plan forward:

About
Blender is the free and open source 3D creation suite. It supports the entirety of the 3D pipeline—modeling, rigging, animation, simulation, rendering, compositing and motion tracking, video editing and 2D animation pipeline.

Blender has a physics engine and OpenVDB support. Here's a demonstration of it's use:

https://www.blendernation.com/2020/0...db-quickstart/

Blender itself has implemented adaptive subdivision which works on similar principles.



A CFD demo reel is above my pay grade, although I understand the principles, I lack the muscle motor memory. I hit a wall early on. But there's your solution — free, runs on low-end hardware but likes big iron.

[aerohead]

Originally Posted by freebeard

Whelp, they're both gone; but both right? Or they agree and they're both wrong?

CFD is useful. The problem is in the software. The big corps throw money at the problem that 'free' software won't resolve. But computational science progresses. It's a modeling problem. The memory and speed constrained solutions are obsolete.

https://en.wikipedia.org/wiki/OpenVDB

https://www.openvdb.org/




I know what they are saying, but I couldn't say it as well myself. I used to work elbow-deep in the MacOS file structure, so I have an understanding of sparse, shallow b-trees. They use some brilliant optimizations.

So what does this mean for the unwashed masses? It has to do with free and open-source software. The best, fastest moving example I know of is Blender. It evolves on a rapid schedule, but with landmark long-term support releases so you can plan forward:


Blender has a physics engine and OpenVDB support. Here's a demonstration of it's use:

https://www.blendernation.com/2020/0...db-quickstart/

Blender itself has implemented adaptive subdivision which works on similar principles.



A CFD demo reel is above my pay grade, although I understand the principles, I lack the muscle motor memory. I hit a wall early on. But there's your solution — free, runs on low-end hardware but likes big iron.

'Commercial / Industrial' CFD competes with full-scale wind tunnels.
Their 'price-point' must be 'competitive' in the marketplace.
If using, say, Dassault's EXA POWERFLOW CFD ( which Tesla Motors used for their Model S development ) can produce output in 48-hours, equal to, and at a cost advantage to full-scale wind tunnel testing, then there's a probability that it will be seen as smart business decision.
If it costs $ 2,000 / day (perhaps the minimum ), it's the same cost as 1/2-hour of wind tunnel time.
And since it requires a supercomputer, or, say, 100 desktops running in parallel, we might not expect 'free' software to offer the degree of realism the 'big-boys and girls' are obtaining.
The CFD Daimler- Benz and BMW use is 99% accurate, compared to 1:1-scale wind tunnel testing.

[jakobnev]

And since it requires a supercomputer, or, say, 100 desktops running in parallel

I was trying try google on DDG how suitable CFD is for parallel computing i.e. would you get close to 100x performance when run on 100 desktop PCs with low bandwidth, high latency networks between them.

[freebeard]

The corps don't get earlier access to free and open source software (OpenVDB) than anyone else. They may be able to capitalize on it faster, absent institutional inertia.

[aerohead]

Originally Posted by jakobnev

I was trying try google on DDG how suitable CFD is for parallel computing i.e. would you get close to 100x performance when run on 100 desktop PCs with low bandwidth, high latency networks between them.

I've tried to contact Dassault in order to ask these types of questions. They won't return a phone call.

[freebeard]

One could come at it from the other direction. OpenVDB lists software that employs it. Likely candidates might be

https://realflow.com/ — FLUIDS & MULTIPHYSICS SIMULATION
https://jangafx.com/ — REAL-TIME VOLUMETRIC FLUID SIMULATIONS FOR GAMES AND FILM

They might be more responsive.

[Cd]

Is CFD software useful ?
Useful for what ?
I have a long defunct CFD program put out many years ago called Falcon.
It is now called something else, and costs thousands of dollars.
I got the free beta program.
The program was apparently created to be portred to tablets for schools.
It not only has a touch friendly interface, but also silly models included in the test folder, such as a car, an SUV a toy spacehip ( ! ) and a pair of SUNGLASSES ( !! )
This leads me to think it was designed for children.

So what about the program ? Well I find it useful to test the direction of airflow on vehicles without tuft testing them.
Is it accurate ?
Not really, but in some areas surprisingly so, given that I am literally running it on a second generation Surface Pro Tablet with maybe 2 GB ram. ( My phone is around twice as fast )
You can see the flow lines over the car in near real time.
It also gives an accurate view of the pressure on the body of the car.
It even gives a Cd after only 20 -30 minutes.
They provide a roughly built model of a Porche 911, and the Cd it gives is actually accurate.
However, models I import in .obj format or any other such as .3ds don't give accurate Cd results.

So if you are just looking to do a virtual tuft test or see if a modification changes the Cd in a good way, then certainly I think it would be useful.
Many of these CFD programs can run on a normal laptop.

If you do find a good CFD program that is accurate and CHEAP, as well as easy ( visual U.I. ) please share the info on it.