[Zdfjpbth]

And what use does consistent nonsense have?
And not knowing how the code they use works means they will get in Honda like situations.
It might be cheap but it certainly isn't top drawer approach, IMO.
It might take time to extensively test and validate the code, but once it's done it's done and it's working as you wish.

I agree that I look to it from the POV of ultimate accuracy and performance, and maybe F1 teams aren't that much interested about this.

I use the word nonsense as it is very easy to believe what CFD tells you if you naively take the results at face value. A lot of engineers can be misled by results as they may not have fully considered the consequences of their modelling technique, as in the CDG. I make the distinction between the model used and its numerical implementation - you can have many combinations of numerical schemes that satisfy the governing equations for a given modelling technique. Obviously numerical schemes can make a difference but you can bet that industrial best practice outlines a certain set to be used, and these will always be a constant in the simulations.

CFD is comparatively cheap. It certainly saves the manufacture of a lot of components that may be of no use when the analysis is done in the wind tunnel. But it comes at a price of reliability; developing an in-house code from scratch that requires a) taking a CAD geometry, b) meshing it to give a high quality computational grid, c) solving the governing equations on that grid to a good degree of accuracy and d) post-processing the results is not the work of a moment. As I said above it takes years (and I'm not kidding either) to do this. No F1 team has that kind of time to burn so buying in commercial codes is clearly the best plan. In addition to the availability of commercial codes (such as Fluent, Star-CD, CFX etc.) the numerical models in those packages are more than adequate for the speed and efficiency of simulations required for industrial CFD. The general rule of thumb is that flow solutions on complex geometries should take no longer than 24 hours to be produced from point a) to point d). That's a pretty big challenge even with today's computer power hence they buy those ridiculously big computers to do the runs.

The point about accuracy is that you can do a fully time-dependent simulation of a F1 car now if you want - you'll only need the biggest supercomputer in the world and about 50 years to run the simulation. Evidently that's impractical so shortcuts must be made through modelling assumptions. The levels of accuracy of these models in CFD are as follows:

1. Euler solutions: Viscosity is assumed to be zero hence boundary layers do not form near solid surfaces. Very crude but useful for high speed flows like that around aircraft. Produces a steady-state (i.e. average) solution. Runs on a coarse grid and very cheap to compute.

2. Reynolds-Averaged Navier-Stokes (RANS): Effects of viscosity included, hence an improvement over Euler methods. Again produces an average flow solution - many physical processes are smeared out. Needs a finer grid than Euler and is more expensive.

3. Large Eddy Simulation (LES): Time-dependent method where the Navier-Stokes equations are solved exactly above some filter width (usually defined by the computational grid). Below this filter width the scales of motion are modelled using an algebraic function. Needs a much better-resolved grid than RANS and is significantly more expensive. However the results obtained from LES have a much better basis in physical reality than the above two methods.

4. Direct Numerical Simulation (DNS): Navier-Stokes equations are solved for all scales of motion. Grid spacing must resolve the smallest turbulent scales in the flow, typically around a micrometer in size. Stupendously expensive. Very accurate.

Academics perform simulations using 4. on simple things like flows in a pipe, turbulence behind a wire grid, simple boundary layers over a flat plate and so on. It's used to study fundamental turbulence and is of absolutely no interest to industry for the next 20-30 years (at least). Technique 3. is just starting to creep into industry as it costs a hell of a lot less than 4. and offers a reasonable prediction of the real flow physics, although industrial engineers mistakenly think there is little benefit in 3. over technique 2. RANS is now almost standard in industry, with 1. being used mainly in aerospace design for "quick-and-dirty" calculations. The point I'm trying to make here is that accuracy and computational turn-around time are very much interwoven. F1 can't afford to do the highly accurate, time-dependent simulations that 3. and 4. could give them, hence they turn to more crude modelling (mainly 2.), but implicit in those modelling assumptions are things than can be misleading if their effects on the flow are not fully appreciated. This is why no CFD code in existence can be trusted implicitly and the results from CFD must be complemented by wind tunnel test results.

Back to the point of the topic and why I raised CFD in the first place; restricting wind tunnel testing will make teams use CFD more, but for the moment there is no substitute for tunnel testing and if it is limited then it will affect their rate of development.

I doubt it, honestly. There are new theories and models being developed each and every day, that will only be implemented in an industrial code in the next years. What if you could have it a few months before the competition has it too?

New models are developed in academic circles, validated for generic cases and then possibly attract the interest of commercial CFD developers if enough of their customer-base wants it included in the package. Industry is notoriously set in its ways when it comes to CFD models so anything new needs to be pretty hot to make them budge.