Firstly, all the teams use CFD as well as wind tunnel testing. Indeed, no doubt the likes of McLaren and Ferrari, etc, will likely (I suspect) be using CFD more extensively than Virgin, such is their resources. So it’s not as if Virgin is being radical in the sense that they’re using some super advanced technology and methodologies. In truth, it’s quite the opposite and they’re being less sophisticated (but are saving time and costs) by not using a wind tunnel/CFD combination.

I feel that some people are thinking of comparing wind tunnel testing to CFD as akin to VHS to DVD or something – i.e. that wind tunnels are a ‘manual labour’, dinosaur way of doing things that the ‘old’ teams are stuck in their ways with, and CFD – being all digital and whizzy computers – is by definition more accurate and better quality. This is the wrong way to look at it. CFD is not analogous to someone who has a calculator (the CFD engineer) being able to do calculations many times quicker than another chap stuck trying to work with an abacus (the wind tunnel tester). Heh, how many more analogies can I conjure up – I’ll stop there!

Peter Murdoch has already well surmised the deeper technicalities of CFD, but in essence CFD is a way of mathematically modelling an aspect of the physical world (in this case fluid flow: air, water, whatever), and then using numerical simulations to understand and predict the behaviour of the fluid in various situations (such as the flow around the endplates of a front wing), and to calculate properties such as drag, (negative) lift force, vorticity, temperature, etc.

However, whenever you mathematically model something you inherently have to make simplifying assumptions of the real world to arrive at a suitable model amenable for analysis; it would be impossibly complicated otherwise. So already you are approximating the real life phenomenon and so already you’re not going to be accurate as if you can see the real thing (such as the flow shapes in a wind tunnel – as long as it’s calibrated and you’re measuring instruments are suitably accurate).

Now, modelling fluid flow is extremely complicated. Except in trivially simple situations, there are no exact solutions to the equations that govern the motion and properties (pressure, etc) of any fluid flow (these are called the Navier-Stokes equations and are basically just a sophisticated advancement of Newton’s laws of motion, i.e. Force = Mass x Acceleration, adapted for fluid flow). And by exact solution I mean it’s not a case where you can plug data into an equation, turn the handle and out pops your answer (unlike, say for an equation to model the growth of your savings account based on the interest rate, where you stick in your bank balance, turn the handle of the equation governing how interest is applied, and bingo, out pops your lovely savings for the month/year)!

With CFD, to obtain solutions you have to approximate (discretise) the equations and solve them numerically on a computer. The key word here is ‘approximate’. In approximating something you’re by definition introducing an error into your answer (e.g. down-force figures), and in F1, errors can mean tenths of a second (or seconds if you’re the new teams – ah that’s unfair)! To obtain more accurate answers requires more computing power since you need to use a finer mesh (i.e. use more data points on the F1 car/component model) and a more complex model to simulate turbulence (cf. Peter Murdoch’s post above), and thus solve more equations.

Now due to the complexity of the equations and geometry of an F1 car, it is currently impossible to use CFD to exactly model every aspect of the car. That is why wind tunnels are still worthwhile. Of course they need to be correctly calibrated, since like with CFD it’s a case of, put garbage in, get garbage out. Perhaps that’s what Wirth means in his “The Eureka moment was when he realised that the CFD numbers were more accurate than the wind tunnel” quote. But as mentioned by others, you need (properly calibrated) wind tunnel data to validate the CFD so that you know your virtual model is behaving true to life.

Finally, here’s a summary of some of the advantages and disadvantages of CFD to wind tunnel testing, so that others hopefully can appreciate the worth of each and why it’s better to have a blend of the two.

Advantages of CFD compared to wind tunnels:
- Can analyse whole domains and measure many parameters (velocity/pressure, etc) simultaneously. Wind tunnels use sensors to measure a single point on the car (just like the sensors hanging off the McLarens in testing). Hence a whole wind tunnel run could be useful for just a handful pieces of data. Also, you have to ensure that these physical sensors in the wind tunnel don’t adversely disturb the flow you’re measuring too, since they won’t be there on track! By definition CFD calculates solutions across the whole domain investigated.
- Cheaper overall since quicker turnover times than having to design, build (and bin!) models. Also parts on the car can be much more easily remodelled on the computer so you can test designs tweaks quicker.
- Better visualisation of results. Computers can produce a wealth of informative graphs/flow patterns etc, and you can focus on design details of particular interest. Wind tunnel photos/videos don’t necessarily provide the same level of detail.

Disadvantages of CFD compared to wind tunnels:
- Results can be erroneous. It is well known that in some circumstances CFD results do not coincide well with reality. This is notably the case for complex turbulent flow (which is pretty prevalent in F1 eh)! Example problems include: overproduction of turbulent energy in wakes and incorrect generation of vortices/ vortex shedding. Thus due to the difficulties of modelling and unpredictive nature of turbulence, a CFD model could predict that the vortices off the front wing endplates move nicely around the front tyres, but then in the wind tunnel and on the track, a completely different flow pattern could be happening, and not just that the CFD results are a little less accurate!
- CFD is limited by computing power and thus the size of the projects are limited, though of course computers just get ever more powerful! A large wind tunnel is less compromised. Also, due to the modelling complexities wind tunnels can handle complex geometries better.

Apologies if some of this is rather dry but I hope it is of use to people. One of the great things about this site is the quality and knowledgeable discussions that go on her compared to other sites. James has done great work!

In order to compare one thing to another you need a frame of reference. In classical science and engineering, empirical formulas are used to make estimates that are correlated to experimental data. This works great for calculations such as the strength of a straight beam and the mass of an object, but for fluid flow problems there are no empirical methods for computing the exact flow over an arbitrary complex shaped body let alone the flow over a F1 car. Hence the only way to determine the definitive flow as well as the lift and drag values on a car is with a calibrated wind tunnel. That is the frame of reference. You then compare the measured results to your CFD simulation and determine the error of the simulation not the other way round. These are basic fundamental principles of experimentation and simulation.

When talking about the accuracy of a CFD simulation you need to consider the accuracy of the turbulence model used in the simulation. The reason being is the lift and drag values on the car are highly sensitive to turbulent flow separation in the boundary layer.
There are several ways to model this turbulence in CFD, ranging from the most accurate DNS (Direct numerical simulation) and LES (Large Eddy Simulation) to the least accurate approximation method, RANS (Reynolds-averaged Navier–Stokes). F1 teams all use the RANS turbulence model.

The reason you ask? Well it’s all to do with the mesh, the mathematical base model where the Navier Stokes equations are solved.
The teams use either a tetrahedral mesh or a trimmed polyhedral mesh, which are generated by an automatic meshing program that requires a relatively low level skill to operate. The biggest problem is you can only run a RANS turbulence model on these tetrahedral meshes. They also require enormous computing resources and worst still produce fictitious results that can be completely misleading. Hence the reasons why F1 teams still keep and use their wind tunnels. They know from experience that a CFD model will generate fake vortices’ and completely miss other crucial ones.

The only way to produce a CFD result with any level of respectability and degree of accuracy is to use a high quality all Hexahedral mesh running a transient LES turbulence simulation. The next obvious question is why don’t the teams use this method? Yes indeed that is a very good high tech question.

Instead, there are a series of hydraulic rams attached. These rams can position the tub in any attitude necessary to simulate G forces on the driver, for example, on it’s side to simulate cornering, and nose down to simulate braking.

The engineering involved is, to calculate from the driver’s inputs the attitude changes to induce. Full hard on the brake pedal while simulating my Smart car would be, say a 15-degree nose-down attitude; stepping out of the simulator and repeating the action in my Smart should feel the same. For a Formula 1 car, the nose-down would be more pronounced, because there are higher G-forces to be simulated.

If the simulator is accurate, the driver’s inputs (all logged, I presume) are representative of the data that would be collected by driving the actual car around a track. To the degree that this is the case, the engineers have correctly understood and solved the problems they were working on.

Accurate simulation is critical to getting accurate inputs from the driver. Accurate inputs from the driver are critical to making correct design choices.

All that said, the only simulator I have had the pleasure to experience was at Disneyland. I must say, even the potholes seemed very real. The technology is amazing.

The more you know about CFD the more you understand its limitations, the less you know the more you imagine it can do.

I know what the simulator has to do to provide some feedback to the driver. However, a simulator is not primarily used for the development of the driver it is for the development of the CAR.

If you go back to my last post, my question remains unanswered i.e. how can such large movements of the simulator provide accurate feedback for changes that are always small increments or improvements of an existing part. I assume that the engineers can measure the comparative differences between parts but I was looking for a more precise explanation.

Just one point – Manor’s HQ is in Rotherham.

It’s not as hard to believe as you might think that a F1 race team and cars will be equipped, based and resourced in the former pit village of Dinnington, Rotherham.

Coincidently, Rotherham is also home to F1 Show Cars – the proud owner of the only 3-seater F1 Simulator to date. Available to hire with virtual reality goggles included – http://www.f1showcars.co.uk/