How a dynamic design tool has grown in F1 importance

There are many terms in motorsport bandied about by many people with only a vague understanding of what they really mean. One such is CFD or Computational Fluid Dynamics. In a way the very term is autological in that it describes itself but, equally, three simple words cannot do justice to an immensely complex subject.

Any physical system can, in theory, have its various states calculated mathematically. Sometimes this is easy, such as the case of an object falling in a vacuum. Sometimes it’s more difficult, such as calculating the stress of an object of complex shape under loading. When it comes to predicting the flow of air over and around an F1 car, the problem becomes immensely complex but not insoluble to a reasonable degree of accuracy.

The calculations involved in solving this problem are based on the Navier-Stokes equations which, when Sir George Stokes added a method of resolving the viscous terms in the mid-19th century, became a viable way of predicting air flow. Unfortunately, before the advent of computers, it was impractical to solve the equations for any meaningful problem.

While work was undertaken on the primitive computers that existed in the 1950s it wasn’t until 1967 that the first paper was published suggesting solutions to 3D fluid-flow problems. During the 1960s and ’70s work at Imperial College, London took the art forward leading to the first commercial code, Phoenics, being released in 1981.

By the late 1980s F1 teams were looking at using CFD to guide simple optimisations. I was at Benetton and we invested in what was then a sophisticated Sun Sparc workstation running at 25MHz and with 64MB of RAM, significantly less than a modern phone. With this and a simplified implementation of CFD called panel methods, we were able to look at some basic surface-pressure profiles of our rear wings. While simple, this was the first time we had moved away from purely empirical aerodynamics.

Although it was exciting to be able at last to gain some understanding and use it to improve performance, the method was extremely limited. It was two dimensional and only really worked to examine flow on the surface. The next step forward came with improved commercial codes in the early 1990s. Computers too were developing rapidly and F1 teams embraced this new technology.

Symonds experienced a primitive early version of CFD in his Benetton days

Photo by: Motorsport Images

While an F1 car achieves phenomenal aerodynamic performance it’s very difficult to simulate because the airflow around the car is far more complex than, say, on an aircraft. Flow detaches 
from numerous parts of the car and breaks up into turbulent eddies and vortices.
 These are even more difficult to predict and we were still hampered by the fact that, while the 3D solvers could handle the more simple parts of the Navier-Stokes equations, they struggled with turbulent flow.

As an industry, F1 was instrumental in pushing the vendors of CFD software to improve the turbulence modelling and it was this which started to make the method deliver reasonable results. The technique was
called RANS, which stands for Reynolds-Number Averaged Navier Stokes and is
 still in common use.

The volume on the surface of the virtual model and the volume of the air around it is
 split up into a mass of virtual cells. The partial differential equations that describe the flow field then have to be solved for each of these cells

Interestingly, one spin-off from the push for better turbulence modelling was that exactly the same problems needed to be solved by wind farms. A single wind turbine is one thing to simulate but a wind farm, where each turbine is operating in a flow field dictated by the turbulent flow from
 its neighbours, is much more complex.
 It’s comforting to know that the push to improve the performance of F1 cars also
 led, indirectly, to better performance of
 wind farms and made a contribution to mitigating climate change.

Even now there are different turbulence models but, as they improved, so too did the accuracy of simulation. Just as importantly, computing power was following Moore’s law of doubling computing power every two years, allowing more detailed simulation of the flow field.

The reason this is important is that the volume on the surface of the virtual model and the volume of the air around it is
 split up into a mass of virtual cells. The partial differential equations that describe the flow field then have to be solved for each of these cells.

For a simple case we’re talking of around 95 million cells
 which, on a single-core laptop, would take around 40 weeks to solve. Hence the need for computing power. Typically a team may be running around 192 cores, bringing the solve time down to a matter of hours.

CFD power has improved significantly since the 2010 Virgin became the first F1 car to be designed exclusively using the technology

Photo by: Sutton Images

Unfortunately, even with improved turbulence modelling, the RANS technique has limitations and the gold standard at present is Direct Numerical Simulation or DNS. This solves the equations directly for all eddies but is so computationally intense as to be impractical for F1 use.

When developing the 2022 F1 car a system known as DES or Detached Eddy Simulations was used, which is effectively a mid-complexity solution which allowed the turbulent wake to be studied in detail while keeping the computing time within reasonable, but still very large, bounds. A further technique known as the Lattice Boltzmann technique or LBM was also used in the studies.

CFD and computing power have developed enormously in a short period of time, such that it’s now a viable proposition to do a significant amount of development in CFD – but what of the future?
 I believe aerodynamic development will expand in two realms.

The first is machine learning – about which I’ll write soon – and also in the area of multi-physics, where a complete system and its environment are co-simulated. For example, the aerodynamic solver will apply loads to the body which will deflect, hence altering the aerodynamics. Even the effects of wet weather could be simulated and full trajectories analysed. It can be done to some extent already, but future developments will accelerate this analysis.

Combining reasonable level of detail with managable computing time is a balancing act

Photo by: Williams