Throughout this Formula 1 season we have seen an increasing display and interest in data analytics through on-screen broadcast graphics. The feedback we have had on the F1 Insights powered by AWS graphic series has been resoundingly positive and, therefore, we as a team are producing more of these insights to help tell the story through the data. My feeling is that in two years’ time, fans will have complete reliance on data analytics to help them understand Formula 1 racing at a deeper level. This is no different to what we do within the Formula 1 teams’ environment - using data analytics to help inform our decisions - and I’m proud that this is also becoming the case also for Formula 1 fans. The next graphic that we are going to look at within the F1 Insights series really tells the story of the heart of the Formula 1 team. As an F1 engineer who has worked in a team setting for many years, I know that car development is the very reason for the technical teams’ existence – to develop a car faster than your Formula 1 rivals. If you can do this, you will not only win races, but also win world championships. It is a simple formula.
Therefore, this new Car Analysis / Team Development graphic is set to go behind the curtain of how teams develop their cars, how quickly they develop their cars, and what the on-track result is. We will look at the important aspects that make up an F1 car performance and break that down into something that is understandable for the F1 fan. We hope to give you an understanding of how the chassis and engine teams develop their cars, and ultimately how quickly they can develop them.
Car development within the Formula 1 team setting can be broken down into four specific elements:
Aerodynamic Downforce – This is the force that pushes the car and, more importantly, the tyres into the ground while the car is cornering. The more the car is pushed into the ground, the higher the lateral acceleration (or mid corner speed) it can generate. Therefore, downforce is an extremely important element of an F1 car.
Aerodynamic Drag – With any aerodynamic entity, such as a wing section, when you create downforce, you will also create some level of drag. The drag is the resisting force; the air force that is trying to slow the car down as it moves around the track. Therefore, the F1 chassis development teams are constantly trying to reduce drag, or at least add downforce with as little additional drag as possible. We call this aerodynamic efficiency, and this is explained in a later section.
Engine/Power Unit Power – This, as I’m certain we all understand inherently, refers to the amount of power that the engine/power unit produces. Quite simply, the more power, the better the lap time. Although this is a very crude analysis, it does give a view on the overall resultant of power. The power unit teams and manufacturers are constantly trying to develop more power within their power units in order to give the chassis teams a better overall lap time. Performance from the power unit is derived principally in a straight line; we call this the power limited section of the track, and the more power we can add at this point, the faster the straight line performance of the car will be.
Tighter Grip – This is another very sensitive performance parameter for a Formula 1 car. The higher the tyre grip, the higher the cornering performance, and therefore the lower the lap time. It is very similar in effect, to the aerodynamic downforce. We should consider that at present we have a single tyre supplier in Formula 1. So, for the purposes of this graphic, it is considered identical across the cars. We only consider the aerodynamic downforce/drag, and the engine power. Of course, we know that there are many cases where the tyre grip between cars for an identical compound on the same track, at the same time, will be slightly different. This can be due to the way the tyre is operated between different cars, and this results in a slightly difference performance. Over the average case, and due to the fact that we have a single tyre supplier, we can consider that the grip should be the same between the cars.
Let’s consider the chassis team and the aerodynamic development in a little more detail. The two main aerodynamic tools that the teams use are the wind tunnel and computational fluid dynamics. Teams are essentially trying to develop more downforce with either the same level of drag, or less. They want to develop a more aerodynamic, efficient car. It is worth noting that within modern day Formula 1, the aerodynamic testing or development is restricted. This regulation was introduced some years ago to reduce the financial burden on the teams in terms of the arms war that had developed around increasing aerodynamic performance of the car. With these clever regulations known as aerodynamic testing restrictions (ATR), the teams are now restricted as to how many wind tunnel runs they can carry out, or how many computational fluid dynamics (CFD) simulations they can carry out, all in the quest of developing their car. This means that they need to be incredibly efficient with their development methods as each and every run must count.
With the calculations and simulations that we have developed, we want to be able to reverse engineer these development cycles across teams, and be able to give a moving picture throughout the season of how each and every team is developing compared to its rivals. This will give us a really clear picture of how the season is emerging from team to team and, even more importantly, with development trends and what could happen in later races. For example, will Red Bull be able to catch Mercedes this season? Does their development trend indicate that this will be the case? Who is winning the midfield battle in terms of development? How quickly can a midfield team develop its aerodynamics compared to a top team? We hope to be able to answer all these questions with this new F1 Insight graphic.
In the section above we explained the performance parameters that the teams center on. But it is also important to try and explain how the teams set the targets for both the downforce and drag parameters. To do this we need to consider how both parameters affect the lap time. In the plots below we express the effect on laptime of the downforce and drag performance indices. These are known respectively as CL and CD which refers to the fact that we are really considering the co-efficient of lift and drag rather than the actual lift and drag which is affected by ambient conditions and more difficult to monitor during development and correlation exercises between the track and the wind tunnel.
What these plots show is the sensitivity of CL and CD to lap time. We express this as percentage of lap time reduction for every “point” increase in either CL (downforce) or CD (drag). A “point” refers to a percentage point so for example if the coefficient of lift, or CL, for a particular team was 4.80 and they brought an upgrade that was worth 1 point increase then their CL value would increase to 4.81.
In the example above, we have highlighted the Mugello track sensitivities with the red bar. This shows that the CL sensitivity is -0.033% lap time per point of CL, whereas drag is defined as 0.068% laptime per point of CD. Let’s look at another example:
A team brings an upgrade which has 5 points more downforce but 3 points more drag to Mugello. Does this make the car faster overall?
• CL – 5 x -0.033 = -0.165%/lap (faster)
• CD – 3 x 0.068 = 0.204%/lap (slower)
• Net effect = -0.165 + 0.204 = 0.039%/lap slower (for a ~1m15s lap time like Mugello, this equates to 0.029s/lap slower)
Therefore, the team would not bring such an upgrade, as this would be deemed “aerodynamically inefficient.” What we look for then is the “break even” (B.E.) efficiency – the amount of drag that you can add in relation to the downforce which would result in the same laptime. This is derived from the simple equation below:
In the case of the Mugello circuit, we have the following, therefore:
This means that for every unit of downforce (or CL) added, we must not add any more than (1/2.06) units of CD. If we want to make the car faster overall, we need to develop the downforce and drag above the B.E. efficiency.
The modelling starts from picking a reference team/car for each circuit. The team could be different race by race, or we can keep it constant. Currently, the most obvious team to take as reference is Mercedes, since they are more often than not the fastest.
By considering the teams lap times and how these compare with respect to the reference, we can infer the difference in aerodynamic downforce, drag and engine power.
The all-important aerodynamic downforce combined with tyre grip informs us of the performance of the car in cornering, while engine power and drag give us information about the performance in a straight line.
For the level of accuracy required in this application, we can consider that every car has the same level of grip for a given time of day of the circuit. That is to say, the simulation accounts for grip across different days but not for differences in tyre grip for different cars at the same time of day for a given circuit. Of course, there are edge cases where a particular car has not been able to generate enough heat (=grip) in a particular tyre set, but these are not accounted for in this particular version of the model.
Whereas there are 10 chassis teams all working on the aerodynamic aspects of their particular cars, with power unit development, we are restricted to just four manufacturers. In terms of the power levels, these are limited to Mercedes, Ferrari, Renault, and Honda. Note that by regulation the performance levels between a “works” team compared to the customer team has to be identical. For example, the Power Unit (PU) performance between the Mercedes GP team and Williams, who also used the Mercedes PU, will be identical as supplied by the PU manufacturer.
In order to evaluate the differences of each team with respect to Mercedes, we simulate a lap that fits the telemetry one (we call this lap simulation, or “lap sim”).
On this initial simulation we tune the model input parameters - downforce, drag, power, in order to obtain the sensitivities. The sensitivities tell us how much the lap time varies as a consequence of the change.
We then compare the telemetry of the particular competitor to the telemetry of Mercedes and thus, we are able to calculate the “delta time” (the time difference between the two speed signals when the x-axis, or domain, represents the distance around the track) in each position of the track. We then reconstruct the delta time using the pre-calculated sensitivities. By using regression techniques and optimizing for each of the configurable performance parameters, we can quickly arrive at a “reverse engineered” car that tells us the differences in downforce, drag, and power with respect to the reference. This is represented in the figure below, in which the measured delta time [lower signal] in red can be compared against the delta time derived from the simulation method (black). It can be noted that the “fit” (how well the two signals match) is very close, which tells us that the optimization method has produced an accurate estimate of the downforce, drag and power differences. Of course, the fit cannot be perfect as the driver also plays a role this – humans will react differently to a simulation but I think you can see that in the case of Formula 1 drivers they are pretty close!
As we mentioned before, the power and drag have a very similar effect in terms of the “delta time.” The method that we have used to account for this is to consider that each track has a slightly different effect for both engine power and aerodynamic drag, and that the power is a function of the engines, while the drag is function of the car. Of course, the “optimization” of these two parameters would remain affected by some uncertainties given the closeness of their effects. Building up a picture over several circuits will help to increase the accuracy and at this stage of the season the models have produced an accurate picture.
In general, the delta performance is better represented by the delta lap time at each qualifying lap and the sensitivities tell us how this delta laptime has been made.
In the plot to the left, we have the deltas to Mercedes for all the teams (for each team we have considered the fastest driver). The deltas are normalized to an 80-second lap, and a positive number means it is slower than the reference, which in this case is Mercedes.
The results of the fitting on the downforce, drag and power are reported in the following plots. The first, CL, is the downforce coefficient delta in percentage points with respect to the reference car - Mercedes (i.e. if the reference car has a value of 5 and the considered car has a value of 4.85, the CL in the plot would be (5.00-4.85)*100=15.These are referred to as points). The second plot shows the same output for CD, which is similar to the CL methodology, but in this case refers to the aerodynamic drag. The final plot, PW, is the delta power in KW of the engines with respect to the reference, Mercedes.
In general, teams have the ability to change the CL and CD – this is usually driven by rear wing changes. Teams do this in order to adapt their cars to the characteristic of the given circuit that is defined by the sensitivities explained earlier. Hence, it is possible that some cars are slightly more competitive in certain circuits rather than others, as an optimization of their CL/CD package could generally work better than others. Still, teams will not be able to run their “best” package in all the circuits, as it may not be optimized for that given circuit. In fact, a circuit could demand a lower CD level than the car’s standard rear wing. As mentioned, the CL and CD are coupled, and lowering the CD could result in a less optimized package for the circuit – it becomes the trade-off between the two effects. For some teams, the trade-off is better optimized by increasing the CD level while a better overall package is sought by others by decreasing the CD. All teams are driven by the circuit characteristics and demands, however, unless they have a big enough performance margin that would allow them to run a non-optimal setting from a simulation point of view. They want to balance requirements to account for driving style or tyre wear, for example.
We can translate CL-CD-PW deltas into delta lap times. This can also be reported in a lap time for an 80s lap (it is simply the lap time in percentage, multiplied by 0.8).
If we focus on the delta between the Red Bull and Mercedes teams, then we can see that there is a trend this season of RBR catching MER. The data has a certain amount of noise, but we should note that this is entirely normal for this kind of data with such a high degree of variability). The noise within the data originates from the aforementioned variables amongst which there is also the driver repeatability.
We can smooth the delta lap time using a moving average over three races and see the trends with less noise.
Being able to process the timing and simulation data in the manner described above, we are able to estimate the key performance elements of each car. This allows us to spot trends during the year and see which cars are gaining and which are losing. Where are the exciting battles going to emerge in the race for precious points? We can visualize the important role that the technical teams play in this behind-the-scenes game of chess.