After more than 25 years in Formula 1, for me, there’s never been a more exciting time. We are at the advent of using profound levels of data analysis in order to shape the future of the sport. In partnership with AWS we are innovating just to this effect and have utilised data analytics to shape the 2021 regulations via HPC cloud computing and leading-edge technology in computational fluid dynamics. Last year, we also launched ‘F1 Insights powered by AWS’, a series of graphics that are bringing data analytics to the live feed of the TV production. In 2019, these three graphics gave key, and previously unseen, insights into the inner workings of Formula 1 and brought them out into full public view for the enjoyment and education of our fans. As the first in a series of graphics which utilise car telemetry, timing data and all the other data input feeds that we have access to in Formula 1, they represent exactly the methodologies used within the teams and so demonstrate to the fanbase the power of analytics and key insights into the decision making of Formula 1 teams. This year, we plan to introduce six brand new F1 Insights graphics via the AWS and Formula 1 partnership. As with the 2019 graphics, these will make key use of car telemetry and timing data and give further insight to our fans. The first of these on-screen graphics that we will introduce in Austria will be Car Performance Scores. This graphical application will get under the skin of the important aspects of Formula 1 car performance. From the very first race of the 2020 season, the fans will be given a much clearer understanding from the very outset of how the different cars perform relatively to each other. This is the first insight of its kind with the aim to convey to the fan a much better understanding of where their favourite car ranks in the Formula 1 field. Enjoy!
Car performance is made up of many different aspects. What we want to try and convey with this latest F1 Insights graphic are the major building blocks that make up car performance; namely cornering performance, straight line performance and car balance or handling. The teams work relentlessly in trying to improve these three aspects. For good cornering performance, you need a lot of down force to have centred the tyres in the correct working window and to have a car that is well balanced throughout the different phases of the corner. Straight-line performance is instead much more about engine power and aerodynamic drag. These are the main building blocks that the teams will be working hard on as they get to Australia in order to maximise each area of the car performance.
Within this graphic we want to convey all the aspects mentioned above. In microscopic detail we intend to use the car data in order to understand the speed of the car through a given corner. We then split this between high, medium and low-speed cornering at certain thresholds for both the low, medium and high-speed cornering aspect. This will give the fan key insight as to where a car is stronger or weaker than the opposition. For example, last year’s Mercedes was incredibly strong in the low speed corner but was less so in the high-speed corner when compared to the Ferrari and in fact the trend completely reversed when we considered straight-line performance. Although this was clearly evident to those of us that understand car telemetry analytics, it was less so to the fan and therefore in my opinion, we missed telling a key story of the season.
Just as with the high, medium and low-speed cornering performance, the straight-line performance also takes key car data and uses that to understand where a particular car is strong in a given, normalised section of the track (i.e. the straights). This will help us understand who has the best engine or lowest aerodynamic drag package which are two of the key performance attributes that the teams and engine manufacturers will be working on relentlessly as we speak.
The final aspect of this graphic is to consider the analysis behind the car handling or balance. All cars handle slightly differently, and it is the job of the designers and engineers to try to build a car that handles perfectly through every phase of the corner. At the turn in phase (the very initial part of the corner) the rear grip is extremely important as this helps to give the driver a lot of confidence in turning the car in. As the car speed reduces and the vehicle moves towards the mid corner, the front grip becomes more dominant and more important. If a car washes out or doesn’t have enough grip on the front axle, we call this understeer. If instead the rear axle doesn’t have enough grip, and therefore the rear axle tends to slide more than what is required to execute the corner, we call this oversteer. On the corner exit, the importance in car handling then swaps to the rear axle. As the driver picks up the throttle and requests power through the rear tyres, he requires a commensurate increase in rear grip to keep the rear stable. The underlying models within the graphics will convey to the fan if the car is understeering or oversteering in the low, medium and high-speed corners and will so help to appraise which car handles better than the others.
All of the above will help to frame the car performance for the fan and they will be able to understand in microscopic detail who has the best car in the low-speed corners, the medium speed corners and the high-speed corners, in a straight-line or who has the best handled car. These are all the main attributes which are needed to make up a fast racing car. So how did we do it? Let’s look at the math and physics of the underlying models.
Low/Mid/High /Straight performance
Some cars are quicker in low speed, others in high speed while others perform best in the very low speed parts of the track. In order to quantify this performance, each track has been cut into different sections; low speed, medium speed, high speed and straight-line speed, using real car telemetry to identify each section.
Using the car speed, lateral acceleration and other parameters, it has been possible to define the start and the end point of each corner and straight and to include them in the specific section. It's important to notice that not the entire track is included in some sections, only the parts which are in respect to the designated requirements; i.e. if a section doesn't have enough lateral acceleration to be included in a corner definition but the driver is still turning too much to be considered a straight-line, this section is not included in any metrics.
With this information in our hands and the distance travelled by each car, it's possible to calculate how much time the cars are spending in these sections and then to rank them accordingly to their performance.
The same concept is applied to the straight-line performance; using the same approach it is possible to identify the main straight-line sections of each track and to calculate the time spent there. These values are then normalized in a range between 0 and 10, using the best possible lap as reference for all comparative scores, with 10 being the best performance possible.
Example, AUS Q3:
Low/Med/High speed balance
Although we often hear a lot of discussion about oversteer and understeer, it's not easy to understand from which condition the car may be suffering from at any given time that is affecting their performance.
In order to help to answer this question and make it clearer for others to understand, we have built a vehicle model which, using some vehicle telemetry channels, is able to calculate how much the front tyres and the rear tyres are sliding.
The goal of the model is to calculate the so-called tyre slip angle, which is the angle between the lateral speed and the longitudinal speed of the tyre; the higher the lateral speed, which means the tyres are sliding massively, the higher the angle. If the front axle is sliding much more than the rear, the car is not turning so much and that is called understeer; if the opposite is true, the behaviour is called oversteer.
In order to calculate these values, we need to start from the relationship between the rotational speed of the car as measured by the gyro sensor and the rotation of a hypothetical point which is turning the corner:
ω_gyro is the car rotational speed as measured by the sensor installed on the car
ω_CoG is the rotational speed of the hypothetical point which is turning the corner
∆ω is the difference between the two, what we are looking for to define the car as under steering or oversteering
ω_CoG is defined as
Where a_y is the lateral acceleration measured by the accelerometer and vCar is the car speed calculated by the wheel speed sensors.
What then we are looking for is ∆ω= ω_gyro- ω_CoG which can give us an indication of the understeer/oversteer level of the car
Then the lateral and longitudinal speed of the car can be calculated, we call them vxCOG Estimated and vyCOG Estimated. As final step, using the lateral and longitudinal speed of the tyre calculated through the yaw rate, it’s possible to calculate the slip angle
The relationship between the value calculated on the front with the value calculated on the rear axle, can give us an idea of the understeer or oversteer level of the car. The more the front is sliding the higher the understeer level, while higher rear sliding is linked to an oversteering car.
In order to simplify the reading of this number, we have created a metrics which is limited between -10 and +10; whenever the number is smaller than 0, the car is understeering, when the number is greater than 0, the car is oversteering.
Example, AUS Q3:
Uncovering the stories through data
As you can see from the previous section, the model output is exactly what we would expect to see with this type of analysis that goes on within the teams. Cornering output and that of the straight line is essentially a time to pass through a given sector of the track. This gives us the ultimate arbitration of who is fast and who is slow and who lies in the middle in these important track sectors. We then take these numbers and normalise them in a scale of 0-10 to give a simple output which is easy to understand. So, for example, the worst case or slowest car that we could ever see would be 0.0. Whereas the best case or minimum time through a particular sector, would be a 10.0. In reality, we will expect to see number between 0 and 10, but never actually 0 or 10. A score of 9.0 or above would indicate that the car is extremely strong in that sector. However, it is all relative as one car could be a 9.5 but the best car could be a 9.8 indicating that it took less time to pass though the given track sector.
To understand further, it must be pointed out that the limits of 0-10 are set the same throughout the race weekend. So, let’s say we expect in a given sector the absolute worst case to be 25 seconds to pass through from the entry to the exit of the sector in question, whilst the absolute best case would be 20 seconds. In this case then, 20 seconds would represent a 10.0 and 25.0 seconds would represent a 0. As we go through the race weekend the best car in FP1 may take 23 seconds to pass through that sector resulting in the following score:
((25 – 23) / (25 – 20)) *10 = Score = 4.0
but as the weekend progresses through to qualifying, that time will reduce. This is caused by the track rubbering in, the driver gaining confidence, the handling getting more towards the drivers preference, the fuel level reducing and the PU performance being turned up to maximum. Therefore, our best car in P1 that took 23 seconds to pass through the sector, by the time qualifying takes place, could be closer to 21 seconds, thus extracting a score of.:
((25 – 21) / (25 – 20)) *10 = Score = 8.0
We should therefore expect the index to increase as the weekend progresses; to be explicit – our best car in the imaginary sector would score a 4.0 in Friday P1 and an 8.0 in Saturday qualifying. P2 and P3 we would expect to lie between these two values as all the afore mentioned parameters move towards the driver being able to exploit a faster lap time. This is the case for high-speed, medium-speed and low-speed cornering performance and straight-line performance.
For the car handling analysis this is a little bit different - instead, we try to convey the levels of understeer and oversteer. As explained in the model section, this output is based on the relationship between how much the front tyres and the rear tyres are sliding relative to each other – thus the ratio between the front and rear axle. However, just as with the straight-line performance and high-speed/medium speed/low-speed cornering performance, we have tried to normalise this in a scale so that each driver and their car can be represented from -10, which represents understeer, to +10, which represents oversteer. It’s important to recognise at this juncture that all cars have some level of understeer and oversteer – the car is never perfectly neutral. However, we’re trying to convey via this graphic and underlying model when a car has too much understeer or too much oversteer. For context, if a car is oversteering badly, it will have a very high positive number, but if a car is much more towards neutral or as us Formula 1 engineers would say, ‘well balanced’ then the car can and will have a small amount of oversteer (a small positive number, between 0 and 3) or a small amount of understeer (a small negative number, between 0 and 3). We should therefore use this normalised scale to try and understand whether a car has excessive understeer or excessive oversteer and is therefore not well balanced, causing the car to be slow in a particular corner type.
All of this allows us to tell stories via the data as to which car is good or bad or somewhere in the middle in the various sectors. By using the normalised scale, we can convey in very simple terms, the best car and the worst car and everything else within that scale. It is hopefully a numerical output that can be understood by everybody. During the designing stage we compared the method to the old-fashioned Top Trumps where the higher number is better. For the cornering performance and straight-line performance, this is definitely the case, although the car handling as explained is slightly more nuanced. However, we believe that once the fan base and commentators are used to these numbers, they will be much more readily able to tell the important stories about car performance.
In summary, we are hopeful that through this latest F1 Insights graphic we have not only been able to convey some really important data points concerning car performance but also that this blog is helpful as an understanding aid for fans and commentators alike. This is the very first of six truly insightful graphics that we will build across the AWS and Formula 1 partnership in 2020. With each graphic we hope to give the Formula 1 fan more insight and more data driven enjoyment, education and engagement in consuming this highly technical sport. For the Formula 1 teams, data has been the life blood for many years. It has helped to shape the sport itself as well as producing the fastest racing cars on the planet. With these insights, we hope that data now becomes the life blood for the consumption of Formula 1. It’s never been a better time to be a Formula 1 fan.