F1 automated data analysis software

[SidSidney]

With the whole Hamilton/Rosberg data sharing spat, as well as comments about a certain driver's ability to understand the data even if he has access to it, I was pondering if that is relevant in the modern F1 age, and how automated data analysis has come along over the past few years.

Here's a couple of images from public releases of the baseline F1 telemetry (appropriately Senna and Hamilton).

1989: Senna at Suzuka


2012: Hamilton at Spa


What is striking is that a) the plots have changed so little over the years, and b) after 23 years they are still drawing conclusions in the garage by writing on a plot in pen.

I would have thought that modern data analysis software could have automated some of that work by now, correlating cause and effect, comparing driver A to driver B, lap 1 to lap 2 etc. to indicate automatically areas of delta, possible causes and solutions. And the output would be a much richer report that offers better insights and a set of conclusions, options and solutions (in terms of car and driver behaviour). Simple things would be:

- simple 80/20 observations e.g. 0.5 seconds slower on main straight is main cause of 0.6 slower overall lap time
- identifying the cause of slower lap times e.g. lower revs in a certain corner, earlier braking, low top end revs
- correlations in data e.g. lower top speed correlated to lower exit speed from previous corner
- anti-correlations e.g. higher exit speed but lower top speed, later braking but slower exit
- trajectory discrepancies e.g. driver A has a faster corner time and takes a different path to driver B
- sub-sector information e.g. he changes gear at meter 250 while you wait until 275
- percentages e.g. you spend 30% of the lap on the brakes, he spends 25%, here's the two corners where the 5% is lost
- cause & effect e.g. same exit speed, same revs at end of straight, 15kph lower top speed -> too much drag, reduce wing

And so on. Of course with experience they can get these results manually, but there must be hundreds of opportunities to get automated conclusions immediately, as part of the basic system output, and present them in a more useful manner that drivers can immediately understand.

I expect they have some systems for this, at least in the back office. Does anybody know of anything?

[Jef Patat]

It's probably not that easy. I have written data capturing software in the past. Capturing, analyzing and diagnosing are whole different things. You are actually talking about diagnosing, you want the software to give you 'a' or 'the' solution. How are you going to bring in differences in for example car setup, driving style, momentary track conditions (decision to take over another car), pushing/cruising, ... Those are all parameters that a human can easily have in mind when looking at a graph but are hard to all input in software.

[Jersey Tom]

You're working under the assumption that those two graphics are representative of all the data analysis work done by the engineers.

Not a good assumption

[knabbel]

With the whole Hamilton/Rosberg data sharing spat, as well as comments about a certain driver's ability to understand the data even if he has access to it, I was pondering if that is relevant in the modern F1 age, and how automated data analysis has come along over the past few years.

Here's a couple of images from public releases of the baseline F1 telemetry (appropriately Senna and Hamilton).

1989: Senna at Suzuka
http://www.creativeapplications.net/wp- ... _Senna.jpg

2012: Hamilton at Spa
http://andyyoungf1.files.wordpress.com/ ... _image.jpg

What is striking is that a) the plots have changed so little over the years, and b) after 23 years they are still drawing conclusions in the garage by writing on a plot in pen.

I would have thought that modern data analysis software could have automated some of that work by now, correlating cause and effect, comparing driver A to driver B, lap 1 to lap 2 etc. to indicate automatically areas of delta, possible causes and solutions. And the output would be a much richer report that offers better insights and a set of conclusions, options and solutions (in terms of car and driver behaviour). Simple things would be:

- simple 80/20 observations e.g. 0.5 seconds slower on main straight is main cause of 0.6 slower overall lap time
- identifying the cause of slower lap times e.g. lower revs in a certain corner, earlier braking, low top end revs
- correlations in data e.g. lower top speed correlated to lower exit speed from previous corner
- anti-correlations e.g. higher exit speed but lower top speed, later braking but slower exit
- trajectory discrepancies e.g. driver A has a faster corner time and takes a different path to driver B
- sub-sector information e.g. he changes gear at meter 250 while you wait until 275
- percentages e.g. you spend 30% of the lap on the brakes, he spends 25%, here's the two corners where the 5% is lost
- cause & effect e.g. same exit speed, same revs at end of straight, 15kph lower top speed -> too much drag, reduce wing

And so on. Of course with experience they can get these results manually, but there must be hundreds of opportunities to get automated conclusions immediately, as part of the basic system output, and present them in a more useful manner that drivers can immediately understand.

I expect they have some systems for this, at least in the back office. Does anybody know of anything?

This is used by some teams:

http://mclarenelectronics.com/Products/Product/ATLAS

What you see written on te paper is only the conclusion off all the analysis of the engineers.

[monsi]

This feels almost too obvious, but it is very clear to me that the more modern example has been through extensive filtering and processing to remove noise and create clear easy to understand traces.

[Jersey Tom]

This feels almost too obvious, but it is very clear to me that the more modern example has been through extensive filtering and processing to remove noise and create clear easy to understand traces.

I'm not sure I'd say that. All the channels shown there (car speed, steering, gear number, throttle, brake, time difference) tend to be pretty smooth with little if any filtering.

Incidentally that "current" plot is from ATLAS, which knabbel linked to.

[SidSidney]

You're working under the assumption that those two graphics are representative of all the data analysis work done by the engineers.

Not a good assumption

I was working under the assumption that there must be there must be better data and decision support software available in F1 nowadays. I just didn't know what it was called and what functions it had.

[SidSidney]

This is used by some teams:

http://mclarenelectronics.com/Products/Product/ATLAS

Thanks for the link. Many standard functions seem to be presentational/combinatorial, but I see they have extensions for simulations and decision support as well.

What you see written on te paper is only the conclusion off all the analysis of the engineers.

Those notes are pretty much directly off the plots - you can see the time +/- trace on the Hamilton shot that shows exactly how much time he lost and where he lost it.

[Jersey Tom]

I was working under the assumption that there must be there must be better data and decision support software available in F1 nowadays. I just didn't know what it was called and what functions it had.

Keep in mind there's nothing stopping anyone from just writing their own support software. If kids in undergrad college engineering can do it, you can bet well-staffed professional race teams can.

There are plenty of commercial / off-the-shelf data logging and analysis software options available of course. Let's just say some are better than others. Or some are less disappointing than others.

[SidSidney]

I was working under the assumption that there must be there must be better data and decision support software available in F1 nowadays. I just didn't know what it was called and what functions it had.

Keep in mind there's nothing stopping anyone from just writing their own support software. If kids in undergrad college engineering can do it, you can bet well-staffed professional race teams can.

Well, precisely. I would expect that side of things to have moved on by enormous amounts, in terms of automating understanding and assisting decisions based on data.

I don't want to build it myself, but I have built and sold a couple of software businesses (the last one sold for $150M) that do similar things in another space that is much more complex to analyze. Those systems capture and analyze about 2-3 terabytes of live data an hour, process it, decide what is wrong and what to fix automatically, in near real-time, and make instantaneous adjustments to about 800 parameters on about 15,000 individual nodes in a very large system that is literally used by several million people every minute. And that is just for one customer.

So I am guessing it is not that complex to download the brain of a race engineer and compare a couple of 10MB files containing data for 10 qualifying laps and draw a few conclusions. I fully expected somebody had already done it.

[knabbel]

This is used by some teams:

http://mclarenelectronics.com/Products/Product/ATLAS

Thanks for the link. Many standard functions seem to be presentational/combinatorial, but I see they have extensions for simulations and decision support as well.

What you see written on the paper is only the conclusion off all the analysis of the engineers.

Those notes are pretty much directly off the plots - you can see the time +/- trace on the Hamilton shot that shows exactly how much time he lost and where he lost it.

The plots with notes is what they show the drivers sometimes, but like I said that is only the result of alot of analysis and not all conclusions are written down. The data is for engineers, 99 out of 100 engineers are visual thinkers, they prefer a visual representation over words, there fore they prefer graphs and specially delta's between logging channels.

There are also a lot of variables why you can virtually never compare two laps directly. Brake points are never the same, they are constantly shifting and depending on the amount of rubber on the track, tire conditions, fuel load, wind direction, wind speed, downforce loss because of a car in front of you and so on.

Keep also in mind that not all data in the pitlane is real time, they have real time info for all important inputs, but they use a high bandwidth burst transmission on the pit-straight for more granular and in depth logging.

But to respond to your points:

- simple 80/20 observations e.g. 0.5 seconds slower on main straight is main cause of 0.6 slower overall lap time
This info is there, they can directly see why you are slower there.

- identifying the cause of slower lap times e.g. lower revs in a certain corner, earlier braking, low top end revs
Data is there but its not that simple. the delta line you show in the graph of Lewis is definitely the best way to show were the differences are on the track.

- correlations in data e.g. lower top speed correlated to lower exit speed from previous corner
Directly visible

- anti-correlations e.g. higher exit speed but lower top speed, later braking but slower exit
This is just a delta of the speed channel

- trajectory discrepancies e.g. driver A has a faster corner time and takes a different path to driver B
Im not sure how precise the GPS information is.

- sub-sector information e.g. he changes gear at meter 250 while you wait until 275
Just a delta of the gear signal.

- percentages e.g. you spend 30% of the lap on the brakes, he spends 25%, here's the two corners where the 5% is lost
This is just a delta of the brake channel

- cause & effect e.g. same exit speed, same revs at end of straight, 15kph lower top speed -> too much drag, reduce wing
Same revs but lower speed is not possible if the gearing is the same if you reduce the wing both revs and speed go up

[SidSidney]

Same revs but lower speed is not possible if the gearing is the same if you reduce the wing both revs and speed go up

Thanks for giving me a way out of that one... Naturally, I was assuming different gearing...

[ImAnEngineer]

I used to work as a vehicle performance engineer at one of the top teams so have a decent amount of experience looking at F1 data. There is a lot of "post" analysis done by various bits of software that can calculate a huge number of different factors - what you see on that plot is the "basics" that can provide easy feedback to the driver.
At the end of the day, vehicle dynamics are very complicated and the best way to analyse data is for an experienced engineer to review it. One of the "tricks" we used was to generate 'best average' laps based on a stint of data. That made it easy to compare the differences between both drivers as you only need to look at one lap.
It is a paper print out with pen marks on it because that is often the easiest way to explain things to someone, and lets them go away and study it while they are eating lunch / sitting on the toilet / lying in bed. Not everything has to be high tech to be effective!