Thursday, October 20, 2016

Speed up your race data analysis

Going through your data from a race weekend, or a test day can take a long time.
Racers do not have a lot of time. So, tough deal, right.

Yep.

Below are some of the tips based on what I have learned over the years to "optimize" how data analysis time is spent. After all, the goal is not, necessarily, to spend less time reviewing your (or another driver's) data - the goal is to get a return on your investment.

What's this return I speak of? Think of it as "Actionable Insight". What does this mean?
Insight - "The driver and/or engineer learned something he or she did not know." Or, perhaps,  confirmed something that was only suspected. Or even better - disproved something  that was suspected, putting an end to some crazy theory.  Examples:
  • "Turn 2 has a lot more impact on lap times than Turn 1"
  • "The car loses power once Oil Temperature crosses 240 degrees"
  • "I did not forget how to drive a race car, all of a sudden - I just need to do better compensating for overheating tires, next time"
  • In left hand corners, at low speed motion range, right rear shock spends more time in compression than in rebound
Actionable - "The driver (or the engineer) is willing and able to do something about what is discovered", - i.e. to take advantage of the Insight gained. Examples:
  • "I will focus on Turn 2 Entry, even if it has to come at the expense of Turn 1 Speed"
  • "I am going to put secondary oil heat exchanger in, and check the data again to see if that "postpones" the power loss until later in the race
  • Next time I am waiting on the grid, I will remind myself  (or ask my crew to remind me) to stay aware of tire conditions if the car changes its handling halfway through the race
  • Ask the crew to turn down compression damping on the right rear, re-evaluate damping dynamics after the next session, as well as changes in the car's behavior in left hand corners
Some of the things one may learn from the data  may appear interesting and even useful, but are not "Actionable Insight". If you want to maximize effectiveness of time spent with data, such discoveries can be classified, generally, as a waste of time. One exception is if you are just learning capabilities of a particular analysis method/approach, so your goal is practicing the process itself, rather than gaining an 'Actionable Insight"

  • I am faster than Carl. Carl sucks at this driving thing
  • Now that I put on the supercharger, I can go 163 mph down the straight
  • My fastest lap is faster than some other lap
  • Carl gets on the throttle much earlier than me, in turns 2 and 6*
*The last one pretends to be an "Actionable Insight"- but it's a fake! Let me explain.
When was the last time you were in the corner, and thought "Hmm. I just don't feel like getting back to the throttle. It's too much work. I'll just wait and see if someone else would do it for me"?
In the majority of cases where there is delayed throttle application, the driver has a reason for it- very often some very good reason, that indicates that something is taking place at a deeper level, and the onion is begging to be peeled! So, strive to peel the damn thing, until you get at that reason - which can be addressed (or, perhaps, it would make clear that there is nothing to be addressed and the focus should be shifted elsewhere), either way, an "Actionable Insight" is likely to follow, for  example:
  • The driver turned in early and failed to rotate the car decisively. Getting to the throttle early will cause understeer, and the car will run out of track at the exit
  • Car is using all of its grip, and adding throttle will cause either front or rear to lose traction
  • The driver thinks that the car is using all its grip and adding throttle would cause the car to lose grip
  •  There is another car in front of the driver
  • The seating position makes it difficult to depress the pedal fully until the lateral force is decreased (YES, THAT HAPPENS!)
  • There is another car behind the driver, and the driver is too distracted watching his or her mirror.
  • The driver is not looking far enough ahead
  • The driver got on the throttle too aggressively initially, had to go off line to "save it", and is now hanging onto the wrong side of the track.
I could go on, I have another dozen lined up :) The point is - not figuring out what it is, that can be addressed to get to the throttle earlier, makes the insight non-actionable. Thus, mostly useless. On the other hand, peeling the onion will make you cry. I mean, will result in you gaining an Actionable Insight (which may or may not make you cry...)

OK. Here are the promised tips:

1. Invest time into learning about capabilities and limitations of your data system, and analysis software you are using. Are you only using 20% of analysis software capabilities? 50%? 90%? Are there more efficient ways to go about your analysis? What if you plugged those numbers into Excel? What if you put this on one screen, and this on the other one? Never assume that what you are able to do today, is the most effective way of achieving your goals. Just like your driving... You got better at driving by methods other than doing the same thing over and over, this is no different.

2. Take notes at the track, and review them before you dive into the data. This can be a huge time saver - in both picking the right sessions/laps too look at, and at setting goals for what kind of  "Actionable Insights" you are after. I, sometimes, get out of the car and take a note "need to check/confirm XYZ in the data". Even if it's weeks before I look at the data from the session, the note is there, so I can focus on what was important (and it often serves as a tool to recall specific events that prompted the note)

3. Same as above, with the video. If you don't have time to review whole session's worth of video before reviewing the date, at least watch 2 or 3 complete laps, and a few seconds for all other laps (this will remind you what kind of traffic you were dealing with, and what the track conditions were).

Note: Some systems will allow you to integrate video with your data analysis software, so as you scroll the data around, the video frames follow (to be clear, I am not talking about tools that just overlay/paste few selected channel values onto the video, to produce a video you are going to show off to your youtube "friends" -  but, rather, a solution where the video becomes one of the analysis channels).
If you can afford time and effort to set this up, it will take your analysis to the next level, guaranteed.


4. Use "Channel report"/Section time summary views. They will save you A LOT of time. Set them up so they show you some measures of braking, cornering, and acceleration performance in critical sections, and you can easily track both consistency across laps/sections, as well as any patterns that may develop lap to lap. I spend as much, if not more, time in various summary/statistic-type views as I do browsing through actual channel traces.
Here's an example of a simple view - for each channel, some measure of its values is summarized for each lap:


Here's more advanced view - now summarization happens per section (more on sections below). I know it looks busy, but you are taking in A LOT of information quickly, if you practice to interpret these, as opposed to "trace" views, where, at best, you are only getting a lap or two's worth of data before you have to scroll around:


5. Related to above - ensure that you have pre-set map sections set up for the tracks you drive the most. Here's the main method I use to set up track sections/segments:
  • Each section begins just before "normal" beginning of braking point. This confines any issues resulting from braking performance to a single section, the same one as the one most affected by such changes.
  • If a corner is followed by a straight with a single upshift, that straightway is part of the same section as the preceding corner.
  • "Long" straights (2 or more shift, or some other separation method), get their own section
  • Corners that are "stringed" together in such a way that changing something in the first one affects how you drive the following one, become parts of the same section, even if there is brake application in between.
The idea behind such method is that any performance (car or driver) patterns specific to a specific type of situation (i.e. braking for a specific corner, accelerating out of left hand corners, issues with a racing line in a specific corner) are confined to corresponding sections. For example if something happens under braking for turn "11" (sharp blue left hander on the map below, with number 6 over it), it would be confined to section 6, as would most of the consequences.
Here's an example, with annotations illustrating the method described:



  I have seen some drivers set up sections so that the sequence of numbers matches official track maps. There's little value in that, compared to the benefit of an approach like the one described.

6. If you are not getting anywhere, watch the video of the session. Nothing works as well to focus your analysis as "immersing" yourself into the actual experience of being in a race car on the track.

7. Check and maintain calibration of all critical sensors. For a discussion of accelerometer correctness, including an explanation of why it is critical, and procedure to correct as needed, see my earlier post

I hope you can find practical use for these. Remember: "Actionable Insights"!