Competitive Intelligence
Zach Catanzareti Photo, via Wikimedia Commons
Known Aliases: Competitor Analysis, Opposition Research
read everything, build tools
Every day as part of my day job, I pay attention to:
what manufacturers are racing, in what series, with which teams
with which sponsors, with which component suppliers
who works for them, their relationships and degrees of separation from me
what people are saying to media, laptimes, published engineering values
When I am working on a particular program, I do deep dives on:
What is the state of the art? Performance levels of existing systems.
How did we get here? The development history of relevant technologies.
Which factors have the biggest influence? Pay attention to when things go right and when things go wrong.
Knowing what is possible is an important step in figuring out how it is done. Specific strategies are determined by making the largest improvements, given the resources and time available. Start with the easiest criteria that are calculated to have the largest effect, while working on others in the long term.
John Carriere’s article and graphs are unbelievably good examples of this process: Coarse To Fine Design
Below is an abbreviated demonstration of reverse engineering performance. The complete process involves gathering all available data, including scraping from videos, or pausing at every key point to type values into a spreadsheet. Then back calculating other factors. Lap simulation of competitors can validate assumptions, reverse engineer missing coefficients, and is critical to Balance of Performance (BOP).
In a presentation a few years ago, I noted the Porsche 919 Hybrid and Toyota TS050 required a Lift/Drag ratio of at least 6.35 to replicate their lap times. This was met with strong skepticism, but much later Toyota came out in the press with a Lift/Drag of 6.5.
Invest whole days into Google Image Search and Excel.
Longitudinal And Lateral Acceleration Targets
2019 FSAE Michigan Endurance winners TU Graz were very competitive in the skidpad event with a best average of 5.085s. Assuming a 18.25m diameter from the rules:
average tangential speed = circumference / lap_time
pi * 18.25m / (5.085s / 1 lap) = 11.28 m/s average speed = 40.59km/h.
average lateral acceleration = velocity^2 / radius
11.28m/s^2 / (18.25m / 2) = 13.93m/s^2
average friction coefficient = (mass * acceleration) / (mass * gravity)
13.93 m/s^2 / 9.81 m/s^2 = 1.42
Which, in my experience, is a very sensible starting assumption for a racing slick.
At 40km/h, I am comfortable negating aerodynamic forces in this example. Furthermore, winning skidpad times have not changed much going back 20 years, before wings were popular. But backing out downforce, side force, and tractive force for your team and for other competitors is a good step in refining tire estimates (even at a point-mass level, though bicycle and double-bicycle will work even better.)
I made a super quick table of 40-90 kW vs 1.2-1.8 CdA for top speed. Based on the peak of 120km/h in the video below, 40kW, 1.8 CdA seems like the best approximation of the 2018 TU Graz car.
Old results (except 2020, with no in-person events and no tech inspection) have team weights. So I also made super quick and rough acceleration estimates to give peak straight speed estimates to the track designers for 2021 and beyond. You can integrate over distance, you can do stepwise distance. In whatever program you’re most comfortable with. For acceleration modeling of a single car, I like to build a lookup table for every km/h. Doing the whole field at once is a little trickier, so using an exponential decay and CdA = 1.8 from the top speed analysis, the best guess for power at the end of the 75m accel event is 50kW for TU Graz.
Hey, look at that: Quick search for the 2018 TU Graz car specifications: 70hp (52kW).
Do a braking calculation, combine with the acceleration and lateral, and you’ve now defined the entire performance envelope of your vehicle, and ideally your fast competitors’.
And if you set the winning times as targets, remember:
they already have a car this fast
they’re supposed to understand why it’s fast (check design event scores, especially finals)
they have a head start working on an even faster car
That’s why it’s not a big risk, though still very, very cool, for them to share the video with telemetry.
Analysis of 2019 FSAE Endurance WinnerS, TU GRAZ
The fastest top speed does not win endurance. The fastest lap does not win endurance. The highest G does not win endurance. The fastest average lap wins endurance. And usually the fuel economy event too, though both drivers’ heavy use of drag reduction helped.
Based on higher steering lock, which is also held longer, it looks like it took the first two laps to get the tires up to temperature.
There are a number of different slaloms:
~40km/h, lateral ~1g
Offset gate before the infield entrance
~70km/h, lateral ~0.8-1.2g
After infield entrance
~50km/h, lateral ~1g
Most everywhere else
One team has done a study on the human limits of minimum period between slalom gates. I hope they study this video as further evidence. The first driver is very fast, but the second is even faster, and with lower lateral spikes. He is so smooth, and never appears to be trying because braking blends into turn in, then throttle blends into exit. There is no waiting to straighten out for throttle, or overdriving by turning in at too high a speed or suddenly applying full throttle. Gossip in the paddock was that one driver really liked and used the car’s rear-wheel steering, while the other did not.
Wings Don’t Work*
*quite as incredibly as hype would suggest
A lot of downforce values are quoted at speeds that are only relevant to oval superspeedways. And a lot of the "drive upside down" example speeds forget the need for drive, braking, and lateral traction.
The fastest corner for the fastest car in the 2019 Michigan Endurance event is around 100km/h and 1.9g sustained.
The downforce increase compared to static:
1.9g (fastest corner) / 1.42g (skidpad event) - 1 = 34%.
Rearranging:
F = ( ( g / mu - 1 ) * mass_with_driver * gravity ) = 0.5 * air_density * ClA * v^2
ClA = ( ( g / mu - 1 ) * mass_with_driver * gravity ) / ( 0.5 * air_density * v^2 )
ClA = ( 0.34 * ( 156 (car) + 80 (driver) ) * 9.81 / ( 0.5 * 1.225 * ( 100 / 3.6 ) ^2 ) = 1.67.
That seems like a small number for FSAE ClA. Plenty of teams claim much higher numbers, (probably including this one) so what’s going on? Considering tire induced drag, I think they might be approaching drag limited top speed around that corner. There may be more lateral grip available, but not enough power to accelerate. There may be an imbalance to the front, preventing more power from being used at the rear tires. Bumps on the track, springing and damping may threaten a loss of control at higher speeds. Conversely, it may still be accelerating like crazy, but the corner is too short to reach steady state.
34% more lateral grip is a big jump, especially at only 100km/h. Achieving that represents a lot of thought, hard work, and development. It is still important to track the resources used, and use lap simulation to weigh the cost in engineering, materials, tooling, mass, etc. vs the total performance gained.
The “usable” ClA could be higher if there is substantial tire falloff due to load sensitivity, perhaps from mu = 1.4 to mu = 1.1 or mu = 0.9. But the second driver’s 1.3g braking down to 50km/h suggests the falloff isn’t substantial. At a glance:
aero drag is probably 0.1g at 50km/h
the front tires probably approach 50% (mass) * 1.34 (aero multiplier) = 67% (load)
probably not quite as high as the outer tires in the fastest corner
those have both high-g load transfer and aero effects
downforce will cancel out some percentage of the lateral load transfer
It’s worth backing load sensitivity values out of a proper braking calculation, and repeating all affected calculations, but I don’t think there’s much load sensitivity in the range those tires are used.
The other alternative is enormous load sensitivity, and the outside tire lateral forces are at their absolute peak.
EV BATTERY C-Rate
EV racing is all about thermal limits. That is the first lesson I was taught. For any battery, controller, or motor, there is a constant power level (with cooling) that reaches a steady state temperature. Running above the constant power level, closer to peak power, causes the component to heat up, and requires time below the constant power to cool back down. Integrate versus time above or below constant power and combine with an efficiency map to manage heat energy.
The constant power of the battery, controller, and motor should be closely matched to optimize the system. By either sizing the battery, controller, and motor for a constant power of 80kW, or strategically using peak power over short periods, it is trivial to build an 80kW EV compared to building an 80kW IC with a 20mm intake restriction. There is the difficulty of finding traction for 80kW, though AWD is also comparably easy in EV. The big issue is supplying enough power * time = energy from the battery to complete the endurance event. Lithium-Ion battery cells are often marketed as Energy Cells prioritizing kWh/kg, or Power Cells prioritizing kW/kg. Both values are important to maximize, but there is a very easy way to decide between the two.
In addition to researching competitor’s total battery capacity, establish a range of C-rates, which describe how quickly a battery is being charged or discharged. The C-rate for the whole battery and for individual cells only differ due to the statistical noise of imperfection. Conceptually, they are 1::1. C-rate is simply 60 minutes divided the charge or discharge time in minutes. A C-rate of 1 discharges the capacity in 1 hour. A C-rate of 2 discharges the capacity in 1/2 hour. A C-rate of 12 discharges the capacity in 5 minutes.
For a low estimate, between 2017 and 2021, the slowest FSAE EV Endurance finisher took 2419 seconds. Assuming a very conservative 60% battery usage from 80% down to 20%, C-rate = 3600s/(2419s/0.6) = 0.9. The fastest 2017-2021 FSAE EV finisher took 1490s. Assuming a very aggressive 90% usage, and a wild guess of 20% increase from regen, the C-rate = 3600s/(1490s/(0.9+0.2regen_in+0.2regen_out)) = 3.1. And those are just the average C-rate. Some teams are using bursts of 80kW under acceleration and regeneration, which on a typical 5kWh to 9kWh FSAE battery represents a peak C-rate of 9 to 16. If this was a professional assignment or if I was a student competitor, I would add as many European FS EV teams as possible to my data collection.
The dividing line between Energy Cells and Power Cells is usually a C-rate of 1. Energy Cells may simply overheat at higher C-rates, or possibly lose so much energy to heat that there is less usable capacity than a Power Cell at the same C-rate. The interaction between lower internal resistance and energy density is beyond the scope of this article. However, understanding the electrochemical causes could help choose between different Power Cell options.
EV AERO Philosophy
In IC, aerodynamic downforce and drag reduction both enable faster lap times. Lap simulation and dynamic testing show an active drag reduction system (DRS) that reduces drag and downforce in a straight line results in slightly faster lap times. FSAE cars don’t spend much time at high speed in a straight line. FSAE cars do spend a lot of time at lower speeds with high lateral and combined acceleration. It’s the same problem as all non-oval racecars: The highest aerodynamic forces occur where they matter the least.*
*Fair enough, it will brake harder and faster at the end of the straight.
The lift::drag (L/d) ratio has dominated road racing for 50 years. An increase in L/d almost always means the laptime saved with faster cornering speed is greater than the laptime increase from going slower down the straights. There are subtle variations: Le Mans wants the highest L/d by chasing the lowest drag, Monaco wants the highest L/d by chasing the highest lift. But EV racing requires another layer of analysis.
For the EV endurance event, energy capacity and management (including thermal) are the deciding factor in almost every design or strategy aspect.
FSAE EV at full 80kW for 20-40 minutes = borderline inconceivable
Using full throttle whenever full traction is available for 20-40minutes on an FSAE EV course = unlikely
Now it’s not only the laptime effects of aerodynamics, it’s the relationship between energy and laptime delta.
Higher cornering speeds decrease laptime.
And may decrease the acceleration energy needed to maintain average speed.
Higher drag at top speed is an abominable energy drain.
And make the straights slower.
Lower top speed and higher drag reduce the amount of energy available for regeneration.
An aerodynamic package for an EV racer needs to focus on energy and efficiency. With significant aerodynamic surfaces, so-called DRS cannot be a minor afterthought.
When traction limited, drag barely matters.
When power (or energy limited), downforce barely matters.
Chasing fixed L/d is obsolete.
Minimum drag that can switch to maximum downforce should be the core target from the start of design.
Additional EV 101:
Build all of your EV powertrain, dynamic, lap simulations in metric. Battery cell suppliers speak in kWh, electric motor and controller suppliers speak in kW. If you try to engineer in horsepower-inches you will take constant damage from unit conversions. If your team already has a good understanding of lap simulation, an interesting question is when the extra energy regenerated and deployed in a FWD car equals the better exit traction of a RWD car. There’s probably a Power::Mass::Drag::Downforce ratio where FWD, AWD, and RWD are well balanced. Maybe we’ll see some good future touring car racing come out of that.
