Isle of Man TT | 2023

Easiest fix? TC/off.
Hook it up to, yet, another button?

 

Need a light on the dash when it is being activated.

 

What did that math end up looking like? The aero must have been nearly impossible to model. The DTC intervention would've been in the data if it was doing it.

 

Actually it's not very hard at all. I decided to ignore changes in aero efficiency, because over a single straight I'd stay rellllatively equally tucked the entire time.

In order to find the aero coefficient, I took advantage of a bike-provided "force" channel (fx_rear_acc, described as the "rear wheel acceleration force"). I didn't trust that it had accurate values, so I didn't want to use it in the HP calculation, but I did decide to trust that even though I thought it was wrong, it was likely wrong by a constant factor. I used that to reverse-model the bike's mass by subtracting out the aero drag force and then dividing by overall acceleration and then raising or lowering the K_aero constant I was using until the calculated Bike Mass stayed constant(ish) over an entire acceleration straight:



Once that was found, I plugged the same aero constant into my own force/power calculations, using a constant 300kg (instead of the calculated channel above, which would've become circular).




The result is this: We're looking at the short straightaway between Quarterbridge and Braddan Bridge, which is very flat and straightforward. I power out of QB in second gear, stand the bike up and get to full throttle the entire straight, going through 3rd and into 4th. The bike, even at only 5% lean, thinks the rear is spinning about 4.1% faster than the front (139mph vs 133mph) and so is giving me only 110N-m of torque when it knows I'm "requesting" 115 (and the throttle valves are opened to a point that'd deliver 125).



Ultimately, I guess this all just goes to show the importance of good setup and good technical help. My teammate, Sam West, has raced BMWs for years and knows all the tricks and tips for tuning these things, but neither of us were prepared for the amount of work it became to find a good chassis setup, a good computer setup, and then try to debug and troubleshoot problems, while at the races. We should've taken these bikes to Spain for a week or two first.

 

FTFY...

 

Hey, I have a question for you nerds: at around what speed does the inside pressure in a typical sport bike airbox reach its maximum? Does it plateau at some point, or does it keep rising indefinitely with speed?

 

Depends on the manufacturer.
It plateaus.

For some perspective...
IIRC from back in the day, Kawasaki's ram air would start taking effect in the 60mph range and peak around 140mph. The Hayabusa's system didn't do much until ~140mph...no idea where it it stopped.
Don't you read RRW?

 

Garbage in garbage out software rule.

 

Kind of.. the channels are all pretty evidently valid and even the channels I don't entirely trust, are extremely self-consistent with one another. My suspicion of them is more that they're not directly-measured values. md_act, for example, is described as "Actual Engine Torque" but it's just a lookup and interpolation of a few programmed-in dyno tables. Because I don't trust the absolute value in that channel, I didn't use it in the power/force calculations - but it's still useful for comparison against the other md_ channels. The power calc is basically entirely based on GPS location (and derivative velocity and acceleration), which I do very much trust.

It definitely plateaus, at atmospheric pressure. Motorcycle "ram air" intakes aren't actually true ram air in the sense of creating higher-than-atmospheric pressures - those only start working as flow speeds begin approaching the speed of sound. What we know as "ram air" is really just the design of the bike not forcing the intake to suck air from the sides of the bike, which is already depressed in pressure due to Bernoulli effects.

 

Absolutely fascinating work Chris, but has anyone ever accused you of taking things too far?

You're really onto something, but you pretty much get their with acceleration, all though it would be nice if you had considered calculating inertia as well. With acceleration sometimes the channels will report raw acceleration (including gravity) and other times only some sort of synthetic values with gravity subtracted. If you check your data when the bike is motionless, look for gravity in X, Y or Z. If it's there then you use the euclidean norm (square root of x squared + y squared + z squared), subtract gravity from that and the remainder is the bias you have to apply to all three channels. It's usually painfully off.

When you look at your x-axis acceleration, it's easy to miss the amount of acceleration projected off axis, especially if the IMU has a cross-axis mounting error which they usually do. You can directly correct it by riding the motorcycle in a perfectly straight line in the paddock and backing the Y axis acceleration into X. A better solution is for you to try the square root of X2 + Y2 as a better replacement to your X channel, but beware it will always be positive, even while braking. If you rotate the Y error back into the X axis and subtract gravity by a secondary axis rotation, you can use the sign on X to give you a very solid view of engine braking and use it to correct your estimation of horsepower delivery, first by averaging both and subtracting one from the other to give you the proportional error since starting and ending at 0 mph means all accelerations and decelerations must sum to zero. Also try to average the raw X axis and note it's difference from zero.

Also, your smoothing function is killing you, take a look at how the instant spikes in ignition of the shifter produce only small dips in power, that's the scale of your smoothing filter and it's completely hidden what's going on.

The wheel speeds are also prone to error, remember that they expand and shrink as you accelerate and brake due to centripetal force on the tire carcass. Tie your tire pressure data to your wheel speed data with a correlation/covariance function and you'll see it right away. The tire also changes diameters considerably with lean angle. Both of these things are not modeled by BMW, so the tire diameter is programmed by you for the maximum diameter all around the center of the tire. But as you lean in and out it varies in diameter, and not equally front to back. The BMW software acknowledges that wheel speed difference is therefore an uncertain subject as lean and/or speed increases and therefore you're supposed to ignore some arbitrary percentage of wheel speed as is. Use a GPS reference you trust as your speed channel and correct both wheel speeds against it at all times.

Which brings me to lean angle, this is not as knowable as the IMU would have you believe. First, its bias can be quite large and is unknowable by BMW because they would have to calibrate the Bosch unit for both XYZ offset in the accelerometers AND the scale factor error, which is not easily measured. The bias aligns with offset error between the chip, it's housing and your mounting and then of course your subframe. There are small errors there, but they accumulate with the accelerometer errors. The third problem is that when you're pitched and leaned, the math gets different and the two start to couple, so the definition of roll when pitched heavily gets a little murkier. Finally, the lean angle is calculated first from gravity in an inertial process known as the erection. Then it "rides the gyros" by integrating the gyro outputs until the next time it smooths out enough to check the accelerometers for gravity.

The error rate on the gyro riding with a Bosch chip like that is absolutely magnificent after even a couple of seconds. So the longer you accelerate or decelerate, the more the lean angle measurement integrates all of the errors and diverges from reality. For all these reasons, and given all the errors in all those measurements, you should probably stop at just watching acceleration and its various divergences, their causes and the solutions.

 

They're supposed to exhibit a pressure differential which is why ram-air on a carb bike require a vent hose to run back to the carb. I think, vaguely, the Honda NSR 500 GP user manual references the process somewhere in there around the petcock.

 

Found it

 

Overthinking things? Never.

All of what you suggest is actually already covered for in the channels available. accx_veh, the "Vehicle Acceleration" channel, is a computed norm of all the x/y/z IMU channels (which are also all individually available).

Inertia definitely is accounted for - see the F=ma part of things. All of my power data, however, is ultimately based on literally only GPS position (and then some empirically-decided constants). The smoothing functions are mostly to eliminate discrete-data artifacts causing massive derivative swings. The GPS logs at 10hz, so I smooth by 0.1s at it turns out well. I'm not interested in the power values by fractions of a second anyway, I ultimately want to know the average power delivered across a 15s straight.

The questions I have remaining are mostly around whether or not there's a factor I'm missing, and whether or not I can draw any further conclusions from the data I have available. The wheelspeed data matches exactly the spin (and thus DTC correction) factors I see. The power levels do *not* match what I'd expect (even with spin corrections 1.04*160=166, which is still wtf low) so that's an area of interest.

 

Non-ram-air bikes exhibit consistent sub-atmospheric airbox pressures because the engine is sucking air out and the intake typically can't "refill" the airbox at the same rate. A front-entry intake fixes that and lets air flow into the box as fast as it's sucked out.. the limiting factor becomes the TB/valves, not the intake tract, basically. There's still not any true "compression" inside the airbox however.

 

In God We Trust, all others bring data.

 

Chris, nice little piece here!

https://www.motorcycle.com/bikes/features/out-and-about-at-the-2023-isle-of-man-tt-part-2-44585140

 

Really enjoyed that video. He seems like a really chill guy and did a nice job explaining the differences.