OK, I decided to do a bit more testing, using the CANBUS data of the car, given that a few people have pointed out that “Indirect TPMS” (Tyre Pressure Monitoring Systems) exist, and use the ABS wheel encoders to infer if a tyre is flat. I was really puzzled by these systems, since it looked like my first round of tests proved that (on my tyres at least) it shouldn’t work.
(In theory Indirect TPMS also uses Fourier analysis of the encoder data to look for resonance. Which I’m sure exists. But I’d be fairly skeptical that normal tyres show enough of a radius change for flatness to be reliably detecting without looking at Fourier stuff)
I did two runs. One with all tyres full, and the other with the rear-left tyre deflated to 150kPa (65% of normal pressure). That level is low enough that I can feel a fairly strong pull as the car “wants” to go to one side, but it’s high enough that I can be fairly sure my experiment won’t break my tyres and cost me money…
I made sure to take the same route, starting and ending point, and used a stretch of road that was quite straight for most of the way. CANBUS frames were logged to my laptop, then decoded in Python.
Here’s the results. As you can see there’s only fairly subtle differences in the speed, but it’s very hard to tell when compared with the speed differences that occur during turns:
A better way to analyze it is to compare the ratio of the wheels (Front Left / Front Right), and see how that varies when compared to the steering wheel angle. Thanks to the CANBUS data I can easily plot that and see where the centre of mass of the data points is:
You can actually see that there probably is a difference, but it’s a damn subtle one. Deflating one tyre almost enough to be a safety issue caused maybe a 1% difference.
Realistically I should do this several times to average out and see if the effect persists. This test involved me driving down the street and having to wait for traffic and lights, with lots of interruptions and turning around. So maybe the results would be different if performed again. But I’m happy to have narrowed down the size of the effect (if any) to be very small.