View Single Post
      02-18-2011, 02:18 AM   #64

Drives: 2014 Shelby GT500
Join Date: Apr 2008
Location: Vancouver, BC

iTrader: (4)

Originally Posted by eeghie View Post
Is the process by hand the elimination of errors or more? Did your integration involve some smoothing or data(n+1) = data(n) + d data(n) / dt * delta t?
Essentially, I couldn't find a string of data points (in an hour or more of data) that would behave sensibly for more than a second or so. You're double integrating, so the process is to integrate from acceleration to velocity, detrend it, then integrate velocity to position and detrend it again. If you pick your start and end points when you can be confident that the suspension is at static height, then you can look at movements away from that static height. You can also look at motion from a "sensibility" perspective to see whether the data says the suspension is in places that physically it just can't go.

For sets longer than a second, you don't get consistent trends for detrending. The accelerometer signal has (what I presume is) low frequency noise components that make the integrated result wander around by 10's of cm over a few seconds.

Originally Posted by eeghie View Post
Looking at it more closely, I see what you mean now: the top of the wheel/suspension curve is flatter than expected from a harmonic signal. From the technical drawings I recon that bump stops nowadays have their own additional spring and damping characteristics, rather then being hard stops but I still would expect a more significant deviation (acceleration spike) from the harmonic line. It could also be related to the shape of the bump (~flat top?) and/or a dry friction effect? Once I have time I wouldn't mind looking into deriving the shocks damping coefficient [Ns/m] and dry friction force from your curves. The mass (or weight), front-rear balance and spring rate of the measured setup would be useful for that (and if willing to share, the measurement data). My goal here is mainly to get a feel for the damping coefficients used in M3 shocks and in the end to quantify the difference in EDC damping settings vs their drive signal.

I started diving into old Matlab stuff and found that it lets you record directly from the PC's line-in, but I would need an accelerometer with the right output signal. Did you consider accelerometers with analog output?
I did a force vs velocity curve for the OEM dampers based on the acceleration data. F=MA, sprung and unsprung masses and known spring rates allow you to compute the forces and velocities of the suspension parts and graph the contribution from the dampers. This is actually easier to do because you only integrate once, and you can get enough samples to see the shape through the noise. It was an interesting exercise, but I can't find the data on my laptop, and I may have chucked it in a fit of cleaning up.

I thought a lot about analog accelerometers because I'm an analog circuits guy from decades ago. That would have been vastly easier for me to set up and calibrate than digital processing, but vastly harder to post process. I'd have been able to bandpass filter it, integrate and detrend it and get meaningful numbers without breaking a sweat. Digitally, it's not so easy for me. If I was starting now, though, I'd take a close look at Bosch BMA180 digital accelerometers - they have built-in bandpass filtering and data cleanup. They might work better.