Accelerometer fun

[nando]

from yesterday's autox:




Gforce-X: + Acceleration, - Deceleration
Gforce-Y: + Left Corner, - Right Corner

I didn't get a log of my last two runs which were a full second faster, laptop battery died.. the big flat ~1g area was a 100ft turning circle, lots of fun. I could have been even faster there if I'd swapped strut mounts for more camber on the outside front wheel, but I didn't bring my jack and there were issues with timing so I didn't think it was worth it.

datalog is noisy because the refresh rate of the accelerometer is way too high at 500hz. 10 or 20hz would be better. I could probably adjust it but it requires soldering of tiny SMD components (needs a cap from the output to ground).

it's not terribly useful at the moment but it's still fun to see actual cornering strength - I can't wait to try it on a track rather than a slow autox.

[yberther]

Can you reformat the data to make a G-G plot and kinda see the traction circle?

On the UW Formula team, we draw data from MoTec and are able to make G-G plots and its somewhat interesting to see the differences from each driver.

:D looks awesome though.

[nando]

hmm that's a good point, I hadn't even thought of representing it that way. Does MoTec spit it out that way or are you using Excel? I basically just have a raw datalog to work with so I can do whatever.

one thing you can see is where I missed 2nd gear - in the first graph I hit -1g braking, in the 2nd graph I didn't, because I lost so much speed on the 1-2 shift. I wish I was a little more advanced with the charting so I could represent RPM as a graph but they're on different scales so I left it as the X-axis label.

[10/10ths]

Good stuff, what tires were you running on?

[nando]

R888 235/50-15 that I've daily driven on for 10 months.. so they're not exactly sticker hoosiers. :p

[Wh33lhop]

hmm that's a good point, I hadn't even thought of representing it that way. Does MoTec spit it out that way or are you using Excel? I basically just have a raw datalog to work with so I can do whatever.

one thing you can see is where I missed 2nd gear - in the first graph I hit -1g braking, in the 2nd graph I didn't, because I lost so much speed on the 1-2 shift. I wish I was a little more advanced with the charting so I could represent RPM as a graph but they're on different scales so I left it as the X-axis label.

For RPM I'd have just made another plot underneath with an identical x-axis. You don't have MATLAB, do you?

[Trucidatio]

Damn, wish I could have been there on sunday... but saturday was definitely fun. Stupid ankle.

How was the course?

[guibo09]

excel allows you to choose a different y axis scale for a second series of data. it will show it on the right side. just select the 2nd data series and the option should be there, im running excel 2007 now so i dont remember exactly where.

a g-g plot should be really easy to do in excel. loose the time and rpm, just list out a series of lateral and long accel data points (2 columns, 1-lateral, 2-long). and then plot it as a x-y scatter with points only. with a little more excel magic, you can plot yourself a theoretical friction circle too.

also, your accelerometers sample rate isnt a problem! never say that your sample rate is too quick, makes my teeth cringe! im my daily work i constantly crave higher sample rates... look into decimation, i dont think excel does it, but matlab has a nice function. basically a low pass filter followed by picking every nth line. this will reduce your sample frequency while keeping your max and min peaks (which can be lost with too low a sample rate).

[Wh33lhop]

also, your accelerometers sample rate isnt a problem! never say that your sample rate is too quick, makes my teeth cringe!

High sample rates are nice, but for the purpose of an automotive accelerometer, 500hz is a bit excessive. ;) There's a lot of superfluous data there and it would be better represented by reporting averages at a lower rate.

im my daily work i constantly crave higher sample rates... look into decimation, i dont think excel does it, but matlab has a nice function. basically a low pass filter followed by picking every nth line. this will reduce your sample frequency while keeping your max and min peaks (which can be lost with too low a
sample rate).

You could easily just convolve the data with a gaussian for a low pass, subsampling isn't super necessary but it's very easy to do in MATLAB too.

[guibo09]

High sample rates are nice, but for the purpose of an automotive accelerometer, 500hz is a bit excessive. ;) There's a lot of superfluous data there and it would be better represented by reporting averages at a lower rate.



You could easily just convolve the data with a gaussian for a low pass, subsampling isn't super necessary but it's very easy to do in MATLAB too.

i dunno, doesnt a gaussian have the tendency to really 'blur' the data. i think the decimate function in matlab is the easiest and best way to downsample.

but yeah, if you dont have matlab, just keep every 50th line of your data and you will pretty much end up with the same result as having an actual 10hz sensor. the decimation thing is mostly for geeks who want perfection

[Wh33lhop]

i dunno, doesnt a gaussian have the tendency to really 'blur' the data.

It's a low pass filter, I'm not sure what you were expecting. :p

Yeah, you may technically lose some very sharp peaks, but how likely is it that that the time you hit 1.4G for 1/500th of a second is a real reading vs. noise? ;) You won't lose any real data; if you want to store acceleration peaks, they should probably be sustained for a good tenth of a second or so anyway before they are valid.

Either way, we are in agreement that MATLAB is a great tool and very easy to manipulate data with. In fact, Nando I doubt you care that much but if you want to send the data over I can fiddle with it a bit and see what I can come up with.

[nando]

noise is a huge issue, I actually hit 1.8g according to the sensor.. yeah.. probably just a nasty bump. that's why I want to try it on a track, it's much smoother and less frantic, I think I'll have a much better idea of cornering force potential in that case, plus the R888s might actually get up to temp. :)

here's the raw datalog, you can open it in Excel:

http://www.325ix.com/pics/2010/04-11-2010/2010-04-11_15.22.57.msl

the columns of interest are gforcex and gforcey. the graphs I posted above are using a simple moving average to get rid of some of the noise, it works for the most part but you still get a lot of spikes. I only have Excel 2003 so I'm not sure there's much more I can do with it.

[E-Thirty]

Some pretty impressive results for an e30. Subscribed to seeing what the graph looks like when you take it to the track. When do you think that'll be? :p

[nando]

june or july probably

[backtrail69]

How was the course? I completely forgot about the auto-x season starting. I heard a few of the formula guys brought V46 down. I need to put my summer tires back on my car and get out there next time, its been way too long.