I have tri-axial data from an accelerometer, however, I know in retrospect that it was not aligned properly when placed on a mechanical arm. I know because when looking at the values at stationary instead of being 0,0 1g they were 0.4g, 0.5g,0.92g. I am trying to calibrate the data and the technique I have used has not completely realigned my data. Has anyone experienced a similar issue? I have read literature and apparently it is not straight forward as it seems ie considering 3 angles around the three axis. I am mainly interested in two axis however so I am prepared to do this approximately.
13
-
$\begingroup$ OK, so you have an rms gain error of 12.1%. You can turn the device around each axis until you get (g_x,0,0), (0,g_y,0) and (0,0,g_z). Those are your three gains. Correct each axis gain, then find a rotation matrix that can explain the remaining gain corrected output. If your device also suffers from offsets, then you need to find a way to make a measurement while it is in free fall. $\endgroup$– CuriousOneCommented Jan 4, 2015 at 23:18
-
1$\begingroup$ @CuriousOne - There are a number of things you can't do to robotic arms. Destructive testing (e.g., putting them into free fall) is in general not a good idea on even the cheapest of robotic arms. Besides, while free fall will catch biases, it not catch misalignment or non-orthogonality errors. $\endgroup$– David HammenCommented Jan 4, 2015 at 23:25
-
$\begingroup$ @DavidHammen: Losen screw, remove accelerometer, drop accelerometer in calibration jig, attach accelerometer back to arm, tighten screw? You were not gentle on him either, though. I like that. :-) $\endgroup$– CuriousOneCommented Jan 4, 2015 at 23:27
-
1$\begingroup$ There are (at least) three causes of misalignment and non-orthogonality: (1) The cheapo MEMS accelerometer one uses is just that, a cheapo MEMS accelerometer. The misalignment and non-orthogonality are built-in into the device. (2) The lab assistant who mounted the device made some dumb mistake and mounted it incorrectly. (Murphy's law at work.) (3) Every device, no matter how well-constructed, no matter how painstaking attached, is still going to have some errors of this sort. They might be small, but they are still present. $\endgroup$– David HammenCommented Jan 4, 2015 at 23:30
-
$\begingroup$ There are a forth and a fifth possibility: the device is broken and/or your readout code is buggy. Other than that you may have an infestation with microscopic black holes... but I would check that hypothesis last. $\endgroup$– CuriousOneCommented Jan 4, 2015 at 23:33
|
Show 8 more comments
1 Answer
$\begingroup$
$\endgroup$
8
Presumably you are calibrating your sensors and your algorithms before you put your device to actual use. (If you aren't, that you are seeing these issues is entirely your fault.) Misalignment and non-orthogonality issues are easily detected at this stage.
Presumably you are using a Kalman filter of some sort, or something more advanced. (If you aren't, I strongly suggest that you don't let your robot near anything of value.) After enough time has elapsed, a properly-constructed and properly-tuned Kalman filter can detect all kinds of errors that slipped past your initial calibration.
Do a literature search. There are articles upon articles that discuss calibration and Kalman filters.
answered Jan 4, 2015 at 23:22
-
$\begingroup$ static accelerometer readings - nothing to do with a Kalman filter even if branny12000 is using one. $\endgroup$ Commented Jan 5, 2015 at 0:30
-
$\begingroup$ @docscience - That's not correct. A Kalman filter can model all kinds of things in addition to position and velocity. The number of states in spacecraft Kalman filters can be astounding. In addition to the standard position, velocity, attitude, and attitude rate, they oftentimes try to estimate misalignments, non-orthogonalities, biases, drift rates, and more. $\endgroup$ Commented Jan 5, 2015 at 0:35
-
$\begingroup$ For a robotics application a KF might be used to 'fuse' multiple sensor readings together to provide a best estimate of joint angle for example. Yes - RE Kalman invented a remarkable filter. But that's not the issue here. This guy's raw readings are off - before they are even fed to the filter, if there is one. His static measure with z vertical shows the other two axes reading nearly half a g. $\endgroup$ Commented Jan 5, 2015 at 0:40
-
$\begingroup$ @docscience - As I said in a parenthetical remark in my answer, they should have done some preflight testing (or the robotic equivalent thereof). A KF can appear to be quite magical, but they only can do so much. The more error one can eliminate beforehand, the better. $\endgroup$ Commented Jan 5, 2015 at 0:44
-
$\begingroup$ Hi guys. I am a little confused as I said I am aware that when the accelerometer was fastened to the arm it wasnt aligned correctly and was rotated slightly and so I didnt begin with 0 0 1g will a KF help me? $\endgroup$ Commented Jan 5, 2015 at 1:05