Nonsensical TCP behavior

  • right now in Ottawa, police is trying to stop (and arrest) anyone supplying protesting truckers with anything, specially fuel....


    so bunch of Canadians are trolling police by walking around with empty fuel canisters. :grinning_face_with_smiling_eyes:

    1) read pinned topic: READ FIRST...

    2) if you have an issue with robot, post question in the correct forum section... do NOT contact me directly

    3) read 1 and 2

  • Place your Ad here!
  • OK, besides Canada must be a province of Mexico (check google map) and follow post #1


    Without having the adapter plates pinned (flange A6 to adapter flange and adapter flange to tool flange) the results are very good - you never told us the calculated tool data - why?


    In order to get better results add the missing pins and do the calculations once more.

    Also check thickness of the adapter flanges (almost killed me abroad) and present your results here

    Could be very helpful


    Concerning Canada

    never take just a piece, take the whole thing

  • I've gotten the EMD from KUKA and un-mastered + remastered my robot with it.

    Sadly, with roughly the same 4 directions and still with painstaking care, I get a worse-than-before uncertainty value of 0.74 mm, and the "rotate around tool A" test still sees the TCP move in the direction of tool "A" (z) axis even more than before.


    I'm puzzled by the concept of calibrating with / without a tool load in my case because the spindle is there all the time.

    I made my calibration with the spindle attached, as this will be the "Permanent" state of the robot ;

    the "tools" for me are just the end-mills which weight will be negligible.


    Then, I wonder if I should enter the spindle weight in the tool load data (which I use for all my tools) since it has already been accounted for in the calibration...


    In short , calibrating with / without load plus entering load data seems redundent, and I wonder if the robot is over-compensating the load in my case.


    I'm off to re-do the calibration, and this time, I'll take pictures of the 4 postures that I use to calibrate the tool XYZ

  • Then, I wonder if I should enter the spindle weight in the tool load data (which I use for all my tools) since it has already been accounted for in the calibration...

    Yes always. There are internal features that use load data to predict robot behaviour in certain conditions. E.g. it is used in elasticity compensation mechanisms. In mastering stuff like this is a more statical correction of such effects (only positional effects in mastering Position can be improved) while load data is also used to additionally correct dynamical effects (errors induced by velocity and acceleration) and positional effects over the whole work envelope by usage of mathematical robot models.


    Quote

    I made my calibration with the spindle attached, as this will be the "Permanent" state of the robot ;


    If you will ever only use this single tool and do not want to be the robot to be interchangeable with another robot of the same type if it breaks for whatever reason this is ok. The calibration without load usually sets the base calibration and the calibration with load than stores the offset to base calibration induced by tool. So even if the tool is mounted and you use the calibration without tool method this is ok under the above conditions. What you should not do in this case use calibration with load, since this only calculates the shift caused by tool with reference to a possibly bad base calibration. So which calibration method did you use? If it was calibration with tool this might explain your higher error compared to before.


    Fubini

  • To be clear my spindle was on the robot when I made the mastering, and I followed "EMD / Standard / Set mastering". I also performed the "Check mastering" sequence, and the deviations were in the order of 0.001° in motor angles, so I guess my mastering is OK.


    I DID NOT do the load compensation procedure (one mastering with load, one without).


    Using my laser, I moved the TCP along the Y axis on a broad LIN movement, and the result is appalling.

    I measuer at both ends and at the middle of the LIN, and this time, I didn't bother trying to align my laser cross with the Y axis, so the end points aren't close, but I wanted to see the "Sag" at the midpoint.

    I get 1.5mm sag downwards at midpoint which is where the arm is the least extended.

    This seems to indicate that the robot is over-compensating the static loads when it is fully extended...

    I made a second run after setting the load data ""weight" to 0 kg, and the result was almost the same.


    Anyways, for my milling application, dynamic loads will probably not make any difference. Milling is a relatively slow process.


    Sorry for the messy video BTW ; difficult to use a pendant and a smartphone at the same time...

  • Looks for me like axis 2/3 needs to revert exactly in the middle of your path, so backlash of axis 2/3 comes into play. You could trace all set and actual values to see if this really is the case. This should get a bit better if absolute accuracy robot is used but still would not be zero.


    Have you an understanding which value is closer to the ideal value (in the sense of absolute accuracy). Middle or stretched out? Inaccuracies at the work envelope bounds are not surprising. Even if models for the robots exist they are never perfect.


    Might also be friction effects. Friction is something that's always difficult to model without information on temperature. Do you see differences when robot is warm or cold?

  • In the calibration position, my spindle doesn't exert any torque around axis 6, so I think it's irrelevant in my case, but I understand how a tool with an offset weight along the flange Y axis couls cause tro

    Think again about that. Everything mounted to axis 6 also influences axis 2 to 5. 😉 This is the reason for dismounting everything from axis 6. Not the torque around axis 6. Best would be to dismount also everything from axis 3, but that I slo don't do normally, because the leverage is short.

    I have seen many not absolute accurate robots with deviations like your robot. I think they often are normal for Kuka robots. As fubini said, there are many influences like manufacturing tolerances, backlash, friction, actual working position, weight of tool and accessories on axis 3....

  • have seen many not absolute accurate robots with deviations like your robot.

    Oh well. I guess I put my 40 000€ on the wrong robot for milling, then.

    I hope I have better luck next time.

    Someone told me something about stereoscopic cameras. Could that help getting closer to a decent calibration ?

  • The first milling I made was back in 1994.

    The important part was (and I believe still is) the exact tool data definition.

    I gave those guys a course about getting the exact tool data and how to check the results afterwards.

    If tool data are correct you may rotate A, B and C and the tool tip will not move (otherwise tool data are incorrect). As fubini mentioned check that at the position without backlash (like A1 not around 0°)


    According your tool (I have seen):

    - use the adapter pins

    - check the exact length of your tool in x, y and z


    At that course my Z-value was a bit off and produced an virtual circle with a radius of about 6 mm


    My suggestion at the first place would be:

    compare the CAD drawing with your tool defintions

  • I like stupid ideas


    Could You do me a favor?


    1- Tell me the calculated tool data

    2 - Rotate around A, move manually to the test position and give me the flange position

    3- repeat (2) for angle B and C


    Also do the same sequence at around A1=0° position


    It could be very helful for me and also for You

  • OK, now that I did a complete mastering with an EMD (200 € ; yikes !), I re-did my XYZ 4 directions calibration, and this time : documented !

    I think my calibration tool is as good as it gets, made from a linear guide ; I measured 0.09mm of runout at 200mm distance from the spindle.

    Well, the result (error = 0.788 mm) is still completely crappy, so I hope you guys can give me advice on how to get a better XYZ.

  • Well, the result (error = 0.788 mm) is still completely crappy, so I hope you guys can give me advice on how to get a better XYZ.

    You can try to move to the 4 calibration positions a few times within the calibration screen (on the newer robots it is possible to move to the stored positions), and make them better, until the result is better.

  • Attached are the documented results...

    It doesn't make sense at all to show the axis and motor values in this case. And the pictures captured from different sights don't make sense in this case. Can't see the orientations in the different pictures in relation to each other, cause the third picture is captured from a different view.

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account
Sign up for a new account in our community. It's easy!
Register a new account
Sign in
Already have an account? Sign in here.
Sign in Now

Advertising from our partners