Robot path from mesurment data

  • Hello,

    I'm working on a robot cell for deburring gears. The cell should have 2 or 3 work tables with rotating clamps/grippers (sorry for my english on this one), then the robot will move from one table to the next and so on.. The problem at our company is that we make specific gears and shafts so not that much repetitive work (from 2 parts to 60 parts in one "serial" production). And the CAD models do not match the end product since the our customer designes all gears and then add a table with the tolerance.

    So I'd like to know if it is possible to create path from laser scans or a 3D camera?

    And also if that is not possible or not accurate enough I was thinking about generating robot path with a PC. So to summ my idea.. we would snap the gear from above to get the rotation data and then the robot would touch the Da and the Df of the gear and also the hight of the gear. Operator would input the other data thru HMI (module, angle alpha, ...). So in what language can we write the program on PC and how to write the program so it would compute a path for each gear and send it to the robot to execute.

    Thanks to all of you in advance, this forum has been very helpfull since I started my career :)

    Best regards


  • AD
  • This depends entirely on the precision required. At the "easy" end, you could simply mount a rotary wire brush to the robot and make a blind, circular path around the gear circumference, with spring-loaded compliance. At the "hard" end of the scale, you would need to trace out the profile of every gear tooth with a 0.1mm diameter chamfer tool at 0.01mm accuracy.

    The normal way to do this would be to generate a path from the CAD model. Attempting to generate such a complex path purely from vision data... should be possible, but I suspect it would require some very specialized software (not to mention high-end vision hardware, probably a laser profile system, not a simple 2D vision system). I'm not aware of an off-the-shelf solution for such a process on the market.

  • First, thank you for this reply! I also tought about the "easy" way first, but our customers want the gears chamferd (1-3 mm at an 45° angle). Yeah we need accuarcy around 0.2-0.4 mm and as I said the CAD models are not accurate compared to the real life gear. So I'll check with Fanuc and Yaskawa tech support team what can the systems do and what are they capable of. Btw Fanuc posted a video on this topic: (16) Robotic Cast Deburring with FANUC Robot, iRVision, Image-to-Path Software & Force Sensing - YouTube So if we try to use the same principle with gears in theory it should work right? :)

    Best regards

  • Hm... I hadn't seen that before. Interesting.

    However, the audio talks about "casting variations," as well as reliance on "simulation in RoboGuide." The video lacks sufficient detail, but based on what I saw, I would be willing to bet that the paths are CAD generated, with the vision-to-path software only making small path tweaks to account for variations from casting to casting.

    And of course, the vision-to-path is of limited accuracy, which leaves the dynamic force-torque sensing to "smooth out" the vision errors. I will note that this is force-based deburring, not precision deburring -- since the robot is prioritizing force over position, a "lump" in the casting will simply push the robot away, leaving the lump deburred but still present.

    That said, approaching Fanuc with your process requirements to inquire if the technology in the video might be able to serve your needs would certainly be worth it.

  • Okay so I asked with Fanuc in my country and they said that it should be possible but the accuarcy is limited to 1 pixel. So with biger teeth and small diameters it shouldnt be a problem, but with smaller teeth and big gear diameter can cause some problems due to position of the camera being far away in order to snap the whole gear.. So thanks again I'll let you know how we solved the problem! :)


  • ... but with smaller teeth and big gear diameter can cause some problems due to position of the camera being far away in order to snap the whole gear ...

    One thought:

    Those gears normally have exact same teeths around the whole cirmuference. So it should be possible to calculate the trajectroy from a picture of one or two teeths.

  • One thought:

    Those gears normally have exact same teeths around the whole cirmuference. So it should be possible to calculate the trajectroy from a picture of one or two teeths.

    Maybe... The issue with only imaging a few teeth is that, while the resolution of data on the teeth increases, the location of those teeth in space becomes less accurate -- "projecting" from a few teeth to the entire circumference of the gear, the accuracy of that projection scales inversely to the circumference of the gear.

    Two ways, off the top of my head, to solve that:

    1. Use two vision systems, or vision system and a laser profiler -- one to establish high-detail data on a small number of teeth, and another to establish the position of the gear center and its orientation. Ideally, the gear would have some clear features to make it possible to determine its angle of rotation with high accuracy.
    2. Fixturing -- if the gear can be held in a particular orientation with high accuracy and reliability, then it becomes more possible to "project" high-detail data from just a few teeth around the entire circumference. This would require a slip-fit fixture with some sort of "clocking" detail, however, like a keyway on the shaft of the gear. For a gear with a perfectly circular central shaft, it would be necessary to have another detail on the gear. The downside of the fixturing approach is the possibility that the the gear features to engage the fixture might themselves be in need of deburring, leading to accuracy issues.
    • Helpful

    We have built a system that works this way, but the path for one tooth is teached the standard way.

    This system was built many years ago when KRC2 where state of the art, and still works.

    The gear just lies centered on the 7. axis table (very big and heavy gears, they have diameters up to 2m). The centering is done manually in less than a minute.

    Before deburring the middle point between two teeths is measured with a simple inductive sensor.

    The spindle is mounted on a "flexible" fixture, to egalize small variations.

    Works like a charm.

    And a "funny fact" about that: if deburring fails (too much metal removed) repairing the gear will cost many thousands of euros.

  • Okay nice to hear about something that works.. thanks a lot! And how do you center the gears manualy in under 1 min? Could you maybe share some pictures of the 7th axis (table) or what messurment tools you are using..? thanks alot to bolth of you! :)

  • Sorry no pictures, . Not I do the centering but the production staff. I am working for a system integrator, so the last time I have seen the system is a long time ago.

    He centers the gear with a mallet (rubber hammer ?) while the 7. axis is turning relatively fast. For measuring the center he uses a dial gauge mounted on a fixture. Sometimes may be he needs 2 minutes;).

  • Okay no problem.. Aha okay.. because at our company we use fine thread bolts from multiple directions (every 10° maybe) that are screwed in to center the gear.. and this is on a CNC for milling gears.

    Thanks again and best regards!

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