Milling problem

I am also working with Sprutcam and if the robot is calibrated properly with the software - no problems with the accuracy at all.
If you are using a milling tool with more teeth - 6000 is fine... actually it is all depends of the material and the surface finish you need.

do you mind to share your tool settings for XYZABC? and also in the Sprutcam - the tool settings there?

yes, it is looks right... that means the the problem is mechanical
do you did the test with the cube?

I know this post is about 2 years old, need to ask some questions.
If Tool is not accurate enough is there any other method to improve it?
The robot is sitting at the top of a square frame without lavelling could this affect the accuracy?
The stock table is not laveled will this also affect accuracy?

Hi Sham,
Tool calibration using laser like Faro? If so this might be a bit expensive.
Looking for an alternative way of calibrating using existing measurement tool let say Haimer Measuring Instrument used in CNC?

Quote from ShaM

The leveling of the robot and table does not effect the accuracy. The calibration solves all construction defects.

OK, this part is confirmed, I do not need to worry that now.

I'll throw a minor caveat in here: the calibration will solve all minor defects. If your robot is mounted, say, 10deg off vertical, however, there will be negative effects on robot accuracy, due to the gravity vector being different than the robot expects.
But, in general, the physical setup of the robot simply has to be Very Good, not Perfect.

For measuring the tool, if you need extremely good accuracy, you're going to need something equivalent to a Faro, at least for the initial setup. If you set the tool up this way properly the first time, you can probably make due afterwards with a simple Haimer or Renishaw-type tool measuring device.
Think of it this way: the initial setup with the Faro would be the equivalent of a ball-bar calibration on a CNC machine: http://www.renishaw.com/en/ballbar-testing-explained--6818. After doing that, you should only need to make minor adjustments to compensate for things like cutter length variances.
Of course, people using CNC machines for very high-accuracy work end up re-doing their ball-bar calibrations on a regular basis -- some customers require it for certification. Likewise, a robot you're using this way may need some degree of regular checking or re-calibration on a long-term interval (1 year, 2 years, etc), and will definitely need it if the robot undergoes mechanical changes or suffers a hard collision.

If a Faro is completely out of reach, you may get satisfactory results by mounting a precision-machined "pointer" in your end effector and mounting a 3-way set of dial indicators where the robot can reach. The pointer will need a spherical tip to let you test the rotation accuracy of the tool, and as long as the long axis of the pointer is aligned exactly with one axis of your TCP (assume Z for this example), you can use the cylindrical shaft of the pointer to align the B and C angles of the TCP.
The weakness of this approach vs a Faro-style laser tracker is that the tracker can work over longer motions, giving you a better baseline. Using the dial indicators, you have to use much smaller motions in each degree of freedom (and mounting a 2-meter long pointer in your effector will probably give you results that are worse than useless).

Thanks Skyfire.
I brought this topic up because I'm facing an unknown reason where excess of poly stock occurs, some say is the accuracy of ABC needed to be adjust this is what I'm trying to look into.
This Robot is new and I believe is calibrated before shipped, did TCP 4 point calibration, Base 3 point calibration, results of the milling is an extra/excess poly stock lies in between of the sculpture head.
The procedure was milling the sculpture face on the Right of the stock then once finished the router bit is moved to the left side of the stock for the rest of the milling(back of the sculpture head). The excess of stock is solved by manipulating the cam software tool length, but still it is not supposed to be like this.

Thanks,

I'm not entirely clear on what you are describing, but it sounds like you have two operations, which have the tool at very different orientations. And that the milling error is apparent where the two different processes meet.

Unfortunately, that's probably due to the inherent nature of articulated robots. Put briefly, a TCP calibration is best when the robot is working in an orientation and position similar to how the TCP calibration was carried out. The further the working orientation is from the TCP calibration orientation, the greater the error can become. Changes in orientation amplify robot positional error faster than changes in position, velocity, or acceleration.

Even though you are using the same physical tool for both operations, you may have to deliberately create two separate TCP calibrations, one fore each operation, with the TCP calibration carried out in a pose as similar as possible to the working pose for that operation. That should help, but....

The simple fact is, robot motion accuracy becomes very problematic while using varying orientations. The TCP positional error probably scales exponentially with the amount of orientation change. Minimizing tool orientation changes is one of the first things to try when encountering spatial errors like this.

Quote from c2c2

Thanks Skyfire.
I brought this topic up because I'm facing an unknown reason where excess of poly stock occurs, some say is the accuracy of ABC needed to be adjust this is what I'm trying to look into.
This Robot is new and I believe is calibrated before shipped, did TCP 4 point calibration, Base 3 point calibration, results of the milling is an extra/excess poly stock lies in between of the sculpture head.
The procedure was milling the sculpture face on the Right of the stock then once finished the router bit is moved to the left side of the stock for the rest of the milling(back of the sculpture head). The excess of stock is solved by manipulating the cam software tool length, but still it is not supposed to be like this.

Thanks,

This is the dumb-guy approach to this - if your X/Y mate when you're milling on one side vs. the other is acceptable and it's your depth of cut that seems off (leaving uncut stock where the two sides should meet up)... which it sounds like because you changed your tool length in CAM to solve it - you could instead adjust your TCP to cut the depth accurately. You can make a number of test cuts and iteratively manually change the TCP to get what you desire.
This only works if you're always cutting left/right sides of statues because more than likely if you do full-motion 5-axis simultaneous (6 axis) milling, you'll see the TCP off in different directions

SkyeFire

What do you think of the LEONI ADVINTEC TCP system, how accurate is this system?

Sadly, I found it hard to work with, and poorly supported and documented by Leoni. If you are in Europe, rather than North America, support may be better.

Running repeated tests on it (measuring the same tool over and over), the repeatability in the measurements didn't seem too bad -- about .25mm, IIRC (unfortunately, I don't have my testing notes anymore). What I never had the chance to do was to check the accuracy of the Advintech against a good laser-tracker TCP calibration.

However, based on general experience, I suspect the Advintech is going to be better-suited for checking your TCP, rather than measuring it. Before I abandoned the Advintech, my plan was to calibrate the TCP as precisely as possible using a laser tracker, then make a baseline measurement using the Advintech. Then, over time, the system was programmed to periodically run a fresh Advintech measurement and compare the new measurement to the baseline. In this way, I was not depending on the Advintech for accuracy, but simply to track if the TCP had changed since the baseline calibration.

For small TCP changes over time, the Advintech might give you good "tweaks" to adjust for them. For larger changes (someone banged the robot into a wall, say), you would probably need to re-do the laser tracker TCP calibration, but the Advintech would at least warn you that something had happened.

One of the biggest issues I had in high-precision applications was that TCP changes too small for the human eye to detect, but still large enough to exceed my process tolerance, could not be detected until after a Very Expensive part had been processed, and then sent out for quality-control measurement. So the Advintech would have been a good warning indicator, if nothing else, but just getting it to work with the robot turned into a huge pain -- Leoni had no North American tech support for it, and the manuals are quite vague on some key items. I eventually just gave up on them entirely.

1. How is the tool calibrated using a laser tracker?I have never seen anyone doing this.

2. The next thing is that if I will have an on-site laser tracker, do I update the Absolutely accurate robot model?I'm shooting a little with this question.Why such a question, the robot was
calibrated at the kuka plant (created model of absolute accuracy) 7 years ago. Now, after 7 years, and over a dozen or so work hours, this model can change.My question is, is the model of
absolute accuracy updated over time?

3. SkyeFire Is it possible to measure like in a movie? https://www.youtube.com/watch?v=wZ8kQxbv2B0

Well, any method for measuring the robot's motion in space, precisely, can be used. A laser tracker is among the easiest, due to its range and lack of any physical obstructions.

TCP calibration with a laser tracker I performed by:
1. Mounting multiple SMRs (reflectors) to the Tool
2. Placing the robot at a known starting position and orientation
3. Measuring the TCP location and orientation with the laser tracker
4. Perform a LIN move along Z axis of the TCP
5. Measure the new location and orientation of the TCP
6. Determine error between actual TCP location&orientation and the programmed location&orientation
7. Adjust TCP B&C values to correct Z motion error
8. Goto Step 2 until Z motion error is within tolerance
9. Repeat steps 2-8 for Y, then X, then A, then B, then C -- each axis must be calibrated one at a time

This achieved quite good results, but:
1. The measuring location&orientation had to be close to the production location&orientation. The greater the difference between the two, the greater the final accuracy error
2. Past a certain limit, reducing the error on one axis increases the error on others, so it was necessary to repeat the entire process multiple times to find a reasonable compromise. In my case, A error could be ignored, and Z error was less important than X or Y.

Now, calibrating the [r]robot[/b] was done after the TCP was fully calibrated, with an SMR attached exactly at the TCP (sometimes, this required creating an extra TCP that matched the physical center of the SMR)
Calibrating the robot was a process of:
1. Create a set of calibration points that spanned the critical working volume of the robot, in both location and orientation -- the density of this pattern depends upon your accuracy requirements. In my case, I had approximately 8000 working positions, and used 1200 reference positions.
2. Move the robot to each reference point and measure the error with the laser tracker
3. Add corrections to each point and goto Step 2 until sufficient accuracy is achieved
4. Build a weighted-sum algorithm that can take each working position, find the nearest matches among the corrected reference points, average their corrections, and apply that average to the working position.

Hello ShaM,
here are some feedbacks based on my experience with milling and 6axis operations with kuka and Sprutcam.

1) TCP calibration is crucial. I found that the longest the calibration tool the better TCP calibration. I recalibrate all my tools using a reference tool 50cm long. This helps to reduce calibration error on the milling tools that normally ar much shorter.
2) If You have A,B and C value different from a perfect straight or normal position to robot flange(0,-90,0), you cannot calculate TCPs only measuring the length difference between tools. You need a system to recalculate X and Y according to A,B, C.
3) Use a height gauge to measure your tools length.
4) Base calibration is crucial. Calibrate X, Y, Z, A, B and C on krc but leave base A,B,C data set to zero in Sprutcam, otherwise you will had additional transformations. Of course this applies only if you import your model in Sprutcam already in the virtual expected position respect to your milling table.
5) Check my old posts, there is a link to a killer calibration procedure that was posted on this forum to improve X,Y and Z calibration after measuring tcp via 4 points
6) I have a KR150 with position accuracy, if you expect precise milling faithful to your drawings (+/-0,..), you cannot mill faster then 2400mm/min.
I hope this helps