Any advice for improving milling accuracy on older hardware?

Well... horizontal circles are going to be among the worst features you can cut. That's b/c, whatever you do, you're reversing Axis 1 through its deadband, twice. Although the 1/4" error does seem a bit excessive. I've never done CNC work with a KUKAbot, but I recall hearing that accuracies down to 0.016" were possible. However, that was with a KRC2 and an Absoluate Accuracy robot, so....

Basically, any path that reverses an axis during its motion is going to create this kind of problem. It's just that A1 is probably the worst, and a circle shows the error in an easily visible and quantifiable way. I don't think there's any easy fix for it -- the CAM software assumes that its G-Code output is going to a perfectly accurate machine, without the degree of backlash that an articulated robot has.

Ideally, you would break up the cut such that the robot made passes all in one direction, without introducing any axis reversals along the way. Of course, along curved cuts, that becomes difficult to impossible. It might come down to a matter of deliberately introducing some under- or over-steer in the corners, so to speak. Or introducing a deliberate tool bias around curves. But all that is going to require hand-tweaking of a process that should, by rights, be automated.

Hopefully, someone with more experience doing CAM work with KUKAbots can speak to this issue more authoritatively.

I recommend this book:
https://amzn.to/2ZLUR2p
It points out most of the things you needed to know on multiaxis milling.

Why dont you try a circle without external software - just krl core with two arcs?
Or even few arcs with known radius?

I wrote a routine to do this without using CIRC commands a while ago:

FOR Index = 1 TO NumberOfPoints
  DegreeIncrement = DegreesOfArc / NumberOfPoints
  MoveArray[Index] = ClockToolXY (CenterOfCircle, Index, DegreeIncrement, CircleRadius)
ENDFOR
;
LIN MoveArray[1]
FOR Index = 1 to NumberOfPoints
  LIN MoveArray[Index] C_DIS ; approximated motion
ENDFOR
;====================================================================
GLOBAL DEFFCT E6POS ClockToolXY(StartPos :IN, ClockIndex: IN, ClockIncrement :IN, Radius :IN)
  DECL INT ClockIndex
  DECL REAL ClockIncrement, Radius
  DECL E6POS NextPos, StartPos
  DECL POS Shift


  Shift.X = (COS((ClockIndex*ClockIncrement)-180)*Radius)
  Shift.Y = (SIN((ClockIndex*ClockIncrement)-180)*Radius)
  NextPos = StartPos : Shift
  RETURN NextPos
ENDFCT

The idea is, you can generate an arc of however many degrees you want (even more than 360), divided across however many points you want. You load an array with these points, and then run the array with approximation for smoothing. As written, it only generates arcs in the XY plane of the Center Position you give it. I've used it for doing circular cuts and adhesive patterns, among other things. The robot's approximation turns it from a polygon into a circle, but also means that you effective radius is smaller than your programmed radius. However, this can usually be tuned out by selecting an appropriate number of points to set the arc's resolution, without going so high as to drive the path planner nuts.
I've got another version around here somewhere for doing programmable ellipses and spirals using the same underlying concept.

Given the amount of error probably the mechanics of the robot are really bad, in case you don't think so, you can always contact any robot calibrating company in your country to get the robot calibrated (as in a high accuracy robot). That would help from the software side (just in case mechanics are ok).

Good luck!

Sent from my BlackBerry 9300 using Tapatalk

Hi i can see that you are going in circles
ok, i will explain about the toolpath: the toolpath defines the movement of the physical end of your tool and it doesn't matter how long you will make it in Sprutcam - the code is still about the movement of end of your tool bit. As long you are doing proper tool calibration on the robot it will follow the same path with different (in the CAM) tool lengths.
This is not an issue in any way, that is normal. And you don't need to have super ultra precise robot model - the software is not writing the rotations/angles of the axes etc - it using the kinematics just to check the collisions, reachability and singularities. The end result is the toolpath which will be fine as long you have well calibrated tool, base and mastered robot

About the tool definition: the X is pointing to the tool orientation vector - in your case down. And your Z should point forward - just as it is in the picture you provided from the Kuka documentation. I think you mixed them up... You can easily check you directions on the robot by moving it XYZABC

The problem is if you have rectangle instead of square for me that means that your robot needs remastering.
It is not difficult and you can do it by yourself even with smaller precision, but is a must if you want to do milling.
Do you have instructions how to master your axes? Send me a PM if you need info about that.

I have to agree with VVelikov -- checking the Mastering of all 6 axes is the first thing to do. The official way to do this is to use an EMT, but I'm willing to bet you don't have one, and they're expensive. Your alternatives are to obtain or build a Dial Mastering adapter, which would let you do the Mastering with a typical machinist's dial gauge, or to use the "field expedient" trick passed down from KUKA serviceperson to serviceperson over the ages.

You'll want to read over the parts of the manual that explain how Mastering works, but basically you're trying to capture the very bottom of the V-notch with the tip of the pin. The EMT does this by slooowly dragging the pin across the v-notch and recording the exact location where the pin bottoms out and reverses direction. With the Dial/manual approach, you're doing the same thing, except by eye or feel, and you need to stop the axis at exactly the bottom center of the v-notch and then hit the Master button. Also, for maximum accuracy, do not hunt back and forth across the notch -- that introduces the deadband of the axis directly into your Mastering. Instead, always keep the axis moving in the negative direction.
If you need to, you can keep doing Check Mastering for practice until you've got it down. Check Mastering doesn't actually change anything, it just shows you the difference between the previous Mastering for that axis and your new mastering. If you do this the same way on all 6 axes, and find one that has a bigger error than the rest, that's probably your culprit.

Quote from happytriger2000

I recommend this book:
https://amzn.to/2ZLUR2p
It points out most of the things you needed to know on multiaxis milling.

Short version of this book http://ma.gnu.ac.kr/course/2013/am/5axis-ParkJH.pdf and http://ru.scribd.com/doc/35582337/S…-Axis-Machining

Hm. In that case, you could still use the "feel" method or the dial gauge, establish the physical zero, then check the Position Monitor. Not as good as Check Mastering, but should be good enough to let you detect any gross errors.

I have to master krc1 every time I start it up using dial gauge option, the robot is already in its mastered position but no need for KRC2.