[mh1470]

Hi there, I've come to this forum many times for answers, but this is my first time posting.

We've built a cnc ourselves, and while most things are going good, we're having great trouble getting it to cut a decent circle. The machine we've built is a table router, 1m x 2.6m. The base is made from steel that's bolted to the ground. Two I-beams are running lengthways for our X-axis, and a length of RHS filled with epoxy and sand is sitting on top for our Y. We have linear rails and bearings (SBR30UU) with rack and pinion for moving our X and Y axis; HIWIN's with a lead screw for our Z axis. Each axis is powered by a Nema 34 stepper motor. On the computer end we're using mach 3, and our g-code is being generated by Cut2dPro.

Photos of machine:
https://ibb.co/KWwK9fn
https://ibb.co/ypx5C78


Our trouble comes when trying to cut circles. Our first attempts yielded the results shown here:
https://ibb.co/kxvfb6k

Close-up:
https://ibb.co/fDLHS5V

Looking at the picture with "Set 1" at the top, left to right is our X-axis and our Y is up and down. The circles on the left are a CW profile cut and the ones on the right are a CCW profile cut. The ones in the middle are pockets cut in a CCW direction. The g-code generated by Cut2dPro for the circle cut it up into 4 arcs. The material we're working on is 16mm melamine. Two of the arcs transition into each other perfectly, but one of them adds a straight line into it, causing the circle to meet back on itself slightly off. The circles on the right were tested at half the feed rate. After searching the forums, I concluded that backlash on the x-axis was probably the culprit for that little straight line added to the circle.

The next day, after doing some other rectangular cuts with no problems, I tested some more circles:
https://ibb.co/LNvTG39

Close-up:
https://ibb.co/tPXtsQJ

This set was done with a CCW profile cut. I calculated the tool paths the same way in Cut2dPro as the first tests. The bottom circle was the first one, and we were quite happy with it. I wasn't sure why it was better this time, but I was happy. For the next one above we tried adding backlash compensation in mach3 to see if that helped at all, but we didn't notice much difference, so for the third one on the top we took it back off.

A couple of days later, I did some more circle tests:
https://ibb.co/tXbVGKq

These ones were all over the place. I generated the same g code in Cut2dPro. Here are some close-ups.

This was a CCW profile cut starting on the left point:
https://ibb.co/8xhWFwq

The one was also a CCW profile but starting on the bottom:
https://ibb.co/16N9TYr

The middle-bottom circle was a CW profile starting on the left. Up until this point all of our circle errors had occurred at the transitions between the arcs. These circles introduced a stepping that occurred halfway through the arcs. The g-code generated by Cut2dPro divides the whole circle into 4 arcs, each starting at the left, bottom, right, and top points of the circle. On these new circles, we now had perfect transitions between the arcs, but the stepping was occurring halfway through some of them. The circle on the top-left gave a similar outcome to our first tests, only this time the starting point was 90 degrees further along the circle and all the other errors occur in relation to that just as they did in our first circles.

Just to see what would happen, I created two semi-circles in Cut2dPro, and ran that:
https://ibb.co/4YXsbYd

Again, I ended up with a step off-path occurring halfway through the arc. I then loaded back up the g-code I used on the set 2 circles, and ran that without changing anything, which resulted in a different circle than the ones we did a previous day:
https://ibb.co/L8ZHK96

Again, in these circles, they were perfect in the points where we previously had problems, but were now stepping over at points which were previously perfect. My final test was with a manually written g-code circle straight from mach3. I told it to just do two arcs, from the point on the left to the point on the right, and back to the left again. The result was similar to the other circle tests in set 3:
https://ibb.co/CQVhjdz

What confuses me is that the results have changed over the different sets. What I did think was backlash has now got me confused. I'm especially confused as to how the machine has started stepping over halfway through the arcs, where it used to cut each arc perfectly and only go wrong in the transitions.

Any help would be much appreciated. Let me know if you need more info.

[ger21]

Definitely backlash. Check for loose belts, or loose setscrews on your pulleys.

[mh1470]

Hi Gerry, do you mind explaining how you came to that conclusion? I'm not saying you're wrong, I'm just a beginner in the cnc world and am curious as to how backlash is causing the issues we're having. I'm confused as to how it could be causing errors in one spot during one test, and then during the next test be causing errors in a different area with the original errors disappearing.

As for the backlash, we can't feel any sort of movement when we try to move our axis by hand, and the one test we did do for backlash using a dial indicator came up with a negligible value. I will try some more tests though to make sure.

Thanks,
Matthew

[ger21]

Something is probably intermittently getting loose. Which results in backlash.

Your results are exactly what you get when you have backlash.

When the issue is repeatable, run the program VERY slowly, and see if you can see what's happening.

[BBMNet]

Hello.

¿May I suggest you do your circles without any tool compensation? That is to make sure that no hardware error is the cause of your problem.

In our shop we ran into such a problem when we were just beginning with the CNC world. The way to solve that problem was to scale our CAD drawings so that the generated path included the tool compensation instead of having the actual machine tool to do it. If instead of circles you get ovals then you can suspect mechanical problems. Maybe even a miscalibration error.

I hope this helps.

Regards.

[machinedude]

the thing you need to keep in mind is when your cutting circles you are actually cutting 4 sections of the circle so every 90 degrees your going to notice the transition to the next segment of the circle. this is where your error will be most noticeable. i agree with the backlash being the cause. when you get into critical size holes Circular Interpolation is not the way to go. you have to do things old school and resort to a boring bar and even with those your limits end up being the spindle run out. it could have some calibration error but i don't think that would cause the start and end point error of the radius of the diameter your cutting. i think that is more on the fine tuning of things if the circles are out of round with the radius segments having a decent match on the start and end points.both are worth looking into but i think the backlash is probably more of the problem. loose coupling would be a major cause of backlash. in general there will be small amounts of backlash just because is not cheap to build a drive with zero but those small amount normal to a DYI build should not be as extreme as what is pictured by any means.

after looking at this some more and the second set i could be more of a machine issue. it looks like the machine does better based on the choice of a conventional cut vs climb milling. this would probably be caused by machine rigidity taking a guess. something to keep in mind when programming if the better of the two is acceptable.

[routalot]

I can see arcs on your spoilboard that look less segmented.While I would go along with the probability of backlash,I would also be checking the software to see how much chordal deviation is allowed.I think a look at Task Manager to see if there are several other processes running would be useful too.

[machinedude]

this is just reaching at straws here. but typically when when climb milling the cutter tends to want to climb up over the cut and on a manual machine you tend to avoid climb milling all together. with the cnc machines the motors hold the axis in place so you can cut either way on a cnc machine because of this. but it got me thinking about how aggressive your cuts are that you are making? i was thinking maybe if the cuts were too aggressive combined with steppers being open loop systems and the combination of timing belts in the system maybe if the cuts got to aggressive you might get enough force to cause the timing belt to jump a tooth? i would maybe try very light easy cuts to see if that would get rid of the issues? maybe plunge a hole with the cutter in a reference position you can come back too after you see the problems to confirm your not loosing steps or loosing position mechanically.

[mh1470]

Thanks for the interest guys.

We've confirmed that the cause is backlash. Not backlash from our rack and pinion, but backlash from the keyway on our motors. We performed some more precise backlash tests and found that the motor on the y axis was able to move slightly while not moving the pulley at all. When we were building our machine we didn't have the equipment to cut a keyway into our pulleys and gears, so we opted to just use grub screws, hoping that they would suffice. We can now see that although the grub screws may have held fine at first, some of them have become loose, especially the one connecting the y-axis stepper motor to our pulley system. The amount of backlash this is causing is almost 3mm, which perfectly accounts for our errors with the circles, and also for some other alignment issues we were having while cutting rectangles. We're purchasing a broach set today and are going to remake all of our motor connections so that the keyways are working as they should.

Hopefully I can update this thread with good news once we get everything back together.