Issue with step servo setup - x-axis cutting skewd

The system is a plasma table running ac brushless servos in position mode. EMC is generating step/dir signals using Mesa cards.

If we setup a square with four holes (one in each corner), the holes will skew in the positive x direction proportionally to the acceleration value given to x. For instance if I set x accel to 10 I get 0.200 of skew in the positive x-direction for those holes. If I increase the accel to 25, it reduces the skew to about ~0.120. If I setup a G64 P0.05 it doesn't noticeably change the result with this accel value. If I reduce the accel speed to 10 and use a G64 P0.05 the skew is reduced to 0.020 but it's still not correct. So basically in open loop mode is doesn't seem to know when to slow down and or by how much.

So I'm very confused about these results and how to fix this. I could hookup the encoder signals back from the servo amps, but I'm unsure if the control would be able to correct this without setting up a PID loop, which I dont know if you can do with step/dir. I could run the drives in torque/velocity mode but the advantage of running the drives in positioning mode is they fully auto-tune.

Any other thoughts?

I don't know where the problem lies. The machine is a Komatsu KCR-0951, when it ran the original komatsu controller, it did not have an issue.

When that control died, we retrofitted it with a Hypertherm Edge control, with the existing dc brush servo drives in velocity mode. The machine showed the same skew (offset) problem as it does now.

Since then we've replaced the drives and installed and setup emc2 and now we're seeing the same issue we had with the edge controller.

Some tests we've run:

1. 100 back to back 0-0.50 on a dial gauge accurate within 0.0001.

2. Jog in 1" increments - random # sometimes 7 - 15" then back up to exactly the same position as before. The laser finder will be on the spot where we started at and the encoder count on the drives will read as it should.

3. Cut a 4" square - accurate within 0.002.

4. Now add 4 0.50" holes in the corner of that square and FAIL. All the holes are offset shifted towards the positive x direction.

Getting very frustrated with this.

Random thoughts:

Isn't your acceleration extremely low? 10-25 IPS^2 is really low for a servo plasma machine
I would expect a number closer to 200-500 IPS^2

Do all holes have the same lead-in direction?

Is there an offset in the machine after the holes are cut? (that is, is this a dynamic motion problem or a static counting error )

All the holes have the same lead-in's.

All the programming is done with sigmanest.

How would I test for an offset in the machine?

When I start a program I write down the encoder outputs on the drives. At the end of the program, I check it again and it's exactly the same after it returns to start position.

So the drives atleast think they are doing their job.

I can try raising the accel. I wasn't sure what a good value should be.

This was the machine running with the original komatsu control and komatsu power source. We had no error at all.



This was the machine running with hypertherm edge2 control, dc brush servo's in velocity mode.



The error we have with the second setup is now present in the current.

Since you had the same error with a different controller and servo hardware. it really seems like a CAM/G64 type error of some kind, as this would change with acceleration. Note that you need to change both MAX_ACCELERATION and the stepgenerators MAX_ACCEL to change acceleration.

So if the encoders are in the same position at the end, is the machine in the exact same position?

It does seem to change with acceleration. That seems to be the only thing that effects it. If I were to cut 10 of those squares, they would all be exactly the same incorrect.

I will look at both values. I've only changed one of the accel values, not both.

I guess I could scribe a point where the laser hits the plate, run the program and see if it returns to exactly that spot. We've done this testing with random jogging and it did.

On the mechanical side, we've replaced all the belts. We've checked all the tracks for debris. We've cleaned everything. It's like a ghost in the machine.

Much appreciate the suggestions!

I raised the MAX_ACCELERATION to 200 on both axis, and it was visibly faster - no issue. I could not find a MAX_ACCEL in any of the .ini or .hal files.

The result is still offset slightly to the right in this picture:

I marked the start point using the laser on the plate and ran a couple programs back to back, the end point is exactly the start point each time.

I don't get it.