Carl,
Remember one change at a time.
Have you tried any of the other suggestions to eliminate components?
What current set resistor do you have on the G201's?
Can you post a picture of you mill?
RICH

RICH,

I started with the easiest test which was to eliminate the possiblity of inductive coupling.  I noticed that the X and Y were gaining steps when run together but were right on when run alone.  Seemed like there must be some coupling but I'm no EE.   When I brought the cables directly out of the top of the box, I did the original test which was a simple x1y1 and then an x0y0 and it worked amazingly even if I did if 50 times.  I then did the test in my last post and was happy when the 10 cycles worked.  But when I added that 1 line things went south.  I'm trying to do one step at a time and test enough before I declare victory.

I just received the new parallel cable and will try to setup a breakout box so I can put a scope on it easily and also connect directly to the drivers.  I wanted to rule out coupling before moving to the parallel cable.  Now I'm just trying to undersand the results of the test.

I have the drivers set for 6A for the X and Y axis using a 270K ohm resistor for each.  The Z axis driver is set for 2A with an 18K ohm resistor.

In any event, here are some pictures of my setup.  Let me know if you need more details.

Best Regards.

Carl

Carl,
Thanks for the pictures as it really helps just knowing it's not some 40 year old mill ( mine is !).
One thing mechanicaly woud be the attachment of the screws ball to the underside of the tables and if the bolting moves you will never get consistantcy. So thats something worth eliminating. This would also be more pronounced at higher travel speeds. I assume you can't rack your tables, when my friend installed his ball screws into same kind of mill
as yours is was a PITA as the anchoring of the nuts had to be just right with the gibs adjusted.

Just a comment on testing, namely keep your moves the same, ie. same pattern.

Reduce your velocity to 10, and keep the acceleration low and maybe make your moves always over say 3" square and accross the corners so you have the same pattern to judge against. It's just an inerita thing.
Just for kicks,lower your tuning, do a test,  then do a test with the  heavy vise removed. This can give you a glue if the nuts are moving.

Not an EE either, so excude if the lingo is off , in terms of noise traveling / being inductively coupled to other wires, the solution can be difficult and sometimes like black magic. If you find something like spreading wires makes a difference just play around with those wires and gage results. ....One thing at a time...... The electrical field around the wires will increase as as current, voltage, and of course frequency is increased. The field can build up over time and at some point could cause interference with a device.  Sort of sounds like what would that one extra line of code do! Shielding will help this as it contains that field, neatly wrapping aluminum foil 100% around each cable is a quick shield without trying a different cable. Additioinaly just have the shield touching against the grounded box and not the stepper.

Just some thoughts,
RICH

Hallo Carl

I had the same problem with a lathe that I converted. My problem Was solved by replacing both the geko drives. I had  the same drives as in your pic`s The drives I have now is Geko g203 v drives red in colour.This is my 2cent`s worth.

Hope it helps
Cobus

The last few post and your pictures do not answer your question which was " Why does the addition of one line cause a 60 thou error"
What you do not say is whether the error was due to missing steps or additional steps. We must assume that it is missing steps.

What this means is you are loosing 6 thou per cycle. I have analysed your program , and in the first program each axis changes direction twice. In the second program, each axis changes direction 4 times.

I have run this into my machine and run it several times. I am not sure but the answer might be "Constant Velocity"

The movement is made up of several sharp changes of direction - i.e. not straight lines back and forth, but sharp 10 degree, 20 degree etc angles and I noticed that with constant velocity on (my normal state), the axis stopped well short of the X-3 Y-1.5 position. It then travelled down to the X3 position and stopped bang on (a 90 degree turn follows). It was similarly short on the zig zag of the additional move entered.
Using my digital calipers I noticed that when running in CV I was 2 or 3 thou out. (measuring one axis (Y) only).
I changed to exact stop and watched the DRO's - the axis stopped at the precise distance. X0Y0 was bang on.

Now I have scratched my head, becasue, in theory, the fact that corners are "rounded off" should not affect the positional DRO's and so the final position should be the same. If you are loosing one or two steps in the sharp corners being rounded, then that well might be where your discrepancy occurs - especially if you are running at speed.

Cobus,
 When you changed Gecko's, did you re-route any wiring ? Or did you put everything back just as it was ?
Just curious,
RC

And btw, is that a Vampire ? (one of my favorites)

Hallo Rc

No the gecko drives can Unplug on the electrical connections and I undid the screws removed the drives. the new drives was fitted in reverse.What I did see when I exchanged the drives at our local supplier, was that there was a stack of these drives that were returned to him. I can now make 30 to 50 components and it keeps the size within 2 to 5 microns .

Cj

Hi all,
If you swapped cables so that you used the x drive for the y drive and vise versa, wouldn't you see a change in returned position if it was a drive or signal path problem?


Put a small dot on your vise. put a toothpick in the chuck, manualy move table to allign toothpick point to dot, lock the quill. Just an easy visual check.
You should be right over the dot after running either of the following.

In absolute and exact stop and using the MDI line for moving:

G01 X1 Y1 F20
G01 X0
G01 X1 Y0
G01 X0

Now swap the x & y cables to the steppers  ( make sure you shutdown the drives before disconnecting the cables)

G01 X-1 Y-1 F20
G01 Y0
G01 X0 Y-1
G01 Y0

This provides for same crossed pattern. (Note this is 4x to 2 y movements, thus you favor an axis.
Make a file and test using constant velocity or exact stop.

RICH

Before you start swapping all the cables about - THE ERROR LIES IN THE CONSTANT VELOCITY, and is a combination of the narrow angle, speed, and relative low number of steps per unit.

If Constant Velocity is engaged, as an axis nears the end of its travel, the computer starts to calculate the next move, and actually implements it. Running the test piece, my axis was stopping some 30 thou short of the mark, and reversing, with constant velocity engaged. On Absolute Stop it ran the full distance.

I have 60 pulses per thousanth on my machine, so it was stopping some 1,800 pulses short of the mark. At a 4 degree angle, the difference in position of two lines at the 1800 mark is 125 - or 2 thou. Let us assume my old eyes are not as quick as they were and we were only 28thou short of the mark, then the distance between two lines at 4 degree would be 1.95 thou.

I do not know the precise method of calculating the CV "offsets", but if we have the old X position advancing up one path and the new X position advancing towards it on a new path, (at a known angle) then the Y distance between them can be calculated, based on the tangent of the angle, and if it exceeds the present Y position, then that is incremented.

The crunch comes as the two positions pass each other (and this only happens for acute angles), then in my example, Y may be incremented by one thou, but is short for the next thou, therefore nearly one thou is not incremented.

Now my machine has at least 6 true steps per thou, so (depending on the CV calcs) could be incremented by 1/6 thou, but if your steps per inch ar small, you will loose out.

Now Mach 3 will gaily carry on the second line, thinking it is on the correct y position ( which again could be being continuously calculated if this is at an angle to the axis) and when X reaches 0, the Y axis says I'm here - but it is a bit short. The same scenario happens to the X axis if the angle of the lines is rotated 90 degrees since the Y axis takes the lead.

I don't really want a discussion about the maths, since I have necessarily picked figures to help my explanation, but the fundamentals work with whatever angles etc, you pick - although the narrower the angle the worse it gets.

I think if you try all your tests in Absolute Stop mode, instead of CV mode then you will get accurate results - certainly more accurate than before - bearing in mind your relatively small number of steps per unit.

If you are in Absolute Stop mode then £$%&*&%$£   

never seen this problem with CV, will have to check it out today and see.
Hood