[SCCoupe]

Im using mach 3 with servo motors and gecko g320's. Also have ballscrews and ballnuts. When repeating a simple circle, the y axis seems to loose ground. After about 30 passes, the center of the circle is gone and im left with horse racing track shape, round on the corners and flat on the sides. I mic the flat sides and they are the same, so the x axis is doing its job, but the y axis doesnt return to the same point. Ive tried with and without backlash compensation, and get the same results. Plus if it were backlash I would expect to see a not so round circle even after 100 passes. Any ideas?

Thanks all.

[[email protected]]

So, you are ending up with a race track.
So you are telling us the race track is what the circle
should be when it comes to the width.
The length of the race track is longer than it should be.
If I am getting this right, numerous times through the
program would eliminate tool pressure and this would
allow the axis that is giving you the longer direction of
the race track to coast further when the tool pressure
is reduced.
I think I would turn the machine off and start pushing
things around by hand with indicators placed here and
there. What you are looking for is when you move an
axis, it shouod come back to the zero you have set.
If you have incremental feed on your control, set up an
indicator and move in one direction until the indicator
shows movement, then while counting, move in the other
direction until the indicator moves. If you find lost motion,
try to isolate the where the lost motion is coming from.
Could be in the ball screw or the ball nut. But it could also
be in the Gibs in the Z axis.
A little more discription regarding the size of the tool, size of
your programmed circle and the length of the race track.
How about that, no cussing or nuttin.
Regards Walt.

[HuFlungDung]

Is it still the case with the Mach software, that it runs open loop servos? This means that the computer does not know that a positioning error has occurred, and some degree of binding on that one axis might be accumulating and creating this out-of-round effect.

[ger21]

Yes, but the Gecko drives will fault if the encoder count is off by 128 counts, I think.
Sounds like either backlash, or noise on the encoder line. Note that backlash comop doesn't work well if your in CV mode. Try running in Exact stop mode with backlash comp.

[[email protected]]

I believe it could be electrical however, I would like to eliminate any
chance of somethine mechanical. If this is electrical, maybe some
electrical components could be swapped and see if the racetrack
changes from one direction to something 90 degrees different.
Regards Walt

[One of Many]

Closed loop systems that result in short travel can be the result of noise getting into the feedback, seen as more counts than actual or slipping between the motor and the lead screw thinking it made it but never actually getting there. Lost feedback would result in excess travel.

DC

[NC Cams]

Welcome to my world - only we had the same deal with servos and learned some real expensive lessons in the process of getting there.

First and foremest, you MUST eliminate hysterisis (aka backlash) PERIOD.

THen you have to make sure that motion in + and -, X and Y are 100% proportional and linear. How well??? How round do you want????

After thousands of dollars of parts in upgrades and improvements and week of work on a decent machine tool to start (not home brewed by any stretch) and finally a call from a machine tool service guy, we learned that we had invested:

bad armature in motor ($1300 rebuild),

sloppy ball screw bearings (like new OEM bearings were replace by $825 EACH axis ball screw bearings and we still had 0.0001 or 0.0002 backlash in the "SYSTEM" even though bearings had 0.0000 backlash at 500lb preload).

sloppy gibs (adjustment to the point of barely no slop with minimal drag, can not get perfection)

1.5 days of machine tool tech ($100/hr plus travel, meals and lodging) tuning the system,

Replace Hard drive (bad sectors for read write of ramdrive for look ahead interpolation)

$250 ball screw reload to eliminate backlash

and weeks, no months of our own time and god knows how much in phone calls not to mention lost business and p/o'd clients.

We're now round to within 0.0001 or 0.0002 and runout of hole is within 0.001 or less when mounded and eccentricity checked with $15k computerized cam measuring equipment/software. Runout and roundness are NOT the same.

With steppers, I don't see how you'll match what we do with servos. Reason, there is no way to make up for electrically lost steps and you have no way to correct - what's randomly lost is screwed forever and uncompensateable.

How much error is 127 steps worth as you will keep running fat dumb and happy as you won't get an shutdown from error until you get over 128 steps behind???

Moreover, if you consistently lose 20-30 steps and it takes you 3600 steps to do a circle (10 steps per deg), you can get pretty lame size and never get an error fault. Never saw steps added unless noise creeps in and that another random mess altogether.

Ok so add encode feedback to do true feedback error correction but that essentially turns you into a servo, but not quite as there still will be lag time in the error correction loop.

If you hve randomly added electrical noise, it will be easier to scratch your nose with your unbroken arm than to to hold roundness - ain't gonna happen.

BTW, a stepper based OD grinder (high dollar one too, based on industrial grinder) simply can't make round parts - they are "round" to the eye but fail even a 0.001" interval runout check. A check with a 0.0001" resolution will make you laugh or cry, depending on which end of the make/reject scale you're on.

Think CNC is easy, huh???

Make a perfectly round circle (within 0.0002 or better) with CNC anything. My neighbor's Haas can barely do it under a thou TIR and a $1.5million CNC grinder can do it witin a few tents to half thou TIR with relative ease.

This does not count "lobing" which is not the same a TIR but does affect "roundness". Perfect roundness costs money. How round do you want/need it and how much time and money can you dedicate to achieve your goal????

EDIT

Sorry, miss read the part about the system driving servos.

Issue remains the same - the computer is NOT getting direct feedback when you simply rely on STEP/DIRECTION to run. Computer still things the Geckos did what they were supposed to do after 43960 steps cause nobody told it that it didn't.

You still have to find/eliminate hysterisis.

You still have to make sure servo gain is IDENTICAL in X versus Y.

Simple trick: swap axis drives, then swap motors. If the problem follows the hardware, you found your problem. If not, look further....

END EDIT

[SCCoupe]

Thanks for the help guys, ill check some of this stuff out when I get home today. I did resolve most of the problem yesterday. I cleaned the y axis ball screw again and replaced the ball nut with another new one. I also rechecked the number of step per inch. I eyeballed this before and was over a quarter inch out. This time I drilled a hole, sent it out 6 inches, drilled another hole and then put the micrometer on it. Every time, I move the machine past where i needed to be and then came back to it so that backlash wasnt in the picture. I have both axis' much much closer now. I find the gibs very hard to adjust, so they are not tight, there is a very small amount of play. Whenever I tighten them even the slightest bit, the table will no longer slide. Is this normal? I need to double check the mach3 modes you all speak of for the backlash. I also need to get ahold of a dial caliper to measure the backlash properly. Can the G320 provide feedback to the PC? This kinda bothers me. I assumed that the controller was taking care of the steps to make sure they were all there. That was my whole reason for going with servos. What good is using an encoder if it doesnt care of there are missed steps?

Thanks all!

[ger21]

Can the G320 provide feedback to the PC? This kinda bothers me. I assumed that the controller was taking care of the steps to make sure they were all there. That was my whole reason for going with servos. What good is using an encoder if it doesnt care of there are missed steps?

Mach cannot miss steps. If Mach's DRO says it moves a step, a step was sent. You can buy an add-on encoder board from www. rogersmachine.net that will let you monitor your encoders position in Mach3 and pause the machine if there is a deviation greater than you specify.

[KTP]

If you are using servos with a Gecko the Gecko will tell your controller if you have missed more than 128 steps by signaling a fault. I think your problem is with noise on the encoder lines. Do you have the encoder lines shielded and have the shield grounded at the Gecko drive? Also, do you have the step/dir lines from the parallel port shielded, and the shield grounded at the computer?

If there is no noise, perhaps your motors are not tuned very well and you do not have a stiff enough response (not enough gain). Try increasing the gain pot (and also increase the damping pot along with it). This will of course make your motors "sing" a bit louder and heat up a bit more, but it is better than having them constantly lagging 20 or 30 steps because they are not tuned properly.

My Pixie P100 controller lets you set the position error from 1 to 20,000 so you could know exactly how much your motor was lagging from true position

Sorry, had to get that plug in there

Rutex drives also allow you to set the position error before fault.

Pixie controller
www.skyko.com

Rutex drives
www.rutex.com

[SCCoupe]

None of the wires I used are shielded at all. Not to the encoder or to the parallel port. Sounds like a project for this weekend. Also, the Gecko's arnt really tuned. I dont have an O scope to do that so they are set to the default that the directions say to start with. So much to do!

[One of Many]

Looking into the G320 manual, I see this is a step and direction run servo system. It does have an output for fault handling(term 5) and there is an application example there. Mach3 would need to monitor this I suppose? The error handling does not correct for the fault as I understand it. It should just halt the operation.

It would be realistic to have better control over lost steps other than a fault output. If the feedback does include noise. the position error(test point, not a terminal) should indicate that as commanded speed verses actual speed. This sounds more like a following error type of condition, but that term is used elswhere in this document as part of the fault indicator.


The motion scale in distance per step will effect it too. I'd imagine this to be in concert with the encoder counts per inch also.

If you have experienced any fault conditions during the move, then this could be mechanical restriction since you mention needing to leave the gibs loose. Backlash will be a constant value and not accumulate each pass.

There are a few other contributors I'd look at too. Step and Direction pulse voltage, encoder voltage and motor current limits. In step and direction or encoder marginal voltages could fade in and out missing portions of the expected pulse stream during a run. Certain noise here could muffle these signals too. Even so, the drive monitoring these should introduce a fault signal, but could reset after 3 seconds as stated in the manual. By then it is too late.

DC

[KTP]

"None of the wires I used are shielded at all. Not to the encoder or to the parallel port" (chair)

I am 90% sure shielding those wires will make some or all of your problem go away.

5% chance your gecko is not tuned very well.

5% chance it is something else.

Let us know how it works when you get some shielding on those wires. I bet if you could hook a scope up to the encoder lines as they come into the G320 you would see noise spikes all over the place. Same with the step line from your computer. Make sure you GROUND the shield on one end.

People on this zone just do not stress enough how important proper shielding is

[NC Cams]

Noise will cause problems - PERIOD. Unless you eliminate it, you're going to have problems maintaining roundness.

Just because a step was SENT by MACH, doesn't mean it was RECIEVED or acted upon properly. Set your error tolerance to 1 and see how well the system responds. If it pukes, keep working on noise abatement until errors don't occur. You'll be surprised at what you learn...

Yes, gibb slop will result in position error (position error is NOT backlash compensateable). The trick is to minimize it or manage it but unless you go to true table position feedback, a STEP/DIRECTION system can't deal with it - it doen't/can't know that it (table slop or tightness) occurs.

We spent a small fortune in time and money getting a servo system, with true computer closed loop feedback, to cut round circles - if you count 0.0002" steps at the direction changes "round".

Fix the noise, fix the mechanical slop, do it systematically. Then, find out the remaining resultant error and fix some more. Been there, done that and it isn't easy or cheap....

[KTP]

"if you count 0.0002" steps at the direction changes "round". "

Quit bragging NC_CAMS

You are doing this on a bridgeport or something IIRC? Doesn't temperature change or a housefly landing on one end of your table affect things almost as much as 0.0002"? That is 5,080 nanometers...only 7x or so from the wavelength of deep red light!

[One of Many]

Yup, I suppose the only thing this type of closed loop system can track is the difference of step pulses from the PC against whatever encoder pulses the Geco sees. +/- 128 seems like a large window of error without correction. Since no direct position feedback to the PC is offered, this barely qualifies for a closed loop control.

Even the closed loop stepper driven systems will make a correction at the end of a run. While that may be fine for straight line moves, interpolated paths could end on an odd correction kirchank in a circle. Yikes!

DC

[NC Cams]

Yes, it is a Bridgeport EzTrak but it is major tuned up version.

Using canned external circle mill EzTrak program, we cut round circle in a blank that we use to make cam grinding masters. We then read roundness of the part on our cam measuring machine which can measure to 0.000001" deviations of profile.

At the 4 direction change points (12, 3, 6 & 9 o'clock), we see profile deviations - these are then smoothed until we see NO deviation. Difference is 0.0002" at the direction change "blips" in the theoretical errors from perfect roundness.

Not trying to brag, just demostrate what our cam measuring machne (an EzCam by Andrews Products) reports which is cross checked against an industry standard cam measuring device (the Adcole) as the "standard" of cam measuring accuracy.

The EzCam uses a Landis CNC ground master for comparable comparison.

Only reporting what the EzCam machine says and we bought the EzCam cam measuring software that was verified against an Adcole cam measuriing standard.

Even the Bridgeport guy who tuned the machine (Brian at BPT Machines, Carol Stream, IL) sat in slack jaw amazement of the performance of the Eztrak's performance after he tuned it.

Not bragging, only trying to point out what it takes to solve the problem of cutting out of round circles via a CNC mill. We've even checked our neighbor's Haas VMC using same methods.

[[email protected]]

Back in the 1900s I was installing electronic bookkeeping machines. These machines were for the most part replacing mechanical bookkeeping machines. We had noise problems one right after the other. It was hard to explain to an electrician that ground "zero volts" was more important that the 115 to 125 volts AC. We ended up running many new circuits from the service box to a receptacle. When the electrican ran the wires "in conduit" we asked them to run three wires. One black, one white and
one green. We also asked them to make the green wire one size bigger that the blk and wht wires. This started more heated discussions than you can think of. This was in the late 1960s. People just did not understand ground, zero volts and quiet lines.
People think it is a big joke when you walk accross the rug and the spark jumps 2 inches to your hand when you reach for the doorknob. That kind of voltage will cause most electronic systems to go bonkers. Might even leave a trail of smoke.
Run a new line, include a ground wire. No not rely on the seven pieces of conduit between your plug and the service box to provide a clean zero volt ground.
Use shielded cables, ground only one end (I can't remember which end) increase the size of the ground wire by one size. If you have access to a scope, hook it up and watch things for a while. Be careful what you ground the scope to because everything you are looking at will be relative to what ever you are grounded to.
If everything changes when the phone rings, you have a bad ground. On one occasion that was the complaint, "Wrong addition when the phone rings". The problem was a loose connection in the service box.
Noise will drive you up the wall until you can provide a quiet source of power to your piece of equipment.
Best of luck, check your ground,check for loose connections, use shielded cable and try to eliminate binds and lost motion. Wow, that is easier said than done.
Regards Walt.