PDA

View Full Version : Need an accurate grid for backlash comp.



Jimno2506
03-03-2008, 10:46 AM
Machinists,

Does anyone know where to find an accurate grid that I could put under my mill spindle camera to adjust backlash and view actual position after a move?

I've done the dial indicator method but for some reason, when I get to cutting, I don't end up where I'm supposed to be (off a few thou. here and there). I'd like to be able to watch the location of the mill spindle and adjust backlash, or whatever is causing the problem, to get accurate locations.

Thanks,
Jimno

SGW
03-03-2008, 10:58 AM
Disassemble a broken glass scale and use a piece of that?
Acu-Rite scales have 1,270 lines per inch.

mochinist
03-03-2008, 02:12 PM
Did a search for "metrology grid" and "calibration grids" and this site popped up.
http://www.maxlevy.com/bulletin1600.cfm


Maybe you could etch one yourself from a printing.

How accurate are most printers?

Evan
03-03-2008, 02:44 PM
Your mill should be quite capable of scribing a test pattern by hand. Work out a way to mount a pin in the spindle and drag it across a piece of polished aluminum with perhaps a .005 spacing for each line. You don't need a crosshatch since all you need to do is rotate it 90 to calibrate the other axis.

clutch
03-03-2008, 03:09 PM
How much backlash do you have using the dial indicator method? If you have a lot of lash the tool can force the table off position. Comp justs puts the axis in the right place considering axis reversals, it doesn't make the system tighter.

Can you see what kind of following error you have?

Do you have any idea of the pitch accuracy of your screw?

Clutch

Jimno2506
03-04-2008, 10:22 AM
I've not been able to verify the accuracy of my, or any other printer. I would think a plotter would have an accuracy value, but have found none.

Perhaps my problem was how I was checking the backlash; I was using the jog feature to move it .001 at a time. I recently read that I should write a g-code to move it like G01x0, then G0x-1 and take the reading from there. Turns out I'm right on with backlash enabled. I didn't have time to run a part and see if all axis returned to x0y0 after finishing, but I will shortly.

Doing the g-code method I find quite a bit of backlash, which surprises me since I just had the x ballscrew repacked and they said the screw was "good".

x backlash .010, y .021. but when I run it with backlash comp I get perfect returns using the dial indicator method at 1 and 1.5" moves in both axis.

I'll get a better idea how I'm doing when I run some parts.

Regards,
Jim

Carld
03-04-2008, 10:27 AM
Are you trying to reinvent the wheel? A DRO would do everything you are trying to do and backlash is not an issue with a DRO.

Jimno2506
03-04-2008, 10:49 AM
The only DRO I have is in the CNC software.

I figured the graph paper might be a inexpensive alternative to a real DRO.

Carl,
Not trying to reinvent the wheel, just find a good alternative. Thought I'd see if the vast knowledge on this forum could be of help.

What kind of hard-wired DRO would incorporate backlash into the positioning?

Regards,
Jimno

Carld
03-04-2008, 01:05 PM
The DRO reads the table movement, not the feed screw movement. I am not a CNC person but I think a servo and sensor reads the screw on a CNC, not the table.

I would think that if I were making a CNC mill I would use a DRO to read the table position rather than work off the feed screws. In that way it wouldn't mater if the screw had a half turn backlash the table would only start moving and the DRO registering movement when the table actually moves and goes to the next stop point. I don't have a clue if there is a program to do that with a PC.

Of course, back lash is a big issue with CNC machines. Will it be CNC or manual?

On the other hand you may be doing this to operate like an optical magnifier while machining since I just saw a thread about this from the past get resurected.

I guess my posts serve no relative purpose to your project.

I reread your first post and perhaps I misunderstood your question. What it seems your asking is when you watch the cut with a PC camera and stop at a certain point the cut is more than you think you stoped at. Well, if your watching a point on the table and stop at that point then the cutter is still under a load it will continue to cut untill it is free. That would cause overtravel in the cut while the table shows the cut to be exact. I still don't see where the backlash in a feed screw has anything to do with where the table stops and where the cutter finally ends up at.

There also may be an issue with paralax in the lens.

clutch
03-04-2008, 01:33 PM
The DRO reads the table movement, not the feed screw movement. I am not a CNC person but I think a servo and sensor reads the screw on a CNC, not the table.

I would think that if I were making a CNC mill I would use a DRO to read the table position rather than work off the feed screws. In that way it wouldn't mater if the screw had a half turn backlash the table would only start moving and the DRO registering movement when the table actually moves and goes to the next stop point. I don't have a clue if there is a program to do that with a PC.

I work on machines that have both encoders and glass scales. It looks like they use the rotary encoder for the rough positioning and the slide mounted scales for the fine work.

As far as the backlash, a scale system would try to fight backlash, but that is going to be a matter of response in reading, computing and positioning.

Considering the investment made in the basic cnc machine, I've often wondered why scales seem to loose out to rotary encoders. Would seem to be a small increment in price for better response.

Clutch

Evan
03-04-2008, 01:38 PM
Carl,

Your average home shop CNC machine has no idea where the table or the lead screw is. They run what is called "open loop" meaning that there is no sensing system at all to report on what position was achieved when movement is commanded. All movement is completely blind unless a limit switch is hit. This is standard practice for small CNC stepper powered CNC equipment.

Even with servo motors instead of steppers the feedback is usually derived from an encoder on the motor or leadscrew, not the table.

Feedback systems are not simple to implement. As soon as you give position feedback a whole new set of possible problems crops up such as position hunting caused by excessive system gain. Even with feedback the system must have some dead band or "hysteresis" to prevent hunting and overshoots. Feedback in a closed loop system can also cause jitter as the digitized position value flickers +- one bit of value. The cure for most of these types of problems is intentionally induced electronic "backlash" even if the mechanical system has none at all. In fact, a closed loop system with zero backlash is the most susceptible to these kinds of problems.

Carld
03-04-2008, 02:28 PM
I see, it tells it to move and then you determine what you got and go from there. For that you should always call short and then finish out by hand.

oldtiffie
03-05-2008, 08:38 AM
I've not been able to verify the accuracy of my, or any other printer. I would think a plotter would have an accuracy value, but have found none.

Perhaps my problem was how I was checking the backlash; I was using the jog feature to move it .001 at a time. I recently read that I should write a g-code to move it like G01x0, then G0x-1 and take the reading from there. Turns out I'm right on with backlash enabled. I didn't have time to run a part and see if all axis returned to x0y0 after finishing, but I will shortly.

Doing the g-code method I find quite a bit of backlash, which surprises me since I just had the x ballscrew repacked and they said the screw was "good".

x backlash .010, y .021. but when I run it with backlash comp I get perfect returns using the dial indicator method at 1 and 1.5" moves in both axis.

I'll get a better idea how I'm doing when I run some parts.

Regards,
Jim

Hi Jim.

Good question.

First the accuracy of a printer and printer versus plotter.

The printer is a plotter (mostly) and is also a computer-controlled stepper-motor-driven machine with an "X" and "Y" axis.

Most if configured correctly are extremely and very surprisingly accurate.

Many CAD packages (ie AutoCAD) have a feature that allows you to plot to the printer/plotter, measure the physical output and if in error to adjust is so that the printed size is exactly as per the electronic size. Say plot a 10" wide X 8" high CAD file. Measure with an accurate engraved drafting scale. If in error re-put the "as drawn/plotted" size and the CAD software will re-scale the plotted output. It also has a feature where you can plot a say 8" circle to the screen, measure the X and Y axis of the circle and if in error re-put it back into the software which will re-adjust your screen circle to be correct accordingly.

Now draw and plot a graph as required and use it on your mill with your web cam.

There is one vital item that is either ignored. not known or over-looked when talking of CNC and DRO accuracy as regards tool paths.

That is that neither the ball-screw nor DRO will pick up any "cross" error or slackness in the fit of the table and its "ways". Tightening gibs etc. may not be enough.

Example.

If there is - say 0.002" - slack or wear in say the "X" travel, neither the "X" nor "Y" DRO will pick it up and the CNC is not aware of it or may not be able to compensate for it.

The only/best way for say the "X" way (table) cross-error (slack or wear) is to put a dial indicator on the side of the table and pointed in the "Y" direction. Now if you pull or push the "X" table in the "Y" direction you will see all or part of the 0.002" wear/slack on the dial indicator.

On a heavier machine under light load "up-milling (Ie not "climb-milling") it may never be a problem. But if the load is increased and/or climb-milling is used it will become very evident with the table and/or the milling cutter "jumping" or "thrashing" and/or the job being "out" by the amount of the wear/slack.

A similar result can emerge in a CNC machine if there is significant back-lash in the ball-screws or if the ball-screw pre-load etc. is insufficient.

The net effect of all this is that the CNC may have stepped correctly and if DRO's are fitted, nor error will be shown - but it is there in varying amounts in any machine as no machine is perfect.

Of course machine and cutter deflection under load are also contributing factors too as is ball-screw "wind-up" under load and release/reduction as the load eases.

I hope this is clear. If not I will try to clarify it.

I expect some comment from some who may question my assertions/comment/advice.

Jimno2506
03-05-2008, 09:30 AM
Thanks so much. Everything you said makes sense to me.

Regards,
Jimno

Paul Alciatore
03-06-2008, 01:17 AM
All of the above is good, but backlash is only one type of error that can occur with an X-Y table. If the gibs are loose, the table can also rotate. This can easily produce 5 or even 10 thousanths of error in a position. If your screws are in fairly good condition, this rotation can easily produce a lot more error than the screws do.

With readouts of any kind, rotation of the table can still produce errors even when the readouts are dead on. For one thing, the readout's scales are almost always mounted at an offset from the center of the table or the axis of the spindle. Even a pair of absolutely perfect scales would only guarantee that "perfect" accuracy at the point where the two scales intersect as viewed from above. That's the point they are actually reading, not where you are drilling or milling. If the table can rotate, it will and your tool will be displaced from that "perfect" location by some error amount.

Proper adjustment of the gibs is very, very important for the best accuracy on any machine. For the best accuracy, any axis of movement that is not actually being moved for a particular cut, should be LOCKED DOWN - TIGHT. This is easier to do with a manual machine as a CNC one would need frequent pauses for operator attention when making most parts.

And there are other sources of error in most machines.

oldtiffie
03-06-2008, 01:51 AM
All of the above is good, but backlash is only one type of error that can occur with an X-Y table. If the gibs are loose, the table can also rotate. This can easily produce 5 or even 10 thousanths of error in a position. If your screws are in fairly good condition, this rotation can easily produce a lot more error than the screws do.

With readouts of any kind, rotation of the table can still produce errors even when the readouts are dead on. For one thing, the readout's scales are almost always mounted at an offset from the center of the table or the axis of the spindle. Even a pair of absolutely perfect scales would only guarantee that "perfect" accuracy at the point where the two scales intersect as viewed from above. That's the point they are actually reading, not where you are drilling or milling. If the table can rotate, it will and your tool will be displaced from that "perfect" location by some error amount.

Proper adjustment of the gibs is very, very important for the best accuracy on any machine. For the best accuracy, any axis of movement that is not actually being moved for a particular cut, should be LOCKED DOWN - TIGHT. This is easier to do with a manual machine as a CNC one would need frequent pauses for operator attention when making most parts.

And there are other sources of error in most machines.

Thanks Paul.

That nicely rounds it off and sums it up.

That stuff that you posted re CNC and the non-locking/clamping of the non-operating slides is what has bothered me with CNC as it flies in the face of traditional practice.

Mustn't forget the quill in the case of all vertical mills, the column clamps on a column mill and the knee clamps on a knee mill.

I am surprised and disappointed at what lateral movement there is between the clamped and free conditions/movement/location of slide-ways (quills included). I expect that a lot of people will assume that it must be OK and don't check.

It is an interesting exercises to say cut a face along the "X" axis (with the "Y" axis clamped) and take a medium-fast cut bot conventional and climb milling and read off a dial indicator reading lateral movement. It's quite a bit actually, and as said no DRO will find or identify it.

barts
03-06-2008, 10:50 AM
Evan is right as usual... using glass scales for feedback requires a _lot_ of mechanical stiffness between motor and scale; any looseness anywhere (ballscrew wear, gibs, connector torsion, etc) will cause resonance if the gain is high (needed for good positional accuracy under load and for multi-axis sync). One useful technique as clutch pointed out is to use both.

One company I worked with back in the 80s used digital encoders on their very accurate positioning stages; once the stage was close they would use A/D converters on the two raw analog sine waves from the encoders and subdivide that signal into 255 positions. They're (Anorad) is still using this technique with even higher resolution, I see from their website. We built a 30' long CMM using equipment from them and others; had to error map the thing using laser interferometers and a rather complex forward kinematics model..... it took months to get it to work right....

- Bart