Servo idea - elimated backlash - zero backlash !

Here is an advanced idea that is quite feasable and completely possible.

When you get backlash on your system, it has to be compensated, and as I understand it, software can compensate for it, but how well, and to what resolution? I can imagine it being even worse when trying to engrave back and forth on something very small and fine.

Usually the servo motor drive has an encoder on the rear, so your software/electronics knows exactly where the motor position is.

The idea is to have the optical encoder not read directly from the motor shaft, but from the gantry position itself.
A simple construction of this is to have an optical encoder operated by a thin tight high tensile wire attached to the gantry and running across to the encoder shaft, on the side of the machine. (less moving wires too!)

Or the wire fixed at each end and running along the axis, and wrapped a couple of times around the encoder shaft (specially made pulley, like in an old wind up clock mech.) The encoder which is mounted on the gantry.

I am sure there are more methods of actual construction, including an optical laser based position sensor? (prone to dust though) and I ask anyone, if they have any thoughts on this, to reply & include them here.

The reason I mention this is that my servo gearbox (3:1) has some noticable backlash within the gearing, and I'm thinking of ways to eliminate it, as I've not yet had any experience with a complete machine. Still building!

My intro: http://www.cnczone.com/forums/showthread.php?t=27905

My gallery: http://www.cnczone.com/gallery/showg...0/ppuser/36476
Thanks

Tony

It has been tried, but I think the results are less than what you would hope for. The reason being, when tuning the servo, trying to create an optimum PID filter, the mechanical backlash amount interferes with the stability that you can achieve in the PID feedback loop. This is because a tiny error in position of the axis is not proportionate to the motor position, ie., there is the inevitable dead spot where there is no response even though the control circuit has attempted to bring the motor into position. Then, all of a sudden after the backlash space has been traversed, small increments in motor position do start to matter.

It is also the case that a tool in the cut can exert enough force when climb cutting that it pulls the machine table across the backlash space, taking a heavy chip, or leaving a mark when an axis direction is reversed during circular motion.

Mechanical backlash should be eliminated with better mechanicals, the electronic compensation has its limits. It will work to a certain extent, but the settling time for the axis to reach true position is much longer, and so the best tuning settings will be kind of spongey, instead of crisp and distinct.

To add a bit to Huflungdung's comment:

Bridgeport tried this (use linear encoder instead of motor encoder) on their early Eztrak mills. It didnt' work for the reasons cited.

The thing that is hard to fathom until you do some math is that you can electrnically do all sots of things digitally or via analog tricks - however, there ineveitably becomes a point where one can not get the mechanicals to differentiate to such small integrals mechanically what you can do digitally or electronically.

Finally, mechanical backlash can be minimized but usually not eliminated. Afterall, you NEED clearnce between mechanical devices for them to move with respect to each other. THus, eliminating ALL sources of mechanical lash is sort of oxymoronic if you want/need the parts to move with respec to each other.

The clearance you need for stuff to move smoothly begets the same problems that you wish to eliminate - if that isn't an oxymoron, what is????

I agree with HuFlungDung and NC_Cams. Mechanical backlash creates it's own set of troubles regardless of what you do in the servo system to compensate for it. Tooling forces can cause the table to move around within the backlash zone even if the servo doesn't move. Also, the inability of the control to close the position loop and remain stable is another problem.

What was somewhat successful in some very high-end machines was a "dual" positioning loop. With a dual loop, you have a rotary position feedback device in the motor, which does the lion's share of the positioning. When the velocity of the servo approaches zero, another linear feedback device, such as a glass scale or inductosyn scale takes over and provides the zero-backlash position feedback. This helps a bit, but it does not eliminate backlash either. These machines always had high quality ballscrews and thrust bearings.

If your ballscrew/ballnut assembly has very little backlash, but there is "pitch error" along the length of the screw, THAT can be compensated for in software very easily by mapping the error along the length of each axis. This works much better than backlash comp.

The biggest mistake people make with backlash comp is to compensate TOO MUCH. When you put an indicator against the slide of a typical CNC machine and step-jog back & forth, you can easily confuse ballscrew "wind-up" with backlash. The best way to measure backlash is to move the slide at least 1/10 inch to position in the + direction, coming up to "zero" on an indicator. Then, move the slide 1/10 inch in the - direction and approach the same position from the other direction. This requires an indicator with some overtravel, but it's more accurate than just stepping back & forth.

The problem with having mechanical backlash in a servo drivetrain is that it becomes impossible to tune the servo system for any kind of performance. That is because as you tighten up the parameters to get decent accel and decel, everything gets thrown off when the drivetrain is in the backlash area where there is very little inertia. Usually the system resonates and jacks off, trying to beat itself and the crummy gearbox to pieces. You can crap up the tuning on it so that it moves, sort of, but then it will only perform as well as a much less expensive alternative.

I ran into this by subbing out engineering on a very expensive retrofit of a special purpose machine, once in a different life. The thing had a LOT of backlash, and turned out to be unsuitable for the servo control installed on it. It didn't need servos, but the whole retro system was built around them and installed, and the machine didn't work. The servos on this system cost over $12,000. What I ended up doing to fix the problem was to invent a timing belt transmission and install it at my expense, to replace the gearbox. The thing worked after that, and I didn't sub out any engineering any more.

Short version of the story is that it can be expensive to discover that servos need tight coupling to the load.

--97T--

Hmmm... thanks everyone for your input and information.
I had wondered if tuning the PID loop might be a problem, and obviously it is :)

Thanks again.. I am at the start of the homemade CNC road - there is alot to learn!

Tony

My mill was originally retrofitten in the manner you state. belt drive servos to ground ballscrews and heidenhain linears on the axis. It actually worked pretty decent. there is about .002 backlash in each axis (Soon to be fixed, thrust bearings were shot) and under the original control it worked pretty well. I have now retrofitted to Mach3 with Pixie controllers. Initially I tried connecting the slides to the pixies (The pixies take encoder feedback and allow you to control +/-10v servo drives with step/dir) but was unstable. the table would move back and forth about .001 as the pixies tried to find home. I ended up installing 2500line encoder on the backs of the motors. I will just use the linears as dros. I think now that if I turned the signal gain down on the drives I could have gotten acceptable performance, but what done is done.

I used to have some very high precision xy tables that had direct drive preloaded ballscrew. feedback was with a laser positioning system directly on the axis.