Pathpilot Backlash Compensation

Hi Folks. I would love to see a discussion on PathPilots new backlash compensation feature.

I start the ball rolling with this introductory video. Please post your thoughts in this thread.

Cliff

https://www.youtube.com/watch?v=KP1k3X57hDY&t=8s

Oh dear!
I've commented here in the past that backlash compensation can work for you as long as you understand the limitations. It CAN also work against you.
The picture most of us have assumes the table movements (lets not consider Z here) lag behind the programmed position by 1/2 backlash in each direction. This is only true for light cuts.
Those familiar with manual mill work will know that climb milling on a manual mill can be problematic, if not dangerous. The mill can dig into the cut and pull the work and table forwards, into the cut. This still happens with a CNC mill but the ballscrews have much less backlash so this effect becomes very small and is mostly ignored. However, the ways on a CNC machine are also very low friction and therefore table is fairly easy to move. This increases the likelihood of the mill climbing up the cut. Under such conditions the table will be moving AHEAD of the programmed position by 1/2 backlash. Again this amount is very small but if backlash correction is enabled this will ADD an additional 1/2 backlash, making the net result worse than without backlash correction! This is most often found on finishing passes around a contour with a high(ish) depth of cut or presumably when machining harder materials (not something I do). How high, or how hard depends on multiple factors.
I use backlash correction sometimes but I use a macro to enable it just for particular ops where I'm sure it will have a positive effect (seems to work!). I disable it for all other operations.
Measuring backlash (or lost motion) should be done under conditions as close to actual milling as possible to reproduce a realistic friction/stiction which will be different to a static measurement without spindle or cutting vibration. I mill a circular boss and measure the diameters just before and after 12:00/06:00 and 03:00/09:00. The difference between the before and after measurements is TWICE the backlash in that axis.
Step

keen mentions that some high end machines use ball screw mapping to improve accuracy. I believe that LinuxCNC supports ball screw mapping but no UI interface to the capability is available in PathPilot.

Quote:

---

Originally Posted by TurboStep

Oh dear!
I've commented here in the past that backlash compensation can work for you as long as you understand the limitations. It CAN also work against you.
The picture most of us have assumes the table movements (lets not consider Z here) lag behind the programmed position by 1/2 backlash in each direction. This is only true for light cuts.
Those familiar with manual mill work will know that climb milling on a manual mill can be problematic, if not dangerous. The mill can dig into the cut and pull the work and table forwards, into the cut. This still happens with a CNC mill but the ballscrews have much less backlash so this effect becomes very small and is mostly ignored. However, the ways on a CNC machine are also very low friction and therefore table is fairly easy to move. This increases the likelihood of the mill climbing up the cut. Under such conditions the table will be moving AHEAD of the programmed position by 1/2 backlash. Again this amount is very small but if backlash correction is enabled this will ADD an additional 1/2 backlash, making the net result worse than without backlash correction! This is most often found on finishing passes around a contour with a high(ish) depth of cut or presumably when machining harder materials (not something I do). How high, or how hard depends on multiple factors.
I use backlash correction sometimes but I use a macro to enable it just for particular ops where I'm sure it will have a positive effect (seems to work!). I disable it for all other operations.
Measuring backlash (or lost motion) should be done under conditions as close to actual milling as possible to reproduce a realistic friction/stiction which will be different to a static measurement without spindle or cutting vibration. I mill a circular boss and measure the diameters just before and after 12:00/06:00 and 03:00/09:00. The difference between the before and after measurements is TWICE the backlash in that axis.
Step

---

Thanks Step - Really good points. I have done the milled boss test also - I should have included it in that video. I rotated it in my lathe with a indicator and found additional, non symmetrical errors, probably caused by slideway 'racking' . Another issue that would be hard to compensate for.
Cliff

Quote:

---

Originally Posted by keen

Thanks Step - Really good points. I have done the milled boss test also - I should have included it in that video. I rotated it in my lathe with a indicator and found additional, non symmetrical errors, probably caused by slideway 'racking' . Another issue that would be hard to compensate for.
Cliff

---

My first thought would have been to use an indicator in the mill. How large are these other effects? I wouldn‘t have expected anything significant, at least I‘ve never seen anything that I couldn‘t (I believed) attribute to anything other than backlash (at light loads). Did you plot the result?
Step

Edit: have you repeated a boss test with backlash correction enabled?

Yes, as I replied to your video, I agree with Step that backlash compensation is likely to worsen backlash in many situations. There are lots of myths and misunderstandings about the perceived benefits. The business of ballscrew mapping keeps coming up too, yet that is quite different to backlash, addressing instead the absolute positional accuracy of the ballscrew pitch.

I'd like to see a way to measure and plot the backlash of a machined test piece (boss or bore), rather than a static, unloaded measurement on one axis. You can't very well use the same machine to probe the end result, as the machine backlash will further modify the measurements. I wondered if using a touch probe to digitise a precision bore gauge or cylinder around the circumference in the same way as you'd do the machining would achieve the desired result. Sounds like a nice idea but unfortunately probing isn't done in the direction of the surface - it's done at right angles to it on a good day.

Perhaps the simplest method would be to make a mechanical version of the "ballbar" used on high end machines. Instead of using an electronic method to measure the error as the "tool" traces a programmed circular path, a DTI could give a direct indication of the difference between the planned locus and the actual. Perhaps a Haimer type 3D DTI tool would be an option, used to probe a bore gauge? Could be expensive....

It's relatively easy to measure static backlash with no load involved but once you start to cut metal, there are some serious forces involved. The actual characteristic is a combination of backlash (hysteresis) and spring. The measurements will differ according to how aggressive the cut is. A ballbar connected to the head rather than the spindle would allow measurements to be made while either real cutting forces or a spring load were present.

Quote:

---

Originally Posted by Muzzer

I'd like to see a way to measure and plot the backlash of a machined test piece (boss or bore), rather than a static, unloaded measurement on one axis. You can't very well use the same machine to probe the end result, as the machine backlash will further modify the measurements.

---

I was thinking of using an indicator in a similar manner to the method shown at around 5:00 in the following video. The mill will bring you very close to the zero position and once centered accurately the indicator can be swept around 360° to roughly plot the deviations. Obviously there will be no accurate measurement of the angle (unless you have a spindle encoder) but that's not really necessary.
Step

https://www.youtube.com/watch?v=sRfpXuBDWt0

Quote:

---

Originally Posted by Muzzer

Yes, as I replied to your video, I agree with Step that backlash compensation is likely to worsen backlash in many situations. There are lots of myths and misunderstandings about the perceived benefits. The business of ballscrew mapping keeps coming up too, yet that is quite different to backlash, addressing instead the absolute positional accuracy of the ballscrew pitch.

I'd like to see a way to measure and plot the backlash of a machined test piece (boss or bore), rather than a static, unloaded measurement on one axis. You can't very well use the same machine to probe the end result, as the machine backlash will further modify the measurements. I wondered if using a touch probe to digitise a precision bore gauge or cylinder around the circumference in the same way as you'd do the machining would achieve the desired result. Sounds like a nice idea but unfortunately probing isn't done in the direction of the surface - it's done at right angles to it on a good day.

Perhaps the simplest method would be to make a mechanical version of the "ballbar" used on high end machines. Instead of using an electronic method to measure the error as the "tool" traces a programmed circular path, a DTI could give a direct indication of the difference between the planned locus and the actual. Perhaps a Haimer type 3D DTI tool would be an option, used to probe a bore gauge? Could be expensive..

---

I agree that ball screw mapping has very little to do with backlash. On relatively low accuracy machines such as a Tormach screw mapping is probably irrelevant; both mapping and backlash compensation has been in LinuxCNC for quite a while but not conveniently accessible to the user. It is great that Tormach has made it easy to play with.

Manually using a DTI is apt to be more useful than probing. Consider that most probes have lobing errors and as you move around the periphery of the circle the probe's orientation is constantly changing. I'll have to look into ballbars!

Can't wait to see what Cliff comes up with next but he will have thought about it very carefully and he comes at this with decades of experience....

This is getting more interesting!
I did the boss test back in 2008 shortly after buying my first Tormach and was disappointed to measure the real world errors.

See attached PDF. I did not record the specific shape errors - but I remember they were not symmetrical.
By "I should have included it in the video" I mean discuss this subject or repeat the test. I wish I had time to get into this more deeply - hopefully soon.
Cliff

Eagerly awaited!

Quote:

---

Originally Posted by keen

This is getting more interesting!
I did the boss test back in 2008 shortly after buying my first Tormach and was disappointed to measure the real world errors.

See attached PDF. I did not record the specific shape errors - but I remember they were not symmetrical.
By "I should have included it in the video" I mean discuss this subject or repeat the test. I wish I had time to get into this more deeply - hopefully soon.
Cliff

---

I made some hex bits for my power screw driver, and there was a thou or two difference across the flats, turned out this didnt make much difference in the use of them as the clearance was greater than the error. I even considered changing the model to take into account the error...... still dont have a roundtuit for that one ha!!

This has come up in a couple newbie posts but I suppose it's worth asking you veterans of machining, what are your expectations from your tormachs? I can say for my part, the more accurate I can measure a thing, the more disappointed I am when it's not perfectly true to size, but even so, I recognize that a total error of .001" (.025mm in Keen's test) is pretty darn good and totally acceptable for most applications. Having worked in the automotive and firearms industries over the course of my career, I can say with some authority that a 0.002" total tolerance is rare and typically avoided whenever possible because of the rapid increase in parts cost that results from putting that on a print. A total tolerance of 0.010" is much more common as it generally has no appreciable affect on a well designed assembly and results in substantial cost savings for each part. That all being said, we're talking about machines here that are among the least expensive options on the market for a "large" working envelope mill. To expect accuracy and precision at a level that big industry knows to be difficult and expensive to achieve seems almost foolish. Yes they are or can be made capable of hitting such goals but to be disappointed in results that hit a total tolerance less than the requirements of major manufacturers just doesn't make sense to me. Maybe that's my newbness talking?

Quote:

---

Originally Posted by soofle616

This has come up in a couple newbie posts but I suppose it's worth asking you veterans of machining, what are your expectations from your tormachs? I can say for my part, the more accurate I can measure a thing, the more disappointed I am when it's not perfectly true to size, but even so, I recognize that a total error of .001" (.025mm in Keen's test) is pretty darn good and totally acceptable for most applications. Having worked in the automotive and firearms industries over the course of my career, I can say with some authority that a 0.002" total tolerance is rare and typically avoided whenever possible because of the rapid increase in parts cost that results from putting that on a print. A total tolerance of 0.010" is much more common as it generally has no appreciable affect on a well designed assembly and results in substantial cost savings for each part. That all being said, we're talking about machines here that are among the least expensive options on the market for a "large" working envelope mill. To expect accuracy and precision at a level that big industry knows to be difficult and expensive to achieve seems almost foolish. Yes they are or can be made capable of hitting such goals but to be disappointed in results that hit a total tolerance less than the requirements of major manufacturers just doesn't make sense to me. Maybe that's my newbness talking?

---

I bought my Tormach to learn the trade, it did a good job of this, it doesnt do aerospace work, but makes decent parts for the money invested.The more you spend, the more accurate the parts, for the most part this is true.

Quote:

---

Originally Posted by soofle616

This has come up in a couple newbie posts but I suppose it's worth asking you veterans of machining, what are your expectations from your tormachs? I can say for my part, the more accurate I can measure a thing, the more disappointed I am when it's not perfectly true to size, but even so, I recognize that a total error of .001" (.025mm in Keen's test) is pretty darn good and totally acceptable for most applications. Having worked in the automotive and firearms industries over the course of my career, I can say with some authority that a 0.002" total tolerance is rare and typically avoided whenever possible because of the rapid increase in parts cost that results from putting that on a print. A total tolerance of 0.010" is much more common as it generally has no appreciable affect on a well designed assembly and results in substantial cost savings for each part. That all being said, we're talking about machines here that are among the least expensive options on the market for a "large" working envelope mill. To expect accuracy and precision at a level that big industry knows to be difficult and expensive to achieve seems almost foolish. Yes they are or can be made capable of hitting such goals but to be disappointed in results that hit a total tolerance less than the requirements of major manufacturers just doesn't make sense to me. Maybe that's my newbness talking?

---

Hi - Well you raise some fair points there.
The problem is certain classes of work, eg mechanical work, often requires accuracy of better than 0.001" For example toolmaking, engine repair, machine and instrument building, (for sub example interpolate milling a pocket for a ball bearing race , a mold core or a die). That ideally needs to be within 0.0002 "... 0.001" is a problem. Some folk do this work from a small workshop with low cost machines. I have been running a business doing this for decades. It can be done if the machine errors are well understood and worked around. See prehardened P20 steel injection mould example. But doing this work on low cost machines is definitely challenging and you have to love that challenge.
Cliff

It doesn't surprise me that there are talented people in the world getting performance from cheap equipment that would make big expensive machines blush. Hell, back in my racer days there was a kid riding a bone stock bike that was blowing everybody away no matter how good their setup was. Ratty leathers, suspension that looked like it was going to bounce him into orbit and body language that everybody thought was going to end up in a gruesome crash one day but there he was, whipping laps at near expert speeds. That sort of thing always makes me wonder though, how much BETTER could that person be given the right tools?

Anyway steering back to the direction of this thread, Your point " I recognize that a total error of .001" (.025mm in Keen's test) is pretty darn good and totally acceptable for most applications." Is a fair point.

Yes If you can get your Tormachs backlash down to that it is pretty good and will be acceptable for many. A high end VMC with care and skill is good to around 0.00025" total positioning error. But is more like 3 or more times the price, and wont fit into, or be powered in most shops anyway.

I said I was disappointed to find my machine was not in that league - but at the same time I knew I was asking too much.

All the same, I think it is a good thing if we can find ways to improve our machines accuracy and get closer to that VMC. We will not get to 0.00025" But we can get closer towards it. I wish I had asked in the video if people would post their backlash in the comments below the video. Going off various things I have heard, it seems some have machine set up errors that give lost motion/backlash greater than 0.002 ..even 0.003" - and this has sometimes really damaged Tormachs reputation, and caused some rants and machine returns.

I think a better understanding of this subject will help many owners, and Tormach.
Cliff

My 1100 always had excessive backlash, I did make a shim for the ballnut spacer and it took most of it out of the Y axis. I dont recommend this for everyone, this was not a correct fix but it did work. I think when the ballnuts are set up they are on the loose side (too much clearance), this is determined by a spacer between the two halves of the ball nut.

It is a low priced machine and I think it does a relatively good job for the price point.