Does anyone have experience using linear glass scale as servo encoders on vertical milling machines.... I recall reading that dro scales such as Anilams will cause servo fault problems due to machine vibration but motor shaft encoders are not affected.

Also has anyone used their servo motor rotary encoders with an Anilam DRO. If so,, how is it done........

Thanks,
Dennis

Heidenhain have a mill system that uses their Scales for position FB. You may get problems if the backlash is high, for precision ground ball screws it is not usually a problem.
Electronically, the scales are the same as rotary encoders.
Al.

Having a scale system is regarded as more accurate than encoders.
Encoders read off of the motor or screw and do not actually give a "real" measurement - if there is backlash the system will not compensate (by itself) and parts will be wrong.
Scale system gives direct measurement of the table travels so moves are always the right size. If machines get too much backlash then the servo tries to position the table, normally the table (when sloppy) will overshoot the target repeatedly back and forth and things get kinda crazy, but at least you know there is a problem.

So it seems that there is some possibility here.. I have a RF 45 equipped with an Anilam wizard 800 dro. I am currently installing ground ball screws that I obtained a while back on the x - y, and will eventually also add servos to cnc it....

Has anyone actually built a system using the linear encoders. If so, can the DRO operate in parallel with the servo boards.. What kind of isolation may be needed ???

You will have to throw the DRO in the garbage and send the scale signals into a controller card and go from there.

Woh,,,

Can't throw a $ 1500 programmable dro in the trash.. Is it possible to read the quadrature output of the glass scales with both devices ???? The scale system is running on 5VDC regardless of the dc source.. The principal idea in this is to capitalise on both the dro as a backup or manual use and also use linear encoder direct measurement for better servo accuracy... If it cant be done I would still leave the dro intact and use rotary encoders on the servo motors...

It could be done, but it involves some fancy (electronic) foot work.
I would power the scales from the DRO, this would mean the DRO would be on at all times, and make a TTL buffer circuit fed by the output of the scales and output the buffer signals and common to the input of my control, keep in mind that the two systems would not be isolated.
It might pay however to use the two independantly and plug the scales into whatever system happens to be in use.
As you would only be using one or the other.
Al.

Bridgeport tried the linear scale route to feedback the Z axis on their Extrak mill. Ultimately, they had to go to motor encoders, supposedly due to vibration, hunting and oscillation problems.

Pretty sure it was because they did NOT have a true ball screw drive on Z but rather a rack and pinion as I recall.

The problem you always have to deal with is hysterisis and/or slop.

If you monitor the motor position, the servo will go where you tell it and stay there. The table might not be exactly where you want it due to slop/backlash/whetever but that is usually manageable.

If you use linear encoders and have backlash in system, the motor will tend to hunt as the table is going to move and the motor WILL try to keep up with fixing/adjusting for the movement.

I'd contend that a sloppily/uneven worn gibb based system would be especially problematic with a scale system due to the slop potential.

If you have linear guides that are slop free, the scales would probably be dead on - but then too, so would a well tuned motor mounted encoder based system.

We had a similar but exagerated problem. 1957 Niles 30 ft. dia. vertical boring mill remanufactured to 4-axis with Siemens 840D control. The size of the machine & workpieces caused the Heidenhain glass scales to vibrate and cause loss of feedback. We replaced the glass scales with Newell Spherosine round tubular rod running thru a reader block that stabilized the rod at the measuring point cancelling the vibration problem. It worked so well that we replaced the other scales on all the VBM.s in that shop, even the manual machines with DRO's

I suspect that successful application of linear scales probably includes a sensor on the motor and careful tuning or sensor fusion. I know people are using sensor fusion for this problem.

Can I ask did you get the ball screws cut for you to size or are you machining them yourself? Where did you get them?


So it seems that there is some possibility here.. I have a RF 45 equipped with an Anilam wizard 800 dro. I am currently installing ground ball screws that I obtained a while back on the x - y, and will eventually also add servos to cnc it....

Has anyone actually built a system using the linear encoders. If so, can the DRO operate in parallel with the servo boards.. What kind of isolation may be needed ???

IMHO, a linear encoder on a CNC is just a way around doing things right... that is having a rigid, vibration damping frame, accurate linear rails/ways, and an accurate ballscrew system.

If all the components are functioning the way they should, even a cheap no-feedback stepper system should give you excellent results with error that's difficult to measure.

Bridgeport Machines tried the glass scale/servo interface on their early Extrak machines and ended up replacing them with motor mounted servo encoders.

Since there always will be some hysterisis and/or lost motion between the servo and the glass scale, it is pretty much impossible to get accuracy and/or vibration free motion due to the lost motion issues.

Both DZASTR"s and ZUMBA's admonitions are pretty much spot on. The use and interfacing of glass scales is tempting (especially with DRO's as they are) but it has been proven that this sort of feedback is not a wheel that needs to be reinvented.

It's not a problem that's impossible to solve. I have a Moog VMC manufactured 15 yrs ago with Heidenhain glass scales on it that will still repeat down to a tenth. I know it's had a less than easy life, so it can't be said that the machine has no wear at all either.

It IS possible, and has been done with great success by a number of manufacturers.

Perhaps it's out of the realm of what a DIY'er can handle, but I'm not even convinced of that.

Here is a Galil video that shows how it is done by tuning the PID in an Advanced dual-loop system.
http://www.galilmc.com/training/v_dual_loop_comp_method.html

Using scales for feedback could be a way to get by with less than perfect ballscrews. I used to run a 1980 vintage mill with 11 feet of x axis travel, and heidenhain glass scales, and a point to point control. The ball screws were not very good. The z axis had .100 inch backlash in the ball screw, but the head counterweight was on the light side, so the backlash was always in the down direction, and was not a problem. The y axis would sometimes overshoot .010 to .012 inch if you tried to rapid up to the desired position. The x axis would occasionally get "stuck" .001 away from the position, the servo motor would just sit there humming for ,what seemed like a minute, until it got there. If the table overshot, it would not go back. It was not a very good machine, by todays standards, but much better than a manual machine.

Here is a Galil video that shows how it is done by tuning the PID in an Advanced dual-loop system.
http://www.galilmc.com/training/v_dual_loop_comp_method.html

Bingo.

I was just about to post that when I replaced the X Axis drive motor on that machine that the servo had an encoder lead, meaning that the control was using BOTH encoders to control the oscillation and overshoot generated by any backlash in the system.

It is very easy to see how a system with only the linear scales and a lot of backlash would hunt for it's end point, and that video lecture (which was quite well done BTW) explains how the control can make use of the second encoder to damp the unwanted motion out.

The isssue of whether or not it can be done is essentially a moot point - Reason: someone can and will have the skill and technology and take the time and effort to do it. The problem comes when you start dealing with the how to and how much does it cost.

Any number of times that I've explained how to get milling machines (ala Bridgeports and/or the clones) to get essentially zero backlash/hysterisis axis', an issue is raised about the cost. Sadly, the cost for precision, zero/near zero compliance ball screw bearings is typically out of the range of affordability for the DIY'er.

Yes, you can use less costly bearings BUT they can not/will not have the performance that the "good ones" do. Result: you will end up with compliance/backlash. Depending on your feedback logic, you may or may not be able to deal with/manage/eliminate the resultant lost motion from the compliance.

Then you get into the issue of motor response. ANY backlash can be dealt with UNTIL/UNLESS you want rapid/high speed moves and do a lot of rapid direction changes. Keeeping up with bidirectional slop becomes a never ending nightmare after a while.

We tried it that way (tuning) and finally gave up. We then set out to get rid of the slop/hysterisis and, after spending lots of time and money, we still had to tune the motors. However, at that point, motor tuning basically dotted the I's and crossed the T"s with regard to smooth and deadly accurate system motion.

I suspect that factory equipped with glass scale equipped servo systems that are deadly accurate with no backlash were engineered to be that way from the get go. They probably have preloaded ball screws and well fitted/adjusted ways or gibbs. They also probably have a very good set of servos and servo drives.

Yes, it has and can be done. It has also been tried by seasoned pro's and abandoned because they couldn't get it to work adequately/properly and/or economically. Like most successful endeavors, it is all in the planning and execution.

Now that I will certainly agree with.

My machine has preloaded ball screws and huge end block mounts, which makes for very nearly zero backlash.

Even in that system you do have to tune the servo drives to each motor (which was a rather large PITA I will add).

If I were to build a machine from scratch I would bypass the linear scales in favor of a conventional rotary encoder and a drive system that eliminates as much lost motion as possible. In the end it will cost just as much, but it'll be quite a lot easier to set-up and tune software wise.

I've been pondering taking a large swing manual lathe and converting it to linear guides, ball screws, etc, but I am afraid that by the time I do a good job on the conversion I will has spent at least as much as what a good used slant bed costs, and I will still be out the ages of time spent on the project.

I guess sometimes you just gotta know when to back off and punt. ;)

Does anyone have experience using linear glass scale as servo encoders on vertical milling machines....

Well not vertical milling machines but diamond turning lathes yes. While still machining a huge difference in technology.


I recall reading that dro scales such as Anilams will cause servo fault problems due to machine vibration but motor shaft encoders are not affected.

Garbage! It is a question of where the vibration is, but if the gain is high enough vibration impacts rotary encoders just as much as linear. Now this doesn't mean that specific mounting arrangements with linear encoders won't make issues with vibration worst, just that it is another engineering problem.

Given all of that vibration would kill the whole machining task in a diamond turning lathe so great lengths are taken to isolate the machine form the environment. Even so when the lathe switched over to rotary encoders it still would react to vibration.


Also has anyone used their servo motor rotary encoders with an Anilam DRO. If so,, how is it done........

I'm not sure why you would want to do this, in fact I'd think it would be regressive. A properly mounted linear encoder is the only way to go with a DRO as it eliminates the problems associated with wear and backlash.


Thanks,
Dennis

The equipment I use to work on had both types of encoders mounted on each axis. The linear encoders for gross movements with resolution of about a micron and the rotary encoders to provide very high resolution at a much slower speed. When running on the rotary encoders the resolution was out to the nanometer range even if the lathe didn't resolve that in the real world.

In any event the control lops had very high gain. Have either one of the encoders go bad ( develop a bas spot on the engraving) and the drives would react to the faulty device. ... And do very bad things!!! Same thing with vibration, at the wrong time and in the wrong place the drives would react to the vibration.

Dave

The isssue of whether or not it can be done is essentially a moot point - Reason: someone can and will have the skill and technology and take the time and effort to do it. The problem comes when you start dealing with the how to and how much does it cost.

This could result if a very long conversation, but one reality that can't be denied is that Rotary encoders are dirt cheap.


Any number of times that I've explained how to get milling machines (ala Bridgeports and/or the clones) to get essentially zero backlash/hysterisis axis', an issue is raised about the cost. Sadly, the cost for precision, zero/near zero compliance ball screw bearings is typically out of the range of affordability for the DIY'er.

Sad costs is an issue but a Bridgeport isn't really a good platform for this sort of conversion. To the original posters question, related to vibration, we need to point out that a rotary encoder has little advantage here.

As to the DIY'ers, I'm of mixed opinion here. From the standpoint of economics many of us are forced to open loop steppers and excessive backlash and slop in a system built this way is just as bad. Given that you have a platform where low backlash and well fitted ways have been implemented, linear encoders can work well. The cost though is much higher for the encoders so economics often pushes rotary encoders for a given machine cost.


Yes, you can use less costly bearings BUT they can not/will not have the performance that the "good ones" do. Result: you will end up with compliance/backlash. Depending on your feedback logic, you may or may not be able to deal with/manage/eliminate the resultant lost motion from the compliance.

The problem is that is you have a sloppy mechanical system driven by servos with mounted rotary encoders you never know about the compliance issues. Such a system goes on its merry way.


Then you get into the issue of motor response. ANY backlash can be dealt with UNTIL/UNLESS you want rapid/high speed moves and do a lot of rapid direction changes. Keeeping up with bidirectional slop becomes a never ending nightmare after a while.

You bring up the issue of system response which highlights another problem with linear encoders or at least linear encoders of reasonable cost. That is the encoder having enough resolution to run the control loop. One problem with DRO scales is that they are not normally high resolution devices. Ideally you want to have your feed back to be at least 4X the desired machine resolution.


We tried it that way (tuning) and finally gave up. We then set out to get rid of the slop/hysterisis and, after spending lots of time and money, we still had to tune the motors.

Tuning the motors is almost a given on any high performance machining environment I've been involved in.


However, at that point, motor tuning basically dotted the I's and crossed the T"s with regard to smooth and deadly accurate system motion.

I suspect that factory equipped with glass scale equipped servo systems that are deadly accurate with no backlash were engineered to be that way from the get go. They probably have preloaded ball screws and well fitted/adjusted ways or gibbs. They also probably have a very good set of servos and servo drives.

Well fitted ways and gibbs are an absolute requirement! No way to get around that as are minimal back lash in the leadscrews. I don't think that achieving this is out of the realm of a DIY'er though. Certainly some skill is involved.


Yes, it has and can be done. It has also been tried by seasoned pro's and abandoned because they couldn't get it to work adequately/properly and/or economically. Like most successful endeavors, it is all in the planning and execution.

I might add 'expectations' to the planning and execution. Rotary encoders win so often with the issue of economics it takes exceptional expectations to justify the use of linear encoders. Like all things cheap though you do give up something in return and that something is knowing in reality where you table is.

Dave

That is the encoder having enough resolution to run the control loop. One problem with DRO scales is that they are not normally high resolution devices. Ideally you want to have your feed back to be at least 4X the desired machine resolution.


How about the Heidenhain 0.1µm line , when this is increased in the control by multiplying the resolution x4, you could end up with 0.025µm
Al.

Linear scales are a great idea for the inexpensive machines. They would cure a host of evils. The idea that the feedback loop is really closed by rotary encoders is just not accurate--a relatively limited portion of the feedback loop is closed. Being able to measure the actual position of the axis and just not whether the shaft has turned the desired amount opens the doorway to greatly improved performance because a lot of real phenomena can be measured directly and accounted for:

- Backlash, as has been mentioned.

- Irregularities in the lead screw, which are always present, but which are worse with less expensive components.

- Thermal expansion characteristics.

- Wear over time or wear that arrives due to retrofitting old machines and using surplus components.

These are all areas where linear scales could help, not hurt, IF THE PROPER SOFTWARE WERE AVAILABLE. It's no accident that I've capitalized the latter, because software is the real issue for this. Mach 3 doesn't really support closed loop operation that is remotely able to take advantage of what linear scales could offer. I'm not aware of any software available at remotely reasonable prices today that would work for this application, and that's the real rub in using linear scales for these applications.

It's a real shame too. The sentiment that they are only appropriate if the rest of the machine is equipped with near perfect ways and ballscrews is much more a reflection of software shortcomings than it is a reflection of the real potential linear scales could have.

Given how cheaply DRO scales are available these days, if Mach 3 actually had the capacity to really use the devices to their full potential, the accuracies achievable with cheap components would be truly suprising. Think I'm wrong? Take a look at the accuracies that have to be achieved with cheap disk drives for computers. They're way in excess of what the cheap Asian tool crowd only dreams about.

Someday it will happen, whether via Mach 3 or some other vehicle. The fact that Mach now has plug-ins may enable some enterprising group to figure out how to offer a Mach upgrade to do it. I would definitely pay for the feature.

The most likely performance challenge I can see are the response times on the cheap scales are not too great. However, we're not talking about building 200ipm rapids machines here, nor do we need to. A real workable machine capable of 60-100ipm and repeatable accuracy of 0.001" or perhaps 0.0005" which be a real step up. The incremental cost for DRO scales and a little software would be very little. Combine it with the upcoming "unstallable stepper" technology and you'd see an amazing new price performance regime established in short order. Add a little active chatter reduction technology (also cheap to implement save for the required software), fill the bases and columns with epoxy granite, and I think folks would be surprised at what these cheap machines could do.

Now who wants to get started writing the Mach 3 plug-in to support it?

Best,

BW

Probably also a bit on the expensive side. The other issue is that I was under the impression that Heidenhain scales are analog which means additional interface circuitry. I could be wrong on this though.

One thing you do have to watch out for with the high resolution device is the data rate back tot he controller. You may get the resolution but the controller might not handle the data at fast axis movement rates.

I know that Pnuemo used high resolution linear scales on some of their precision machines but they are not very fast at all. That may not be completely related to data rate either but it is almost funny to watch a normal CNC lathe run and then look at a Pnuemo.

How about the Heidenhain 0.1µm line , when this is increased in the control by multiplying the resolution x4, you could end up with 0.025µm
Al.

The linear quad signal scales are made exactly the way the rotary are except of course in a Linear fashion.
Heidenhain make both low level sine wave and Standard TTL line driver outputs.
This one thing to watch for when buying Heidenhain, otherwise you have to either buy their expensive interface or make up a recieving circuit for the Sine wave type.
Al.

I read this thread rather quickly, but felt I need to throw in my thoughts on this.

The problem is in the rate feedback NOT the position feedback. If you consult control literature in the 90's you find this to be a research issue. It is termed "co-location of rate sensor and actuator".

If you study the attached picture, you'll find two masses attached by a spring. Let mass m1 represent the motor, m2 represent the table, x1 and x2 the positions respectively. The spring represents a "sloppy" connection; in this case--backlash.

To stabilize this system, you will need position feedback and rate feedback of mass m2; the table. However, because of the sloppy connection, the rate feedback MUST come from m1; the motor, hence, the term co-location of rate sensor and actuator. If, in your system, the rate feedback in interpolated from the linear scale, the motor (actuator) will hunt position all the time. If you could install a rate sensor on the motor and if your machine allows for separate feedbacks of position and rate, the problem should go away.

I hope I did not confuse anyone. I am not sure I explained this well, but it is late here, so I'm going to sleep.

Salah Zenieh
Ph.D. Control engineering

reading this thread, I think it is the way to go.

rotary encoder on the servo or stepper
linear scale on the table
cheap rolled ballscrews
proper table guideways
proper ballscrew bearing

hardware able to collect the 2 feedbacks

Mach 3 plug in to sort it all out.

I am looking forward to this day to come

Daniel.

The HARDWARE combo outlined in post #27 is sort of a "dreamware" scenario. It has been a machine tool designer's dream to be able to use cheap hardware in concert with electronic interfaced with "magic software" to make this errorless machine that has no real cost associated with it.

In spite of all the software that is out there, this simply hasn't materialized and the reason is simple. Creating a program that can compensate for unrepeating unknowns is not something mankind has been ablt to create yet. Even the most sophisticated adaptive software must "learn" from prior mistakes so that it can't/won't repeat them in the future.

We sort of learned this lesson ourselves when we were doing our Extrak development. In our case, we assumed that the computer knew how to cut circular tool paths. We therefore assumed that the deviation from a true circle was a combination of tuning and/or hysterisis errors in the hardware. We then set out to fix the mechanical stuff as we KNEW how to do this and had the means to do it as well.

Once we got the mechanics fixed, repeatable, we then called for and got help with the servo tuning - the software was never touched and it is/was mid-90's generation CNC servo driving code running in a DOS environment. Once the servo tuning was "fixed" the machine ran in essentially a flawless fashion and in a fashion that was well beyond the theoretical capabilitis of a new machine.

This judgement was passed by a factory technicial who was VERY familiar with the machine, its capabilities and the stuff they did at the factory to get the thing to work as advertised - ours surpassed the best he'd ever seen (it had some high buck fixes and hours upon hours of TLC).

When you try to use semi precision hardware, you'll have a hard time trying to make software correct and compensate for non linear, non repeating hardware. If you don't pay for the precision in the hardware, be prepared to pay for it in the software, especially if the software will have to do some horribly sophisticated error correction in an adaptive fashion.

The way I see it is this: buy the best hardware you can't afford. It will only make the software work to its full potential and you won't have to keep waiting for some new computer/software advancement/plug-in to fix what can easily be fixed today by simply using good, precision, properly fit and assembled parts.

First of all, a lot of the errors are entirely repeatable. They produce the same effect over and over. Leadscrew error is a wonderful example, and you can do leadscrew mapping today in Mach 3.

Imprecision in the alignment of axes relative to one another (i.e. their squareness) is another example. Both are going to be common in these kinds of machines and both can be eliminated relatively simply with software and linear scales.

Second, there are error sources that change so slowly that relatively simple software together with linear scales can once again make the difference.
Thermal problems are not precisely repeatable, but their effect changes so gradually that a control system that includes linear scales can still make a big difference without much sophistication. Machine wear and gib adjustment are two others that fall into this category.

A machine could have a mode to calibrate itself for flexure once the table is loaded, if it had the right sensors. Spring effects due to cutting loads can also be measured and compensated.

Even backlash, that great bugaboo that seems only susceptible to more expensive mechanical components can be compensated for provided the commanded move doesn't require a direction reversal while cutting. Yes, circles are problematic. But, how often are you machining something that is essentially blocky in nature that wouldn't need to reverse directions while cutting? Most of the time if the projects I see are any indication. Backlash in this case can be precisely compensated for. Even where it can't, having everything but the backlash compensated very well and backlash down to a thousandth or two leaves a very productive work envelope.

If the cost to get there is some cheap linear DRO scales and a Mach 3 plug-in, why not? The chief obstacle remains lack of software, not any inherent impossibility in doing these things.

The astronomy world has been going through these revelations for a little while now. It used to be that all they cared about was bigger mirrors and sturdier mountings. However, very clever software has allowed big telescopes to outperform the theoretical limitations of atmospheric viewing by actively compensating for unpredictable non-linear changes. It doesn't get much more unpredictable or non-linear than to try to deal with a fluctuating atmosphere from the ground up to over 100,000 feet! This same active image compensation technology is making its way all the way down to the amateur level and it is quite effective.

Don't underestimate the power of a system that has active intelligence to do things a lot more cheaply than just throwing more iron at it!

Best,

BW

But it's a lot easier to start with good precise iron.

I read this thread rather quickly, but felt I need to throw in my thoughts on this.

The more the better.


The problem is in the rate feedback NOT the position feedback. If you consult control literature in the 90's you find this to be a research issue. It is termed "co-location of rate sensor and actuator".

I believe this would be an all digital control loop where rate info comes from the encoder. Not all systems in discussion here work that way, but the point is taken.


If you study the attached picture, you'll find two masses attached by a spring. Let mass m1 represent the motor, m2 represent the table, x1 and x2 the positions respectively. The spring represents a "sloppy" connection; in this case--backlash.

This I have to disagree with. The spring is springyness in in the mechanics. Backlash is not in anyway this, as it is the accumulation of clearances in the mechanics.

For example some time ago I worked on lathes who's only purpose was to cut very precise curves. One issue we had to deal with is the lead screw untwisting as the diamond cut through center. That is as the velocity of the lead screw changed to a lower value (approached zero) the lead screw would relax. This changed the form of the curve a few microns. A change that was very predictable but different from lathe to lathe. In effect the lead screw is a spring.

Interestingly at this point in the operation the servo was getting feed back from a very high resolution rotary encoder attached to the lead screw.


To stabilize this system, you will need position feedback and rate feedback of mass m2; the table. However, because of the sloppy connection, the rate feedback MUST come from m1; the motor, hence, the term co-location of rate sensor and actuator. If, in your system, the rate feedback in interpolated from the linear scale, the motor (actuator) will hunt position all the time. If you could install a rate sensor on the motor and if your machine allows for separate feedbacks of position and rate, the problem should go away.

You have in a nut shell described the problem of linear encoders on poorly fitted up machines.


I hope I did not confuse anyone. I am not sure I explained this well, but it is late here, so I'm going to sleep.

I should be asleep now my self.


Salah Zenieh
Ph.D. Control engineering

Dave
Far from a Ph.D.

But it's a lot easier to start with good precise iron.

It is an Imperative to have precise iron if you expect precise operation of the machine.

Hi Bob;

I tried responding to your message the other day but Firefox crashed on me. So I try again.

Linear scales are a great idea for the inexpensive machines. They would cure a host of evils.

Actually they don't cure anything fully and add their own problems.


The idea that the feedback loop is really closed by rotary encoders is just not accurate--a relatively limited portion of the feedback loop is closed.

This is true in one sense but in the electrical sense the rotary encoders do close the position loop and may close the velocity loop. What is happening mechanically out at the load is another matter.


Being able to measure the actual position of the axis and just not whether the shaft has turned the desired amount opens the doorway to greatly improved performance because a lot of real phenomena can be measured directly and accounted for:

- Backlash, as has been mentioned.

Backlash is always a problem and may be more so if the linear encoder is used to establish velocity feedback.


- Irregularities in the lead screw, which are always present, but which are worse with less expensive components.

This is where I see the biggest impact. But the cost of linear encoders can be very high relative to buying better lead screws. You still need to shoot for low or no backlash leadscrews.


- Thermal expansion characteristics.

Also significant. Especially on lower cost drives where lead screws might not be as efficient.

- Wear over time or wear that arrives due to retrofitting old machines and using surplus components.

This one I have a bit of trouble with. Looseness is always bad in CNC machinery. leadcrews with glitches and grabs in them cna really upset a control loop.


These are all areas where linear scales could help, not hurt, IF THE PROPER SOFTWARE WERE AVAILABLE.

The need for special software is not really there. But I don't see linear scales solving all the problems you bring up.


It's no accident that I've capitalized the latter, because software is the real issue for this. Mach 3 doesn't really support closed loop operation that is remotely able to take advantage of what linear scales could offer. I'm not aware of any software available at remotely reasonable prices today that would work for this application, and that's the real rub in using linear scales for these applications.

I'm not at all familiar with MACH so I can't comment on that software. What I can say is that control loops have and do run off linear encoders.


It's a real shame too. The sentiment that they are only appropriate if the rest of the machine is equipped with near perfect ways and ballscrews is much more a reflection of software shortcomings than it is a reflection of the real potential linear scales could have.

This I have to disagree with. Your machine has to be up to the demands of the tolerance you want to keep. You can't expect a machine to hold a couple of microns if you have ten microns of slop in the gibbs. Software can not possibly make up for a poorly assembled machine that is free to wander around on its own.

Linear scales do do wonder for thermal issues in leadscrews and also lead errors.


Given how cheaply DRO scales are available these days, if Mach 3 actually had the capacity to really use the devices to their full potential, the accuracies achievable with cheap components would be truly suprising.

You still have to have a machine capable of the accuracies you need. Electronics can not pull mechanical accuracy out of thin air. That being said though you can get good results with linear scales if they are used to eliminate the sources of error that rotary encoders can't.

In any event those linear scale, from cheap DRO readouts, are not really all that accurate for CNC work. They still cost more than a rotary encoder too.


Think I'm wrong? Take a look at the accuracies that have to be achieved with cheap disk drives for computers. They're way in excess of what the cheap Asian tool crowd only dreams about.

Those drives are moving read heads around that weigh almost nothing and are actually flying over the disk surface. Beyond the fact that the controller is highly optimized for the head it really isn't a fair comparison.

So while I'm not willing to give you that, I will be the first to admit that the right electronics could be very cheap if a volume production could be had.


Someday it will happen, whether via Mach 3 or some other vehicle. The fact that Mach now has plug-ins may enable some enterprising group to figure out how to offer a Mach upgrade to do it. I would definitely pay for the feature.

If Mach does encoder feed back now you could hook up a linear encoder to it. It may or may not work depending on the mechanics and the servo loop.


The most likely performance challenge I can see are the response times on the cheap scales are not too great. However, we're not talking about building 200ipm rapids machines here, nor do we need to. A real workable machine capable of 60-100ipm and repeatable accuracy of 0.001" or perhaps 0.0005" which be a real step up.

If you want to resolve to half a thou you will need a fairly good encoder.


The incremental cost for DRO scales and a little software would be very little. Combine it with the upcoming "unstallable stepper" technology and you'd see an amazing new price performance regime established in short order.

WHile I don't know about the unstallable stepper, I do have to agree that the potential for extremely cheap CNC is getting better and better everyday.

Add a little active chatter reduction technology (also cheap to implement save for the required software), fill the bases and columns with epoxy granite, and I think folks would be surprised at what these cheap machines could do.

Now who wants to get started writing the Mach 3 plug-in to support it?

Best,

BW

Dave

Actually they don't cure anything fully and add their own problems.


Not so, Wizard. By your own admission they can cure leadscrew errors and thermal issues. The primary bone of contention seems to be backlash, but you have not addressed my assertion that a system like this can cure backlash too except when changing direction while cutting.

This is true in one sense but in the electrical sense the rotary encoders do close the position loop and may close the velocity loop. What is happening mechanically out at the load is another matter.

That's a pretty convenient way of saying ignorance at the load is bliss! LOL

This is where I see the biggest impact. But the cost of linear encoders can be very high relative to buying better lead screws.

Scales are cheap, even relatively decent ones, when you compare to the cost of NCCams 24TAC47 bearings at $800 per axis and expensive ground ballscrews. What does that add to the cost of a real machine? $3000? Minimum? Now what does a nice DRO system cost that has reasonable resolution?

This one I have a bit of trouble with. Looseness is always bad in CNC machinery. leadcrews with glitches and grabs in them cna really upset a control loop.

Wear is a lot more gradual than this for a long time.

The need for special software is not really there. But I don't see linear scales solving all the problems you bring up.

So far the only one I don't see being solved is backlash when changing directions while cutting. You've not provided a counter argument or example to the other cases.


This I have to disagree with. Your machine has to be up to the demands of the tolerance you want to keep. You can't expect a machine to hold a couple of microns if you have ten microns of slop in the gibbs. Software can not possibly make up for a poorly assembled machine that is free to wander around on its own.

The question boils down to which errors are systematic and which ones are truly random. I have provided two counter examples that are real applications that seem to fly in the face of this wisdom--disk drives and telescopes.

You still have to have a machine capable of the accuracies you need. Electronics can not pull mechanical accuracy out of thin air. That being said though you can get good results with linear scales if they are used to eliminate the sources of error that rotary encoders can't.

Which is all I have proposed!

In any event those linear scale, from cheap DRO readouts, are not really all that accurate for CNC work. They still cost more than a rotary encoder too.

I have not proposed replacing the rotary encoder with a linear scale. In fact, you may even need or want both given the slow response of DRO scales. I have proposed that linear scales will enable lower cost leadscrews, ways, and machine frames to perform a lot better. I have not stated that it will enable them to perform better than much higher quality leadscrews, ways, and frames. When you say they're not really all that accurate for CNC work, you need to quantify that accuracy. How accurate do they need to be for the work you are proposing? I have suggested one could create a lower cost machine accurate to 0.001" for less money using this technology than to do so with higher cost leadscrews and mechanical construction. I'll stick by that argument with the proviso I've already made that we're not talking about eliminating backlash when you reverse direction while cutting.

Those drives are moving read heads around that weigh almost nothing and are actually flying over the disk surface. Beyond the fact that the controller is highly optimized for the head it really isn't a fair comparison.

I don't see anything in what you've said that makes this an unfair comparison. It's an excellent example, in fact. All of the effects we've talked about have to be handled by these drives. To make matters worse, their accelerate/decelerate curves and precision are far in excess of what I've talked about for this application. Backlash, thermal expansion, and leadscrew inaccuracy are all handled by these controls and all of those effects will prevent the head from being accurately positioned over the proper data track at the right time if they're not handled.

There are many more examples where proper sensors and software have made huge differences to various devices. Another is image stabilized optics. Look at the advent of digital music over analog media such as vinyl LP's and tapes. The conventional wisdom from the analog world was bigger, better, more expensive components was all that mattered. Now high quality music is available much more cheaply because you can do things in the digital domain to overcome the need for bigger, better, and more expensive components. All I'm proposing is that the same could happen with CNC.

The biggest issue is that there are precious few people even working on these sorts of problems for a variety of cultural reasons. Namely, computer guys are not going to be very high up the pecking order at most machine design firms which are run by the MechE crowd. If they're even remotely involved, they're going to be off at a CAD/CAM company or working on the sexy 5-axis stuff. Or, they're off making the big bucks designing disk drives and such.

It's going to take crossover thinking from someone who understands both the mechanical problems and how to apply the digital domain to them. Someone will crack the nut, though. And market forces drive costs inexorably down. Whoever does figure it out, and gets the patents, will make a tidy sum and a large segment will move in this direction. Interestingly, the last interview I read with Gene Haas indicated his feeling that the market was all about making the machines cheaper because that's what Asia wants to buy.

BTW, we've already seen a ton of mechanical problems overcome by CNC. In 1960, we could as easily have been arguing over whether digital electronics would make taper attachments on lathes obsolete.

Cheers!

BW

BW and Wizard,

It is great to read your posts...anywhere on this portal; I especially like the current discussion.

My first post attempted to point out what I believe is an inherent problem in the modeling of the "problem". The "spring" connection between two masses is meant to "model" a problematic connection, be it, a thermal problem, backlash (hysteresis), compliance (springiness) in the screw...etc. What one needs to see is that the system is being stabilized ASSUMING the connection is "rigid". Rigid here means more than lack of springiness--it includes backlash. Since it is not, the degrees of freedom in this dynamical system is 2 not 1. From a control engineering point of view, one must assign the motor shaft a coordinate separate from the table. So we have two positions, theta, an angle which could be measured by a rotary encoder attached to the motor, and x-a position of the table. To get this system to do what you want regardless of the "rigidness", "sloppiness", of the connection, one needs to feedback both the table position and rate, and the motor angle and rate. What I am saying is the control loop must be re-engineered to accept both measurements and the system stabilized based on a full state feedback. I am pretty sure you can do away with the rate feedback from the table. You should be able to successfully get zero errors at the table position from feeding back table position, motor encoder angle, and motor rate.

I know that the current hardware is not setup to do that. But that's the way to do it. Bob's software suggestions will have to be written to implement this control loop not finds ways to fix the current inherent problem in the engineering of the system.

What do you guys think?

Salah Zenieh

Yes, you can "map" the system and once you do that, theoretically the software should be able to take care of "repeatable" errors. HOWEVER, from my prior experience with machine tool bearings, the errors can be both cumulative and definitely NOT repeatable.

EXAMPLE: there is a BIG difference between REPETITIVE and NON REPETITIVE runout. This was learned when the first HDD's were being created in the late 60's and early 70's. At the time, even ABEC 7 grade bearings were not adequate as you could put data on a HDD but it wouldn't be there the next time (revolution) when you went to look for it.

Over time, they learned that the combined runouts from raceways, ball diameters and a bunch of other variables caused the inner ring of a bearing to "orbit" as opposed to rotating in a repeatable fashion about an axis.
Without careful parts selection and Q/C, it is determined that the orbit didn't repeat until at least 15 to 17 revolutions.

The actual value was a function of raceway diameter, ball pitch diameter, ball count, ball size and runout deviations and ball count. This created a whole new technology for HDD bearings as ball bearings, as accurate as they were, they were NOT adequate for the REPETITION needed to store and recover data from the magnetized disk.

Now, lets say you only move the ball screw 7 revs and, in doing so, you unload the preload in the ball screw bearings. The error that you may have mapped is now going to be different and your software won't be any the wiser. Hence, non repetitive error.

At this point, i'd be inclined to say that almost ANY software can create the machine code to cut a round circle. It can also account for REPEATABLE errors. However, as we learned in the development of our iron (good high dollar stuff, probably well above the quality level of most DIY'ers), we learned that the software was being limited by the iron.

Since we weren't going to get ANY help from Bridgeport or EMI with regard to DOS based and compiled BMDC software development (analogous to the missing plug ins for MACH), we turned to the only solution we had and was viable - namely, fixing the iron.

Somewhere, someday today's vaporware will be developed to take low cost, imprecise pieces and integrate them into an adaptive system that can/will do wonders in machine control. But, as long as the code writers are having to continually adapt to ever changing O/S's (Win 9x to Xp then to Vista then ???) and M/S doesn't simply offer a STABLE (as in non-changing) machine control focused instead of aftermarket adaptationg of non-machine tool control based O/S's, I don't see where plug-ins will appear anytime soon.

Would it be nice to have system that had motor encoders AND net table feedback to totally close the speed, velocity and position loop for CNC control? DEFINITELY. Yet is anybody doing or has anybody done that? I dunno but I think not. They can get adequate performance from traditional servo's so why bother????

Until that happens and when It becomes affordable, I'll look forward to buying it. Until then, sadly, I must rely on $800 ball screw bearings, preloaded precision ball screws and well tuned and/or tricked up hardware to provide the machine tool controls that I need for my use.

NOTE: some of the high buck stuff we used/did to take our mill and turn it into a psuedo jig mill was done out of need. We also did it because we were told we couldn't do it and yet, by goodness, we did. Even so, we use our neighbor's Haas to do the master milling because it does it faster and easier and just as accurately than setting up the Bridgeport while running essentially the same G code.

BW is correct, it IS doable. BUT at this time, I simply don't see the aftermarket or the OEM's putting forth the effort into doing it. The machine tool market, and especially the DIY aftermarket, simply doesn't have the collective desire or impetus or market demand to support the effort.

When/if the academia who are gazing into the heavens with their tricked up telescopes take to doing machine tool controls, it will happen but not until. chances are the guys with the smarts to do it already work for Haas or other machine tool control housed and they can't/won't offer up their expertise to the hobbyist for what should be obviuo$ rea$on$.

I believe the DIY CNC crowd out there are closer to solving this problem than they may think. The effort must come from the Linux based EMC developers. They develop the code of a very powerful CNC system that can handle closed loop servo architecture. The reason they can fix this is that all control loops are open source in that firmware, so, what is needed is a decent group of control engineers re-writing the tracking algorithm to be based on full state feedback using a linear encoder feedback from the table plus a rotary encoder feedback on the motor. Sure, an added cost in electronic hardware is to be expected, but it should be far less costly than buying premium quality mechanical hardware.

Salah Zenieh

I re-read some of the earlier posts on this thread and felt the need to post again.

The title "Why feedback?" is the first statement I heard in the first control engineering class post-introduction to the general concepts of control engineering.

Feedback is costly, and complicates system design greatly. So, why feedback?

If one lived in a perfect world, where leadscrew errors can be exactly mapped regardless of thermal expansion variations, where backlash either did not exist or is known the last micron, where machining forces were exactly known at every microsecond of machining time, one would never need feedback. All one needs to do, is to apply the right voltage at the right microsecond to develope the correct current profile that will develop the absolute correct torque profile that will then drive the table with the exact amount of force vs. time profile to get the perfect job done. This, off cource, is a dream world.

All gain scheduling algorithms, leadscrew mappings, backlash compensation routines remind me of my first computer programs. A nightmare of nested if-then-else statements that with countless do-loops that promise to be the mother of all computer codes that solves the problem at hand.

So, why feedback? It provides system robustness in the face of uncertainty, hence, compensating for unknown errors. You do not need to map the lead screw errors or measure backlash and then compensate for it. The control loop, when properly engineered will do it for you. It will also reject disturbances due to unknown machining forces. This is the reason we feedback and it is not done properly in the CNC world.

Salah Zenieh

The human body is the epitome of a closed loop system with well developed feedback. The brain sees what's going on via sight, sound, touch and feel and compensates in real time for the desired effect. In spite of many efforts to try to duplicate this with artificial intelligence, often at a cost is no object fashion by/for the government, it still hasn't been done to the desired levels of satisfaction or even at all.

The theory and technology does, to a very large extent, exist to do the feedback needed to affect good/decent machine control. We have the glass scale encoders that attach to tables and the digital encoders that attach to the motors already. We have the ability to make slop free drive systems and robust electronic amps.

We have the resultant micron level resolution. Obviously, the encoders and feed back systems are NOT the problem -they exist already and are SITTING there awaiting implementation. WE have the technology to do closed loop servo driving. The technology was developed long ago.

Bridgeport did it in DOS in the early/mid 90's and, by super servo tuning and adding superb iron tuning, we can get near jig mill performance out of a DOS based BMDC controller using a mill that is several steps better than a ultra superb drill press and darn close to a TM1 at a fraction of the cost. Truly a case of putting lipstick on a pig.

If the technology exists to read both encoder and micron level glass scales and drive servos at 0.0001" levels of perfection (and less than that on high buck CNC grinders that we're aware of), why hasn't it been done yet/alread?

I'd contend that the mind set of the industry is such that the existing control algorythms (sic) are of the "why both? What we have is good enough" level. The impetus is simply not there to "redundantly" drive the servo with the typical encoder system and then close the position loop with TRUE table position feedback.

I'd venture a reasonable estimate would be to take the cost of an existing servo system and simply add the cost of glass encoders, wiring and some interconnection BOB's. The key then becomes the amount of input channels you'd need to input the data in concert with the processing speed needed to process and number crunch the data.

As is pretty much always the case, we now have bottleneck issues with the processor and the software. It would be wonderful if EMC or ???? were to become the defacto standard. But, without economic incentives, who can afford to develop the code? This whole issue of altuistic versus capitalic attitudes was discussed in a separate open source thread some time ago.

As i recall, the decision was that the guys who can do it already do so as a livelihood and they don't make a point of giving their livelihood away. Since the software guys often can and do make a good living from their intellectual property rights, they tend to sell the stuff as oppose to making it open source and thus giving it away. I can't argue with that in the slightest.

Yes, there are open source projects and they can do well. However they can also become disjointed and lose focus as each developer does their own thing with the code. About the time the thing gets going, it either loses focus, the developers move onto something else or M/S introduces a new O/S and people jump on a new band wagon - after all, they do have to make a living.

Heck, even Wozniac is set for life and he's not giving away any of his old code. For that matter, none of the true M/S DOS code has gone open source even though M/S doesn't support or even sell the darn stuff anymore.

There are a number of paths to go down given the cross roads we're at now.

Good iron and electronic hardware and adequate software or (the ultimate goal being discussed herein) wherein generic harware is given manners via table mounted scale feedback in concert with motor based encoder feedback managed by vaporware software.

I already KNOW that 10+ year old DOS based code can and will run well built and tuned iron - been there and done that. I know that some darn good servo based CNC controls could and have been made from legacy, essentially FOC (free of charge) DOS O/S in concert with "surplus" hardware that ciould EASILY be reconstituted via some relatively easy to do reverse engineering.

Sadly, however, PROVEN code that already exists that would run the stuff is NOT documented although it is compiled and has been well circulated in circles that use the software. Even though the "system" that runs this code is beyond obolete, it is still QUITE adequate for most DIY use - we're milling cam masters with mid 1990's software and you can't tell the diffence between what our masters grind and CNC ground ones off of a $1.5million machine.

Yet, the chances of the source code becoming available for open source development, redevelopment or adaptation of this DOS code to something that can/will ruin on EMC or XP or whatever are slim and none. Yet, even today, equipment baed on this hard and softare would be a DIY'ers dream as you could run a mill, lathe or 4 axis VMC with the exact same interface boards.

Having grown up in the automotive aftermarket doing technical marketing and engineering, I know first hand that the techniques used there (engineering, reverse engineering, aftermarket servicing of OEM applications) can and will work in the DIY and some of the small machine shop machine tool industries.

Perhaps, someday, the guys who have the skills to do the systems integration for CNC will learn that there is a lot of money to be made by exploiting the market that is out there as opposed to trying to invent an all new adaptation of a more sophisticated wheel for computer systems that are intentionally in a constant state of redevelopment.

Wow!

Bob, As you know I'm a software and electronics guy first, and a machinist second. I don't really agree with your idea that you can compensate for a weak machine with feedback and code. Particularly if the feedback ONLY comes from the table movement (glass slides, motor encoders or whatever)

When I jump up and down here in Atlanta, realllllllly good software and equipment might allow you to measure the movement in California. Compensating for it is another thing completely. Especially if the sensors are here and not in your backyard.

As you know, the microsecond your tool hits the workpiece, EVERY part in the mechanical-support-foodchain is compressed or tensioned.

The tool bends, the spindle bearings get loaded to one side, the z is pushed upon, the frame / head gets a push, the leadscrew gets it push or pull, the servo gets it's it end load, the leadscrew gets it's spring winding and the part moves relative with respect to where you wanted to be cutting.

Figure out how to sense the point of contact between tool and workpiece and you MIGHT have chance of creating a system that compensates, but you are still COMPENSATING.

Once you've compensated for a lack of rigidity and precision you will still have to pay for not having precision to start with. If there is play, then those parts will wear faster, If the machine is just lacking iron you will pay for that strain in the long run. (notice that I have not even mentioned resonance or vibration)

I envision some Chinese piece of crap with a complex compension system that initially works well, but wrecks itself in short order.

best regards,

Barry

.

If the technology exists to read both encoder and micron level glass scales and drive servos at 0.0001" levels of perfection (and less than that on high buck CNC grinders that we're aware of), why hasn't it been done yet/alread?

The answer is simple: it has been done already!

Heidenhain's white paper provides a useful introduction to some of what is available off the shelf today:

http://www.heidenhain.com/wcmsmimefiles/349_843_20_15180.pdf

In this paper, they talk about compensating for a great deal of what I've proposed here:

- Backlash

- Wear

- Thermal Effects

- Machine structure rigidity

They do point out a number of limitations of the approach, such as Barry's issue about how cutting force stresses may not be entirely measurable with nothing but linear scales. At the same time, they show the scales help there as well. In the main, their view of the applicability of linear scales on modern vmc's is for thermal effects. Such machines are already using very high quality ballscrews, bearings, and frame components, so they are not really even trying to address the same issues I am for cheap machines.

My contention has NOT been that one can achieve 0.0001" accuracy on a cheap Chinese mill using this sort of technology. Rather, it has been that repeatable 0.001" accuracy ought to be straightforward if the classic servos and ballscrews approach is applied with the addition of linear scales and proper software compensation.

Why do I think so? Perhaps more important to understanding the idea I'm trying to get across is to understand why I differ from what many of you have said. As I've already said, it isn't that I disagree that expensive components and beefier frames will make a better machine. It is simply that I believe there are enough systematic and reproducible errors in the Asian mills that we can reliably achieve 0.001" accuracies before we ever wind up dealing with most of the chaotic non-linear issues you're concerned with.

I think the Chinese machines suffer most from poor fit and quality control. Axes are not square to one another. Machines are assembled without precision squaring after enduring a rough shipping process. Tolerances are poor on leadscrews. Lathes are not leveled, or have twists in the beds. These are all classes of error that swamp what's being discussed here and that are easily correctable with true closed loop linear scales in software.

Why do I think this is largely the issue? Because I've read too many cases where people were able to tune up this "junk" to achieve decent results on the order of what I'm talking about. The same happens with worn out Olde American Iron as well. Read for example this fellow's exploits:

http://rick.sparber.org/Articles/sb2/sb2.htm

I have no trouble believing his claim having experimented with Z-axis accuracy on my own Chinese mill, and said accuracy augmented with a DRO scale. He's achieving high accuracy at the lowest levels on the food chain with a small round column mill. The reason he can do this is that the biggest terms in his error budget are not incurables such as unpredicatable flexure in the machine structure or weird not repeating effects in the spindle bearings. Those effects are there, but they are buried under much grosser phenomena that are readily addressable.

I've also seen the precision that can be achieved by something like the Widgitmaster's CNC Mini-Router. He's assembled those machines on a fairly mass produced basis using Asian machine tooling even to turn the leadscrews. There are no $800 ballscrew bearings nor ballscrews at all. He's using a Chinese Birmingham lathe to cut the leadscrews on these machines which no end of Internet personalities have labeled junk unsuitable for any level of precision at all. The Widgitmaster knows better. He's a machinist. He knows how to work around whatever problems there may be to get the job done. That's all we're really asking the linear scales to do, and they are up to the task if the software is written for it.

Barry's example about jumping up and down in Atlanta is apt. We don't really care about jumping up and down in Atlanta with these machines because we're sitting in a moving car and there are worse vibration issues much closer to us than Atlanta.

These machines need not "wreck themselves" either, as Barry has suggested. I am not proposing we run them at 100's of IPM travels or suddenly start bolting on huge cutters they are wholly unsuited for. I'm only calling for an increase in precision, and there is nothing about the way the machine will operate with that increase that will increase wear at all. To what extent there is wear, a machine designed in this way will be much better able to self-tune those effects away. There is an example of this in the Heidenhain white paper as well.

Lastly, I'll be the first to admit this is all rather academic and that in the end we may all wind up just agreeing to disagree until someone does it or doesn't. This would not be the first time I've been told I was wrong about something, or even the first time I really was wrong! LOL

I make my living largely by doing things others have deemed technically impossible with computers. For example, I made quite a tidy sum selling software that uses genetic algorithms to automatically generate QA test scripts for software. While I was raising the venture capital for this one of the world's foremost experts on genetic algorithms said he didn't think it could be done. Needless to say, the VC's who brought him in didn't fund me, but some others did, and 10 months later the software was working and the company was sold.

Cheers!

BW

"The point of philosophy is to start with something so simple as to seem not worth stating, and to end with something so paradoxical that no one will believe it." Bertrand Russell (from wiki philosophy)

Bob,

I too have achieved striking results with my crappy Enco. I've also pushed it to the point where my part was destroyed.

I agree that you could "easily" create a dual feedback system that would allow you to achieve excellent results with mediocre equipment at the expense of time.

The other problems that we all acknowledge do need to be addressed and the lowest common denominator worked at until it is no longer a factor.

If your lead screw has slop and the mill pulls your part in, then you need to fix that problem. The same goes for all of the other gotchas.

You also need patience, skill, and experience to achieve a widgitmaster level of excellence.

"good, fast, cheap - pick any two"
Best regards,

Barry

"good, fast, cheap - pick any two"
Best regards,

Barry


Accurate and cheap. Fast will have to be sacrificed. Accurate and Fast are the domain of commercial vmc's. Fast and cheap without accuracy seem not to be very useful.

Best,

BW

Look at the block diagram of a Gecko to get the basics of how the servo loop works.

Understand that the motor (position from the encoder is stored in a counter chip) is never very far away from the commanded position (also stored in a counter).

As the position gets off due to commanded position, inertia or load, the sum from the subtractor grows; that sum is converted to a voltage which goes to the PID filter and power amp which pours the power to the motor to get the two counters back to the equal position. If either counter overflows, an error is generated and the servo faults.

Imagine another quadrature decoder that connects to your glass slide. That decoder goes to another up/down counter (just like the one already on the Gecko)

Hack the signal from the exsisting shaft encoder counter.

Insert another adder chip. This adder adds the two errors. (if you like you can adjust the scale of the inputs to adjust the significance of each encoder - since the slide will be coarser you probably want to give it a multiple by shifting its bits to effect a multiply)

The output of the adder would go to the existing subtractor. You would lower the overall gain to account for the extra error signal.

The motor would run until both counts equaled.

You could also use seperate D/A converters and do the add in analog. That way you could also add an extra PID filter for the slide (but I don't think that you would have to...)

Barry

Barry, excellent! Skip the software altogether so as to avoid waiting on it.

It would be an interesting experiment to try, to be sure. It sounds like it could be implemented on a little PC board upstream of the Gecko drive.

I am fond of saying that software development goes through cycles:

Stage One: It is impossible! We seem to have made it through that stage, or at least there is light at the end of the tunnel.

Stage Two: It is easy! This is where we need to be very careful not to get overly confident, but it is also the fun stage of the cycle. Some experiments and demos on this definitely ought to be conducted. One gets the demo made and what we learn from it brings us screechingly into the realities of the last stage.

Stage Three: It can be done, but it will be late and over budget! We now see all the things we didn't consider rearing their ugly heads in the demo. We are confident it can be done, but we no longer see how to do it on schedule.

Perhaps your electronics projects follow a similar pattern!

From a software standpoint, my thought had been to try to piggyback on the existing Mach 3 leadscrew mapping feature. To do so, we would add a feedback loop that updates that map continuously based on readings from the linear scale. It is admittedly a crude and inelegant hack, but I was thinking along these lines for several reasons:

- I wanted to avoid disturbing Mach 3's architecture too much. Piggybacking on existing functionality seemed to fit that bill.

- I was concerned that the response rates of the linear scales might be pretty slow. In fact, perhaps to slow to allow updates during rapids, but perhaps good enough to work during a slow cut.

I had also thought to provide a "fault" mode whereby if the linear scales disagreed too much with the commanded position at the end of each line of g-code, the system would fault out and halt.

It had also occured to me that a "calibration pass" might make sense. This would effectively be a "learning mode" where the first part was cut and the system would compare actual position as measured by scales with commanded position. After the calibration pass a new leadscrew map would be generated that was appropriate to that particular machining operation so that subsequent runs would be more accurate.

Another thought is that Mach 3 already deals with a pretty "springy" interface to the hardware when the GRex is being used. Careful study of the GRex plug-in might suggest a more optimal way to use linear scales. The G-Rex also provides the interface to read the scales.

It may turn out that a hybrid that does some things slowly in software along the lines I mention and some things in real time using simple hardware works best.

Best,

BW

We learned some lessons during the tuning of our $800 bearing equipped Bridgeport (we didn't pay that due to "connections" but that's what they'd cost to replace them). These were recognized as being too costly soe we came up with some relatively cheap A/C"s in other posts - however, even so, "cheap" and "A/C" are oxymoronic when used in the same sentence.

Interestingly, the Bridgeport lesson came AFTER we did some glass scale/DRO equipped tuning of our Excello manual mill. We used it to make our indexing fixtures for cams as well as our drive dog places wherein we built in the lobe center offset. This mill was used because it had DIRECT table feedback and we needed/wanted that for "jig accuracies" for the drive dogs we were making. Result: the stuff wasn't accurate even though the DRO said we were RIGHT ON to the tenth.

As part of the learing process, we went back and forth between the Excello and the BPT - both highly reputed machines and in relatively good (Excello) and pristine (BPT) condition. Yet, accuracies sucked.

NOT being computer programming savvy nor being a machine tool builder by trade but having machine tool bearing experience, we went about eliminiting the readily visible or measureable sources of hysterisis/slop.

We learned first hand that slop is readily and easily remedied by proper adjustment of the machines' gibbs and ways - by doing this alone ( BTW the Excello has acmes and stock bearings otherwise), we could readily hold tenths in position accuracy during subsequent drilling/boring/reaming operations.

Thus, by doing simply machine tuning and slop elimination (well within the capability of the limited budget DIY'er), we got a machine that EASILY could meed the 0.001" target accuracy of the discussion. All this in a machine that I have less than $3000 invested into (including a full motor rebuild with armature balance for good measure).

Once we learned what to look for and how to do it, we then went ballistic on the BPT and threw high buck bearing technology at it willy nilly.

In both cases, however, residual slop in the systems allowed the cutters to either push off the material and shove the part away. Or, it grabbed hold of the material and cut it more deeper (sic) than we wanted/needed. That's when we got serious about the bearings and ball screws in the BPT.

After measuring the TRUE compliance under load (two guys at 200lbs plus each) by pushing and shoving on the table did we find that the "precision" BPT wasn't as precision as we thought. By doing this we both quantified and located the source of the slop.

It WAS to spec but not adequate for what we wanted to do. Figuring that the CNC controller worked well, we attacked the slop hysterisis factor in the fashion we could - preloaded bearings and other known machine tool/high end machine tool tricks.

Asside from the high buck ball screw bearings and low buck ball screw nut preload reload, the BPT is TOTALLY stock as is the software. Once we got rid of the slop, the mid 90's software ran amazingly fine. We'd hoped to be able to mill cam masters but the REAL/ACTUAL intent for doing all the tricks to the CNC side of the machine was to mill golf club putters.

The client admitted that the kluged up A/C's we had in it and shim preloaded were MORE than adequate for putter quality machining. The high $ bearings were added to get into 0.0001" accuracies necessary for our cam masters.

The way I see it is this: the human mind can readily and adaptively compensate for the wear (slop) in the front end of a car as it wanders due to worn ball joints and sloppy tie rod ends. HOwever, if you fix the worn components, the computer (driver) can spend time doing other things than correcting for stuff that really needs to be or should be fixed anyway.

I'd contend that the technology exists to create and make deadly accurate machine control systems. Although adaptive software is great if you have the where with all to create and understand and adapt it - sadly I don't. But I, like many others so mechanically inclined can deal with and eliminate mechanical slop creaed by poorly chosen and/or poorly assembled parts.

I'd contend, however that in light of the hardware and knowledge availability to the community, some computer solutions involve complexities that are beyond the skill and/or education levels that some of us have.

The neat part about this hobby/business is that there are many paths to the same water hole.

I read the Heidenhain white paper mentioned by BW. It is clear that this is the correct solution to this problem; feedback loops from the linear scale for position and rate feedback from the motor encoder. Position feedback is not needed from the motor; the motor shaft angle may be wherever it wants to be as long as the table is where it is supposed to be. Rate sensor and actuator MUST be co-located. This is also clear from the Heidenhain implementation.

The "piggyback" approach from the software guys is an implementation issue. It may work. But it wont be the proper way to do it. The loop should be closed properly once and for all and the software must implement the properly engineered control law. The software solution proposed is adhoc. Or as BW mentioned "not elegant". The "feedback" loop suggested by BW is not really feedback in the true sense of the word. It sounds like an attempt to update lead screw data on the fly and "learn" from errors during machining; this is a dangerous burden on an already loaded real-time code trying to stabilize motors and track pre-defined tranjectories. It should be done properly as NCCam mentioned; extra input channels to the PC and the processing of this data as part of the real-time code-proper; the CNC core software.

Great discussion! Problems are more than software for the DIY market. They are also hardware for our fantasy system we need very accurate linear scales. There are also interface and data flow considerations. As well as issues such as affordabily.
US digital puts out a good interface though it is about $600 with connectors etc.
It is a combined hardware and software solution!

Not so, Wizard. By your own admission they can cure leadscrew errors and thermal issues. The primary bone of contention seems to be backlash, but you have not addressed my assertion that a system like this can cure backlash too except when changing direction while cutting.

Linear scales can certainly help with classes of lead screw errors. One of those is thermal expansion. But they can not compensate for all thermal expansion in a machine tool.


That's a pretty convenient way of saying ignorance at the load is bliss! LOL

In a sense we are always slightly ignorant of where to tool is with respect to the work piece.


Scales are cheap, even relatively decent ones, when you compare to the cost of NCCams 24TAC47 bearings at $800 per axis and expensive ground ballscrews. What does that add to the cost of a real machine? $3000? Minimum? Now what does a nice DRO system cost that has reasonable resolution?

Yes but the DRO is just part of a CNC solution. Even then the scales will not compensate for all the issues that could impact the relationship of the tool to the work piece.



Wear is a lot more gradual than this for a long time.



So far the only one I don't see being solved is backlash when changing directions while cutting. You've not provided a counter argument or example to the other cases.

Nope backlash wont be solved. As to the other issues, the ability to machine to superior tolerances involves being able to maintain the position of the axises under load. This requires superior fit up of ways and gibbs.




The question boils down to which errors are systematic and which ones are truly random. I have provided two counter examples that are real applications that seem to fly in the face of this wisdom--disk drives and telescopes.

Well I still think disk drives are a poor choice due to the technology used there.

As for telescopes the usage there is similar to lead screw mapping which can be done today. The problem with lead screw mapping is that we need to know that the axises won't wobble around in their ways for it to be reliable.



Which is all I have proposed!

There is nothing wrong with some for the stuff that you have proposed, such as lead screw mapping. What I'm trying to get at is that the technology does nothing for you if the rest of the machine is not up to the task.

A little bit of work fitting the machines up mechanically can pay off by allowing you to realize the specs you want on the finished product. I believe that one of NCCams posts highlight the issue. That is the error in a machine may not even register on a linear scale.



I have not proposed replacing the rotary encoder with a linear scale. In fact, you may even need or want both given the slow response of DRO scales. I have proposed that linear scales will enable lower cost leadscrews, ways, and machine frames to perform a lot better.

I guess that it depends on what you mean by perform better. If you want nice feel good numbers showing up on a display then yeah it may appear to perform better.

That is not to say that the machines we are talking about can't be made to perform better, just that you first need to address the machine mechanically.

I have not stated that it will enable them to perform better than much higher quality leadscrews, ways, and frames. When you say they're not really all that accurate for CNC work, you need to quantify that accuracy. How accurate do they need to be for the work you are proposing? I have suggested one could create a lower cost machine accurate to 0.001" for less money using this technology than to do so with higher cost leadscrews and mechanical construction. I'll stick by that argument with the proviso I've already made that we're not talking about eliminating backlash when you reverse direction while cutting.

Admittedly it is one of those issues that does require definitions but the reality is that the control system can't compensate for everything. That no matter what the accuracy you want to achieve. My point remains that in order to benefit from the sorts of systems you are talking about the mechanics have to be right. That doesn't mean that the cheap Chiniese mills are out of the running just that a rebuild may be in order.


I don't see anything in what you've said that makes this an unfair comparison. It's an excellent example, in fact. All of the effects we've talked about have to be handled by these drives. To make matters worse, their accelerate/decelerate curves and precision are far in excess of what I've talked about for this application. Backlash, thermal expansion, and leadscrew inaccuracy are all handled by these controls and all of those effects will prevent the head from being accurately positioned over the proper data track at the right time if they're not handled.

The big problem with the disk drive example is the servo system just isn't comparable. First there are no variable reaction loads to deal with. Second in some of these disk drives servo information was printed right on the disk.


There are many more examples where proper sensors and software have made huge differences to various devices. Another is image stabilized optics. Look at the advent of digital music over analog media such as vinyl LP's and tapes. The conventional wisdom from the analog world was bigger, better, more expensive components was all that mattered. Now high quality music is available much more cheaply because you can do things in the digital domain to overcome the need for bigger, better, and more expensive components. All I'm proposing is that the same could happen with CNC.

From what I can see it is happening right now. That is where it is feasible.


The biggest issue is that there are precious few people even working on these sorts of problems for a variety of cultural reasons. Namely, computer guys are not going to be very high up the pecking order at most machine design firms which are run by the MechE crowd. If they're even remotely involved, they're going to be off at a CAD/CAM company or working on the sexy 5-axis stuff. Or, they're off making the big bucks designing disk drives and such.

The last thing we want is these guys doing proprietary work someplace. In any even I think you are over estimating just what can be compensated for in the physical world by digital control.


It's going to take crossover thinking from someone who understands both the mechanical problems and how to apply the digital domain to them. Someone will crack the nut, though. And market forces drive costs inexorably down. Whoever does figure it out, and gets the patents, will make a tidy sum and a large segment will move in this direction. Interestingly, the last interview I read with Gene Haas indicated his feeling that the market was all about making the machines cheaper because that's what Asia wants to buy.

I'm not convinced that digital will make machines cheaper. Having worked on some very precise machines that have implemented some forms of compensations I still maintain that you have to know what your problem is before you can compensate. Or maybe better stated you have to be able to measure the problem before you can compensate.


BTW, we've already seen a ton of mechanical problems overcome by CNC. In 1960, we could as easily have been arguing over whether digital electronics would make taper attachments on lathes obsolete.

Yes I understand the digital has done wonders. In many cases obsoleting whole store rooms full of attachments, fixtures and stuff.

Cheers!

BW

As a side note a few years ago I had the good fortune to spend part of my career working on diamond turning hardware in the optics industry. This was high volume work to tolerances of a couple of microns on radius's, with optical quality finishes. Invariably when ever the lathe started to have problems the first thing to be looked at was the ways and the machines alignment. Even if tuning was determened to be an issue, the mechanics where checked before the amps where tuned. It was determined the hard way that ignoring the mechanics was a poor course of action. This on a state of the art control system.

In fact the mechanics are so important that we ended up sending out the ways to be hand scraped by a local well known machine tool builder. In the case of these machines, the way where of high quality to begin with. But lowering the friction and stiffing up the assembly via hand fitting went a long ways to improving the machines.

What gets you is the slop factor in the ways and saddles. No matter how good your controls are they can only respond to those machine imperfections they can measure. Or in the case of lead screw mapping something that was measured. We would try to maintain a couple of microns as the maximal amount of slop in the ways. Not easy to do on production machine.

Dave

This discussion gets much simpler then, as I'll avoid poking at areas that still remain pretty vague among the naysayers and stick to what isn't vague at all.

I'm not convinced that digital will make machines cheaper.
...
Or maybe better stated you have to be able to measure the problem before you can compensate.

I've never argued anything different--linear scales are all about measuring problems that were previously unmeasurable with rotary encoders or worse open loop stepper systems. In fact, it may be that for cheap Asian machine tools, it is even more important to have the scales than rotary encoders precisely because of the nature of their shortcomings.

My contention is that a lot of the more egregious problems with Asian machine tools are entirely measurable and straightforward to correct with linear scales and software. I will readily agree that if you can't measure it, you can't correct it and have never said all the errors will be corrected. I have simply said that there are enough correctable errors available in the error budget to bring things to 0.001" repeatable accuracy if we had the added benefit of linear scales and that this may well be the cheapest way to get there.

I've quantified in pretty fair detail what I think the addressable areas are as well as providing 3rd party information on how their products are correcting these areas.

The remaining question, it seems to me, is not how diamond turning machines work. I'm not proposing remotely this level accuracy and I have real doubt about whether the construction, performance envelope, design goals, and detailed breakdown of the error budget of those machines has anything to offer us in terms of understanding how to improve cheap Asian mills. We can't turn an Asian mill into a diamond turning machine, so why try?

Looking at the world of diamond turning machines and commercial VMC's that want much higher performance than I have proposed, isn't it also possible that they need a radically different strategy than I am suggesting precisely because their error budgets are made up quite differently? There are reasons why the Concorde looks so different than a Cessna 182 and why it doesn't make sense to try to turn a Cessna 182 into a Concorde, but this does not mean the 182 couldn't be improved nor does it mean that the best way to improve a 182 is equally what you'd try first on a Concorde. These arguments about how high end machinery works need to be put aside and the root issues with the cheap machinery must be analyzed directly before progress will be made.

The real question, then, should be whether there is ample reason to believe the error budget can be constructed so that by eliminating enough measurable correctable errors a good result can be had despite there remaining some errors that were not measurable/correctable.

As I've already stated, I have seen too many cases where people are able to achieve the desired level of repeatable accuracy to conclude that we just have to throw up our hands because, OMG, the ways and gibs are so sloppy the machine could move 6 feet in any direction at all and we'd never know it!

We know there isn't that much slop because of the repeatable accuracies that can be achieved through manual machining efforts and because of the types of cutters and operations we can successfully run on these machines. If the slop in the ways, gibs, and leadscrews were as bad as all that, people wouldn't be running face mills on these machines (which they are) and could scarce get a 1/2" end mill to cut right. In fact, if you're willing to live with a little less extreme depth of cut which is dictated by the beef in the frame, these machines perform pretty nicely.

You have to accept that for the most part, these machines do not have that much random slop with even a little gib adjustment and no expensive bearings and way rescraping whatsoever. I know for an absolute fact that while the quill on my IH mill is junk for precision (and this has been written about many times in these forums) and repeats only around 0.008", having put a DRO scale on it, I can get 0.001" repeatable accuracy from it all day long.

Given as many examples as one reads and experiences, it is not so hard to see that scales could make a positive difference here, and likely are a better price/performance return than completely rebuilding the machine with radically more expensive components.


Cheers,

BW

hello,
Given as many examples as one reads and experiences, it is not so hard to see that scales could make a positive difference here, and likely are a better price/performance return than completely rebuilding the machine with radically more expensive components.

I agree 100%
the body of a man is not high precision and the musles tendon +bones flexibility are far to be high precision it's all feedback that make it working also we can look at the rigidity of laser disc reader at the begining they where mechanical piece of art now they are pretty loosy meanwhile the positionning of a hdvd is much better than a cd player of 1980 the difference is feedback
the combination of faster chips and high resolution scale+rotary encoder can perform a perfect close loop who can compensate many default bu the way all the diamond turning machines use also this type of device so we can encourage initiative like 6 axes closed loop of cncbrain
http://www.safeguardrobotics.com/default.aspx?tab=cncbrain
if you do not have brain you must have legs
if you do not have expensive mechanic you must have nrain

There is certainly nothing new about the idea of using linear scales for the feedback device in a cnc. Early machines of the 60's used linear inductosyn scales for feedback.

There are many problems with this technique notably the phase shift introduced into the control loop and the fact that this problem has been proven to to be statistically unsolvable and is very nonlinear.

Under the right circumstances this may help, many times the results will be worse than simply using encoders on the motors. Simple point to point moves are one thing, adapting to variations while cutting is a different story altogether.

This is like the Holy Grail of motion control and on the surface appears very simple and logical. The best control engineers in the world have worked on this problem for decades but still machine builders rely on accurate and stiff assemblies. Do you really think these guys don't know what they are doing?
Bob

This is like the Holy Grail of motion control and on the surface appears very simple and logical. The best control engineers in the world have worked on this problem for decades but still machine builders rely on accurate and stiff assemblies. Do you really think these guys don't know what they are doing?
Bob


but they have solve the problem the problem is the price they sale it
know a little story
why the car always misse the dove (if the dove is healthy)
because the dove analyse the vision 10 time faster than us
now let's take an encoder 1000 lines for a screw of 4mm/turn
and a linear encoder of 1 micron 250 lines/mm
let's take a backslash of 1mm
let's take a servo driver working at maximum speed 75 khz
let assume that the encoders can work at 75 khz
the max distance which can be controled in 1 sec is
75000/1000=75 mm
so the time to run 1mm is 1sec/75=13.331/100th. of sec
the time we have to run 1micrometer is 13.3 microseconde
let's talk about a chip working at 500 mhz
this give us 6666 tic between each step
lets take an average of 4 tic per instruction (very long !!!)
it give us the time to run a program of 1666 instructions (very very long)
so if we have a strong motor at half speed we can corect a backslash of 1mm (+-.5mm)in 26/100th of sec
remark you never machined at half speed(not for metal for hobbyst at least)
so with a fast electronic circuit you can make the job
full closed loop with 2 encoders
taking each 1/4 of line of the scale as an ending point and add a nuber of step to close the gap
the case of 1mm nean full reverse which always occur after deceleration and acceleration (so slower) we have plenty time IF we work at a slower speed than the maximum speed of the motor

Hi Folks

Somewhere in this thread it was noted that the software does not exist in the homeshop world to try this.

Check out this link referencing a EMC2 retrofit on a LARGE machine with rotary and linear encoders:

http://jmkasunich.com/cgi-bin/blosxom/shoptask/wichita-trip-02-20-08.html

Scroll down to the bottom and read the last paragraph.

It may be possible to try this today if your willing to spend the money on the necessary linear feedback position transducer.

Having seen this done before (But for a insane amount of dollars) and experiencing the benefits, it is high on my list.

Al

hello,
about price
glass scales quadrature encoder 250 lines mm(250*4=1micron) 400mm useful 174 us$ +freight(china)
1 encoder CUI amt 102-103 about 40us $ from 100 to 2048 lines per turn (you can choose your resolution by switch)
one uhu hp servo controler (kit )180us $+15 $ for UHU chip
one good servo motor 90V 11 A 150$
so 200+40+200+150 max 600$/axe
*3=1800 $ if you add 500$ for cncbrain make 2300$
to be very nicely oversized
and with a full closed loop (when cncbrain will be operationnal)
let say that you add 350$ per axe comparing to the minimum you have to spend
with 350 $ in precision mechanical stuff per axe you have peanuts !!!

Hi

That looks like a great start. I assume 400mm is the length of the scale?

glass scales quadrature encoder 250 lines mm(250*4=1micron) 400mm useful 174 us$ +freight(china)
(!!!

In my usecase I need scales up to 96 inches long, do you have a link to any suppliers for scales in that range?

Thanks
Al

ask this guy

Hello rokag3,

I can offer very inexpensive scales to you that either have A and B signals or A/ A not & B/ B not signals. The scales are 5 volt TTL square wave out of the encoder. The resolutions are 5um/.0002 inch or 1um/.00005 inch. My e-mail address is LMSC@pacbell.net if you want more information.

Regards,
Tim
***************


You can see the scale information at http://www.digitalreadoutsystem.com/Jenix/jenix_dro.htm.

for the price ask the guy

We learned some lessons during the tuning of our $800 bearing equipped Bridgeport (we didn't pay that due to "connections" but that's what they'd cost to replace them). These were recognized as being too costly soe we came up with some relatively cheap A/C"s in other posts - however, even so, "cheap" and "A/C" are oxymoronic when used in the same sentence.

Interestingly, the Bridgeport lesson came AFTER we did some glass scale/DRO equipped tuning of our Excello manual mill. We used it to make our indexing fixtures for cams as well as our drive dog places wherein we built in the lobe center offset. This mill was used because it had DIRECT table feedback and we needed/wanted that for "jig accuracies" for the drive dogs we were making. Result: the stuff wasn't accurate even though the DRO said we were RIGHT ON to the tenth.

As part of the learing process, we went back and forth between the Excello and the BPT - both highly reputed machines and in relatively good (Excello) and pristine (BPT) condition. Yet, accuracies sucked.

NOT being computer programming savvy nor being a machine tool builder by trade but having machine tool bearing experience, we went about eliminiting the readily visible or measureable sources of hysterisis/slop.

We learned first hand that slop is readily and easily remedied by proper adjustment of the machines' gibbs and ways - by doing this alone ( BTW the Excello has acmes and stock bearings otherwise), we could readily hold tenths in position accuracy during subsequent drilling/boring/reaming operations.

Thus, by doing simply machine tuning and slop elimination (well within the capability of the limited budget DIY'er), we got a machine that EASILY could meed the 0.001" target accuracy of the discussion. All this in a machine that I have less than $3000 invested into (including a full motor rebuild with armature balance for good measure).

Once we learned what to look for and how to do it, we then went ballistic on the BPT and threw high buck bearing technology at it willy nilly.

In both cases, however, residual slop in the systems allowed the cutters to either push off the material and shove the part away. Or, it grabbed hold of the material and cut it more deeper (sic) than we wanted/needed. That's when we got serious about the bearings and ball screws in the BPT.

After measuring the TRUE compliance under load (two guys at 200lbs plus each) by pushing and shoving on the table did we find that the "precision" BPT wasn't as precision as we thought. By doing this we both quantified and located the source of the slop.

It WAS to spec but not adequate for what we wanted to do. Figuring that the CNC controller worked well, we attacked the slop hysterisis factor in the fashion we could - preloaded bearings and other known machine tool/high end machine tool tricks.

Asside from the high buck ball screw bearings and low buck ball screw nut preload reload, the BPT is TOTALLY stock as is the software. Once we got rid of the slop, the mid 90's software ran amazingly fine. We'd hoped to be able to mill cam masters but the REAL/ACTUAL intent for doing all the tricks to the CNC side of the machine was to mill golf club putters.

The client admitted that the kluged up A/C's we had in it and shim preloaded were MORE than adequate for putter quality machining. The high $ bearings were added to get into 0.0001" accuracies necessary for our cam masters.

The way I see it is this: the human mind can readily and adaptively compensate for the wear (slop) in the front end of a car as it wanders due to worn ball joints and sloppy tie rod ends. HOwever, if you fix the worn components, the computer (driver) can spend time doing other things than correcting for stuff that really needs to be or should be fixed anyway.

I'd contend that the technology exists to create and make deadly accurate machine control systems. Although adaptive software is great if you have the where with all to create and understand and adapt it - sadly I don't. But I, like many others so mechanically inclined can deal with and eliminate mechanical slop creaed by poorly chosen and/or poorly assembled parts.

I'd contend, however that in light of the hardware and knowledge availability to the community, some computer solutions involve complexities that are beyond the skill and/or education levels that some of us have.

The neat part about this hobby/business is that there are many paths to the same water hole.

If we were to look at a machine as being analog and control electronics as being digital, then it does seem reasonable that a mathematical model could be made representing a machines mechanical errors as a coordinate sytem between axis when interpolating axis movement precisely from linear scales.

If the analog parameters are not changing with load and speed then a digital result for axis positioning could be achieved with a known accuracy and repeatability. There are some high end digital contol sytems on CNC machines that now do just that, allowing for less expensive ballscrews to be employed on CNC mills.

I would find it interesting if in real time a digital control system could respond fast enough to compensate for changing on the fly variables and yet respond correctly or with enough consistancy when required to maintain a known result.

Encoders with scales if required should not be part of the close loop for axis positioning, but should only provide error correction and fault detection for the motor and amplifier. A pre-mapped system with digital control response
might be of some benefit to small machine tools with leadscrews and backlash and squareness issues, but then again this seems to me somewhat like a man trying to walk on artificial legs. I think I would rather have good legs to begin with not have to rely on the artifical ones.

My thought is build a better machine and you will then find a better solution.

My thought is build a better machine and you will then find a better solution.


It's just a question of money the closed loop is mainly software and a few electronique .
Software take a lot of time (it's a hobby too) but no money
I am going to try to make a little box based with an arduino and some external circuit to supervise one linear encoder one rotary encoder check the step in and dir in
check that the rotary encoder has moved
if yes check the load encoder(scale) has moved too (and in which dir)
(i will take encoders working with the same resolution 1micron/tic) if the load encoder has not given a tic in the same dir then i will generate a tic at the entry of the servodrive
if the load encoder has moved and the rotary encoder has not i will calculate the direction and send the dir and step to the servo drive.
if there is a need for a correction then i will put a flag to calculate the timing between 2 pulses send by the pc and insert my corection at a medium rate between 2 pulses since i will use the error feature which is inside the servo driver
i am sure that it's not going to be magic since a good work should be done by rerwrite the PID of the servo completely
But this will corect errors "on the fly" and be less performant for reducing large backslash but maybe I can make some fuzzy logic to recognise some situations and use a strategy of correction preregistered(later)

It's just a question of money the closed loop is mainly software and a few electronique .
Software take a lot of time (it's a hobby too) but no money
I am going to try to make a little box based with an arduino and some external circuit to supervise one linear encoder one rotary encoder check the step in and dir in
check that the rotary encoder has moved
if yes check the load encoder(scale) has moved too (and in which dir)
(i will take encoders working with the same resolution 1micron/tic) if the load encoder has not given a tic in the same dir then i will generate a tic at the entry of the servodrive
if the load encoder has moved and the rotary encoder has not i will calculate the direction and send the dir and step to the servo drive.
if there is a need for a correction then i will put a flag to calculate the timing between 2 pulses send by the pc and insert my corection at a medium rate between 2 pulses since i will use the error feature which is inside the servo driver
i am sure that it's not going to be magic since a good work should be done by rerwrite the PID of the servo completely
But this will corect errors "on the fly" and be less performant for reducing large backslash but maybe I can make some fuzzy logic to recognise some situations and use a strategy of correction preregistered(later)

This seems like fun, but if this were a new plane design, I would be one of the first to be building the parachute, given the complexity of what you are suggesting. Yet I would love to see what progress you might make.

This seems like fun, but if this were a new plane design, I would be one of the first to be building the parachute, given the complexity of what you are suggesting. Yet I would love to see what progress you might make.


This complex !!!?

I was expecting critic but on the other way basically use the linear encoder to fix an ending point and adding step at the entry of the servo drive this is very very simplistic wrting a PID is by far more complex !!!

rokag3,
I like the fuzzy logic idea. I think you'll find it very helpful for this type of control.

Remember that you'll be putting out multiple corrections as the errors will be larger than one count. You'll need to limit how fast you push the corrections at the amp as the mechanical time constant of the system will cause a delayed response and if you correct too fast you'll overshoot your target.

You may find it helpful to view the dual-loop control video on the Galil website to see how this is done with an analog servo system.
Bob