CNC Conversion Accuracy and Performance

This page contains the results of extensive research I did on the accuracy and performance of CNC conversions. This is a long page without a lot of action or engaging pictures. Feel free to skip it unless you are just hungry for knowledge.

What Are Your Goals?

The first step in planning your CNC conversion (or purchase of a new CNC machine) is to determine what your goals will be.  It’s all fine and well to do the normal male thing, grab all the spec sheets, go down all the columns, and decide yours will have the best value in each column.  Just consider whether you really need all that (i.e. are you over engineering for your purpose?), whether your skills are up to those levels, and whether you can afford the expense.  Also please remember the immutable laws of nature which boil down to:

If I want something twice as good it will cost me four times as much and be at least four times harder to achieve.

Your goals should be measured in several ways.  First, there is the capacity question.  How large will your work need to be?  You’ve probably already made this choice in advance if you own the machine and are doing a CNC conversion on it.  It will be very hard to increase the capacity of the machine beyond what it was originally designed for.

Second, you must determine what you want to do with the machine as far as CNC is concerned.  This is somewhat a function of what your software is capable of, but that software must be chosen with your aspirations in mind too.  Are you going to basically use CNC to do the same kinds of things you could already do manually?  Are you going to do things in CNC, such as 3D profile milling, that are impossible to do manually?  Are you prototyping, or attempting to manufacture parts efficiently?  Make sure your software supports whatever you are up to.  Think about the materials you will be machining.  Wood requires very little precision and power, hence the prevalence of gantry-style router machines.  Aluminum and plastic are much easier to cut than steel, and dictate different design tradeoffs.

Lastly, you must determine the degree of accuracy and speed you are shooting for.  On this you must be brutally realistic.  It sounds great to think you are going to hold to a tenth of a thousandth and have speeds in excess of 100 inches per minute, but that will not be an easy goal to reach, and do you really need it for what you are doing?  Sometimes it isn’t obvious.  If you are going to 3D profile mill a design in metal, and you want to do so efficiently from a production standpoint, tighter tolerances and higher speeds may mean a surface finish that minimizes the need for separate finishing steps in your manufacturing process.

Here are some roughly guidelines where accuracy is concerned:

• 0.020”:  Beginnings of “hobby class” machining accuracy.
• 0.010”:  Most hobby work is doable except for engines and complex parts.
• 0.005”:  Some engine work is now doable and parts are beginning to have a fairly finished appearance.
• 0.001”:  You can build about any model engine or complex tool with the exception of turbines and other very high rpm close tolerance work.  For me, this is an ideal target and I find I can hold work to 0.001” in my manual machine work if I’m careful and think about what I’m doing.  You will probably have to do some careful adjustment on your Asian tools as well as have the right measuring instruments and good techniques to achieve this goal.
• <0.001”:  Now you are into the serious stuff.  If you can tread in the tenths of a thousandth, you aren’t afraid of much.  If your machines can do this reliably, they are well set up and in good condition and you should be doing something other than trying to learn anything from me!

In my case, I want to use my tools to prototype a variety of parts that may be related to hot rodding, guns, or virtually anything else.  Realistically this means I have to deal with steel, and can’t assume machines suitable for wood, plastic, or even aluminum will be acceptible.  My work envelopes are determined already by my machines.  I purchased an Industrial Hobbies bed mill and a Lathemaster 9x30 lathe.  Both of these tools have largish envelopes so far as the hobby machine spectrum is concerned.  I am shooting for 0.001” accuracy and a high degree of repeatability.  If I can even get close to 0.001”, my CNC capabilities will be the equal of my old manual machining capabilities.  If I actually hit 0.001”, I will be able to do even better with CNC than I could manually. 

I am less concerned about speeds, so will go with steppers rather than servos in all likelihood.  Speed is important for production applications, but I am unlikely to use these machines for production purposes, although I may make a few small runs from time to time.

Rigidity

A lot of the performance potential of your machine is going to be baked in by its rigidity. This is one reason why the best machine tools weigh so much--there simply is no other way to keep them rigid than to use a lot of structure. Cast iron is heavy, rigid, and also has good vibration dampening characteristics. This all contributes to rigidity and the performance of the machine. Some industrial machines even use granite as part of the structure, for example the bed or column of a mill. It is extremely rigid, and has several times the vibration dampening of cast iron. I have wondered about incorporating inexpensive granite surface plates into some machine designs.

There are things you can do to your machine to increase rigidity. A common modification to Asian lathes is to replace the compound clamp with a stronger 4-bolt variety. I won't spend too much time on this page dwelling on how to improve this issue. You can find plenty of that elsewhere and ultimately you can wind up remanufacturing your machine if it gets to be too much of a Holy Grail.

Beware some of the materials you may be tempted to use to obtain rigidity. Cold rolled steel, for example, warps easily if you machine the skin off one side in an attempt to make that side true. Aluminum is not as strong, but since it does not have this property a lot of CNC machine builders are using aluminum. Cast iron also does not have this property, but is sometimes expensive and can be difficult to machine.

I will leave you with a parting thought. Sometimes we can trade speed for rigidity in a CNC machine and its a good bargain. If your machine won't cut 0.125" on a pass, it may be capable of cutting 0.0125" and doing so over the course of 10 passes. Since its automated, we can live with it when we have to. Machines that are great at hogging aluminum or plastic may need to take much more shallow cuts on steel to get where they are going.

Friction vs Accuracy vs Backlash

Having decided on the broad design goals for your CNC project, you will shortly descend through the looking glass and into the myriad of conflicting opinions and details about the design choices needed to realize those goals.  Before we go there, we need to discuss a little bit about friction, accuracy, and backlash.  Consider this a background of understanding needed before we can discuss the actual gadgetry with any authority.

Let’s start with backlash.  While there are precise engineering definitions, let’s keep it a bit more informal.  Think of it as lost motion of your machine along one axis.  It can be due to many factors.  An input is given to the axis that is lost, and does not move the axis.  You experience it whenever you change directions with your handwheels during manual machining.  There will be a brief period when turning the handwheel does not move the axis right as you reverse direction.  The distance that would have been moved by the handwheel is the backlash.  On my Lathemaster lathe, this value is somewhere in the 0.004 – 0.006” or possibly even 0.008” region.  It can be precisely measured, but let’s not worry about it for the moment. 

Backlash comes about for a variety of reasons.  On an ACME screw with a single nut, there is some inherent play between the threads on the screw and the nut engaging.  The support system (bearing or bushings and ancillary components) may allow the screw to move axially back and forth as well, which adds to backlash.  Backlash is often not such a problem when manually machining because we’re all used to taking up the slop with our handwheels well before we reach the point of cutting.  As you can imagine, it’s necessary to take out that slop any time you reverse directions on an axis. 

The manual operations one can undertake almost by definition do not involve reversing direction while cutting unless the reversal is intended to pull the tool out and stop cutting.  It would be very hard to freehand cut a circle on a mill by turning the X and Y handwheels just the right amounts, but if you did, you would see glitches in the circle at the direction change points due to backlash.

For certain operations, backlash can induce chatter and other undesirable effects.  Imagine that instead of cutting that circle manually, you are using a fly cutter in the mill.  The forces on the cutter are very similar to the manual circle cutting as the fly cutter travels around its circle.  If the tables are jittering back and forth under those forces due to backlash, the fly cutting will not go well.  Most of the time, the mass of the machine together with the friction, will provide enough resistance to minimize this effect on manual machines.

Now let’s consider the CNC case.  CNC software, such as Mach, often has backlash compensation built in.  It’s a rule of thumb sort of thing—you have to measure your backlash, and the software will do something similar to a manual operator in making sure the slack is taken up before cutting proceeds.  It doesn’t work quite as well as for the manual operator, but it isn’t bad.  Unfortunately, the CNC software rarely can exercise the judgment and experience of that manual operator.  Sometimes a tool path is generated that calls for a direction reversal that just isn’t accommodated well by the backlash compensation.  Even worse, CNC now allows us to contemplate doing things a manual operator would be hard pressed to follow.  Cutting that circle should be child’s play for CNC, assuming the machine is up to it and doesn’t choke due to backlash.  Imagine some of the engraving and profiling (think sculpture-style carving) that can be done.  Lots of direction reversals going on there.  Just carving or engraving an alphabet makes you think how often your pen reverses direction when you write down the letters.

Backlash compensation really can’t compensate for cutting that involves a direction reversal.  There’s just no way to take out the slack fast enough without moving the cutter for it to be practical.  If you want to do these kinds of operations, you will have to minimize the backlash in your machine.

Lathes have it a little better than mills because the profiling operations that reverse direction seem to be less common there.  Unless you are making nozzles or chess sets with flowing curves, most shaft work can probably avoid direction reversal.  For the mill, backlash is a hard problem.  Based on what I read in the forums, if you want nice 3D profiling, you had better be able to get down to 0.001” or less backlash.

Now let’s get back to the friction and accuracy issues.  We’ve already mentioned that friction can be helping to hold things in place and fight chatter.  It dampens errant motions, in other words.  Unfortunately, friction is bad in most other respects.  It’s a crude force that has to be overcome.  You can imagine on a tiny scale that as the machine pushes against the force of friction, the axis will suddenly break free of the friction and start moving.  Anyone who has ever played with friction understands this stick/slip phenomenon and it isn’t helpful to precise CNC operations.  It can make very slow precise motions jerky, and in the worst case, can be a source of chatter. 

CNC has very little means of sensing what’s really going on (we’ll talk about encoders and limit switches in a minute, but they are no match for a human operator’s senses, or even a good DRO!).  Because they lack this fine feedback (even servo systems with encoders to an extent), they depend on the machine always doing the same thing if they issue the same commands to it.  This insensitivity of the computer (frustratingly literal devices that they are), has been dealt with largely by dramatically increasing the precision of the machines, which also involves lowering their friction.  Ballscrews and linear slides, much beloved arcana of the CNC community, are all about increasing accuracy and reducing friction. 

Now for the ugly secret that you must have surmised by now: low friction requires zero backlash!  Without friction, backlash is left free to wreak maximum havoc on our operations.  The tool cutter can potentially jitter around on every cut along every axis within the backlash spec if we let it.  That would be very bad!  Other forms of errant motion must also be precisely controlled if we eliminate the damping effects of friction.  Tormach, for example, argues that very low friction linear bearings are best used either for small CNC machines cutting wood and plastic, or massive industrial CNC machines that have rigidity and don’t need the damping.  They argue that for a medium sized case (most hobby CNC conversions fall here), if you want to cut metal, you will have high cutting forces and will benefit from a little bit of damping.

Another aspect of accuracy is the accuracy of the screws themselves, which we’ll call Lead Accuracy.  The threads will not move the nut exactly the same distance per turn on all places on the screw.  This is another case (much like backlash!) where the CNC control commanded an input to the axis and it didn’t wind up where it was expected to.  Lastly, too much friction results in having to apply a lot of force to the axis, which may in turn deform the screw or some other part of the machining—another change in positions that the CNC control did not ask for and cannot compensate for.  Let me say it loudly and clearly, lowering friction and backlash almost always improve accuracy.

Okay, so now we understand approximately the relationship between backlash, friction, and accuracy.  What we can conclude is that our worst enemy is backlash.  It is never good, always causes trouble, and can only be compensated for in a limited number of circumstances and then not necessarily very well.  They used to say when you buy a stereo, spend most of the money on the speakers.  I would say that if you are building a machine tool spend most of your money getting backlash under control.  Note that I said, “under control” and not “eliminated”.  Your backlash needs to be less than the accuracy you are striving for, potentially a lot less.  Cutting wood to an accuracy of 0.010” can obviously live with a lot more backlash than cutting steel to 0.001”.  Following the backlash, our next enemy is lead accuracy, and then perhaps friction.

ACME Leadscrews vs Ground Ballscrews vs Rolled Ballscrews

Now that we’ve educated ourselves a little bit on the vagaries of friction, backlash, and accuracy, let’s delve into one of our first design choices for our CNC conversion.  Specifically, do we want to use standard ACME leadscrews (probably already on our machine in the event of conversion or much cheaper to purchase if building from scratch), rolled ballscrews, or ground ballscrews (in approximate increasing order of cost and desirability)?

This is an important question with respect to cost as precision ground ballscrews can be extremely expensive, even when scrounged on eBay.  In addition, the effort required to convert a machine from the leadscrews that came with it to a set of ballscrews properly mounted can be very large as well.  We had better not set off in search of precision ground ballscrews out of sure desire to have bragging rights!

The differences in these choices boil down to some of our old friends: efficiency (aka friction), accuracy, and backlash.  What a surprise!  Let’s summarize these choices:

Clearly, if you can afford them, ground ballscrews are the superior choice.  Yes, you can get very high precision ACME’s, but they will have extremely high friction and will need an anti-backlash nut that adds even more friction.  All of that will lead to increased wear.  That wear is going to stress your machine and it will be uneven, introducing varying amounts of error across the range of travel that are hard to compensate out. 

I can’t see the benefit to the ACME’s when you can get rolled ballscrews at a decent price unless you actually want the friction for some reason, or already have the ACME’s and are trying to decide if you can “live with them.”  Living with them has to be a function of how much your application is susceptible to backlash problems, what accuracies you hope to achieve, and so on.

I can imagine some scenarios where living with an ACME screw makes sense.  We’ve already talked about how reversing direction is a prime problem area for backlash, and how mills are probably much more sensitive to backlash than lathes.  If you are willing to forgo profiling operations, you can also worry less.  I think also plasma tables and router tables probably care less either because they are working in wood and don’t need high precision (though some applications will) or because plasma cutting isn’t a high resolution operation.  Lastly, there may be a preload situation on an axis that squeezes out the backlash automatically.  Some folks have even attached springs for this purpose.  This is easy to do, for example, on a lathe cross slide, where a spring may be used to force the tool against the work piece creating a bias against the backlash.  It may be that the Z-Axis for some tools would be fine with some backlash because gravity will drag a heavy spindle down against the backlash.  The head on my Industrial Hobbies mill weighs over 200 lbs, for example!  OTOH, most recommend counterbalancing these tools will improve their performance.

I’m going to try out the ACME approach on my CNC lathe conversion and see how it goes.  It’s easy to drive the existing screws and hard to fit ballscrews.  For reasons described above, I believe lathes are less subject to the backlash morass than mills and the worse case is I will do a conversion later.  For my mill conversion, I cannot see even starting out with ACME screws.  It’s down to the issue of rolled versus ground ballscrews.  So what are the pros and cons of those two approaches?

If we can get ground ballscrews at a reasonable price, we have the best of all worlds.  What if we can’t?  Let’s explore how well we can do with the rolled ballscrews.  Rolled versions can be had fairly cost effectively from a variety of sources.  They have a track record of successful use as well.  One fellow on CNC Zone uses a Bridgeport retrofitted with rolled ballscrews to do custom CNC cams for engines.  He reports they have achieved nearly 0 backlash with the rolled screws.  His error on direction changes for circles is 0.0001”.  As you can imagine, machining cam lobes requires precision!  He achieved this through a combination of preloaded ballnuts (more on these in a moment) to get to 0.0002” and careful tuning for the rest.  We can see from this that it is possible to create a rolled screw configuration that does away with all but imperceptible amounts of backlash. 

I’ll get on to the techniques needed to reduce backlash on rolled screws in a bit, but first, what about the accuracy of rolled ballscrews?  The accuracy refers to the fact that the threads on the screw may not precisely move the nut according to specification.  In fact, one turn of the screw may move the nut different distances depending on the starting point of the nut on the screw.  We’ve discussed how backlash compensation doesn’t work as a panacea.  It appears that leadscrew mapping has the potential to be much more successful in dealing with these accuracy problems.  What’s done is to create a map of the inaccuracies and let your cnc software use the map to compensate for errors.

Creating these maps involves varying degrees of difficulty.  One could use a high precision DRO if you have one available.  A fellow on one of the boards was using 4” job blocks and a tenths indicator to laboriously check each 4” of travel.  The pros use a laser system to setup commercial CNC machines and measure to very exacting tolerances in a very short time.  With some judicious tuning of the compensation map, you can keep your lead errors to 0.001” or less. I have also heard of cases where it made sense to focus the compensation on a small portion of the center of the work envelope in order to achieve very high accuracy at the expense of the extremes. I'm not clear how those trade-offs work, but I note as something to consider for further research.

Anti-Backlash Nuts

For those who want to stick with ACME leadscrews and wonder about backlash compensating nuts, this section is for you. For those who have rolled ballscrews because the ground screws were too expensive or hard to find, this section is also for you.

Blah, blah, blah!

Ballscrew Mounting

Blah, blah, blah!

Linear Rails vs Dovetail Ways

Blah, blah, blah!

Steppers vs Servos

Blah, blah, blah!

Encoders and Closing the Loop

Blah, blah, blah!

Machine Adjustments

Blah, blah, blah!

How Good Are the "Pro" Machines?

The figure I see quoted most often is that shops are comfortable that a good machine will hold 0.0005" all day long without too much trouble. With special care, they may do better. Here are some anecdotes I've collected:

- Makino VMC with boring head holds 0.0005" for press fit bushings.

- A Toyoda horizontal mill cut bearing bores in cast iron all day long to 0.0007".

- Mori Seiki SL15 manual lathe holds 0.0002" all day long except for tool wear and will turn to that accuracy a 6" long 2" diameter cylinder with no center. Okuma CNC lathes will do this, but Haas will not, according to these posts.

Back to CNC Cookbook Home...