cnc,linear encoder design question

I have read recently the post regarding the concept of using linear
encoders for a do-it-yourself CNC conversion. It didn't seem to me
that I saw clear answer to the questions I have had, myself,
regarding this concept. Let me state my idea here in detail and
solicit some specific answers to specific questions.



My design principle is this: I have a Rong-Fu 35 milling
machine that I wish to convert to CNC operation. My prior experience
is with an open loop stepper system on a Sherline CNC machine. What
I wish to do is to use a PM servo motor which drives through a belt
reduction system, to turn a zero backlash precision ballscrews to
move the X, Y,Z axes of the machine. Directly to the table will be
mounted a linear incremental encoder. my intention is to use a U.S.
digital system which is composed of a 360 line per inch Mylar encoder
strip and read head. I am likely to use gecko drivers and as a
result the encoders will tie directly to the servo drivers of each
axis.

I heard much talk regarding resolution and accuracy. As I
see it from an engineering standpoint the accuracy of any given
system is equal to or less than the most inaccurate part in that
system. In a system of a servo motor, a belt drive, thrust bearings,
ballscrew, and ballscrew Mount, the most inaccurate part in this
system is likely to be the ballscrew. From the specifications I have
seen ballscrews range in accuracy from approximately 0.004 inches per
foot, to significantly more accurate. However in the world of do-it-
yourself CNC the more accurate ballscrews are significantly out of
the price range of the average user. The ballscrews I had
entertained using were Thompson units available through Reid tool and
had a lead screw accuracy of 0.004 inches per foot. That means that
no matter what I drive the ballscrew with, the accuracy of the system
can be no greater than the inaccuracy of the screw. Any attempted
extremely high resolution servo motors and encoders is a waste of
time, as they will very accurately locate an inaccurate part
(relatively speaking). For instance in the above example of a lead
screw accuracy of 0.004 inches per foot the lead screw accuracy is
approximately 3/10,000 of an inch per inch. A servo motor with a
rotary encoder of 1024 counts per revolution is capable of resolving
approximately 2/10,000 of an inch per inch assuming a five turn per
inch lead screw. If this servo motor where to be running and 5 to 1
belt reduction the accuracy of the system has not increased by a
factor of five, but because of the fact that the lead screw is no
more accurate than 3/10,000 of an inch, then the system regardless of
any engineering prior to the ballscrew, can never resolve greater
than this

The highest order of accuracy, it would appear, in the world
of CNC would be the use of an encoder to directly measuring table
movement. In a system such as this the inaccuracys downstream from
the actual table movement are irrelevant, as what is finally measured
is table movement as opposed to lead screw rotation. As I see it
only one significant problem exists and that problem is a backlash
since table movement and the servo motor movement are not
specifically and directly tied together as they would be with a
system where a rotary encoder is mounted to the servo motor. then a
situation of oscillation could potentially exist where a driving
voltage to a motor would not immediately result in table movement as
backlash in the screw would need to be absorbed prior to actual
encoder movement. The logical counter this problem would be to
design a system with zero backlash therefore any movement of the
motor would be table movement itself.

The question I have is: in the actual operation on the
milling machine will the physical shaking on the machine induce
sufficient jumping of the encoder signal to induce oscillation. This
question could likely be restated as: is a ballscrew driving a table
sufficiently rigid so as to allow one near zero actual table movement
even when forces are applied assuming that the ballscrew is
completely incapable of rotation. If indeed this system is
sufficiently rigid then the system will assume that the linear
encoder being read is indeed the rotary encoder mounted to the back
of the servo motor. Now the resolution of the system is purely a
function of only one variable and that is the resolution of the
encoder. With the encoder I intend on using the resolution is; 360 x
4 (as this is a quadrature encoder), or approximately 7/10,000 of an
inch. If my part is one inch long, or one foot long I should be able
to hold this resolution regardless of length of the part because the
cumulative error of the ballscrew does not exists and the final
feedback to the hardware is indeed actual table movement and not
encoder movement tied to a servo motor. I realize that of course any
error in the encoder will be replicated in the part, however encoders
are known to be of extremely high resolution and accuracy.

Any questions, comments, or general gripes with this overall
design scheme ,I would be interested in your opinions. Especially if
you give them to me fairly soon as I intend on amassing the parts as
soon as possible.


Bill H