This is one of my CNC projects and it is currently a work in progress.

CNC Controller

Table of Contents

1. Introduction

The overall CNC controller architecture is to break the controller into a motion board and buffer board.

Motion Board
The motion board is controlled via a standard PC parallel port using step and direction commands to control the motion of up to 4 stepper motors. The step and direction commands are sufficiently standardized that alternative motion control systems (e.g. [Camtronics] or [Gecko]) can be substituted in without breaking the overall architecture.
Buffer Board
The buffer board contains a dedicated microcontroller that talks to a high speed serial port (or SimpliciNet hub; see below) and provides some command buffering. If there is enough buffering, it should be possible to run the CNC equipment from a standard desktop operating system without requiring something as specialized as real time Linux [RTLinux].
There are a total of three configurations:
Parallel Port Mode
In parallel port mode, the motion board is directly connected to the parallel port of the host processor.
Serial Port Mode
In serial port mode, the buffer board is connected to the host processor via a high speed serial port and the motion board is connected to the buffer board.
SimpliciNet Mode
In SimpliciNet mode, the buffer board is connected to the high speed serial port of the host processor via a SimpliciNet hub [SimpliciNet] and the motion board is connected to the buffer board as before.
These three configurations are shown diagramatically below:
CNC Controller architecture

2. Motion Board Design

Before I get into the design issues of the motion control board, I would like to point out that the various revisions of the board are kept in a separate motion control directory.

The motion board takes standard step and direction signals for up to 4 axes and generates the corresponding coil excitations for up to 4 stepper motors. The motion board is designed around the L298 [L298] dual H-bridge which provides up to 2 amperes of current per coil with a maximum voltage of 46 volts (actually only 40 volts due to my choice of Schottky diode.)

2.1 Parallel Port

The motion board is designed to operate with the Enhanced Motion Control [EMC] software. Thus, the pin-outs on the parallel port are compatible with EMC pinouts. The pinouts are copied from a table put together by Lawrence Glaister [Glaister] The differences is that I added another axis called A and I have changed the way that limits are dealt with:

Pin EMC Signal Motion Board
Number Name Direction
1 /C0 Out n/a n/a
2 D0 Out X Direction X Direction
3 D1 Out X Step X Step
4 D2 Out Y Direction Y Direction
5 D3 Out Y Step Y Step
6 D4 Out Z Step Z Step
7 D5 Out Z Step Z Direction
8 D6 Out n/a A Step
9 D7 Out n/a A Direction
10 S6 In Probe (Polarity=1) Probe (Polarity=1)
11 /S7 In Probe (Polarity=0) Probe (Polarity=0)
12 S5 In X, Y, Z Home (Polarity=1) A, X, Y, Z Home (Polarity=1)
13 S4 In X, Y, Z Limit- (Polarity=1) Ground
14 C1 Out n/a n/a
15 S3 In X, Y, Z Limit+ (Polarity=1) Ground
16 C2 Out n/a n/a
17 /C3 Out n/a n/a
18-25 Ground Ground Ground Ground

2.2 Wave Tables

Typically, systems that use L298's also use the corresponding L297 [L297]. The L297 is a stepper motor controller chip that provides a step and direction interface to drive the L298 H-bridge. There are two issues with the L297:

My preference is to use a less expensive Microchip microcontroller that is dedicated to do essentially the same task as the L297. The revision A uses a PIC16C505 and the revision B uses a PIC16F628. (The revision A design had a serious design flaw.)

In order download the programs into the microcontroller, a small PIC programmer is placed in one corner of the board. (I hope I have the board space!)

There are three kinds of wave tables that make sense and these are listed below:

Wave Stepping
In wave stepping, only one coil is active at a time. First coil A is tuned on, then coil B, then coil A in the opposite direction, then coil B in the opposite direction and repeat. This mode consumes the least amount of power and provides the lowest amount of torque.
Full Stepping
In full stepping, the two coils are always turned on. First A and B are forward, then A in inverted, B is inverted, A is pushed forward and repeat. The mode consumes the most power and provides the most torque (i.e. double the torque of wave stepping.)
Half Stepping
In half stepping, the coils are turned on in a sequence that provides twice as many steps as the previous two modes. With half stepping first one coil is turned on, then two, then back to one then back to two. It is basically interleaved wave and full stepping.

All three tables are listed below:

Wave Stepping
Phase CoilA+ CoilB+ CoilA- CoilB-
0 1 0 0 0
1 0 1 0 0
2 0 0 1 0
3 0 0 0 1
Full Stepping
Phase CoilA+ CoilB+ CoilA- CoilB-
0 1 1 0 0
1 0 1 1 0
2 0 0 1 1
3 1 0 0 1
Half Stepping
Phase CoilA+ CoilB+ CoilA- CoilB-
0 1 0 0 0
1 1 1 0 0
2 0 1 0 0
3 0 1 1 0
4 0 0 1 0
5 0 0 1 1
6 0 0 0 1
7 1 0 0 1

It turns out that the microcontroller only implements the half stepping wave table. The other two tables can be `synthesized' by simply using all even phase numbers (Wave Table) or all odd phase numbers (Full Stepping).

2.3 Cabling

To simplify cabling, the motion board uses a DB25 cable to connect from the controller to the stepper motors, limit/home switches, and linear/shaft encoders. Since DB25 cables are primarily designed to support signal transmission rather than power transmission, a total of 8 wires are dedicated for each coil (4 for each side.) Thus, a total of 16 wires are used for stepper motor power. The remaining 9 wires (= 25 - 16) are used for limit/home switch detection and linear/shaft encoders. The pin-outs are listed below:

Pin
Number Name Direction
1 Home In
2 EncoderPhase1+ In
3 CoilB+1 Power
4 CoilB+2 Power
5 CoilB-1 Power
6 CoilB-2 Power
7 Limit+ Current Loop
8 CoilA+1 Power
9 CoilA+2 Power
10 CoilA-1 Power
11 CoilA-2 Power
12 +5V Power
13 EncoderPhase2+ In
14 Ground1 Ground
15 EncoderPhase1- In
16 CoilB+3 Power
17 CoilB+4 Power
18 CoilB-3 Power
19 CoilB-4 Power
20 Limit- Current Loop
21 CoilA+3 Power
22 CoilA+4 Power
23 CoilA-3 Power
24 CoilA-4 Power
25 EncoderPhase2- In
26 Ground2 Ground

When it comes to limit switches there are two basic strategies:

Advisory Strategy
The adviosory strategy is that a limit switch event is forwarded to the software that in turn responds by shutting the whole machine down in a fairly graceful fashion.
Depower Strategy
The depower strategy is that the moment that a limit switch event occurs, power is immediately removed from all drive motors and spindle motors.
It should be noted that these two strategies are not mutually exclusive. One set of limit switches can be set up for an advisory strategy and another set of limit switches can be set up for the depower strategy. If the software does not get the machine shut down when the advisory limit switch is triggered, the depower limit switch can depower the whole system.

EMC [EMC] implements the advisory strategy for limit switches. I am real nervous about relying exclusively on the advisory strategy because it relies on some fairly complex software doing the "right thing" in all circumstances. The next option is the "let's support both". There are two problems with supporting both -- first, the it takes up even more cable conductors and board space, and second it allows people to be sloppy about the depower switches, since `after all they are really only there if the first set does not work.' Ultimately, I decided that I only wanted to support the immediate depower strategy. This causes all limit switches to be hooked in series with a current passing through all of them to a relay that supplies drive and spindle motor power. If any cable gets damaged, unplugged, or a limit switch is triggered, the current throught the relay is cut off and the whole system powers down. This is real simple and real safe.

What this means for EMC is that the Limit- and Limit+ are tied to ground. Thus, as far as EMC is concerned, it will never get a limit switch advisory signal.

The DB25 connector used to connect to each axis has 4 wires set aside for transimitting standard quadrature signals from linear/shaft encoders. The two quadrature phases are sent using RS-422 levels to ensure that there is high noise rejectcion due from the power pulses being sent down the the same 25-wire cable. The Revision A and Revision B versions of the motion board are based on the EMC-style advisory limit switches. The Revision C is based on depower style of limit switch.

The rest of the information about the motion control board is kept in the motion control directory.

3. Buffer Board Design

{The Buffer Board is still in the early design phase.}

3.1 Buffer Board Theory

This section discusses the theory behind the buffer board.

In the process of using CNC to manufacture a part, a series of software and hardware tools are used to transform the idea into physical reality. In theory, a CAD system is used to design the part and a CAM system is used to take the design and produce the tool path that is used to manufacture the part. The tool path is most typically represented as a file in RS-274 format, which is also known as `G-codes'.

In the process of generating the toolpath, the CAM system has to take into account the various characteristics of the CNC equipment being used and the properties of the material being machined. Thus, the toolpath for an identical part being machined on a CNC Bridgeport knee mill might be quite a bit different from the toolpath on a CNC Sherline table top mill. The additional power available in the Bridgeport spindle, the increased tool rigidity and increased X, Y, and Z axis power allow the Bridgeport to take significantly deeper and faster cuts over its smaller Sherline cousin. None the less, the resulting part can be, for all intents and purposes, identical.

Once the toolpath has been generated it is up to CNC controller to take that G code and translate it into a series of precisely timed and coordinated moves on the CNC machine. The coordinate space of G codes is in either English or Metric units measured to fractions of inch or fractions of a millimeter. The coordinate space of the CNC machine is integral units where each unit is typically something small like .00025 of an inch or some such. The CNC controller is responsible for converting from the G code coordinate space to the machine coordinate space.

From an abstract point of view, the CNC controller takes the G-Code and reduces it to a sequence of timed events of the base form `at time t, move to position (x, y, z)'. Combining the time time into a four-tuple results in a sequence of 4-tuples (t, x, y, z). (This example is for a 3 axis machine, each additional axis would lengthen the tuple appropriately.)

Historically the conversion of G codes to a sequence of timed events has been done by a single dedicated processor. However, it is possible separate the generation of the timed event sequence from the actual processing of the events on the CNC machine. The timed event sequence can be generated on any general purpose computer and the processing of the timed events can be performed by specialized hardware that can carefully emit the control signals at exactly the right time. I call this specialized hardware a buffer, since it is responsible for storing up a bunch of motion commands and emitting them at just the right rate.

The question arises, how big does the buffer need to be? At one extreme the buffer can be large enough to store all of the tuples before even the first one is executed. At the other extreme, the buffer can be of size zero and we are back to the situation where the processor that is computing the tuples is responsible for the timing as well. In the middle, there is a buffer that is large enough to ensure that the machine never runs out of timed events (i.e. buffer under run) and is not prohibitively expensive.

Before the question of how big the buffer is can be answered, it is necessary to get a handle on how much bandwidth is required between the G-code to timed event converter and the buffer board.

The timed event stream is amenable to a large amount of compression. The first thing to note is that the times are monotonically increasing. The next thing to note is that when the machine is moving from machine coordinate X1 to Xn it must traverse all intermediate coordinates. Thus, delta encoding will result in substantial bandwidth savings. Thus, the sequence (123657, 23459, 5432, 98623), (123663, 23460, 5432, 98623), (123668, 23460, 5431, 98623) can be encoded as an initial value of (123657, 23459, 5432, 98623) followed by delta values of (6, 1, 0, 0) and (5, 0, -1, 0). Note, that the first element must always be positive and the remaining elements must be -1, 0, or 1.

The rate at which these event tuples are processed depends upon the maximum stepping rate of the machine in question. If each machine axis is capable of being ramped up to 100,000 steps per second, there are potentially up to 3 × 100,000 event tuples per second that need to be processed, since the steps may not be occuring in sequence. If we set aside 6 bits for the time delta, and 2 bits for the each axis delta, that results in 10 bits per delta or 3 × 100,000 × 10, or 3Mbps. While, this is not a small number, it is nice to have an upper bound on the required bandwidth.

Are there additional ways of compressing the information? Certainly. The way that looks the most promising to me is to break the event stream up on a per axis basis. By noting that the axis motor is limited by its maximum acceleration and deceleration, it should be possible to come up with a much more compact representation of the axis event stream. For example, if the motor is running at a constant velocitiy, a simple run code compression can be used. For run code compression, a sequence of evently spaced events can be replaced by just counting how many equally space events there are. Similarly, when the motor speed is being changed, a code that describes the amount of acceleration (or deceleration) can compress a whole bunch of motor events. Once the per axis event streams have been computed, they can be mixed back together to form a single unified stream.

One possible command stream might look as follows:

Axis Id (3-bits)
For a 3 or 4 axis machine, only 2-bits are needed. For a full 6 axis machine 3 bits are needed.
Acceleration/Deceleration (5-bits)
This is a 5-bit signed two's complement number. It is zero when the motor is running at a constant speed. It is negative to decelerate and positive to accelerate.
Event Count (8-bits)
The repeat count specifies how many event counts are covered by this command.
The time is implied as the end of the previous command for this axis. For example to deccelerate the X axis from one step every 100 µS, to one step every 200 µS, a command of (0, -2, 100) would do the trick. To keep the motor going for 1000 steps, 4 commands of (0, 0, 250) would do the trick.

Yes, there are lots of details to work out. However, I am quite hopeful that the bandwidth required to keep a CNC machine happy can be well under the 3Mbps mentioned above, preferably under the 115Kbps that most asynchronous serial lines need.

The next step is to get some more solid numbers. In order to do this, my plan is to take the Enhanced Machine Controller [EMC] and modify it to produce an timed event stream. After that, I intend to experiment a little to figure out what kinds of compression work well on the stream.

3.2 Buffer Board Features

This is still pretty rough!

The buffer board has the following features:

The following features did not make the cut:

Control Panel
A control panel allows the user to manually control the CNC machine. This is best done with a separate board that connects via a SimpliciNet connection.
Power supply
The power supply should just be purchased separately.

As usual, the shaft encoders use a dedicated PIC to keep track of the position...

3.3 Buffer Board Design

Before I get into the buffer board design, I would like to point out that the individual revisions of the buffer board are kept in a buffer board revision directory..

{Remaining design issues go here.}

4. Acknowledgements

I have received useful feed back from `Ballendo' and Doug Fortune about mistakes in this document that I have attempted to correct. In addition, I have found the CAD_CAM_EDM_DRO group at Yahooo [E] to be extremely informative. Other useful tidbits of information have been mined from the rec.crafts.metalworking newsgroup.

5. References

[CAD_CAM_EDM_DRO]
CAD_CAM_EDM_DRO group at Yahoo. URL: http://groups.yahoo.com/group/CAD_CAM_EDM_DRO/
[Camtronics]
Frames URL: http://www.seanet.com/~dmauch/. Non-Frames URL: http://www.seanet.com/~dmauch/site_map.htm.
[EMC]
Enhanced Machine Controller URL http://www.isd.mel.nist.gov/projects/emc/ developed by the National Institute of Standards and Technology The source code is available from SourceForge at URL: http://www.sourceforge.net/projects/emc/.
[CNCKits]
Frames URL: http://cnckits.com/ Non-frames URL: http://web.cuug.ab.ca/~fortuned/cnckits/
[Geckodrive]
URL: http://www.geckodrive.com/.
[Glaister]
Lawrence Glaister's CNC pages. Main URL: http://members.shaw.ca/cncstuff/ Bridge circuits URL: http://members.shaw.ca/cncstuff/bridges.html. EMC parallel port pinouts URL: http://members.shaw.ca/cncstuff/emcio.html
[L297]
Stepper Motor Controllers by ST.
[L298]
Dual Full Bridge Driver by ST.
[LS7266R1]
The LS72661R1 is a 2 channel quadrature encoder chip from LSI Computer Systems Inc.
[Mauch1999]
Three-Axis Chopper, Step Motor Controller for Computer Numerical Control (CNC) Applications, Part 2 by Dan Mauch June 1999 (pp. 6-8) issue of Nuts and Volts magazine (Vol. 20, No. 6). Part 2 of a 4 part series in the May through August issues.
[PIC16C505]
14-pin 8-bit CMOS Microcontroller by MicroChip.
[RTLinux]
URL: http://www.fsmlabs.com/.
[Rutex900]
Rutex RT900 Motion Control IC.
[SimpliciNet]
SimpliciNet URL: http://web.gramlich.net/projects/simplicinet/index.html.
[Wedemeyer1999]
Author: Hans Wedemeyer. Bi-Polar Motor Driver.
URL: http://hans-w.com/cnc.htm (about half way down.) Image Only URL: http://hans-w.com/9908-09A.gif.

Appendix A: CNC Notes

This appendix contains some fairly free form notes about CNC as I slowly start to figure all this stuff out.

Mill Types
There appear to be two styles of vertical mill -- a knee mill where the spindle is kept rigid with respect to the floor and the other kind (name?) where the spindle is moved veritcally in the Z axis. Thus, the table for a knee mill moves in the X, Y, and Z axes. The other kind (name?) only moves the table in the X and Y axises and the spindle moves in the Z axis. Knee mills seem to be bigger, so they have more Z axis travel. The larger Z axis travel translates into fewer hassles with getting tools into and out of the spindle.
Backlash
Leadscrew backlash is the slop in the system as the table pushes agains one side of the lead screw of the other. An ideal milling system has no backlash. Unfortunately, real milling systems can have a great deal of backlash. There are at least four ways (probably many more) of dealing with backlash:
Ball Screws
A ball screw is magical device that has ball bearings that push against both sides of the lead screw at the same time. By properly grinding and hardening the lead screw and ball screw races, wear is greatly reduced. In addition to having essentially no backlash, ball screws have very little friction as well. The only drawback is that they cost a small fortune. They are not really an option for a table top mill right now.
SuperNuts
I'm not sure I have the terminology right here. Apparently there are nuts that can be placed on a lead screw to take out all of the backlash. I suspect that they are somethting as simple as two threaded nuts with a spring between them. I need to find out more here.
Pre Loading
Anyhow, the concept is to apply a constant force to one side of the lead screw. There are several ways of accomplishing this task -- weight and pullys, springs, etc. A less clunky looking solution is to mount another nut on the lead screw and put a stiff spring between the two. Any way it is done greatly increases the friction and leadscrew wear. The good news is that the hobbyist lead screws are pretty cheap, so for hobbyist use, it is probably OK to shorten their lifetime. Another drawback is that the cutting tool can temporarily overcome the side load force and cause surface imperfections. The big down side for side loading is the need for larger motors to overcome the increased friction.
Compensation
The more common strategy is to compensate for backlash in software. This works pretty well as long as the tool is being pushed against one side of the lead screw. Whenever there is an inflection point in tool cutting path the software quickly steps to the other surface of the lead screw. But while this is happening the cutting bit is free to flop around a little. Again this results in a surface imperfection.
Coupling Nut
A coupling nut is a longer threaded nut made out of plastic material that is at least 3 times longer than the diameter of the leadscrew. It is inexpensive, does not appear to add much wear and does not appear to add much drag while significantly reducing backlash.
As near as I can tell, most systems compensate for backlash and deal with any resulting problems. I suspect that there is some serious issue with the side loading solution that I do not understand, because I do not know of anybody that uses it.

There is an additional issue of uneven wear on the lead screw. Since most travel occurs in the middle of the lead screw, that is where the most slop is introduced. For small machines, the lead screw is cheap and should be replaced when it gets unevently worn.

Encoders
An encoder is used to measures where the X, Y, and Z axes are. There are two kinds:
Linear encoder
The linear encoderss are attached to the table proper and measure the exact X, Y, and Z location in relation to the spindle. The less expensive encoders seem to be clocking in at .001" to .00025". The linear system can compensate for thermal expansion and contraction in the table.
Shaft encoder
A shaft encoder sits on the lead screw and measure thes amount of lead screw rotation. A shaft encoder can be be accurate up to 2048*4 counts per rotation. Given that a lead screw might be 20 TPI (threads per inch) this results in a theoretical resolution of .0000061035". The big disadvantage of shaft encoders is that they only measure the lead screw position, excluding any backlash. Backlash will eat up the accuracy of a shaft encoder.
SPC
SPC stands for Statistical Process Control. The concept is to measure your parts and notice when they are getting out of kilter and replace the dull tool that is causing the problem before they get too far out of kilter. For the small hobbyist kind of guy what this really means is that your measurement tools have a digital output that can be fed into a computer to be used. The bad news is that there is not that much standardization on the connectors, wires, voltages, and protocol. Sigh, that means a lot of custom adaptors.

There are two kinds of drive sytems:

Stepper Motor
A stepper motor is a motor without a commutator that is controlled by selectively turning coils on and off. Stepper motors cost more per lbf (pounds force) than DC motors.
Servo Motor
A servo motor is a standard DC motor coupled with some feedback circuitry to detect position errors an correct them. DC motors cost less per lbf (pounds force) than stepper motors.
There are pros and cons between using stepper motors and servo motors to control everything. Right now stepper motors seem to be holding the low end of the market and servos are holding the high end. For the same amount of torque, stepper motors seem to be more expensive. For a large machine, a large stepper motor will cost a lot. Stepper motors can be run open loop. Servo motors always have to have some sort of feedback in the system. In theory, when a stepper motor is not moving, it is being held with a very large torque. When a servo motor is not moving, its torque may be bouncing on and off a little as the tool is pushed ever slightly off its mark. On some servos systems there is a certain roughness that results. Other people claim this is really just an improperly tuned servo system (and I tend to agree.) The smaller mills seem to be standardizing on NEMA 23 mounts for stepper motors.

Stepper motors coils are driven in cyclical fashion. There are four ways to drive a stepper motor:

4 steps per cycle, one coil active (wave)
This cycle is N, E, S, W and repeats.
8 steps per cycle (half stepping)
This cycle is N, NE, E, SE, W, SW, W, NW and repeats. The NE, SE, SW, and NW cylces have two coils simultaneously active thereby doubling the torque on those steps.
4 steps per cycle, two coils active (full stepping)
This cycle is NE, SE, SW, NW and repeats. This cycle has twice the torque of the first N, E, S, W cycle. Given that stepper motors cost a lot for a given torque, any technique that will double the torque is pretty important.
Microstepping
With microstepping the current is varied through the currently active two coils to obtain an additional level of positioning accuracy.

If too much current flows through a stepper motor coil it will over heat and cause a motor failure. There are four basic solutions.

Constant Voltage
The constant voltage system is the simplest and slowest. If the stepper motor is rated for N volts, no more than N volts is every supplied to the coil. This is simple and results in the lowest performance.
Series Resistor
A resistor can be placed in series with the coil. A higher voltage is applied to the coil. The series resistor is picked to limit the current through the coil using Ohms law (V = I * R). Essentially when the coil is fully on, the voltage drop across the resistor is same as the constant voltage. When a coil is first activated, the current can ramp up quickly because there is no voltage drop across the resistor.
Chopper stablized
Chopper stablized is the next one up. A higher voltage is supplied. The voltage is turned on and off depending upon whether the current is too high or not. The current is sensed using a small current sensing resistor near ground.
Variable voltage
I haven't seen this one done yet, but it is pretty straight forward. The current is measured and voltage is increased or decreased appropriately. The voltage can be supplied using a switching supply so that it is not a power hog.
Right now the middle two seem to be the most used solutions. Frankly, the variable voltage solution should be given a try.

There are two kinds of stepper motors:

Bipolar
A bipolar stepper motor has two coils. The each coil can be individually activated. Call one coil the north south coil and the other the East West coil. This gives 8 possible activation combinations - N, NE, E, SE, W, SW, W, and NW. A bipolar stepper motor as a total of four wires -- 2 wires for the north south coil and the 2 wires for the east west coil. Bipolar stepper motor coils need to be driven with an H bridge to be able to control which direction the current flows through the coil.
Unipolar
A unipolar stepper motor is one that that has two coils, but the each coil has a center tap. Furthermore, the windings on one side of the centertap are clockwise and on the other side they are counter clockwise. What this means is that current from the centertap to one side will activate the coil in one direction (say north) and current from the centertap to the othe size will activate the coil in the opposite direction (in this case south.) A unipolar stepper motor can connect the centertap to a positive voltage and use one transitor on either end of the coil to control energization direction. Thus, a unipolar stepper motor does not need an H-bridge to control current direction. In general, a unipolar stepper motor of a given mass has half the torque of a similar mass bipolar stepper motor. This is because only half the coil is active at a time.
Interestingly enough, it appears to be possible to use unipolar stepping motors in a bipolar mode. The only catch is you have to cut the current in half because there are twice as many turns active at a time. Very interesting.

There are at least three ways of connecting the motor to the lead screws:

Direct drive
Direct drive hooks the motor shaft directly to the lead screw shaft. Done properly, there is no additional backlash introduced.
Belt drive
The motor shaft is attached to the lead screw shaft with a drive belt. A reduction ratio can be added to slow down a faster DC motor and get correspondingly more torque. There is an opportunity to add some backlash into the system with belt drive.
Gear Drive
A gear drive uses a gear train to connect the motor shaft to the lead screw shaft. One simple strategy is to use a worm gear. Like belt drives, backlash can be added to the system.

Thermal expanstion and contraction rears its ugly head in metalworking. As the temperature changes, the metal part, the cutting tool, and the machine tool expand and contract as the temperatur changes. If the cutting tool overheats, it will frequently be rendered unusable. Whenever you are working on part that need that level of accuracy, trying to keep the temperatures constant starts to matter. The big professional machining centers actually will refrigerate the bearings, motors, etc. to keep everything at a constant temperature. The poor mans equivalent is to run everything in an air conditioned shop; not as good, but not as expensive either.

There seem to be at lest five kinds of cooling:

None
Basically, everything is ether run slow so heat does not build up; or more counter intuitively, things get done fast so that the chips that are thrown out carry most of the heat away.
Flood Cooling
A flood of cooling fluid is drenched over the work piece. Everything is sprayed all over the place. This is very effective and quite messy. You need to have a drip pan to collect the run off, a filter to keep the chips out the pump, a pump to pump it up again, etc. Lastly, the fluid can mess up some the optical encoders out there.
Mist Cooling
A fine mist of coolant is sprayed onto the piece. Seems to work fine. There are lots of complains about the mist getting all over the shop and into people's lungs. Yech.
Drip Cooling
A trickle of cooling fluid is dripped onto the piece. There is much less splatter.
Hybrid Mist/Drip
There seems to be hybrid between mist and drip whereby somebody has a fine mist wand that is sprayed directly onto the part. The lung irration complaints are way down.
Vortec Cooling
Vortec cooling uses swirling presurized air and separates it into hot air and cool air. The cool air is blown onto the part. It is quite loud.
Lastly, the various liquid coolant systems can mess up encoders real fast. Care needs to be taken to keep the encoders sealed gains coolant. It tends to be easier to seal a rotary encoder than a linear encoder.


Where does that leave us? The first issue to deal with is backlash control. For table top systems, ball screws are out because they are too expensive. That leaves conpensation vs. preloading. With preloading friction is way up, so the motors need to be correspondingly larger to overcome the friction. However, this means that the system can be run open loop with no encoders. If encoders are added, they can be the rotoary kind and sealed against any coolant.

If there is going to be no preloading, then backlash compensation is a necessity. Open loop backlash compensaton only works up to a point. Linear DRO's will allow the system to compensate out any backlash. They also compensate out lead screw wear. The linear DRO's need to be sealed agains any coolant.

For small hobby mills, both stepper and servo systems will work. For now, it is probably better to go with stepper motors in NEMA23 frames since that seems to be the better supported solution. It would be tempting at some point in the future to try out an inexpensive gear drive motor directly coupled to the shaft with an inexpensive shaft DRO.

For stepper motors, bipolar motors with either chopper drive or variable voltage seems to be the way to go. Unipolars can be used in bipolar mode.


Copyright (c) 2001-2002 by Wayne C. Gramlich. All rights reserved.