Guerrilla guide to CNC machining

1. About the series

2. Milling? That's so old school!

• The ability to work in a wide variety of cheap, non-proprietary materials, from soft waxes to hard steels. CNC mills are capable of processing plaster, wood, elastomers, hard plastics, metal - or even PCBs and ice blocks - and do so at roughly the nominal cost of stock material itself.
  General purpose additive technologies are restricted to a small set of proprietary materials, and produce parts within a very narrow range of mechanical properties. This limits the applications quite significantly, and also imposes a hefty price tag on even the smallest parts manufactured.
• The ability to produce intricate, high-precision components. Many general-purpose CNC mills boast resolutions of 0.001 mm or so in all axes, and can deliver silk smooth surface finishes in all directions on even a very complex part.
  Most general-purpose additive devices are one or two orders of magnitude behind in Z axis, and offer several times lower X and Y resolutions. This often results in staircase patterns on models; such patterns must be removed through time-consuming and complex sanding or sandblasting processes later on.
• The ability to make parts fast. A typical cutting process takes between 2 minutes and 5 hours on a medium size CNC mill, depending on part complexity, size, and process decisions. When it's done, it's done: little or no finishing or postprocesisng is required.
  Most additive technologies take much, much longer to produce usable parts of comparable size (stereolithography takes from several hours to several days on average), and in many cases, additional work to remove support structures and harden the resin must be performed, further multiplying the time necessary.
• Low initial setup cost. Benchtop CNC mills start around $1,300 or so, not much more than high quality manual milling machines and drill presses, and several orders of magnitude below additive systems of comparable work envelopes ($50,000-$200,000 is common there).

3. Selecting the right mill

• The number of axes supported:
  In the most basic design, general-purpose milling machines operate in three axes (X, Y, Z), with the tool lowered into the workpiece parallel to Z axis. This limits the geometries that may be cut in one step to patterns that could be represented by a two-dimensional "depth map" projected from top onto the workpiece: the cutter will be moved higher in some X-Y locations, and lower in others, but it will not enter the workpiece from any other side.
  The following pictures illustrate the behavior. The shape on the left may be easily machined by moving the cutting head in three axes; the two shapes on the right, not so:

           \_/                             \_/                             \_/
            :   OK                          :   NOT OK                      :   NOT OK
            :                               :                               :
                       __                       ____   _______                        _
        ^      ___    /  |                     /  __|  \     /              _________/ \_
        |   __/   |__/   |                  __/  /______\   /_             |   _     ____\
        z  |_____________|                 |__________________)            |__/ \___|
      
           x-->                            (Undercuts)                     (Features on multiple sides)
  

  Tool path generation software is designed to fail securely; any features that could not be reached without ruining an essential part of the workpiece would not be cut. As such, with the cutter descending in parallel to Z axis, the actual results of three-axis CNC cutting for these shapes would be:

                       __                       ____    _____                         _
               ___    /  |                     /    |  |     |              _________/ \_
            __/   |__/   |                  __/     |__|     |             |             \
           |_____________|                 |__________________\            |______________|
  

  An example of a 3-axis milling process is shown on this Youtube video.
  This is not to say that a basic three-axis unit is incapable of cutting more sophisticated shapes - the thing is, to do so, some manual operations, such as flipping or rotating the workpiece, or splitting the part into segments and then joining them together, might be necessary. These operations, when done well, can be performed with relative ease and high precision using registration pins - but they take your time, increase the likelihood of operator errors, and for some complex shapes, may simply become annoyingly cumbersome.
  The process of manually flipping sides for 3-axis milling is demonstrated on this video.
  Because of these downsides, some of the more advanced machines support one or more rotary axis, implemented either through the rotation of the workpiece itself (clamped in a special holder and attached to a motor), or of the spindle assembly in which the end mill is mounted. Two most common rotation axis configurations are A (rotation around X axis) and B (around Y). Some designs also use C axis - a horizontal table rotating around Z. Affordable benchtop machines use three or four axes pretty much exclusively.
  An example of 4-axis cutting is shown on this video.
  The price tag on four-axis machines is higher, from $100 extra for manual indexers (where the rotation angle still needs to be dialed in manually by the operator, but without unclamping the workpiece - which is already pretty convenient), and from $1,000 for fully automatic, computer-controlled units.
  Do you need a rotary axis? You will need to weigh the pros and cons as you go through this guide, and make the call; in all likelihood, 90% of robot-related mold-based work will not significantly benefit from having a rotary axis; keep in mind that most thermoforming and metal punching dies in the industrial world are "depth map" projections with no undercuts anyway, otherwise it would not be possible to remove shaped material from them after the process. That said, there are some exceptions - and having this extra degree of freedom certainly makes it easier to be creative. If you are into jewelry or general model making, you might find it particularly handy - and chances are, with a CNC mill in your workshop, you will be tempted to explore.
  As such, if you can afford it, it is not a bad idea to get one; if you do not want to overspend, it is wise to keep your options open, and select a machine that supports an optional unit that may be purchased at a later date. Note that there might be third-party, aftermarket alternatives available for many mills (both manual indexers and automatic devices), so you might be able to make some savings there.
  NOTE: For automatic rotary axis units, make sure that all four axes may be simultaneously used and addressed. Some el cheapo units may offer rotary axis functionality at the expense of one of the linear dimensions (such setups will be designated as AXZ, BYZ, CXZ). These "lathe" mode devices are dedicated for working with cylindrical shapes, for example in jewelry making, where the workpiece is continuously rotated. In mold machining and other general applications, you will more frequently have an use for performing three-axis cutting from several predefined angles, and for this, you need is a device capable of AXYZ or BXYZ milling; lacking this, a manual indexer is probably better.
• Mechanical movement ranges (travels) in linear axes:
  Greater X-Y-Z movement ranges directly translate into the ability to process larger geometries in a single pass - and within a given price range, you are likely to see some significant variation in this area. Size matters: although most work may be broken up into segments, the process quickly gets cumbersome if the number of segments is high.
  For robot work and other mechanical designs, keep in mind that you will never be cutting a complete design in one large chunk: individual parts such as leg segments would be machined separately, then joined using ball bearings or other moving connectors; and even within a single segment, it is often preferable to have several elements connected with appropriate fasteners for improved servicability and extendability. In addition, some of the simple, large structural components are often not worth machining in the first place - it is faster, easier, and just as aesthetic and structurally sound to saw aluminium pipes or shafts to desired lengths, then machine plastic top and bottom connectors to hold these in a particular arrangement and attach them to other components in arbitrarily sophisticated ways.
  In my experience, for most tasks, about 15 x 10 cm of X-Y movement is a sane entry point, allowing most small and medium scale projects to be pursued easily - although you might find it necessary to work around these limitations in some designs, and you will be limited in the number of parts you can cut at once. X-Y ranges of about 30 x 20 cm are pretty generous and cover almost all bases imaginable; around and over 50 x 40 cm or so, any differences are largely immaterial for most hobbyist engineering tasks.
  As far as Z axis goes, the expectations there are fairly modest. In my experience, merely about 4 cm is required to account for most mid-size robot-related work, as almost all parts consist of flat planes or relatively small protrusions and curves, as far as this direction is considered. About 8 cm or so gives a comfortable margin for virtually all future projects. Most machines meet this criteria, although note that there are some types of CNC mills - often called routers - meant specifically for 2D contouring, and these may lack the travel range required.
  In any case, you are unlikely to utilize more than 10-15 cm or so in Z axis, so it probably makes no sense to overspend there.
• Precision and mechanical resolution in all axes:
  These parameters together describe the ability for a machine to mill fine details and produce smooth, snap-fit surfaces.
  Mechanical resolution reflects the physical axis addressing capability for positioning motors, and is used by the embedded controller to interpolate all movements between subsequent locations requested by the NC program. If this resolution is too low, movements in multiple axes may produce jagged and unattractive curved surfaces.
  Precision, on the other hand, tells how trusted the mechanism may be to assume the same position in a repeatable manner (taking effects such as mechanical backlash into account). If this value is too low, manufactured parts may randomly deviate from desired specs, and surfaces of any type may end up being uneven or rough.
  To illustrate, the drawing on the left shows an interpolated movement between points A and B with high precision and high resolution. The next image shows low resolution, but high precision - a staircase pattern. Finally the last image shows possibly high resolution, but poor precision - random deviations from the expected path.

               / B                       | B                         / B
              /                         _|                           |
             /                         |                            _/
            /                         _|                         __/
           /                         |                          /
        A /                        A |                        A |
  

  Mechanical resolution is often provided in machine specs, and commonly falls anywhere between 0.001 and 0.1 mm. If for any of the linear axes, this value is higher than 0.05 mm or so, the machine is probably less suitable for high-quality, precision work; values of 0.01 mm and more are desirable, as they enable very smooth motion.
  Precision is a more complex topic. The most common way to express it is through a parameter called repeat accuracy - a somewhat ill-defined spec that describes the error in resuming the exact originating position over and over again following movements of some sort. The problem is, with benchtop mills, manufacturers often do not bother to tell you the value - and when they do, they get creative. Some provide the worst-case accuracy expected to be maintained through a complex and lengthy cutting process, meaning that in practice, you will almost always see cutting precision better than specified (and so a value of 0.05 mm might be perfectly acceptable, and translate into a typical actual error of 0.01 mm or less). Others come up with convenient and uninformative tests designed to show behavior near their best-case scenarios, such as relatively small movements in one axis only - meaning that in practice, cutting precision will almost always be worse than specified (so, a value of 0.05 mm might be too low for precision work, and translate into a typical error of 0.15 mm).
  And then some manufacturers just go wild and claim repeat accuracy identical to motor resolution, which likely means they never bothered to check in practice.
  Anyhoo... common sane repeat accuracy values you might see on benchtop mills are between 0.01 and 0.3 mm. Because of the aforementioned problem with testing methods, it is probably best to dismiss unrealistic claims right away; if a manufacturer of a $1,500 lightweight mill with stepper motors and acme lead screws claims to have a repeat accuracy better than 0.05-0.1 mm, it's likely not an honest estimate; the device will not outperform a $15,000 0.05 mm mill with digital high accuracy servos, ball screw mechanisms, and the like. At the same time, be wary of mills where the manufacturer readily admits a repeat accuracy worse than 0.15 mm or so - these are probably not well suited for precision work.
  What else? Oh - for machines with a rotary axis, the values for this axis will be specified as an angular distance. You may trivially convert them to linear dimensions for a particular workpiece size, of course. For example, if the value specified is 0.05°, and the surface of a piece 10 cm in diameter is being machined on a rotary axis, the linear value is 100 mm * π / 360° * 0.05°, or about 0.044 mm.
• Movement speed, acceleration, and spindle speed (RPM):
  To understand these parameters and their impact on milling speed, it is important to grasp how an end mill does the cutting. The most common variety used for all multipurpose work superficially resembles a drill bit - it's a round shaft with a sharp cutting blade (or blades) wrapped around in a spiral fashion. Unlike a drill bit, however, end mills are meant to cut in all directions, and do bulk of their work being moved sideways.
  Simplifying slightly, a Z axis cross-section of a two-flute end mill may look like this:

                         _
  	              / /   -.
                       / |_     \
  	            /    \     |  
  	            |    |     | direction of rotation
                      \_   /     |
           swarf -->    | /     /
         clearance     /_/   <-'
  	  
                      ^
                      `- cutting edge
  

  Because of this design, per each rotation, a given end mill is physically capable of removing only a fixed surface area of material that could immediately fit in the space underneath the cutting edge. The linear speed at which it is advanced over the workpiece in every turn may not exceed a certain value - going past that point will simply drag a non-cutting tool clogged with removed material through the workpiece, damaging it or the cutter, whichever yields first. Naturally, it makes sense to operate the cutter near this maximum chip load most of the time, as to remove as much material as possible as quickly as possible.
  A CNC mill that moves the cutter fast, but offers low RPM settings only, may not deliver top cutting performance in easily machinable media - maximum chip loads might be hit at relatively low linear speeds. Likewise, an end mill with very high RPM but slow linear motion may be unable to get anywhere near reasonably high chip loads, just shuffling air most of the time.
  Note that this model describes material removal in terms of the surface area (in X-Y); the actual volume of material that could be removed in each pass depends on how deeply the cutter can be driven into the workpiece, and corresponds to several constants, such as flute helix geometry, tool strength, and so forth. Neither of these parameters is adjustable for a given cutting process, though, so we can ignore them for now.
  All right - a proper balance of high RPM and fast movement speeds is advisable for optimal material removal performance. Simple enough? Well, there are some gotchas:
  • If you start with optimal parameters, further increases to linear speed alone might not result in faster high-load cutting - but they still translate to faster traverse speeds when the machine is NOT cutting, and simply moving freely from one cutting spot to another. Likewise, with fast linear movements, the performance may be improved when finishing and smoothing out surfaces after the initial removal of bulk of the material, at which point chip loads are negligible. This matters for some, but not all, cutting processes.
  • On the other hand, high linear movement speeds may be just a marketing gimmick if the mill is not able to accelerate and decelerate quickly enough. Since most automatically generated cutting operations for complex shapes involve relatively short movements, a mill with an acceleration rate of 0.02 G (circa 200 mm/s²) and a maximum speed of 150 mm/s will never attain that speed during a stop-go-stop movement of 5 cm; in fact, the average speed would be only 50 mm/s.

              ^
            v | - - - - - - - - - - - - Vmax
              |
              |          . - .
              |        .       . 
              |      .           .  
              |    .               .  
              |  .                   .
              |.                       .
              +-------------------------->
               t = 0              t = 1s
    

    Look-ahead processing might lessen the impact of acceleration deficiencies, however. In absence of look-ahead processing, the mill has to come to a full stop after every consecutive movement, as there is no guarantee that the next movement would not require a nearly complete reversal of directions. With look-ahead, the controller will notice that in some cases, no full stop in all axes is required to smoothly enter some subsequent movements - and will optimize braking to account for this where appropriate by peeking forward into the received code.
  • RPM increases come with a catch, too. As rotational speed increases, so does friction. Heat will be produced faster, but dissipation will not improve nearly as well, leading to an increased cutter and workpiece temperatures. Once a threshold is exceeded, the material may start to be negatively affected, leading to poor finish at best, and catastrophic cutter failures at worst (for example, when molten plastic solidifies around the cutter, preventing further material from being removed). This is mitigated by the use of sharp carbide tools with proper friction-reducing coatings, such as titanium aluminium nitride (TiAlN), or the introduction of cooling liquids.
  • Another problem encountered at very high rotational speeds is that some end mills may lose rigidity due to centrifugal forces, or start oscillating while driven into holes and other confined spaces ("chatter"), causing temporary loss of cutting precision.
  Yes, it's complex. Machine performance is an important usability consideration when comparing competing models, and quite possibly the aspect where devices within a given price range diverge the most. Slow operation is annoying, and also means you would have to put more hours on spindle bearings and motor, and wear them off faster, to get the same job done. So, less bang for your buck. How to choose wisely? Here are some ballpark figures that should suit the types of cutting recommended in this guide well:
  • Choose a machine with spindle speeds between 6,000 and 20,000 RPM. Values lower than 4000 will slow you down very significantly with many prototyping materials when using miniature cutters. Spindle speeds over 25000 will almost certainly require messy cooling to be fully utilized in many materials. Typical values seen on affordable CNC mills are 3000 to 18000 RPM. Within these bounds, the more the better.
    Adjustable spindle speed is a plus, as the ability to operate at lower speeds is useful with large diameter cutters, certain difficult materials, or for reducing chatter in tight spots.
  • Choose a machine with linear speeds of at least 1/10th of its movement range per second - for example, if X movement range is 300 mm, at least 30 mm/s would be advisable. The more the better. Below that, it might be agonizing to watch it move between distant locations on full-size workpieces. Most benchtop CNC mills boast speeds of 10 mm/s to 200 mm/s here.
  • Aim for a ratio of 500:1 to 100:1 between RPM and linear movement speeds - for example, if maximum RPM is 15000, and linear movement speed is 100 mm/s, the ratio is 150:1. Outside that range, it might be difficult to fully utilize higher RPM, or higher movement speeds, when working in typical prototyping materials with precision cutters.
  • When comparing mills within these specs, count a 2x increase in movement speed as roughly comparable to a 1.6x increase in maximum spindle RPM. This is an unscientific guesstimate, but probably a good rule of thumb for work in prototyping materials with small diameter cutters.
  • When a linear speed of more than 50 mm/s or so is claimed, be sure to find out the acceleration rate. Not all manufacturers provide it, but it might be wise to inquire. Values under 0.1 G or so are a warning sign that the top speed might be seldom achieved in some cutting processes; 0.05 to 0.5 G are common, the more the better.
    Look-ahead processing is a major plus, although in practice, a mill doing 0.05 G with look-ahead processing will still be worse than one capable of 0.5 G in "dumb" mode.
• Spindle motor and linear motor power:

  At some point, you might be tempted to work directly in softer metals (aluminium, silver, gold, tin, copper, etc) or in some of the particularly tough plastics (e.g., varieties of polyester). For many hobbyist uses, there is no need to, to be sure: it is considerably easier to manually cast hard resins into machined molds made of easy to process materials - and for short runs and small to medium size parts, some of these casting resins match lighter metals, both performance- and price-wise. Mold-based processes are also much faster and cause less tool wear, and many properties of metals, from metallic luster to ferromagnetism and electrical conductivity, may be approximated using filled resins.

  There are some situations where metals perform considerably better, naturally, most notably in elevated temperatures: many plastics lose much of their rigidity past 100-150° C, and even most resistant ones (Bakelite, silicone rubbers) yield at about 250° C - whereas most metals easily go over 500° C. If you dream of a jet-powered robot, this might matter a lot.
  All these materials, and even some fairly hard steels, can be processed with just about any CNC mill - but on some underpowered units, tough workpieces might have to be cut very slowly, as pushing the cutter too quickly into the workpiece in an attempt to improve chip loads could stall motors. Slow cutting is not a concern when making a 5 mm gearwheel, but will be a pain when working on larger structural parts. To achieve a decent performance there, you need a powerful spindle motor and competent axis feed motors, too.
  As far as spindle power goes, the bare minimum for soft metals seems to be about 75-100 W, and 300-500 W seems to be a sweet spot. For axis motors, somewhere around 30-60 W is desirable (although because of various transmission rates possible, it might be wiser to compare linear thrusts instead - only not all vendors bother to disclose them).
  If you anticipate doing some metal milling or other taxing jobs, be sure to look at these parameters carefully.
• Tool Z0 sensor support, standard or optional:
  With most of the low-cost tool holder systems available, it is impossible to replace an end mill and maintain a perfect, desired "free" length sticking out of the holder. Yet, in many cutting processes, it is advisable to alternate between 2-3 different end mills to complete cutting with minimum effort (for example, employ a large diameter cutter to quickly shape a large surface area, then a very small end mill to drill tiny screw holes). If a tool that extended 20 mm from the holder is replaced with one that extends 24 mm, a milling disaster could occur unless the difference is properly measured and compensated for - so it is important to somehow detect this offset.
  Tool height sensor is a very useful device that does just that, in a very accurate manner. A typical design is simply a precisely machined, flat tipped metal block of known dimensions, placed in a known reference location (surface of the table, workpiece top, or somesuch). Whenever the need to recalibrate tool length occurs, the cutter is taken to a X-Y location occupied by the sensor, and then slowly lowered until the tool comes into contact with the top surface of the block, closing an electrical circuit and prompting the machine to retract the tool and calculate the result.
  Since the mill knows the exact position of the cutting head, and the expected location of the top of the sensor, the length of a tool may be trivially computed based on this data.
  It is possible to do the same using various manual tricks, though only with an accuracy of around 0.3 mm or less, which is sometimes too low; plus, nothing beats the safety and convenience of a Z0 sensor integrated with the machine. So, if a particular unit supports Z0 sensing by default, consider it a major plus; if one is available as an option, seriously consider getting it - the price is unlikely to be prohibitive.
• Languages and software supported:
  A computer-controlled machine is obviously useless without the right software to manage it. Although almost every manufacturer bundles in some basic proprietary CAM software to control the mill, you will likely find these programs somewhat lacking - and they may also suddenly stop working with the next generation of your workshop PC, with no support in sight. Given that CNC mills age much slower than computers these days, and last longer than companies might, this is a pretty important concern.
  The purpose of CAM software is to convert 3D meshes prepared with a variety of CAD applications (or general-purpose 3D modeling programs) into series of paths the cutter needs to follow to reproduce the desired shape. Once these paths are ready, they need to be broken down into thousands of painfully basic movement commands (such as "set speed to A mm/s" or "move to x=A y=B z=C") and sent to the embedded controller on board of the mill. The first steps are largely machine-agnostic - but the knowledge of how to produce these basic commands in the last stage is clearly hardware-specific.
  Read the specs to see what control language the machine speaks natively. A common quasi-standard is G-code, an ancient yet rich language used by many hobbyist and industrial systems alike. There are some significant variations in the dialects of G-code used by each manufacturer, so consider it a bonus if the machine includes a detailed reference of the variant used - but even without this, G-code spells at least rudimentary compatibility (or easy integration) with almost all third-party CAM software on the market.
  From time to time, manufacturers come up with their own languages. This is not necessarily a bad sign - some of these are documented and widely embraced (Roland RML-1 is a prime example) - but whenever you see a vague mention of a proprietary control protocol, or a name of a language that gets very few hits outside the manufacturer's site, be wary.
  Some of the big names in "budget" CAM software include Mayka, Deskproto, VisualMILL, and some more; try to browse their websites or download trial versions to see if a particular mill is natively supported - if yes, this is a good sign that the communication protocol is both well-documented, and popular enough for others to care about it. If the manufacturer provides their own plugins for these programs for a machine that has been on the market for a while - well, it makes you wonder.
• End mill mounting options:
  The end mills you will most likely find most useful for precision work in the range relevant to robot building come in two primary shank diameter varieties: 3 and 6 mm; in addition, some manufacturers, such as Hanita, also use 4 mm for some of their precision cutters. These two or three values will almost certainly cover all your needs, but no single option will be sufficient - so it is important to figure out how (and if) these end mill diameters could be mounted on the unit.
  Common mounting options you might see:
  • Fixed diameter (collet-less) chuck. Found on some of the low cost units, this option might add a hidden price to the purchase, and prove to be rather annoying: to accommodate tools of various shank diameters, you will need to purchase several collet / spindle assemblies, and replace them as needed (which is cumbersome). It's not a terrible solution, to be sure - but be sure to consider the extra cost and effort.
  • "Legacy" collets (R8, Morse taper, etc). Commonly used in the States, these are cheap, easy, but som might be less accurate for precision work because of how they grip the tool. Standard types are available through a large number of independent retailers.
  • ER16 collets. Possibly the best option for small CNC mills: these collets remain very affordable ($15 a piece or so, for example from Techniks USA), but grip the tool more uniformly, and accommodate a greater range of tool diameters. Fancy variants, such as Techniks "DNA" collets (twice the price, but at least in my experience, better finish quality with small tools) are also available.
  • Easy-release and automatic tool changer chucks / collects. Almost guaranteed to be rather expensive and probably not worth the extra benefit of being able to replace the tool in 3 seconds, rather than 30.
  Other options are more common on industrial machines - but if you encounter something else on benchtop mills, simply investigate to get a good understanding of that technology's limitations and costs.
• Unit size, weight, power needs:
  Benchtop units span from small, fully enclosed devices the size of an inkjet printer, to units weighing over 100 kg and taking almost 1 x 1 m of desk space. When going for some of the larger models, be sure to fully account for physical characteristics before making a decision (looks are often deceptive, especially on the web). Double check that you will be able to fit your machine of choice through the door - some doors are merely 60-70 cm wide, so fitting a mill with 40 cm working area and a 25 cm gantry-style frame around it is not given.
  Be aware that many CNC mills have thick frames cast from solid aluminium or other metals for extreme rigidity; because of this, even a small mill may be remarkably heavy. Verify that you have a piece furniture large and stable enough to host it. Note that the surface must not only be able to withstand the static load, but also must not be wobbly, as parts of the machine will start and stop moving rapidly, and will exert some horizontal shear forces on the furniture.
  Benchtop mills usually run on standard, single-phase 110 / 230 VAC supply, but some of the larger units might be power-hungry and require several amps in peak (typically not more than a hairdryer or a vacuum cleaner - but if the circuit is already shared with these appliances and an electric kettle or somesuch, you might eventually trip a breaker or burn the place down). Double check the specification and make sure you would be able to supply power from an adequate circuit.

• Table clearances and workpiece weight limits:
  Mostly irrelevant, in the ranges seen on typical general-purpose benchtop devices. Table clearances are easily confused with axis movement ranges, but these parameters simply specify the largest workpiece the unit will be able to accommodate - and offer no guarantee that the entire piece would be machinable. On most mills, there is a margin of 2 to 15 cm between the end of a machinable region, and the physical boundary of the working table - and the practical difference between having 2 and having 15 is almost none.
  One important caveat is that if you intend to use a rotary axis, then the maximum radius of a rotated piece must not exceed 1/2 of Z axis clearance; there is usually no need to rotate particularly wide pieces in typical robot work, but it might be a consideration for artistic projects or whatnot.
• Positioning accuracy:
  Usually expressed as a deviation over a specified movement distance (most often 300 mm), this describes how close the machine is expected to match a calibrated reference of a given length. If positioning accuracy is 0.1 mm per 300 mm, a milled part of 300 mm may be off by 0.1 mm compared to what 300 mm is truly, scientifically defined as.
  In practice, the impact is typically marginal: with the aforementioned accuracy, a hole 10 mm in diameter may actually measure somewhere between 9.997 and 10.003 mm, depending on the build of this particular unit (the error will be repeated for all the parts cut on that mill, so snap fits between parts you made on the same machine would not be affected no matter how badly off it is).
  Whenever a higher accuracy is required, the machine may be carefully calibrated, and software compensation may be implemented. In practice, however, for a vast majority of applications, the errors introduced are too small to be of any concern.
• Origin reproducibility:
  While the machine is powered, it is able to keep track of its location with the accuracy specified elsewhere, by the virtue of knowing what commands were issued to motors to date. If equipped with closed loop controls, it is also able to monitor current axis angles and compare them to internal tracking data. When turned off and on, however, the device assumes that parts of it might have been moved or tampered with while inactive, so it fully reinitializes self and tries to detect reference points in all axes by retracting fully, until an appropriate home position sensor is triggered. This process might be somewhat inaccurate, as the sensing mechanism is often relatively simple (e.g., a regular microswitch).
  For most intents and purposes, it does not matter a lot - the only situation where it comes into play is if the machine needs to be restarted due to an unrecoverable error (such as an invalid instruction in the data sent) arising in the middle of a cutting process, or when power is lost due to external causes. Neither of these situations is common, and so the parameter does not warrant close scrutiny.
• Software resolution:
  Hardware resolution plays an important role in determining the quality of machined surfaces, as it is used to interpolate movements between program-specified locations using series of discrete steps. Software resolution, on the other hand, is less significant: it determines how finely the endpoints for such a moment may be specified before the interpolation takes place. Almost all mills boast software resolutions of 0.01 mm or more, and differences between 0.001 mm and 0.01 mm are very unlikely to have a noticeable effect.
  (It is an interesting warning sign, however: if a manufacturer makes unbelievably fine software resolution, higher than the actual hardware resolution, a selling point - the other claims made warrant a closer look, too.)
• List of millable materials:
  Just about any material softer than the end mill itself can be machined. Manufacturers often come with short lists of recommended media, but the criteria for inclusion on such a list are not set in stone, and so such lists are a matter of vendor's discretion and honesty, and not much more.
  Any CNC mill should be capable of handling a broad variety of materials, including elastomers, hard plastics, wood, wax, plaster, and a whole lot of other stuff, too. Almost any mill should be capable of milling soft metals, too - and to see how well suited it is, it is always better to look at spindle motor power instead (see earlier discussion). On the other hand, most benchtop mills are poorly suited for rapid machining of ferrous metals, regardless of what the manufacturer claims - simply because motor powers, cooling requirements, and chip containment needs are a wholly different animal in such processes.
• Motor control loop type:
  With open control loop motors, the machine does not have a way to measure its position, so there is no immediate feedback on whether a move succeeded or not (due to a collision with something in the working area, for example). Most units feature crude controls to detect overload conditions (current sensing) - but otherwise, fingers crossed. If you accidentally bump the cutting head hard enough with a vacuum cleaner pipe, your bare hand, or some other material that should not be there in the first place, the cutting process will continue at an unexpected offset, possibly ruining the workpiece or breaking the end mill.
  Closed control loop motors have built-in sensors to precisely measure actual position and counter misalignment. Because of this, the machine is considerably more resistant to these unforeseen circumstances, and will either correct itself as needed, or shut down with an error whilst minimizing workpiece or cutter loss following an incident of this type.
  Closed loop systems are more expensive, and in most cases, not strictly necessary; it is a nice feature to have, but do not consider it a significant advantage for amateur, attended work. Attach your workpieces securely, set sane cutting parameters, check twice, keep the work area clear - and chances are, you will not have to worry.
  (Closed loop systems are often more precise, too, but this should be accounted for by other metrics discussed earlier.)
• Automatic tool changer support:
  Some of the more expensive machines come with automatic tool changer units (ATCs), or have them available as options. Although this technology is immensely useful for unattended, production-grade milling processes, not so in hobbyist uses - with a good Z0 sensor, it takes about 30 seconds to switch tools manually anyway.
  Not only the price tag is fairly high, but specialized tool holders, as well as a supply of compressed air, are typically required.
• Coolant support:
  Useful for high-speed metal cutting, for work in some thermoplastics (such as polycarbonate), and for mills that operate at very high RPM. There is no need for cooling in the processes discussed in this guide, and since the solutions available are fairly messy to begin with (cooling oils are inevitably at least partly dispersed as an aerosol, so they may prove to be a nuisance in home environments), you probably do not want to go there.
• Scanner support:
  Some machines may come with 3D scanning heads, most commonly with a piezoelectric needle-shaped touch sensor. Although this sounds somewhat cool in principle ("scan and reproduce any shape you want!"), in practice it is almost always easier to measure and manually copy mechanical parts and other geometric shapes, so the usability of such a device is dubious at best. You will scan a couple of coins and pendants, then get bored.
• Noise ratings:
  Chances are, noise ratings are not what you think they are: these specs correspond to the noise produced by the unit when NOT cutting through the material. Even when two values are supplied, these refer to "standby" (all motors off) and "not cutting" (motors on, but moving freely) states only. Virtually all benchtop units are likely to be reasonably quiet when just idling (that is, they seldom make more noise than a small hairdryer) - and when cutting begins, the amount of noise generated is almost strictly a function of the material machined, cutter geometry, RPM, and feed rates in all axes.
  If you get a powerful and fast mill, with some materials and under certain conditions, it may produce loud, high-pitched noises that warrant wearing earmuffs or vacating the room until the cutting is done (these sounds will likely not propagate well through walls and closed doors, though). This will be the case regardless of what the specifications say, however, and you will be able to control it at the expense of cutting a bit slower where necessary.

A final good tip is to find the actual PDF instruction manual for your mill of choice before you make the final call. These are often either available for download from manufacturer's website, or may be bootlegged somewhere; if not, you can often talk resellers or existing users into giving you a copy. Manuals let you get a good grip of how it feels to operate the unit and what additional limitations it might have.

In most cases, once you order the machine, you will have to wait a couple of weeks for it to be manufactured and shipped to you; relatively few models are in continuous stock. This is actually pretty good, however, as it gives you time to shop for tools, software, cutters, and machinable media, so that you can unwrap the device and start having fun right away - instead of staring blankly at the device once you unpack it, then getting discouraged by rushed experiments.

4. 3D CAD software

5. 3D CAM software

• Rotary axis machining modes:
  If your machine does not support a rotary axis, this is obviously an unnecessary expense - but if it does, be sure to confirm that the version of a CAM program you are considering will be able to accommodate the types of machining you anticipate. There are three primary types of fourth axis cutting - and not all applications support them all.
  • Indexed fourth axis cutting (X-Y-Z motion, A constant):
    In this mode, the application lets you specify a fixed angle from which which you want to cut the part at the moment, then performs regular three-axis X-Y-Z milling without further rotating the workpiece through the process. Once done, a different discrete angle might be assumed or manually dialed in, and cutting from that new angle may take place. This repeats until all sides of the part are complete.
    This mode is particularly useful when cutting parts with two or more distinct, discrete faces - for example, a two-sided PCB, a tire, or a cube with screw holes on multiple sides. This is the most rudimentary type of fourth axis milling, and one that may be easily approximated without native support for such operations in your software, too: you can simply rotate your 3D model in a CAD application to a desired angle, save the data, load it into the CAM program, make sure that the same angle is dialed in on the rotary axis, and commence cutting - then rinse and repeat until all the desired angles are done. This fully manual approach might be sometimes cumbersome and is somewhat error-prone - but certainly doable.
    This illustration shows an example part that would respond well to indexed rotary cutting from several discrete sides:

                 |      |    
                 |______|    ^
    	       |  |      |
    	       \  /      z
    		||       
                    ::      (x)  y -->
    		::
    		
    		Side A
                                _
              __   ____________| |_
             |  |_|                |              -.
             |                     |_               \  
    	 |_______     (O)       _|  Side B       :  A = const (while side is machined)
                     \             |                /
                      \____________|             <-'
    
                    Side C
    

  • Lathe-type fourth axis cutting (X-Z-A motion, Y constant):
    In this mode, the application will rotate the workpiece in A axis while the cutter, locked to a single Y position, moves sideways and up / down along the rotation axis, removing material from the top.
    This mode of operation is similar to the concept of a lathe, and makes it possible to quickly achieve high-quality, seamless finish on cylindrical shapes and other objects that do not have well-defined, planar faces. Examples include shafts, threads, rings, and so forth. The mode is somewhat more useful for artistic work, and less so for everyday robotics, although does come handy from time to time, for example to make custom worm drives - and unlike the previous rotary axis cutting method, it cannot be conveniently implemented without proper software setup (though it can be often approximated by cutting the workpiece from anywhere between 4 to 16 sides, albeit imperfectly).
    This illustration shows a shape that would respond well to continuous rotation X-Z-A cutting:

                       |      |
    		   |______|   ^
                         |  |     |
                         \  /     z
    		      ||
                          ::     (x)  y = const
    		      ::
          
                      .--. .--.
                    .'   |_|   '.        -.
    	       /             \__       \
    	      |      (O)       _)       :  A
    	      |               |        /
    	       \       _____.-'     <-'
    	        `.___.'
    

    Some programs further permit the fixed tool position in Y axis to be offset from the center of rotation by a certain amount. This is useful for operations where more sophisticated off-center features need to be machined without switching back and forth between lathe and indexed cutting.
  • Fully simultaneous four axis cutting (X-Y-Z-A motion):
    In this mode, the tool moves in all linear directions, and the part is being rotated simultaneously, making it possible to achieve all types of complex geometries, particularly various eccentric shapes, off-center holes, and so forth. Compared to varieties of milling restricted to motion in three axes, generation of optimal paths for true four axis milling is far more algorithmically challenging, however - so not all applications offer it.
  Having full support for the first two modes, and the ability to mix them to achieve an approximation of true four axis cutting is important to fully utilize your rotary axis. A real support for simultaneous four axis toolpaths is a nice touch, too, as it may simplify some processes by the virtue of not having to manually define machining angles, offsets, and regions for complex parts - but don't count on having a real use for it very often.
• 3D milling strategies and optimizations in three axes:
  There is a large number of various milling strategies available in contemporary CAM programs - starting with 2D motion methods for engraving and cutting out shapes in sheet material; through 2.5D options, where in addition to moving in two dimensions, the cutter also assumes two or more preprogrammed Z positions to cut at various depths; and finally to 3D methods that enable seamless modeling of complex features by unconstrained motion in all axes.
  Of all these, if you have a semi-competent CAD program, only 3D strategies are of significance for typical robot-related work - and only some of them, too, as CAM applications tend to boast largely redundant variations of similar algorithms as wholly separate methods. To make comparisons harder, naming conventions for these toolpath strategies are not normalized, and every vendor comes up with own lingo to refer to the same set of concepts.
  Regrettably, "lite" versions of most CAM programs have crippling limitations on the types of toolpaths generated, so figuring out what is available in the product you are considering is an important and difficult step. The three core methods that are of great value in robot work and many other types of machining are:
  • "Overhead" 2D toolpath projection:
    In this mode, a two-dimensional X-Y path of some shape (a set of parallel lines, a spiral or zigzag pattern, etc) is projected on top of the workpiece. The tool then follows this two-dimensional route, moving higher and lower in Z axis depending on the desired shape of the machined part in the spot underneath.
    This toolpath type works great for flat regions and light slopes, where it is computationally inexpensive and removes material very efficiently, with relatively simple and quick motions. More steep surfaces, and vertical wall edges, might not be reproduced with much fidelity, however. For example, if toolpath sweeps the area in X axis, and a vertical edge of the shape shown below on the left must be cut, the outcome may not be satisfactory:

            ____                                          ____
              - \ - - - - - -.                                (
             .-  \  - - - - -'                                 (
      	 `- - \ - - - - -.                                  (
      	 .- -  \  - - - -'                                   (
     	 `- - - \ - - - - -> (*)                              (
     	 ________\                                     ________(
     	
     	 Toolpath shape and edge                       Final effect of the
     	 to be reproduced                              sweep
    

    The problem may be addressed either by reducing toolpath pitch (but this might be wasteful, as it significantly increases the time needed to complete the operation), or aligning the toolpath so that edges are aligned with the direction of a toolpath (but this might be very time consuming or virtually impossible to arrange for complex models). Because of these problems, this should never be the only type of a toolpath supported by your program - but rest assured, while very basic, it remains one of the more useful ones.
    A good program should support scan lines parallel to X axis and to Y axis; it is sometimes also good to have support for arbitrary scan line angles between X (0 degrees) and Y (90 degrees), square or round spiral, or radial (star-like) and zig-zag patterns (but you can live without them). Everything else is probably purely a novelty.
  • Z axis slicing:
    In this mode, a set of horizontal (X-Y) planes is intersected with the model at a constant Z-axis interval to slice it into multiple layers. At each of these planes, the intersection produces a set of boundary curves around model features. Once offset for tool size, and checked for collisions, these curves can be then traced around by the cutter to very faithfully reproduce vertical and sloped patterns.
    This method ensures a great and efficient reproduction of complex vertical features, such as gear teeth, perfectly round holes, and so forth. It fares relatively poorly with gently sloped surfaces, however, as it produces a staircase pattern unless Z plane pitch is chosen to be very small (which, again, is wasteful). That said, together with the previous method, they form a formidable duo - and any application that does not support both, or limits their use in any way, is in all likelihood not a good choice for robot work.
    A good program should support constant Z axis interval cutting (sometimes referred to as waterline or horizontal) with no strings attached - and preferably, should generate such paths quickly (some programs, such as Deskproto, tend to take forever). It is a plus if the program additionally makes it possible to auto-detect horizontal planes on the model and adaptively fill them with projection-type toolpaths without human intervention.
  • Constant surface distance milling:
    If your goal is simply to make sure that all surfaces of a complex, curvy shape are perfectly smooth, setting up a clever combination of the two aforementioned strategies does the trick - but it takes time and effort to manually define which region should be machined which way. This is where smarter, surface-analyzing algorithms come into play.
    One of such algorithms relies on "draping" a 2D toolpath on the surface of a part - so that the distance between tool passes never exceeds a certain maximum value in any axis, and any curvature of the model is automatically accounted for.
    Relatively few programs offer this strategy (one example is VisualMILL, under the name of 3D offset pocketing), but where available, it should be considered a time-saving plus.
  Well, that's the most important trio; everything else is likely a variation of these methods.
  For all cutting strategies, it is also worth determining what types of toolpath optimization and grouping is enabled in the program. The following illustrates the process of Z slice machining of two holes in a workpiece; the process on the left is how a nominal cutting order would look like with no optimization (it involves a large number of back-and-forth movements between two cutting sites); the process on the right is optimized more wisely:

                   -- 2 -->
                  <-- 4 --
                   -- 6 -->                          -- 3 -->
        ____       _______       ___      ____       _______       ___
            | -1- |       | -3- |             | -1- |       | -4- |
  	  |     |       |     |             |     |       |     |
  	  | -5- |       | -7- |             | -2- |       | -5- |
  	  |     |       |     |             |     |       |     |
            `-----'       `-----'             `-----'       `-----'	  
  

  Many, but not all programs feature such optimizations. DeskProto, for example, does not.
• Selective machining features:
  In addition to having control over how the material is being removed, it is also very important to be able to tell where the process should take place to begin with - and which regions should be off limits for the cutter. For example, after cutting the entire part with a large diameter tool, you may want to mount a small one just to machine some holes and narrow valleys on the part. Since the second cutter is small and fragile (so it needs to be operated slowly and toolpath pitch must be small), it would take forever and be a huge waste of time to run it over the entire part just to work on features limited to less than 1% of part's surface area.
  It is very useful to have a program that supports these milling region definition strategies:
  • Manual region selection:
    At the very least, the application should have the ability to restrict toolpaths to a custom rectangular X-Y-Z region on the model; and it is very beneficial if the application offers other X-Y section definition functions, such as freehand curves, circular regions, etc - or lets you import such region definitions as curves from your CAD application. The ability to exclude sections within a selection is also useful (supported by VisualMILL, but not Mayka).
    This is a critical feature; if absent from the tool, you will have a great difficulty working with large molds for multiple parts, and you will have to dedicate a lot of forethought to compensate for its absence during part design stage.
  • Angle- or feature-dependent region selection:
    Another relatively useful feature is the ability to specify surface analysis property ranges that need to be satisfied in order for the region to be automatically selected for machining. Some applications permit you to restrict machining to horizontal or vertical surfaces only (e.g., Mayka); other tools may simply let you specify an arbitrary range of angles to include in or exclude from the process (VisualMILL).
    Several tools can also automatically machine features such as sharp corners or narrow valleys, which is sometimes useful for accurate finishing operations.
    Not having any of these automatic region detection features is not a disaster, but they tend to save time every now and then.
  • Residual machining:
    Several programs let you keep track of the material removed in previous operations, and only machine the parts in regions that were not adequately machined before, and where residual material is still present; so, if you want to process the same material with two tools, one of which has a smaller diameter and may be able to reach new locations, you do not have to select them manually.
    In theory, this is a great feature; in practice, your mileage may vary. Some programs tend to be fairly inaccurate with the workpiece models they use for such calculations, making it not particularly useful for high precision work; other programs are often just unbelievably slow once this feature is engaged. It makes sense to try it out in demo versions, and see how it fares. If well, consider it a plus (but one that takes a back seat to constant surface distance machining or feature-dependent region selection mechanisms mentioned earlier); if it does not work well for you, do not lose sleep over it, and do not be tempted to overpay.
• Arc interpolation:
  G-code contains provision for circular, helical, and spiral movements of the tool, and many CNC controllers understand and interpret these instructions in a manner optimal for the hardware in question. A majority of consumer CAM programs, however, defaults to outputting linear movements only - meaning that every arc movement is broken up into dozens or hundreds of small linear steps, inflating NC program size, impairing look-ahead processing on machine controllers, and potentially introducing needless vibration.
  Depending on the part machined, the ability to use arc interpolation (codes G02, G03) may speed up the cutting process by 10% or so, and make it easier on the equipment. So, some of the higher quality applications (VisualMILL, Mastercam) permit you to either simplify toolpaths by detecting and merging together arc-like movements in already generated toolpaths, or to natively output arc motion instructions for certain toolpath strategies.
  If your machine does not support G-code, or does not support arc interpolation, you simply have to live with linear interpolation only; but if hardware support is present, it makes sense to shop for a program capable of utilizing it.
• Cutter shape and collision detection management:
  There are three most common types of cutting tools that you will be using almost exclusively - and these should unconditionally be supported by your software of choice:
  • Square tip end mills (also referred to as: flat or cylindrical). Used to efficiently work on models with dominant flat horizontal and vertical surfaces, where they produce great surface aspect even with very generous toolpath pithes; you will be almost certainly using them the most.
  • Ball tip end mills (aka: round, hemispherical). Great for quickly approximating complex curved surfaces without producing a staircase effect. They also can be pushed harder with less vibration, by the virtue of not having protruding flute ends on the very bottom. These cutters will be of use in some projects, and may also serve as a good tool for initial roughing, double as drills in soft materials, etc.
  • Corner radius end mills (aka: bullnose). An all-purpose compromise between the two aforementioned types: a square cutter with a flat bottom, but corners slightly rounded. They perform comparably well to square cutters on flat surfaces, and can be used for curved surfaces with reasonable efficiency, too. Some of the very useful small-diameter, long-reach cutters come in this variety only from some manufacturers.

             square tip         corner radius      ball tip
          
             |   |              |   |              |   |
             |   |              |   |              |   |
    	 |   |              |   |              |   |
    	 |   |              |   |              |   |
    	 |___|              `---'              `._.`
    

  There are several other types of mill types that might be supported by CAM applications, including conical mills (aka v-mills, vee mills) used primarily for engraving and jewelry work; disk (face) cutters used for stone work and slotting; tapered end mills that offer improved strength (but cannot be used to navigate narrow slots and holes); burrs for finishing sharp edges; reamers and drills for hole cutting; and more. Neither of these are particularly useful in all-around work, and so having software support for them is of little or no significance.
  Tip shape aside, once an end mill is mounted in the machine, the whole setup may actually look this way:

         |      |  <- tool holder
         |______|
           |  |                                           -.
           |  |    <- end mill shank                       |
  	 |  |                                            |
  	 \  /    <- shoulder angle                       |
  	  ||                         -.                  |- tool extension length
            ||     <- end mill neck     |                  |
  	  ||                          |-- reach length   |
            ::     <- cutting length    |                  |
  	  ::                         -'                 -'
  

  For many operations, no consideration needs to be given to the overall geometry of this assembly, operator error detection aside: if you account for cutter length properly when designing the part, there is no need for the software to keep track of it. If you do not, and non-cutting parts could end up colliding with the workpiece, you will be forced to redo the design anyway, regardless of whether the software stopped the cutter from wrecking your workpiece or not.
  Having a method to quickly check for problems is useful, to be sure, but in my experience, mistakes are not common - plus, there are many other considerations for which the program will not account for (such as collisions with workpiece holders, mill table, rotary axis unit, Z0 sensor, loose hand tools, detached workpieces, vacuum cleaner heads, etc), so you still need to be careful.
  That said, there are some operations where spindle assembly collision detection is indispensable - particularly if the program has only limited abilities to define freehand cutting regions and the like. In cutting processes that use multiple tools, some of which may be longer and some maybe shorter, it is very helpful to be able to avoid collisions without having to define operation boundaries very carefully by hand, and without having to design all of the part for the shortest cutter used.
  Because of this, it is nice if the program permits a precise shape of the cutter and tool holder to be programmed in, and is capable of carrying out collision detection to automatically stay clear of trouble; however, for this to be meaningful, you should be able to define all of the following:
  • Cutting length,
  • Total tool reach length,
  • Total tool extension length,
  • Shank diameter,
  • Shoulder angle,
  • Tool holder diameter.
  In many cases, it's the collision with tool shoulder that helps the most; a program that detects tool holder collisions only is of less value here, so read the fine print.
• Cutting direction control:
  Although of lesser significance when working in easily machinable prototyping plastics, the ability to carefully control how the flutes cut the material may noticeably reduce stress and vibration, and improve surface finish, in light metals, rigid plastics, and other more demanding stock.
  In conventional (upcut) mode, CAM software tries to generate toolpaths in a way that ensures that the material comes to contact with the cutter on the left side, relative to cutter motion direction (assuming the tool rotates clockwise). This causes uncut material in front of the cutter to be removed more gradually and smoothly, which requires less machine or tool rigidity - but has certain undesirable side effects, most notably increased friction and worse surface finish.
  In climb (downcut) mode, the software would aim to get the workpiece contact the cutter on the right - which is more demanding, as cutting begins more abruptly - but if the material, tool, and mill are rigid enough, affords higher machining speeds and better finish quality.
  Lastly, in meander (bidirectional) mode, the software simply generates toolpaths with as little air travel as possible, paying no attention to cutting direction. This is sometimes faster, but also more likely to leave subtle tool marks because of the differences in how the tool interacts with the surface depending on its motion pattern.
• Workpiece approach method:
  By default, many programs will generate toolpaths that simply have the tool plunge vertically to begin machining a new region. Unfortunately, in contrast to drills, end mills are generally far more efficient cutting sideways than moving down, so the initial impact, if done at the speed that is normally safe while the cutter is moving sideways, may break the tool, or just leave an unnecessary tool mark.
  There are two basic ways to compensate for potential problems; one is to let the user specify a reduced approach feed rate for these entry operations, so that the approach is more gradua; another is to make the entry at a certain angle, over an area that is already cut (or will be shortly removed). It's good for the software to support at least one of these methods.
• Cutting visualization:
  Most CAM programs contain some tools to visualize the result of a cutting process, and quickly check it for uncut areas and other errors. This is an extremely useful feature that saves time and money, and sometimes alerts you to geometry errors, software bugs in CAM software, or just typos made when setting up cutting processes. It can be made even more useful in three important ways:
  • Having the ability not only to see the final result, but also to see the cutting process in time lapse mode, makes it possible to spot problems such as excessive chip loads, incorrect ordering of cutting methods, etc.
  • The ability to color-code areas where the difference between final machined shape, and the input mesh, exceeds a certain threshold, is great for spotting small uncut regions you forgot to run through with a smaller tool, or areas where incorrectly entered process parameters resulted in the cutter plunging through an important feature.
  • The ability to run simulations not only for planar three-axis cutting, but also for rotary axis processes, is very valuable. This requires the program to use more sophisticated logic (such as mesh models instead of 2D voxel depth maps), and hence is not included in some of the lower-end applications.
• Industry-standard input format support: Make sure that the program will be capable of importing data produced by your CAD applications. Most common formats include IGES and STEP (vendor-agnostic), DXF and DWF (AutoCAD), 3DM (Rhinoceros), and STL (3D Systems, open standard). Most of them support complex non-linear 3D primitives such as arcs or NURBS objects that enable complex shapes to be reproduced very faithfully; but STL and some less common formats are limited to triangulated mesh data only.
  Internally, almost all programs convert input data to meshes upon load anyway.

6. Buying cutters

                .----------------------- total length ------------------------.
                |                                                             |
                                                               .- cut length -|
                                                               |              |
            _    __________________________
           |    |                          \                       
     shank |    |                           --------------------~~~~~~~~~~~~~~.    -.
  diameter |    |                                                \ \ \ \ \ \ \|     | cutting diameter
           |    |                           --------------------~~~~~~~~~~~~~~'    -'
           |_   |__________________________/                                  `------ n flutes (helix)
                                                                                      angle specified
                                            |                                 |
                                            `--------- reach length ----------'

• Material. End mills are typically made either from very strong cobalt steel alloys (denoted as HSS), or from tungsten carbide in a cobalt lattice. The latter option is considerably harder and more wear-resistant - and today, priced only about 25% higher than HSS cutters in the ranges of interest to benchtop manufacturing. The downside of carbide is that it is more brittle than high-speed steels - so if dropped or subjected to excessive loads, it will fail more spectacularly. Still, since carbide offers longer tool life and often better surface finish quality, it's pretty much a no brainer - shop for carbide.
• Coatings. To reduce surface friction and further improve hardness of carbide cutters, their surface may be additionally coated with thin layers of ceramic materials. The most common contemporary coating is titanium aluminium nitride (TiAlN, aka AlTiN), which gives the tool a dark bluish-gray hue, and enables machining speeds up to 15-20% higher than with uncoated tools - usually at a price premium of under 5%. Some of the other, less common coatings of slightly different properties are titanium nitride (TiN), titanium carbon nitride (TiCN), or chromium nitride (CrN). It is a good idea to go for coated tools where possible, as there is pretty much no downside.
  A special class of remarkably expensive tools uses diamond coatings, which is said to improve tool life very dramatically - but these come at ridiculous premiums (500% - 1000%).
• Tip geometry. As discussed earlier, the tools of use in precision workflows are flat tip, ball end, and corner radius cutters, preferably with no taper nor other inventions of this sort (taper improves tools rigidity, but is of limited use in complex molds). Flat tip tools are definitely of most use for mechanical parts.

           SQUARE                       BALL                   CORNER  RADIUS                  CORNER RADIUS DETAIL
      
        :          :                :          :                :          :                |                        |
        :          :                :          :                :          :                |     corner             |  Flat section diameter
        |          |                |          |                |          |                |.... radius             |  = tool diameter - 2 *
        |          |                |          |                |          |                 \  :                   /     corner radius
        |          |                 \        /                 |          |                  `-:________________.-'
        |__________|                  `-.__.-'                  `.________.'                   
  

• Cutting diameter. End mills come in cutting diameters from 0.05 mm (thinner than human hair) to 50 mm and more. For most intents and purposes, cutters below 0.4 mm or so are just not very practical, and over 8-10 mm, too big to mount and meaningfully use on benchtop mills. I use 1 mm cutter for various types of finishing work (fine detail, screw holes, etc); a robust 3 mm cutter for all general-purpose roughing, contouring, and finishing of molds; and a nearly indestructible 6 mm cutter for quick roughing of large-volume parts.
• Shank diameter. Diameter of the tool on the side it is mounted in the holder. On almost all the cutters in ranges of use to the tasks discussed in this guide, this is either 3, 4, or 6 mm (see notes on CNC mill selection criteria). Most cutters have perfectly cylindrical shanks, but some may be shaped to accommodate various additional slippage-minimizing grip devices. For benchtop milling with small to medium diameter tools, such features do not matter a lot, though.
• Cutting height. The height of cutting blades determines how far the tool may be driven into material in Z axis in a single pass. Most smaller tools are typically driven less than their diameter into the workpiece in roughing processes, as going much further may exert too much force on the cutter, and would require speed and chip loads to be reduced. Since the part with a blade is structurally weaker than the remainder of the shaft, keeping it only as long as necessary in fragile small diameter tools makes sense.
• Reach length. This parameter describes the height of the cutter through which the tool does not exceed its nominal cutting diameter (and hence may be safely driven down narrow, vertical holes or slots).

         ___________________                                
        |                   `---------------~~~~~~~~~.      This tool has a long neck that makes it capable of reaching deeper than the
        |                                   \ \ \ \ \|      height of the cutting blade itself.
        |___________________.---------------~~~~~~~~~'                                 
              
      
        .-----------------------------------~~~~~~~~~.      This tool has a cutting diameter identical to shank diameter, making it capable
        |                                   \ \ \ \ \|      of reaching to a depth equal to how far the tool extends from the holder.
        `-----------------------------------~~~~~~~~~'                                 
  
         __________________________________
        |                                  `~~~~~~~~~.      
        |                                   \ \ \ \ \|      This tool has no neck, making its maximum reach equal to cutting height.
        |__________________________________.~~~~~~~~~'                                 
  

  Since much of your work may involve machining deep, straight pockets or vertical part sides, it is important to get tools of sufficient reach. In my experience, for all-around cutters (3 and 6 mm), at least 30-60 mm is strongly desirable (less if you have a mill with a small Z movement range, of course).
  For tools used to drill fine holes and other detail, at least 10-15 mm is appropriate. Be careful, though - note that small-diameter tools become wobbly and prone to breakage if made too long. Reach of 10-15 mm on a 1 mm tool is acceptable; 30 mm, on the other hand, requires the tool to be operated extremely slowly to prevent breakage.
• Total length. Self explanatory. With small precision cutters, the only considerations here are that if your mill has a limited Z clearance, you obviously do not want to buy a tool that would further restrict it by sticking out too far and obstructing work area when spindle is fully retracted in Z axis; and consequently, if you have a generous Z clearance, you might want to keep the collet at a comfortable distance from the workpiece, rather than sticking with stub length tools.
  Typical mold milling cutters have lengths of 63 to 75 mm; variants of 80, 100, 150 mm, or more, are also available, if more pricey, but you probably don't have to go there. Shortest cutters still useful for working on top surfaces or in gently curved valleys are about 38 mm long - but that means only about 2 cm between tool tip and a large and unwieldy collet assembly.
• Number of flutes. The fewer flutes, the less cutting action can be carried out per every tool rotation, and the more coarse surface finish will be. The more flutes, the more cutting takes place - but the fewer space is available to evacuate removed material. When working with easily machinable prototyping materials, 4 flutes are often optimal (where available). For very hard materials that need to be machined slowly (and hence produce a very low volume of swarf), up to 8 flutes for all operations make sense. When machining aluminium, elastomers, wood, or other materials that produce lots of pronounced, large-size swarf, two or three flutes might be more suitable for best performance (and some single flute tools are available, too!).
• Helix angle. Here, zero degrees means perfectly straight, vertical flutes. Low angles (0-15 degrees) evacuate material fast, which is important for aluminium and elastomers, and sometimes produce better surface finish in malleable materials. Medium angles (30-50 degrees) reduce cutter stress and vibration, sometimes enabling it to work faster and achieve better surface finish in general-purpose uses. High angles (60-70 degrees) are for gentle machining of very tough materials.
• Center cutting ability. Most of the cutters in the ranges you will be dealing with have bottom blades that extend through the entire mill diameter, and are capable of competently (though slowly) removing material if driven straight down into a workpiece (for example, to drill a hole). That said, some models, particularly miniature cutters (below 0.2 mm or so), may not have a geometry that permits this type of cutting, and need to enter the workpiece at an angle. These will be almost certainly clearly denoted as "not center cutting" in catalogs. This limits their applications, and requires careful software compensation to minimize the risk of tool damage - so avoid such tools unless absolutely necessary.
  (Confusion exists on some CNC forums on whether some types of mills - for example with a particular number of flutes - are always / never center cutting. There is no such dependency in contemporary end mills, however.)

• Hanita - die mold range carbide cutters
• Hanita - generic carbide finishers
• Nachreiner - carbide finishers
• Cobra Carbide - full end mill range
• Harvey Tool - all tools

• Hanita 7N2201021 - square, 2 flutes, TiAlN, diameter 1 mm, reach 10 mm, length 50 mm
  Price: $35
  This is a good, albeit somewhat pricey, cutter for low-speed, precision machining of small features, such as gearwheel teeth, very small screw holes (1.2 - 1.8 diameters needed for 1.5 and 2 mm screws), slots, motor shaft mounting holes (1 - 2 mm), and other fine detail. Reach length of 10 mm is very generous, but small enough to make the cutter reasonably easy to handle. It does not tolerate experimentation well, though - play by the book or lose $35. Variants with 8, 12, and 15 mm reach lengths are also available.
• Cobra Carbide 24452 - square, 4 flutes, TiAlN, diameter 3 mm, reach 75 mm, length 75 mm
  Price: $15
  The baseline cutter for almost all operations in my workshop. A great and robust choice for roughing and finishing in prototyping materials. It can take a lot of abuse, can reach real deep, and removes material efficiently. You can get two, as you will likely be tempted to experiment a bit - and breaking this one won't be too painful.
• Cobra Carbide 25454 - ball, 4 flutes, TiAlN, diameter 3 mm, reach 75 mm, length 75 mm
  Price: $23
  Good all-around cutter for working on organic shapes, or as a roughing end mill for molds where using a square cutter all the way through results in undesirable tool marks.

• Hanita 7N2202041 - square, 2 flutes, TiAlN, diameter 2 mm, reach 30 mm, length 75 mm
  Price: $35
  A great cutter for medium precision work, including 3 - 4 mm screw holes, many slotting operations, etc. It is too large for making small gearwheels and so forth, but can be operated much faster, and cut much deeper, than 1 mm cutters. You might also want to consider 1.8 mm diamter, 20 mm reach cutter instead.
• Hanita 463200400RT - square, 2 flutes, TiAlN, diameter 0.4 mm, reach 1.5 mm, length 38 mm
  Price: $15
  A low-cost cutter for (shallow) machining of very fine features and routing in relatively thin material. Might be of limited use in medium scale projects, but it will give you a good opportunity to explore the fine machining capabilities of your unit, and to experiment with very small designs powered by pager motors, etc. Note that unless you play by the book, you are almost guaranteed to break it right away. There are also 0.4 mm diameter, 4 mm reach; or 0.6 mm diameter, 4 or 6 mm reach cutters available from Hanita, more expensive but more useful for more complex gears.
• Cobra Carbide 24540 - square, 4 flutes, TiAlN, diameter 6 mm, reach 75 mm, length 75 mm
  Price: $28
  A good roughing cutter for large size, large volume objects, assuming you want to do this type of work. This cutter is nearly impossible to destroy when working in prototyping materials on medium power mills, so buying one is enough.
• Cobra Carbide 25534 - ball, 4 flutes, TiAlN, diameter 6 mm, reach 75 mm, length 75 mm
  Price: $28
  Good roughing and finishing cutter for medium to large volume objects with curved, organic shapes. Like the previous one, nearly indestructible.

7. Getting stock material

One reseller I had an excellent experience with is Walco Materials; another web-friendly place is Freeman Manufacturing and Supply. In Poland, Milar is a good source, and in the United Kingdom, Mould Life might be a decent pick, too.

          .- silicone poured in                                                    .- polyurethane poured in
          |                                                                        |
  __      v     __                         __________                              v
  ::|    __    |::                        |@@@@@@@@@@|         rotate          ___    ___                          ___
  ::|___/##|___|::           ->           |@@@/  |@@@|           ->           |@@@|__/@@@|           ->           |##/  
  ::::::::::::::::                                                            |@@@@@@@@@@|
      
  Positive form (PU)               Negative cast (silicone)                                                       Final part (PU)

7.1. Silicone rubbers for molds

• Condensation cure compositions use organic tin compounds as catalysts (commonly dibutyltin dilaurate) to trigger polymerization. These are the most common, affordable mold making silicones, although their performance might be lower than that of the addition cure resins, viscosity higher, and they might have limited shelf life once polymerized.
  Organotin compounds are also somewhat irritant or harmful if misused, and have a slight but not particularly pleasant smell (sometimes likened to stale beer). In the quantities used, nothing to lose sleep over, as long as you follow certain basic rules laid out later.
• Addition cure RTV-2 rubbers use platinum catalysts to trigger polymerization. Cure rate may be accelerated or decelerated quite spectacularly by varying temperature within a convenient range.
  These compositions generally enjoy superior physical characteristics, including much lower shrinkage while curing and longer shelf life of the polymerized product, and their components are non-toxic and odor-free - but they are more sensitive to contaminants and incompatible chemicals until cured. So for example, if you want to dye them, you will likely have to use non-reactive, powdered organic pigments (see later).

• Tensile strength and elongation at break:
  When it comes to mold making, you want the rubber to be flexible enough to permit parts to be removed effortlessly - but it must be stiff enough to maintain dimensional accuracy when laid on a surface and filled with a liquid resin.
  There are several indicators of stiffness, not all of which are consistently advertised by all manufacturers; a ballpark estimate, however, may be trivially derived from specified tensile strength and elongation at break values.
  Tensile strength by itself documents how much force per surface area needs to be applied to a standardized test specimen to break it apart. My recommendation is to look for silicones with tensile strength of at least 3-4 MPa, as it ensures good all-around properties for cast parts - but the parameter itself is of secondary significance (tear strength is more important).
  NOTE: pounds per square inch are used in some datasheets instead of Pascals; 1 MPa is approximately 145 psi.
  The other relevant parameter, elongation at break, denotes deformation at maximum tensile strain, usually expressed as a percent (0% is no elongation). You can convert it to a ratio (1 + elongation at break / 100%). If this ratio is then divided by tensile strength, you will achieve a semi-decent indicator of how strongly the rubber deforms in proportion to force applied.
  I advise shopping for silicones where this factor is in the 0.8-1.4 range. Rubbers that get closer to 2 are still usable, but might require more careful mold engineering, and make it harder to pull out certain feats, such as having a protrusion 15 mm tall and 1 mm in diameter or so.
• Hardness:
  This parameter is closely tied to stiffness in silicone rubbers, and is often the primary key by which they are sorted in catalogs.
  Free-flowing silicone resins typically come with hardness between 3 and 50 Shore A for the cured rubber (Shore A is a durometer scale devised for elastomers); some harder compositions exist, but they become progressively less useful for general applications due to worse physical characteristics and very high viscosity. On the lower end of this spectrum, you have chewing-gum-like substances used primarily for special effects, low load shock absorbing pads, and the like (they may even use Shore 00 scale, meant for gels); they almost certainly fail the aforementioned stiffness criteria, however.
  On the upper end, you will find fairly stiff rubbers ideal for the job, and also great for making softer tires, transmission belts, and similar elastic components. In theory, anything around 20-25 Shore A should be already stiff enough for basic mold making uses - but I would strongly recommend looking into the 40-45 Shore A range if possible, because they offer far superior form dimension stability for precision uses.
  Also note that hardness and stiffness of polyaddition resins can be trivially lowered by mixing in some silicone oil (though at the expense of tensile strength), so if you select a very hard composition, you will be still able to produce soft parts where needed; going the other way round is not trivial.
• Tear strength:
  This parameter is a close relative of tensile strength, but describes the force needed to break a standardized specimen by applying a non-uniform stress to the material, simulating the process of tearing it at the edge. It is a better indicator of mold and part longevity than tensile strength itself, as the force needed to break the mold this way is much lower, and more likely to be accidentally applied.
  Roughly speaking, silicone resins can be divided into "low tear" and "high tear" varieties, denoting how well they respond to this type of abuse. "Low tear" compositions are somewhat cheaper (10-20%), but molds made from them often barely last enough to make several parts in a row, and they are not a very good material for tires and other final components that are expected to withstand abuse.
  For moldmaking rubbers in the 20-40 Shore A range, tear strength of 10-30 kN/m is desirable (10 kN/m = 57 ppi, pounds per inch).
• Catalyzed viscosity:
  Flow and wetting characteristics of liquids are difficult to parametrize, yet very important for casting resins - preparations that flow poorly, wet molds weakly, and exhibit high surface tension, will be significantly more difficult to work with, as they would trap air in tight spots, and prevent bubbles from rising to the surface and bursting fast enough. Dynamic viscosity is usually the only parameter given by manufacturers, and within a specific composition, it usually gives you a good picture of how the resin will perform - but it does not really enable you to compare the ease of use across vastly different types of resins.
  With silicones, you can generally expect viscosities of 10,000 mPa*s to be very easy to work with, and largely self-degassing. Viscosities closer to 20,000 or 30,000 will require careful application, possibly with a brush, to reproduce intricate details (or a quick round of in-mold degassing with a vacuum pump). Compositions getting closer to 50,000 - 100,000 end of the spectrum will require vacuum, or a lot of really mundane and error-prone work; they make sense only if the composition is transparent, and allows you to spot and suck out bubbles as needed.
• Pot life (also known as gel time):
  Mixed resin begins gradually polymerizing right away, which results in increasing viscosity. The effect becomes noticeable around 25% through the "pot life" period; and by the time 100% is reached, the resin is no longer free-flowing and self-leveling, which makes it nearly impossible to work with.
  For silicones, something around 25 minutes might be OK for low viscosity compositions; closer to 40-90 minutes is much better when dealing with high viscosity variants. In most cases, cure can be easily accelerated if needed, but it's more difficult to slow down - so stay on the safe side.
• Shrinkage:
  As the molecular structure of the resin changes when curing, some changes in volume may occur, most commonly resulting in a slight reduction in overall volume of the substance. Typically, prototyping resins are carefully engineered to minimize this effect, and so you are unlikely to have to worry - but it does not hurt to check.
  Different manufacturers seem to have different testing protocols, but the most common methodology test a specimen of 500 x 50 x 5 mm or so, and examine reduction in length expressing it either as a percentage, or as mm/m ratio (multiply this by ten to get percentage, obviously). Some companies tend to cheat a bit (for example, using 4 mm specimens instead, which results in 20% lower apparent shrinkage, as the dependency is roughly linear) - so be wary.
  In general, condensation cure silicones should exhibit shrinkage under 0.6%; addition cure compositions are usually under 0.1%. Try to stick to products that meet or exceed this criteria.
• Color:
  This property is important if you intend to make anything other than molds from the resin itself (e.g., rubber tires). Some casting resins are white, light gray, or beige; a small number of compositions might be translucent - and all these varieties are easy to pigment any way you please.
  In contrast, compositions that are readily dyed blue, red, or other such color (as an indicator of proper mixing) obviously don't give much freedom in this department. You can still pigment them black, but that's about it. I would not consider color a matter of top importance, as polyurethane elastomers are a cheaper way to make rubber-like parts anyway - but something to be aware of.

Non-vacuum users in the States may consider Quantum Silicones QM 237, which seems to be a rough equivalent of MM 242; as mentioned, I wasn't able to actually order it in a timely manner and a sane quantity, but you might have more luck. Other than QM 237, I am yet to find a comparable alternative on this market. Polytek PlatSil 71-40 is a transparent resin with more than twice the viscosity, but may still work well - never tried it. Rhodia / Bluestar V-340 + CA-45 or CA-55 catalysts (44 and 45 Shore A, respectively) is a comparable opaque formulation, too.

7.2. Polyurethane elastomers

7.3. Rigid polyurethanes

• Stated purpose:
  Amusingly, as with polyurethane elastomers, this is often more telling than any measurements. If the manufacturer hints that the product is meant to simulate engineering plastics (e.g., ABS) for prototype parts - great. If they instead talk about tooling fixtures, scale models, and so forth - be wary, as the resin might have some shortcomings that are not evident when looking at cold, hard numbers alone. For example, flexural strength in thin sections may be disproportionately low.
  For the same reason, try to stay clear of resins designated for vacuum casting. On paper, they may seem no different from other compositions, and may have very favorable specs - but they are very demanding in terms of mixing, degassing, cure profiles, etc. I tried vacuum grade resins from three different manufacturers, and all of them were extremely annoying to work with.
• Pot life:
  Many polyurethane resins react very quickly - pot life between 30 seconds and 4 minutes is pretty common. This may sound like enough if you just want to pour the resin in a simple mold - but if you need to degas it after mixing, then apply it with a syringe to a narrow, multi-part mold - you will run out of time. Quickly curing resins are also more exothermic, limiting the maximum size of a casting you can make without overheating the system.
  To complicate your life a bit, while silicone have a mostly linear cure profile, polyurethanes are less predictable: some may exhibit a long period of low viscosity followed by a snap cure, while others may get difficult to handle 3/4 through the nominal pot life. All in all, I strongly recommend opting for systems that give you at least around 7-8 minutes to wrap things up; 10-20 minutes is even better. Don't overdo it, though: while there are some resins with pot lives of 40, 60, or even 100 minutes, they take forever to fully cure, and tend to be annoyingly sensitive to moisture.
  The total curing time before the part is ready to be used might sound discouraging on these slower compositions (8-24 hours to demold, plus several days to reach final properties) - but once the resin is reasonably solid, further polymerization can be safely accelerated by carefully increasing temperature. There are two challenges involved: firstly, if the resin is still liquid, excess heat may increase cure-related shrinkage, or cause it to develop bubbles as gas solubility decreases, or some components of the mix begin to boil - so the system should be allowed to polymerize most of the way through before you crank up the heat. The other problem is the phenomenon of thermal expansion - the volume of matter increases with temperature, with very few exotic exceptions. The usual rate for silicones and polyurethanes is around 0.005 to 0.02% per ° C in each axis. In complex molds with thin sections, this expansion may introduce stresses that could permanently stretch or otherwise warp the part. For this reason, you should remove the part from its mold when going over 50° C or so; but you can't do this until the part is cured well enough to withstand this operation.
  In the end, you will need to work out the right cure schedule depending on the resin you're using; for IE-3075, one of the resins I use, 60 minutes at room temperature, followed by 30 minutes at 30° C and 30 minutes at 40° C, seemed to work well for initial cure. The mold should be then allowed to cool down (putting it on a fan speeds this up significantly; cold water is not recommended, as it causes a very rapid and uneven cooling, which may contribute to warping). Once this is done, the part can be removed and put in an oven at 60° C for 30 minutes, 80° C for another 30, and 100° C for the last 2-4 hours, to achieve good working properties.
• Mixed viscosity:
  You can find systems with viscosities starting in the 50-100 mPa*s range, but there might be some strings attached (such as higher shrinkage or reactivity). Resins around 500-1500 mPa*s are common, and very convenient to work with. I do not recommend going much past 3000 mPa*s or so; polyurethanes behave somewhat differently than silicones, so, say, 9000 mPa*s is pretty challenging to degas cast right in the comparatively shorter timeframe allotted.
• Hardness:
  For rigid plastics, a different scale, Shore D, is commonly used. Typical polyurethane resins span between 50 and 90 Shore D; I recommend sticking to the 80-90 range, as this usually maps to excellent rigidity, dimensional stability, and scratch resistance. Plastics in this range are also usually fairly easy to polish to luster.
  (For the curious: a basic comparison of durometer scales with example materials can be seen here.)
• Presence of fillers:
  The addition of plastic, mineral, or metal fillers may improve hardness, thermal conductivity, or abrasion resistance of the resin - and also reduce shrinkage. On the other hand, heavily filled systems may be difficult to degas and pigment, and may be more brittle in thin sections; shorter shelf life due to the filler settling down is also common. Because of this, your primary resin probably should not be a filled one - but there are some uses where it could come handy.
• Flexural strength and flexural modulus:
  Flexural strength describes the bending force (per surface area) at which a standardized test specimen yields - the lower it is, the more fragile the material. Flexural modulus describes the ratio of the bending force to the degree of elastic deformation observed in a test specimen. The higher it is in proportion to flexural strength, the stiffer the part. A component with a high flexural modulus but low flexural strength is rigid, but brittle (a good example is a biscuit); a component with low flexural modulus and high flexural strength will not break easily, but will deform significantly under load (think plastic PETG bottles).
  Polyurethane resins vary greatly here, depending on the intended purpose and the additives used. Flexural strength usually varies between 40 and 120 MPa. Aim high: resins under 80 MPa may be fragile enough to be a major limiting factor in some designs; and given the choice, there is little or no reason to settle for less than 100 (which is roughly equivalent to acrylic glass / polycarbonate - try breaking a CD to get a picture).
  When it comes to flexural modulus, chances are, you want your rigid components to, well, stay rigid. Acrylic glass, ABS, and polycarbonate all are in the 2.5-3 GPa range here; polyurethanes that go into the 2.7-3.1 GPa zone are available and worth exploring. Filled systems that go up to 5-6 GPa are not unheard of, too!
  (For the curious: you can see some data on known polymers and composites on this page.)
  Both flexural strength and modulus can be further greatly improved by constructing composite materials using glass or carbon fibers (or shredded strands), as discussed briefly in later sections. It is a time-consuming process, though.
• Shrinkage:
  Optimally under 0.1%, with the same gotchas as described for silicone rubber.
• Color:
  There is absolutely no reason to buy a polyurethane resin that cannot be colored. All white or slightly off-white (light beige or gray, translucent amber) resins with no mineral fillers should respond very well to appropriate pigments. You most certainly want to have this liberty - so avoid resins that are already "helpfully" dyed blue, black, or somesuch; as well as ones with strong natural hues. If the manufacturer says the resin has a color such as "caramel", "cognac", or "tan", inquire to see how easy it would be to mask this color for your applications, as it's impossible to tell from the description alone.

In Europe, the situation is different - although Innovative Polymers do not have any resellers that I know of, Huntsman's RenPIM 5219, also sold as RenCast 5146, is pretty close to the IPI product, and fairly cheap. It has a hardness around 85 Shore D, ~100 MPa flexural strength, modulus of ~3 GPa, 20 minutes pot life, and a neutral white color that is easy to dye as desired.

RenCast 6491, 5146 / 5219, as well as IPI IE-3075, are all very moisture-sensitive, and benefit from the addition of molecular sieves to the mix (see later for an explanation what it is, and where to find some). Their cure profile is otherwise excellent: great surface cure, long low-viscosity stage, good tolerance for heat-accelerated curing.

One of the aforementioned products should fulfill most of your part manufacturing needs, but there are several other resins worthy a mention. For example, you might want to have a look at Inovative Polymers TP-4020. It is roughly comparable to IE-3075 on many counts, if a bit more brittle (flexural stength 110 MPa) - but because of a composite filler, it has an impressive flexural modulus of around 6 GPa, resulting in unmatched rigidity; and because of a polyurea chemistry, also a very high temperature resistance. It is especially useful for lightweight frames, precision gears, hot-running motor mounts, and so forth. It costs about $40 per liter, so it's not exactly cheap in comparison, though.

Other resins worth exploring are transparent compositions; they cost up to twice as much as their translucent or opaque friends, but they have an array of obvious, decorative or quasi-decorative applications; nothing beats a transparent gearbox. My favorite transparent resin is Innovative Polymers OC-7086 - excellent mechanical properties (80 Shore D, 110 MPa flexural strength, 2.4 GPa flex mod), nice cure profile, no mercury or other surprises, about $30 a gallon, slightly under 20 minutes pot life. Axson PC 521 is comparable, but mercury-catalyzed.

7.4. Alternative plastics: epoxies and polyesters

7.5. Adding colors to resins

• It is easy to get dull earthy colors from more vivid pigments by mixing them together or adding neutral opacifiers, but it is not possible to make dull colors more punchy. So begin by completing a competent palette of brilliant, vivid colors before moving on to more artsy, exotic choices. The aforementioned studio colors are a good starting point.
• It is harder than it may sound to accurately mix high-yield pigments in small quantities to get exactly the color you want. One of the most notoriously difficult tasks is getting a decent shade of ruby or violet by mixing red and blue. In addition, some pigments can't be mixed at all: fluorescent dyes exhibit an interesting quenching effect, where they lose their special properties if used in excess or mixed with each other. So, don't try to save a buck by buying just sorta-primary red, blue, and yellow dyes, and assuming you would be able to get every other color easily.
• Most synthetic, organic pigments are safe to use in plastic resins (e.g., phthalocyanate, thraquinone, pyrrole, perylene, alizarin, dioxazine based ones). This also includes metals and other chemicals encapsulated in acrylic or phenolic resins. Some organic dyes will dissolve or form translucent dispersions in transparent resins and give beautiful, rich colors. Some will make the resin opaque.
• Many natural organic pigments made from plants or animals are a gamble - and not to mention, they are pretty expensive. I would advise against going there.
• Some inorganic pigments are safe, particularly the ones based on silica and zirconium silicate doped with various metals, but a good number of them is not. Titanium white and most iron oxide pigments are OK; Prussian blue is not. Powdered metals are a risk, as some of them may prematurely catalyze the resin.
• Be wary of pigments that are either known to be outright harmful (check labels), or are based on particularly toxic / bioaccumulating compounds; the latter, even if not soluble or harmful in their original state, may eventually break down and migrate if exposed to UV radiation or certain chemicals, so it's best not to take risks. More about this in the "health" section later on.
• Stay away from any ready-made dilutions, pastes, liquid suspensions, etc, unless specifically recommended for polyurethanes - or be prepared for some nasty surprises.
• Read descriptions for any unusual properties, such as limited compatibility with carrier media, poor lightfastedness, the need for wetting agents, etc. Be alert.

• Fluorescent palette, for the road-sign-grade punchy colors:
  Kremer 56150 (lemon yellow), 56100 (light green), 56400 (pink), 56300 (orange-red) is a good basic set; additionally, you might want to consider 56450 (violet), 56050 (light blue), 56200 (golden yellow). Pigment 56000 (white) is an interesting animal, too: it's actually translucent and not masking, but it acts as a powerful optical whitener (and adds the ability to glow in UV light) when added to some other colors.
  All these cost about $20 (EUR 10) for 100 g, an amount that should last for years, and are very easy to uniformly disperse in a resin.
  Note that there is no "true" daylight fluorescent blue pigment. Most manufacturers produce it by adding optical whiteners and UV-fluorescent components, resulting in worse performance in daylight compared to yellow or red. Blue, violet, and green pigments may be hence less spectacular under a tungsten lamp on your desk - they will, however, light up nicely in UV or blue light. A must-have for party robots!
• Studio colors, for the low-cost general, crisp "plastic grade" hues:
  Kremer 55100 (lemon yellow), 55300 (dark red), 55500 (light blue), 55600 (dark blue), 55700 (light green), 55900 (violet), plus maybe 55470 (pink) or 55450 (bordeaux) depending on your personal taste. These generally cost $10 for 100 g (EUR 2-3), with comparable efficiency. They are less striking than fluorescent pigments (exactly the shades you see in most plastic products around you), and do not glow in UV - but a wider palette and darker colors are available. Very easy to disperse.
• Opacifiers:
  Kremer 46200 (titanium white), 48400 (Mars black). Both have an insanely high yield, so you will never ever run out of them if you get 100 g or so ($10 / EUR 4). You can use to control the brightness and opacity of a resin ("natural" opaque polyurethane is translucent in thin sections), to mask out some pigmenting mistakes, and to easily achieve shades of gray. Note that they will be fairly taxing to disperse uniformly in a resin (2-5 minutes of mixing might be required).

• Assorted individual pigments:
  There are several interesting organic and inorganic pigments that nicely complement these basic palettes. One example is DPP red (Kremer 23180, CI pigment red 254), a deep and pretty hue of red historically used for Ferrari cars, and really striking when polished to luster; another is alizarin violet (23750, CI pigment violet 5), a rich cherry-like shade and forms transparent solutions in polyurethanes. Dioxazine violet (23451, pigment violet 37) is a very dark, steel shade, also transparent; and anthraquinone blue (23100, pigment blue 60) is an amazing steel blue. As a general rule, they will be tricky to disperse, and sometimes, making a paste with a minimal quantity of a resin or a surfactant might be the only way to properly wet them.
  If you shop for these, it's best to read up on standard pigment naming schemes ("Color Index") to avoid duplication, and to look up additional samples on the web. Some useful notes on various pigments may be found at this site. It deals primarily with watercolor painting, but has an extensive database of pigment descriptions, CI numbers, and so forth.
• Transparent dyes:
  Only if you intend to make clear polyurethane castings. If you want to experiment, grab Kremer 94400 (ORASOL lemon yellow), 94402 (ORASOL gold yellow), 94406 (ORASOL orange), 94412 (ORASOL red), and 94414 (ORASOL blue). You will pay about $10 (EUR 4) for 5 g, but this amount should suffice for several bucketfuls of a resin. Note that these dyes are pretty challenging to disperse uniformly, and 5-10 minutes of constant mixing might be required.
  Alternatively, you can buy, say, Eager Plastics EP7701: scarlet red, violet, royal blue, lime green, bright orange, lemon yellow (should be $5 a pop, with a slightly lower yield) - but as noted, these premade solutions will almost certainly not play well with polyaddition silicones.
• Fancy pigments and glitter:
  If you have some money to spare, and are curious: Kremer 56550 is a phosphorescent blue dye, 56650 is red; 50440 is a red-metallic pearlescent pigment that looks pretty stunning, and 50180 is gold. Aluminium-based glitters are 50800 for gold, 50701 for silver; 54850 is fine copper powder (looks rusty at first, but turns shiny metallic once polished); etc. Transparent powdered glass pigments give pretty interesting effects in transparent resins, too. Be sure to check out Paint with Pearl store, too - they have some very interesting pearlescent and metallic pigments for less.
  Note that some of the large-grain glass and metal pigments would inevitably settle in a mixture, given enough time. You can either apply them directly to the sides of a mold, where they would stay; or you would need to wait for the resin to achieve a thicker consistency to prevent this effect (10000-20000 mPa*s).

7.6. Making plastics look like metal

I will not be providing any detailed instructions in this document, but particularly if you are thinking of doing any jewelry work, it might be worth looking into. Electroless silvering is particularly hassle-free, requiring only distilled water, stannous chloride, ammonium hydroxide, sodium hydroxide, silver nitrate, and some glucose. Ready-made nickel, cobalt, and tin kits are available e.g. from Caswell Plating, too - but boy, they can be expensive.

7.7. Creating reinforced plastics

8. A word on solvents

• Isopropyl alcohol: for various cleanups, removal of excess grease, soldering paste, water-soluble contaminants, etc. Safe on almost all plastics, although might reversibly soften some types of polyurethane rubber (use this to restore it to original shape if deformed under strain). Low toxicity in normal use, minimal environmental burden.
  Ethanol is a superior alternative, but a pure variety is considerably more expensive due to extra taxes. In some markets, denaturated alcohol is a good choice; elsewhere, it may contain dyes or aromatizers that make it less useful or nauseating to work with.
• Light naphtha: non-polar solvent for waxes, paints, uncured silicone and other uncured resins, etc. Will reversibly (and often quite grotesquely!) swell silicone rubber. Toxicity will differ depending on contaminants and exact composition (e.g., presence of benzene), but OTC solvent varieties should be only modestly harmful; they do have a more pronounced environmental impact, however.
  Tip: if applied to silicone rubber topically, makes it possible to glue it using cyanoacrylate glues, otherwise an impossible task. Some companies would actually sell you grossly overpriced naptha as a "surface activator" for joining silicone rubbers (I mean, $10 for 20 ml).
  Heavy napthas are a good alternative, but take considerably more time to dry.
• Ethyl acetate: dissolves many plastics, some very quickly (polystyrene, PMMA), some at a slower pace (polyurethane). Useful for preparing plastic-plastic bonds, etc. Reasonably low toxicity and not a major environmental burden. It is also a good tool for speeding up the degassing process for polyurethane resins.
  Acetone and methyl ethyl ketone are good replacements, but attack plastics more aggressively. They also happen to be (sloppily) regulated in some markets as precursors to some drugs, sometimes requiring silly in-store paperwork to purchase them.

9. Release agents

10. Casting tools and other workshop stuff

• Small bow saw with metal and wood blades,
• Scalpels or box cutters,
• Heavy duty scissors,
• Small wire cutters,
• Precision pliers, tweezers, etc,
• A set of precision screwdrivers,
• A set of needle files,
• A decent vise and clamps,
• Some reasonably small hobby drill,
• Good, precision soldering iron (Weller WMRP rocks, if you need inspiration).

• Precision weighing scale:
  Mixing ratios for all casting resins are typically given in parts by weight, rather than parts by volume, simply because weight can be measured more reliably and is not significantly affected by air bubbles, temperature differences, etc. You probably have a kitchen scale already, but for consistent results, it is strongly advisable to get a more precise device. Digital scales with 0.1 or 0.05 g resolution, and capacity up to 1 kg start at about $10-$20 (say, from there), and work surprisingly well for pretty much all the uses imaginable; when you need to measure 1 kg or more, which is pretty unlikely, you can fall back to a kitchen scale.
  I generally do not recommend settling for resolution of less than 0.05 g or so, as less precise devices would limit your ability to work with small quantities of resins. For example, on a scale with 1 g resolution, where you need to mix 10 g of an addition cure silicone resin with 1 g of catalyst, but may easily end up preparing 0.6 g or 1.4 g (both of which will read as "1"), grossly exceeding the specs. This would force you to prepare, say, 50-100 g minimum even for the smallest job, and throw away the excess - so much for the savings.
  Polyaddition silicones and several types of polyurethanes are very sensitive to mix ratios, and will not cure well with as little as 1% off; other resins are more forgiving, but you should aim to never exceed 2% or so.
• Vacuum pump and hoses:
  Casting resins contain small amounts of dissolved atmospheric gases. The solubility of these gases may be significantly reduced as the resin polymerizes, and so the material may develop tiny air bubbles that may reduce its strength and affect surface finish. Furthermore, as a result of pouring and stirring, more viscous formulations may develop suspended air bubbles that take longer to emerge and burst than the resin actually remains liquid; and reactions with moisture in the air may contribute some more.
  A way to address all these problems is a process called degassing, where ambient pressure in a container with the resin is quickly and significantly decreased, as to facilitate a rapid evacuation of all gases from the solution itself; existing bubbles will expand to large sizes and raise rapidly, and dissolved gases will leave the resin likewise. Although many resins give reasonably good results without degassing, I strongly recommend degassing all the components prior to mixing, and doing another quick round with catalyzed resins too, to improve physical properties and mold reproduction without the need for careful application of the resin.
  Atmospheric pressure is about 1.01 bar (101 kPa), and perfect vacuum is 0 bar. Vacuum pumps might have their performance expressed in terms of the maximum pressure differential they are capable of producing (for example -800 mbar, meaning that the usual final vacuum will be about 210 mbar on a typical day), or the ultimate vacuum attainable regardless of fluctuations in air pressure (for example, 50 mbar absolute). Be careful to get this parameter right - the difference between -900 mbar relative and 900 mbar absolute is obviously pretty dramatic; for proper degassing, you will need a pump capable of getting to at least 20 mbar absolute / -990 mbar relative or so, preferably in less than a minute; 5-10 mbar absolute might be better.
  Depending on your budget and space available, you can either get a dual-stage rotary vane vacuum pump, or a two- to four-stage diaphragm one. Rotary vane units are more common, fairly cheap, noisy, and somewhat bulky (well, ever seen a portable compressor?), but they also enjoy higher flow rates in a particular price range. They usually run wet - with mineral oil serving as a gas displacement medium and a lubricant - and need to have the oil replaced every few months. The cost is very low - maybe $5 or so - but there is some hassle involved. The advantage of being protected with an oil is that any resin aerosols or dusts accidentally sucked into the pump are less likely to stick to pump components, and can be drained with the next oil change, however.
  Diaphragm pumps, on the other hand, are small, quiet, maintenance free, and can sometimes deliver a higher vacuum - but are more pricey, slower, and much more sensitive to contamination. They are easier to disassemble and clean if necessary, but it is definitely a more time-consuming and tricky process than just a simple oil change; you might want to consider an appropriate intake filter with the pump to prevent this from happening.
  Rotary vane units are available from multiple sources, as they are commonly used for HVAC servicing. A suitable device should cost around $100 if you look around - for example, VIOT offers a nice VPB1.5 unit for $117; I use their VPD5 unit, which goes for about $170. VIOT folks seem to be nice and ship promptly.
  When it comes to diaphragm pumps, you can expect to pay $200-$300; Gardner Denver Thomas has a nice selection, such as their 70110037 (7011ZVR/2,2/VVN/DC) four-stage unit, which goes to 10 mbar absolute.
  Once you select the pump you want, you also need to get a degassing chamber - a container where a cup with your resin can be placed, and vacuum applied, in a safe manner. You can get one for about $70 from a place such as Ted Pella, along with a suitable wire-reinforced PVC or polyamide hose. You can also try building a makeshift chamber, for example from a medium size, robust mason jar - but be wary of the risk of an implosion; if using glass or rigid plastic, wrap the container in thick cloth or put it in a wooden box. By the way - don't expect "vacuum" food storage containers to be actually capable of withstanding proper vacuum - been there, done that, had to collect the pieces.
  Along with the pump, consider getting a silicone-based vacuum grease, such as Dow Corning 891-7. You can use many other greases to seal joints and lubricate o-rings in vacuum chambers, but these formulations last longer and work better - and because of their high viscosity, non-reactivity, and superb purity, also work very well for non-vacuum applications (e.g., temporarily sealing multi-part molds, resealing resin containers, etc).
• Magnetic stirrer (optional):
  There are some cases where automated resin mixing is a godsend - for example, when dealing with stubborn pigment dispersions that need constant mixing for 10 minutes or so; or when trying to minimize air entrapment. Another less obvious case is when you need to degas a resin quickly and thoroughly - as mixing or otherwise agitating the resin under vacuum greatly facilitates the process by seeding bubble formation, and bursting bubbles that already reached the surface. The trick saves you a total of about 2-3 minutes of resin's pot life - you no longer have to mix it very thoroughly prior to placing it in a vacuum chamber for final degassing, and you need to degas for a shorter time. If you ended up buying any fast-curing resins (pot life under 8 minutes or so), a stirrer can help tremendously.
  One example of a decent stirrer is Hana Instruments HI190M (under $90); get it and a dozen of stir bars (e.g., disposable Teflon 2" ones - they are very much reusable, but you will probably sooner or later lose some by not retrieving them from a resin in time, so better safe than sorry). The device works in a very simple manner - it has a base unit where a rotating magnetic field is produced (either mechanically, by just rotating a magnet attached to a small motor; or electronically, using solid-state electromagnets switched on and off in a sequence); and two small magnets embedded in Teflon or other non-reactive, non-stick plastic that are thrown into the liquid to be mixed, and rotate in sync with the external field - and then can be recovered and wiped clean. This allows the liquid to be stirred in a contact-free manner, including through the bottom of many degassing chambers, in a very simple and inexpensive way.
  If you don't feel like spending another $100+ on this gadget, you can also consider other options: cheap battery-operated hobby paint mixers, such as this one from Badget ($6) could be probably installed inside a vacuum chamber; milk frothers with a modified mixing rod could work too. Finally, a small solenoid or an off-balance motor could be placed inside the chamber to vibrate the container and promote degassing, likewise.
• Temperature controlled oven:
  At room temperature, many resins take between 4 and 24 hours to become safe to (gently!) handle; this is sometimes described as "demolding time" in manufacturers docs - although small and complex parts might be actually still unsafe to remove from molds at that time. Another 2-14 days must pass for the resin to achieve final physical properties. This process can be greatly accelerated by placing the part in a higher temperature: a ballpark estimate for polyurethanes is that every 10° C cuts the curing time in half; polyaddition silicones are accelerated even more.
  So, to get your parts in just a couple of hours, you almost certainly want to be able to experiment with increased temperatures. Unfortunately, the safe range for many resins is very narrow: 60 degrees might be OK; at 90, shrinkage of the resin, or warping due to thermal expansion of the mold, might be excessive; and clear epoxy or polyester resins may begin to develop a yellowish hue. At as little as 150 degrees, some plastics may begin to swell due to residual uncured components - and thermal decomposition takes center stage at 180-200.
  You might have some luck using your regular convection oven, assuming it has basic temperature control (well, baking chemicals where you later prepare food is discouraged), or with one of these cheap tabletop toaster ovens with adjustable temperature (maybe $80 at Wal-Mart) - but it is a bit of a gamble, as many of these devices offer very sloppy temperature control in the range we need the most - 40 to 90°. For example, my convection oven heats up to 90 degrees when set to 60. More importantly, these devices are really not designed to run for, say, five hours, or to keep heat very well - so you're wasting energy and risking equipment failure or worse.
  Still, if you want to take this route, you can - just be sure to buy a meat thermometer (an old-school, analog, all-metal one) and put it inside the oven, then try to be around through the process.
  A better option is to get a proper heater. Laboratory and industrial ovens are bulky and expensive, but one of the more affordable and portable options are hot air sterilizers / dryers for medical puporses. One such device is Tau Portable. The unit should be selling for about $300. These devices offer very precise temperature control, and design that permits them to run for long hours in a very power-efficient manner. Be sure to examine sterilizer chamber dimensions - at least 30 x 15 x 6 cm is a good idea for typical mold sizes.
• Digital calipers and a micrometer:
  You will absolutely, positively need to accurately measure many third-party components such as motors, shafts, ball bearings, or even wires, to ensure proper fits and troubleshoot problems; manufacturer specs will be often unavailable or not worth that much, particularly when dealing with surplus components. If you do not have one already, it's a good idea to get small digital calipers (15 cm or so is fine), necessarily with a resolution of 0.01-0.02 mm. Ball bearings and motor shafts must have perfect press-fits, so do not try to be a cheapskate. There is a broad selection even on Amazon, and some $15 should get you a decent one.
  Furthermore, from time to time, you have an use for more precise measurements - so consider getting a digital micrometer, too. These are more expensive and less convenient to use, but offer resolutions of 0.001 mm. Typical price tag is $70-$100 for digital units (and again, Amazon - surprisingly - has some decent deals).
• Jigsaw:
  Cutting prototyping boards by hand is very time-consuming and annoying. A decent 600-800W corded jigsaw from any hardware store, with blades designed for fast cutting of hard woods, would work wonders. Try to get an unit that could be attached to a vacuum cleaner, to minimize the amount of airborne dust produced - and make sure that the blades purchased are long enough for the thickness of your prototyping boards.
  Budget brand jigsaws are best to be avoided; not only they tend to break, but are also generally pretty underpowered. The difference between a 200W no-name and 700W Hitachi jigsaw is pretty significant. Wireless saws are also usually a poor idea, as you need to pack some punch to cut through a 50 mm solid plastic board with a reasonable speed. You should be able to get a decent one in the $80-$150 range (do not go crazy for features such as tilt bases, guides, etc - your CNC machine will do the dirty work, you just need to feed it at least sorta-rectangular stock).
  (Jigsaws are one of the safer power saws, but it is still possible to cut off a finger or two if you misuse them. Read and follow manufacturer's safety guidelines, or learn about survival of the fittest the hard way.)
• Machine vise (if you can find one):
  Some CNC machines come with vises or other clamping systems (e.g., T-slot, rotary axis clamp) installed. On units that miss such features, many materials can be still reliably secured with proper double-sided tape, screws, etc - but you might want to look into machine table-mounted vises, too, particularly if you want to be able to do rough and dirty work in blocks of metal.
  I could not find any solution that would work well for me - most of the flat profile, affordable precision vises are very small, and large units are not only insanely pricey, but also weight in excess of 30-40 kg, which is pretty ridiculous and above the table load limits on most benchtop mills. As such, I cannot offer a meaningful advice here - other than saying you will do well without one, but if you spot some sweet lightweight (aluminium?), low-profile units, let me know.
• Deburring tools:
  These simple, swivel-action manual devices are used to quickly, conveniently, and safely remove extra burr that may be left in some materials around edges, or appear on mold joints. These tools have pretty unassuming appearance - but you will soon find they work great for all types of minor manual finishing jobs for plastics and light metals, and are much more useful than files and knives, particularly on curved shapes.
  There are several companies that make such devices, and you just need to get an entry-level, general use one with a blade intended for plastics and soft metals; Noga has RB1000; C. P. Enterprises has DT-2; and just about anything along these lines from any other manufacturer should do.
  Another useful tool is a scraper, which is a very fine fixed cutter that allows small amounts of material to be removed safely, without the risk of cutting too deeply or damaging nearby surfaces; one such example is Noga SC8000.
• Fine sanding paper, possibly a rock tumbler:
  Stock up on several grades of waterproof carbide sanding paper. Hardware stores usually have pretty ridiculous prices, but you can get 50-sheet packs on Amazon for cheap from sellers such as Industrial Supply. Try to get about eight sheets of each gradation (or something close): 100, 320, 600, 1200, 2500. These, when used progressively, can be used to bring any opaque plastic surface to satin shine, and will be indispensable for correcting occasional tool marks, reducing friction in sleeve bearings, etc.
  You might be also tempted to look around in hobby supply stores and find sheets of 4000, 6000 and 12000 sanding paper (or, alternatively, just a soft cloth and an appropriate polishing paste). With the help of these gradations, every opaque or transparent plastic, as well as metal, can be brought to perfect luster. Kingsley North has some 4000, 6000, and 8000 sheets, and polishing pastes, in stock; and several slightly more expensive types of 12000 cloth can be found on Amazon.
  If you are feeling frisky, you can buy, or build yourself, a polishing chamber. The idea is to have a rubber-lined can or jar, rotated or vibrated by a motor. When filled with an appropriate abrasive powder, it eventually brings any part placed inside to luster, even in hard-to-access spots you would not be able to finish well with sandpaper. Kingsley North has such devices ("rock tumblers") starting at around $70, along with suitable media and fillers. Rotating tumblers tend to shape parts, wearing out sharp edges more aggressively than other surfaces; vibrational tumblers do not exhibit this behavior, and work faster - but tend to be noisy in comparison. Your call.
  One thing you should not waste your money on are sanding or polishing attachments for a Dremel-like power tool. Although they work very well on wood and metals, they suck for plastic molds - and most of them are also too big to safely use for this purpose. You can still use the tool with thin rolls of hand-rolled sanding paper to gently reach into tight spots, or use some non-abrasive felt or cloth attachments for buffing waxes, but that's about it.
• Cyanoacrylate glue and a hot glue gun:
  Glues would be pretty essential for some of your work, and cyanoacrylates ("instant glues") are one of the best all-around choices for plastic parts. Ditch grocery store instant glues, however, and shop for real engineering cyanoacrylates: you will find compositions specialized for a variety of uses (binding plastics, metals, flexible compositions for rubber, etc), in various viscosities (including thixotropic variants), with specified adhesion and tensile strengths, and with a wide variety of curing times. They also come in very convenient, larger containers with decent applicators and caps.
  One company I am familiar with is Loxeal, but there are many alternatives (including Freeman-branded products or Loctite). Some quick recommendations: Loxeal 14 is a good all-purpose glue with lower reactivity and a manageable viscosity (very low viscosity glues might be too hard to apply); 17 is a slow-curing, gap filing variant; and 29 is a flexible elastomer-filled compositions for rubbers. Interestingly, they do not cost more than the grocery store variety, per ml.
  You might also want to consider getting a small hot glue gun from your local hardware store, plus a pack of adhesive bars. Just look for the cheapest, reasonably sized type - you do not need fancy temperature controls or so, $10 will do. Hot melt adhesives are very useful for temporarily attaching and encapsulating elements, etc - and in most cases, can be removed later on.
• Lithium grease:
  For periodic maintenance of ball screws and other moving parts of your machine (follow manufacturer's advice).

• Canned air:
  Polyurethanes are humidity-sensitive; some compositions to a greater extent, some to a lesser. In any case, it is advisable to keep opened containers under a blanket of a heavier-than-air, chemically inert gas - nitrogen is the primary choice in industrial applications, but the gases used in canned air dusters would work just as well, and are easier to obtain and safer to handle. The additional benefit of stocking on a couple of cans is that whenever working with a resin that must be exposed to air while polymerizing - e.g., as a glossy top coat on a previously molded item - the gas can be used to displace air from a makeshift "process chamber" (be it a cup, a jar, or even a cardboard box lined with foil).
  Go with the cheapest duster you can find - just be sure it's not lacd with Bitrex or other additives meant to discourage inhalant abuse; it's not that Bitrex would react with the resin in any way, but it tastes pretty bad and gets on your hands, clothes, etc. Stoner offers a Gust brand duster with no additives, in 12-can packs at $4 per can, complete with free shipping.
• Molecular sieve, grade 4A:
  Sounds serious, but it's a term for porous, powdered zeolite. The material is excellent for trapping certain molecules on its surface because of its highly uniform pore size; grade 4A is particularly suitable for removing water without interfering with the chemistry of the resin. This fine powder has a neutral color, costs relatively little ($30 for 2 kg) - and when a small amount is baked at about 200-250° C, then quickly added to a mixed resin, it dramatically reduces resin's susceptibility to moisture. The only downside: you can't use it with transparent resins, as it would give them a milky appearance instead.
• Thick, transparent, heavy duty plastic sheeting:
  When loosely draped over the mill, well clear of rotating parts and radiators / cooling inlets, this serves as an excellent and low-cost airborne dust and projectile protection mechanism that provides easy access to the unit and visibility from all angles. Thick varieties of sheeting stay in place, do not wrinkle, and are unlikely to interfere with the cutting process (but you don't get to sue me if this turns out to be a filthy lie).
  You may also be tempted to build a permanent case from wood, acrylic glass or polycarbonate, of course.
• Thin double-sided adhesive tape:
  A thin, non-foam based double-sided adhesive tape will come handy for affixing prototyping boards to machine table. Get some from a hardware store (but really, a film-thin one used for wallpapers, not a foam variety!). Tesa 4970 is an industrial tape with excellent strength, commonly used for amateur CNC work.
• Syringes and veterinary needless:
  Locate an on-line medical / veterinary supply store and order a box of 100 10 or 20 ml, two-element, single use syringes. Pick possibly the cheapest set ($20-$40 in the States, much less elsewhere), preferably with no rubber seals or other features of this nature (seals do not survive the chemistry of polyurethane resins). Syringes are remarkably useful for getting components out of containers, carefully measuring them, and applying catalyzed resin to hard-to-reach spots on the mold. The 10-20 ml range is optimal; much larger syringes are nearly impossible to operate with viscuous liquids.
  You might want to also get a box of 100 oversize, 2-2.5 mm x 80-120 mm veterinary needles, also single use, for about the same price. These will be useful for reaching into resin containers without having to tip them over (you really, really do not want to spill isocyanates). Go for the cheapest set you can find.
  Lastly, consider getting a yet another box of 100 1.2-1.8 mm x 20-50 mm needles, again single use. These might come handy for filling very narrow holes in the mold with resins, or sucking out any trapped air spotted after pour.
  Also, if you live in one of the eleven US states that require a prescription to buy syringes, congratulations - your elected officials are morons.
• Large, deep stainless steel soup spoon (L-shaped variety):
  Silicone resins typically come in relatively large cans. In my experience, many robot part molds need about 100-300 ml of silicone resin (for 30-60 ml of polyurethane), it is somewhat cumbersome to use a 20 ml syringe to put it in the mixing container (and if you get a 100 ml syringe, you will need three men to operate it, given the viscosity of a typical silicone resin).
  So, soup spoons are a perfect solution. Just be sure to get a smooth, high-quality, plated one (i.e., shiny). This reduces the chance of any unexpected interactions with addition cure resins. You can also use a plastic cup or some other means, of course, but a spoon is definitely the best option.
• "Specimen" sample containers:
  Made of thick, chemically resistant polypropylene, these are ideal for mixing resins of all sorts. Ted Pella has a good selection of Tri-Stir polypropylene cups for about 10-20 cents a piece. I would recommend against using some of the flimsier beverage cups instead - made of polystyrene and other lower grade plastics, these are not guaranteed to take aggressive resins well (plus, are easy to accidentally puncture or crush).
• Tongue depressors or chopsticks:
  All resins must be mixed thoroughly to initiate proper polymerization. In my experience, this is best done with single-use wooden chopsticks, although some people like tongue depressors more; Ted Pella is a good place to go for the latter (two cents a piece), restaurant supply stores might have the former in stock.
• A set of small, cheap brushes:
  For applying viscous resins to molds, and applying demolding agents where needed.
• Airbrush (optional):
  Some demolding agents (e.g., polyvinyl alcohol) and some lacquers achieve perfect, glossy surface aspect only when evenly sprayed, and brushes will unavoidably leave some streaks. If you have some money to spend, consider purchasing a quality dual-action, gravity-fed airbrush, plus a reasonably small compressor capable of at least 4 bar intermittent pressure and about 25 l/min of air flow.

  A decent airbrush should cost about $90 to $130 (Paasche Talon and Badger 100 LG are both a decent choice; I use Sagyma SW770). The compressor will add another $150 or so (sample selection).

  Using airbrushes properly takes some learning, but there are quite a few tutorials for model makers and painters on the web, so look around. They also double as a superior and in the long run, more economical alternative to compressed air for cleaning or drying molds.

• About 50-100 pieces of polypropylene, acrylic glass, or similar plastic, cut to size (1-3 mm thick):
  Cut to a dimension roughly comparable to the scale of parts you would be making. When casting polyurethane or epoxies in one-sided molds, putting them on top of a mold ensures a smooth surface and eliminates denting on foaming that may occur in humid environments due to isocyanate - water reactions. The benefit of acrylic glass is higher rigidity and good availability; polypropylene, on the other hand, is sometimes cheaper, is very resistant to chemicals, and does not stick to most resins.
  Acrylic glass and many other plastics in arbitrary sizes can be ordered in bulk for cheap from multiple places that supply them for advertising industry (for signage, displays, etc); just get the cheapest variety from your local outlet. Online, a good source is Professional Plastics, although it might be not the cheapest one.

• Containers and tools used for silicone oil, wax demolding agents, etc: salvageable after rinsing with water and a detergent, then drying out. Waxes and silicone are also non-toxic and inert, so you can also just wipe the tool dry and reuse it, even with residual material.
• Containers and tools used for non-catalyzed epoxy, polyurethane, polyester, and silicone components: salvageable only after thorough rinsing with a solvent. Since the solvent then needs to be disposed of, and has a worse environmental footprint than a plastic cup, it is advisable to just throw the stuff away. Some uncured epoxies are water-soluble, but everything else isn't.
  You can, however, reuse a container or a syringe for the same component a number of times, assuming the substance is not particularly dangerous or sticky: just be sure to label the item clarly, and wipe it reasonably clean to avoid contaminating next batches - or the air in your workshop - with residues.
• Containers and tools used for catalyzed silicone resins: pour out excess resin prior to setting, wait for it to cure, peel away film. Note that if material was not mixed sufficiently well, some uncured regions may be exposed (particularly common with polyaddition resins). In this case, throw the container away, or wipe as much as possible and reuse only for the same resin.
• Containers and tools used for other mixed / catalyzed resins: pour out excess resin prior to setting, wipe carefully with paper towels. Wait for any excess to cure, try to peel away film if present (this usually works with polypropylene cups, but not PET or PS ones). If complete removal is not possible, and the remainder interferes with operation of the tool / container, throw away.

11. Safety and health

11.1. Power tool safety

11.2. Noise considerations

11.3. Dust considerations

11.4. Dealing with chemicals

11.4.1. Silicone rubbers

11.4.2. Polyurethanes

• Phthalate plasticizers: there is some controversy around certain phthalate plasticizers, most notably dibutyl phthalate (DBP). These plasticizers are fairly uncommon in polyurethanes, but appear in a handful of compositions - e.g., Innovative Polymers IE-9080 rubber. Acute toxicity of DBP is modest, but some scientists are concerned about long-term, subclinical exposures to extremely low concentrations of DBP leaching out of cured plastics, and disrupting certain hormonal processes in human body. While phthalates are still very commonly used (and therefore, your exposure is largely independent of your hobby work), their use is restricted in products such as children toys. In case these fears are warranted, it may be safer to stay away from DBP when producing parts that are meant to be extensively handled, worn, or come in contact with foodstuffs.

• Mercury catalysts: another controversial (but increasingly rare) additive is phenylmercuric neodecanoate (or some other mercury carboxylate). Used in minutiae quantities (0.05 - 0.4%), this substance served as a highly selective polyurethane catalyst that greatly favors the isocyanate-polyol reaction to the undesirable isocyanate-water one - hence imparting superior humidity resistance. It is now mostly phased out, but still making some appearances in transparent or elastic resins, especially on the US market, where it is relatively unencumbered by blanket environmental regulations. Examples of mercury-catalyzed resins include Smooth-On Crystal Clear 2xx, Freeman 1090, or RenCast 6401.
  The catalyst is one of the less harmful organomercury compounds, and is present in quantities small enough to be of very little concern during normal casting operations; but, it is not clear what the long-term fate of mercury in the cured resin might be. In the 60s, researchers assumed that the large molecule would be immobilized in the polymer matrix, but recent studies of polyurethane rubber floors catalyzed with phenylmercuric acetate revealed that after several decades of use, some of them give off elemental mercury vapors at an unexpectedly high rate. This could be due to manufacturing or application errors, UV degradation, or something else - but it's a good reason to refrain from using mercury-catalyzed resins to make toys, jewelry, and other everyday items.

11.4.3. Epoxies

11.4.4. Polyester resins

11.4.5. Pigments and dyes

That said, not all is roses. Key risk factors to look out for:

• Most pigments are finely powdered and may easily become airborne. As with any generic dust, do not inhale, and wear at least a basic single-use dust mask when doing something particularly messy, or working in confined spaces with poor ventilation,
• Some pigments are based on powdered silica as a way to encapsulate otherwise chemically active dyes, or to achieve desired visual properties. If so, additional respiratory risk is present, as outlined in earlier sections. In most cases, powders consist of amorphous silica particles 40 - 200 micrometers in diameter, which is believed to be reasonably safe - but double-check this, and exercise caution anyway,
• As noted earlier, some pigments, particularly inorganic ones, are either known to be outright harmful (check MSDSes), or are based on particularly toxic / bioaccumulating compounds that may be released when the pigment degrades or is attacked by resin components.
  General guidance is hard to provide. In principle, lead, mercury, and arsenic are top offenders, and should be avoided at all costs. Pigments based on cadmium, chrome, nickel are problematic, too. Cobalt, manganese, barium, antimony, and copper may be somewhat harmful in some compositions, depending on solubility, reactivity, manufacturing contaminants, etc. Zinc, tin, iron, aluminium, magnesium, titanium, and common alkali metals are of least concern. Google around for MSDSes and health warnings, particularly if you intend to make toys, jewelry, or other items that would be extensively handled or worn by humans. There are also some pages listing potentially dangerous pigments, for example this one.
  As usual, there is no need to overreact, but it is also unwise to buy and use potentially harmful pigments if equally good non-toxic equivalents exist.

It is particularly important to avoid harmful pigments when spraying coats with an airbrush or a paint gun. For such applications, stick religiously to reasonably safe compositions.

In addition, not every pigment is guaranteed to be compatible with every possible resin; be sure to perform initial tests on a small quantity of material, and immediately contain and discard the sample, then cease to use the dye, if an unexpected reaction is evident (through color change, premature polymerization, inhibited polymerization, foaming, smoke, etc).

11.4.7. Solvents, glues, demolding agents

11.5. Staying legal

12. What next?