• MMS Youtube
  • MMS Facebook
  • MMS Linkedin
  • MMS Twitter
9/1/1996 | 18 MINUTE READ

Advanced Controls For High Speed Milling

Facebook Share Icon LinkedIn Share Icon Twitter Share Icon Share by EMail icon Print Icon

This CNC expert covers the key factors behind making controls and the entire milling process move faster.


Facebook Share Icon LinkedIn Share Icon Twitter Share Icon Share by EMail icon Print Icon

What can a high speed control do for you? In simple terms, it gives you the ability to finish one task faster and move along to the next sooner. In drilling and tapping, this can result in faster hole-to-hole times, quicker spindle reversals for tapping, and substantial cycle-time reductions. The most dramatic benefits, though, come in 3D contouring. Few, if any, drilling and tapping jobs require a million lines of machine codes. In molds, dies, patterns, and prototypes, complex surfaces comprising a million or more line segments are not at all uncommon. Saving just a fraction of a second per move can result in substantial cycle-time improvements.

As an example, the part that formerly took 3 hours and 43 minutes to mill accurately on a machining center. The part is being cut in wax, but is actually commonly milled in carbon as an electrode to burn cashew-shaped gates for plastic injection molds. A control retrofit enabled that same machine to mill a more accurate part in just 17 minutes. The user enjoys greater productivity from his people and machinery. He also enjoys a distinct competitive advantage when selling his work, because he can deliver an equally good or better job for less money, in less time.

In a broader sense, high speed creates many other benefits. Improved accuracy, fit, finish and cutter life are the most commonly reported peripheral benefits. Customers share the benefits of high speed through the entire manufacturing process, not just to produce more work in less time, but also improving the accuracy and finish and reducing polishing and fitting time. They do this by using the high speed to reduce the stepovers and the tolerances. Tools simply last longer because their chip load is much more consistent.


"High speed machining" lured us with a promise of going faster. But faster than what? Each of us has a different perspective on what truly constitutes high speed machining, depending on our experience, applications and our needs.

The relative connotation of high speed is evidenced by talking with different people who utilize high speed milling on a daily basis. Going from 10 ipm roughing pre-hardened tool steel to 30 ipm, or from 30 ipm to 180 ipm, can be high speed in the mold business. Still, if your application is the milling of foam patterns for automotive stamping dies, 800 ipm may not be fast enough.

There is no clear threshold where plain old milling becomes high speed milling. High speed is relative, based on your perspective, your materials and your needs.

Surfaces To Points To Surfaces

The evolution of CAD/CAM into a powerful tool for 3D surface creation is the main reason we talk about high speed milling for molds and dies. CAD works with entities and surfaces. Points, lines, arcs, cylinders, spheres, planes, and so on, all join in CAD to create surfaces. CAM then translates those surfaces into point meshes of data for machines. Once that data is passed to the CNC, it executes one point at a time to reconstruct the surface. In order to do this efficiently, points are not milled at random, but rather organized into a data flow along slices or flow lines. These may be along any axis or across a combination of axes, or even along a constantly changing detail on a surface. In any case, the flow of data is from one point to another along the flow line or slice, then typically to step over and repeat that slice either in a single direction box-type cycle or in a zigzag of back and forth movements along the flow.

Point-to-point then is the process of creating surfaces by milling from one point to another in succession. As we apply point-to-point to high speed milling, that succession from one point to another should ideally be quite fast.

Chordal Deviation

CAM typically creates points with various distances between them. Why? If we step back a moment and just consider milling an arc by single point moves, we see that the arc really becomes a series of line segments. Those line segments deviate from the arc by a value commonly called chordal deviation. This value is set in your CAM system to define what the acceptable deviation from the surface will be, based on the accuracy required for the application. The natural desire is to be as accurate as possible, but stating a deviation value that is too tight can result in enormous file sizes and high data density that can be difficult to handle. Chordal deviation must be properly set to lend balance to productivity and required job accuracy.

The result of chordal deviation from CAM is a series of chord segments that are now becoming commonly referred to as "point departures." This term simply means the distances between successive point-to-point moves.


In 3D contouring the cutter must flow through the points without dwelling, but also must not inadvertently overshoot its intended path and consequently gouge the workpiece. Most numerically controlled milling machines take from 0.100 to 0.200 inch to stop from a move at 100 ipm. If a CNC and machine are instructed to flow through data at high feed rates, yet point departures are short, gouges can result at points of abrupt changes in the contour. Look again at the shape shown earlier. There are several close data segments at the bottom of the contour. This is an area of great danger for gouging. The longer line segments going from left to right might easily permit high feed rates. Without look-ahead, the CNC might be surprised by the abrupt change in direction over a short move of only 0.010 inch. If the feed rate is too high to stop in that distance, the result will be an overshoot.

Look-ahead must evaluate data many blocks ahead to prevent such gouges. In most applications one or two, or even ten or 20-block look-ahead is not enough. The amount of look-ahead needed varies based on contours, feed rates and machine performance. In general, look-ahead cannot be limited to any arbitrary value, because conditions are constantly changing. Ideally, look-ahead should be dynamic, varying the distance and number of program blocks based on the part profile and the desired milling feed rate.

Look-ahead is now offered by some companies either as a preprocessing step or as a part of a DNC system. To achieve the effect of look-ahead, the system must add data segments at varying slower feed rates. In this way, gouging and overshoots can be prevented. The drawback is that by adding program lines, the data throughput problem is exacerbated, greatly reducing productivity.

High Feed Rates

How accurate can a CNC really be at high feed rates? We can answer this by analyzing the entire machine system, starting at the CAD/CAM system and ending with the machine iron. Because we are talking specifically about the controls here, let's assume that everything else is in order, and that the control is the only issue.

Servo cycle time is the amount of time a CNC takes for each measuring and command cycle. In other words, if the control's servo cycle time is 20 ms (milliseconds, or thousandths of a second), then the axis positions are measured and a new direction commanded by the control 50 times a second. Though 20-ms servo cycles were thought to be good just 10 years ago, servo cycle times over 4 ms are now considered inadequate. At a fairly common 3-ms servo cycle time, positions are being measured and corrected 333 times per second. A machine moving at 100 ipm is moving 1.66 inches per second, so each time the axes are measured, the machine should be moving 0.005 inch.

This might be alarming to you if you are trying to hold tolerances to 0.0001 inch or so, because your machine is basically out of control for 0.005-inch increments at a time. The accuracy problem gets worse when attempting to mill at 400 ipm, where a 3-ms servo cycle results in 0.020-inch moves between measurement and correction commands from the control.

Chart 1 shows a few sample servo cycle times, measuring speeds and distances at feed rates. This chart demonstrates that to mill as accurately at 1200 ipm as at 100 ipm, the control must indeed be very fast.

Feed rates will continue to increase, and the need for faster servo cycle speeds will continue to grow. Cutter technologies are proving capable of supporting amazing speeds and feeds. The other supporting technologies like high speed spindles, end mill holders, and so on are all enabling amazing speeds and feeds. Machines with linear motors are now available with traverse rates to 3,000 ipm and more. They can accelerate to 3,000 ipm faster than most machines today can get to 300.

Block transfer timethat is, the number of blocks-per-second the control executesshould not be confused with the servo cycle time. Ideally, the servo cycle time should be faster than the block transfer time. Still, it is possible for controls that execute a high number of blocks-per-second to execute at slower servo cycle times. In these cases, the number of blocks-per-second is misleading, indicating a higher speed than can really be achieved if each block is an actual discrete motion block. The combination of fast block transfer time with a still faster servo cycle time ensures high data throughput, with optimal accuracy.

The DNC Bottleneck

Now that we've looked at the control's ability to mill accurately at high speeds, we have a new dilemma: how to get the program information to the CNC fast enough to avoid data starvation. Anyone who has milled 3D contours has watched as the CNC has stopped and waited to fill the buffers again to continue program execution. Loading the program into the control helps it run faster, yet that can often be impractical with the small CNC memories or slow communications speeds.

First, let's consider DNC, the most common communications in use for CNCs today. DNC stands for distributed numerical control, the distribution of numeric cutter path data to CNC machinery, or direct numerical control, the "drip feeding" or dynamic downloading of numeric cutter path data "on-the-fly" as the CNC executes it.

DNC is typically performed through a serial communications link at data rates of 110 to 38,400 baud or bits per second. Most common is 9600 baud, resulting in potential throughput of up to 960 characters per second.

Program information for the CNC usually is in blocks or lines of program data averaging approximately 20 characters per line. For example:

G1 X123.456 Z234.567
<Enter> <Linefeed>

Even spaces and invisible "control" characters like the carriage return and the linefeed take time for transmission. The addition of 3-, 4- or 5-axis definitions, line numbers and feed rates simply add to the overhead in data transmission. Given a communications rate of just 960 characters per second, the CNC is then limited by DNC to just 48 blocks per second. In reality, DNC overhead commonly results in still lower performance, generally about half the theoretical potential. At this rate of 24 blocks per second with 0.010-inch point departures, the resulting DNC speed limit is just 14.4 ipm. That is too slow for high speed milling no matter where the observer's relativity is based.

There are several techniques used to speed DNC performance:

  • Use incremental data where points are measured from the last location, rather than absolute where all points are measured from one fixed location.
  • Use integer data, where a dimension like 0.012 is instead shown as 120, eliminating the decimal point, but adding trailing zeros.
  • Data compression, using mathematical algorithms to make data, takes less space.
  • Preprocessing, to reduce data tolerances and eliminate unneeded data.
  • Don't use cutter compensation or fixture offsets, mirror imaging or scale factors.

These techniques increase the operator workload, increase the opportunities for errors, and reduce the operator's flexibility with the machine. Networking your DNC computer may help work flow, but does not solve the data flow problem. Many DNC computers are networked to get the data from the CAD/CAM to the DNC system fast, but the data flow is still restricted between the DNC PC and the CNC control by the limited bandwidth of RS-232 communications.

Direct CNC Networking

Direct CNC networking, or DCN, offers a better solution to the data flow problems in high speed milling. DCN uses existing networking architectures to provide a direct network link from the CAD/CAM to the CNC, eliminating the DNC system entirely. DCN is normally 1,000 times faster than DNC. This can be illustrated by the fact that a 10-MB file, which takes 3 hours to transfer by DNC at 9600 baud, takes less than a minute by DCN.

While Ethernet is the most common network architecture in use today, Arcnet, Token Ring and Fast Ethernet are examples of other common networks in use. Software protocols used include Novell's IPX/SPX specification, TCP/IP and NFS as found in most UNIX systems, Netbeui as used by Microsoft's LAN Manager, and more.

Networking architectures are already commonly available with data rates of 100 megabits per second or more, even ten times faster than standard Ethernet. Direct CNC networking eliminates any data bottlenecks.

DCN also provides infrastructure for growth, a foundation for expansion into the future. On PC-based (personal computer) controls, interesting possibilities include running CAD/CAM, job control and quality control software right in the controller.

Digital Signal Processing

The technology that allows "plain old PCs" to act as high performance CNC controls is known as digital signal processing, or DSP. Digital signal processing uses special dedicated processors to convert and interpret digital signals at very high speeds. Using DSP, a single board can control up to eight axes at the fastest servo cycles discussed earlier. Not so long ago, a much larger board would have been required for a single axis, operating at speeds 200 times slower.

The incredible power of DSP for specific tasks is illustrated by the fact that a DSP chip can execute a multiply-accumulate (MAC) instruction, a fundamental operation, in a single clock cycle. This same operation on a current Pentium processor chip takes 11 clock cycles. Obviously a 120-MHz Pentium will still take nearly four times as long as a 40-MHz DSP processor.

Earlier, we saw how dramatically the control's servo cycle time can impact the accuracy of the control. DSP is the key to fast servo cycles. DSP also influences the acceleration and deceleration "ramps" of the CNC control. Traditional CNCs simply set up a ramp rate for accelerations. Because of machine dynamics and drive systems, the machine suffered "following error". With so much power, DSP systems allow tuning the acceleration for real conditions and specific machine characteristics. Acceleration is no longer simply a straight-line ramp, but rather is tuned as a sort of "bell" shape. This optimization minimizes following error, reduces strains on the machine, yet provides greater acceleration overall. DSP's speed allows better accuracy and speed, yet is gentler on the machines it controls.

There are many varieties of DSPs, and those choices give control builders choices in their priorities for control functions. The variety of DSPs available means that you as a user can obtain different control performance, functions and capabilities to meet your specific needs.

DSP also has the ability to handle a variety of servo amplifier and measuring scenarios. While most traditional CNCs are limited to interfacing with one specific drive type, many DSP boards allow the integrator the flexibility to work with several, or most available interfaces. Given the concept that technology will change over the life of a machine, this flexibility is reassuring.

One of the other great benefits of DSP is that the main CPU (central processing unit) in the PC is still free to perform other tasks. In reality, a PC equipped with a DSP for machine control is using multiple processors, gaining a tremendous performance advantage. The DSP can be measuring and commanding axis positions while the main CPU is handling the receipt of network data and preprocessing that data for look-ahead to optimize milling speeds and part accuracies at the same time.

Open Systems Architectures

The use of an Intel x86 series processor in itself does not mean that a control is open architecture. Even using a PC-based design does not always mean that the architecture is really open. Open architecture can have many interpretations, none necessarily more right or wrong than another.

To me, open architecture implies two key components: an ability to be serviced and an ability to be changed without proprietary parts and/or knowledge. This means that open architecture should empower the owner to have choices in service and in expansion and updates.

There is a general convergence in the industry toward PC-based controls. This gives everyone the widest possible choice of design options and expansion flexibility. Personal computers, serviced by virtually anyone, in a variety of environments, have arguably proven to be as reliable or more reliable than the best of "hardened" CNC controls. The amazing development of PC features and performance has left the industrial CNC control market reeling to stay remotely in step with developments.

For use as a CNC, the PC may be used in a "shop hardened" configuration, available from a limited number of specialty PC suppliers. Alternately, virtually any common PC may be implemented within a variety of shop enclosures, providing protection from the contaminants in the shop atmosphere. If the air we breathe in a shop is worthy of our health risks, though, perhaps a PC need not even be protected from the shop atmosphere to be reliable. Many users of PCs for DNC would argue that plain old office-type PCs are perfectly fine for shop use.

One advantage which some industrial PCs offer is a "watchdog timer." This is an output that may be incorporated in the Emergency Stop circuit for a machine to add to the CNC's safety. A dedicated circuit in the computer constantly checks the integrity of the computer processes at very high speeds to ensure that the system is operating normally. If the system does not check out, the output can disable the machine motion by shutting off the drives as an Emergency Stop. This PC watchdog timer does not replace the DSP's watchdog timer, but rather complements it with additional security.

Multiprocessor Strategies

Earlier we talked about the fact that a PC teamed with DSP is actually a multiprocessing computer. This concept can be carried further in several ways.

Multiprocessors can give the CNC operator the ability to multitask. He may simply edit programs that are about to be run, or carry that further with graphical verification of programs submitted by the CAD/CAM department for operation. The operator may even use shopfloor programming to prepare cutter paths from IGES surface data right in the CNC control.

Within the PC marketplace, there now exists a number of computers that are equipped with or can be updated with multiple processors. For controls operating under Windows NT or UNIX, the multiple processors may offer performance benefits and the opportunity to use other programs effectively within the control while cutting.

An alternative is the use of entirely separate computers, controlled by the same keyboard and monitor. There are several CNCs in the marketplace today using this concept to maintain a high level of CNC control integrity and security while still performing other tasks. Multiple computers may be implemented in a number of different ways. Several companies have long offered dual-processor options, using a physical switch to select the "MS-DOS" mode or the CNC. Others use keyboard commands to select between the computers. Still others obscure the second computer, using one computer with multitasking as the user interface, and having it pass commands to the machine control computer.

The multiple computer strategy can be very simple to implement, gives the operator multitasking capabilities, and offers security for the CNC process. Moreover, multiprocessors offer another level of performance for high speed milling, allowing the operators to perform multiple tasks quickly and efficiently at their machine control station.

A Look At The Future

As we consider the future, the certainty is that CNC will continue to change. CNC will have to be able to accommodate the evolutionary changes in the industry without requiring the complete replacement of those controls as has been the norm with traditional CNC designs.

The example that I like to use to illustrate this is the word processor. When first introduced, it was an impressive contraption, consisting of a large workcenter with a green display screen, keyboard, printer, and built-in logic, the firmware. When new features were introduced, the entire word processor required replacement at a cost of tens of thousands of dollars, and the old system was traded in for pennies on the dollar. Through evolution, today's word processor is now a shrink-wrapped piece of software that may be used on virtually any personal computer. Costs run in the tens or hundreds of dollars, and updates for new features are less than that. Installation of the update is performed by the user, generally in just minutes. Again, the evolution of the word processor parallels developments in many other computer industries, and is a good example of where CNC is likely to go.

Shrink-wrapped CNC? Is it possible? Imagine buying a fancy new CNC machine, and selecting your CNC as a software package for it, not an entire custom interface.

Think about new features evolving. They have historically made your CNC machine obsolete. Now imagine again getting updates at low cost, as with the word processor, or with your CAD/CAM system. We call these incremental advances in the technology. Simple, low cost changes will keep your present equipment and entire company competitive.

All this is not just a pipe dream, but the developments are taking place right now to make it a reality. Change is the one thing we can count on in the future. CNC must be able to accommodate that change.

Chart 1--Distances Moved At Given Feed Rates And Servo Cycles

Distance Traveled
100 ipm 400 ipm 1200 ipm
20 50 0.0333" 0.1333" 0.4000"
10 100 0.0166" 0.0667" 0.2000"
3 333 0.0050" 0.0200" 0.0601"
1 1,000 0.0016" 0.0066" 0.0200"
0.4 2500 0.0007" 0.0026" 0.0080"
0.1 10,000 0.0002" 0.0007" 0.0020"