Motion Control Boards Take Mach3 From Hobby Class to Industrial Grade, Part 1

Dec 23, 2011 // by Bob Warfield // Blog, Products, Software, Techniques // 12 Comments

Many a forum post has carried the derisive view that Mach3 can’t be an industrial grade control because you can’t build “industrial grade” on top of Microsoft Windows. The logic goes on to say that because Windows is not a real time operating system, it’ll never be able to do industrial grade work. There are various other arguments against Mach3, such as the idea that the servo closed loop isn’t accessible to Mach3, but the big issue is the real time operating system issue. How severe is this criticism of Mach3 and Windows?

It turns out the idea that an industrial grade CNC controller can’t be built on top of Windows is false. We can dispense with that objection right away with an existence proof. Consider the Siemens A2100 control, used on machines like the Cincinnati Arrow. Some say it was one of the best CNC controllers ever made, but whether you agree with that assessment or not, it was clearly an industrial grade controller, and it was based on Windows NT. There are many other examples of industrial grade CNC controls based on Windows.

How did they manage to get great results despite Windows and what’s the deal with Mach3?

The key answer in both cases is that they separated the functions of low level motion control and the user interface and g-code interpreter. Look at a CNC controller as consisting of 3 layers:

CNC Controller Block Diagram

At the top level is the human interface, the part that talks to the operator through the control panel. The middle level is the g-code interpreter. Give it a block of g-code (uploaded by the User Interface through a USB key, serial feed, or other means), and it knows how to convert that g-code into simple motions. Those simple motions go to the motion control portion of the program which is responsible for converting the simple motions into driver control signals. For example, if you want to X0Y0Z0 and the machine is at X1Y0Z0, the X-axis has to move -1 to get to 0. Let’s say your machine moves in 0.0005″ steps. It then has to move 1 / 0.0005 = 2000 steps in the correct direction.

Let’s look at how this works with vanilla Mach3, running with a parallel port. What that means is the Mach3 software must generate a smooth series of 2000 pulses on a parallel port pin to accomplish that motion. If we want to move at 100 IPM, which is not very fast by industrial standards, we have to produce those 2000 pulses in 0.6 seconds. For each pulse, we only have 0.0003 seconds to get the job done. Put another way, we need to move at 3,333 pulses per second. It turns out Mach3 can pulse up to 6 axes at whatever the kernel speed allows, but Mach has to do more than a few unnatural acts to pull that off.

Any minor interruption or distraction for the computer, may result in a brief pause between sets of the pulses. It turns out that Mach processes groups of buffered pulse requests. For example, it may have 5 requests that each ask for 0 to 5 pulses, depending on whether the axis is maintaining speed, slowing down, or accelerating. Lets say we have perhaps 25 pulses from 5 requests of 5 each. If, instead of getting a following on request for 25 pulses, that follow on is lost, it may take a whole new kernel timer cycle before the pulses are delivered. In the meanwhile, it’s as though the lost cycle had instead been heard but requested the axis to slam on the breaks to 0 speed. On average, we got 25 pulses instead of 50, so we ran a lot slower, but the acceleration/deceleration makes it that much worse. Imagine trying to cut smoothly and with good surface finish if you command your cutter to move 1″ at 100 IPM, but at 2 points during the move it actually slows down to 33 IPM. Things are not quite that bad because we have inertia to smooth it out and the axis doesn’t start and stop instantaneously, but you can start to understand the problem here.

A so-called “real-time” operating system, which Windows is not, has the capacity to put out a “do not disturb” sign during critical operations. If it was making that 1″ move, it might very well want to be left alone to finish that task in perfect timing and at exactly the right speed.

The problems of distracting the controller, resulting in poor surface finish or even worse problems, can be compounded because Mach3 runs on a general purpose Windows PC. The owner may have installed all sorts of other software on the machine. They might be running the MP3 player so they have music while they’re machining. There are a number of factors that can cause one of these timer ticks to be disturbed including:

- Multimedia Software. I’m told Quicktime is a known offender.

- DMA. If your machine accesses disk too much, it interferes as the disk takes control of data transfer through a process called “Direct Memory Access”.

- Normal OS operations can also disable the interrupts involved from time to time.

- Lots of other kinds of software.

For this reason, you frequently see recommendations not to connect a Mach3 PC to the Internet, not to allow any extraneous software onto the controller machine, and so on. This is a pity, because it is almost entirely due to the parallel port’s limitations and we can work around those as we shall see in a moment. Personally, I hate the idea of being cut off from the Internet in my shop. There’s just too much useful information to be had there that I can use in my work.

How do the Industrial Grade Controllers get around this problem?

The answer is pretty simple. They move those parts of the software that are sensitive to these timing issues to another piece of hardware. Imagine a configuration like this:

CNC Controller w/ PC and Motion Control Board

Now we have a separation of tasks that forms an assembly line. As the Windows PC finishes converting each g-code line to some sort of motion commands, it hands them off to the Motion Control Board and goes on to the next line. The Motion Control board is a piece of dedicated hardware whose sole purpose is converting motions to driver control signals. It can’t really be interrupted because it doesn’t do anything else, which is great in terms of providing smoother and higher speed motion.

The G-Code Interpreter is happier too. It takes a fair amount of machine cycles to produce all those motion control pulses smoothly. By handing that task off to another piece of hardware, the G-Code Interpreter finds more time to do a better job itself. Whether that’s simply processing g-codes faster, or using the time for more sophisticated “look-ahead” and trajectory planning is a function of the controller, but the important thing is that more time is available.

Having this 2-part architecture is the secret sauce in creating an industrial grade CNC control from Microsoft Windows. It isn’t the only secret sauce, but it is an extremely important starting point.

Is it hard to use a Motion Control Board with Mach3 instead of the Parallel Port? Is it worth it?

It turns out to be pretty easy to use a Motion Control Board with Mach3, and it is well worth the effort. In Part 2 of this series, we’ll take a look at some of the Motion Control Boards that are available and we’ll also talk about the results.

*** This article is slightly revised from the original version. I benefited greatly from a generous side conversation with Henrik Olsson, who gave me quite a lot of additional information and even took the trouble to hook his logic analyzer up in order to prove that Mach3 can pulse 6 axes as easily as it pulses 1. Thanks Henrik!