[JohnJW]

How do the various HSM and look-ahead functions from the various cnc controllers compare? I am curious because the PC based controller, ie Hurco, Mach 3 and others, looks more impressive on paper than the embedded system of Haas or Fanuc.

It's a very complicated subject and I am just guessing here, but assuming a pulse every .0001" a 1 Khz controller can have a max speed of .1"/sec, a 1Mhz controller can have max speed of 100"/sec or roughly 6000ipm. Assuming a finer encoder, overhead resources, multiple clock cycle per instruction depending on chip architecture, my overly simplified theoretical max for a even lowly modern industrial cpus running in the sub Ghz range are still beyond the max cutting speed of the cnc machines I see.

Have we progressed far enough in electronics where the bottle neck is in mechanical system, ie servo motor and machine design rather than the electronics? I am leaning toward that conclusion because I see wide discrepancy in the specs of the cnc controllers without much market differentiation. . . . or because I'm a Haas owner and need to convince myself that my Haas controller is just as good as or better than PC based controller

I also wonder how Haas controller can hold it's own against dedicated cnc controller manufacturer's like Fanuc or Siemen? Haas controller looks the same over the years but I think they go through hardware/cpu upgrade every few years, but how come the specs for HSM and look-ahead seems to stay the same?

Will HSM work with program running from USB? Even the lowly USB 1.0 has a theoretical max transfer rate of 12Mbits/sec vs 115Kbit/sec for RS-232. I am asking because with limited funds, should HSM be favored over 16mb upgrade. .. or is HSM even necessary for the newer processors?

In addition, how do the various controllers compare, not so much in interface eye candy but in actual raw machine performance?

I know, a lot question in one post but all this recent talk about Haas controller options really brought out the my curious side. . .

[Xenomorph]

I have been looking for the same answers.

http://www.cnczone.com/forums/showth...highlight=haas

[mactec54]

Hi JohnJW

I have the Haas lookahead turned on it does help but is not that great, The best system out there is a control by softservo systems,It is very fast, I have had one for 9years,

It has 1000 cycle three-dimensional dynamic look-ahead contour control (3D-DLACC) with pre-interpolation acceleration for high-speed, high-precision milling,You can also use 17 bit upto 20 bit encoders with this control,

This is the only true HSM control that there is in the market today, the other controls kind of got it, about half right even the the big name controls, Go to there site & check it out

[Xenomorph]

Mactec54,
I am trying to get this. Why is this different from what Haas does? Haas states that they do 1000 blocks per second block execution (same as ServoSoft). Haas states that they do "acceleration before interpolation" (I think the same, just worded differently). This all sound the same to me. What am I missing in all of this?

http://www.haascnc.com/MAIN_high-prod_HSM.asp
http://www.haascnc.com/MAIN_HaasControl.asp#haascontrol

Hi JohnJW

I have the Haas lookahead turned on it does help but is not that great, The best system out there is a control by softservo systems,It is very fast, I have had one for 9years,

It has 1000 cycle three-dimensional dynamic look-ahead contour control (3D-DLACC) with pre-interpolation acceleration for high-speed, high-precision milling,You can also use 17 bit upto 20 bit encoders with this control,

This is the only true HSM control that there is in the market today, the other controls kind of got it, about half right even the the big name controls, Go to there site & check it out

[mactec54]

Xenomorph

There is a big difference Soft Servo hold the patent for this process, If you put the 2 systems side by side then you will see the difference, a 1000 line look ahead does not mean very much, the word they don't use is true execution, the soft servo control can execute the true 1000 blocks of code, where the other controls can not, you need to have at least a 2Ghz processor & 256mb to 512 mb of ram to make this happen, there are not many commercial controls that even come close to the required amount of computer power needed to do this

[Xenomorph]

Mactec54,
I am really getting confused with this. On the Haas web site I found this statement:
"The Haas control performs servo loop calculations up to 4,000 times per second for each of six motors, and provides up to 1,000 blocks per second of program execution speed."

I did go to the ServoSoft web site but could not find a good definition of what it is that they do differently than anyone else. Maybe I am just looking in the wrong place. I also tried to find the price for their software but could not. If it is inexpensive I would just buy it and then take a look at their source code as it is open source.

Thanks Mac I appreciate you help

Xenomorph

There is a big difference Soft Servo hold the patent for this process, If you put the 2 systems side by side then you will see the difference, a 1000 line look ahead does not mean very much, the word they don't use is execution, the soft servo control can execute the true 1000 blocks of code, where the other controls can not, you need to have at least a 2Ghz processor & 128mb to 256 mb of ram to make this happen, there are not many commercial controls that even come close to the required amount of computer power needed to do this

[mactec54]

Xenomorph
You have to think the one you are looking at I have, there system as well, there top speed is 833in/min & only 2D, they have 2 processors to do this,Soft servo is 3D 2400in /min 1 processor, there software around $3000 They mostly only sell to OEM machine builders,

In the last quote you must of copyied it before I had done some changes, the word is True Execution of 1000 blocks of code, the word that not many can use is True

[JohnJW]

Hi JohnJW

I have the Haas lookahead turned on it does help but is not that great, The best system out there is a control by softservo systems,It is very fast, I have had one for 9years,

It has 1000 cycle three-dimensional dynamic look-ahead contour control (3D-DLACC) with pre-interpolation acceleration for high-speed, high-precision milling,You can also use 17 bit upto 20 bit encoders with this control,

This is the only true HSM control that there is in the market today, the other controls kind of got it, about half right even the the big name controls, Go to there site & check it out

I wonder if we are at a stage where the hardware has not yet caught up to the software. I think the coldfire processor used in the newer Haas controllers are running at maybe a few hundred Mhz, and the chip before the coldfire, like my 2006 Haas, is probably in the sub hundred Mhz range. However, I haven't heard of any complaints about the later Haas controller not being able to keep up. I would image the difference between a Ghz chip and a lower range Mhz ship should be pretty significant, yet I haven't heard of too much discussion on that area so I assume the limitation is with the mechanical hardware rather than electronics.

Just briefly glancing over the specs of softservo, I wonder how they can compete with embedded controller on speed if they are relying on ethernet as a mean of communication. Won't that add more latency and resources to the systemn. . . . I guess that can be solved with the ever faster electronics. . .

[tobyaxis]

Maybe the HAAS APPS guy will drop in and shed some light on your confusion.
Just a thought but just how fast do you think a machine is going to move from point to point over a .005 distance?? That .005 is like 100 miles when it comes to tool&die accuracies.

Just a thought

[mactec54]

JohnJW

Softservo does not just have ethernet, they have many options, my control has fiber optic cables, they have there new Canopen which is better than the ethernet, it just depends on what drives you go with as to what interface works best with that drive, & no the hardware & machines are matching the controls,for speed with the use of linear motors that run at 3 to 6 G that is more than enough to handle the 300m/min rapids

[JohnJW]

Maybe the HAAS APPS guy will drop in and shed some light on your confusion.
Just a thought but just how fast do you think a machine is going to move from point to point over a .005 distance?? That .005 is like 100 miles when it comes to tool&die accuracies.

Just a thought

I don't know. Not having much experience I will wonder about how much backlash and how fine is the encoder resolution? Assuming a encoder resolution of .0001", moving .005" is only 50 counts so I think the harder part is mechanics in moving that massive table exactly .005 accurately and fast.

From what I've read, there's HSM as in High end HSM where machine rigidity and cutting tools are factor in and then there's the HSM where it's just getting the machine to move fast accurately and that's the part I wonder about. A lot of this is just to satisfy my curiosity because so far I am the limiting factor, not the machine.

However, it is nice to know more about the controller and how it functions. I prefer an embedded industrial system over a pc based system simply because I think most pc are not RTOS based and all the extra memory and speed of a pc based system are being used up by the fancy gui and the extra resources of a pc based system, instead of going into machine control. I go with HSM under the assumption that if I want 3d contouring I need hsm and that's what I've been told from talking to sales and other end users. The answer is usually either "yes hsm is better" or something similar to what I can read in Haas's short paragraph on HSM.

Of course, it's quite possible that my assumptions are wrong and I'm simply throwing money away on HSM or prefering an embedded system over a PC based system. It'll be great if Haas can shed some light on these issues.

When I compared Haas to Hurco, it was very tempting to go with Hurco base on controller specs, but I have a soft heart for products made here in the US and by working with HFO sales I was able to configure a Haas machine that was comparable in price with the Hurco machine.

[BobWarfield]

All the controllers are on pretty even footing when they're moving in a straight line. It's when they have to turn the differences set in.

In all likelihood, any industrial control can rely on some extra hardware to keep up with maintaining the signals to the servos. It doesn't take an awful lot. Look at how much the cheap little Smoothstepper will improve Mach3.

But that's just the straight lines. The motion controller has to do what's called trajectory planning to get through the curves. The reason is your machine can't turn on a dime. If the controller accelerates it too fast, it won't go around the corner properly, just like a race car moving too fast when it enters a turn.

That lookahead is all about the trajectory planner understanding what lies ahead in terms of turns, and figuring out how fast it can let the machine get going before it has to slow it down to make the turn.

If you're driving a race course, you practice until you know the course cold. You know exactly how fast you can take your car through every turn, and you set up for it in terms of when you brake and accelerate. These machines don't have that luxury. They only lookahead a little ways.

This is all compounded when your g-code is simulating some complicated motion as a whole bunch of tiny line segments. If you want accuracy to a tenth, you suddenly have to provide as many as 10 g-code segments per thousandth of an inch (that's worst case for a really tight twist).

tobyaxis's 0.005 now uses up 50 blocks of lookahead.

Add to that all the other little considerations: backlash comp on all axes with a different value in every direction, trying to read scales and other sensors to do temperature compensation, each axis has differing acceleration characteristics, what is the state of servo tune, yada, yada. Even cheaper machines like Robodrills have parameters to tell it things like whether you have the table fully loaded. Fancy machines sense how much weight is on the table, and every other darned thing.

All the while you're trying to make full use of your 1500 IPM and zillion rpm spindle. Yikes! That's a lot of computer action.

Let's say you're going pretty slow by some HSM standards. Maybe 100 IPM.

So each inch in the worst case needs 10,000 g-codes for tenths accuracy. Man, a 1000 block lookahead is gone through in 1/10 of an inch of motion which happens in 1/1000 of a minute or about 0.06 seconds. Not much time for the poor controller to get a clue what it should do.

Time flies when you're having fun.

Cheers,

BW

PS There was a fascinating thread on PM where they looked at machine execution times based on acceleration versus raw rapids speed. Acceleration was the most important factor!

[Drassk]

Add into that the fact that a crazy fast servo-update loop would be running at 30,000Hz (and nearly anything out there running at fractions of that) and you only have 1800 reference points in there at that speed...so your servo loop can't actually keep up if you're doing a direction change 10,000 times in that span!

The SoftServo system has a 2000Hz servo update rate over EtherCAT and most systems are even slower. Of course, it's rather rare that one would have to do 10,000 consecutive orthogonal moves!

The 30Khz update speed is a Galil Accelera running 2 axes on a limited instruction set. They can only run 15Khz on three or four axes on that set, and go to half that again if using a full instruction set (though they wouldn't need the full set for machining). Dynomotion's G-FLOP does 10Khz at a fraction of the cost of the Galil cards. You need to be able to program your own motion control on both, though.

[Edster]

My fanuc 31iA control's encoders have 16 million pulses per revolution.

My okuma osp100e control's encoders have 1,296,000 pulses per revolution.

I'm not sure of the pitch of the screws for either machine, but you can bet there is more than 1 pulse for each .0001 moved

I don't know much about the servosoft system, but if it were that great the top machine builders would be using them. I've never heard of makino or mori seiki or mitsui seiki or okuma or using a servosoft control.

[Drassk]

My fanuc 31iA control's encoders have 16 million pulses per revolution.

My okuma osp100e control's encoders have 1,296,000 pulses per revolution.

I'm not sure of the pitch of the screws for either machine, but you can bet there is more than 1 pulse for each .0001 moved

Pulses per revolution is independent of servo update rate. You can think of encoder count as how accurate your speedometer is and servo refresh rate as how fast your foot can move the gas pedal. More/better information does you no good without the ability to apply it and change course.

Once you've got ten or twenty pulses for your minimum programmable move, your servo loop is what's holding back your accuracy at speed.

[samco]

It really IMO comes down to the maximum acceleration and velocity of the machine. Emc2 can run 2000hz servo loop (even faster) - it also will process a g-code program and cue up the motion relatively quickly - way faster than it takes to actually run.

It comes down to how fast the machine will actually accelerate and move.

sam

[samco]

Take a look at these 2 screenshots. This is a simple spiral shape made up of short line segments. The gcode program has a feedrate of 200ipm and a tolerance of .005". The first screenshot has an acceleration of 20in/s^2. Even at the largest radius - it barely gets above 80 ipm. The second screenshot has an acceleration of 100in/s^2. It approaches 170ipm. I also set the acceleration to 200in/s^2 and it pretty much stayed at 200ipm the whole way.
sam

[Xenomorph]

IMO I really do not think the control CPU clock cycles have much to do with its ability to execute motion. I do not know how many clock cycles the Haas control can do, but lets assume 100 megahertz (probably low). Now they say on their web site that they can execute 1000 blocks per second - that would be one kilohertz. With these assumptions the CPU would make about 100,000 clock cycles to execute one block of motion. This really indicates that there is something I am missing. I think it is more in line with how the control handles changes in direction. The key is for it to maintain as much velocity as it can and still be accurate. This must be some form of motion algorithm. I am really interested in a control that can be accurate and execute 3D motion as fast as possible and how they accomplish this (how they define their high speed motion).

I have also considered the look-ahead function of a control. Based on the simple math above I would still think that the 100 megahertz CPU would be able to calculate dramatic changes in direction with a minimal look-ahead. It will simply keep buffering new commands as it completes a block of motion.

I do not know how much influence acceleration and velocity have on high speed 3D motion. I think it may be more a function of keeping velocity as fast as possible during changes in direction. This still points to some sort of motion algorithm and how much velocity is carried through the motion.

Thoughts? Am I way off base?

[Edster]

Pulses per revolution is independent of servo update rate. You can think of encoder count as how accurate your speedometer is and servo refresh rate as how fast your foot can move the gas pedal. More/better information does you no good without the ability to apply it and change course.

Once you've got ten or twenty pulses for your minimum programmable move, your servo loop is what's holding back your accuracy at speed.

So what are you trying to say, the largest cnc control builder Fanuc's top of the line control with nano interpolation, ai contour control, and nano smoothing isn't as good as a servosoft system?

So all you need is 10 or 20 pulses for the minimum programmable increment? Hmmm....I guess fanuc better just throw in the towel. I'd be willing to bet that if they have encoders with 16 million pulses per rev the rest of the hardware is on par with the encoders.

[JohnJW]

How does Haas controller compare with Fanuc or Hurco or Siemen or ? and what exactly is in Haas hsm? The spec on it doesn't seem to change, while the main cpu has been upgrade a few times already.