What's new
What's new

Considering LinuxCNC, what limits rapid speeds?

huleo

Hot Rolled
Joined
Feb 12, 2014
Location
UT
We have a couple older machines that are worth recontrolling. I am not immediately confident Linux is the answer because it seems to written by non-machinist types that may not understand what makes a great control. I think about anything can "make it move", but its the little things and special features that make them jam.

If we upgrade a couple machines, one of the priorities is the take a close look at turning up speeds as much as possible. I realize there are physical and sometimes electrical limitations to what can be done. We have one machine that runs Yaskawa drives/servos/spindle. It runs 787ipm rapids and 8K spindle. We want to take a close look at where the limits are. I know ballscrews and thrust bearings could get hot if pushed too hard but I seem to remember a lot of the limits of 20yrs ago had more to do with the ability of the motion controller to read the encoders? I know we checked on one machine and the MTB most certaintly pulled every bit they could from the system at the time, but I want to determine if today's tech can allow a bit more from the same amps and servos?

I realize machines were built with "everyday/all day" in mind, and they seem to take some objective away. For instance, rapids and accelerations are fixed whether you have 100lbs or 2000lbs on the table, and that can change the load to the servos considerably. As well, spindles seem to be fixed to run "safe" at full speed all day long.


What we are looking at is possibly we can improve acceleration for our own needs or possibly we cannot escape acceleration limits, BUT, max rapid speed could be turned up because we can now read encoders 10x faster? Maybe our machine is fixed with 2G accel to account for max table load, BUT if we are light, maybe we can push to 3G accel, or remain at 2G but push rapids from 787 to 1000?

Regarding the spindle, maybe we can't push the spindle to a higher "safe" limit, BUT maybe we can consider a "little more" with a certain duty cycle in mind? Usually higher RPM is used with smaller tools and maybe we only need that extra speed for one tool, for 10min/hr.

Now, I intend to fully inspect the mechanical limitations, but I want to look more at the electrical limitations. Where might we find those limits? Is the switching fast enough in older amps?
 
I don't think you can gain very much without re-engineering the whole system.

The MTB is always trying to get as much as possible out of as little as possible to the point that many times such things as spindle horsepower become almost outright lies.

If you are after increased performance with a specific part, that is possible by sometimes re-tuning the servo system to handle the specific load. This only works for a specific part setup and then must be redone on a different part.

Increased processing horsepower can help in increasing error sampling and correcting but you are still limited by the physics of what the mechanical servo system is capable of. Unless you are willing to re-engineer the whole system, your gains will be limited and even then the existing machine design such as the mass of the table and saddle will still establish practical performance boundaries.

If it were me, I wouldn't waste my time on this and just buy the faster machine.
 
After having integrated a number of PC based systems in my younger years and reviewing the Linux based systems available today - and seeking to apply a control and never wanting to have a call-back . . . we only integrate Siemens now when it comes to CNC applications (which by the way, Sinumerik controllers have a Linux based NCU).

As Ziggy stated - there is a lot of engineering that goes into defining the maximum rapid speed and accelerations.

Most basic considerations include critical speed for the ball screw, motor terminal voltage (Back EMF) and what capability the drive might have to field weaken (you mention Yaskawa so forget about it - you are stuck with whatever max speed the motor is rated for). . . also peak torque capability at the motor may or may not be constrained by available current from the servo amplifier (drive) and you mention encoder rates - while this may be a limitation for some really cheap systems, this generally hasn't been a limitation for the last two decades.

I have integrated a number of Delta Tau based CNCs and have two in my shop that we use every day - those were the last DT systems we will deploy in a CNC application. In the last decade, we have done more and more Sinumerik 840D CNCs and it is hands down the most flexible, most powerful, and most robust system we have deployed and the best part is that once you are done with it, it just runs and runs and runs. We additionally get quite a bit of work modifying other's 840D implementations (especially the HMI) to facilitate easier operation of complex machines. Our largest machine was a 54-axis carbon fiber layup machine and our goal was to make it so that someone with a high school diploma could operate it - and for the most part we were able to make it very easy to operate.

On the topics above - the 840D Solution Line uses Sinamics S120 drives which can field weaken permanent magnet brushless servo motors using BEFM Suppression - so the feedrate axes now are limited by available torque at extended speed (and again, critical speed for the ball screw).

Extended Spindle Speed operation is more a bearing / drivetrain issue. Perhaps you need to upgrade bearings and lube to an oil-mist system.

In short - you have to look at each axis and evaluate every component from the table, bearings, ball screw, drivetrain, servo motor / amplifier, encoder feedback and finally the controller and its ability to manage the drive/motor combo if considering field weakening . . . (and perhaps even dynamic servo loop tuning as motor tuning changes when you enter into the field weakened operating zone). Again, the Sinumerik control excels in all of these areas and gives you rock solid performance managing all of this out of the box . . . you just have to figure out how to integrate and exploit these features which can be a lot of work and a lot of learning over several years worth of projects.
 
huleo
If you DO decide to go with Linuxcnc, you won't get anywhere with questions of Linuxcnc on this forum, other than negative response.
Best bet is go over to that other 'cnc' forum 'zone' or the Linuxcnc website where
they have a users forum.
 
Last edited:
Ah, I did not think this through thoroughly I guess. Correct me if I am wrong here but as I understand things, the back EMF is a known glitch with IGBT driven motors in which the returned flux is shunted at the IGBT due to the gate being shut. This same EMF has been exploited in sensorless vector drives?

As you increase speed, EMF increases, to the threshold of the drive? If you push beyond the rated back EMF ability of the amp, you must weaken the field to control the EMF? Plus, when decelerating, you then have to consider the extra power being returned between max and full rated RPM? I am guessing even if the are not exceeding the max accel current of the amp, it could still see an over current condition during decel unless steps are taken to mitigate that?

This kind of brings me to my next question of how more modern machine tools are commonly configured? In my mind, the safest/most economical way to increase machine performance would be to select a coarser or faster screw pitch, and use a larger servo motor to hold the load better? Would it be outside the norm for smaller machine tools to even exploit field weakening techniques?


I could certainly see advantages in large gantry machines for servo over speeding to get to their destination but maybe not with smaller machines?


NOw, since the goal here is to utilize the factory drives/servos, I might ask which criteria we should be considering to ensure at least all the electrical abilities have been exploited? IE, as long as we can see that max servo speed is already running at max rated speed, we simply cannot push beyond that without major hurdles? However, if max current is not being utilized, assuming we have headroom in the mechanics, we might push accel rates a touch?
 
huleo
If you DO decide to go with Linuxcnc, you won't get anywhere with questions of Linuxcnc on this forum, other than negative response.
Best bet is go over to that other 'cnc' forum 'zone' or the Linuxcnc website where
they have a users forum.

Thanks. that may be the case but I know there are real machine owners here that might highlight the real downfalls of Linux, at least at this stage of development. It seems to be a common choice for hobby machines but we are not messing around at the hobby level. I have yet to see a real VMC/HMC really running Linux. All I see is yay!, it did a tool change!
 
My bridgeport had a dead Anilliam crusader controller on it, it got swapped to LINUX cnc and i love it. But its not nore ever will be a vmc, just a really useful feature to add to the long list of quirks a Bridgeport can already do.

IMHO a EMC retrofit done well might gain you a lot, done badly its a heep of poo like any badly done retrofit. Theres a lot of advanced axis tuning and error allowance now available in it, again what you do with it is up to you. But you now very much can do dead nuts machining or roughing and just getting within X allowance of a given movement point which really lets you hall ass on roughing.

Don't mistake rapid for acceleration, acceleration matters far more on small moves on small parts than a high top end speed. Watch the you tube videos of brother VMC's running V Hass on the same parts and the same programmes and look at the time difrences, faster accelerations gain you a massive lead very quickly and saves a lot more time than another hundred inches a minute possibly would. Fast rapids are only a benefit on bigger parts. If you know what your rapids and accelerations are break out the calculator and do the maths as to just how big a move your machine needs before its at rapid, look at your work size, programme style then ask the question of just what point will rapid get lost to acceleration speed.
 
Thanks. that may be the case but I know there are real machine owners here that might highlight the real downfalls of Linux, at least at this stage of development. It seems to be a common choice for hobby machines but we are not messing around at the hobby level. I have yet to see a real VMC/HMC really running Linux. All I see is yay!, it did a tool change!

Most of the knockers of Linuxcnc here have never used it. You can make a perfectly viable VMC retrofit using Linuxcnc, if you haven't found examples, then your not trying hard enough. Go to Linuxcnc.org

If your trying to make a machine faster to save time/increase profits etc, have you considered the time required to implement a retrofit, debug servo tune etc etc? The time you spend on the retrofit will never be gained back by any incremental improvements you expect to gain.
 
I pretty much agree. Realizing a machine rarely gets to full rapid speed before decelerating when down in a cut, acceleration is a bigger part. One in which very few MTBs even publish, like HAAS!!

There is no way I am going to waste 6 months trying to dig about 400ipm out of the rapids, but it seems obvious that if a machine can handle its full acceleration rate at max table capacity, there is room to push it if we are no where near capacity. I figure MTBs likely don't even want to mess with toggles in the control for estimated weight on the table. However, I could be missing something because it would seem obvious that if I just use the servo and do a quick jog, I can determine the exact mass based on the inertia! I know if I was in the business, I would exploit that little bit to make that machine accelerate as much as it could.

It is possible that this gets pretty simple though, Push the accel rate right to the max current threshold and since the control or motion controller will get feedback of that current, it should be able to adjust this acceleration on the fly in the microseconds to keep it at max obtainable accel based on the current limit parameter.

I am probably overthinking this! It is possible that all we can do is determine the preset max current values allowable at the amps are being utilized and just focus on maybe improving the tool change speed a bit, and live with the new control improvement alone! I mean bring a tool to the new age of USB, ethernet, webcams, remote operation, probes, etc. Not to mention we will never have a program size or digestion problem again!
 
Thanks. that may be the case but I know there are real machine owners here that might highlight the real downfalls of Linux, at least at this stage of development. It seems to be a common choice for hobby machines but we are not messing around at the hobby level. I have yet to see a real VMC/HMC really running Linux. All I see is yay!, it did a tool change!

It is also one thing to get it running but then you have the issue of keeping it running. Not to imply the Linux approach has to be unreliable but the personalities that do the retrofit and set everything up probably won't be around when you need them for support.

Ask yourself,"Do I want to be personally married to this machine for the next X of years?"

If you are a small shop then this might not be a big deal but if you are a moderate size or larger it can be a major roadblock to keeping the machines making chips.

There is a reason that companies like to specify certain controls and or components throughout a facility, it makes support so much easier.

I would suggest that you have a set down with someone like Motion Guru that does this for a living.
 
I hear you, and one we have discussed in the shop. We see this as a test....to see if we should do other machines. We are a company that does 98% of our own repairs. Board level chip replacement, spindles, etc. I am sure we don't have the "best" tools compared to specialized companies BUT, what we have realized is we can diagnose and sometimes repair a machine way before a tech can even get in the door. Then you look at the cost savings. The second a machine goes down, we are on it!

The biggest hurdle we seem to face is a proprietary control systems. I guess since we build and repair our own computers, it seems logical that it could work really well for us. I guess time will tell. I think some of this is almost just intriguing and we like a challenge!


Quick story. Had a spindle drive go dead 2yrs ago. Right in the middle of a job, ALWAYS! Call MTB, no parts/service, replace only. $6K, no warranty unless tech installs, tune parameters, etc. Down for 7-10 days. Call a repair shop, $2500 for repair with warranty, with the repair and road trips, still 7-10 days. Pull drive, determine capacitor failures in the processor board, not even in the big power section. Next day'd all new capacitors for everyone for $10, back up and running in 2 days, still running.

Love soldering I guess.....lol
 
It isn't impossible to get under the covers of anyone's controller including Haas if you want to learn how to alter acceleration parameters. If you are really creative, you can create your own user variables and set acceleration limits on the fly with most controllers such that if you know your load is 1/3rd of capacity, you can increase your acceleration by 3x drawing the rated current from the drive that the machine was designed to deliver under full load.

I watched my son and his college buddies hack into the CanBus network on a Toyota Prius to get data they were using to test a vehicle guidance system - if you have the time and energy, why not spend it learning the control you have . . . go to tech school at Haas or go to a Siemens class or two.
 
It isn't impossible to get under the covers of anyone's controller including Haas if you want to learn how to alter acceleration parameters. If you are really creative, you can create your own user variables and set acceleration limits on the fly with most controllers such that if you know your load is 1/3rd of capacity, you can increase your acceleration by 3x drawing the rated current from the drive that the machine was designed to deliver under full load.

I watched my son and his college buddies hack into the CanBus network on a Toyota Prius to get data they were using to test a vehicle guidance system - if you have the time and energy, why not spend it learning the control you have . . . go to tech school at Haas or go to a Siemens class or two.

I did not mean to imply that we could not just retune our current control. We are not so much chasing a little better acceleration as much as trying to get some things we just can't get from the old control. Well, we probably can but.... at some point it is just old. All we wanted was a color LCD that we can actually see, some extra programmable outputs, and be able to run HSM tool paths. That will cost us over $10K..... What we want just isn't practical from the old control. The hardware just doesn't have the capacity.
 
Your limitations are as motion-guru stated complex.

However, modern sw based controllers are extremely good, fast and reliable.. given appropriate hw.
Most retrofits esp. done by amateurs use very cheap, crap, hw which kills the systems performance.
Your limitations are likely to be hw.

Your limitations will not be linux.
Not be reliability.
Not be speed or acceleration.

But the total system complexity is exponential, and good IO in and out and docs and safety is critical in business.
And it will be vastly cheaper to buy new servos as a kit.
Yes it will.

There are about 1000 discrete problems, in a VMC system.
90% are not documented, many wont work the same-as-now, and will need some level of custom PLC/hw/sw work from someone who knows what they are doing, more or less.
You don´t need an expert, but a good mechanic on that stuff be it sw or hw.

SO..
it is not too hard.. I suspect from Your description You could perfectly manage to solve every problem, and get a real working running full-blown VMC.
I suggest it will take you 3 good guys, 1 year, and 3 x 1 yr salaries + 10k$ in cash in fiddly bits to get it done.

I had a long reply drafted - junked it.
If You want, I can do a commercial-work list of things that need doing.

Short:
It´s not hard - but no-one has ever done it in the diy world, fully.
There is about 6000 hours work worth of gotchas in it.

And all the gotchas depend on a stack (2-3) of sw and hw components from various vendors and suppliers.
Exponential issues, like machine axis, n (issues) x b (issues).
And of course 90% is not documented, and neither the gotchas nor the errata, nor the actual working parts are well documented.

There is an excellent reason motion guru works with siemens.
Or others work with other systems.
Because their controllers are documented. Stable.
So the refit installer does not need to test, document, train, verify, have liability on complex controller features.
 
Well, lets assume 3 of us work blind folded so it takes us a whole year to control one machine.

Lets then relook at things. How much is a Siemens 840d kit? Assuming all we want is the control unit and software?
 
How much is a Siemens 840d kit? Assuming all we want is the control unit and software?

Depending on what machine you're redoing, there are less expensive options than the 840d. That's their flagship CNC and can handle lots of axes.
 
I think if we go through the hassle, we at least want 5 axis in mind but right now the machine has 3, and we have a rotary we want to throw at it.


I am open to other ideas. I just hate jumping through hoops and an hr on the phone just to hear "$50K-ish". NOPE. No way we will invest near that in these VMCs, thus the reason for looking at cheaper options.

However, the machines we have are GOOD machines. Just too good to walk away from.
 
The biggest thing I see- if you do linux cnc / the free stuff you better love that little devil because you likely just killed any trade/ resale available. The servo motors/ drives/ screws are the single most expensive part of recontrolling a machine tool- if you have to put bigger motors on to jack up the rapids/ give it more balls/ whatever other reason most of the advantage of a retrofit is gone IMHO. Until you hit the $ 300k replacement value mark I don't think you can afford to retro, and even then you'll spend $150k doing a retro on an old machine ( that may still have issues when you get done).

You can make money making parts or repairing machines..... but sometimes what seems like a worthwhile repair is a never ending task......without a warranty. I understand the financial side,,,,, but sometimes it just is not worth it.
 
2. No idea, at all.

1.
It is not blindfolded, at all.
3 guys can get one machine somewhat running in 1-2-3 weeks, if they have some skill.
It is the mickey-mouse stuff, edge cases, secondaries, that take all the time.

Do you have good MPGs ? How .. Via what ? What latency ? How many ?
Toolchangers ? How many ?
Probing ? Re-probe, point clouds, format, multi-probe to get it done faster, stl, etc etc..


It is because, as a clear and present fact, 95% of the 1000 issues, are not dealt with by the sw controllers, today, well, standardised, and documented.
Most are easy to solve, and document. Many are slightly different to your current control.
If YOU do not document it, You cannot sell it, as an ongoing business concern.

E.
Each feature must have clear specs.
We do probing via "xxx" by "yyy". Fine.
The feature "probing" as an example might be old, limited, simple, but it exists, has specs, works if done your way.

Most retrofits, if not all, are
1. hobby-style stuff of little commercial value, often high practical usage-value
2. or professional-industrial Siemens xxx type stuff.
3. diy self-built stuff of vastly varying capacity and quality


Many issues are also more difficult - because there is little-none of docs, experience, references, guides.
99% of installs don´t have toolchangers.
None, zero, of the very very few industrial refits have full IO from the toolchangers implemented. Afaik.

Full IO is lots of inputs, some not-hard programming per issue but with people who understand industrial cnc and programming.
All of this takes time.
And testing, and verifying.

And some few things have cascading dependencies, some on critical stuff.

I know of a few (2-5, no more.) installs where experts, really smart guys, have VMCs etc. running on their own stuff via sw controllers and various motion-control boards.
It took the experts about 1-3 years, to get stuff working ok.

Note:
The sw controllers work really well ! Core.
IF using good hw.
Much better on new hw, mostly, than the original stuff.
Or good old hw (much harder and more costly to get running well).

I am not *anti* linuxcnc.
It does not matter what platform You pick.

1.
Well, lets assume 3 of us work blind folded so it takes us a whole year to control one machine.

2.
Lets then relook at things. How much is a Siemens 840d kit? Assuming all we want is the control unit and software?
 
While Linux can be used for anything including running rockets, it's not the best solution for many things.
its hard enough retrofitting a machine with out figuring out the operating system.
MG likes Seimans, I like Bosch.
The 50k price is easily burned in support and wasted time.
 








 
Back
Top