Advanced Search

I've attached an A-axis probing routine - it's from PathPilot.  I think it's suitable for cylindrical stock along A, not really for prismatic stock.

I've not tested it, but it'd probably be a good starting point for someone to test/tweak and then submit to the ProbeBasic devs for inclusion in PB mainline.

2. Run the macro — it probes at A=0° and A=180° in Y, calculates the midpoint, and sets G52 Y to the center of rotation

For this to work the following must be true:

• You have some sort of TCP enabled, or
• You adjust rotational point to stock offset in CAM after probing, or
• You have very simple parts, with very simple programs, and
  • an internal macro/subroutine 'adjusts' the Z/Y positions for every single point in the program based on the probed offset.

I think the easiest way to explain why (I think) your proposal won't work is to make a diagram.
 


If you have TCP that takes care of everything.
If you adjust CAM after probing, that also takes care of everything.
But just probing and shifting Y will result in bad parts unless you're only machining at one A-position.

I didn't draw the other problem... but not only will features not line up, the part will orbit the actual CoR, which means at A90/A-90 the Z-position will be higher/lower than CAM expects it to be.  Maybe crashy-crashy.

The only reason to probe is if you're doing 2nd op parts and have to line up features with existing surfaces.  Even then you'll probe, adjust CAM (or TCP), and pray.

No skim cuts, no calipers, no manual offset math — roughly 30 seconds per part.

Or use a work stop that gets the stock centered enough to take care of a little excess material - roughly 1 second per part.

The cause has been identified to be a side effect of the bug fix mentioned above. You can revert this by finding 'gremlin.py' on your system (I sorry but I do not know where that is on a package installation) and change this
 

to this:

 

This will revert the bug fix.

I'll have a look at getting this fixed without regression.

 

Thanks for the detailed writeup — very helpful, especially the encoder index homing tip (I use A6 Servos with absolute encoders myself)!

For the daily workflow with changing stock I would like to find an easier way to avoid the manual indicating and skim cuts: a small probing macro that finds the rotary center automatically and sets G52.

The idea:
1. Clamp the stock (no need to center it carefully)
2. Run the macro — it probes at A=0° and A=180° in Y, calculates the midpoint, and sets G52 Y to the center of rotation
3. Swap to the cutting tool and run the program

With the WCS origin set to the rotary axis in CAM (Y0 = center of rotation), one single setup covers all rotation angles. No skim cuts, no calipers, no manual offset math — roughly 30 seconds per part.

G52 persists across resets in LinuxCNC by default, so the machine-level calibration (the fixed offset between G54 and the rotary axis) only needs to be done once after homing. The macro then corrects only for the actual clamping offset of each individual part.

What do you think?

Might be worth adding something like this to the ProbeBasic probing tab — the Rotary Axis tab seems like the natural home for it.

Please let me know if you make any progress , I am planning on using two cia 402 motors for the tool changer but don't know where to begin.

Hi,

Yes, the internal homing routine needs to be triggered one more time; after that, the axes will remain permanently referenced.
I am using internal homing method 4.
Once that is complete, you can comment out or delete everything related to homing.
My home and limit switches run via EtherCAT. This allows me to save a significant amount of cabling.
My XML file is nothing special.

It seems LinuxCNC runs great on the majority of PCs. I've had a lot of network and USB problems with several min-pc's I bought new, cheap, for around $100 USD. With desktop, gaming, or workstation type PCs you will find occasional models that refuse to cooperate but most work.

I've test run LinuxCNC on my GMKTek mini-pc with a Ryzen 7 5825U cpu, with good results, but I recall someone recently complaining about a similar model.

If you use 1-10 year old hardware I'd expect a good chance of success. But that number range is no strict rule in any way. I'm typing this on a ~11 year old pc that tested fine on LinuxCNC, currently running straight Debian 13. With a ~9 year old graphics card and 16GB RAM it is still very usable.

Even brand new less than 1 year old often works great, you just run a greater chance of not-yet-fully supported hardware.

A couple days ago I installed LinuxCNC on A Dell Precision T5820 for a customer, an 8 year old server grade workstation with massive overkill computing power. Everything works so far except the new 8-port non-branded network card bought separate from Amazon was junk.

forum.linuxcnc.org/18-computer/39371-res...or-use-with-linuxcnc

forum.linuxcnc.org/18-computer/54979-new-pc-build-asrock-n100dc

forum.linuxcnc.org/18-computer/57073-hp-...or-use-with-linuxcnc

forum.linuxcnc.org/18-computer/38838-har...ency-tests-used-pc-s

forum.linuxcnc.org/18-computer/54369-use...mesa-ethernet-boards
The thing with laptops is, despite getting the same exact model, still might not work. Happened to me only once, so it can happen.

Hey everyone! 

I'm pretty new to LinuxCNC and I'm trying to figure out what kind of computer or hardware setup works best for running it smoothly.

I've been reading some posts about latency issues and it got me kinda confused lol. Like, does it matter a lot what CPU or motherboard you use? And is a regular desktop PC always better than something like a mini PC or a single board computer (like Raspberry Pi)?

Here's kind of my situation:
- I don't have a huge budget, so I want something reliable but not super expensive
- I'll mostly be using a parallel port or maybe Ethernet-based motion controller
- Low latency is really important I think, but I don't fully understand how to test for it yet

I've been browsing some options online, maybe these can help others too:

www.amazon.com/s?k=mini+pc+intel+n100

www.ebay.com/sch/i.html?_nkw=dell+optiplex+linuxcnc

etechdevices.com/collections/pc-servers-software

If anyone can share what PC or hardware setup they're using and whether it runs well, that would be super awesome! Even just a quick "this works great" or "avoid this" would help a lot. 

Thanks so much in advance, you guys seem really helpful here!

It sounds like you either need to (1) make your stock a little bigger and *mill* it true after mounting, or (2) probe the part and update your origin in CAM. It would be cool if there were a kinematics layer that could do the G52 equivalent in a rotating frame.
If you're doing a production run, could you put in the equivalent of a vice stop on your chuck, so that every part is at least in the *same* place?

Thanks for the quick replyfind attached my ini file and the Gcode, you will see in the Gcode in the first few lines that I have added G07, with this in the Gcode everything works as expected (and how it worked in linuxcnc 2.8.4)
revoce the G07 from the Gcode and the 'bug' will be self evident when saving the modified Gcode file

I program (F360) using the center of rotation as Y0Z0.

Your example shows a self-centering vise, and I assume it'll be connected to the axis with a zero-point system.  If your adapter plate is adjustable you can indicate the adapter & zero-point chuck in very close.  From that point on the vise should self-center close enough.

For round parts (3-jaw chuck) treat it like a 'set-true' lathe chuck.  Chuck up the part and indicate it concentric with the axis of rotation (dial test indicator). Loosen the chuck on the backplate and tap it around.

Once it's turning concentric the Probe Basic 'Ridge & Valley' probing routine (in Y) will find Y-axis center.  Once you have Y-center, probe in Z and then adjust Z-offset based on stock diameter.

As far as probing the center of rotation for a prismatic part - assuming you've adjusted your vise & zero-point to be on center...

• Z-axis
  • clamp a piece of stock in the self-centering vise
  • Skim cut one side using Y-axis to establish a flat plane - doesn't need to be a huge area
  • Rotate 180
  • Skim cut opposite side at the same Z-position - preferably without changing Z-position between cuts
  • Measure part thickness (and checking for taper to ensure 180 degrees is actually 180!)
  • Probe part in Z, setting Z0 to top of part (probably automatic in probing routine)
  • Adjust Z-offset by half the measured thickness.  Z0 should now be center of rotation
• Y-axis
  • Probe your vise on one side (jaws horizontal)
  • Rotate vise 180
  • Probe same vise surface again, approach from the opposite Y-direction
  • Difference between these two is the Y-axis center.
  • From that point use a scale, calipers or a jig to center the stock in the vise along the jaws.
• Bonus
  • Write down the machine coordinates once you've found center of rotation in Y and Z
  • Next time you home the machine, repeat the above indicating and compare the numbers.  Any difference is the error window for your machine's homing position.
• Double-bonus
  • Avoid removing your 4th axis at all costs so you don't have to re-indicate
    

For repeat prismatic parts, you can machine or 3D print a fixture/jig to help center the part in the vise jaws.  This will get things close, and from there just program a little extra material in your stock.  You will air-cut a bit, but this will avoid taking a heavier cut than desired.

The above headache of indicating is why I switched to encoder index homing on my A-axis, and am in the process of doing the same thing on my XYZ axes.

Most important: You have to modify the postprocessor to enable tcp.
With 5 Axis you really want to use tcp.

Looking for some community advice:
I’m planning to add a 4th axis (rotary A-axis) to my milling machine. In principle, that’s not a problem. What I’m wondering is how to best handle situations where the stock isn’t clamped perfectly centered in the chuck/vise.I think it’s unrealistic to expect the workpiece to be mounted absolutely concentric every single time.I’m using the Probe Basic interface. I checked the probing functions to see if I could probe and compensate with G52, but that doesn’t seem to be possible.
How would you approach this? Would you probe the offset somehow, use a different work coordinate system, or handle it another way?I’d appreciate any tips or suggestions!

I have a rack-based toolchanger/toolsetter at the back of my CNC router.  My custom M6 gcode saves the X,Y location at the start, and moves back there at the end (at max Z, of course).  Often the next thing that happens is a rapid to a different X,Y.  I'm wondering if there's a safe way to not restore the position after the M6, or if there's some magic to restore it only if the next move is anything other than a G0XnYn move (i.e. G1 or G0Zn)?

Not sure, works for me:

mesaflash --device 7i96s --addr 10.10.10.10 --write 7i96s2p_7i74d.bin --reload

md5sum 7i96s2p_7i74d.bin
b77b9faf82730e3acf82c2cdcba6e022  7i96s2p_7i74d.bin

mesaflash --device 7i96s --addr 10.10.10.10 --readhmid > 7i96s2p_7i74d.txt

gives me: