Tool Length Automatic Touch Off Probe Error Analysis LinuxCNC
ฝัง
- เผยแพร่เมื่อ 29 มิ.ย. 2024
- Automatic tool length probing avoids manual measurements and touch offs and helps verify lengths already in the tool table. Probing tool-length offsets is done into LinuxCNC with an automatic tool length sensor. The sensor has two electrical connections to a Mesa 7i76e board for detecting the tool and for protecting against over travel, and an air solenoid controls an air blast to clear chips between the probe and tool. Reproducibility on a CNC retrofitted mill is high but accuracy is so-so. A good amount of the video is the analysis of sources of error.
Channel Love: Likes 👍 are free. paypal.me/KentVanderVelden
Business Inquiries, I'm a freelance scientific programmer: kent@confluencerd.com
Chapters
00:00 - Introduction
03:08 - Initial testing
04:27 - Backlash and homing error
07:07 - Spindle nose error
08:28 - Modeling tilt error
10:06 - Tramming
12:05 - Probe elevation
13:07 - Following-error
15:14 - Error modeling
18:21- Closing thoughts
github.com/kentavv/kvvcreates
www.notion.so/Tool-length-pro...
Parts from Amazon (affiliate links):
Automatic Tool Sensor: amzn.to/31exhhR
Air solenoid: amzn.to/3l5plHE
Silicone Tubbing 2mm ID x 4mm OD: amzn.to/3iYTF4M
Parts from McMaster:
51835K232 D.O.T. Brass Compression Tube Fitting for Air, Straight Adapter, for 1/4" Tube OD x 1/4 NPTF Male
62005K133 Air Flow Control Valve, Elbow, 1/4" Push-Connect Female Inlet x 1/4 NPT Outlet
5097T41 D. O. T. Hard Nylon Plastic Tubing for Air Opaque Black, 11/64" ID, 1/4" OD, 25 ft.
4429K412 Low-Pressure Brass Threaded Pipe Fitting, Bushing Adapter with Hex Body, 3/8 Male x 1/4 Female NPT
4464K562 Low-Pressure 302 Stainless Steel Plug with Hex Drive, 1/4 NPTF
5975K12 Aluminum Right-Angle Flow Rectangular Manifold, Anodized, 4 Outlets on 1 Side, 3/8 NPT, 6-1/4" Long
4450K3 Muffler, 3/8 NPT Male, Steel, 23 scfm @ 100 PSI Maximum Flow Rate
47865K22 Brass On/Off Valve with Lever Handle, 3/8 NPT Female
44615K463 Standard-Wall Steel Pipe Nipple, Threaded on Both Ends, 3/8 NPT, 4" Long
Parts from Northern Tools:
Milton M-Style Safety Coupler - Push Button, 1/4in. MNPT, Model# S99706
#KvvCreates, #LinuxCNC, #CNC - วิทยาศาสตร์และเทคโนโลยี
This video is pretty heavy on numbers, but the TL;DR version is that inexpensive tool length probes are surprisingly good and well worth checking out. Could be used for primary measurements or verification. (I've been known to transpose numbers in tool lengths when entering them into the tool table.)
its amazing how deep you went in to this :) in my opinion the biggest source of error is the probe it self. it has some springy action happening there. 1 it is triggering different at different locations, 2 it might not return exactly to the same point. why not making a rigid probe perhaps with laser sensor above it to slow down the decent for probing. this way you could face top of the probe flat and measure all sorts of tools. great video tho! lots of very interesting ideas.
Hi Fox, this drove me nuts. I measured the same set of tools so many times. The probed length is so repeatable. All can be shut down, restarted, rehomed and essentially the same measurement comes out if touching probe at the same position. But, I have one very-very last test to do... :)
In this video, he uses a laser to find the rough length quickly, and then a copper plate for the fine probing. It's pretty cool
th-cam.com/video/v-S4RXMdlrY/w-d-xo.html
The very-very last test results: After reloading a tool, and repositioning, between each measurement, the repeatability is still +/- 5um. I thought for sure that reloading the tool (hitting the drawbar with the hammer, cranking down the drawbar) would be the source.
@@kentvandervelden omg.. i forget you have drawbar. i have drawbar in my manual mill and i replace it with a drill chuck. draw bar takes ages to change. i think you reached the limits of the machine.
I’m really enjoying all of this CNC related content you’ve been making lately. Keep up the awesome work!
Thanks Justin! Already on to the next project and video. Hopefully, a getting a little better with every upload! Peace :)
This is something i have been thinking about to implement in my cnc mill.
Now that I have watched the whole video, I have to tell you that I really liked it.
Hi Aleks, definitely! Good addition if frequently changing tools, and some interesting routines to experiment with.
Bravo! I just purchased one of these setters for my router and now very excited to duplicate your tests and crunch the numbers. Thanks for sharing.
Thank you, and I hope yours works well. It's been nice having a tool length probe, first for switching drill but more recently when switching tools in a secondary high-speed spindle. The high speed has been getting a lot of use recently making wooden toys.
Great video as always Kent. I love how deep you go to figure things out. I just bought one of those tool Touch offs myself. 😁
Thank you! I about fell over when I saw the cost of the official Tormach one, but the
That was a wonderful rabbit hole. Thank you for the great distraction.
Thank you, I try for distraction on par with discussion, good or bad :)
That was quite an extensive rabbit hole. I picked up one of those probes a while back but am yet to get it setup. Amazed to see someone go that far into optimising the probing cycle. Thoroughly enjoyed the video. Keep up the good work
Hi Jonny, thank you! I should have shown the g-code for the probing routines. The updated ones will always be in the GitHub repo.
@@kentvandervelden I think you showed plenty. Still getting my head around linuxcnc as an escapee from mach3. I appreciate your attention to detail but practically dont think I will go to the depths you did. For me its overkill for the level of precision I require. Most entertaining certainly.
What a simply superb video. I’m just starting in cnc milling (haven’t switched my machine on yet!) and was wondering about where errors come from when measuring tool height - your deep dive into this particular method has a lot of take home messages for other methods too. Super interesting and thoughtful.
Tying down errors here, in tramming, work holding, and work position measuring - will go a long way to ensure cutting with multiple cutters - will register sequentially as one would like, and seem super important to understand and pick Methods within ones budget so as to minimise errors in each.
Thank you, I'm really glad you found the video useful. Having a small machine some problems are more pronounced and so it's better suited to the work than a heavy rigid machine. I'd like to understand where error comes from and how it might be corrected on a budget, in hardware or software. I guess it's a curiosity to have an excuse to learn more about mechanics :) After a machine is reasonably good, you might check out IPE's videos on software error correction:
th-cam.com/play/PLm4_p2x39CHgF_gRtyBFJgPCrsmbsnhPW.html
Uncanny - I just received one of these yesterday :) After a Linux Distro update (still on Stretch) and moving to 2.8, this is next on the list. Thanks for posting your config; I've known I needed to add some debounce to my existing probe setup for awhile. Referencing your config will definitely be welcome!
Hi Ron, that's pretty neat :) If you hook up the airline, I listed the parts that I used in the video description, mostly McMaster stuff. I run the air solenoid at 90PSI, but put a small restricter on the inlet. 2mm ID silicone hose worked well, with 1/8" NPT barb fitting, with copper wire used to secure the hose. At full 90PSI the silicone tubing would balloon up :)
@@kentvandervelden Thanks. I hooked up a number of extra solenoids when I plumbed in my ATC a year or so ago. I think I'm stepping down shop air to 60PSI for its operation but need to check. (also blew my Mesa card because I didn't think I'd need flyback diodes for the solenoid :( . The restricter doesn't seem a bad idea regardless)
Very interesting. Thank you! I have a renishaw touch probe on my lcnc machine (flat carbide contact surface). As you found the errors come from every direction and all we can do is try and reduce them overall. I suggest a protection cover for your probe and stmbl drives with real servos 😁
thanks for the information, very good
Thank you for commenting. Glad video was helpful :)
Quite a significant analysis. Have to say though, parts that requires that sort of tolerances wouldn't be made on such thermally stressed and rigidity impaired machine-tool anyway. A tolerance of 0.03mm is normally sufficient for hobby spacecraft making.
great video very informative, i have the exact same tool setter for the price i think it outperforms its real world monetary value, +1 subscriber
Great job，Thanks！
Great Video!
Glad it was helpful :)
@@kentvandervelden Will you show the settings, how to configure it?
Or just simply use a probe input?
Sorry that I overlooked that. All my configuration files are on GitHub, and link should be in video description. Mostly, just connect the tool length probe to the probe input in LinuxCNC, add a lowpass filter, and everything seemed to work. If I make a video on the Drewtronics probe, I'll be sure to explain the necessary code. It's the same code as the tool length probe.
You are pure awesomeness
Thank you brother :)
Accuracy in probing is of course needed If precision is required in the final parts. But try this: load any tool in your mill, set up a dial indicator between the mill column and the tool and then apply force, by manually pushing the tool towards the column. My guess is, that you will see significant deflection of the mill, and compared to this, chasing um when probing might not be worth the effort. Anyway, I really liked your video. I learned a lot. Thank you for all your awesome videos
Thank you, no doubt I put time into a measurement that'll be overwhelmed during machining. I just wanted to see how close I could get with an inexpensive probe :) Pushing reasonably hard, left, right, top, bottom, I can get +/- 5 thou movement of the head. Oh, and the spindle deflects 2 thou too, but in only one position. That will be a future repair for sure. With this deflection, would a 1-2 thou error measuring length manually matter? I've learned to live with some of the deflection, for instance knowing that when profiling, I must raise a couple a couple thou else the floor will be cut too. I would like to someday work with a big machine and see if stuff just works out of the box or if similar tweaks are needed. Let me know your suggestion of the next improvement to make. I finally have some dedicated time now.
Wow, that actually surprises me. I had the same machine once, and I was able to flex the tip of the tool almost 0.2mm by pushing it - and I already had the column bolted to the wall to add further stiffness. I can see you don't use the original spindle, maybe that is the difference (and maybe my machine had a production fault. I bought it new).
I now have an old Aciera F4, which is much stiffer (and heavier).
I see lots of videos on the topic of upgrading to cnc and chasing precision in the new parts and the electronics. 1000's of dollars are spent on ballscrews, spindles etc. But the frame is often extruded aluminium profiles or light weight cast iron and I often feel much of the precision (and dollars) must be wasted. I would love to see the mechanical topic investigated
For repeatable homing you could rely on switch + encoder index, if you have access to the encoder signal of the motor. That’s one of the changes I need to do on my machine.
Hi Vasi, great idea! I just received some HD15 breakouts to hack into the encoder signals :) The driver manual does not show an index signal, but maybe the encoder provides an unused one? Will soon test.
Very nice
Thank you :)
are you sure about the "overrun signal"? i think you're using the wrong switch for the measurement ;) usually those tool length sensors are supposed to work the other way around: the tool moves at high speed until the first switch contact signals the "initial approach distance". then the movement slows down and continues until the second contact which is the actual measurement... as a protection - you know how much travel the sensor has until hard stop, so limit the move to that distance after the first contact. and if you get second signal without the first in advance - the first switch failed, then stop immediately. still has to be enough overtravel to decelerate from rapid speed.
Please send an example of the probe you are describing. From what I've seen, the first trip use a kinematics mount while the over travel uses a basic switch. The only tear down I've found is at martyscncgarage but he may be mistaking the over-travel switch for the primary switch. Without taking this probe I have apart, I rely on the manufacturer's instructions, and in which case I've wired and am using this probe correctly. I've not tested the reproducibility of the over-travel switch. Thank you.
th-cam.com/video/3HR2bPmyirs/w-d-xo.html
@@kentvandervelden oh, ok, if so - i may be wrong. i was retrofitting an old mazak with llinuxcnc a while ago and it had a pneumatically actuated tool length sensor arm with 2 switches working the way i described. but the plunger had much lager travel and distance between switches. like 50mm. you could hit it with pretty fast, gently decelerate and creep to the measurement point... thought this thing works the same way. my bad :)
Great deep-dive on this. Did you consider the runout introduced with the TTS interface between the tool and spindle nose as a potential factor? I did some exploration of error in the TTS/spindle interface several years back and found that clocking of the holder in the spindle produced measurable changes in runout. Drawbar tension was also another factor. This seems like it would potentially translate to error with tool length probing unless you could guarantee clocking of the holder, drawbar tension, etc. I suspect there's also additional variability introduced with the position of the collet inside the holder, tension on the collet closer nut, etc. I think your tramming of your mill column and head using the ground rod in the tool holder could also be subject to tool holder/spindle errors - in other words, if you trammed it perfectly with one holder/collet/rod setup, I suspect if you put the same ground rod into another holder/collet setup, you'd probably see error. It's a deep rabbit hole - great to go down it to understand the sources of error, but also important to find a realistic point to stop chasing errors on a hobby machine and just move on with making some parts! Cheers!
Also, if you missed it, Cliff/Threadexpress in NZ has done a lot of interesting analysis on these tool setter probes. TH-cam doesn't seem to let me post the links, but check his channel and search for "Toolsetters."
Thank you! For each setup, a complete set of measurements was taken: all tools were measured, never removed from their holders, always clocked (within a few degrees), and drawbar torqued. While torqueing the drawbar, I subjectively applied a counter force to the mill head. I also tried touching off stacked gauge blocks of various lengths to learn about the ballscrew error. I was trying to not need high measurement accuracy. If I could get high reproducibility (which seemed to be so), I could model and correct the error (which I failed to do.) There's the troubling rabbit-hole question: with high reproducibility, what am I doing wrong to model the error? Of course you're right, in the end, the accuracy is well beyond what I can make use of :)
Here are Cliff's videos, parts 1 and 2:
th-cam.com/video/wDGvQHevN9Y/w-d-xo.html
th-cam.com/video/pOCxNe3c4x0/w-d-xo.html
Exactly. So many potential sources of error from every direction. Use best practice, best tools you can afford, make and measure! If you know what is happening and why you can keep moving in the right direction...
Very nice video! I have the same issue that the probe trips during machining like you mentioned in the video. How did u fix this?
A few ideas you could try are: 1) disable the probe input if the program is running; 2) filter the probe input; and 3) use a switch on the probe. In LinuxCNC you could and2 the probe input with halui.program.is-running and halui.mode.is-auto. This might not work though... if a probing operation appears as a running program. The filtering operation is probably the simplest and a LinuxCNC debounce module could be used but it adds latency (so move slowly and all will be fine.) The switch could be physical, or use a digital output and2'd with the probe input, where the digital output (the switch) is triggered by m62-65. Probably others ways... I'd start with filtering and if that does not work well, use the switch. Of course, use shielded cable and prefer higher voltages if possible. Hope something here was helpful.
@@kentvandervelden thanks for the reply! i will try these solutions soon.
Did you consider the influence of the debounce? In 20ms delay the mill travels 67µm with 200mm/min search speed.
Hi Robert, the tool's final sensing speed is 10mm/min, so maybe 3.35um of uncertainty? The final speed might have always been 1/10 of the initial speed, and I tested different approach speeds, but I'm not seeing anything in my notes or the video about explicitly testing different final speeds. I certainly could have overlooked what you're suggesting. Would be embarrassing :) Thank you for asking
I dropped my touch-off probe which looks identical to yours. The center piece popped off and it appears it was only held on by a non-precision glob of glue.
Not to be confused with a precision glob of glue :) Could you see and identify the primary trigger mechanism?
@@kentvandervelden I could not see the trigger mechanism (or any way to open it up further). Just appeared to be a hard disk glued to the softer top. Sorry I didn't take a photo before regluing.
You talked about the convexity of the surface of the tool length setter. You even plotted that convexity.
Could you lap the surface of the probe to make it flatter? Maybe within a micron?
Thanks! :)
The surface may be very flat but the sensing mechanism have a non-flat response surface. If the pressing off center, maybe the surface tilts slightly and registers a different length than if pressing in the center. I suggest this only because of relative difficulty of making a surface flat vs. making a mechanism move perfectly parallel. Thank you
@@kentvandervelden that makes sense. Looks like there's a lot of room out there to develop a much better tool height setter.! A lot of room.....
@@artmckay6704 its like 40 gbp us around 50 dollars, what kind of accuracy are you expecting at this price point lol
I would be interested to see the tool diameter, tool length and tool weight graphed in the same order as the chart at 17:06 to see if there is any correlation.
Hi Matt, I like the way you think! While not the same as a graph, the r^2 for a linear regression between error and diameter is 0.41 (after dropping the fly cutter, Shear-Hog, and facemill) and between error and length is 0.61 (including all tools.) I dropped the three tools in the first regression because they were way off the regressed line (and dragged r^2 to 0.7). I don't have weight measurements.
Kent VanderVelden is it possible that your ball-screw has travel distance error or fluctuation across that distance? Or differs from the range that you calibrated the axis movement to?
@@molitovv Absolutely! When I get home, I'll stack up some gage blocks and test their heights. Most of the parts I make are
The test was to use a Haimer and the LinuxCNC DRO to measure the distance from the top of the vise fixed jaw plate to the top of gage blocks, always approaching from the top. (The gage blocks have length specs way better than I could verify.) At least over 4" the mill measured lengths are surprisingly good. Was hoping you were on to something about the ballscrew, though I sure would not want to need to fix that mechanically :) (may be able to correct it in software) Would be glad to test other ideas. This was a great one!
Fixed jaw plate: 0.0000"
0.1" block: 0.1002"
0.5" block: 0.4999"
1.0" block: 1.0004"
2.0" block: 1.9994"
3.0" block: 2.9995"
4.0" block: 4.0000"
Kent VanderVelden I think that I would remove the tools from the equation in order to reduce the variables. Perhaps zero the spindle nose to the tool length sensor then move your height gauge from the surface plate to the bed of the mill. With the height gauge probe zero’d and contacting underneath the spindle nose, you can check repeatability of the whole system at any different Z heights. You can then move the mill to any position and at any acceleration and check if your Height Gauge agrees with the LinuxCNC DRO. I think that this would make it very easy to set up an iterative process by which you could discover the root cause.
You can prevent switch bounce problems by using a normally closed switch
Good video, Kent. I was following it along and was ready to comment "what is the probe surface or response isn't uniform!", but you saw that coming. ;) I have one last probably wrong set of suggestions. First is are you using the same reference surface between when you have the tool mounted in the machine vs when you have it on the surface table? The second is that you are doing the tests differently. In one, the tool hangs and in the other, the tool is the opposite--it stands up. How much error do you expect due to gravity? I imagine that would vary based on stiffness, mass distribution, and length. Maybe a little based on geometry, too. What was it the mech-E's are always telling me? Oh, yeah, everything is a spring!
Hi David, not easy questions... The same surface is used in both cases. The tool holder rests on the surface plate on the same face that pulls against the spindle nose. I'm less sure about error due to mass... but the error seems to have no qualitative correlation to error. I didn't weight each tool and do the regression, but looking at the errors, I see the big and small tools towards the high end of error.
@@kentvandervelden That's good enough for me. Thank you for considering my silly ideas.
@@davidwillmore Not silly at all. I was reading a paper today on sources of errors in machines. In their diagram of sources were multiple springs at each joint :)
@@kentvandervelden I guess I should keep listening to all the MEs, then. ;)
I used the mesa thcad 2 A/D converter, a touchplate , g38 commands, and the ohmic.comp in linuxcnc to read frequency put out by the mesa card when 24v is going through the circuit so one lead goes to the cutter that has 24v on it and one lead goes to the touchplate from the mesa as the tool gets closer to the touch plate the frequency starts to go up the closer the 2 leads get, I set a threshold for when it reaches a full 24v and the g38 commands do the rest I think this method gives you the most accurate way to set tool heights would you be able to run some tests to find out
Im applying a component from QtPlasmaC torch height control dubbed hypersensing its using ohmic sensing and software instead of relays to trigger the surface contact g38.3 probes to the piece and g38.5 slowly moves away from the surface and the moment contact is lost that's the height I think its more accurate because it removes the time a relay takes to trigger I could be wrong I can send you the custom hal file i used if your interested
hello sir I also want a tool like this plz can u let me know the connections from this tool to vmc
Did you connect the yellow wire to resistors?what type of resistor did you use?I'm using 2.4k ohm resistors and its not working,,help me
Attaching the probe that I received to a Mesa 7i76e didn't require an external resistor at 24V. The diagram in the instructions show a resistor in probe. If your multimeter uses a voltage greater than the Vf of the LED, you can measure if the probe has a resistor by measuring resistance across the probe light leads. If meter reads zero ohms, try flipping the leads. If reads zero ohms both directions then meter probably uses too low of a voltage to get the LED to conduct.
Hello kent, maybe a stupid question but, the yellow wire is the signal wire and the green wire a ground wire? i'm realy confused with this probe
There are two common varieties of this probe: a four wire and a six wire version. The four-wire version can be thought of having two switches (while the six-wire version is more likely an inductive sensor.) With the four-wire version that I have, the yellow and green wires together connect to the main probe switch. The other two wires connect to the normally-closed over-travel switch. The electrical symbols in the diagram near 1:11 are a little confusing... perhaps there is a four-wire version with an LED? If you ignore that and think of it as a switch, the wiring may be easier. Regardless, honor the direction of the LED just in case. Let me know if this did not help and if you need me to double check wiring on a Mesa 7i76e.
@@kentvandervelden Hey kent, thank's! i found out that my board is damaged. i thought i did something wrong
I just bought one of these and i cant read the manual and the wire colors dont match the few english words in the manual. whats an easy way to determine circuits without the control board being online. im still waiting on my 9014 frame but i do have my gecko g540 controller and all electronics but i have to install 4 gx-16s for the homing switches and the tool probe and i have to wire in the relay for the spindle control. I have everything but the frame but its all in boxes and i havent started anything yet because i still have to weld up a table. Im making a full steel table to be able to ground the machine through its base to the outside ground.
If there are four wires, and no LED, you could try using a continuity tester. The probe that I have has a tool-length switch and an overtravel switch. With the probe electrically disconnected from anything else, try attaching the continuity tester to each pair of wires and gently pressing the probe down. After the first switch is found, you'll know this method should work and can look repeat to find the overtravel pair.
@@kentvandervelden thank you
Hi..
What type of homing sensor did you use? is it Proxymity sensor?what type?
Thanks
CB
They are inductive proximity switches, PNP, NC. Reproducible and small. Search for AHS-CP-1A from Automation Direct. Here's a video about their installation. Thank you
th-cam.com/video/rrgeKOQWOGA/w-d-xo.html
@@kentvandervelden thank you so much good sir, thank you !
I really wish there was more detail in the actual setup. Haha.
If I would have spent anymore time on it, my remaining hair would have been pulled out :)
@@kentvandervelden lol. I know the feeling. Complete newbie to Linux and CNC in general, so this PrintNC build I’ve been working on has been more stressful than anything.