Candle
Candle copied to clipboard
Denvi
Height map irregularities
Hi Denvi, Many thanks for a great piece of software. I'm using Candle 1.1.7 with a Vevor 3018 CNC mill to mill PCBs.
I undertake a height map before any PCB mill, and tick the box in Candle to use the map, but I'm finding that the height map doesn't seem to be providing the height corrections that it should. The cutting tip (0.1mm, 30deg) cuts down through the copper layer in some areas, but then only cuts part-the-way-through, or not at all, in others.
I decided to check the consistency of the height map process, and ran 10 consecutive height maps over exactly the same area (10 point x 10 point, height map probe feed =10), and looked at the .map files these runs created in Excel.
I've been concerned to find some dramatic and variable differences between the (what-should-be identical) height map coordinates.
I selected the first (of 10) height map as a reference, and then subtracted each subsequent height map array from the reference, and looked at the resulted difference arrays.
These different height map arrays showed fluctuating differences, both in their magnitudes and distribution. While most deviations fell in the -30 micron to +30 micron range (still surprisingly large, considering that the copper layer on a 1 oz PCB is 35 micron), several difference maps showed some points that were 200 to 900 micron deviant(!).
These latter figures are clearly in error, but it seems they all might be, as no two deviation arrays were the same. I would have expected to see typical machine and randomisation errors of 0~5 micron, perhaps, but not 20-30 micron, and certainly not 200-900 micron.
Could I ask your comment on what you think may be happening here? I can send the Excel file with the height map difference arrays, if it's useful.
Many thanks for your help, Glen
Increase step accuracy and decrease point homing speed.
Because how many steps do you have per rotation?
Hi Jarewa, Thanks for your advice.
How do I increase step accuracy, and where do I find the steps/rotation?
The point homing speed is (default) 10(mm/sec?), which seems fairly slow already. Do you have a recommendation that typically works well? Thx…
On Sun, 24 Apr 2022 at 11:48 pm, Jarewa @.***> wrote:
Increase step accuracy and decrease point homing speed.
Because how many steps do you have per rotation?
— Reply to this email directly, view it on GitHub https://github.com/Denvi/Candle/issues/550#issuecomment-1107845487, or unsubscribe https://github.com/notifications/unsubscribe-auth/AY3ZUNUYKJXYPQJZGZOTZN3VGVGLVANCNFSM5UGGYDGA . You are receiving this because you authored the thread.Message ID: @.***>
Send the command $$ and write what you have under the items $100 $101 $102
10(mm/sec) the slower the better
Hello, The Feed rate is specified in "units per minute". For G21 it is "mm per minute".
@104TMR , I would recommend to check the accuracy of the probe process itself. Repeat the following command sequence in console:
G21 G91 G38.2 Z-40 F10 ( do the probe and check the last number in response [PRB:0.000,0.000,-1.031:1] Z=-1.031 )
G21 G91 G0 Z1 ( move the spindle 1mm up )
Check the dispersion of Z probe response. If it is stable enough, then it makes sense to continue your investigation. ( I've got -0.011 .. -0.013 for first 5 probes, then it changes to -0.028 .. -0.031 for latter 6. But I'm using V-cut as a probe for copper PCB and each probe may push in a copper layer a bit. )
The probe process is sensitive to feed rate: the lower feed rate the more accurate result. Candle 1.1.7 does not use "Heightmap probing feed" setting value at all. The last movement feed rate is used. So, try manually execute the sequence above right before the heightmap probe and control the feed rate in a left-top corner. Consider unofficial v1.1.9 with fixed "Heightmap probing feed" setting usage and reasonable small feed rate configured (10 - 30 mm/min).
$101=$102=$103=800. The probe feed rate is at 10mm/sec.
However, I suspect I may have found a contributor to the problem: the polarity of the probe electrical connections (to pins A5) were reversed from what I had thought I’d connected them: positive was on the work piece and negative was on the probe.
I decided to check the electrical signals when I noticed that one of the height map probe excursions only went part-way down towards the PCB surface, but then pulled back and recorded (an incorrectly high z-value), without actually having reached the copper surface.
Clearly, the A2 pins had received some (bogus) electrical signal that told the controller that the probe had reached the surface, when in physical reality it had not.
So I put an oscilloscope on the A2 pins to try and see what electrical behaviour might be occurring to cause this, and then noticed the reverse polarities.
I’ve since reversed the polarities of the A2 connections to the probe and work piece, and run another set of 10 identical height maps, and found that the unrealistic (200-900 micron) differences between them are not present anymore.
I am seeing maximum differences between height maps of 10~20 micron, which is still a bit concerning, compared to the 35 micron thickness of the copper, but nothing like the outliers seen previously.
I've also just tried milling a PCB using a new height map on a new board, but the results are not looking good. Please see attached pic. Shows severe undercutting and overcutting in various areas.
Further thoughts appreciated.
On Mon, 25 Apr 2022 at 9:25 pm, Jarewa @.***> wrote:
Send the command $$ and write what you have under the items $100 $101 $102
10(mm/sec) the slower the better
— Reply to this email directly, view it on GitHub https://github.com/Denvi/Candle/issues/550#issuecomment-1108447286, or unsubscribe https://github.com/notifications/unsubscribe-auth/AY3ZUNQKVCY5D6QXSEWSL6LVGZ6MBANCNFSM5UGGYDGA . You are receiving this because you authored the thread.Message ID: @.***>
Thanks for your advice, Max. You will see I have just posted an update, describing the polarity reversal issue I discovered. But there still seems to be some significant height map deviations. I'll investigate your suggestions tomorrow and will advise.
On Mon, Apr 25, 2022 at 11:36 PM Max @.***> wrote:
Hello, The Feed rate is specified in "units per minute". For G21 it is "mm per minute".
@104TMR https://github.com/104TMR , I would recommend to check the accuracy of the probe process itself. Repeat the following command sequence in console:
G21 G91 G38.2 Z-40 F10 ( do the probe and check the last number in response [PRB:0.000,0.000,-1.031:1] Z=-1.031 ) G21 G91 G0 Z1 ( move the spindle 1mm up )
Check the dispersion of Z probe response. If it is stable enough, then it makes sense to continue your investigation. ( I've got -0.011 - 0.013 for first 5 probes, than it changes to -0.028 - 0.031 for latter 6. But I'm using V-cut as a probe for copper PCB and each probe may push in a copper layer a bit. )
The probe process is sensitive to feed rate: the lower feed rate the more accurate result. Candle 1.1.7 does not use "Heightmap probing feed" setting value at all. The last movement feed rate is used. So, try manually execute the sequence above right before the heightmap probe and control the feed rate in a left-top corner. Consider unofficial v1.1.9 https://github.com/mar0x/Candle/releases/tag/v1.1.9 with fixed "Heightmap probing feed" setting usage and reasonable small feed rate configured (10 - 30 mm/min).
— Reply to this email directly, view it on GitHub https://github.com/Denvi/Candle/issues/550#issuecomment-1108585288, or unsubscribe https://github.com/notifications/unsubscribe-auth/AY3ZUNWSEEEMNQQQAXXDFQLVG2NVTANCNFSM5UGGYDGA . You are receiving this because you were mentioned.Message ID: @.***>
Hi Max, I tested your two lines of code, and got z-depth readings that stayed within 1 micron of the 1st reading over a set of 10 readings - which is a great result! Question still remains: Why does a set of repeated, identical height maps produce differences of up to ~20 micron? However, I've done some reading, and noted that it looks like the max. steps/mm for the stepper motors can be up to 1600. So, I've set z-steps/mm to 1600 ($102=1600, leaving $100=$101=800) and rerun the multiple (10) height maps. Remarkably, the array of differences of each height map from the average of all height maps are showing maximum and minimum values of less than 7 um, which is 3x better than what I was achieving previously. I have since done a new height map and run a PCB routing run using the new steps/mm, and the cutting depth consistency is much better. Proof of the pudding...
got z-depth readings that stayed within 1 micron
So, the probe itself works fine when the probe runs at 10 mm/min feed rate. Perfect!
Have you checked the actual feed rate during height map producing (at the top left corner, next to F/S) ?
Grbl $102 setting is the number of steps for stepper motor to move spindle for 1 mm along Z axis (source). It is conditioned by your stepper motor and driver (the number of steps per rotation) and mechanics of you machine (lead screw step). If you have 800 steps/mm and your Z-axis movement is 100% accurate it is incorrect to set $102=1600 just because your Z axis starts moving twice longer (in mm). This is it. There will be no double precision etc. And you've got better probe results just because movements along Z-axis are doubled. Use the ruler and measure the height after "G21 G91 G0 Z20" command (there should 20mm difference).
Yes, I see what you're saying, Max. Indeed, setting $102=1600 makes the z-displacement 2x what it should be. Clearly, I'm a novice with all of this. So, I've reset $102=800, set the probe speed to 2 mm/s (and Yes, F/S reads 2 (or whatever speed I set it to in the Height Map Settings, so it does seem to get it's value from there)), and rerun 10 height maps, and find much the same as previously: differences of +/- ~17 um. So, why can the machine get such repeatable precisions of 1~2 um in a single test mode (using G21 G91 G38.2 Z-40 F10, etc.), but such lousy precisions when doing a height map (even with a very slow probe speed of 2 mm/s)?
I'm using Lightburn 1.1.03, when I have a layer set to line to outline shapes or text, the laser does not turn off during "Traversal Moves". Is there a setting in the "$" commands that will fix this?
On Wed, Apr 27, 2022 at 6:44 AM 104TMR @.***> wrote:
Sorry, "12 um" should be "1~2 um"...
— Reply to this email directly, view it on GitHub https://github.com/Denvi/Candle/issues/550#issuecomment-1110902424, or unsubscribe https://github.com/notifications/unsubscribe-auth/AXPLRFWXSU7JQ7AFOPZ6LYLVHESB7ANCNFSM5UGGYDGA . You are receiving this because you are subscribed to this thread.Message ID: @.***>
Hello @chfgwd ,
Actually this is the issue tracker for Candle and in this issue we're trying to understand and solve heightmap probe inconsistency. You are welcome to participate in discussion too!
The answer for you question is $32=1 and we're expecting your instant contribution.
Best regards.
I've been doing a number of height map tests under different conditions, but haven't got to the bottom of why z-differences are so small when measured at one point, and why they start varying so much when tested between varying x-y positions. Following are some descriptions and plots of the results. Interpretations to follow. The height map scans were a 50x50mm square of 4 points (ie. each corner). The 4 depths recorded at each corner were averaged over the 10 scans, and then each of the 4 points for each scan were subtracted from the average (to create differences from the mean), and the minimum and maximum values for each scan plotted vs scan number.
- I wondered if the z-probe trigger signal may be affected by RFI from the z-stepper motor (I notice that the nearby AM radio gets heavy RFI when the stepper motors run, so I put a 0.22 uF capacitor across the A5 terminals, to dampen any RFI that could affect it. Didn't make any difference.
Image shows maximum and minimum deviations from the mean for 10 scans.
Shows min/max ranges of approx. +/- 15 um.
What does stand out from the plot is the apparently strong correlation between max and min values across the scans.
This seems to imply that when the maximums are going up, so are the minimums, and vice-versa.
If we were dealing with a purely random effect causing the variations, we would not expect to see correlation between the max and mins - instead random effects would make the values independent of each other across the scans (disagree if you like...!). - To check this further, I disconnected the capacitor, and ran three more 2x2 scans of 10 scans each (ie. 30 scans in total), to see if the max/min correlations continued. It looks like they do:
- So then I thought that perhaps a systematic offset may be associated with one axis drive (X or Y). So I did a set of 10 height scans between two points on the X-axis, and then the same for the Y-axis. Plots below show the max/min differences vs scan number, and the correlation seems to have disappeared:
- I then did several height tests at single points (ie. not moving the X/Y position), and found that the height differences reduced to 1~3 um, as found previously. This seems to be showing that when you don't move the X/Y dimensions, height values are very precise. However, as soon as you start moving around in the X/Y dimensions, variations creep in, and make the height map fairly unreliable. I don't know what else to interpret from all this. Further thoughts appreciated from anyone interested in this topic. Does anyone else experience PCB isolation routing showing unreliable depth cutting - or does everyone else's projects work perfectly and mine's the only dud?! ;-) Thanks for any thoughts...
Not sure. Their outside diameter is 8mm (hence the "8" in "TR8"?), and they move the head 4 mm with 1 turn.
On Fri, Apr 29, 2022 at 1:52 PM Jarewa @.***> wrote:
What screw do you have, TR8x2?
— Reply to this email directly, view it on GitHub https://github.com/Denvi/Candle/issues/550#issuecomment-1112852142, or unsubscribe https://github.com/notifications/unsubscribe-auth/AY3ZUNXBZT6HJCDL4XGTIVLVHNMGHANCNFSM5UGGYDGA . You are receiving this because you were mentioned.Message ID: @.***>
104TRM,
I think that you are not alone. I've been pulling my hair for a week or so because of inconsistencies in isolation routing. My case is a bit different in that sometimes I get really deep mills, and therefore wider mills. For a while the tip was barely touching the pcb and just engraving it. I made a very basic circuit to repeat the test.
In my case all the isolation maps increase in height as the probing progresses from bottom left to top right. This means that my spoilboard is not level but the thing is that I just surfaced it with a 1 inch surfacing bit.
Do you mind sharing the height map parameters that go with the charts you shared above? Also the excel you are using for graphing? I'll attempt to run the same tests generate the graphs and report back.
I'm new to CNCs but have several years of experience with 3d printers so I understand the mechanics of the steppers motors and controllers. Is a different mindset when it comes to the relationship between the tip of the bit and the surface and having two sets of coordinates to deal with.
I have a Sainsmart Genmitsu 3020 Pro Max if that's important to anyone.
Hi Madbyte3D,
I guess it's good, but also disappointing, to hear I'm not the only one with clumps of hair on the floor over this issue!
I still haven't been able to get reliable height mapping working.
The following recent results show the overcutting of tracks, despite running and applying a height map:
I'm also attaching the Excel file I use to do the height map difference comparisons:
ExampleHeightMapDifferencePlots-2x2-MultiRuns.xlsx
The way you use it is as follows:
- Run a series of height maps, saving each one with its own file name as you go. The Excel file is setup for a 2x2 (ie. 4 points) map, and I generally use a 1/4, 3/4 layout, eg. on a 100x100mm board, I would map at (25,25), (25,75), (75,75), (75,25).
- Open each height map (Notepad, Wordpad, etc.), and drag over and copy the last two lines in the height map, that contain the height measurements at each probe point (values are separated by semicolons).
- Go to ExampleHeightMapDifferencePlots, drag over the 4 cells coloured orange, and hit Paste, and the 4 height values should paste straight into the cells. Do this for all 10 height maps.
- The means of the 4 height points are calculated in AF3:AG4, and the array formulas, based on the formula in B8:C9 calculate the deviations of each height map from the mean.
- The charts show various measures of the deviations using different metrics, which I think should become fairly self explanatory when you examine the cells and look at what each chart is plotting.
It will be interesting to see how you go applying Max's formula at a single point:
G21 G91 G38.2 Z-40 F10 ( do the probe and check the last number in response [PRB:0.000,0.000,-1.031:1] Z=-1.031 )
G21 G91 G0 Z1 ( move the spindle 1mm up )
My machine showed very good repeatability using this at a single point, but it all seems to go to pieces when you start trying it across a range of x,y positions.
Let me know how you go!
Cheers...
Hello guys!
How are you going with this problem?
I've tested all in this issue, and recreate all the tests, and got the same results with no fix
So i decided to add 1/32 microsteps, and got some better results, but no success yet
So i tried to link my oscilloscope to see whats happening, i saw alot noise.
I added a pull up with divider to trigger SCL, creating a route for noise escape
and i got a nice result
3 times sequentially of the same point
This occured on 8 tries, but sometimes keep falling
but it's reduced by 80~90% of errors
Maybe it is a direction. I need work better with this resistor values
Hi Bruno, Good on you for persevering with this, and it looks like you may have a good technique for improving the height map performance.
Personally, I have moved away from routing, and instead mounted a 40W laser on the router head, and am now spray painting the blank PCB, letting it dry, then cutting through the paint layer with the laser, then drilling the through holes with the router & chuck, and then etching with H2O2 + HCl, then dissolving the paint off with acetone, and the PCB is done.
Works very well.
I might try routing again at some stage, but for now the laser/paint method allows me to produce the boards I want without too much hassle, so I’ll stick with it for now.
Cheers, Glen
On Wed, 31 Aug 2022 at 11:26 am, Bruno Pitteli Gonçalves < @.***> wrote:
Hello guys!
How are you going with this problem?
I've tested all in this issue, and recreate all the tests, and got the same results with no fix
So i decided to add 1/32 microsteps, and got some better results, but no success yet
So i tried to link my oscilloscope to see whats happening, i saw alot noise.
I added a pull up with divider to trigger SCL, creating a route for noise escape
[image: image] https://user-images.githubusercontent.com/2607849/187570251-d6e679cf-252a-405a-8021-dbb08547d1e0.png
[image: image] https://user-images.githubusercontent.com/2607849/187571239-f3dd5967-805e-4489-b4e6-97e6fe57b736.png
and i got a nice result
3 times sequentially of the same point
[image: image] https://user-images.githubusercontent.com/2607849/187571307-e9b667d2-7c7d-4ab8-90b1-537ed1f1cd58.png [image: image] https://user-images.githubusercontent.com/2607849/187571348-cc10f160-53f9-44ce-bc83-5b0514cddcbc.png [image: image] https://user-images.githubusercontent.com/2607849/187571377-23ddef17-5faa-4b04-a75c-e2866f0ed456.png
This occured on 8 tries, but sometimes keep falling
[image: image] https://user-images.githubusercontent.com/2607849/187571829-9708a4f7-356f-4a74-b46c-6a486024f1ad.png
but it's reduced by 80~90% of errors
Maybe it is a direction. I need work better with this resistor values
— Reply to this email directly, view it on GitHub https://github.com/Denvi/Candle/issues/550#issuecomment-1232342907, or unsubscribe https://github.com/notifications/unsubscribe-auth/AY3ZUNRAJJZCZGAUVVRI4JTV32YENANCNFSM5UGGYDGA . You are receiving this because you were mentioned.Message ID: @.***>
Here is a video I found interesting, A quick demonstration on how much overshoot happens when you use continuity based probing.
From what I understand, the overshoot is a result of the feedrate and stepper acceleration, The faster the acceleration and slower the feed, the less overshoot.
Nice video. But i think he show the precision of probe, not the different results of the same points (you can see he moving the plate with hands alot time)
Anyway, today i'll do some tests with acceleration, becase i done alot with speed, bot not with accel
Thank you
I found a loooooonnng discution of that https://github.com/gnea/grbl/issues/96
And the summary: https://github.com/gnea/grbl/wiki/Wiring-Limit-Switches
I ordered the optos, and will test with probe as i can, and post results
Hi Bruno, Interesting info on limit switches, although I’m not sure how this relates to the topic of this thread - ie. height map irregularities? Cheers…
On Sat, 17 Sep 2022 at 2:26 am, Bruno Pitteli Gonçalves < @.***> wrote:
I found a loooooonnng discution of that gnea/grbl#96 https://github.com/gnea/grbl/issues/96
And the summary: https://github.com/gnea/grbl/wiki/Wiring-Limit-Switches
I ordered the optos, and will test with probe as i can, and post results
— Reply to this email directly, view it on GitHub https://github.com/Denvi/Candle/issues/550#issuecomment-1249559690, or unsubscribe https://github.com/notifications/unsubscribe-auth/AY3ZUNU22TXQ6LX6PO363HLV6SNTBANCNFSM5UGGYDGA . You are receiving this because you were mentioned.Message ID: @.***>
Sorry, but i think you dont understand your first problem, and dont read the related links.
All this problem is causes by noise and the links is about cut that, isolating contacts with opto-isolation
The height map irregularities is not caused by software problem, but by measurement
Hmmm…ok, I’ll look forward to seeing what the eventual solution is. Thx…
On Mon, 19 Sep 2022 at 8:40 pm, Bruno Pitteli Gonçalves < @.***> wrote:
Sorry, but i think you dont understand your first problem, and dont read the related links.
All this problem is causes by noise and the links is about cut that, isolating contacts with opto-isolation
The height map irregularities is not caused by software problem, but by measurement
— Reply to this email directly, view it on GitHub https://github.com/Denvi/Candle/issues/550#issuecomment-1250853260, or unsubscribe https://github.com/notifications/unsubscribe-auth/AY3ZUNST4LSAFLEAEG2GU6TV7A7J3ANCNFSM5UGGYDGA . You are receiving this because you were mentioned.Message ID: @.***>
I have been reading and following your comments re the height map inconsistencies. I recently bought a 3018 pro and also tried to cut a simple circuit PCB 108x30 cm. I have almost broken every bit In the process to try and auto levelI. I am a retired hardware engineer. Sometimes the hardware is not what it should be and the battle of the software versus the hardware is still evident! I put a scope on the input if the Z probe and to my amazement saw quite a lot of bounce. The problem as I see it is on the second reading and again I'm not sure what the software is doing. But it seems that the first plunge gets a clean low upon touching the copper. Then the probe rises and descends to get a more accurate height reading. This is where lots of bounce occurs. As the smaller steps just kiss the copper, lots of high resistance connections appear causing irregular connections untill rock bottom connection occurs. A simple solution would be to step down, denounce for a given time, read, and repeat. Another solution would be to increase the pull up on the sense input to much greater than 10k maybe even 100k . I have captures of this but don't know how to share it with you. I'm happy to do so if you show me the way.🙂
Ok I missed the obvious share photo! The first image is the first plunge, 2nd and 3rd is on the final smaller incremental steps to Zero.
Hi MOT007, Those are interesting results you’ve got for the height probe resistance measurements, and Yes, likely that it has something to do with the uncertainties that are throwing the calibrations off.
However, as mentioned in an earlier post, I have found the laser scribing of painted PCB (and H2O2 +HCl etchant afterwards) to be an accurate and reliable method for creating the PCBs that I bought the 3018 to do. So I haven’t been pursuing further analysis of the variable height-mapping problem.
Your results however, do seem to indicate that the software probably could be modded somewhat to try and correct the effects you’ve identified. However, I’m not sure who would do that - unless your into playing with the source code yourself?
Cheers…
On Tue, Aug 8, 2023 at 9:41 PM Mot007 @.***> wrote:
Ok I missed the obvious share photo! The first image is the first plunge, 2nd and 3rd is on the final smaller incremental steps to Zero.
— Reply to this email directly, view it on GitHub https://github.com/Denvi/Candle/issues/550#issuecomment-1669453125, or unsubscribe https://github.com/notifications/unsubscribe-auth/AY3ZUNRW76X5NJ5US5J77A3XUIQU7ANCNFSM5UGGYDGA . You are receiving this because you were mentioned.Message ID: @.***>