2 potential Bed Leveling options

Discussion in 'OpenBeam Kossel Reprap and OpenBeam Kossel Pro' started by Taylor, Jul 1, 2015.

?

Which bed leveling solution are you most interested in for your machine?

  1. Aluminum/Conductive bed

    11.8%
  2. FSR

    61.8%
  3. Either one, as long as it works

    26.5%
  1. Taylor

    Taylor Moderator Staff Member

    Hello Everybody,

    We (MatterHackers) have been working on improving the bed leveling probe and G29 sequence since we got our machines in December. Like Terence said in his last update, things have definitely improved in the past few months. As usual, much of the issue is simply the combination of a bunch of little things - a little slop in the probe, variance in this or that, slight build differences, etc.

    We think 99% of the problems arise from 2 related issues. The main issue (independent variable) is that the effector isn't parallel to the bed throughout the build envelope. Because of this, the probe offset introduces error. An offset probe wouldn't be an issue if you could guarantee that the effector was always parallel to the bed. As it is, it is very difficult to get the Kossel Pro (or any delta), to stay parallel throughout the entire build envelope, and there is simply no way to properly account for the offset of the probe if the effector isn't parallel. I won't go into details here, but if you want more info/explanation as to why, send me a pm...

    So, we've been playing with potential upgrades.

    We're getting close to having an option available as an upgrade, but wanted to share what we've done thus far for those interested.

    Feedback and some additional machine hours if you want to implement one of the options on your machine would be greatly appreciated.

    Option 1: Conductive Bed/Hot end

    We had a mic-6 aluminum plate machined (250mm diameter, 5/16" thick). We placed that on top of the heated bed. We attached one lead of the probe endstop to the aluminum plate. The other lead was attached to the heater block using a ring terminal.

    2015-06-30 13.01.25.jpg


    Now the tip of the nozzle is the probe, and it doesn't really matter much if your effector is parallel or not.

    A few notes about this configuration:

    A) Because there isn't currently an easy way to save the G29 probe values like there is with Smoothie, there is some difficulty (and potential danger) with calibrating the Z-offset.

    B) Since you can't print on aluminum very easily, we run G29, and then place the glass bed on top of the plate when finished while the nozzle heats up.

    C) It's crucially important that the nozzle is clean and clear of any filament before probing. If not, it will drive into the bed until it grounds and you'll either have gouged aluminum, a bent hot end, or other issues.

    That said, this configuration produces the most consistent and repeatable results. Ideally, you would be able to run it once, save the G29 values in the eeprom/firmware, and be done with it until you need to recalibrate.

    Were exploring the idea of a conductive coating/film (specifically, indium tin oxide and flourine-doped tin oxide) for glass, as well as some other print surface options that would work with this configuration. If you have some ideas or experience with anything relevant, please let us know.


    Option 2: FSR

    We installed Force Sensing Resistors (FSR). under the heated bed. We used the kit from Ultibots.com

    We designed new plates for the heated bed that have a place for the FSR (STL attached).

    There are 3 FSR and they all plug into a single board. The board plugs into the "Bed Probe" slot on the brainwave, next to the Extruder Motor.
    BedProbe Plug.jpg
    The FSR board has LEDs that light up when triggered (1 for each sensor). It also has 3 sensitivity settings. It's a good idea to play with them a little bit to make sure they are triggering properly before you send a G29 command.

    The FSR setup requires a little tuning of the Z-Offset because the amount of pressure required to trigger is more than the conductive setup, but it still produces nice, even results.

    This is a 180mm diameter calibration print. There are few spots it's slightly smashed, but it's significantly better than the standard probe.

    2015-06-30 17.58.44.jpg

    2015-06-30 17.58.33.jpg


    This is getting a bit long, so I'll wrap it up for now.

    We'll be posting our G29 values with the standard probe as well as with these 2 options. We'll also show some prints done with each setup for comparison.

    If you have comments, suggestions, or feedback, please let us know.

    We're working to have an easy drop-in replacement as soon as possible, so your feedback will help greatly.

    Cheers,

    Taylor

    Attached Files:

  2. KDog

    KDog Member

    Wow! Thanks Taylor! This is really nice work. Great to see that there are some folks doing active development on improvements for the KP. The new force sensor kit sounds great. Am I reading it right that it comes with silicone pads that allow the sensors to work with the heated bed?

    KDog
  3. Taylor

    Taylor Moderator Staff Member

    It does come with silicone pads for heated beds. Given their placement on the KP, there isn't much heat to worry about, but they do provide a good, even surface to ensure good contact with the FSR.
  4. Def

    Def Member

    FSRs that work with the heated bed sounds like the most elegant solution.

    The conductive probing solution sounds cool, but I see a whole raft of failure modes lurking out there (you alluded to some already). So it sounds like a recipe for more of a headache in the long run than FSRs.
  5. AndyG

    AndyG Member

    Awesome stuff @Taylor . I just put in an order for an FSR kit from Utilibots. Thanks for the design file. I'll give this a shot.

    With the new K-head, FSR leveling, and hopefully some new firmware soon the KP will be solid.
  6. David Boyd

    David Boyd Active Member

    Yeah the conductive option just seems too prone to a short causing other issues.
    The other issue is you store the calibration which could be changed by a simple move of the
    printer and/or temp differences in the room.

    The force sensor sounds good but I am looking forward to seeing some of the repeatability and accuracy
    data mentioned. Also, it would be good to know the exact amount of force required to trigger the sensor.
  7. Well, at least you're asking about it ahead of time.

    Is option 3 - FSR on the effector itself - off the table?
  8. Matt

    Matt New Member

    what about using the switch built into the end effector like in this design?

    https://github.com/BatteryPower/MG-Kossel/blob/master/nozzle.stl


    works similar to shown here





    where it uses the hall effect sensor and the tip of the extruder? the bonus here is this should support the e3d j head.


    this is the style that came with my makergeeks mini kossel kit, I haven't been able to find the original source, but it seems like tooling something up for this option might be a good idea? it seems like it is a more reliable option than the FSR. and the conductive bed just seems like a recipe for disaster.
  9. Taylor

    Taylor Moderator Staff Member

    There are the same potential issues with a hall effect sensor - namely, if it's not properly adjusted, it can crash into the bed. But it certainly is another option.

    We aren't really looking at anything that would involve significant changes to the effector because it complicates things significantly. It also introduces other variables related to the printed parts and mounts, etc, that would be needed to make that change.

    It absolutely can be done, and could likely produce good repeatability, but someone else would need to take that project on.

    I'm cleaning up the spreadsheet with the data from the 3 probe styles and will be posting that shortly.
  10. Def

    Def Member

    According to the guy, he's just using a microswitch and mounting the hot end on a spring and linkage. So not a hall effect sensor unless I'm missing something.

    I don't see how a microswitch failure with the actual end effector is any different than with a bed probe. If the microswitch fails to go closed, they're both going to crash the head into the bed.
  11. I'm taking option 4 and sitting this one out. I have doubts about longevity/reliability. Good luck with whatever path you choose.

    If you redesign the bed clamp setup, please *don't* make them reversible because that really screwed the HBP owners.
  12. Taylor

    Taylor Moderator Staff Member

    Ok, so we performed 10 G29 routines with each probe method - Stock probe, FSR, and conductive bed. We knew there was a significant difference because of the print results we got, but we're a bit surprised by how much of a difference there is.

    To clarify a few things:

    1) Precision (repeatability), not accuracy, is the most important thing for our use. It would obviously be nice to have both, but as long as the probe is repeatable, we can get good leveling results.

    2) Because of the offset for the standard probe, it's impossible to determine what the nozzle position will be due to the tilting of the effector. We have no good way of dealing with this other than using the tip of the nozzle as the probe position. Because of this, when we look at the the standard probe data points - already clearly the least precise of the 3 - it becomes obvious why there has been such wild variation from machine to machine and print to print.

    3) The bed was at different heights for each method. The Standard used the stock glass plate. The FSR used the glass plate as well, but with no probe offset. The Conductive had the 5/16" aluminum plate. So the Raw data look strange compared side by side, but we normalized it for the comparison in the graph by subtracting average of each data set from the reported reading for the given point.

    For example, in Round 1 of the Standard probe, the first measured point, X-64 Y-64, reported 9.53 (shown in Standard Probe Raw Data sheet cell C3.
    The average of all 10 Rounds for point 1 was 9.577 (cell M3)
    9.53 - 9.577 = -.047 (cell C5 in Normalized Data sheet)

    And now the data...

    Standard Probe raw data
    Standard - Raw.png

    FSR Raw
    FSR - Raw.png

    Conductive Raw
    Conductive - Raw.png


    Normalized Data

    Standard
    std norm.png

    FSR
    FSR norm.png

    Conductive
    cond norm.png

    And now the fun graph
    graph.png

    Preliminary Conclusions:

    Basically, what we're seeing is that with the standard probe, with 27 points in the G29 procedure, there's a very high likelihood to have at least 4 points greater than 1 standard deviation above the mean and at least 4 points less than -1 std deviation.

    That means it's very likely to have a difference of at least .1mm across your build surface.
    It will be common to have a difference of ~.2mm.

    Now, this difference could be between points right next to each other which would be a drastic change, or it could be spread across the width of the bed. It just depends on the specific measured points in any given G29 sequence.

    But put simply, .2mm is a whole layer. Obviously a significant issue. regardless if it's between 2 close points or across the diameter of the G29 grid.


    The FSRs are about twice as precise as the standard probe.

    So it's very likely to have about .05mm difference.
    And common to have .1mm.

    .05mm is within the tolerances of extruding filament on FFF machines, so the large majority of the time, the FSRs will be totally acceptable, but there's still a decent chance on any given G29 that it will be a bit off.

    The Conductive setup is remarkably precise.
    1 std dev is .0082
    That means there's a 99.7% chance of having a .05mm (or less) difference.

    This would obviously be fantastic for every machine.


    It wouldn't be difficult to perform 25 or even 50 G29s with the standard probe, and then take the average of those and use them as the values in the firmware. This would produce essentially the same results, statistically, as the conductive setup. It wouldn't, however, account for the tilt in the effector. The probe would have to be changed, like the video posted a above, to probe at the tip of the nozzle, in order to get an even print across the bed.

    All 3 probes would benefit (though it would be minimal for the conductive) if the firmware allowed for an easy way to probe each point 25-30 times, take the average, and then store those permanently.

    I've attached the spreadsheet if you want to look at the data yourself.

    Thoughts/input appreciated.

    We're happy with the results of our testing and we're confident we're close to having a solution that will make every Kossel Pro "just work" with minimal effort/modification.

    Have a happy 4th everyone.

    Attached Files:

    Last edited: Jul 3, 2015
  13. Terence Tam

    Terence Tam Mr. OpenBeam Staff Member Vendor

    You may have better results with the FSR setup, if the probe was rigidly coupled to the end effector. It is held in place currently by a wire form spring to take up tolerances in the grove mount manufacturing.

    -=- Terence
  14. Taylor

    Taylor Moderator Staff Member

    Yes, we had already done that. Printed a groove mount piece so that it's easy to access the PTC fitting.

    2015-07-02 20.12.04.jpg
  15. That was my concern. The FSR is better, but only somewhat ... incrementally so. In a broad sense, the FSR is "kinda" better than the stock probe, and you'd think it could do better.

    Is it possible that the silicone pads are a source of variance? Every time you tap the bed, the entire thing rearranges slightly?

    Good data though.

    Hate to be the guy that goes, "but you should have tried!", but if the conductive probe really is that precise, I wonder if it could be used to test this. Basically, put the metal bed on there simultaneously with the FSR bet setup, and run both probes against it. Specifically, run the conductive probe, then the FSR probe, and alternate, and see what happens to the conductive probe's data.
  16. Taylor

    Taylor Moderator Staff Member

    The conductive probe testing was run on top of the FSRs but the FSRs were not run on top of the mic6 plate. We have a Rostock Max setup with the FSRs and conductive plate as well and ran the probing. That machine only has an aluminum bed and the FSRs performed virtually identically.

    As for the results, while it is basically an incremental improvement in precision of the probe, it is probably an order of magnitude better than the standard probe because it accounts for effector tilt. There's no way to measure/quantify this effectively because we would have to measure extruded filament extruded on the bed, but it is very easily seen. The FSRs have passed the 'eye test' on everything we printed.

    Further, the only thing 'necessary' to make the FSRs as good as the conductive is a way to easily store the g29 grid to eeprom. We could then run a simple macro that would probe each spot 15-20 times, take the average, and store them. Then the user would not need to run G29 before each print, only whenever they changed something, moved the machine, or needed to recalibrate.

    As it is, statistically, there's around an 80% chance that any given G29 with the FSRs will produce error within the acceptable tolerances.

    While not ideal, it would be very easy to hit print, watch the first layer, and cancel/restart if it wasn't right. Chances are it would be good on the 2nd attempt.
  17. Def

    Def Member

    The distribution of the FSR data is pretty good, to where you could run a G29 say 30 or 40 times, and throw out the lowest and highest 5 or 10 values at each point. This will give you a chopped off roughly normal distribution, and should have a very low standard deviation.

    Basically take a large enough sample size that you can then throw out the outliers and have good confidence you're still close to the actual data point.
  18. Interesting; I guess it depends on whether or not the conductive probe puts as much force on the platform as the FSR probe while it's doing it's job.

    Sounds reasonable; I didn't consider statistically massaging the results.
  19. Taylor

    Taylor Moderator Staff Member

    The conductive definitely does not put as much force on the bed because as soon as there's physical connection, the switch is closed. The FSR lowest sensitivity is .80v difference from ambient. If im interpreting the FSR data coorrectly, that translates to approximately .5psi on the bed. So there's definitely a difference in the pressure on the bed.
  20. bytecounter

    bytecounter Member

    @Taylor Is the groove mount part you're using available/released?

Share This Page