Project Logs

Collapse

• v3 Testing

  mulcmu • 05/15/2024 at 01:37 • 0 comments

  After some tweaks to the touch sensor measurement interval and threshold some more probe data was collected.  For 88% of the probes the range was 0.00 or 0.01 and 99% were 0.02 or less.  Much improved.  The mechanical changes are confounded by the software change from ESP32 to ESP32-S3 processor.  

  As the copper pad engages to the bed the mylar does a good job holding it flat until the flex in the mylar causes it to touch the bed too.  With the copper pad mounted to the bottom of mylar, each arm contacts the bed away from the center point, resulting in the mylar wanting to raise the pad in the center.  Might need to change arm mounting holes to get mylar / nozzle height to sweet spot.

• Mylar V3

  mulcmu • 05/10/2024 at 23:51 • 0 comments

  With the ESP32 dev module hanging off the umbilical and a long wire lead leading down to the copper probe, it was difficult to avoid small movements of the probe wire during operation.  The small changes in wire position and long wire was negatively impacting sensitivity and repeatability.  The copper foil also seemed to be wearing/bunching from the nozzle contact.

  V3 redesigned the mylar mount to use a 15x15x0.3mm shim.  This is pretty thin but rigid enough to keep from bending.  The shim mounts on the bottom of the mylar with double faced tape.  There is a hole in the mylar for the nozzle to pass through an make contact.  An ESP32-S3-Zero dev module is mounted to the rear mylar arm.  This lets the wire lead from the board to the probe be short and independent of any umbilical motion during movements.
  

  The Arduino code was updated for testing.  The S3 touch sensors operate a bit differently than the plain ESP32.  No major problems getting the code back up and running, now with an RGB addressable led for feedback.  Initial testing looks like there is improvement.  Next step is to power the dev board via the printer probe connector.  The USB-C cable hanging off the side of the mylar isn't ideal.

• Lots of probes later

  mulcmu • 03/16/2024 at 21:04 • 0 comments

  Working on getting the printer calibration optimized.  Probing a hexagon grid, range of 10x samples at each point is still a little higher than desired.  The optimization logic seems to be working much better than with the stock probe.
  

• Sensitivity

  mulcmu • 03/09/2024 at 17:33 • 0 comments

  Took a bit of a break to use printer for making stuff..

  After reinstalling probe there seemed to be a regression in performance.  Initial testing started using glass plate on top of the magnetic PEI to make sure bed surface was planar as  possible for testing some calibration ideas.  Center probed okay with higher variance between probe, edge probes were very noisy and triggered the probe consistency limit.  Went through some troubleshooting to figure out root cause.  Cleaned nozzle better, cleaned foil, cleaned bed, tried without glass plate, double checked all connections to the ESP, removed sock...  Variance still persisted.  Tried probe with a new piece of foil on top of the existing foil.  There was marking under the nozzle location but no remotely concerning damage.  Anyway the extra foil seemed to get probe performing like it was previously.  Not sure if any of the other actions were necessary too.  So still prototype is still pretty fiddly at this stage but at least there wasn't a significant difference between results probing on PEI spring steel and glass plate.

• Mylar V2

  mulcmu • 02/28/2024 at 03:05 • 0 comments

  A Y-shaped section of black mylar was cut on the laser cutter with some holes to attached to clips printed from PLA.  A section of copper foil tape was placed in center of the Y for contact with the nozzle.  With the probe clipped onto the printer, the copper foil hovers a few mm below the nozzle.  When it approaches the bed the mylar levels and conforms to the bed surface nice and flat while the nozzle approaches.  Minimal extra external force on the effector is an added bonus.

  The slow drift was seemed to converge at a stable value over hundreds and hundreds of relative probes in the same location.  The probe speed was 1 mm/sec and the lift after contact was 1 mm as well so during the probe it just moves up and down by 1mm.  Previously the initial few probes were found to be consistent after homing the printer.  Instead of homing the printer was commanded back to near the top and back down.  Nearly same travel as home movement but without triggering or resetting stepper positions.  This process cleared the observed drift and returned back to the initial probe reading.  Smaller changes in Z height as well as just travel in X and Y also reset the drift to initial value.

  The copper foil and the mylar are about 0.3 mm thick so a smidge thicker than paper 0.15mm.  A hole in the mylar for the nozzle to pass through with two layers of the foil tape on the bottom of the mylar would get this to be truly a digital paper test.  The mylar should be good for contact with heated bed, contact with heated nozzle might cause problems with plastic ooze and deformation from the heat.

• Iterations

  mulcmu • 02/24/2024 at 15:32 • 0 comments

  Went through several iterations of testing the accuracy with minor tweaks to the Arduino ESP32 code.  Activation by ISR, different types and levels of filtering, and changes to the touchRead() default configuration.  The default (0x1000) measurement clock (ULP) cycles should result in about 50 measurements per 0.025 mm interval while lowering.  This was decreased by factor of 8 (to 0x200) for 400 measurements per 0.025 mm interval.  The quicker measurements reduce the magnitude of the touchRead() return value but was high enough to still distinguish between triggered and not triggered.  

  During the first 2 or 3 probe accuracy cycles with 100 sample points the trigger height will slowly drift downward by 0.01 to 0.02 mm for each 100 cycle iteration.  This was also observed with a micro switch fixed to the bed and actuated by the nozzle.  At this point it seems like the capacitive sensor is triggering consistent when compared to a physical microswitch.  Not sure yet if this is a klipper quirk, mechanical or electrical issue.  Haven't run it enough times to see if it ever stops drifting lower.  After home command it does revert to the original height. 

  Anyway, starting to rework the clip design to use copper foil tape and mylar sheet for the probe.  This will get probe offset (thickness) down to 0.20 mm range.  There are still a few things to improve the code like doing measurement and filtering on separate core to improve sampling frequency with the trigger logic and serial output on the other core and tweaking the high and low charge voltages.

• Side Project

  mulcmu • 02/17/2024 at 02:00 • 0 comments

  A digital bubble level was built to measure the platform tilt over the travel range.  This consists of an ESP32 dev board with 1.9" LCD and Sparkfun ICM-20948 breakout board.  The base was small enough to mount on the effector platform (with hot end removed).  Some python code was used to have Klipper move the platform around and record the x and y tilt from the bubble level.  
  

• Initial Testing

  mulcmu • 02/03/2024 at 16:32 • 0 comments

  An Arduino ESP32 was setup using touchRead() in the main loop to toggle the onboard led and stream the read value out over Serial.  A single 15mm x 15mm x 1mm pad was tested with dial indicator to get a feel for the trigger range.  The led was toggled consistently at same location, the range of motion was less than 0.001-inch, maybe 3 to 5 tenths, so about 0.01 mm.  Looking good so far...

  Code was modified to use an ESP32 output to trigger the probe input on the printer.  Testing started with the clip mount with an ungrounded 40mm x 40mm x .5mm pad on the bottom of the 3mm thick adhesive foam.  The touchRead() values were plotted with serial plotter, 5ms delay was added to slow down the stream of data.  Manually controlling the z-height, the probe would drop 2 to 3 units with a quick transition from values above 50 to values less than 20.  Cycling z-offset ±0.0005 triggers output on and off.  This was very consistent and repeatable.

  Testing was started with Klipper PROBE_ACCURACY command.  The range was way higher than expected 0.1 mm or more.  Testing to get this to improve was started.  The delay in the loop was removed, different trigger points were tested, the output was latched to stay on after initial trigger for 250 ms, different filtering was tried, grounding the bottom pad was tested, different springs on the probe holder were tried, rigidly attaching the sensors to the bed, eliminating the 3mm foam section, changing probe speeds and retract distances, ground connection to the nozzle, removing silicone sock, cleaning nozzle, unplugging heatsink cooling fan.  This improved things, 10 probes would typically trigger over a range of 0.02 to 0.04.  Consistent 0.00 to 0.01 mm is the goal.

  Some testing with the stock probe micro switch and another smd microswitch soldered to perfboard and actuated by nozzle seemed to perform a bit better,  Pretty consistent sometimes but sometimes range was in 0.02 to 0.04 as well.
  

  I cleared out the old delta calibration and started from scratch with DELTA_CALIBRATE using the clip probe to see how things changed.  When probing near the C tower the range increased drastically when the wires bumped into the tower.  Rerouted the wires to eliminate the contact and keep them as fixed as possible.

  Testing still in progress, things to look into:  trace on the trigger voltage (is it getting pulled down quickly / consistently), recheck position bias, (is this digital paper making it better or gone), test probe position bias with the smd switch (less parts and code if this  would work), put Klipper on another delta and test (maybe something odd mechanically going on).

• Prototype

  mulcmu • 01/28/2024 at 04:56 • 0 comments

  A prototype design for the probe was created that will clip loosely around the effector platform.  Springs will force the probe assembly downward into the bed as the platform is lowered.  A 15mm x 15mm copper plate will be used as a capacitive sensor.  A 3mm thick foam pad will connect the plate to the 3d printed clip.  Springs will engage under the effector platform and force the foam pad to contact the bed as the platform is lowered.  Once the platform takes up the spring travel the nozzle will contact the copper plate and triggering capacitive sensor.

View all 9 project logs