The offset of the score on the left side side of the longer design was about 0.010"
and about the same for the cut line. Harder to measure on the right side, but less than 0.010". Does the alignment always converge to the same result or do random factors cause different runs of aligning the same part give different results? Would it be possible to report how far off the alignment may be? Clearly, the method can align such that there is no noticeable difference in the two aligned paths. How does one know when the alignment is good before executing the cut?
Thank you for this additional report and the thought’s you’ve shared!
Regarding the app sharing how accurate the alignment result might be ahead of time, the app doesn’t have that ability currently, but that’s a great idea for a feature! I’ll share the idea with our product team with a note that it came from a Glowforge owner.
I also spoke with one of our engineers to see if they had any additional advice to get even better results during your Pro Passthrough prints, and they shared the feedback that wider prints are more likely to have even more accurate alignment between slices than narrow prints. If a design is particularly narrow, it may help to add some symbols to the left and right of it and set them to score during the print. Please let me know if this information helps!
Thanks for the update.
Adding fiduciary marks to facilitate that alignment makes perfect sense
as that is often done in many manufacturing processes. The next
question would be, what is the best shape to facilitate the alignment?
Lines parallel, orthogonal or diagonal to the long axis of the passthrough
direction? My naive guess would be to have an elongated rectangle along
the passthrough axis on either side of the design to be cut with diagonal lines
connecting the corners of the rectangle. Possibly a stack of these along the
length of the design would be best. At least one user was disgruntled by the
suggestion of adding extra alignment features to facilitate the alignment of
their low feature design. They suggested that it was a revisit of the snapmarks
that did not work for them. I never had the snap marks feature, but I don’t think
that this suggestion is unreasonable.
I assume that the machine vision alignment method relies upon ML methodologies
at some level and possible gets better as it is used more by the devices in the
field. That would be ideal especially how fast ML is impoving so many things.
Maybe the alignment code could suggest when the design is too sparse in
alignment elements to get a reliable alignment and suggest patterns to add to
the design before it is cut. Not as clean as one would like, but given the nature
of the optical system and random variations in system performance, I don’t think that
is unreasonable. The extra score marks would waste some material, but that is not
unheard of in building and manufacturing to get the desired functional and cosmetic outcome.
The machine vision alignment was clearly much more difficult
to implement reliably than first thought, otherwise it would have been available much earlier.
As they say, “You can do almost anything in software, all it takes is time and money…”
As far as adding additional visual features to the design or material, any shapes can be used and even drawing on the material itself with a pen has been a strategy we’ve seen people use successfully in the past.
My naive guess would be to have an elongated rectangle along the passthrough axis on either side of the design to be cut with diagonal lines connecting the corners of the rectangle. Possibly a stack of these along the length of the design would be best.
This suggestion of yours is an excellent idea! I think that if you do the above it will give you the best possible shot at getting a great result. Will you please let me know how it goes?
I also appreciate your detailed thoughts about alignment methodologies, which I’ll be sharing with our product team.
