Posted 02 July 2012 - 07:10 AM
Posted 03 July 2012 - 03:32 PM
I'll say a few things here, since you asked... But I do have a discussion group somewhere on-line about the whole concept.
The machines I'm building are quite complicated in detail but the overall principle is fairly simple - iterative feedback systems.
On the grinding/polishing/figuring side I'm constructing a set of machines with capacity to 24" on the first two and about 30" on the next pair. They fully automate all 3 stages of the process, but the interesting part is figuring.
On the testing side two machines suffice for the time being - a computer controlled maskless Foucault tester that gives quick hundred-point readings across one axis of the mirror, and a Shack-cube interferometer that provides full surface mapping of the finished mirrors. Since my mirrors don't show any irregularities of revolution (at the go-finish stage anyway) testing on a single axis is sufficient to bring them to completion.
The key here is iteration and using the results of testing to feedback into the next stage of figuring. With complete computer control of the tool motion in time and space, as well as high resolution information about the developing curve on the glass, it's possible to make small adjustments to the figuring strokes that converge on the desired figure, to well below any practical limits of performance. Plus the figuring system can learn by experience, becoming able to make its own decisions about procedure.
It's pretty much what I've been doing by hand and eye for years, only taken to the next level.
Veritas Optics - new name for a new approach. "Veritas" in this case is chosen for the root meaning of truth, as in verily or verity. Production and testing that supports a consistent product that I've already seen a few times making them "the old fashioned way."
Once in a while a perfect mirror came along, and those are fairly easy to detect on the bench. They take any expectation about optical performance and turn it on its ear. It's something I'm doing because I think it can be done. The whole project is intended to capture these somewhat mythical mirrors and turn them out much more easily.
Posted 04 July 2012 - 07:49 AM
I searched for a thread on your new robotic mirror making system and came up empty. Could you provide a link to where this information has been discussed? Thanks, Bob
Posted 04 July 2012 - 07:14 PM
Posted 05 July 2012 - 05:47 AM
I found some information you wrote in September 2009 about where you have been going for the past three years.
"You raise some good points, and some of them have been discussed from time to
time. I know there was a figuring simulator project discussed on the ATM List a
I can only speak from my own experience approaching these topics, so in no
particular order, here goes:
First (and in pride of place) is deterministic figuring. This was foremost in
my mind when I started working out a methodology to follow to reach an easily
stated goal - mirrors figured and measured entirely by machine, with minimal
human intervention. Any practical system has to make an estimate of how to
proceed for the next step based on the measured results of the last step. This
is crucial, as it eliminates a great amount of the uncertainty you're talking
about. The system is set in motion by applying the "default" figuring patterns
and measuring the first results.
Such a system doesn't need to be AI, but it needs to be at least a dim AI - an
expert system of some sort. It can add to its repertoire by making small
experiments and integrating the results into its map of possibilities. To my
mind, it should suggest what it thinks the right thing to do next is, and be
instructed otherwise if wrong. But that's easily handled. I leave out the
topic of how to represent the instructions, at this point.
Second, no less important though, is that the figuring method itself has to lend
itself to deterministic manipulation easily. This for me was the really tough
part. I could see that I didn't want to try to emulate a typical MOM approach
(although one could, certainly). What I wanted was a method that adapts readily
to an XY(Z) table, one that can implement any arbitrary stroke as required.
I've figured likely hundreds of mirrors now entirely by hand using the method I
came up with, which involves sub-diameter polishers with fixed weight loads,
moved in a progressive (expanding) pattern over distance against time, for each
figuring session, followed by rotation of the mirror over a 24-point orientation
during the entire figuring duration (4-8 hours or so of polisher-on-mirror
The method isn't easy to do, though, and requires a lot of attention to do
manually. I've proven it out, and if properly applied it works to perfection.
It doesn't involve any deliberate randomization, ie, I'm applying it as
accurately as humanly possible. The machine is precise to about 1 mm and 1 ms
so I expect it to do a much better job.
Third, simulating the figuring process. Although I understand the reason, I'm
replacing this by determinism. This requires consistency in polisher
preparation, temperature control, slurry control, and the like, but it only has
to get "so close" to be useful. The machine on the next iteration, always close
in time and condition to the previous one, can correct for drift in the working
qualities of the materials. That is, it pretty much knows what to do under
optimal conditions to get a particular result, and it goes there first. No muss,
no fuss - default mode is that it does what I'd do looking at a Ronchi curve or
Fourth, I'm not sure how to put this one though. The key to me is that the
figuring method adopted must work the entire mirror consistently, and
progressively, to be successful. I allow and expect for the center on fast
mirrors to require touchup, but that's a minor issue. I require the zonal
progress to move along consistently and at expected rates that end with the
mirror outside the center reaching full correction simultaneously, ideally.
Anything else requires going in with small polishers or polisher-edge work
(localized pressure) to finesse, and this is not what the machine is good at.
What the machine is good at is adjusting the overall pattern in time and 2D
space to lead or lag on certain areas, measure the results, and adjust the
deviations to bring leading or lagging zones into alignment.
Fifth and I think last for now, in an XY(Z) table (the Z is subdiam polisher
rotation under machine control) the mirror does not need to move, if held on a
truly flat support with minimal flotation (the ubiquitous throw-rug backer /
foam mesh shelf liner). The machine can rotate the figuring pattern upon the
mirror, emulating rotation while working by hand at a bench. Astigmatism
doesn't result from this.
I don't doubt that there are multiple solutions to doing this kind of work,
just as there are multiple telescope designs.
I thought it would be helpful to provide a context for the very interesting direction you seem to be going in. Keep up the good work. Many of us revel in watching others push the envelope from the known to the unknown. Kind of reminds me of the search for the Higgs Boson<g>.