Engineering: The subtle art of constantly discovering you’re doing it wrong.—@tensory
Seven months and twenty-two days ago I posted the beginning of our interferometry odyssey. We are hardly at the end, but it seems time to send out an update(!). The near continuous work on the mirror and the tests of its surface with the Bath interferometer are well described by the title of this post. When doing something for the first time, you should frequently get your head up and look around to see if you can find evidence that you are on the right track. You should also expect that you will make simple errors – it takes practice to avoid those.
From September until December our intrepid team worked on learning the art of making interferograms and, from our learning, changing configurations of the instrument and the way we made the interferograms themselves. During that time we were working in Mark’s shop where we could comfortably set up the apparatus at the Radius of Curvature (ROC), which is twice the focal length of the primary mirror. We experimented with lasers (settling on a 4.5 mW lab laser with circular beam), diverging lenses (focal lengths of 40mm, 15mm and 10mm – sticking with the 10mm), positions of the components (converged finally on components crowded as close to each other as we could get them) and camera lenses and cameras. Mark’s 20 Megapixel sensor with 50 mm lens has given the best results so far.
An interferogram is not much use without the software to analyze it. We are using Open Fringe, written by Dale Eason, and supported on the Yahoo interferometry group. During that time Dale made changes unrelated to our work that updated the software from version 12 to version 14. He was adding features and fixing bugs, and we were learning how complex this game is and why the software is so complex to deal with it. The software requires you to take a raw interferogram image through a process that sets configurations for the analysis, establishes an ellipse that corresponds to the edge of the mirror, standardizes its size and resolution, computes a fast fourier transform (FFT) of the fringes and then computes the surface of the mirror from that data. With the surface characterized, you can adjust how it is displayed and how it is analyzed and then save different kinds of data points and reports to share. Along the way, we learned many ways to get unhelpful results from the software and then corrected what we had learned with the support of the interferometry community.
In early January the team decided to move the mirror back to Steve’s shop. By this time we were becoming confident enough in our testing that we trusted what we were seeing in the test reports. I wouldn’t say that we had solved all of the problems that would cause us to get somewhat different results from test to test. But we did find enough consistency to be willing to go back to work on the mirror. The star tests we did over the summer indicated astigmatism in the figure, and the tests were showing the same result – only with the needed detail about where it was.
We had also already begun to understand the value of taking interferograms (igrams, for short) at different rotations of the big mirror. Test stands can introduce astigmatism of their own as the mirror, a big piece of glass that you think of as a solid, immutable object, bends under its own weight. Starting with this phase we began to take igrams of eight different rotations, with four igrams at each rotation point. They were, as close as we could get them, at 45 degree angles to each other. Since each igram itself has some variation due to air currents in the room and vibration of the apparatus, we averaged the four together at each rotation point, and then averaged them across all the rotations, after correcting the igrams to have the same rotation orientation. During this phase we also learned how to better standardize our analysis process so that we could compare session to session without fudging. Standardizing made it clear that we had been fast and loose with our understanding of image reversals in the instrument – all of the raw images were reversed left to right due to the star diagonal effect of the beam splitter.
We started in January thinking that we would work on the surface of the mirror in Steve’s shop, and then transport the mirror back to Mark’s shop to test. Not ideal, but attempts to get to the ROC in Steve’s shop were frustrated by that distance, about 24 feet, just a little bigger than the usual 20 foot by 20 foot shop dimension. It wasn’t feasible to do the tests as we had done the Foucault tests, wheeling the mirror on its test stand down to the end of the driveway, and setting up the light source and razor in the shop. The Bath interferometer is much more sensitive to wind currents and temperature gradients. Dismayed by the need to make this commute with the mirror in the coming months, we did the experiment of setting up the mirror on its test stand in the darkness of the shop, and then opening the door of the shop that led to the house, and setting up the apparatus at the ROC inside Steve’s kitchen. We were concerned at first that even the temperature differences and convection currents between the house and the shop would cause us problems, but after a few sessions this became a routine and workable setup. We owe a big debt to Steve’s family for accommodating this arrangement!
The conventional wisdom of the mirror-making community is that when you have astigmatism in a mirror surface, you use the big tool to return to a sphere (and, by the way, wipe out any figuring you have done so far). It’s a reboot. By unanimous vote, none of us were willing to do that. Instead, we wanted to try and make corrections more locally to see if we could remove the astigmatism and other defects that way. You can see in the January test results that, in addition to the astigmatism that shows itself in the 50% radius as variations in depth, we have a central mound that needed work. We decided to use the small tool and to gently work the surface to avoid going too far, since you can always take a little more off, but it’s hard to put glass back.
Once or twice a week we worked the surface and tested. We analyzed the igram images and decided to continue the same way. The drop box was filling with igrams and processed images and reports and we weren’t making progress. It was so puzzling we began to re-examine our process for using the software and the test results. We did a hot sponge test. The software can easily show the surface inverted, and it is difficult to tell if you are looking at the real surface or the inversion of the surface, and thus be doing bad things with your mirror work. By heating a part of the mirror with a hot sponge and taking igrams in that state, you can easily see the hump in the glass caused by that thermal expansion. In frustration, we appealed to the community. The Open Fringe author, Dale Eason, kindly analyzed a session of igrams and gave us many tips, confirmations and corrections to what we were doing. With that data in hand we embarked on two sessions of mirror work that finally resulted in changes in the central protrusion.
We are getting better now in the way we do the iterations in testing and working on the mirror and are steadily improving in our ability to get consistent surface analysis from the igrams. The April report shows that the hill is much reduced in size. If the mirror surface were a figure of revolution, then the set of profiles to the right (think of a knife edge cutting through the mirror at 16 different orientations, each separated by 22.5 degrees) would coincide with each other. We know that this image does have test stand astigmatism removed, and that some of the astigmatism is caused by what are known as unwrap errors. If you look at the October igram, you can see the fringes very clearly in the center of the disk. Toward the edges, though, it just looks fuzzy. With sufficient zooming you would see there are very tiny fringes still there, right to the edge. When there is insufficient resolution in the image, the software can lose the fringe as it analyzes it, causing an unwrap error.
Our next step is to reconfigure the Bath interferometer. Community members have suggested that a more common configuration exchanges the position of the laser and the camera. This should allow us to both remove the left-right inversion and to move just outside of ROC to get larger fringes at the edges, smaller fringes in the center. We’ve also ordered a 6 mm focal length diverging lens that should give us a bigger igram at the same distance from the apparatus. We are experimenting with the resolution of the images being analyzed, too, again trying to reduce the number of unwrap errors, this time by affecting the way they are laid down in pixels in the image. Once we have settled on the test process with these adjustments, we will continue trying to make local corrections to remove the astigmatism. Our back up plan is, of course, to spherize if we can’t get the astig to drop below acceptable levels that way.
As a team, we have certainly been getting our patience bone exercised with this work! The secret to getting the telescope to the point that we can install it in the observatory will be continuing to look for how we are doing it wrong, however subtly, and correcting our errors.